Removed the libs directory containing win32 compiled versions of libpng, libtiff and liblcms. Added a thirdparty directory to include main source files of libtiff, libpng, libz and liblcms to enable support of these formats in the codec executables. CMake will try to statically build these libraries if they are not found on the system. Note that these third party libraries are not required to build libopenjpeg (which has no dependencies).

26 files changed, 29722 insertions, 0 deletions

diff --git a/thirdparty/liblcms2/src/cmscam02.c b/thirdparty/liblcms2/src/cmscam02.c
new file mode 100644
index 00000000..08eea16a
--- /dev/null
+++ b/thirdparty/liblcms2/src/cmscam02.c
@@ -0,0 +1,483 @@
+//---------------------------------------------------------------------------------
+//
+//  Little Color Management System
+//  Copyright (c) 1998-2010 Marti Maria Saguer
+//
+// Permission is hereby granted, free of charge, to any person obtaining 
+// a copy of this software and associated documentation files (the "Software"), 
+// to deal in the Software without restriction, including without limitation 
+// the rights to use, copy, modify, merge, publish, distribute, sublicense, 
+// and/or sell copies of the Software, and to permit persons to whom the Software 
+// is furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in 
+// all copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 
+// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO 
+// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE 
+// LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 
+// OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 
+// WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+//
+//---------------------------------------------------------------------------------
+//
+
+#include "lcms2_internal.h"
+
+// CIECAM 02 appearance model. Many thanks to Jordi Vilar for the debugging.
+
+// ---------- Implementation --------------------------------------------
+
+typedef struct  {
+    
+    cmsFloat64Number XYZ[3];
+    cmsFloat64Number RGB[3];
+    cmsFloat64Number RGBc[3];
+    cmsFloat64Number RGBp[3];
+    cmsFloat64Number RGBpa[3];
+    cmsFloat64Number a, b, h, e, H, A, J, Q, s, t, C, M;
+    cmsFloat64Number abC[2];
+    cmsFloat64Number abs[2];
+    cmsFloat64Number abM[2];
+    
+} CAM02COLOR;
+
+typedef struct  {
+    
+    CAM02COLOR adoptedWhite;
+    cmsFloat64Number LA, Yb;
+    cmsFloat64Number F, c, Nc;
+    cmsUInt32Number surround;
+    cmsFloat64Number n, Nbb, Ncb, z, FL, D;
+    
+	cmsContext ContextID;
+
+} cmsCIECAM02;
+
+
+static
+cmsFloat64Number compute_n(cmsCIECAM02* pMod) 
+{
+    return (pMod -> Yb / pMod -> adoptedWhite.XYZ[1]);
+}
+
+static
+cmsFloat64Number compute_z(cmsCIECAM02* pMod) 
+{
+    return (1.48 + pow(pMod -> n, 0.5));
+}
+
+static
+cmsFloat64Number computeNbb(cmsCIECAM02* pMod) 
+{
+    return (0.725 * pow((1.0 / pMod -> n), 0.2));
+}
+
+static
+cmsFloat64Number computeFL(cmsCIECAM02* pMod) 
+{
+    cmsFloat64Number k, FL;
+    
+    k = 1.0 / ((5.0 * pMod->LA) + 1.0);
+    FL = 0.2 * pow(k, 4.0) * (5.0 * pMod->LA) + 0.1 *
+        (pow((1.0 - pow(k, 4.0)), 2.0)) *
+        (pow((5.0 * pMod->LA), (1.0 / 3.0)));
+    
+    return FL;
+}
+
+static    
+cmsFloat64Number computeD(cmsCIECAM02* pMod) 
+{
+    cmsFloat64Number D;
+
+    D = pMod->F - (1.0/3.6)*(exp(((-pMod ->LA-42) / 92.0)));
+
+    return D;
+}
+
+
+static
+CAM02COLOR XYZtoCAT02(CAM02COLOR clr) 
+{
+    clr.RGB[0] = (clr.XYZ[0] *  0.7328) + (clr.XYZ[1] *  0.4296) + (clr.XYZ[2] * -0.1624);
+    clr.RGB[1] = (clr.XYZ[0] * -0.7036) + (clr.XYZ[1] *  1.6975) + (clr.XYZ[2] *  0.0061);
+    clr.RGB[2] = (clr.XYZ[0] *  0.0030) + (clr.XYZ[1] *  0.0136) + (clr.XYZ[2] *  0.9834);
+    
+    return clr;
+}
+
+static
+CAM02COLOR ChromaticAdaptation(CAM02COLOR clr, cmsCIECAM02* pMod) 
+{
+    cmsUInt32Number i;
+
+    for (i = 0; i < 3; i++) {
+        clr.RGBc[i] = ((pMod -> adoptedWhite.XYZ[1] *
+            (pMod->D / pMod -> adoptedWhite.RGB[i])) +
+            (1.0 - pMod->D)) * clr.RGB[i];
+    }
+
+    return clr; 
+}
+
+
+static
+CAM02COLOR CAT02toHPE(CAM02COLOR clr) 
+{
+    cmsFloat64Number M[9];
+        
+    M[0] =(( 0.38971 *  1.096124) + (0.68898 * 0.454369) + (-0.07868 * -0.009628));
+    M[1] =(( 0.38971 * -0.278869) + (0.68898 * 0.473533) + (-0.07868 * -0.005698));
+    M[2] =(( 0.38971 *  0.182745) + (0.68898 * 0.072098) + (-0.07868 *  1.015326));
+    M[3] =((-0.22981 *  1.096124) + (1.18340 * 0.454369) + ( 0.04641 * -0.009628));
+    M[4] =((-0.22981 * -0.278869) + (1.18340 * 0.473533) + ( 0.04641 * -0.005698));
+    M[5] =((-0.22981 *  0.182745) + (1.18340 * 0.072098) + ( 0.04641 *  1.015326));
+    M[6] =(-0.009628);
+    M[7] =(-0.005698);
+    M[8] =( 1.015326);
+    
+    clr.RGBp[0] = (clr.RGBc[0] * M[0]) +  (clr.RGBc[1] * M[1]) + (clr.RGBc[2] * M[2]);
+    clr.RGBp[1] = (clr.RGBc[0] * M[3]) +  (clr.RGBc[1] * M[4]) + (clr.RGBc[2] * M[5]);
+    clr.RGBp[2] = (clr.RGBc[0] * M[6]) +  (clr.RGBc[1] * M[7]) + (clr.RGBc[2] * M[8]);
+    
+    return  clr;
+}
+
+static
+CAM02COLOR NonlinearCompression(CAM02COLOR clr, cmsCIECAM02* pMod) 
+{
+    cmsUInt32Number i;
+    cmsFloat64Number temp;
+
+    for (i = 0; i < 3; i++) {
+        if (clr.RGBp[i] < 0) {
+
+            temp = pow((-1.0 * pMod->FL * clr.RGBp[i] / 100.0), 0.42);
+            clr.RGBpa[i] = (-1.0 * 400.0 * temp) / (temp + 27.13) + 0.1;
+        }
+        else {
+            temp = pow((pMod->FL * clr.RGBp[i] / 100.0), 0.42);
+            clr.RGBpa[i] = (400.0 * temp) / (temp + 27.13) + 0.1;
+        }
+    }
+    
+    clr.A = (((2.0 * clr.RGBpa[0]) + clr.RGBpa[1] + 
+        (clr.RGBpa[2] / 20.0)) - 0.305) * pMod->Nbb;
+
+    return clr;
+}
+
+static
+CAM02COLOR ComputeCorrelates(CAM02COLOR clr, cmsCIECAM02* pMod) 
+{
+    cmsFloat64Number a, b, temp, e, t, r2d, d2r;
+
+    a = clr.RGBpa[0] - (12.0 * clr.RGBpa[1] / 11.0) + (clr.RGBpa[2] / 11.0);
+    b = (clr.RGBpa[0] + clr.RGBpa[1] - (2.0 * clr.RGBpa[2])) / 9.0;
+    
+    r2d = (180.0 / 3.141592654);
+    if (a == 0) {
+        if (b == 0)     clr.h = 0;
+        else if (b > 0) clr.h = 90;
+        else            clr.h = 270;
+    }
+    else if (a > 0) {
+        temp = b / a;
+        if (b > 0)       clr.h = (r2d * atan(temp));
+        else if (b == 0) clr.h = 0;
+        else             clr.h = (r2d * atan(temp)) + 360;
+    }
+    else {
+        temp = b / a;
+        clr.h = (r2d * atan(temp)) + 180;
+    }
+    
+    d2r = (3.141592654 / 180.0);
+    e = ((12500.0 / 13.0) * pMod->Nc * pMod->Ncb) * 
+        (cos((clr.h * d2r + 2.0)) + 3.8); 
+    
+    if (clr.h < 20.14) {
+        temp = ((clr.h + 122.47)/1.2) + ((20.14 - clr.h)/0.8);
+        clr.H = 300 + (100*((clr.h + 122.47)/1.2)) / temp;
+    }
+    else if (clr.h < 90.0) {
+        temp = ((clr.h - 20.14)/0.8) + ((90.00 - clr.h)/0.7);
+        clr.H = (100*((clr.h - 20.14)/0.8)) / temp;
+    }
+    else if (clr.h < 164.25) {
+        temp = ((clr.h - 90.00)/0.7) + ((164.25 - clr.h)/1.0);
+        clr.H = 100 + ((100*((clr.h - 90.00)/0.7)) / temp);
+    }
+    else if (clr.h < 237.53) {
+        temp = ((clr.h - 164.25)/1.0) + ((237.53 - clr.h)/1.2);
+        clr.H = 200 + ((100*((clr.h - 164.25)/1.0)) / temp);
+    }
+    else {
+        temp = ((clr.h - 237.53)/1.2) + ((360 - clr.h + 20.14)/0.8);
+        clr.H = 300 + ((100*((clr.h - 237.53)/1.2)) / temp);
+    }
+    
+    clr.J = 100.0 * pow((clr.A / pMod->adoptedWhite.A), 
+        (pMod->c * pMod->z));
+
+    clr.Q = (4.0 / pMod->c) * pow((clr.J / 100.0), 0.5) *
+        (pMod->adoptedWhite.A + 4.0) * pow(pMod->FL, 0.25);
+    
+    t = (e * pow(((a * a) + (b * b)), 0.5)) /
+        (clr.RGBpa[0] + clr.RGBpa[1] + 
+        ((21.0 / 20.0) * clr.RGBpa[2]));
+
+    clr.C = pow(t, 0.9) * pow((clr.J / 100.0), 0.5) *
+        pow((1.64 - pow(0.29, pMod->n)), 0.73);
+
+    clr.M = clr.C * pow(pMod->FL, 0.25);
+    clr.s = 100.0 * pow((clr.M / clr.Q), 0.5);
+    
+    return clr;
+}
+
+
+static
+CAM02COLOR InverseCorrelates(CAM02COLOR clr, cmsCIECAM02* pMod) 
+{
+    
+    cmsFloat64Number t, e, p1, p2, p3, p4, p5, hr, d2r;
+    d2r = 3.141592654 / 180.0;
+    
+    t = pow( (clr.C / (pow((clr.J / 100.0), 0.5) *
+        (pow((1.64 - pow(0.29, pMod->n)), 0.73)))), 
+        (1.0 / 0.9) );
+    e = ((12500.0 / 13.0) * pMod->Nc * pMod->Ncb) *
+        (cos((clr.h * d2r + 2.0)) + 3.8);
+    
+    clr.A = pMod->adoptedWhite.A * pow(
+           (clr.J / 100.0),
+           (1.0 / (pMod->c * pMod->z)));
+    
+    p1 = e / t;
+    p2 = (clr.A / pMod->Nbb) + 0.305;
+    p3 = 21.0 / 20.0;
+    
+    hr = clr.h * d2r;
+    
+    if (fabs(sin(hr)) >= fabs(cos(hr))) {
+        p4 = p1 / sin(hr);
+        clr.b = (p2 * (2.0 + p3) * (460.0 / 1403.0)) /
+            (p4 + (2.0 + p3) * (220.0 / 1403.0) *
+            (cos(hr) / sin(hr)) - (27.0 / 1403.0) +
+            p3 * (6300.0 / 1403.0));
+        clr.a = clr.b * (cos(hr) / sin(hr));
+    }
+    else {
+        p5 = p1 / cos(hr);
+        clr.a = (p2 * (2.0 + p3) * (460.0 / 1403.0)) /
+            (p5 + (2.0 + p3) * (220.0 / 1403.0) -
+            ((27.0 / 1403.0) - p3 * (6300.0 / 1403.0)) *
+            (sin(hr) / cos(hr)));
+        clr.b = clr.a * (sin(hr) / cos(hr));
+    }
+    
+    clr.RGBpa[0] = ((460.0 / 1403.0) * p2) + 
+              ((451.0 / 1403.0) * clr.a) +
+              ((288.0 / 1403.0) * clr.b);
+    clr.RGBpa[1] = ((460.0 / 1403.0) * p2) - 
+              ((891.0 / 1403.0) * clr.a) -
+              ((261.0 / 1403.0) * clr.b);
+    clr.RGBpa[2] = ((460.0 / 1403.0) * p2) -
+              ((220.0 / 1403.0) * clr.a) -
+              ((6300.0 / 1403.0) * clr.b);
+    
+    return clr;
+}
+
+static
+CAM02COLOR InverseNonlinearity(CAM02COLOR clr, cmsCIECAM02* pMod)
+{
+    cmsUInt32Number i;
+    cmsFloat64Number c1;
+    
+    for (i = 0; i < 3; i++) {
+        if ((clr.RGBpa[i] - 0.1) < 0) c1 = -1;
+        else                               c1 = 1;
+        clr.RGBp[i] = c1 * (100.0 / pMod->FL) *
+            pow(((27.13 * fabs(clr.RGBpa[i] - 0.1)) /
+            (400.0 - fabs(clr.RGBpa[i] - 0.1))),
+            (1.0 / 0.42));
+    }
+    
+    return clr;
+}
+
+static
+CAM02COLOR HPEtoCAT02(CAM02COLOR clr) 
+{
+    cmsFloat64Number M[9];
+    
+    M[0] = (( 0.7328 *  1.910197) + (0.4296 * 0.370950));
+    M[1] = (( 0.7328 * -1.112124) + (0.4296 * 0.629054));
+    M[2] = (( 0.7328 *  0.201908) + (0.4296 * 0.000008) - 0.1624);
+    M[3] = ((-0.7036 *  1.910197) + (1.6975 * 0.370950));
+    M[4] = ((-0.7036 * -1.112124) + (1.6975 * 0.629054));
+    M[5] = ((-0.7036 *  0.201908) + (1.6975 * 0.000008) + 0.0061);
+    M[6] = (( 0.0030 *  1.910197) + (0.0136 * 0.370950));
+    M[7] = (( 0.0030 * -1.112124) + (0.0136 * 0.629054));
+    M[8] = (( 0.0030 *  0.201908) + (0.0136 * 0.000008) + 0.9834);;
+    
+    clr.RGBc[0] = (clr.RGBp[0] * M[0]) + (clr.RGBp[1] * M[1]) + (clr.RGBp[2] * M[2]);
+    clr.RGBc[1] = (clr.RGBp[0] * M[3]) + (clr.RGBp[1] * M[4]) + (clr.RGBp[2] * M[5]);
+    clr.RGBc[2] = (clr.RGBp[0] * M[6]) + (clr.RGBp[1] * M[7]) + (clr.RGBp[2] * M[8]);
+    return clr;
+}
+
+
+static
+CAM02COLOR InverseChromaticAdaptation(CAM02COLOR clr,  cmsCIECAM02* pMod) 
+{
+    cmsUInt32Number i;
+    for (i = 0; i < 3; i++) { 
+        clr.RGB[i] = clr.RGBc[i] /
+            ((pMod->adoptedWhite.XYZ[1] * pMod->D / pMod->adoptedWhite.RGB[i]) + 1.0 - pMod->D);
+    }
+    return clr;
+}
+
+
+static
+CAM02COLOR CAT02toXYZ(CAM02COLOR clr) 
+{
+    clr.XYZ[0] = (clr.RGB[0] *  1.096124) + (clr.RGB[1] * -0.278869) + (clr.RGB[2] *  0.182745);
+    clr.XYZ[1] = (clr.RGB[0] *  0.454369) + (clr.RGB[1] *  0.473533) + (clr.RGB[2] *  0.072098);
+    clr.XYZ[2] = (clr.RGB[0] * -0.009628) + (clr.RGB[1] * -0.005698) + (clr.RGB[2] *  1.015326);
+    
+    return clr;
+}
+
+
+cmsHANDLE  CMSEXPORT cmsCIECAM02Init(cmsContext ContextID, const cmsViewingConditions* pVC)
+{
+	cmsCIECAM02* lpMod;
+
+	_cmsAssert(pVC != NULL);
+
+	if((lpMod = (cmsCIECAM02*) _cmsMallocZero(ContextID, sizeof(cmsCIECAM02))) == NULL) {
+		return NULL;
+	}
+
+	lpMod ->ContextID = ContextID;
+
+	lpMod ->adoptedWhite.XYZ[0] = pVC ->whitePoint.X;
+	lpMod ->adoptedWhite.XYZ[1] = pVC ->whitePoint.Y;
+	lpMod ->adoptedWhite.XYZ[2] = pVC ->whitePoint.Z;
+
+	lpMod -> LA       = pVC ->La;
+	lpMod -> Yb       = pVC ->Yb;
+	lpMod -> D        = pVC ->D_value;
+	lpMod -> surround = pVC ->surround;
+
+	switch (lpMod -> surround) {
+
+
+	case CUTSHEET_SURROUND:
+		lpMod->F = 0.8;
+		lpMod->c = 0.41;
+		lpMod->Nc = 0.8;
+		break;
+
+	case DARK_SURROUND:
+		lpMod -> F  = 0.8;
+		lpMod -> c  = 0.525;
+		lpMod -> Nc = 0.8;
+		break;
+
+	case DIM_SURROUND:
+		lpMod -> F  = 0.9;
+		lpMod -> c  = 0.59;
+		lpMod -> Nc = 0.95;
+		break;
+
+	default:
+		// Average surround
+		lpMod -> F  = 1.0;
+		lpMod -> c  = 0.69;
+		lpMod -> Nc = 1.0;
+	}
+
+	lpMod -> n   = compute_n(lpMod);
+	lpMod -> z   = compute_z(lpMod);
+	lpMod -> Nbb = computeNbb(lpMod);
+	lpMod -> FL  = computeFL(lpMod);
+
+	if (lpMod -> D == D_CALCULATE) {
+		lpMod -> D   = computeD(lpMod);
+	}
+
+	lpMod -> Ncb = lpMod -> Nbb;
+
+	lpMod -> adoptedWhite = XYZtoCAT02(lpMod -> adoptedWhite);
+	lpMod -> adoptedWhite = ChromaticAdaptation(lpMod -> adoptedWhite, lpMod);
+	lpMod -> adoptedWhite = CAT02toHPE(lpMod -> adoptedWhite);
+	lpMod -> adoptedWhite = NonlinearCompression(lpMod -> adoptedWhite, lpMod);
+
+	return (cmsHANDLE) lpMod;
+
+}
+
+void CMSEXPORT cmsCIECAM02Done(cmsHANDLE hModel)
+{
+    cmsCIECAM02* lpMod = (cmsCIECAM02*) hModel;
+
+	if (lpMod) _cmsFree(lpMod ->ContextID, lpMod);
+}
+
+
+void CMSEXPORT cmsCIECAM02Forward(cmsHANDLE hModel, const cmsCIEXYZ* pIn, cmsJCh* pOut)
+{    
+    CAM02COLOR clr;
+    cmsCIECAM02* lpMod = (cmsCIECAM02*) hModel;
+    
+	_cmsAssert(lpMod != NULL);
+	_cmsAssert(pIn != NULL);
+	_cmsAssert(pOut != NULL);
+
+    clr.XYZ[0] = pIn ->X;
+    clr.XYZ[1] = pIn ->Y;
+    clr.XYZ[2] = pIn ->Z;
+    
+    clr = XYZtoCAT02(clr);
+    clr = ChromaticAdaptation(clr, lpMod);
+    clr = CAT02toHPE(clr);
+    clr = NonlinearCompression(clr, lpMod);
+    clr = ComputeCorrelates(clr, lpMod);
+    
+    pOut ->J = clr.J;
+    pOut ->C = clr.C;
+    pOut ->h = clr.h;         
+}
+
+void CMSEXPORT cmsCIECAM02Reverse(cmsHANDLE hModel, const cmsJCh* pIn, cmsCIEXYZ* pOut)
+{
+    CAM02COLOR clr;
+    cmsCIECAM02* lpMod = (cmsCIECAM02*) hModel;
+    
+	_cmsAssert(lpMod != NULL);
+	_cmsAssert(pIn != NULL);
+	_cmsAssert(pOut != NULL);
+
+    clr.J = pIn -> J;
+    clr.C = pIn -> C;
+    clr.h = pIn -> h;
+    
+    clr = InverseCorrelates(clr, lpMod);
+    clr = InverseNonlinearity(clr, lpMod);
+    clr = HPEtoCAT02(clr);
+    clr = InverseChromaticAdaptation(clr, lpMod);
+    clr = CAT02toXYZ(clr);
+    
+    pOut ->X = clr.XYZ[0];
+    pOut ->Y = clr.XYZ[1];
+    pOut ->Z = clr.XYZ[2];
+}
+
diff --git a/thirdparty/liblcms2/src/cmscgats.c b/thirdparty/liblcms2/src/cmscgats.c
new file mode 100644
index 00000000..d41c15a2
--- /dev/null
+++ b/thirdparty/liblcms2/src/cmscgats.c
@@ -0,0 +1,2655 @@
+//---------------------------------------------------------------------------------
+//
+//  Little Color Management System
+//  Copyright (c) 1998-2010 Marti Maria Saguer
+//
+// Permission is hereby granted, free of charge, to any person obtaining 
+// a copy of this software and associated documentation files (the "Software"), 
+// to deal in the Software without restriction, including without limitation 
+// the rights to use, copy, modify, merge, publish, distribute, sublicense, 
+// and/or sell copies of the Software, and to permit persons to whom the Software 
+// is furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in 
+// all copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 
+// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO 
+// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE 
+// LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 
+// OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 
+// WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+//
+//---------------------------------------------------------------------------------
+//
+
+#include "lcms2_internal.h"
+
+
+// IT8.7 / CGATS.17-200x handling -----------------------------------------------------------------------------
+
+
+#define MAXID        128     // Max lenght of identifier
+#define MAXSTR      1024     // Max lenght of string
+#define MAXTABLES    255     // Max Number of tables in a single stream
+#define MAXINCLUDE    20     // Max number of nested includes
+
+#define DEFAULT_DBL_FORMAT  "%.10g" // Double formatting
+
+#ifdef CMS_IS_WINDOWS_
+#    include <io.h>
+#    define DIR_CHAR    '\\'
+#else
+#    define DIR_CHAR    '/'
+#endif
+
+// Symbols
+typedef enum { 
+
+        SNONE,
+        SINUM,      // Integer
+        SDNUM,      // Real
+        SIDENT,     // Identifier
+        SSTRING,    // string
+        SCOMMENT,   // comment
+        SEOLN,      // End of line
+        SEOF,       // End of stream
+        SSYNERROR,  // Syntax error found on stream
+
+        // Keywords
+
+        SBEGIN_DATA,
+        SBEGIN_DATA_FORMAT,
+        SEND_DATA,
+        SEND_DATA_FORMAT,
+        SKEYWORD,
+        SDATA_FORMAT_ID,
+        SINCLUDE
+
+    } SYMBOL;
+
+
+// How to write the value
+typedef enum { 
+
+        WRITE_UNCOOKED,
+        WRITE_STRINGIFY,            
+        WRITE_HEXADECIMAL,
+        WRITE_BINARY,
+        WRITE_PAIR
+
+    } WRITEMODE;
+
+// Linked list of variable names
+typedef struct _KeyVal {
+
+        struct _KeyVal*  Next;
+        char*            Keyword;       // Name of variable
+        struct _KeyVal*  NextSubkey;    // If key is a dictionary, points to the next item
+        char*            Subkey;        // If key is a dictionary, points to the subkey name
+        char*            Value;         // Points to value
+        WRITEMODE        WriteAs;       // How to write the value
+
+   } KEYVALUE;
+
+
+// Linked list of memory chunks (Memory sink)
+typedef struct _OwnedMem {
+
+        struct _OwnedMem* Next;
+        void *            Ptr;          // Point to value
+
+   } OWNEDMEM;
+
+// Suballocator
+typedef struct _SubAllocator {
+
+         cmsUInt8Number* Block;
+         cmsUInt32Number BlockSize;
+         cmsUInt32Number Used;
+ 
+    } SUBALLOCATOR;
+
+// Table. Each individual table can hold properties and rows & cols
+typedef struct _Table {
+        
+        int            nSamples, nPatches;    // Cols, Rows
+        int            SampleID;              // Pos of ID
+        
+        KEYVALUE*      HeaderList;            // The properties
+        
+        char**         DataFormat;            // The binary stream descriptor
+        char**         Data;                  // The binary stream        
+
+    } TABLE;
+
+// File stream being parsed
+typedef struct _FileContext {
+        char           FileName[cmsMAX_PATH];    // File name if being readed from file
+        FILE*          Stream;                   // File stream or NULL if holded in memory
+    } FILECTX;
+
+// This struct hold all information about an open IT8 handler. 
+typedef struct {
+
+        char SheetType[MAXSTR];               // The first row of the IT8 (the type)
+
+        cmsUInt32Number  TablesCount;                     // How many tables in this stream
+        cmsUInt32Number  nTable;                          // The actual table
+
+        TABLE Tab[MAXTABLES];
+
+        // Memory management
+        OWNEDMEM*      MemorySink;            // The storage backend
+        SUBALLOCATOR   Allocator;             // String suballocator -- just to keep it fast
+
+        // Parser state machine
+        SYMBOL         sy;                    // Current symbol
+        int            ch;                    // Current character
+      
+        int            inum;                  // integer value
+        cmsFloat64Number         dnum;                  // real value
+        char           id[MAXID];             // identifier
+        char           str[MAXSTR];           // string
+
+        // Allowed keywords & datasets. They have visibility on whole stream
+        KEYVALUE*     ValidKeywords;
+        KEYVALUE*     ValidSampleID;
+              
+        char*          Source;                // Points to loc. being parsed
+        int            lineno;                // line counter for error reporting
+       
+        FILECTX*       FileStack[MAXINCLUDE]; // Stack of files being parsed
+        int            IncludeSP;             // Include Stack Pointer
+
+        char*          MemoryBlock;           // The stream if holded in memory
+
+        char           DoubleFormatter[MAXID];// Printf-like 'cmsFloat64Number' formatter
+
+        cmsContext    ContextID;              // The threading context
+
+   } cmsIT8;
+
+
+// The stream for save operations
+typedef struct {
+
+        FILE* stream;   // For save-to-file behaviour
+
+        cmsUInt8Number* Base;
+        cmsUInt8Number* Ptr;        // For save-to-mem behaviour
+        cmsUInt32Number Used;
+        cmsUInt32Number Max;
+
+    } SAVESTREAM;
+
+
+// ------------------------------------------------------ cmsIT8 parsing routines
+
+
+// A keyword
+typedef struct {
+
+        const char *id;
+        SYMBOL sy;
+
+   } KEYWORD;
+
+// The keyword->symbol translation table. Sorting is required.
+static const KEYWORD TabKeys[] = {
+ 
+        {"$INCLUDE",               SINCLUDE},   // This is an extension!
+        {".INCLUDE",               SINCLUDE},   // This is an extension!
+
+        {"BEGIN_DATA",             SBEGIN_DATA },
+        {"BEGIN_DATA_FORMAT",      SBEGIN_DATA_FORMAT },
+        {"DATA_FORMAT_IDENTIFIER", SDATA_FORMAT_ID},
+        {"END_DATA",               SEND_DATA},
+        {"END_DATA_FORMAT",        SEND_DATA_FORMAT},
+        {"KEYWORD",                SKEYWORD}
+        };
+
+#define NUMKEYS (sizeof(TabKeys)/sizeof(KEYWORD))
+
+// Predefined properties
+
+// A property
+typedef struct {
+        const char *id;    // The identifier
+        WRITEMODE as;      // How is supposed to be written
+    } PROPERTY;
+
+static PROPERTY PredefinedProperties[] = {
+
+        {"NUMBER_OF_FIELDS", WRITE_UNCOOKED},    // Required - NUMBER OF FIELDS
+        {"NUMBER_OF_SETS",   WRITE_UNCOOKED},    // Required - NUMBER OF SETS
+        {"ORIGINATOR",       WRITE_STRINGIFY},   // Required - Identifies the specific system, organization or individual that created the data file.
+        {"FILE_DESCRIPTOR",  WRITE_STRINGIFY},   // Required - Describes the purpose or contents of the data file.
+        {"CREATED",          WRITE_STRINGIFY},   // Required - Indicates date of creation of the data file.
+        {"DESCRIPTOR",       WRITE_STRINGIFY},   // Required  - Describes the purpose or contents of the data file.
+        {"DIFFUSE_GEOMETRY", WRITE_STRINGIFY},   // The diffuse geometry used. Allowed values are "sphere" or "opal".
+        {"MANUFACTURER",     WRITE_STRINGIFY},
+        {"MANUFACTURE",      WRITE_STRINGIFY},   // Some broken Fuji targets does store this value
+        {"PROD_DATE",        WRITE_STRINGIFY},   // Identifies year and month of production of the target in the form yyyy:mm.
+        {"SERIAL",           WRITE_STRINGIFY},   // Uniquely identifies individual physical target.
+
+        {"MATERIAL",         WRITE_STRINGIFY},   // Identifies the material on which the target was produced using a code
+                               // uniquely identifying th e material. This is intend ed to be used for IT8.7
+                               // physical targets only (i.e . IT8.7/1 a nd IT8.7/2).
+
+        {"INSTRUMENTATION",  WRITE_STRINGIFY},   // Used to report the specific instrumentation used (manufacturer and
+                               // model number) to generate the data reported. This data will often
+                               // provide more information about the particular data collected than an
+                               // extensive list of specific details. This is particularly important for
+                               // spectral data or data derived from spectrophotometry.
+
+        {"MEASUREMENT_SOURCE", WRITE_STRINGIFY}, // Illumination used for spectral measurements. This data helps provide
+                               // a guide to the potential for issues of paper fluorescence, etc.
+
+        {"PRINT_CONDITIONS", WRITE_STRINGIFY},   // Used to define the characteristics of the printed sheet being reported.
+                               // Where standard conditions have been defined (e.g., SWOP at nominal)
+                               // named conditions may suffice. Otherwise, detailed information is
+                               // needed.
+
+        {"SAMPLE_BACKING",   WRITE_STRINGIFY},   // Identifies the backing material used behind the sample during
+                               // measurement. Allowed values are “black”, “white”, or {"na".
+
+        {"CHISQ_DOF",        WRITE_STRINGIFY},   // Degrees of freedom associated with the Chi squared statistic
+
+       // below properties are new in recent specs:
+
+        {"MEASUREMENT_GEOMETRY", WRITE_STRINGIFY}, // The type of measurement, either reflection or transmission, should be indicated 
+                               // along with details of the geometry and the aperture size and shape. For example, 
+                               // for transmission measurements it is important to identify 0/diffuse, diffuse/0, 
+                               // opal or integrating sphere, etc. For reflection it is important to identify 0/45, 
+                               // 45/0, sphere (specular included or excluded), etc.
+
+       {"FILTER",            WRITE_STRINGIFY},   // Identifies the use of physical filter(s) during measurement. Typically used to 
+                               // denote the use of filters such as none, D65, Red, Green or Blue.
+
+       {"POLARIZATION",      WRITE_STRINGIFY},   // Identifies the use of a physical polarization filter during measurement. Allowed 
+                               // values are {"yes”, “white”, “none” or “na”.
+
+       {"WEIGHTING_FUNCTION", WRITE_PAIR},   // Indicates such functions as: the CIE standard observer functions used in the 
+                               // calculation of various data parameters (2 degree and 10 degree), CIE standard 
+                               // illuminant functions used in the calculation of various data parameters (e.g., D50,
+                               // D65, etc.), density status response, etc. If used there shall be at least one 
+                               // name-value pair following the WEIGHTING_FUNCTION tag/keyword. The first attribute 
+                               // in the set shall be {"name" and shall identify the particular parameter used.
+                               // The second shall be {"value" and shall provide the value associated with that name. 
+                               // For ASCII data, a string containing the Name and Value attribute pairs shall follow 
+                               // the weighting function keyword. A semi-colon separates attribute pairs from each 
+                               // other and within the attribute the name and value are separated by a comma.
+
+       {"COMPUTATIONAL_PARAMETER", WRITE_PAIR}, // Parameter that is used in computing a value from measured data. Name is the name 
+                               // of the calculation, parameter is the name of the parameter used in the calculation 
+                               // and value is the value of the parameter.
+
+       {"TARGET_TYPE",        WRITE_STRINGIFY},  // The type of target being measured, e.g. IT8.7/1, IT8.7/3, user defined, etc.
+
+       {"COLORANT",           WRITE_STRINGIFY},  // Identifies the colorant(s) used in creating the target.
+
+       {"TABLE_DESCRIPTOR",   WRITE_STRINGIFY},  // Describes the purpose or contents of a data table.
+
+       {"TABLE_NAME",         WRITE_STRINGIFY}   // Provides a short name for a data table.
+};
+
+#define NUMPREDEFINEDPROPS (sizeof(PredefinedProperties)/sizeof(PROPERTY))
+
+
+// Predefined sample types on dataset
+static const char* PredefinedSampleID[] = {
+        "SAMPLE_ID",      // Identifies sample that data represents
+        "STRING",         // Identifies label, or other non-machine readable value. 
+                          // Value must begin and end with a " symbol
+
+        "CMYK_C",         // Cyan component of CMYK data expressed as a percentage
+        "CMYK_M",         // Magenta component of CMYK data expressed as a percentage
+        "CMYK_Y",         // Yellow component of CMYK data expressed as a percentage
+        "CMYK_K",         // Black component of CMYK data expressed as a percentage
+        "D_RED",          // Red filter density
+        "D_GREEN",        // Green filter density
+        "D_BLUE",         // Blue filter density
+        "D_VIS",          // Visual filter density
+        "D_MAJOR_FILTER", // Major filter d ensity
+        "RGB_R",          // Red component of RGB data
+        "RGB_G",          // Green component of RGB data
+        "RGB_B",          // Blue com ponent of RGB data
+        "SPECTRAL_NM",    // Wavelength of measurement expressed in nanometers
+        "SPECTRAL_PCT",   // Percentage reflectance/transmittance
+        "SPECTRAL_DEC",   // Reflectance/transmittance
+        "XYZ_X",          // X component of tristimulus data
+        "XYZ_Y",          // Y component of tristimulus data
+        "XYZ_Z",          // Z component of tristimulus data
+        "XYY_X"           // x component of chromaticity data
+        "XYY_Y",          // y component of chromaticity data
+        "XYY_CAPY",       // Y component of tristimulus data
+        "LAB_L",          // L* component of Lab data
+        "LAB_A",          // a* component of Lab data
+        "LAB_B",          // b* component of Lab data
+        "LAB_C",          // C*ab component of Lab data
+        "LAB_H",          // hab component of Lab data
+        "LAB_DE",         // CIE dE
+        "LAB_DE_94",      // CIE dE using CIE 94
+        "LAB_DE_CMC",     // dE using CMC
+        "LAB_DE_2000",    // CIE dE using CIE DE 2000
+        "MEAN_DE",        // Mean Delta E (LAB_DE) of samples compared to batch average
+                          // (Used for data files for ANSI IT8.7/1 and IT8.7/2 targets)
+        "STDEV_X",        // Standard deviation of X (tristimulus data)
+        "STDEV_Y",        // Standard deviation of Y (tristimulus data)
+        "STDEV_Z",        // Standard deviation of Z (tristimulus data)
+        "STDEV_L",        // Standard deviation of L*
+        "STDEV_A",        // Standard deviation of a*
+        "STDEV_B",        // Standard deviation of b*
+        "STDEV_DE",       // Standard deviation of CIE dE
+        "CHI_SQD_PAR"};   // The average of the standard deviations of L*, a* and b*. It is
+                          // used to derive an estimate of the chi-squared parameter which is
+                          // recommended as the predictor of the variability of dE
+
+#define NUMPREDEFINEDSAMPLEID (sizeof(PredefinedSampleID)/sizeof(char *))
+
+//Forward declaration of some internal functions        
+static void* AllocChunk(cmsIT8* it8, cmsUInt32Number size);
+
+// Checks if c is a separator
+static
+cmsBool isseparator(int c)
+{
+        return (c == ' ') || (c == '\t') || (c == '\r');
+}
+
+// Checks whatever if c is a valid identifier char
+static 
+cmsBool ismiddle(int c)
+{
+   return (!isseparator(c) && (c != '#') && (c !='\"') && (c != '\'') && (c > 32) && (c < 127));
+}
+
+// Checks whatsever if c is a valid identifier middle char.
+static
+cmsBool isidchar(int c)
+{
+   return isalnum(c) || ismiddle(c);
+}
+
+// Checks whatsever if c is a valid identifier first char.
+static
+cmsBool isfirstidchar(int c)
+{
+     return !isdigit(c) && ismiddle(c);
+}
+
+// Guess whether the supplied path looks like an absolute path
+static
+cmsBool isabsolutepath(const char *path)
+{
+    char ThreeChars[4];
+
+    if(path == NULL)
+        return FALSE;
+    if (path[0] == 0) 
+        return FALSE;
+
+    strncpy(ThreeChars, path, 3);
+    ThreeChars[3] = 0;
+
+    if(ThreeChars[0] == DIR_CHAR)
+        return TRUE;
+
+#ifdef  CMS_IS_WINDOWS_
+    if (isalpha((int) ThreeChars[0]) && ThreeChars[1] == ':')
+        return TRUE;
+#endif
+    return FALSE;
+}
+
+// Makes a file path based on a given reference path
+// NOTE: this function doesn't check if the path exists or even if it's legal
+static 
+cmsBool BuildAbsolutePath(const char *relPath, const char *basePath, char *buffer, cmsUInt32Number MaxLen)
+{
+    char *tail;
+    cmsUInt32Number len;
+
+    // Already absolute?
+    if (isabsolutepath(relPath)) {
+
+        strncpy(buffer, relPath, MaxLen);
+        buffer[MaxLen-1] = 0;
+        return TRUE;
+    }
+
+    // No, search for last 
+    strncpy(buffer, basePath, MaxLen);  
+    buffer[MaxLen-1] = 0;
+
+    tail = strrchr(buffer, DIR_CHAR);
+    if (tail == NULL) return FALSE;    // Is not absolute and has no separators??
+
+    len = (cmsUInt32Number) (tail - buffer);
+    if (len >= MaxLen) return FALSE;
+
+    // No need to assure zero terminator over here
+    strncpy(tail + 1, relPath, MaxLen - len);   
+
+    return TRUE;    
+}
+
+
+// Make sure no exploit is being even tried
+static
+const char* NoMeta(const char* str)
+{
+    if (strchr(str, '%') != NULL) 
+        return "**** CORRUPTED FORMAT STRING ***";
+
+    return str;
+}
+
+// Syntax error
+static
+cmsBool SynError(cmsIT8* it8, const char *Txt, ...)
+{
+    char Buffer[256], ErrMsg[1024];
+    va_list args;
+
+    va_start(args, Txt);
+    vsnprintf(Buffer, 255, Txt, args);
+    Buffer[255] = 0;
+    va_end(args);
+
+    snprintf(ErrMsg, 1023, "%s: Line %d, %s", it8->FileStack[it8 ->IncludeSP]->FileName, it8->lineno, Buffer);
+    ErrMsg[1023] = 0;
+    it8->sy = SSYNERROR;
+    cmsSignalError(it8 ->ContextID, cmsERROR_CORRUPTION_DETECTED, "%s", ErrMsg);
+    return FALSE;
+}
+
+// Check if current symbol is same as specified. issue an error else.
+static
+cmsBool Check(cmsIT8* it8, SYMBOL sy, const char* Err)
+{
+        if (it8 -> sy != sy)
+                return SynError(it8, NoMeta(Err));
+        return TRUE;
+}
+
+// Read Next character from stream
+static
+void NextCh(cmsIT8* it8)
+{
+    if (it8 -> FileStack[it8 ->IncludeSP]->Stream) {
+
+        it8 ->ch = fgetc(it8 ->FileStack[it8 ->IncludeSP]->Stream);
+
+        if (feof(it8 -> FileStack[it8 ->IncludeSP]->Stream))  {
+
+            if (it8 ->IncludeSP > 0) {
+
+                fclose(it8 ->FileStack[it8->IncludeSP--]->Stream);
+                it8 -> ch = ' ';                            // Whitespace to be ignored
+
+            } else
+                it8 ->ch = 0;   // EOF
+        }              
+    }
+    else {
+        it8->ch = *it8->Source;
+        if (it8->ch) it8->Source++;
+    }
+}
+
+
+// Try to see if current identifier is a keyword, if so return the referred symbol
+static
+SYMBOL BinSrchKey(const char *id)
+{
+    int l = 1;
+    int r = NUMKEYS;
+    int x, res;
+
+    while (r >= l)
+    {
+        x = (l+r)/2;
+        res = cmsstrcasecmp(id, TabKeys[x-1].id);
+        if (res == 0) return TabKeys[x-1].sy;
+        if (res < 0) r = x - 1;
+        else l = x + 1;
+    }
+
+    return SNONE;
+}
+
+
+// 10 ^n
+static
+cmsFloat64Number xpow10(int n)
+{
+    return pow(10, (cmsFloat64Number) n);
+}
+
+
+//  Reads a Real number, tries to follow from integer number
+static
+void ReadReal(cmsIT8* it8, int inum)
+{
+    it8->dnum = (cmsFloat64Number) inum;
+
+    while (isdigit(it8->ch)) {
+
+        it8->dnum = it8->dnum * 10.0 + (it8->ch - '0');
+        NextCh(it8);
+    }
+
+    if (it8->ch == '.') {        // Decimal point
+
+        cmsFloat64Number frac = 0.0;      // fraction
+        int prec = 0;           // precision
+
+        NextCh(it8);               // Eats dec. point
+
+        while (isdigit(it8->ch)) {
+
+            frac = frac * 10.0 + (it8->ch - '0');
+            prec++;
+            NextCh(it8);
+        }
+
+        it8->dnum = it8->dnum + (frac / xpow10(prec));
+    }
+
+    // Exponent, example 34.00E+20
+    if (toupper(it8->ch) == 'E') {
+
+        int e;
+        int sgn;
+
+        NextCh(it8); sgn = 1;
+
+        if (it8->ch == '-') {
+
+            sgn = -1; NextCh(it8);
+        }
+        else
+            if (it8->ch == '+') {
+
+                sgn = +1;
+                NextCh(it8);
+            }
+
+            e = 0;
+            while (isdigit(it8->ch)) {
+
+                if ((cmsFloat64Number) e * 10L < INT_MAX)
+                    e = e * 10 + (it8->ch - '0');
+
+                NextCh(it8);
+            }
+
+            e = sgn*e;
+            it8 -> dnum = it8 -> dnum * xpow10(e);
+    }
+}
+
+
+
+// Reads next symbol
+static
+void InSymbol(cmsIT8* it8)
+{
+    register char *idptr;
+    register int k;
+    SYMBOL key;
+    int sng;
+    
+    do {
+        
+        while (isseparator(it8->ch))
+            NextCh(it8);
+        
+        if (isfirstidchar(it8->ch)) {          // Identifier
+                        
+            k = 0;
+            idptr = it8->id;
+            
+            do {
+                
+                if (++k < MAXID) *idptr++ = (char) it8->ch;
+                
+                NextCh(it8);
+                
+            } while (isidchar(it8->ch));
+            
+            *idptr = '\0';
+            
+            
+            key = BinSrchKey(it8->id);
+            if (key == SNONE) it8->sy = SIDENT;
+            else it8->sy = key;
+            
+        }
+        else                         // Is a number?
+            if (isdigit(it8->ch) || it8->ch == '.' || it8->ch == '-' || it8->ch == '+')
+            {
+                int sign = 1;
+                
+                if (it8->ch == '-') {
+                    sign = -1;
+                    NextCh(it8);
+                }
+                
+                it8->inum = 0;
+                it8->sy   = SINUM;
+                
+                if (it8->ch == '0') {          // 0xnnnn (Hexa) or 0bnnnn (Binary)
+                
+                    NextCh(it8);
+                    if (toupper(it8->ch) == 'X') {
+
+                        int j;
+                        
+                        NextCh(it8);
+                        while (isxdigit(it8->ch))
+                        {
+                            it8->ch = toupper(it8->ch);
+                            if (it8->ch >= 'A' && it8->ch <= 'F')  j = it8->ch -'A'+10;
+                            else j = it8->ch - '0';
+                            
+                            if ((long) it8->inum * 16L > (long) INT_MAX)
+                            {
+                                SynError(it8, "Invalid hexadecimal number");
+                                return;
+                            }
+                            
+                            it8->inum = it8->inum * 16 + j;
+                            NextCh(it8);
+                        }
+                        return;
+                    }
+                    
+                    if (toupper(it8->ch) == 'B') {  // Binary
+                    
+                        int j;
+                        
+                        NextCh(it8);
+                        while (it8->ch == '0' || it8->ch == '1')
+                        {
+                            j = it8->ch - '0';
+                            
+                            if ((long) it8->inum * 2L > (long) INT_MAX)
+                            {
+                                SynError(it8, "Invalid binary number");
+                                return;
+                            }
+                            
+                            it8->inum = it8->inum * 2 + j;
+                            NextCh(it8);
+                        }
+                        return;
+                    }
+                }
+                
+
+                while (isdigit(it8->ch)) {
+
+                    if ((long) it8->inum * 10L > (long) INT_MAX) {
+                        ReadReal(it8, it8->inum);
+                        it8->sy = SDNUM;
+                        it8->dnum *= sign;
+                        return;
+                    }
+                    
+                    it8->inum = it8->inum * 10 + (it8->ch - '0');
+                    NextCh(it8);
+                }
+                
+                if (it8->ch == '.') {
+                    
+                    ReadReal(it8, it8->inum);
+                    it8->sy = SDNUM;
+                    it8->dnum *= sign;
+                    return;
+                }
+                
+                it8 -> inum *= sign;
+
+                // Special case. Numbers followed by letters are taken as identifiers
+
+                if (isidchar(it8 ->ch)) {
+
+                    if (it8 ->sy == SINUM) {
+
+                        sprintf(it8->id, "%d", it8->inum);
+                    }
+                    else {
+
+                        sprintf(it8->id, it8 ->DoubleFormatter, it8->dnum);
+                    }
+
+                    k = (int) strlen(it8 ->id);
+                    idptr = it8 ->id + k;
+                    do {
+                
+                        if (++k < MAXID) *idptr++ = (char) it8->ch;
+                
+                        NextCh(it8);
+                
+                    } while (isidchar(it8->ch));
+            
+                    *idptr = '\0';                    
+                    it8->sy = SIDENT;
+                }
+                return;
+                
+            }
+            else
+                switch ((int) it8->ch) {
+
+        // EOF marker -- ignore it
+        case '\x1a':
+            NextCh(it8);
+            break;
+
+        // Eof stream markers
+        case 0:
+        case -1:
+            it8->sy = SEOF;
+            break;
+            
+            
+        // Next line            
+        case '\n':
+            NextCh(it8);
+            it8->sy = SEOLN;
+            it8->lineno++;
+            break;
+            
+        // Comment
+        case '#':
+            NextCh(it8);
+            while (it8->ch && it8->ch != '\n')
+                NextCh(it8);
+            
+            it8->sy = SCOMMENT;
+            break;
+            
+        // String.
+        case '\'':
+        case '\"':
+            idptr = it8->str;
+            sng = it8->ch;
+            k = 0;
+            NextCh(it8);
+            
+            while (k < MAXSTR && it8->ch != sng) {
+                
+                if (it8->ch == '\n'|| it8->ch == '\r') k = MAXSTR+1;
+                else {                                    
+                    *idptr++ = (char) it8->ch;
+                    NextCh(it8);
+                    k++;
+                }
+            }
+            
+            it8->sy = SSTRING;
+            *idptr = '\0';
+            NextCh(it8);
+            break;
+            
+            
+        default:
+            SynError(it8, "Unrecognized character: 0x%x", it8 ->ch);            
+            return;            
+            }
+            
+    } while (it8->sy == SCOMMENT);
+
+    // Handle the include special token
+
+    if (it8 -> sy == SINCLUDE) {
+
+                FILECTX* FileNest;
+
+                if(it8 -> IncludeSP >= (MAXINCLUDE-1)) {
+
+                    SynError(it8, "Too many recursion levels");
+                    return;
+                }
+
+                InSymbol(it8);
+                if (!Check(it8, SSTRING, "Filename expected")) return;
+
+                FileNest = it8 -> FileStack[it8 -> IncludeSP + 1];
+                if(FileNest == NULL) {
+
+                    FileNest = it8 ->FileStack[it8 -> IncludeSP + 1] = (FILECTX*)AllocChunk(it8, sizeof(FILECTX));
+                    //if(FileNest == NULL)
+                    //  TODO: how to manage out-of-memory conditions?
+                }
+
+                if (BuildAbsolutePath(it8->str, 
+                                      it8->FileStack[it8->IncludeSP]->FileName, 
+                                      FileNest->FileName, cmsMAX_PATH-1) == FALSE) {
+                    SynError(it8, "File path too long");
+                    return;
+                }
+
+                FileNest->Stream = fopen(FileNest->FileName, "rt");
+                if (FileNest->Stream == NULL) {
+
+                        SynError(it8, "File %s not found", FileNest->FileName);
+                        return;
+                }
+                it8->IncludeSP++;
+
+                it8 ->ch = ' ';
+                InSymbol(it8);    
+    }
+                
+}
+
+// Checks end of line separator
+static
+cmsBool CheckEOLN(cmsIT8* it8)
+{
+        if (!Check(it8, SEOLN, "Expected separator")) return FALSE;
+        while (it8 -> sy == SEOLN)
+                        InSymbol(it8);
+        return TRUE;
+
+}
+
+// Skip a symbol
+
+static
+void Skip(cmsIT8* it8, SYMBOL sy)
+{
+        if (it8->sy == sy && it8->sy != SEOF)
+                        InSymbol(it8);
+}
+
+
+// Skip multiple EOLN
+static
+void SkipEOLN(cmsIT8* it8)
+{
+    while (it8->sy == SEOLN) {
+             InSymbol(it8);
+    }
+}
+
+
+// Returns a string holding current value
+static
+cmsBool GetVal(cmsIT8* it8, char* Buffer, cmsUInt32Number max, const char* ErrorTitle)
+{
+    switch (it8->sy) {
+
+    case SIDENT:  strncpy(Buffer, it8->id, max); 
+                  Buffer[max-1]=0;
+                  break;
+    case SINUM:   snprintf(Buffer, max, "%d", it8 -> inum); break;
+    case SDNUM:   snprintf(Buffer, max, it8->DoubleFormatter, it8 -> dnum); break;
+    case SSTRING: strncpy(Buffer, it8->str, max); 
+                  Buffer[max-1] = 0;
+                  break;
+
+
+    default:
+         return SynError(it8, "%s", ErrorTitle);
+    }
+
+    Buffer[max] = 0;
+    return TRUE;
+}
+
+// ---------------------------------------------------------- Table
+
+static
+TABLE* GetTable(cmsIT8* it8)
+{        
+   if ((it8 -> nTable >= it8 ->TablesCount)) {
+
+           SynError(it8, "Table %d out of sequence", it8 -> nTable);
+           return it8 -> Tab;
+   }            
+
+   return it8 ->Tab + it8 ->nTable;
+}
+
+// ---------------------------------------------------------- Memory management
+
+
+// Frees an allocator and owned memory
+void CMSEXPORT cmsIT8Free(cmsHANDLE hIT8)
+{
+   cmsIT8* it8 = (cmsIT8*) hIT8;
+
+    if (it8 == NULL)
+        return;
+
+    if (it8->MemorySink) {
+
+        OWNEDMEM* p;
+        OWNEDMEM* n;
+
+        for (p = it8->MemorySink; p != NULL; p = n) {
+
+            n = p->Next;
+            if (p->Ptr) _cmsFree(it8 ->ContextID, p->Ptr);
+            _cmsFree(it8 ->ContextID, p);
+        }
+    }
+
+    if (it8->MemoryBlock)
+        _cmsFree(it8 ->ContextID, it8->MemoryBlock);    
+
+    _cmsFree(it8 ->ContextID, it8);
+}
+
+
+// Allocates a chunk of data, keep linked list
+static
+void* AllocBigBlock(cmsIT8* it8, cmsUInt32Number size)
+{
+    OWNEDMEM* ptr1;
+    void* ptr = _cmsMallocZero(it8->ContextID, size);
+
+    if (ptr != NULL) {
+
+        ptr1 = (OWNEDMEM*) _cmsMallocZero(it8 ->ContextID, sizeof(OWNEDMEM));
+
+        if (ptr1 == NULL) {
+
+            _cmsFree(it8 ->ContextID, ptr);
+            return NULL;
+        }
+
+        ptr1-> Ptr        = ptr;
+        ptr1-> Next       = it8 -> MemorySink;
+        it8 -> MemorySink = ptr1;
+    }
+
+    return ptr;
+}
+
+
+// Suballocator.
+static
+void* AllocChunk(cmsIT8* it8, cmsUInt32Number size)
+{
+    cmsUInt32Number Free = it8 ->Allocator.BlockSize - it8 ->Allocator.Used;
+    cmsUInt8Number* ptr;
+
+    size = _cmsALIGNLONG(size);
+
+    if (size > Free) {
+
+        if (it8 -> Allocator.BlockSize == 0)
+
+                it8 -> Allocator.BlockSize = 20*1024;
+        else
+                it8 ->Allocator.BlockSize *= 2;
+
+        if (it8 ->Allocator.BlockSize < size)
+                it8 ->Allocator.BlockSize = size;
+
+        it8 ->Allocator.Used = 0;
+        it8 ->Allocator.Block = (cmsUInt8Number*)  AllocBigBlock(it8, it8 ->Allocator.BlockSize);
+    }
+            
+    ptr = it8 ->Allocator.Block + it8 ->Allocator.Used;
+    it8 ->Allocator.Used += size;
+
+    return (void*) ptr;
+    
+}
+
+
+// Allocates a string
+static
+char *AllocString(cmsIT8* it8, const char* str)
+{
+    cmsUInt32Number Size = (cmsUInt32Number) strlen(str)+1;
+    char *ptr;
+
+
+    ptr = (char *) AllocChunk(it8, Size);
+    if (ptr) strncpy (ptr, str, Size-1);
+
+    return ptr;
+}  
+
+// Searches through linked list
+
+static
+cmsBool IsAvailableOnList(KEYVALUE* p, const char* Key, const char* Subkey, KEYVALUE** LastPtr)
+{
+    if (LastPtr) *LastPtr = p;
+
+    for (;  p != NULL; p = p->Next) {
+
+        if (LastPtr) *LastPtr = p;
+
+        if (*Key != '#') { // Comments are ignored
+
+            if (cmsstrcasecmp(Key, p->Keyword) == 0)
+                    break;
+        }
+        }
+
+    if (p == NULL)
+        return FALSE;
+
+    if (Subkey == 0)
+        return TRUE;
+
+    for (; p != NULL; p = p->NextSubkey) {
+
+        if (LastPtr) *LastPtr = p;
+
+        if (cmsstrcasecmp(Subkey, p->Subkey) == 0)
+                    return TRUE;
+        }
+
+    return FALSE;
+}
+
+
+
+// Add a property into a linked list
+static
+KEYVALUE* AddToList(cmsIT8* it8, KEYVALUE** Head, const char *Key, const char *Subkey, const char* xValue, WRITEMODE WriteAs)
+{
+    KEYVALUE* p;
+    KEYVALUE* last;
+
+
+    // Check if property is already in list 
+
+    if (IsAvailableOnList(*Head, Key, Subkey, &p)) {
+
+        // This may work for editing properties
+
+        //     return SynError(it8, "duplicate key <%s>", Key);                                        
+    }
+    else {
+
+        last = p;
+
+        // Allocate the container
+        p = (KEYVALUE*) AllocChunk(it8, sizeof(KEYVALUE));
+        if (p == NULL)
+        {
+            SynError(it8, "AddToList: out of memory");        
+            return NULL;
+        }
+
+        // Store name and value
+        p->Keyword = AllocString(it8, Key);
+        p->Subkey = (Subkey == NULL) ? NULL : AllocString(it8, Subkey);
+
+        // Keep the container in our list
+        if (*Head == NULL) {
+            *Head = p;
+        }
+        else
+        {
+            if (Subkey != NULL && last != NULL) {
+            
+                last->NextSubkey = p;
+
+                // If Subkey is not null, then last is the last property with the same key,
+                // but not necessarily is the last property in the list, so we need to move
+                // to the actual list end
+                while (last->Next != NULL) 
+                         last = last->Next;
+            }
+
+            if (last != NULL) last->Next = p;
+        }
+
+        p->Next    = NULL;
+        p->NextSubkey = NULL;
+    }
+
+    p->WriteAs = WriteAs;
+
+    if (xValue != NULL) {
+
+        p->Value   = AllocString(it8, xValue);
+    }
+    else {
+        p->Value   = NULL;
+    }
+
+    return p;
+}
+
+static
+KEYVALUE* AddAvailableProperty(cmsIT8* it8, const char* Key, WRITEMODE as)
+{
+    return AddToList(it8, &it8->ValidKeywords, Key, NULL, NULL, as);
+}
+
+
+static
+KEYVALUE* AddAvailableSampleID(cmsIT8* it8, const char* Key)
+{
+    return AddToList(it8, &it8->ValidSampleID, Key, NULL, NULL, WRITE_UNCOOKED);
+}
+
+
+static
+void AllocTable(cmsIT8* it8)
+{
+    TABLE* t;
+
+    t = it8 ->Tab + it8 ->TablesCount;
+
+    t->HeaderList = NULL;
+    t->DataFormat = NULL;
+    t->Data       = NULL;
+   
+    it8 ->TablesCount++;
+}
+
+
+cmsInt32Number CMSEXPORT cmsIT8SetTable(cmsHANDLE  IT8, cmsUInt32Number nTable)
+{
+     cmsIT8* it8 = (cmsIT8*) IT8;
+
+     if (nTable >= it8 ->TablesCount) {
+
+         if (nTable == it8 ->TablesCount) {
+
+             AllocTable(it8);
+         }
+         else {
+             SynError(it8, "Table %d is out of sequence", nTable);
+             return -1;
+         }
+     }
+
+     it8 ->nTable = nTable;
+
+     return nTable;
+}
+
+
+
+// Init an empty container
+cmsHANDLE  CMSEXPORT cmsIT8Alloc(cmsContext ContextID)
+{
+    cmsIT8* it8;
+    int i;
+
+    it8 = (cmsIT8*) _cmsMallocZero(ContextID, sizeof(cmsIT8));
+    if (it8 == NULL) return NULL;
+
+    AllocTable(it8);
+    
+    it8->MemoryBlock = NULL;
+    it8->MemorySink  = NULL;
+    
+    it8 ->nTable = 0;
+
+    it8->ContextID = ContextID;
+    it8->Allocator.Used = 0;
+    it8->Allocator.Block = NULL;
+    it8->Allocator.BlockSize = 0;  
+
+    it8->ValidKeywords = NULL;
+    it8->ValidSampleID = NULL;
+
+    it8 -> sy = SNONE;
+    it8 -> ch = ' ';
+    it8 -> Source = NULL;
+    it8 -> inum = 0;
+    it8 -> dnum = 0.0;
+
+    it8->FileStack[0] = (FILECTX*)AllocChunk(it8, sizeof(FILECTX));
+    it8->IncludeSP   = 0;
+    it8 -> lineno = 1;
+
+    strcpy(it8->DoubleFormatter, DEFAULT_DBL_FORMAT);
+    strcpy(it8->SheetType, "CGATS.17");
+
+    // Initialize predefined properties & data
+    
+    for (i=0; i < NUMPREDEFINEDPROPS; i++)
+            AddAvailableProperty(it8, PredefinedProperties[i].id, PredefinedProperties[i].as);
+
+    for (i=0; i < NUMPREDEFINEDSAMPLEID; i++)
+            AddAvailableSampleID(it8, PredefinedSampleID[i]);
+
+
+   return (cmsHANDLE) it8;
+}
+
+
+const char* CMSEXPORT cmsIT8GetSheetType(cmsHANDLE hIT8)
+{
+        cmsIT8* it8 = (cmsIT8*) hIT8;
+
+        return it8 ->SheetType;
+
+}
+
+cmsBool CMSEXPORT cmsIT8SetSheetType(cmsHANDLE hIT8, const char* Type)
+{
+        cmsIT8* it8 = (cmsIT8*) hIT8;
+
+        strncpy(it8 ->SheetType, Type, MAXSTR-1);
+        it8 ->SheetType[MAXSTR-1] = 0;
+        return TRUE;
+}
+
+cmsBool CMSEXPORT cmsIT8SetComment(cmsHANDLE hIT8, const char* Val)
+{
+    cmsIT8* it8 = (cmsIT8*) hIT8;
+
+    if (!Val) return FALSE;
+    if (!*Val) return FALSE;
+
+    return AddToList(it8, &GetTable(it8)->HeaderList, "# ", NULL, Val, WRITE_UNCOOKED) != NULL;
+}
+
+
+
+// Sets a property
+cmsBool CMSEXPORT cmsIT8SetPropertyStr(cmsHANDLE hIT8, const char* Key, const char *Val)
+{
+    cmsIT8* it8 = (cmsIT8*) hIT8;
+
+    if (!Val) return FALSE;
+    if (!*Val) return FALSE;
+
+    return AddToList(it8, &GetTable(it8)->HeaderList, Key, NULL, Val, WRITE_STRINGIFY) != NULL;
+}
+
+
+cmsBool CMSEXPORT cmsIT8SetPropertyDbl(cmsHANDLE hIT8, const char* cProp, cmsFloat64Number Val)
+{
+    cmsIT8* it8 = (cmsIT8*) hIT8;
+    char Buffer[1024];
+   
+    sprintf(Buffer, it8->DoubleFormatter, Val);
+
+    return AddToList(it8, &GetTable(it8)->HeaderList, cProp, NULL, Buffer, WRITE_UNCOOKED) != NULL;    
+}
+
+cmsBool CMSEXPORT cmsIT8SetPropertyHex(cmsHANDLE hIT8, const char* cProp, cmsUInt32Number Val)
+{
+    cmsIT8* it8 = (cmsIT8*) hIT8;
+    char Buffer[1024];
+   
+    sprintf(Buffer, "%d", Val);
+
+    return AddToList(it8, &GetTable(it8)->HeaderList, cProp, NULL, Buffer, WRITE_HEXADECIMAL) != NULL;    
+}
+
+cmsBool CMSEXPORT cmsIT8SetPropertyUncooked(cmsHANDLE hIT8, const char* Key, const char* Buffer)
+{
+    cmsIT8* it8 = (cmsIT8*) hIT8;    
+    
+    return AddToList(it8, &GetTable(it8)->HeaderList, Key, NULL, Buffer, WRITE_UNCOOKED) != NULL;
+}
+
+cmsBool CMSEXPORT cmsIT8SetPropertyMulti(cmsHANDLE hIT8, const char* Key, const char* SubKey, const char *Buffer)
+{
+    cmsIT8* it8 = (cmsIT8*) hIT8;
+
+    return AddToList(it8, &GetTable(it8)->HeaderList, Key, SubKey, Buffer, WRITE_PAIR) != NULL;
+}
+
+// Gets a property
+const char* CMSEXPORT cmsIT8GetProperty(cmsHANDLE hIT8, const char* Key)
+{
+    cmsIT8* it8 = (cmsIT8*) hIT8;
+    KEYVALUE* p;
+
+    if (IsAvailableOnList(GetTable(it8) -> HeaderList, Key, NULL, &p))
+    {
+        return p -> Value;
+    }
+    return NULL;
+}
+
+
+cmsFloat64Number CMSEXPORT cmsIT8GetPropertyDbl(cmsHANDLE hIT8, const char* cProp)
+{
+    const char *v = cmsIT8GetProperty(hIT8, cProp);
+
+    if (v) return atof(v);
+    else return 0.0;
+}
+
+const char* CMSEXPORT cmsIT8GetPropertyMulti(cmsHANDLE hIT8, const char* Key, const char *SubKey)
+{
+    cmsIT8* it8 = (cmsIT8*) hIT8;
+    KEYVALUE* p;
+
+    if (IsAvailableOnList(GetTable(it8) -> HeaderList, Key, SubKey, &p)) {
+        return p -> Value;
+    }
+    return NULL;
+}
+
+// ----------------------------------------------------------------- Datasets
+
+
+static
+void AllocateDataFormat(cmsIT8* it8)
+{
+    TABLE* t = GetTable(it8);
+
+    if (t -> DataFormat) return;    // Already allocated
+
+    t -> nSamples  = (int) cmsIT8GetPropertyDbl(it8, "NUMBER_OF_FIELDS");
+
+    if (t -> nSamples <= 0) {
+
+        SynError(it8, "AllocateDataFormat: Unknown NUMBER_OF_FIELDS");
+        t -> nSamples = 10;
+        }
+
+    t -> DataFormat = (char**) AllocChunk (it8, (t->nSamples + 1) * sizeof(char *));
+    if (t->DataFormat == NULL) {
+
+        SynError(it8, "AllocateDataFormat: Unable to allocate dataFormat array");
+    }
+
+}
+
+static
+const char *GetDataFormat(cmsIT8* it8, int n)
+{
+    TABLE* t = GetTable(it8);
+
+    if (t->DataFormat)
+        return t->DataFormat[n];
+
+    return NULL;
+}
+
+static
+cmsBool SetDataFormat(cmsIT8* it8, int n, const char *label)
+{
+    TABLE* t = GetTable(it8);
+
+    if (!t->DataFormat)
+        AllocateDataFormat(it8);
+
+    if (n > t -> nSamples) {
+        SynError(it8, "More than NUMBER_OF_FIELDS fields.");
+        return FALSE;
+    }
+    
+    if (t->DataFormat) {                
+        t->DataFormat[n] = AllocString(it8, label);
+    }
+
+    return TRUE;
+}
+
+
+cmsBool CMSEXPORT cmsIT8SetDataFormat(cmsHANDLE  h, int n, const char *Sample)
+{
+        cmsIT8* it8 = (cmsIT8*) h;
+        return SetDataFormat(it8, n, Sample);
+}
+
+static
+void AllocateDataSet(cmsIT8* it8)
+{
+    TABLE* t = GetTable(it8);
+
+    if (t -> Data) return;    // Already allocated
+
+    t-> nSamples   = atoi(cmsIT8GetProperty(it8, "NUMBER_OF_FIELDS"));
+    t-> nPatches   = atoi(cmsIT8GetProperty(it8, "NUMBER_OF_SETS"));
+
+    t-> Data = (char**)AllocChunk (it8, (t->nSamples + 1) * (t->nPatches + 1) *sizeof (char*));
+    if (t->Data == NULL) {
+
+        SynError(it8, "AllocateDataSet: Unable to allocate data array");
+    }
+
+}
+
+static
+char* GetData(cmsIT8* it8, int nSet, int nField)
+{
+    TABLE* t = GetTable(it8);
+    int  nSamples   = t -> nSamples;
+    int  nPatches   = t -> nPatches;
+    
+    if (nSet >= nPatches || nField >= nSamples)
+        return NULL;
+
+    if (!t->Data) return NULL;
+    return t->Data [nSet * nSamples + nField];
+}
+
+static
+cmsBool SetData(cmsIT8* it8, int nSet, int nField, const char *Val)
+{
+    TABLE* t = GetTable(it8);
+
+    if (!t->Data)
+        AllocateDataSet(it8);
+
+    if (!t->Data) return FALSE;
+
+    if (nSet > t -> nPatches || nSet < 0) {
+
+            return SynError(it8, "Patch %d out of range, there are %d patches", nSet, t -> nPatches);            
+    }
+
+    if (nField > t ->nSamples || nField < 0) {
+            return SynError(it8, "Sample %d out of range, there are %d samples", nField, t ->nSamples);
+            
+    }
+    
+    t->Data [nSet * t -> nSamples + nField] = AllocString(it8, Val);
+    return TRUE;
+}
+
+
+// --------------------------------------------------------------- File I/O
+
+
+// Writes a string to file
+static
+void WriteStr(SAVESTREAM* f, const char *str)
+{
+    cmsUInt32Number len;
+
+    if (str == NULL) 
+        str = " ";
+    
+    // Lenghth to write
+    len = (cmsUInt32Number) strlen(str);
+    f ->Used += len;
+    
+
+    if (f ->stream) {   // Should I write it to a file?
+
+        if (fwrite(str, 1, len, f->stream) != len) {
+            cmsSignalError(0, cmsERROR_WRITE, "Write to file error in CGATS parser");
+            return;
+        }          
+        
+    }
+    else {  // Or to a memory block?
+                        
+        if (f ->Base) {   // Am I just counting the bytes?
+            
+            if (f ->Used > f ->Max) {
+
+                 cmsSignalError(0, cmsERROR_WRITE, "Write to memory overflows in CGATS parser");
+                 return;
+            }
+            
+            memmove(f ->Ptr, str, len);
+            f->Ptr += len;
+        }
+                        
+    }   
+}
+
+
+// Write formatted
+
+static
+void Writef(SAVESTREAM* f, const char* frm, ...)
+{
+    char Buffer[4096];
+    va_list args;
+
+    va_start(args, frm);
+    vsnprintf(Buffer, 4095, frm, args);
+    Buffer[4095] = 0;
+    WriteStr(f, Buffer);
+    va_end(args);
+
+}
+
+// Writes full header
+static
+void WriteHeader(cmsIT8* it8, SAVESTREAM* fp)
+{
+    KEYVALUE* p;
+    TABLE* t = GetTable(it8);
+
+    
+    for (p = t->HeaderList; (p != NULL); p = p->Next)
+    {
+        if (*p ->Keyword == '#') {
+
+            char* Pt;
+            
+            WriteStr(fp, "#\n# ");
+            for (Pt = p ->Value; *Pt; Pt++) {
+
+
+                Writef(fp, "%c", *Pt);                
+
+                if (*Pt == '\n') {
+                    WriteStr(fp, "# ");
+                }
+            }
+        
+            WriteStr(fp, "\n#\n");
+            continue;
+        }
+
+
+        if (!IsAvailableOnList(it8-> ValidKeywords, p->Keyword, NULL, NULL)) {
+
+#ifdef CMS_STRICT_CGATS
+            WriteStr(fp, "KEYWORD\t\"");
+            WriteStr(fp, p->Keyword);
+            WriteStr(fp, "\"\n");
+#endif
+
+            AddAvailableProperty(it8, p->Keyword, WRITE_UNCOOKED);
+        }
+
+        WriteStr(fp, p->Keyword);
+        if (p->Value) {
+
+            switch (p ->WriteAs) {
+
+            case WRITE_UNCOOKED:
+                    Writef(fp, "\t%s", p ->Value);                    
+                    break;
+
+            case WRITE_STRINGIFY:
+                    Writef(fp, "\t\"%s\"", p->Value );
+                    break;
+
+            case WRITE_HEXADECIMAL:
+                    Writef(fp, "\t0x%X", atoi(p ->Value));
+                    break;
+
+            case WRITE_BINARY:
+                    Writef(fp, "\t0x%B", atoi(p ->Value));
+                    break;
+
+            case WRITE_PAIR:
+                    Writef(fp, "\t\"%s,%s\"", p->Subkey, p->Value);
+                    break;
+
+            default: SynError(it8, "Unknown write mode %d", p ->WriteAs);
+                     return;
+            }
+        }
+
+        WriteStr (fp, "\n");
+    }
+
+}
+
+
+// Writes the data format
+static
+void WriteDataFormat(SAVESTREAM* fp, cmsIT8* it8)
+{
+    int i, nSamples;
+    TABLE* t = GetTable(it8);
+
+    if (!t -> DataFormat) return;
+
+       WriteStr(fp, "BEGIN_DATA_FORMAT\n");
+       WriteStr(fp, " ");
+       nSamples = atoi(cmsIT8GetProperty(it8, "NUMBER_OF_FIELDS"));
+
+       for (i = 0; i < nSamples; i++) {
+
+              WriteStr(fp, t->DataFormat[i]);
+              WriteStr(fp, ((i == (nSamples-1)) ? "\n" : "\t"));
+          }
+
+       WriteStr (fp, "END_DATA_FORMAT\n");
+}
+
+
+// Writes data array
+static
+void WriteData(SAVESTREAM* fp, cmsIT8* it8)
+{
+       int  i, j;
+       TABLE* t = GetTable(it8);
+
+       if (!t->Data) return;
+
+       WriteStr (fp, "BEGIN_DATA\n");
+
+       t->nPatches = atoi(cmsIT8GetProperty(it8, "NUMBER_OF_SETS"));
+
+       for (i = 0; i < t-> nPatches; i++) {
+
+              WriteStr(fp, " ");
+
+              for (j = 0; j < t->nSamples; j++) {
+
+                     char *ptr = t->Data[i*t->nSamples+j];
+
+                     if (ptr == NULL) WriteStr(fp, "\"\"");
+                     else {
+                         // If value contains whitespace, enclose within quote
+
+                         if (strchr(ptr, ' ') != NULL) {
+
+                             WriteStr(fp, "\"");
+                             WriteStr(fp, ptr);
+                             WriteStr(fp, "\"");
+                         }
+                         else
+                            WriteStr(fp, ptr);
+                     }
+
+                     WriteStr(fp, ((j == (t->nSamples-1)) ? "\n" : "\t"));
+              }
+       }
+       WriteStr (fp, "END_DATA\n");
+}
+
+
+
+// Saves whole file
+cmsBool CMSEXPORT cmsIT8SaveToFile(cmsHANDLE hIT8, const char* cFileName)
+{
+    SAVESTREAM sd;  
+    cmsUInt32Number i;
+    cmsIT8* it8 = (cmsIT8*) hIT8;
+
+    memset(&sd, 0, sizeof(sd));
+
+    sd.stream = fopen(cFileName, "wt");
+    if (!sd.stream) return FALSE;
+    
+    WriteStr(&sd, it8->SheetType);
+    WriteStr(&sd, "\n");
+    for (i=0; i < it8 ->TablesCount; i++) {
+
+            cmsIT8SetTable(hIT8, i);
+            WriteHeader(it8, &sd);
+            WriteDataFormat(&sd, it8);
+            WriteData(&sd, it8);
+    }
+    
+    if (fclose(sd.stream) != 0) return FALSE;
+
+    return TRUE;
+}
+
+
+// Saves to memory
+cmsBool CMSEXPORT cmsIT8SaveToMem(cmsHANDLE hIT8, void *MemPtr, cmsUInt32Number* BytesNeeded)
+{
+    SAVESTREAM sd;  
+    cmsUInt32Number i;
+    cmsIT8* it8 = (cmsIT8*) hIT8;
+
+    memset(&sd, 0, sizeof(sd));
+
+    sd.stream = NULL;
+    sd.Base   = (cmsUInt8Number*)  MemPtr;
+    sd.Ptr    = sd.Base;
+
+    sd.Used = 0;
+
+    if (sd.Base) 
+        sd.Max  = *BytesNeeded;     // Write to memory?
+    else 
+        sd.Max  = 0;                // Just counting the needed bytes
+   
+    WriteStr(&sd, it8->SheetType);
+    WriteStr(&sd, "\n");
+    for (i=0; i < it8 ->TablesCount; i++) {
+
+            cmsIT8SetTable(hIT8, i);
+            WriteHeader(it8, &sd);
+            WriteDataFormat(&sd, it8);
+            WriteData(&sd, it8);
+    }
+    
+    sd.Used++;  // The \0 at the very end
+
+    if (sd.Base)
+        sd.Ptr = 0;
+
+    *BytesNeeded = sd.Used;
+
+    return TRUE;
+}
+
+
+// -------------------------------------------------------------- Higer level parsing
+
+static
+cmsBool DataFormatSection(cmsIT8* it8)
+{
+    int iField = 0;    
+    TABLE* t = GetTable(it8);
+
+    InSymbol(it8);   // Eats "BEGIN_DATA_FORMAT"
+    CheckEOLN(it8);
+
+    while (it8->sy != SEND_DATA_FORMAT &&
+        it8->sy != SEOLN &&
+        it8->sy != SEOF &&
+        it8->sy != SSYNERROR)  {
+        
+            if (it8->sy != SIDENT) {
+            
+                return SynError(it8, "Sample type expected");                    
+            }
+        
+            if (!SetDataFormat(it8, iField, it8->id)) return FALSE;
+            iField++;
+                
+            InSymbol(it8);
+            SkipEOLN(it8);
+       }
+
+       SkipEOLN(it8);
+       Skip(it8, SEND_DATA_FORMAT);
+       SkipEOLN(it8);
+
+       if (iField != t ->nSamples) {
+           SynError(it8, "Count mismatch. NUMBER_OF_FIELDS was %d, found %d\n", t ->nSamples, iField);
+
+           
+       }
+
+       return TRUE;
+}
+
+
+
+static
+cmsBool DataSection (cmsIT8* it8)
+{
+    int  iField = 0;
+    int  iSet   = 0;
+    char Buffer[256];
+    TABLE* t = GetTable(it8);
+
+    InSymbol(it8);   // Eats "BEGIN_DATA"
+    CheckEOLN(it8);
+
+    if (!t->Data)
+        AllocateDataSet(it8);
+
+    while (it8->sy != SEND_DATA && it8->sy != SEOF)
+    {
+        if (iField >= t -> nSamples) {
+            iField = 0;
+            iSet++;
+          
+        }
+
+        if (it8->sy != SEND_DATA && it8->sy != SEOF) {
+
+            if (!GetVal(it8, Buffer, 255, "Sample data expected"))
+                return FALSE;
+
+            if (!SetData(it8, iSet, iField, Buffer))
+                return FALSE;
+
+            iField++;
+            
+            InSymbol(it8);
+            SkipEOLN(it8);           
+        }
+    }
+
+    SkipEOLN(it8);
+    Skip(it8, SEND_DATA);
+    SkipEOLN(it8);
+ 
+    // Check for data completion.
+
+    if ((iSet+1) != t -> nPatches)
+        return SynError(it8, "Count mismatch. NUMBER_OF_SETS was %d, found %d\n", t ->nPatches, iSet+1);
+
+    return TRUE;
+}
+
+
+
+
+static
+cmsBool HeaderSection(cmsIT8* it8)
+{
+    char VarName[MAXID];
+    char Buffer[MAXSTR];
+    KEYVALUE* Key;
+
+        while (it8->sy != SEOF &&
+               it8->sy != SSYNERROR &&
+               it8->sy != SBEGIN_DATA_FORMAT &&
+               it8->sy != SBEGIN_DATA) {
+
+
+        switch (it8 -> sy) {
+
+        case SKEYWORD:
+                InSymbol(it8);
+                if (!GetVal(it8, Buffer, MAXSTR-1, "Keyword expected")) return FALSE;                
+                if (!AddAvailableProperty(it8, Buffer, WRITE_UNCOOKED)) return FALSE;
+                InSymbol(it8);
+                break;
+
+
+        case SDATA_FORMAT_ID:
+                InSymbol(it8);
+                if (!GetVal(it8, Buffer, MAXSTR-1, "Keyword expected")) return FALSE;                
+                if (!AddAvailableSampleID(it8, Buffer)) return FALSE;
+                InSymbol(it8);
+                break;
+
+
+        case SIDENT:
+                strncpy(VarName, it8->id, MAXID-1);
+                VarName[MAXID-1] = 0;
+                
+                if (!IsAvailableOnList(it8-> ValidKeywords, VarName, NULL, &Key)) {
+
+#ifdef CMS_STRICT_CGATS               
+                 return SynError(it8, "Undefined keyword '%s'", VarName);
+#else
+                    Key = AddAvailableProperty(it8, VarName, WRITE_UNCOOKED);
+                    if (Key == NULL) return FALSE;
+#endif
+                }
+
+                InSymbol(it8);
+                if (!GetVal(it8, Buffer, MAXSTR-1, "Property data expected")) return FALSE;
+
+                if(Key->WriteAs != WRITE_PAIR) {
+                    AddToList(it8, &GetTable(it8)->HeaderList, VarName, NULL, Buffer, 
+                                (it8->sy == SSTRING) ? WRITE_STRINGIFY : WRITE_UNCOOKED);
+                }
+                else {
+                    const char *Subkey;
+                    char *Nextkey;
+                    if (it8->sy != SSTRING)
+                        return SynError(it8, "Invalid value '%s' for property '%s'.", Buffer, VarName);
+
+                    // chop the string as a list of "subkey, value" pairs, using ';' as a separator
+                    for (Subkey = Buffer; Subkey != NULL; Subkey = Nextkey)
+                    {
+                        char *Value, *temp;
+                
+                        //  identify token pair boundary
+                        Nextkey = (char*) strchr(Subkey, ';');
+                        if(Nextkey)
+                            *Nextkey++ = '\0';
+
+                        // for each pair, split the subkey and the value
+                        Value = (char*) strrchr(Subkey, ',');
+                        if(Value == NULL)
+                            return SynError(it8, "Invalid value for property '%s'.", VarName);
+
+                        // gobble the spaces before the coma, and the coma itself
+                        temp = Value++;
+                        do *temp-- = '\0'; while(temp >= Subkey && *temp == ' ');
+
+                        // gobble any space at the right
+                        temp = Value + strlen(Value) - 1;
+                        while(*temp == ' ') *temp-- = '\0'; 
+
+                        // trim the strings from the left
+                        Subkey += strspn(Subkey, " ");
+                        Value += strspn(Value, " ");
+
+                        if(Subkey[0] == 0 || Value[0] == 0)
+                            return SynError(it8, "Invalid value for property '%s'.", VarName);
+                        AddToList(it8, &GetTable(it8)->HeaderList, VarName, Subkey, Value, WRITE_PAIR);
+                    }
+                }
+                
+                InSymbol(it8);
+                break;
+
+        
+        case SEOLN: break;
+
+        default:
+                return SynError(it8, "expected keyword or identifier");
+        }
+
+    SkipEOLN(it8);
+    }
+
+    return TRUE;
+
+}
+
+
+static
+cmsBool ParseIT8(cmsIT8* it8, cmsBool nosheet)
+{
+    char* SheetTypePtr = it8 ->SheetType;
+
+    if (nosheet == 0) {
+
+    // First line is a very special case.
+
+    while (isseparator(it8->ch))
+            NextCh(it8);
+    
+    while (it8->ch != '\r' && it8 ->ch != '\n' && it8->ch != '\t' && it8 -> ch != -1) {
+
+        *SheetTypePtr++= (char) it8 ->ch;
+        NextCh(it8);
+    }
+    }
+
+    *SheetTypePtr = 0;
+    InSymbol(it8);
+   
+    SkipEOLN(it8);
+
+    while (it8-> sy != SEOF &&
+           it8-> sy != SSYNERROR) {
+
+            switch (it8 -> sy) {
+
+            case SBEGIN_DATA_FORMAT:
+                    if (!DataFormatSection(it8)) return FALSE;
+                    break;
+
+            case SBEGIN_DATA:
+
+                    if (!DataSection(it8)) return FALSE;
+                    
+                    if (it8 -> sy != SEOF) {
+
+                            AllocTable(it8);
+                            it8 ->nTable = it8 ->TablesCount - 1;
+                    }
+                    break;
+
+            case SEOLN:
+                    SkipEOLN(it8);
+                    break;
+
+            default:
+                    if (!HeaderSection(it8)) return FALSE;
+           }
+
+    }
+
+    return (it8 -> sy != SSYNERROR);
+}
+
+
+
+// Init usefull pointers
+
+static
+void CookPointers(cmsIT8* it8)
+{
+    int idField, i;
+    char* Fld;
+    cmsUInt32Number j;
+    cmsUInt32Number nOldTable = it8 ->nTable;
+
+    for (j=0; j < it8 ->TablesCount; j++) {
+
+    TABLE* t = it8 ->Tab + j;
+
+    t -> SampleID = 0;
+    it8 ->nTable = j;    
+
+    for (idField = 0; idField < t -> nSamples; idField++)
+    {
+        if (t ->DataFormat == NULL){
+            SynError(it8, "Undefined DATA_FORMAT");
+            return;
+        }
+
+        Fld = t->DataFormat[idField];
+        if (!Fld) continue;
+
+
+        if (cmsstrcasecmp(Fld, "SAMPLE_ID") == 0) {
+
+                    t -> SampleID = idField;
+                
+        for (i=0; i < t -> nPatches; i++) {
+
+                char *Data = GetData(it8, i, idField);
+                if (Data) {
+                    char Buffer[256];
+                
+                    strncpy(Buffer, Data, 255);
+                    Buffer[255] = 0;
+                                       
+                    if (strlen(Buffer) <= strlen(Data))
+                        strcpy(Data, Buffer);
+                    else
+                        SetData(it8, i, idField, Buffer);
+
+                }
+                }
+        
+        }
+
+        // "LABEL" is an extension. It keeps references to forward tables
+         
+        if ((cmsstrcasecmp(Fld, "LABEL") == 0) || Fld[0] == '$' ) {
+                                        
+                    // Search for table references...
+                    for (i=0; i < t -> nPatches; i++) {
+
+                            char *Label = GetData(it8, i, idField);
+                           
+                            if (Label) {                                
+                                
+                                cmsUInt32Number k;
+
+                                // This is the label, search for a table containing 
+                                // this property
+
+                                for (k=0; k < it8 ->TablesCount; k++) {
+
+                                    TABLE* Table = it8 ->Tab + k;
+                                    KEYVALUE* p;
+
+                                    if (IsAvailableOnList(Table->HeaderList, Label, NULL, &p)) {
+
+                                        // Available, keep type and table
+                                        char Buffer[256];
+
+                                        char *Type  = p ->Value;
+                                        int  nTable = k;                                        
+
+                                        snprintf(Buffer, 255, "%s %d %s", Label, nTable, Type );
+                                                                                
+                                        SetData(it8, i, idField, Buffer);
+                                    }
+                                }
+
+
+                            }
+
+                    }
+
+
+        }
+
+    }
+    }
+
+    it8 ->nTable = nOldTable;
+}
+
+// Try to infere if the file is a CGATS/IT8 file at all. Read first line
+// that should be something like some printable characters plus a \n
+
+static
+int IsMyBlock(cmsUInt8Number* Buffer, int n)
+{
+    int cols = 1, space = 0, quot = 0;
+    int i;
+
+    if (n < 10) return FALSE;   // Too small
+
+    if (n > 132)
+        n = 132;
+
+    for (i = 1; i < n; i++) {
+
+        switch(Buffer[i])
+        {
+        case '\n':
+        case '\r':
+            return quot == 1 || cols > 2 ? 0 : cols;
+        case '\t':
+        case ' ':
+            if(!quot && !space)
+                space = 1;
+            break;
+        case '\"':
+            quot = !quot;
+            break;
+        default:
+            if (Buffer[i] < 32) return 0;
+            if (Buffer[i] > 127) return 0;
+            cols += space;
+            space = 0;
+            break;
+        }
+    }
+
+    return FALSE;
+
+}
+
+
+static
+cmsBool IsMyFile(const char* FileName)
+{
+   FILE *fp;
+   cmsUInt32Number Size;
+   cmsUInt8Number Ptr[133];
+
+   fp = fopen(FileName, "rt");
+   if (!fp) {
+       cmsSignalError(0, cmsERROR_FILE, "File '%s' not found", FileName);
+       return FALSE;
+   }
+
+   Size = (cmsUInt32Number) fread(Ptr, 1, 132, fp);
+
+   if (fclose(fp) != 0) 
+	   return FALSE;
+
+   Ptr[Size] = '\0';
+
+   return IsMyBlock(Ptr, Size);
+}
+
+// ---------------------------------------------------------- Exported routines
+
+
+cmsHANDLE  CMSEXPORT cmsIT8LoadFromMem(cmsContext ContextID, void *Ptr, cmsUInt32Number len)
+{
+    cmsHANDLE hIT8; 
+    cmsIT8*  it8;
+	int type;
+
+	_cmsAssert(Ptr != NULL);
+	_cmsAssert(len != 0);
+
+    type = IsMyBlock((cmsUInt8Number*)Ptr, len);
+    if (type == 0) return NULL;
+    
+    hIT8 = cmsIT8Alloc(ContextID);
+    if (!hIT8) return NULL;
+
+    it8 = (cmsIT8*) hIT8;
+    it8 ->MemoryBlock = (char*) _cmsMalloc(ContextID, len + 1);
+
+    strncpy(it8 ->MemoryBlock, (const char*) Ptr, len);
+    it8 ->MemoryBlock[len] = 0;
+
+    strncpy(it8->FileStack[0]->FileName, "", cmsMAX_PATH-1);
+    it8-> Source = it8 -> MemoryBlock;
+
+    if (!ParseIT8(it8, type-1)) { 
+       
+        cmsIT8Free(hIT8); 
+        return FALSE; 
+    }
+
+    CookPointers(it8);
+    it8 ->nTable = 0;
+
+    _cmsFree(ContextID, it8->MemoryBlock);
+    it8 -> MemoryBlock = NULL;
+
+    return hIT8;
+
+
+}
+
+
+cmsHANDLE  CMSEXPORT cmsIT8LoadFromFile(cmsContext ContextID, const char* cFileName)
+{
+
+     cmsHANDLE hIT8; 
+     cmsIT8*  it8;
+	 int type;
+	 
+	 _cmsAssert(cFileName != NULL);
+
+	 type = IsMyFile(cFileName);
+     if (type == 0) return NULL;
+
+     hIT8 = cmsIT8Alloc(ContextID);
+     it8 = (cmsIT8*) hIT8;   
+     if (!hIT8) return NULL;
+
+
+     it8 ->FileStack[0]->Stream = fopen(cFileName, "rt");
+
+     if (!it8 ->FileStack[0]->Stream) {         
+         cmsIT8Free(hIT8);
+         return NULL;
+     }
+     
+
+    strncpy(it8->FileStack[0]->FileName, cFileName, cmsMAX_PATH-1);    
+    it8->FileStack[0]->FileName[cmsMAX_PATH-1] = 0;
+
+    if (!ParseIT8(it8, type-1)) { 
+    
+            fclose(it8 ->FileStack[0]->Stream);
+            cmsIT8Free(hIT8); 
+            return NULL; 
+    }
+
+    CookPointers(it8);
+    it8 ->nTable = 0;
+
+	if (fclose(it8 ->FileStack[0]->Stream)!= 0) {
+		    cmsIT8Free(hIT8); 
+            return NULL; 
+	}
+
+    return hIT8;
+
+}
+
+int CMSEXPORT cmsIT8EnumDataFormat(cmsHANDLE hIT8, char ***SampleNames)
+{
+	cmsIT8* it8 = (cmsIT8*) hIT8;
+	TABLE* t;
+
+	_cmsAssert(hIT8 != NULL);
+
+	t = GetTable(it8);
+
+	if (SampleNames) 
+		*SampleNames = t -> DataFormat;
+	return t -> nSamples;
+}
+
+
+cmsUInt32Number CMSEXPORT cmsIT8EnumProperties(cmsHANDLE hIT8, char ***PropertyNames)
+{
+    cmsIT8* it8 = (cmsIT8*) hIT8;
+    KEYVALUE* p;
+    cmsUInt32Number n;
+    char **Props;
+    TABLE* t;
+	
+	_cmsAssert(hIT8 != NULL);
+
+	t = GetTable(it8);
+
+    // Pass#1 - count properties
+
+    n = 0;
+    for (p = t -> HeaderList;  p != NULL; p = p->Next) {
+        n++;
+    }
+
+
+    Props = (char **) AllocChunk(it8, sizeof(char *) * n);
+
+    // Pass#2 - Fill pointers
+    n = 0;
+    for (p = t -> HeaderList;  p != NULL; p = p->Next) {
+        Props[n++] = p -> Keyword;
+    }
+
+    *PropertyNames = Props;
+    return n;
+}
+
+cmsUInt32Number CMSEXPORT cmsIT8EnumPropertyMulti(cmsHANDLE hIT8, const char* cProp, const char ***SubpropertyNames)
+{
+    cmsIT8* it8 = (cmsIT8*) hIT8;
+    KEYVALUE *p, *tmp;
+    cmsUInt32Number n;
+    const char **Props;
+    TABLE* t;
+	
+	_cmsAssert(hIT8 != NULL);
+
+
+	t = GetTable(it8);
+
+    if(!IsAvailableOnList(t->HeaderList, cProp, NULL, &p)) {
+        *SubpropertyNames = 0;
+        return 0;
+    }
+
+    // Pass#1 - count properties
+
+    n = 0;
+    for (tmp = p;  tmp != NULL; tmp = tmp->NextSubkey) {
+        if(tmp->Subkey != NULL)
+            n++;
+    }
+
+
+    Props = (const char **) AllocChunk(it8, sizeof(char *) * n);
+
+    // Pass#2 - Fill pointers
+    n = 0;
+    for (tmp = p;  tmp != NULL; tmp = tmp->NextSubkey) {
+        if(tmp->Subkey != NULL)
+            Props[n++] = p ->Subkey;
+    }
+
+    *SubpropertyNames = Props;
+    return n;
+}
+
+static
+int LocatePatch(cmsIT8* it8, const char* cPatch)
+{
+    int i;
+    const char *data;
+    TABLE* t = GetTable(it8);
+
+    for (i=0; i < t-> nPatches; i++) {
+
+        data = GetData(it8, i, t->SampleID);
+        
+        if (data != NULL) {
+
+                if (cmsstrcasecmp(data, cPatch) == 0)
+                        return i;
+                }
+        }
+
+        // SynError(it8, "Couldn't find patch '%s'\n", cPatch);                               
+        return -1;
+}
+
+
+static
+int LocateEmptyPatch(cmsIT8* it8)
+{
+    int i;
+    const char *data;
+    TABLE* t = GetTable(it8);
+
+    for (i=0; i < t-> nPatches; i++) {
+
+        data = GetData(it8, i, t->SampleID);
+
+        if (data == NULL) 
+            return i;
+
+    }
+
+    return -1;
+}
+
+static
+int LocateSample(cmsIT8* it8, const char* cSample)
+{
+    int i;
+    const char *fld;
+    TABLE* t = GetTable(it8);
+
+    for (i=0; i < t->nSamples; i++) {
+
+        fld = GetDataFormat(it8, i);
+        if (cmsstrcasecmp(fld, cSample) == 0)
+            return i;
+    }
+   
+    return -1;
+
+}
+
+
+int CMSEXPORT cmsIT8FindDataFormat(cmsHANDLE hIT8, const char* cSample)
+{
+    cmsIT8* it8 = (cmsIT8*) hIT8;
+
+	_cmsAssert(hIT8 != NULL);
+
+    return LocateSample(it8, cSample);
+}
+
+
+
+const char* CMSEXPORT cmsIT8GetDataRowCol(cmsHANDLE hIT8, int row, int col)
+{
+    cmsIT8* it8 = (cmsIT8*) hIT8;
+
+	_cmsAssert(hIT8 != NULL);
+
+    return GetData(it8, row, col);
+}
+
+
+cmsFloat64Number CMSEXPORT cmsIT8GetDataRowColDbl(cmsHANDLE hIT8, int row, int col)
+{
+    const char* Buffer;
+
+    Buffer = cmsIT8GetDataRowCol(hIT8, row, col);
+    
+    if (Buffer) {
+
+        return atof(Buffer);
+        
+    } else
+        return 0;
+
+}
+
+
+cmsBool CMSEXPORT cmsIT8SetDataRowCol(cmsHANDLE hIT8, int row, int col, const char* Val)
+{
+    cmsIT8* it8 = (cmsIT8*) hIT8;
+    
+	_cmsAssert(hIT8 != NULL);
+
+    return SetData(it8, row, col, Val);        
+}
+
+
+cmsBool CMSEXPORT cmsIT8SetDataRowColDbl(cmsHANDLE hIT8, int row, int col, cmsFloat64Number Val)
+{
+    cmsIT8* it8 = (cmsIT8*) hIT8;
+    char Buff[256];
+
+	_cmsAssert(hIT8 != NULL);
+
+    sprintf(Buff, it8->DoubleFormatter, Val);
+    
+    return SetData(it8, row, col, Buff);        
+}
+
+
+
+const char* CMSEXPORT cmsIT8GetData(cmsHANDLE hIT8, const char* cPatch, const char* cSample)                        
+{
+    cmsIT8* it8 = (cmsIT8*) hIT8;
+    int iField, iSet;
+
+	_cmsAssert(hIT8 != NULL);
+
+    iField = LocateSample(it8, cSample);
+    if (iField < 0) {       
+        return NULL;
+    }
+
+    iSet = LocatePatch(it8, cPatch);
+    if (iSet < 0) {          
+            return NULL;
+    }
+
+    return GetData(it8, iSet, iField);
+}
+
+
+cmsFloat64Number CMSEXPORT cmsIT8GetDataDbl(cmsHANDLE  it8, const char* cPatch, const char* cSample)
+{
+    const char* Buffer;
+
+    Buffer = cmsIT8GetData(it8, cPatch, cSample);
+    
+    if (Buffer) {
+
+        return atof(Buffer);
+        
+    } else {
+        
+        return 0;
+    }
+}
+
+
+
+cmsBool CMSEXPORT cmsIT8SetData(cmsHANDLE hIT8, const char* cPatch, const char* cSample, const char *Val)
+{
+    cmsIT8* it8 = (cmsIT8*) hIT8;
+    int iField, iSet;
+    TABLE* t;
+	
+	_cmsAssert(hIT8 != NULL);
+
+	t = GetTable(it8);
+
+    iField = LocateSample(it8, cSample);
+
+    if (iField < 0) 
+        return FALSE;
+
+    if (t-> nPatches == 0) {
+
+        AllocateDataFormat(it8);
+        AllocateDataSet(it8);
+        CookPointers(it8);
+    }
+
+    if (cmsstrcasecmp(cSample, "SAMPLE_ID") == 0) {
+
+        iSet   = LocateEmptyPatch(it8);
+        if (iSet < 0) {
+            return SynError(it8, "Couldn't add more patches '%s'\n", cPatch);                        
+        }
+
+        iField = t -> SampleID;
+    }
+    else {
+        iSet = LocatePatch(it8, cPatch);
+        if (iSet < 0) {
+            return FALSE;
+        }
+    }
+
+    return SetData(it8, iSet, iField, Val);
+}
+
+
+cmsBool CMSEXPORT cmsIT8SetDataDbl(cmsHANDLE hIT8, const char* cPatch,
+								   const char* cSample,
+								   cmsFloat64Number Val)
+{
+	cmsIT8* it8 = (cmsIT8*) hIT8;
+	char Buff[256];
+
+	_cmsAssert(hIT8 != NULL);
+
+	snprintf(Buff, 255, it8->DoubleFormatter, Val);
+	return cmsIT8SetData(hIT8, cPatch, cSample, Buff);
+}
+
+// Buffer should get MAXSTR at least
+
+const char* CMSEXPORT cmsIT8GetPatchName(cmsHANDLE hIT8, int nPatch, char* buffer)
+{
+	cmsIT8* it8 = (cmsIT8*) hIT8;
+	TABLE* t;
+	char* Data;
+
+	_cmsAssert(hIT8 != NULL);
+
+	t = GetTable(it8);
+	Data = GetData(it8, nPatch, t->SampleID);
+
+	if (!Data) return NULL;
+	if (!buffer) return Data;
+
+	strncpy(buffer, Data, MAXSTR-1);        
+	buffer[MAXSTR-1] = 0;
+	return buffer;
+}
+
+int CMSEXPORT cmsIT8GetPatchByName(cmsHANDLE hIT8, const char *cPatch)
+{
+	_cmsAssert(hIT8 != NULL);
+
+    return LocatePatch((cmsIT8*)hIT8, cPatch);
+}
+
+cmsUInt32Number CMSEXPORT cmsIT8TableCount(cmsHANDLE hIT8)
+{
+	cmsIT8* it8 = (cmsIT8*) hIT8;
+
+	_cmsAssert(hIT8 != NULL);
+
+	return it8 ->TablesCount;
+}
+
+// This handles the "LABEL" extension. 
+// Label, nTable, Type
+
+int CMSEXPORT cmsIT8SetTableByLabel(cmsHANDLE hIT8, const char* cSet, const char* cField, const char* ExpectedType)
+{
+    const char* cLabelFld;
+    char Type[256], Label[256];
+    int nTable;
+   
+    _cmsAssert(hIT8 != NULL);
+
+    if (cField != NULL && *cField == 0)
+            cField = "LABEL";
+
+    if (cField == NULL) 
+            cField = "LABEL";
+
+    cLabelFld = cmsIT8GetData(hIT8, cSet, cField); 
+    if (!cLabelFld) return -1;
+    
+    if (sscanf(cLabelFld, "%255s %d %255s", Label, &nTable, Type) != 3)
+            return -1;
+    
+    if (ExpectedType != NULL && *ExpectedType == 0)
+        ExpectedType = NULL;
+
+    if (ExpectedType) {
+
+        if (cmsstrcasecmp(Type, ExpectedType) != 0) return -1;
+    }
+
+    return cmsIT8SetTable(hIT8, nTable);    
+}
+
+
+cmsBool CMSEXPORT cmsIT8SetIndexColumn(cmsHANDLE hIT8, const char* cSample)
+{
+	cmsIT8* it8 = (cmsIT8*) hIT8;
+	int pos;
+
+	_cmsAssert(hIT8 != NULL);
+
+	pos = LocateSample(it8, cSample);
+	if(pos == -1)
+		return FALSE;
+
+	it8->Tab[it8->nTable].SampleID = pos;
+	return TRUE;
+}
+
+
+void CMSEXPORT cmsIT8DefineDblFormat(cmsHANDLE hIT8, const char* Formatter)
+{
+    cmsIT8* it8 = (cmsIT8*) hIT8;
+
+    _cmsAssert(hIT8 != NULL);
+
+    if (Formatter == NULL)
+        strcpy(it8->DoubleFormatter, DEFAULT_DBL_FORMAT);
+    else
+        strcpy(it8->DoubleFormatter, Formatter);
+}
+
diff --git a/thirdparty/liblcms2/src/cmscnvrt.c b/thirdparty/liblcms2/src/cmscnvrt.c
new file mode 100644
index 00000000..8dadc875
--- /dev/null
+++ b/thirdparty/liblcms2/src/cmscnvrt.c
@@ -0,0 +1,1039 @@
+//---------------------------------------------------------------------------------
+//
+//  Little Color Management System
+//  Copyright (c) 1998-2010 Marti Maria Saguer
+//
+// Permission is hereby granted, free of charge, to any person obtaining 
+// a copy of this software and associated documentation files (the "Software"), 
+// to deal in the Software without restriction, including without limitation 
+// the rights to use, copy, modify, merge, publish, distribute, sublicense, 
+// and/or sell copies of the Software, and to permit persons to whom the Software 
+// is furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in 
+// all copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 
+// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO 
+// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE 
+// LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 
+// OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 
+// WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+//
+//---------------------------------------------------------------------------------
+//
+
+#include "lcms2_internal.h"
+
+
+// Link several profiles to obtain a single LUT modelling the whole color transform. Intents, Black point 
+// compensation and Adaptation parameters may vary across profiles. BPC and Adaptation refers to the PCS
+// after the profile. I.e, BPC[0] refers to connexion between profile(0) and profile(1)
+cmsPipeline* _cmsLinkProfiles(cmsContext     ContextID, 
+                              cmsUInt32Number nProfiles,
+                              cmsUInt32Number Intents[], 
+                              cmsHPROFILE     hProfiles[], 
+                              cmsBool         BPC[],
+                              cmsFloat64Number AdaptationStates[],
+                              cmsUInt32Number dwFlags);
+                                                        
+//---------------------------------------------------------------------------------
+
+// This is the default routine for ICC-style intents. A user may decide to override it by using a plugin. 
+// Supported intents are perceptual, relative colorimetric, saturation and ICC-absolute colorimetric
+static 
+cmsPipeline* DefaultICCintents(cmsContext     ContextID, 
+                               cmsUInt32Number nProfiles,
+                               cmsUInt32Number Intents[], 
+                               cmsHPROFILE     hProfiles[], 
+                               cmsBool         BPC[],
+                               cmsFloat64Number AdaptationStates[],
+                               cmsUInt32Number dwFlags);
+
+//---------------------------------------------------------------------------------
+
+// This is the entry for black-preserving K-only intents, which are non-ICC. Last profile have to be a output profile
+// to do the trick (no devicelinks allowed at that position)
+static
+cmsPipeline*  BlackPreservingKOnlyIntents(cmsContext     ContextID, 
+                                          cmsUInt32Number nProfiles,
+                                          cmsUInt32Number Intents[], 
+                                          cmsHPROFILE     hProfiles[], 
+                                          cmsBool         BPC[],
+                                          cmsFloat64Number AdaptationStates[],
+                                          cmsUInt32Number dwFlags);
+
+//---------------------------------------------------------------------------------
+
+// This is the entry for black-plane preserving, which are non-ICC. Again, Last profile have to be a output profile
+// to do the trick (no devicelinks allowed at that position)
+static
+cmsPipeline*  BlackPreservingKPlaneIntents(cmsContext     ContextID, 
+                                           cmsUInt32Number nProfiles,
+                                           cmsUInt32Number Intents[], 
+                                           cmsHPROFILE     hProfiles[], 
+                                           cmsBool         BPC[],
+                                           cmsFloat64Number AdaptationStates[],
+                                           cmsUInt32Number dwFlags);
+
+//---------------------------------------------------------------------------------
+
+
+// This is a structure holding implementations for all supported intents.
+typedef struct _cms_intents_list {
+
+    cmsUInt32Number Intent;
+    char            Description[256];
+    cmsIntentFn     Link;
+    struct _cms_intents_list*  Next;
+
+} cmsIntentsList;
+
+
+// Built-in intents
+static cmsIntentsList DefaultIntents[] = { 
+
+    { INTENT_PERCEPTUAL,                            "Perceptual",                                   DefaultICCintents,            &DefaultIntents[1] },
+    { INTENT_RELATIVE_COLORIMETRIC,                 "Relative colorimetric",                        DefaultICCintents,            &DefaultIntents[2] },
+    { INTENT_SATURATION,                            "Saturation",                                   DefaultICCintents,            &DefaultIntents[3] },
+    { INTENT_ABSOLUTE_COLORIMETRIC,                 "Absolute colorimetric",                        DefaultICCintents,            &DefaultIntents[4] },
+    { INTENT_PRESERVE_K_ONLY_PERCEPTUAL,            "Perceptual preserving black ink",              BlackPreservingKOnlyIntents,  &DefaultIntents[5] },
+    { INTENT_PRESERVE_K_ONLY_RELATIVE_COLORIMETRIC, "Relative colorimetric preserving black ink",   BlackPreservingKOnlyIntents,  &DefaultIntents[6] },
+    { INTENT_PRESERVE_K_ONLY_SATURATION,            "Saturation preserving black ink",              BlackPreservingKOnlyIntents,  &DefaultIntents[7] },
+    { INTENT_PRESERVE_K_PLANE_PERCEPTUAL,           "Perceptual preserving black plane",            BlackPreservingKPlaneIntents, &DefaultIntents[8] },
+    { INTENT_PRESERVE_K_PLANE_RELATIVE_COLORIMETRIC,"Relative colorimetric preserving black plane", BlackPreservingKPlaneIntents, &DefaultIntents[9] },
+    { INTENT_PRESERVE_K_PLANE_SATURATION,           "Saturation preserving black plane",            BlackPreservingKPlaneIntents, NULL } 
+};
+
+
+// A pointer to the begining of the list
+static cmsIntentsList *Intents = DefaultIntents;
+
+// Search the list for a suitable intent. Returns NULL if not found
+static 
+cmsIntentsList* SearchIntent(cmsUInt32Number Intent)
+{
+    cmsIntentsList* pt;
+
+    for (pt = Intents; pt != NULL; pt = pt -> Next)
+        if (pt ->Intent == Intent) return pt;
+
+    return NULL;
+}
+
+// Black point compensation. Implemented as a linear scaling in XYZ. Black points 
+// should come relative to the white point. Fills an matrix/offset element m
+// which is organized as a 4x4 matrix.
+static
+void ComputeBlackPointCompensation(const cmsCIEXYZ* BlackPointIn, 
+                                   const cmsCIEXYZ* BlackPointOut,
+                                   cmsMAT3* m, cmsVEC3* off)
+{ 
+  cmsFloat64Number ax, ay, az, bx, by, bz, tx, ty, tz;
+   
+   // Now we need to compute a matrix plus an offset m and of such of
+   // [m]*bpin + off = bpout
+   // [m]*D50  + off = D50
+   //
+   // This is a linear scaling in the form ax+b, where
+   // a = (bpout - D50) / (bpin - D50)
+   // b = - D50* (bpout - bpin) / (bpin - D50)
+
+   tx = BlackPointIn->X - cmsD50_XYZ()->X;
+   ty = BlackPointIn->Y - cmsD50_XYZ()->Y;
+   tz = BlackPointIn->Z - cmsD50_XYZ()->Z;
+
+   ax = (BlackPointOut->X - cmsD50_XYZ()->X) / tx;
+   ay = (BlackPointOut->Y - cmsD50_XYZ()->Y) / ty;
+   az = (BlackPointOut->Z - cmsD50_XYZ()->Z) / tz;
+
+   bx = - cmsD50_XYZ()-> X * (BlackPointOut->X - BlackPointIn->X) / tx;
+   by = - cmsD50_XYZ()-> Y * (BlackPointOut->Y - BlackPointIn->Y) / ty;
+   bz = - cmsD50_XYZ()-> Z * (BlackPointOut->Z - BlackPointIn->Z) / tz;
+
+   _cmsVEC3init(&m ->v[0], ax, 0,  0);
+   _cmsVEC3init(&m ->v[1], 0, ay,  0);
+   _cmsVEC3init(&m ->v[2], 0,  0,  az);
+   _cmsVEC3init(off, bx, by, bz);
+
+}
+
+
+// Approximate a blackbody illuminant based on CHAD information
+static
+cmsFloat64Number CHAD2Temp(const cmsMAT3* Chad)
+{
+    // Convert D50 across CHAD to get the absolute white point
+     cmsVEC3 d, s;
+     cmsCIEXYZ Dest;
+     cmsCIExyY DestChromaticity;
+     cmsFloat64Number TempK;
+
+    s.n[VX] = cmsD50_XYZ() -> X;
+    s.n[VY] = cmsD50_XYZ() -> Y;
+    s.n[VZ] = cmsD50_XYZ() -> Z;
+
+    _cmsMAT3eval(&d, Chad, &s);
+
+    Dest.X = d.n[VX];
+    Dest.Y = d.n[VY];
+    Dest.Z = d.n[VZ];
+
+    cmsXYZ2xyY(&DestChromaticity, &Dest);
+
+    if (!cmsTempFromWhitePoint(&TempK, &DestChromaticity))
+        return -1.0;
+
+    return TempK;
+}
+
+// Compute a CHAD based on a given temperature
+static
+void Temp2CHAD(cmsMAT3* Chad, cmsFloat64Number Temp)
+{
+    cmsCIEXYZ White;
+    cmsCIExyY ChromaticityOfWhite;
+    
+    cmsWhitePointFromTemp(&ChromaticityOfWhite, Temp);  
+    cmsxyY2XYZ(&White, &ChromaticityOfWhite);
+    _cmsAdaptationMatrix(Chad, NULL, cmsD50_XYZ(), &White);
+
+}
+
+// Join scalings to obtain relative input to absolute and then to relative output.
+// Result is stored in a 3x3 matrix
+static
+cmsBool  ComputeAbsoluteIntent(cmsFloat64Number AdaptationState,
+                               const cmsCIEXYZ* WhitePointIn,    
+                               const cmsMAT3* ChromaticAdaptationMatrixIn,
+                               const cmsCIEXYZ* WhitePointOut,
+                               const cmsMAT3* ChromaticAdaptationMatrixOut,
+                               cmsMAT3* m)
+{
+    cmsMAT3 Scale, m1, m2, m3;
+
+    // Adaptation state
+    if (AdaptationState == 1.0) {
+
+        // Observer is fully adapted. Keep chromatic adaptation. 
+        // That is the standard V4 behaviour
+        _cmsVEC3init(&m->v[0], WhitePointIn->X / WhitePointOut->X, 0, 0);
+        _cmsVEC3init(&m->v[1], 0, WhitePointIn->Y / WhitePointOut->Y, 0);
+        _cmsVEC3init(&m->v[2], 0, 0, WhitePointIn->Z / WhitePointOut->Z);
+
+    }
+    else  {
+
+        // Incomplete adaptation. This is an advanced feature.
+        _cmsVEC3init(&Scale.v[0], WhitePointIn->X / WhitePointOut->X, 0, 0);
+        _cmsVEC3init(&Scale.v[1], 0,  WhitePointIn->Y / WhitePointOut->Y, 0);
+        _cmsVEC3init(&Scale.v[2], 0, 0,  WhitePointIn->Z / WhitePointOut->Z);
+
+        m1 = *ChromaticAdaptationMatrixIn;
+        if (!_cmsMAT3inverse(&m1, &m2)) return FALSE; 
+        _cmsMAT3per(&m3, &m2, &Scale);
+
+        // m3 holds CHAD from input white to D50 times abs. col. scaling
+        if (AdaptationState == 0.0) {
+
+            // Observer is not adapted, undo the chromatic adaptation
+            _cmsMAT3per(m, &m3, ChromaticAdaptationMatrixOut);
+
+        } else {
+
+            cmsMAT3 MixedCHAD;
+            cmsFloat64Number TempSrc, TempDest, Temp;
+
+            TempSrc  = CHAD2Temp(ChromaticAdaptationMatrixIn);  // K for source white
+            TempDest = CHAD2Temp(ChromaticAdaptationMatrixOut); // K for dest white
+
+            if (TempSrc < 0.0 || TempDest < 0.0) return FALSE; // Something went wrong
+
+            if (_cmsMAT3isIdentity(&Scale) && fabs(TempSrc - TempDest) < 0.01) {
+
+                _cmsMAT3identity(m);
+                return TRUE;
+            }
+
+            Temp = AdaptationState * TempSrc + (1.0 - AdaptationState) * TempDest;
+
+            // Get a CHAD from D50 to whatever output temperature. This replaces output CHAD
+            Temp2CHAD(&MixedCHAD, Temp);
+
+            _cmsMAT3per(m, &m3, &MixedCHAD);
+        }
+
+    }
+    return TRUE;
+
+}
+
+// Just to see if m matrix should be applied
+static
+cmsBool IsEmptyLayer(cmsMAT3* m, cmsVEC3* off)
+{
+    cmsFloat64Number diff = 0;
+    cmsMAT3 Ident;
+    int i;
+    
+    if (m == NULL && off == NULL) return TRUE;  // NULL is allowed as an empty layer
+    if (m == NULL && off != NULL) return FALSE; // This is an internal error
+
+    _cmsMAT3identity(&Ident);
+    
+    for (i=0; i < 3*3; i++)
+        diff += fabs(((cmsFloat64Number*)m)[i] - ((cmsFloat64Number*)&Ident)[i]);
+
+    for (i=0; i < 3; i++)
+        diff += fabs(((cmsFloat64Number*)off)[i]);
+
+
+    return (diff < 0.002);
+}
+
+
+// Compute the conversion layer
+static
+cmsBool ComputeConversion(int i, cmsHPROFILE hProfiles[], 
+                                 cmsUInt32Number Intent, 
+                                 cmsBool BPC, 
+                                 cmsFloat64Number AdaptationState, 
+                                 cmsMAT3* m, cmsVEC3* off)
+{
+
+    int k;
+
+    // m  and off are set to identity and this is detected latter on
+    _cmsMAT3identity(m);
+    _cmsVEC3init(off, 0, 0, 0);
+
+    // If intent is abs. colorimetric,
+    if (Intent == INTENT_ABSOLUTE_COLORIMETRIC) {
+
+        cmsCIEXYZ WhitePointIn, WhitePointOut;
+        cmsMAT3 ChromaticAdaptationMatrixIn, ChromaticAdaptationMatrixOut;  
+
+        _cmsReadMediaWhitePoint(&WhitePointIn,  hProfiles[i-1]);
+        _cmsReadCHAD(&ChromaticAdaptationMatrixIn, hProfiles[i-1]);
+
+        _cmsReadMediaWhitePoint(&WhitePointOut,  hProfiles[i]);
+        _cmsReadCHAD(&ChromaticAdaptationMatrixOut, hProfiles[i]);
+
+        if (!ComputeAbsoluteIntent(AdaptationState, 
+                                  &WhitePointIn,  &ChromaticAdaptationMatrixIn, 
+                                  &WhitePointOut, &ChromaticAdaptationMatrixOut, m)) return FALSE;
+
+    }
+    else {
+        // Rest of intents may apply BPC.
+
+        if (BPC) {
+
+            cmsCIEXYZ BlackPointIn, BlackPointOut;
+
+            cmsDetectBlackPoint(&BlackPointIn,  hProfiles[i-1], Intent, 0);
+            cmsDetectBlackPoint(&BlackPointOut, hProfiles[i], Intent, 0);
+
+            // If black points are equal, then do nothing
+            if (BlackPointIn.X != BlackPointOut.X ||
+                BlackPointIn.Y != BlackPointOut.Y ||
+                BlackPointIn.Z != BlackPointOut.Z) 
+                    ComputeBlackPointCompensation(&BlackPointIn, &BlackPointOut, m, off);
+        }
+    }
+
+    // Offset should be adjusted because the encoding. We encode XYZ normalized to 0..1.0,
+    // to do that, we divide by MAX_ENCODEABLE_XZY. The conversion stage goes XYZ -> XYZ so
+    // we have first to convert from encoded to XYZ and then convert back to encoded. 
+    // y = Mx + Off
+    // x = x'c
+    // y = M x'c + Off
+    // y = y'c; y' = y / c
+    // y' = (Mx'c + Off) /c = Mx' + (Off / c) 
+
+    for (k=0; k < 3; k++) {
+        off ->n[k] /= MAX_ENCODEABLE_XYZ;
+    }
+
+    return TRUE;
+}
+
+
+// Add a conversion stage if needed. If a matrix/offset m is given, it applies to XYZ space 
+static
+cmsBool AddConversion(cmsPipeline* Result, cmsColorSpaceSignature InPCS, cmsColorSpaceSignature OutPCS, cmsMAT3* m, cmsVEC3* off)
+{
+    cmsFloat64Number* m_as_dbl = (cmsFloat64Number*) m;
+    cmsFloat64Number* off_as_dbl = (cmsFloat64Number*) off;
+
+    // Handle PCS mismatches. A specialized stage is added to the LUT in such case
+    switch (InPCS) {
+
+        case cmsSigXYZData: // Input profile operates in XYZ
+
+            switch (OutPCS) {
+
+            case cmsSigXYZData:  // XYZ -> XYZ
+                if (!IsEmptyLayer(m, off))
+                    cmsPipelineInsertStage(Result, cmsAT_END, cmsStageAllocMatrix(Result ->ContextID, 3, 3, m_as_dbl, off_as_dbl));
+                break;
+
+            case cmsSigLabData:  // XYZ -> Lab
+                if (!IsEmptyLayer(m, off))
+                    cmsPipelineInsertStage(Result, cmsAT_END, cmsStageAllocMatrix(Result ->ContextID, 3, 3, m_as_dbl, off_as_dbl));
+                cmsPipelineInsertStage(Result, cmsAT_END, _cmsStageAllocXYZ2Lab(Result ->ContextID));
+                break;
+
+            default:
+                return FALSE;   // Colorspace mismatch
+                }
+                break;
+
+
+        case cmsSigLabData: // Input profile operates in Lab
+
+            switch (OutPCS) {
+
+            case cmsSigXYZData:  // Lab -> XYZ
+
+                cmsPipelineInsertStage(Result, cmsAT_END, _cmsStageAllocLab2XYZ(Result ->ContextID));
+                if (!IsEmptyLayer(m, off))
+                    cmsPipelineInsertStage(Result, cmsAT_END, cmsStageAllocMatrix(Result ->ContextID, 3, 3, m_as_dbl, off_as_dbl));
+                break;
+
+            case cmsSigLabData:  // Lab -> Lab
+
+                if (!IsEmptyLayer(m, off)) {            
+                    cmsPipelineInsertStage(Result, cmsAT_END, _cmsStageAllocLab2XYZ(Result ->ContextID));        
+                    cmsPipelineInsertStage(Result, cmsAT_END, cmsStageAllocMatrix(Result ->ContextID, 3, 3, m_as_dbl, off_as_dbl));
+                    cmsPipelineInsertStage(Result, cmsAT_END, _cmsStageAllocXYZ2Lab(Result ->ContextID));
+                }
+                break;
+
+            default:
+                return FALSE;  // Mismatch
+            }
+            break;
+
+
+            // On colorspaces other than PCS, check for same space
+        default:
+            if (InPCS != OutPCS) return FALSE;
+            break;
+    }
+
+    return TRUE;
+}
+
+
+// Is a given space compatible with another?
+static
+cmsBool ColorSpaceIsCompatible(cmsColorSpaceSignature a, cmsColorSpaceSignature b)
+{
+    // If they are same, they are compatible.
+    if (a == b) return TRUE;
+
+    // Check for XYZ/Lab. Those spaces are interchangeable as they can be computed one from other.
+    if ((a == cmsSigXYZData) && (b == cmsSigLabData)) return TRUE;
+    if ((a == cmsSigLabData) && (b == cmsSigXYZData)) return TRUE;
+
+    return FALSE;
+}
+
+
+// Default handler for ICC-style intents
+static
+cmsPipeline* DefaultICCintents(cmsContext       ContextID, 
+                               cmsUInt32Number  nProfiles,
+                               cmsUInt32Number  TheIntents[], 
+                               cmsHPROFILE      hProfiles[], 
+                               cmsBool          BPC[],
+                               cmsFloat64Number AdaptationStates[],
+                               cmsUInt32Number  dwFlags)
+{
+    cmsPipeline* Lut, *Result;
+    cmsHPROFILE hProfile;
+    cmsMAT3 m;
+    cmsVEC3 off;
+    cmsColorSpaceSignature ColorSpaceIn, ColorSpaceOut, CurrentColorSpace; 
+    cmsProfileClassSignature ClassSig;
+    cmsUInt32Number  i, Intent;
+
+	// For safety
+	if (nProfiles == 0) return NULL;
+
+    // Allocate an empty LUT for holding the result. 0 as channel count means 'undefined'
+    Result = cmsPipelineAlloc(ContextID, 0, 0);
+    if (Result == NULL) return NULL;
+
+    CurrentColorSpace = cmsGetColorSpace(hProfiles[0]);    
+
+    for (i=0; i < nProfiles; i++) {
+
+        cmsBool  lIsDeviceLink, lIsInput;
+
+        hProfile      = hProfiles[i];
+        ClassSig      = cmsGetDeviceClass(hProfile);
+        lIsDeviceLink = (ClassSig == cmsSigLinkClass || ClassSig == cmsSigAbstractClass );
+      
+        // First profile is used as input unless devicelink or abstract
+		if ((i == 0) && !lIsDeviceLink) {
+			lIsInput = TRUE;
+		}
+		else {
+		  // Else use profile in the input direction if current space is not PCS
+        lIsInput      = (CurrentColorSpace != cmsSigXYZData) &&
+                        (CurrentColorSpace != cmsSigLabData);
+		}
+
+        Intent        = TheIntents[i];
+
+        if (lIsInput || lIsDeviceLink) {
+
+            ColorSpaceIn    = cmsGetColorSpace(hProfile);
+            ColorSpaceOut   = cmsGetPCS(hProfile);
+        }
+        else {
+
+            ColorSpaceIn    = cmsGetPCS(hProfile);
+            ColorSpaceOut   = cmsGetColorSpace(hProfile);
+        }
+
+        if (!ColorSpaceIsCompatible(ColorSpaceIn, CurrentColorSpace)) {
+
+            cmsSignalError(ContextID, cmsERROR_COLORSPACE_CHECK, "ColorSpace mismatch");
+            goto Error;
+        }
+
+        // If devicelink is found, then no custom intent is allowed and we can 
+        // read the LUT to be applied. Settings don't apply here.       
+        if (lIsDeviceLink) {
+
+            // Get the involved LUT from the profile
+            Lut = _cmsReadDevicelinkLUT(hProfile, Intent);
+            if (Lut == NULL) goto Error;
+
+            // What about abstract profiles?
+             if (ClassSig == cmsSigAbstractClass && i > 0) {
+                if (!ComputeConversion(i, hProfiles, Intent, BPC[i], AdaptationStates[i], &m, &off)) goto Error;
+             }
+             else {
+                _cmsMAT3identity(&m);
+                _cmsVEC3init(&off, 0, 0, 0);
+             }
+        
+
+            if (!AddConversion(Result, CurrentColorSpace, ColorSpaceIn, &m, &off)) goto Error;
+
+        }
+        else {
+
+            if (lIsInput) {
+                // Input direction means non-pcs connection, so proceed like devicelinks
+                Lut = _cmsReadInputLUT(hProfile, Intent);       
+                if (Lut == NULL) goto Error;
+            }
+            else {
+
+                // Output direction means PCS connection. Intent may apply here
+                Lut = _cmsReadOutputLUT(hProfile, Intent); 
+                if (Lut == NULL) goto Error;
+
+
+                if (!ComputeConversion(i, hProfiles, Intent, BPC[i], AdaptationStates[i], &m, &off)) goto Error;
+                if (!AddConversion(Result, CurrentColorSpace, ColorSpaceIn, &m, &off)) goto Error;
+
+            }
+        }
+
+        // Concatenate to the output LUT
+        cmsPipelineCat(Result, Lut);
+        cmsPipelineFree(Lut);                            
+
+        // Update current space
+        CurrentColorSpace = ColorSpaceOut;              
+    }
+
+    return Result;
+
+Error:
+
+    if (Result != NULL) cmsPipelineFree(Result);
+    return NULL;
+
+    cmsUNUSED_PARAMETER(dwFlags);
+}
+
+
+// Wrapper for DLL calling convention
+cmsPipeline*  CMSEXPORT _cmsDefaultICCintents(cmsContext     ContextID, 
+                                              cmsUInt32Number nProfiles,
+                                              cmsUInt32Number TheIntents[], 
+                                              cmsHPROFILE     hProfiles[], 
+                                              cmsBool         BPC[],
+                                              cmsFloat64Number AdaptationStates[],
+                                              cmsUInt32Number dwFlags)
+{
+    return DefaultICCintents(ContextID, nProfiles, TheIntents, hProfiles, BPC, AdaptationStates, dwFlags);
+}
+
+// Black preserving intents ---------------------------------------------------------------------------------------------
+
+// Translate black-preserving intents to ICC ones
+static
+int TranslateNonICCIntents(int Intent)
+{
+    switch (Intent) {
+        case INTENT_PRESERVE_K_ONLY_PERCEPTUAL:
+        case INTENT_PRESERVE_K_PLANE_PERCEPTUAL:
+            return INTENT_PERCEPTUAL; 
+
+        case INTENT_PRESERVE_K_ONLY_RELATIVE_COLORIMETRIC:
+        case INTENT_PRESERVE_K_PLANE_RELATIVE_COLORIMETRIC:
+            return INTENT_RELATIVE_COLORIMETRIC;
+
+        case INTENT_PRESERVE_K_ONLY_SATURATION:
+        case INTENT_PRESERVE_K_PLANE_SATURATION:
+            return INTENT_SATURATION;
+
+        default: return Intent;
+    }
+}
+
+// Sampler for Black-only preserving CMYK->CMYK transforms
+
+typedef struct {
+    cmsPipeline*    cmyk2cmyk;      // The original transform
+    cmsToneCurve*   KTone;          // Black-to-black tone curve
+
+} GrayOnlyParams;
+
+
+// Preserve black only if that is the only ink used
+static
+int BlackPreservingGrayOnlySampler(register const cmsUInt16Number In[], register cmsUInt16Number Out[], register void* Cargo)
+{
+    GrayOnlyParams* bp = (GrayOnlyParams*) Cargo;
+
+    // If going across black only, keep black only
+    if (In[0] == 0 && In[1] == 0 && In[2] == 0) {
+
+        // TAC does not apply because it is black ink!
+        Out[0] = Out[1] = Out[2] = 0;
+        Out[3] = cmsEvalToneCurve16(bp->KTone, In[3]);
+        return TRUE;
+    }
+
+    // Keep normal transform for other colors
+    bp ->cmyk2cmyk ->Eval16Fn(In, Out, bp ->cmyk2cmyk->Data);
+    return TRUE;
+}
+
+// This is the entry for black-preserving K-only intents, which are non-ICC
+static
+cmsPipeline*  BlackPreservingKOnlyIntents(cmsContext     ContextID, 
+                                          cmsUInt32Number nProfiles,
+                                          cmsUInt32Number TheIntents[], 
+                                          cmsHPROFILE     hProfiles[], 
+                                          cmsBool         BPC[],
+                                          cmsFloat64Number AdaptationStates[],
+                                          cmsUInt32Number dwFlags)
+{
+    GrayOnlyParams  bp;
+    cmsPipeline*    Result;
+    cmsUInt32Number ICCIntents[256];
+    cmsStage*         CLUT;
+    cmsUInt32Number i, nGridPoints;
+
+
+    // Sanity check
+    if (nProfiles < 1 || nProfiles > 255) return NULL;
+
+    // Translate black-preserving intents to ICC ones
+    for (i=0; i < nProfiles; i++) 
+        ICCIntents[i] = TranslateNonICCIntents(TheIntents[i]); 
+
+    // Check for non-cmyk profiles
+    if (cmsGetColorSpace(hProfiles[0]) != cmsSigCmykData ||
+        cmsGetColorSpace(hProfiles[nProfiles-1]) != cmsSigCmykData) 
+           return DefaultICCintents(ContextID, nProfiles, ICCIntents, hProfiles, BPC, AdaptationStates, dwFlags);
+
+    memset(&bp, 0, sizeof(bp));
+
+    // Allocate an empty LUT for holding the result
+    Result = cmsPipelineAlloc(ContextID, 4, 4);
+    if (Result == NULL) return NULL;
+
+    // Create a LUT holding normal ICC transform
+    bp.cmyk2cmyk = DefaultICCintents(ContextID, 
+        nProfiles,
+        ICCIntents, 
+        hProfiles, 
+        BPC,
+        AdaptationStates,
+        dwFlags);
+
+    if (bp.cmyk2cmyk == NULL) goto Error;
+    
+    // Now, compute the tone curve
+    bp.KTone = _cmsBuildKToneCurve(ContextID, 
+        4096, 
+        nProfiles,
+        ICCIntents, 
+        hProfiles, 
+        BPC,
+        AdaptationStates,
+        dwFlags);
+    
+    if (bp.KTone == NULL) goto Error;
+
+    
+    // How many gridpoints are we going to use?
+    nGridPoints = _cmsReasonableGridpointsByColorspace(cmsSigCmykData, dwFlags);
+    
+    // Create the CLUT. 16 bits
+    CLUT = cmsStageAllocCLut16bit(ContextID, nGridPoints, 4, 4, NULL);
+    if (CLUT == NULL) goto Error;
+
+    // This is the one and only MPE in this LUT
+    cmsPipelineInsertStage(Result, cmsAT_BEGIN, CLUT);
+
+    // Sample it. We cannot afford pre/post linearization this time.
+    if (!cmsStageSampleCLut16bit(CLUT, BlackPreservingGrayOnlySampler, (void*) &bp, 0)) 
+        goto Error;
+    
+    // Get rid of xform and tone curve
+    cmsPipelineFree(bp.cmyk2cmyk);
+    cmsFreeToneCurve(bp.KTone);
+
+    return Result;
+
+Error:
+
+    if (bp.cmyk2cmyk != NULL) cmsPipelineFree(bp.cmyk2cmyk);
+    if (bp.KTone != NULL)  cmsFreeToneCurve(bp.KTone);
+    if (Result != NULL) cmsPipelineFree(Result);
+    return NULL;
+
+}
+
+// K Plane-preserving CMYK to CMYK ------------------------------------------------------------------------------------
+
+typedef struct {
+
+    cmsPipeline*     cmyk2cmyk;     // The original transform
+    cmsHTRANSFORM    hProofOutput;  // Output CMYK to Lab (last profile)
+    cmsHTRANSFORM    cmyk2Lab;      // The input chain
+    cmsToneCurve*    KTone;         // Black-to-black tone curve
+    cmsPipeline*     LabK2cmyk;     // The output profile
+    cmsFloat64Number MaxError;
+
+    cmsHTRANSFORM    hRoundTrip;               
+    cmsFloat64Number MaxTAC;
+    
+
+} PreserveKPlaneParams;
+
+
+// The CLUT will be stored at 16 bits, but calculations are performed at cmsFloat32Number precision
+static
+int BlackPreservingSampler(register const cmsUInt16Number In[], register cmsUInt16Number Out[], register void* Cargo)
+{
+    int i;
+    cmsFloat32Number Inf[4], Outf[4];
+    cmsFloat32Number LabK[4];   
+    cmsFloat64Number SumCMY, SumCMYK, Error, Ratio;
+    cmsCIELab ColorimetricLab, BlackPreservingLab;
+    PreserveKPlaneParams* bp = (PreserveKPlaneParams*) Cargo;
+    
+    // Convert from 16 bits to floating point
+    for (i=0; i < 4; i++) 
+        Inf[i] = (cmsFloat32Number) (In[i] / 65535.0);
+
+    // Get the K across Tone curve
+    LabK[3] = cmsEvalToneCurveFloat(bp ->KTone, Inf[3]);
+    
+    // If going across black only, keep black only
+    if (In[0] == 0 && In[1] == 0 && In[2] == 0) {
+
+        Out[0] = Out[1] = Out[2] = 0;
+        Out[3] = _cmsQuickSaturateWord(LabK[3] * 65535.0);
+        return TRUE;
+    }
+    
+    // Try the original transform, 
+    cmsPipelineEvalFloat( Inf, Outf, bp ->cmyk2cmyk);  
+    
+    // Store a copy of the floating point result into 16-bit
+    for (i=0; i < 4; i++) 
+            Out[i] = _cmsQuickSaturateWord(Outf[i] * 65535.0);
+
+    // Maybe K is already ok (mostly on K=0)
+    if ( fabs(Outf[3] - LabK[3]) < (3.0 / 65535.0) ) {
+        return TRUE;
+    }
+    
+    // K differ, mesure and keep Lab measurement for further usage
+    // this is done in relative colorimetric intent
+    cmsDoTransform(bp->hProofOutput, Out, &ColorimetricLab, 1);
+    
+    // Is not black only and the transform doesn't keep black.
+    // Obtain the Lab of output CMYK. After that we have Lab + K
+    cmsDoTransform(bp ->cmyk2Lab, Outf, LabK, 1);
+
+    // Obtain the corresponding CMY using reverse interpolation 
+    // (K is fixed in LabK[3])
+    if (!cmsPipelineEvalReverseFloat(LabK, Outf, Outf, bp ->LabK2cmyk)) {
+
+        // Cannot find a suitable value, so use colorimetric xform
+        // which is already stored in Out[]
+        return TRUE;
+    }
+
+    // Make sure to pass thru K (which now is fixed)
+    Outf[3] = LabK[3];
+       
+    // Apply TAC if needed    
+    SumCMY   = Outf[0]  + Outf[1] + Outf[2];
+    SumCMYK  = SumCMY + Outf[3];      
+
+    if (SumCMYK > bp ->MaxTAC) {
+
+        Ratio = 1 - ((SumCMYK - bp->MaxTAC) / SumCMY);
+        if (Ratio < 0)
+            Ratio = 0;
+    }
+    else
+       Ratio = 1.0;
+
+    Out[0] = _cmsQuickSaturateWord(Outf[0] * Ratio * 65535.0);     // C
+    Out[1] = _cmsQuickSaturateWord(Outf[1] * Ratio * 65535.0);     // M
+    Out[2] = _cmsQuickSaturateWord(Outf[2] * Ratio * 65535.0);     // Y
+    Out[3] = _cmsQuickSaturateWord(Outf[3] * 65535.0);
+
+    // Estimate the error (this goes 16 bits to Lab DBL)
+    cmsDoTransform(bp->hProofOutput, Out, &BlackPreservingLab, 1);  
+    Error = cmsDeltaE(&ColorimetricLab, &BlackPreservingLab);                   
+    if (Error > bp -> MaxError) 
+        bp->MaxError = Error;
+
+    return TRUE;
+}
+
+// This is the entry for black-plane preserving, which are non-ICC
+static
+cmsPipeline* BlackPreservingKPlaneIntents(cmsContext     ContextID, 
+                                          cmsUInt32Number nProfiles,
+                                          cmsUInt32Number TheIntents[], 
+                                          cmsHPROFILE     hProfiles[], 
+                                          cmsBool         BPC[],
+                                          cmsFloat64Number AdaptationStates[],
+                                          cmsUInt32Number dwFlags)
+{
+    PreserveKPlaneParams bp;
+    cmsPipeline*    Result = NULL;
+    cmsUInt32Number ICCIntents[256];
+    cmsStage*         CLUT;
+    cmsUInt32Number i, nGridPoints;    
+    cmsHPROFILE hLab;
+
+    // Sanity check
+    if (nProfiles < 1 || nProfiles > 255) return NULL;
+
+    // Translate black-preserving intents to ICC ones
+    for (i=0; i < nProfiles; i++) 
+        ICCIntents[i] = TranslateNonICCIntents(TheIntents[i]); 
+
+    // Check for non-cmyk profiles
+    if (cmsGetColorSpace(hProfiles[0]) != cmsSigCmykData ||
+        cmsGetColorSpace(hProfiles[nProfiles-1]) != cmsSigCmykData) 
+           return  DefaultICCintents(ContextID, nProfiles, ICCIntents, hProfiles, BPC, AdaptationStates, dwFlags);
+
+    // Allocate an empty LUT for holding the result
+    Result = cmsPipelineAlloc(ContextID, 4, 4);
+    if (Result == NULL) return NULL;
+
+   
+    memset(&bp, 0, sizeof(bp));
+
+    // We need the input LUT of the last profile, assuming this one is responsible of
+    // black generation. This LUT will be seached in inverse order.
+    bp.LabK2cmyk = _cmsReadInputLUT(hProfiles[nProfiles-1], INTENT_RELATIVE_COLORIMETRIC);
+    if (bp.LabK2cmyk == NULL) goto Cleanup;
+
+    // Get total area coverage (in 0..1 domain)
+    bp.MaxTAC = cmsDetectTAC(hProfiles[nProfiles-1]) / 100.0;
+
+    // Create a LUT holding normal ICC transform
+    bp.cmyk2cmyk = DefaultICCintents(ContextID,
+                                         nProfiles,
+                                         ICCIntents, 
+                                         hProfiles, 
+                                         BPC,
+                                         AdaptationStates,
+                                         dwFlags);
+
+    // Now the tone curve
+    bp.KTone = _cmsBuildKToneCurve(ContextID, 4096, nProfiles,
+                                   ICCIntents, 
+                                   hProfiles, 
+                                   BPC, 
+                                   AdaptationStates,
+                                   dwFlags);
+
+
+    // To measure the output, Last profile to Lab
+    hLab = cmsCreateLab4ProfileTHR(ContextID, NULL);
+    bp.hProofOutput = cmsCreateTransformTHR(ContextID, hProfiles[nProfiles-1], 
+                                         CHANNELS_SH(4)|BYTES_SH(2), hLab, TYPE_Lab_DBL, 
+                                         INTENT_RELATIVE_COLORIMETRIC, 
+                                         cmsFLAGS_NOCACHE|cmsFLAGS_NOOPTIMIZE);
+
+    // Same as anterior, but lab in the 0..1 range
+    bp.cmyk2Lab = cmsCreateTransformTHR(ContextID, hProfiles[nProfiles-1], 
+                                         FLOAT_SH(1)|CHANNELS_SH(4)|BYTES_SH(4), hLab, 
+										 FLOAT_SH(1)|CHANNELS_SH(3)|BYTES_SH(4), 
+                                         INTENT_RELATIVE_COLORIMETRIC, 
+                                         cmsFLAGS_NOCACHE|cmsFLAGS_NOOPTIMIZE);
+    cmsCloseProfile(hLab);
+
+    // Error estimation (for debug only)
+    bp.MaxError = 0;
+
+    // How many gridpoints are we going to use?
+    nGridPoints = _cmsReasonableGridpointsByColorspace(cmsSigCmykData, dwFlags);
+
+    
+    CLUT = cmsStageAllocCLut16bit(ContextID, nGridPoints, 4, 4, NULL);
+    if (CLUT == NULL) goto Cleanup;
+
+    cmsPipelineInsertStage(Result, cmsAT_BEGIN, CLUT);
+
+    cmsStageSampleCLut16bit(CLUT, BlackPreservingSampler, (void*) &bp, 0);
+
+Cleanup:
+
+    if (bp.cmyk2cmyk) cmsPipelineFree(bp.cmyk2cmyk);
+    if (bp.cmyk2Lab) cmsDeleteTransform(bp.cmyk2Lab);       
+    if (bp.hProofOutput) cmsDeleteTransform(bp.hProofOutput);
+    
+    if (bp.KTone) cmsFreeToneCurve(bp.KTone);   
+    if (bp.LabK2cmyk) cmsPipelineFree(bp.LabK2cmyk);
+
+    return Result;
+}
+
+// Link routines ------------------------------------------------------------------------------------------------------
+
+// Chain several profiles into a single LUT. It just checks the parameters and then calls the handler
+// for the first intent in chain. The handler may be user-defined. Is up to the handler to deal with the 
+// rest of intents in chain. A maximum of 255 profiles at time are supported, which is pretty reasonable.
+cmsPipeline* _cmsLinkProfiles(cmsContext     ContextID, 
+                              cmsUInt32Number nProfiles,
+                              cmsUInt32Number TheIntents[], 
+                              cmsHPROFILE     hProfiles[], 
+                              cmsBool         BPC[],
+                              cmsFloat64Number AdaptationStates[],
+                              cmsUInt32Number dwFlags)
+{
+    cmsUInt32Number i;
+    cmsIntentsList* Intent;
+
+    // Make sure a reasonable number of profiles is provided
+    if (nProfiles <= 0 || nProfiles > 255) {
+         cmsSignalError(ContextID, cmsERROR_RANGE, "Couldn't link '%d' profiles", nProfiles);
+        return NULL;
+    }
+
+    for (i=0; i < nProfiles; i++) {
+
+        // Check if black point is really needed or allowed. Note that 
+        // following Adobe's document:
+        // BPC does not apply to devicelink profiles, nor to abs colorimetric, 
+        // and applies always on V4 perceptual and saturation.
+
+        if (TheIntents[i] == INTENT_ABSOLUTE_COLORIMETRIC)
+            BPC[i] = FALSE;
+
+        if (TheIntents[i] == INTENT_PERCEPTUAL || TheIntents[i] == INTENT_SATURATION) {
+
+            // Force BPC for V4 profiles in perceptual and saturation
+            if (cmsGetProfileVersion(hProfiles[i]) >= 4.0)
+                BPC[i] = TRUE;
+        }      
+    }
+
+    // Search for a handler. The first intent in the chain defines the handler. That would
+    // prevent using multiple custom intents in a multiintent chain, but the behaviour of
+    // this case would present some issues if the custom intent tries to do things like
+    // preserve primaries. This solution is not perfect, but works well on most cases.
+
+    Intent = SearchIntent(TheIntents[0]);
+	if (Intent == NULL) {
+		cmsSignalError(ContextID, cmsERROR_UNKNOWN_EXTENSION, "Unsupported intent '%d'", TheIntents[0]);
+		return NULL;
+	}
+
+    // Call the handler
+    return Intent ->Link(ContextID, nProfiles, TheIntents, hProfiles, BPC, AdaptationStates, dwFlags);
+}
+
+// -------------------------------------------------------------------------------------------------
+
+// Get information about available intents. nMax is the maximum space for the supplied "Codes" 
+// and "Descriptions" the function returns the total number of intents, which may be greater 
+// than nMax, although the matrices are not populated beyond this level.
+cmsUInt32Number CMSEXPORT cmsGetSupportedIntents(cmsUInt32Number nMax, cmsUInt32Number* Codes, char** Descriptions)
+{
+    cmsIntentsList* pt;
+    cmsUInt32Number nIntents;
+
+    for (nIntents=0, pt = Intents; pt != NULL; pt = pt -> Next)
+    {
+        if (nIntents < nMax) {
+            if (Codes != NULL) 
+                Codes[nIntents] = pt ->Intent;
+
+            if (Descriptions != NULL) 
+                Descriptions[nIntents] = pt ->Description;
+        }
+
+        nIntents++;
+    }
+
+    return nIntents;
+}
+
+// The plug-in registration. User can add new intents or override default routines
+cmsBool  _cmsRegisterRenderingIntentPlugin(cmsPluginBase* Data)
+{
+    cmsPluginRenderingIntent* Plugin = (cmsPluginRenderingIntent*) Data;
+    cmsIntentsList* fl;
+
+    // Do we have to reset the intents?
+    if (Data == NULL) {
+    
+       Intents = DefaultIntents;
+       return TRUE;
+    }
+
+    fl = SearchIntent(Plugin ->Intent);
+
+    if (fl == NULL) {
+        fl = (cmsIntentsList*) _cmsPluginMalloc(sizeof(cmsIntentsList));
+        if (fl == NULL) return FALSE;
+    }
+
+    fl ->Intent  = Plugin ->Intent;
+    strncpy(fl ->Description, Plugin ->Description, 255);
+    fl ->Description[255] = 0;
+
+    fl ->Link    = Plugin ->Link;
+
+    fl ->Next = Intents;
+    Intents = fl;
+
+    return TRUE;
+}
+
diff --git a/thirdparty/liblcms2/src/cmserr.c b/thirdparty/liblcms2/src/cmserr.c
new file mode 100644
index 00000000..55f03580
--- /dev/null
+++ b/thirdparty/liblcms2/src/cmserr.c
@@ -0,0 +1,428 @@
+//---------------------------------------------------------------------------------
+//
+//  Little Color Management System
+//  Copyright (c) 1998-2010 Marti Maria Saguer
+//
+// Permission is hereby granted, free of charge, to any person obtaining 
+// a copy of this software and associated documentation files (the "Software"), 
+// to deal in the Software without restriction, including without limitation 
+// the rights to use, copy, modify, merge, publish, distribute, sublicense, 
+// and/or sell copies of the Software, and to permit persons to whom the Software 
+// is furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in 
+// all copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 
+// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO 
+// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE 
+// LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 
+// OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 
+// WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+//
+//---------------------------------------------------------------------------------
+
+#include "lcms2_internal.h"
+
+// I am so tired about incompatibilities on those functions that here are some replacements
+// that hopefully would be fully portable.
+
+// compare two strings ignoring case
+int CMSEXPORT cmsstrcasecmp(const char* s1, const char* s2)
+{
+         register const unsigned char *us1 = (const unsigned char *)s1,
+                                      *us2 = (const unsigned char *)s2;
+
+        while (toupper(*us1) == toupper(*us2++))
+                if (*us1++ == '\0')
+                        return (0);
+        return (toupper(*us1) - toupper(*--us2));
+}
+
+// long int because C99 specifies ftell in such way (7.19.9.2)
+long int CMSEXPORT cmsfilelength(FILE* f)
+{
+    long int p , n; 
+
+    p = ftell(f); // register current file position 
+
+	if (fseek(f, 0, SEEK_END) != 0) {		
+		return -1;
+	}
+
+    n = ftell(f);
+    fseek(f, p, SEEK_SET); // file position restored 
+
+    return n;    
+}
+
+
+// Memory handling ------------------------------------------------------------------
+//
+// This is the interface to low-level memory management routines. By default a simple
+// wrapping to malloc/free/realloc is provided, although there is a limit on the max
+// amount of memoy that can be reclaimed. This is mostly as a safety feature to 
+// prevent bogus or malintentionated code to allocate huge blocks that otherwise lcms
+// would never need.
+
+#define MAX_MEMORY_FOR_ALLOC  ((cmsUInt32Number)(1024U*1024U*512U))
+
+// User may override this behaviour by using a memory plug-in, which basically replaces
+// the default memory management functions. In this case, no check is performed and it 
+// is up to the plug-in writter to keep in the safe side. There are only three functions 
+// required to be implemented: malloc, realloc and free, although the user may want to 
+// replace the optional mallocZero, calloc and dup as well.
+
+cmsBool   _cmsRegisterMemHandlerPlugin(cmsPluginBase* Plugin);
+
+// *********************************************************************************
+
+// This is the default memory allocation function. It does a very coarse 
+// check of amout of memory, just to prevent exploits
+static
+void* _cmsMallocDefaultFn(cmsContext ContextID, cmsUInt32Number size)
+{
+    if (size > MAX_MEMORY_FOR_ALLOC) return NULL;  // Never allow over maximum
+
+    return (void*) malloc(size);
+
+    cmsUNUSED_PARAMETER(ContextID);
+}
+
+// Generic allocate & zero
+static
+void* _cmsMallocZeroDefaultFn(cmsContext ContextID, cmsUInt32Number size)
+{
+    void *pt = _cmsMalloc(ContextID, size);
+    if (pt == NULL) return NULL;
+
+    memset(pt, 0, size);
+    return pt;
+}
+
+
+// The default free function. The only check proformed is against NULL pointers
+static
+void _cmsFreeDefaultFn(cmsContext ContextID, void *Ptr)
+{
+    // free(NULL) is defined a no-op by C99, therefore it is safe to
+    // avoid the check, but it is here just in case...
+
+    if (Ptr) free(Ptr); 
+
+    cmsUNUSED_PARAMETER(ContextID);
+}
+
+// The default realloc function. Again it check for exploits. If Ptr is NULL, 
+// realloc behaves the same way as malloc and allocates a new block of size bytes. 
+static
+void* _cmsReallocDefaultFn(cmsContext ContextID, void* Ptr, cmsUInt32Number size)
+{
+
+    if (size > MAX_MEMORY_FOR_ALLOC) return NULL;  // Never realloc over 512Mb
+
+    return realloc(Ptr, size);
+
+    cmsUNUSED_PARAMETER(ContextID);
+}
+
+
+// The default calloc function. Allocates an array of num elements, each one of size bytes
+// all memory is initialized to zero.
+static
+void* _cmsCallocDefaultFn(cmsContext ContextID, cmsUInt32Number num, cmsUInt32Number size)
+{
+    cmsUInt32Number Total = num * size;
+
+    // Preserve calloc behaviour
+    if (Total == 0) return NULL;
+
+    // Safe check for overflow.
+    if (num >= UINT_MAX / size) return NULL; 
+
+    // Check for overflow
+    if (Total < num || Total < size) {
+        return NULL;
+    }
+    
+    if (Total > MAX_MEMORY_FOR_ALLOC) return NULL;  // Never alloc over 512Mb
+
+    return _cmsMallocZero(ContextID, Total);
+}
+
+// Generic block duplication
+static
+void* _cmsDupDefaultFn(cmsContext ContextID, const void* Org, cmsUInt32Number size)
+{
+    void* mem;
+    
+    if (size > MAX_MEMORY_FOR_ALLOC) return NULL;  // Never dup over 512Mb
+
+    mem = _cmsMalloc(ContextID, size);
+
+    if (mem != NULL && Org != NULL)
+        memmove(mem, Org, size);
+
+    return mem;
+}
+
+// Pointers to malloc and _cmsFree functions in current environment
+static void * (* MallocPtr)(cmsContext ContextID, cmsUInt32Number size)                     = _cmsMallocDefaultFn;
+static void * (* MallocZeroPtr)(cmsContext ContextID, cmsUInt32Number size)                 = _cmsMallocZeroDefaultFn;
+static void   (* FreePtr)(cmsContext ContextID, void *Ptr)                                  = _cmsFreeDefaultFn;
+static void * (* ReallocPtr)(cmsContext ContextID, void *Ptr, cmsUInt32Number NewSize)      = _cmsReallocDefaultFn; 
+static void * (* CallocPtr)(cmsContext ContextID, cmsUInt32Number num, cmsUInt32Number size)= _cmsCallocDefaultFn;
+static void * (* DupPtr)(cmsContext ContextID, const void* Org, cmsUInt32Number size)       = _cmsDupDefaultFn;
+
+// Plug-in replacement entry
+cmsBool  _cmsRegisterMemHandlerPlugin(cmsPluginBase *Data)
+{
+    cmsPluginMemHandler* Plugin = (cmsPluginMemHandler*) Data;
+
+    // NULL forces to reset to defaults
+    if (Data == NULL) {
+
+        MallocPtr    = _cmsMallocDefaultFn;
+        MallocZeroPtr= _cmsMallocZeroDefaultFn;
+        FreePtr      = _cmsFreeDefaultFn;
+        ReallocPtr   = _cmsReallocDefaultFn; 
+        CallocPtr    = _cmsCallocDefaultFn;
+        DupPtr       = _cmsDupDefaultFn;
+        return TRUE;
+    }
+
+	// Check for required callbacks
+	if (Plugin -> MallocPtr == NULL ||
+        Plugin -> FreePtr == NULL ||
+		Plugin -> ReallocPtr == NULL) return FALSE;
+
+    // Set replacement functions
+    MallocPtr  = Plugin -> MallocPtr;
+    FreePtr    = Plugin -> FreePtr;
+    ReallocPtr = Plugin -> ReallocPtr;
+
+    if (Plugin ->MallocZeroPtr != NULL) MallocZeroPtr = Plugin ->MallocZeroPtr;
+    if (Plugin ->CallocPtr != NULL)     CallocPtr     = Plugin -> CallocPtr;
+    if (Plugin ->DupPtr != NULL)        DupPtr        = Plugin -> DupPtr;
+
+    return TRUE;
+}
+
+// Generic allocate
+void* CMSEXPORT _cmsMalloc(cmsContext ContextID, cmsUInt32Number size)
+{
+    return MallocPtr(ContextID, size);
+}
+
+// Generic allocate & zero
+void* CMSEXPORT _cmsMallocZero(cmsContext ContextID, cmsUInt32Number size)
+{
+    return MallocZeroPtr(ContextID, size);
+}
+
+// Generic calloc
+void* CMSEXPORT _cmsCalloc(cmsContext ContextID, cmsUInt32Number num, cmsUInt32Number size)
+{
+    return CallocPtr(ContextID, num, size);
+}
+
+// Generic reallocate
+void* CMSEXPORT _cmsRealloc(cmsContext ContextID, void* Ptr, cmsUInt32Number size)
+{
+    return ReallocPtr(ContextID, Ptr, size);
+}
+
+// Generic free memory
+void CMSEXPORT _cmsFree(cmsContext ContextID, void* Ptr)
+{
+    if (Ptr != NULL) FreePtr(ContextID, Ptr);
+}
+
+// Generic block duplication
+void* CMSEXPORT _cmsDupMem(cmsContext ContextID, const void* Org, cmsUInt32Number size)
+{
+    return DupPtr(ContextID, Org, size);
+}
+
+// ********************************************************************************************
+
+// Sub allocation takes care of many pointers of small size. The memory allocated in
+// this way have be freed at once. Next function allocates a single chunk for linked list
+// I prefer this method over realloc due to the big inpact on xput realloc may have if 
+// memory is being swapped to disk. This approach is safer (although thats not true on any platform)
+static
+_cmsSubAllocator_chunk* _cmsCreateSubAllocChunk(cmsContext ContextID, cmsUInt32Number Initial)
+{
+    _cmsSubAllocator_chunk* chunk;
+
+    // Create the container
+    chunk = (_cmsSubAllocator_chunk*) _cmsMallocZero(ContextID, sizeof(_cmsSubAllocator_chunk));
+    if (chunk == NULL) return NULL;
+
+    // Initialize values
+    chunk ->Block     = (cmsUInt8Number*) _cmsMalloc(ContextID, Initial);
+    if (chunk ->Block == NULL) {
+
+        // Something went wrong
+        _cmsFree(ContextID, chunk);
+        return NULL;
+    }
+
+    // 20K by default
+    if (Initial == 0)
+        Initial = 20*1024;
+
+    chunk ->BlockSize = Initial;
+    chunk ->Used      = 0;
+    chunk ->next      = NULL;
+
+    return chunk;
+}
+
+// The suballocated is nothing but a pointer to the first element in the list. We also keep
+// the thread ID in this structure.
+_cmsSubAllocator* _cmsCreateSubAlloc(cmsContext ContextID, cmsUInt32Number Initial)
+{
+    _cmsSubAllocator* sub;
+
+    // Create the container
+    sub = (_cmsSubAllocator*) _cmsMallocZero(ContextID, sizeof(_cmsSubAllocator));
+    if (sub == NULL) return NULL;
+
+    sub ->ContextID = ContextID;
+
+    sub ->h = _cmsCreateSubAllocChunk(ContextID, Initial);
+    if (sub ->h == NULL) {
+        _cmsFree(ContextID, sub);
+        return NULL;
+    }
+
+    return sub;
+}
+
+
+// Get rid of whole linked list
+void _cmsSubAllocDestroy(_cmsSubAllocator* sub)
+{
+    _cmsSubAllocator_chunk *chunk, *n;
+
+    for (chunk = sub ->h; chunk != NULL; chunk = n) {
+
+        n = chunk->next;
+        if (chunk->Block != NULL) _cmsFree(sub ->ContextID, chunk->Block);
+        _cmsFree(sub ->ContextID, chunk);
+    }
+
+    // Free the header
+    _cmsFree(sub ->ContextID, sub);
+}
+
+
+// Get a pointer to small memory block.
+void*  _cmsSubAlloc(_cmsSubAllocator* sub, cmsUInt32Number size)
+{
+    cmsUInt32Number Free = sub -> h ->BlockSize - sub -> h -> Used;
+    cmsUInt8Number* ptr;
+
+    size = _cmsALIGNLONG(size);
+
+    // Check for memory. If there is no room, allocate a new chunk of double memory size.   
+    if (size > Free) {
+
+        _cmsSubAllocator_chunk* chunk;
+        cmsUInt32Number newSize;
+
+        newSize = sub -> h ->BlockSize * 2;
+        if (newSize < size) newSize = size;
+
+        chunk = _cmsCreateSubAllocChunk(sub -> ContextID, newSize);
+        if (chunk == NULL) return NULL;
+
+        // Link list
+        chunk ->next = sub ->h;
+        sub ->h    = chunk;
+
+    }
+            
+    ptr =  sub -> h ->Block + sub -> h ->Used;
+    sub -> h -> Used += size;
+
+    return (void*) ptr;
+}
+
+// Error logging ******************************************************************
+
+// There is no error handling at all. When a funtion fails, it returns proper value.
+// For example, all create functions does return NULL on failure. Other return FALSE
+// It may be interesting, for the developer, to know why the function is failing.
+// for that reason, lcms2 does offer a logging function. This function does recive
+// a ENGLISH string with some clues on what is going wrong. You can show this 
+// info to the end user, or just create some sort of log.
+// The logging function should NOT terminate the program, as this obviously can leave
+// resources. It is the programmer's responsability to check each function return code
+// to make sure it didn't fail.
+
+// Error messages are limited to MAX_ERROR_MESSAGE_LEN
+
+#define MAX_ERROR_MESSAGE_LEN   1024
+
+// ---------------------------------------------------------------------------------------------------------
+
+// This is our default log error
+static void DefaultLogErrorHandlerFunction(cmsContext ContextID, cmsUInt32Number ErrorCode, const char *Text);
+
+// The current handler in actual environment
+static cmsLogErrorHandlerFunction LogErrorHandler   = DefaultLogErrorHandlerFunction;
+
+// The default error logger does nothing.
+static
+void DefaultLogErrorHandlerFunction(cmsContext ContextID, cmsUInt32Number ErrorCode, const char *Text)
+{
+    // fprintf(stderr, "[lcms]: %s\n", Text);
+    // fflush(stderr);
+   
+     cmsUNUSED_PARAMETER(ContextID);
+     cmsUNUSED_PARAMETER(ErrorCode);
+	 cmsUNUSED_PARAMETER(Text);
+}
+
+// Change log error
+void CMSEXPORT cmsSetLogErrorHandler(cmsLogErrorHandlerFunction Fn)
+{
+    if (Fn == NULL) 
+        LogErrorHandler = DefaultLogErrorHandlerFunction;
+    else
+        LogErrorHandler = Fn;
+}
+
+// Log an error 
+// ErrorText is a text holding an english description of error.
+void CMSEXPORT cmsSignalError(cmsContext ContextID, cmsUInt32Number ErrorCode, const char *ErrorText, ...)
+{
+    va_list args;
+    char Buffer[MAX_ERROR_MESSAGE_LEN];
+
+    va_start(args, ErrorText);
+    vsnprintf(Buffer, MAX_ERROR_MESSAGE_LEN-1, ErrorText, args);
+    va_end(args);   
+
+    // Call handler
+    LogErrorHandler(ContextID, ErrorCode, Buffer);
+}
+
+// Utility function to print signatures
+void _cmsTagSignature2String(char String[5], cmsTagSignature sig)
+{
+    cmsUInt32Number be;
+
+    // Convert to big endian
+    be = _cmsAdjustEndianess32((cmsUInt32Number) sig);
+
+    // Move chars 
+    memmove(String, &be, 4);
+
+    // Make sure of terminator
+    String[4] = 0;
+}
+
diff --git a/thirdparty/liblcms2/src/cmsgamma.c b/thirdparty/liblcms2/src/cmsgamma.c
new file mode 100644
index 00000000..db156c75
--- /dev/null
+++ b/thirdparty/liblcms2/src/cmsgamma.c
@@ -0,0 +1,1138 @@
+//---------------------------------------------------------------------------------
+//
+//  Little Color Management System
+//  Copyright (c) 1998-2010 Marti Maria Saguer
+//
+// Permission is hereby granted, free of charge, to any person obtaining 
+// a copy of this software and associated documentation files (the "Software"), 
+// to deal in the Software without restriction, including without limitation 
+// the rights to use, copy, modify, merge, publish, distribute, sublicense, 
+// and/or sell copies of the Software, and to permit persons to whom the Software 
+// is furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in 
+// all copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 
+// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO 
+// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE 
+// LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 
+// OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 
+// WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+//
+//---------------------------------------------------------------------------------
+//
+#include "lcms2_internal.h"
+
+// Tone curves are powerful constructs that can contain curves specified in diverse ways. 
+// The curve is stored in segments, where each segment can be sampled or specified by parameters.
+// a 16.bit simplification of the *whole* curve is kept for optimization purposes. For float operation, 
+// each segment is evaluated separately. Plug-ins may be used to define new parametric schemes, 
+// each plug-in may define up to MAX_TYPES_IN_LCMS_PLUGIN functions types. For defining a function, 
+// the plug-in should provide the type id, how many parameters each type has, and a pointer to
+// a procedure that evaluates the function. In the case of reverse evaluation, the evaluator will 
+// be called with the type id as a negative value, and a sampled version of the reversed curve 
+// will be built.
+
+// ----------------------------------------------------------------- Implementation
+// Maxim number of nodes 
+#define MAX_NODES_IN_CURVE   4097
+#define MINUS_INF            (-1E22F)
+#define PLUS_INF             (+1E22F)
+
+// The list of supported parametric curves
+typedef struct _cmsParametricCurvesCollection_st {
+
+    int nFunctions;                                     // Number of supported functions in this chunk
+    int FunctionTypes[MAX_TYPES_IN_LCMS_PLUGIN];        // The identification types
+    int ParameterCount[MAX_TYPES_IN_LCMS_PLUGIN];       // Number of parameters for each function
+    cmsParametricCurveEvaluator    Evaluator;           // The evaluator
+
+    struct _cmsParametricCurvesCollection_st* Next; // Next in list
+
+} _cmsParametricCurvesCollection;
+
+
+// This is the default (built-in) evaluator 
+static cmsFloat64Number DefaultEvalParametricFn(cmsInt32Number Type, const cmsFloat64Number Params[], cmsFloat64Number R);
+
+// The built-in list
+static _cmsParametricCurvesCollection DefaultCurves = { 
+    9,                                  // # of curve types
+    { 1, 2, 3, 4, 5, 6, 7, 8, 108 },    // Parametric curve ID
+    { 1, 3, 4, 5, 7, 4, 5, 5, 1 },      // Parameters by type
+    DefaultEvalParametricFn,            // Evaluator
+    NULL                                // Next in chain
+};
+
+// The linked list head
+static _cmsParametricCurvesCollection* ParametricCurves = &DefaultCurves;
+
+// As a way to install new parametric curves
+cmsBool _cmsRegisterParametricCurvesPlugin(cmsPluginBase* Data)
+{
+    cmsPluginParametricCurves* Plugin = (cmsPluginParametricCurves*) Data;
+    _cmsParametricCurvesCollection* fl;
+    
+    if (Data == NULL) {
+    
+          ParametricCurves =  &DefaultCurves;
+          return TRUE;
+    }
+
+    fl = (_cmsParametricCurvesCollection*) _cmsPluginMalloc(sizeof(_cmsParametricCurvesCollection));
+    if (fl == NULL) return FALSE;
+
+    // Copy the parameters
+    fl ->Evaluator  = Plugin ->Evaluator;
+    fl ->nFunctions = Plugin ->nFunctions;
+
+    // Make sure no mem overwrites
+    if (fl ->nFunctions > MAX_TYPES_IN_LCMS_PLUGIN)
+        fl ->nFunctions = MAX_TYPES_IN_LCMS_PLUGIN;
+
+    // Copy the data
+    memmove(fl->FunctionTypes,  Plugin ->FunctionTypes,   fl->nFunctions * sizeof(cmsUInt32Number));
+    memmove(fl->ParameterCount, Plugin ->ParameterCount,  fl->nFunctions * sizeof(cmsUInt32Number));
+
+    // Keep linked list
+    fl ->Next = ParametricCurves;
+    ParametricCurves = fl;
+
+    // All is ok
+    return TRUE;
+}
+
+
+// Search in type list, return position or -1 if not found
+static
+int IsInSet(int Type, _cmsParametricCurvesCollection* c)
+{
+    int i;
+
+    for (i=0; i < c ->nFunctions; i++)
+        if (abs(Type) == c ->FunctionTypes[i]) return i;
+
+    return -1;
+}
+
+
+// Search for the collection which contains a specific type
+static
+_cmsParametricCurvesCollection *GetParametricCurveByType(int Type, int* index)
+{
+    _cmsParametricCurvesCollection* c;
+    int Position;
+
+    for (c = ParametricCurves; c != NULL; c = c ->Next) {
+
+        Position = IsInSet(Type, c);
+
+        if (Position != -1) {
+            if (index != NULL) 
+                *index = Position;
+            return c;
+        }
+    }
+
+    return NULL;
+}
+
+// Low level allocate, which takes care of memory details. nEntries may be zero, and in this case 
+// no optimation curve is computed. nSegments may also be zero in the inverse case, where only the
+// optimization curve is given. Both features simultaneously is an error
+static
+cmsToneCurve* AllocateToneCurveStruct(cmsContext ContextID, cmsInt32Number nEntries, 
+                                      cmsInt32Number nSegments, const cmsCurveSegment* Segments, 
+                                      const cmsUInt16Number* Values)
+{
+    cmsToneCurve* p;
+    int i;
+
+    // We allow huge tables, which are then restricted for smoothing operations
+    if (nEntries > 65530 || nEntries < 0) {
+        cmsSignalError(ContextID, cmsERROR_RANGE, "Couldn't create tone curve of more than 65530 entries");
+        return NULL;
+    }
+
+    if (nEntries <= 0 && nSegments <= 0) {
+        cmsSignalError(ContextID, cmsERROR_RANGE, "Couldn't create tone curve with zero segments and no table");
+        return NULL;
+    }
+
+    // Allocate all required pointers, etc.
+    p = (cmsToneCurve*) _cmsMallocZero(ContextID, sizeof(cmsToneCurve));
+    if (!p) return NULL;
+
+    // In this case, there are no segments
+    if (nSegments <= 0) {
+        p ->Segments = NULL;
+        p ->Evals = NULL;
+    }
+    else {
+        p ->Segments = (cmsCurveSegment*) _cmsCalloc(ContextID, nSegments, sizeof(cmsCurveSegment));
+        if (p ->Segments == NULL) goto Error;
+
+        p ->Evals    = (cmsParametricCurveEvaluator*) _cmsCalloc(ContextID, nSegments, sizeof(cmsParametricCurveEvaluator));
+        if (p ->Evals == NULL) goto Error;
+    }
+
+    p -> nSegments = nSegments;
+    
+    // This 16-bit table contains a limited precision representation of the whole curve and is kept for
+    // increasing xput on certain operations.
+    if (nEntries <= 0) {
+        p ->Table16 = NULL;
+    }
+    else {
+       p ->Table16 = (cmsUInt16Number*)  _cmsCalloc(ContextID, nEntries, sizeof(cmsUInt16Number));
+       if (p ->Table16 == NULL) goto Error;
+    }
+   
+    p -> nEntries  = nEntries;
+    
+    // Initialize members if requested
+    if (Values != NULL && (nEntries > 0)) {
+
+        for (i=0; i < nEntries; i++) 
+            p ->Table16[i] = Values[i];
+    }
+
+    // Initialize the segments stuff. The evaluator for each segment is located and a pointer to it
+    // is placed in advance to maximize performance.
+    if (Segments != NULL && (nSegments > 0)) {
+
+        _cmsParametricCurvesCollection *c;
+
+        p ->SegInterp = (cmsInterpParams**) _cmsCalloc(ContextID, nSegments, sizeof(cmsInterpParams*));
+        if (p ->SegInterp == NULL) goto Error;
+
+        for (i=0; i< nSegments; i++) {
+
+            // Type 0 is a special marker for table-based curves
+            if (Segments[i].Type == 0)
+                p ->SegInterp[i] = _cmsComputeInterpParams(ContextID, Segments[i].nGridPoints, 1, 1, NULL, CMS_LERP_FLAGS_FLOAT);
+
+            memmove(&p ->Segments[i], &Segments[i], sizeof(cmsCurveSegment));
+
+            if (Segments[i].Type == 0 && Segments[i].SampledPoints != NULL)
+                p ->Segments[i].SampledPoints = (cmsFloat32Number*) _cmsDupMem(ContextID, Segments[i].SampledPoints, sizeof(cmsFloat32Number) * Segments[i].nGridPoints);
+            else
+                p ->Segments[i].SampledPoints = NULL;
+
+
+            c = GetParametricCurveByType(Segments[i].Type, NULL);
+            if (c != NULL)
+                    p ->Evals[i] = c ->Evaluator;
+        }
+    }
+    
+    p ->InterpParams = _cmsComputeInterpParams(ContextID, p ->nEntries, 1, 1, p->Table16, CMS_LERP_FLAGS_16BITS);
+    return p;
+
+Error:
+    if (p -> Segments) _cmsFree(ContextID, p ->Segments);
+    if (p -> Evals) _cmsFree(ContextID, p -> Evals);
+    if (p ->Table16) _cmsFree(ContextID, p ->Table16);
+    _cmsFree(ContextID, p);
+    return NULL;
+}
+
+
+// Parametric Fn using floating point
+static
+cmsFloat64Number DefaultEvalParametricFn(cmsInt32Number Type, const cmsFloat64Number Params[], cmsFloat64Number R)
+{
+    cmsFloat64Number e, Val, disc;
+
+    switch (Type) {
+
+    // X = Y ^ Gamma
+    case 1:
+        if (R < 0) 
+            Val = 0;
+        else
+            Val = pow(R, Params[0]);
+        break;
+
+    // Type 1 Reversed: X = Y ^1/gamma
+    case -1:
+        if (R < 0)
+            Val = 0;
+        else
+            Val = pow(R, 1/Params[0]);
+        break;
+
+    // CIE 122-1966
+    // Y = (aX + b)^Gamma  | X >= -b/a
+    // Y = 0               | else
+    case 2:
+        disc = -Params[2] / Params[1];
+
+        if (R >= disc ) {
+
+            e = Params[1]*R + Params[2];
+
+            if (e > 0)
+                Val = pow(e, Params[0]);
+            else
+                Val = 0;
+        }
+        else
+            Val = 0;
+        break;
+
+     // Type 2 Reversed
+     // X = (Y ^1/g  - b) / a
+     case -2: 
+         if (R < 0)
+             Val = 0;
+         else
+             Val = (pow(R, 1.0/Params[0]) - Params[2]) / Params[1];
+
+         if (Val < 0)
+              Val = 0;                            
+         break;
+
+
+    // IEC 61966-3
+    // Y = (aX + b)^Gamma | X <= -b/a
+    // Y = c              | else
+    case 3:
+        disc = -Params[2] / Params[1];
+        if (disc < 0)
+            disc = 0;
+
+        if (R >= disc) {
+
+            e = Params[1]*R + Params[2];  
+
+            if (e > 0)
+                Val = pow(e, Params[0]) + Params[3];
+            else
+                Val = 0;
+        }
+        else
+            Val = Params[3];
+        break;
+
+
+    // Type 3 reversed
+    // X=((Y-c)^1/g - b)/a      | (Y>=c)
+    // X=-b/a                   | (Y<c) 
+    case -3:
+        if (R >= Params[3])  {
+            
+            e = R - Params[3];
+
+            if (e > 0)
+                Val = (pow(e, 1/Params[0]) - Params[2]) / Params[1];
+            else 
+                Val = 0;
+        }
+        else {
+            Val = -Params[2] / Params[1];
+        }
+        break;
+
+
+    // IEC 61966-2.1 (sRGB)
+    // Y = (aX + b)^Gamma | X >= d
+    // Y = cX             | X < d
+    case 4:
+        if (R >= Params[4]) {
+
+            e = Params[1]*R + Params[2];
+
+            if (e > 0)
+                Val = pow(e, Params[0]);
+            else
+                Val = 0;
+        }
+        else
+            Val = R * Params[3];
+        break;
+
+    // Type 4 reversed
+    // X=((Y^1/g-b)/a)    | Y >= (ad+b)^g
+    // X=Y/c              | Y< (ad+b)^g
+    case -4:
+        e = Params[1] * Params[4] + Params[2];
+        if (e < 0)
+            disc = 0;
+        else
+            disc = pow(e, Params[0]);
+
+        if (R >= disc) {
+
+            Val = (pow(R, 1.0/Params[0]) - Params[2]) / Params[1];
+        }
+        else {
+            Val = R / Params[3];
+        }
+        break;
+
+
+    // Y = (aX + b)^Gamma + e | X >= d
+    // Y = cX + f             | X < d
+    case 5:
+        if (R >= Params[4]) {
+
+            e = Params[1]*R + Params[2];
+
+            if (e > 0)
+                Val = pow(e, Params[0]) + Params[5];
+            else
+                Val = 0;
+        }        
+        else
+            Val = R*Params[3] + Params[6];
+        break;
+
+
+    // Reversed type 5
+    // X=((Y-e)1/g-b)/a   | Y >=(ad+b)^g+e), cd+f
+    // X=(Y-f)/c          | else
+    case -5:
+
+        disc = Params[3] * Params[4] + Params[6];
+        if (R >= disc) {
+
+            e = R - Params[5];
+            if (e < 0) 
+                Val = 0;
+            else
+                Val = (pow(e, 1.0/Params[0]) - Params[2]) / Params[1];
+        }
+        else {
+            Val = (R - Params[6]) / Params[3];
+        }
+        break;
+
+
+    // Types 6,7,8 comes from segmented curves as described in ICCSpecRevision_02_11_06_Float.pdf
+    // Type 6 is basically identical to type 5 without d
+    
+    // Y = (a * X + b) ^ Gamma + c
+    case 6:    
+        e = Params[1]*R + Params[2];
+
+        if (e < 0) 
+            Val = 0;
+        else 
+            Val = pow(e, Params[0]) + Params[3];
+        break;
+
+    // ((Y - c) ^1/Gamma - b) / a                        
+    case -6:
+        e = R - Params[3];
+        if (e < 0)
+            Val = 0;
+        else 
+        Val = (pow(e, 1.0/Params[0]) - Params[2]) / Params[1];
+        break;
+
+
+    // Y = a * log (b * X^Gamma + c) + d
+    case 7:                   
+
+       e = Params[2] * pow(R, Params[0]) + Params[3];
+       if (e <= 0)
+           Val = 0;
+       else
+           Val = Params[1]*log10(e) + Params[4];
+       break;
+
+    // (Y - d) / a = log(b * X ^Gamma + c)
+    // pow(10, (Y-d) / a) = b * X ^Gamma + c
+    // pow((pow(10, (Y-d) / a) - c) / b, 1/g) = X 
+    case -7:
+       Val = pow((pow(10.0, (R-Params[4]) / Params[1]) - Params[3]) / Params[2], 1.0 / Params[0]);
+       break;
+
+
+   //Y = a * b^(c*X+d) + e          
+   case 8:
+       Val = (Params[0] * pow(Params[1], Params[2] * R + Params[3]) + Params[4]);
+       break;
+
+
+   // Y = (log((y-e) / a) / log(b) - d ) / c
+   // a=0, b=1, c=2, d=3, e=4,
+   case -8:
+             
+       disc = R - Params[4];
+       if (disc < 0) Val = 0;
+       else 
+           Val = (log(disc / Params[0]) / log(Params[1]) - Params[3]) / Params[2];         
+       break;
+
+   // S-Shaped: (1 - (1-x)^1/g)^1/g                    
+   case 108:
+      Val = pow(1.0 - pow(1 - R, 1/Params[0]), 1/Params[0]);
+      break;
+
+    // y = (1 - (1-x)^1/g)^1/g
+    // y^g = (1 - (1-x)^1/g)
+    // 1 - y^g = (1-x)^1/g
+    // (1 - y^g)^g = 1 - x
+    // 1 - (1 - y^g)^g
+    case -108:
+        Val = 1 - pow(1 - pow(R, Params[0]), Params[0]);
+        break;
+
+    default:
+        // Unsupported parametric curve. Should never reach here
+        return 0;
+    }
+
+    return Val;
+}
+
+// Evaluate a segmented funtion for a single value. Return -1 if no valid segment found .
+// If fn type is 0, perform an interpolation on the table 
+static
+cmsFloat64Number EvalSegmentedFn(const cmsToneCurve *g, cmsFloat64Number R)
+{
+    int i;
+
+    for (i = g ->nSegments-1; i >= 0 ; --i) {
+
+        // Check for domain
+        if ((R > g ->Segments[i].x0) && (R <= g ->Segments[i].x1)) {
+
+            // Type == 0 means segment is sampled
+            if (g ->Segments[i].Type == 0) {
+
+                cmsFloat32Number R1 = (cmsFloat32Number) (R - g ->Segments[i].x0);
+                cmsFloat32Number Out;
+
+                // Setup the table (TODO: clean that)
+                g ->SegInterp[i]-> Table = g ->Segments[i].SampledPoints; 
+
+                g ->SegInterp[i] -> Interpolation.LerpFloat(&R1, &Out, g ->SegInterp[i]);
+                
+                return Out;
+            }
+            else
+                return g ->Evals[i](g->Segments[i].Type, g ->Segments[i].Params, R);
+        }
+    }
+
+    return MINUS_INF;
+}
+
+
+// Create an empty gamma curve, by using tables. This specifies only the limited-precision part, and leaves the
+// floating point description empty.
+cmsToneCurve* CMSEXPORT cmsBuildTabulatedToneCurve16(cmsContext ContextID, cmsInt32Number nEntries, const cmsUInt16Number Values[])
+{
+    return AllocateToneCurveStruct(ContextID, nEntries, 0, NULL, Values);
+}
+
+static
+int EntriesByGamma(cmsFloat64Number Gamma)
+{
+    if (fabs(Gamma - 1.0) < 0.001) return 2;
+    return 4096;
+}
+
+
+// Create a segmented gamma, fill the table
+cmsToneCurve* CMSEXPORT cmsBuildSegmentedToneCurve(cmsContext ContextID, 
+                                                   cmsInt32Number nSegments, const cmsCurveSegment Segments[])
+{
+    int i;
+    cmsFloat64Number R, Val;
+    cmsToneCurve* g;
+    int nGridPoints = 4096;
+        
+    _cmsAssert(Segments != NULL);
+
+    // Optimizatin for identity curves. 
+    if (nSegments == 1 && Segments[0].Type == 1) {
+
+        nGridPoints = EntriesByGamma(Segments[0].Params[0]);
+    }
+
+    g = AllocateToneCurveStruct(ContextID, nGridPoints, nSegments, Segments, NULL);
+    if (g == NULL) return NULL;
+
+    // Once we have the floating point version, we can approximate a 16 bit table of 4096 entries
+    // for performance reasons. This table would normally not be used except on 8/16 bits transforms.
+    for (i=0; i < nGridPoints; i++) {
+
+        R   = (cmsFloat64Number) i / (nGridPoints-1);
+
+        Val = EvalSegmentedFn(g, R);
+
+        // Round and saturate
+        g ->Table16[i] = _cmsQuickSaturateWord(Val * 65535.0);
+    }
+
+    return g;
+}
+
+// Use a segmented curve to store the floating point table
+cmsToneCurve* CMSEXPORT cmsBuildTabulatedToneCurveFloat(cmsContext ContextID, cmsUInt32Number nEntries, const cmsFloat32Number values[])
+{
+    cmsCurveSegment Seg[2];
+
+    // Initialize segmented curve part up to 0
+    Seg[0].x0 = -1;
+    Seg[0].x1 = 0;
+    Seg[0].Type = 6;
+
+    Seg[0].Params[0] = 1;
+    Seg[0].Params[1] = 0;
+    Seg[0].Params[2] = 0;
+    Seg[0].Params[3] = 0;
+    Seg[0].Params[4] = 0;
+
+    // From zero to any
+    Seg[1].x0 = 0;
+    Seg[1].x1 = 1.0;   
+    Seg[1].Type = 0;
+
+    Seg[1].nGridPoints = nEntries;
+    Seg[1].SampledPoints = (cmsFloat32Number*) values;
+
+    return cmsBuildSegmentedToneCurve(ContextID, 2, Seg);
+}
+
+// Parametric curves
+//
+// Parameters goes as: Curve, a, b, c, d, e, f
+// Type is the ICC type +1
+// if type is negative, then the curve is analyticaly inverted
+cmsToneCurve* CMSEXPORT cmsBuildParametricToneCurve(cmsContext ContextID, cmsInt32Number Type, const cmsFloat64Number Params[])
+{
+    cmsCurveSegment Seg0;
+    int Pos = 0;
+    cmsUInt32Number size;
+    _cmsParametricCurvesCollection* c = GetParametricCurveByType(Type, &Pos);
+
+    _cmsAssert(Params != NULL);
+
+    if (c == NULL) {
+         cmsSignalError(ContextID, cmsERROR_UNKNOWN_EXTENSION, "Invalid parametric curve type %d", Type);     
+        return NULL;
+    }
+
+    memset(&Seg0, 0, sizeof(Seg0));
+
+    Seg0.x0   = MINUS_INF;
+    Seg0.x1   = PLUS_INF;
+    Seg0.Type = Type;
+
+    size = c->ParameterCount[Pos] * sizeof(cmsFloat64Number);
+    memmove(Seg0.Params, Params, size);
+
+    return cmsBuildSegmentedToneCurve(ContextID, 1, &Seg0);
+}
+
+
+
+// Build a gamma table based on gamma constant
+cmsToneCurve* CMSEXPORT cmsBuildGamma(cmsContext ContextID, cmsFloat64Number Gamma)
+{
+    return cmsBuildParametricToneCurve(ContextID, 1, &Gamma);
+}
+
+
+// Free all memory taken by the gamma curve
+void CMSEXPORT cmsFreeToneCurve(cmsToneCurve* Curve)
+{
+    cmsContext ContextID;
+    
+    if (Curve == NULL) return;
+
+    ContextID = Curve ->InterpParams->ContextID;
+
+    _cmsFreeInterpParams(Curve ->InterpParams);
+    
+    if (Curve -> Table16)
+        _cmsFree(ContextID, Curve ->Table16);
+
+    if (Curve ->Segments) {
+
+        cmsUInt32Number i;
+        
+        for (i=0; i < Curve ->nSegments; i++) {
+
+            if (Curve ->Segments[i].SampledPoints) {
+                _cmsFree(ContextID, Curve ->Segments[i].SampledPoints);
+            }
+
+            if (Curve ->SegInterp[i] != 0) 
+                _cmsFreeInterpParams(Curve->SegInterp[i]);
+        }
+
+        _cmsFree(ContextID, Curve ->Segments);
+        _cmsFree(ContextID, Curve ->SegInterp);
+    }
+
+    if (Curve -> Evals)
+        _cmsFree(ContextID, Curve -> Evals);
+
+    if (Curve) _cmsFree(ContextID, Curve);
+}
+
+// Utility function, free 3 gamma tables
+void CMSEXPORT cmsFreeToneCurveTriple(cmsToneCurve* Curve[3])
+{
+
+    _cmsAssert(Curve != NULL);
+
+    if (Curve[0] != NULL) cmsFreeToneCurve(Curve[0]);
+    if (Curve[1] != NULL) cmsFreeToneCurve(Curve[1]);
+    if (Curve[2] != NULL) cmsFreeToneCurve(Curve[2]);
+
+    Curve[0] = Curve[1] = Curve[2] = NULL;
+}
+
+
+// Duplicate a gamma table
+cmsToneCurve* CMSEXPORT cmsDupToneCurve(const cmsToneCurve* In)
+{   
+    if (In == NULL) return NULL;
+
+    return  AllocateToneCurveStruct(In ->InterpParams ->ContextID, In ->nEntries, In ->nSegments, In ->Segments, In ->Table16);
+}
+
+// Joins two curves for X and Y. Curves should be monotonic.
+// We want to get 
+//
+//      y = Y^-1(X(t)) 
+//
+cmsToneCurve* CMSEXPORT cmsJoinToneCurve(cmsContext ContextID, 
+                                      const cmsToneCurve* X,
+                                      const cmsToneCurve* Y, cmsUInt32Number nResultingPoints)
+{
+    cmsToneCurve* out = NULL;
+    cmsToneCurve* Yreversed = NULL;
+    cmsFloat32Number t, x;
+    cmsFloat32Number* Res = NULL;
+    cmsUInt32Number i;
+
+
+    _cmsAssert(X != NULL);
+    _cmsAssert(Y != NULL);
+
+    Yreversed = cmsReverseToneCurveEx(nResultingPoints, Y);
+    if (Yreversed == NULL) goto Error;
+
+    Res = (cmsFloat32Number*) _cmsCalloc(ContextID, nResultingPoints, sizeof(cmsFloat32Number));
+    if (Res == NULL) goto Error;
+    
+    //Iterate
+    for (i=0; i <  nResultingPoints; i++) {
+
+        t = (cmsFloat32Number) i / (nResultingPoints-1);
+        x = cmsEvalToneCurveFloat(X,  t);
+        Res[i] = cmsEvalToneCurveFloat(Yreversed, x);
+    }
+
+    // Allocate space for output
+    out = cmsBuildTabulatedToneCurveFloat(ContextID, nResultingPoints, Res);
+    
+Error:
+
+    if (Res != NULL) _cmsFree(ContextID, Res);
+    if (Yreversed != NULL) cmsFreeToneCurve(Yreversed);
+
+    return out;
+}
+
+
+
+// Get the surrounding nodes. This is tricky on non-monotonic tables 
+static
+int GetInterval(cmsFloat64Number In, const cmsUInt16Number LutTable[], const struct _cms_interp_struc* p)
+{   
+    int i;
+    int y0, y1;
+    
+    // A 1 point table is not allowed
+    if (p -> Domain[0] < 1) return -1;
+
+    // Let's see if ascending or descending. 
+    if (LutTable[0] < LutTable[p ->Domain[0]]) {
+
+        // Table is overall ascending
+        for (i=p->Domain[0]-1; i >=0; --i) {
+
+            y0 = LutTable[i]; 
+            y1 = LutTable[i+1];
+            
+            if (y0 <= y1) { // Increasing
+                if (In >= y0 && In <= y1) return i;
+            }
+            else
+                if (y1 < y0) { // Decreasing
+                    if (In >= y1 && In <= y0) return i;
+                }
+        }
+    }
+    else {
+        // Table is overall descending
+        for (i=0; i < (int) p -> Domain[0]; i++) {
+
+            y0 = LutTable[i]; 
+            y1 = LutTable[i+1];
+
+            if (y0 <= y1) { // Increasing
+                if (In >= y0 && In <= y1) return i;
+            }
+            else
+                if (y1 < y0) { // Decreasing
+                    if (In >= y1 && In <= y0) return i;
+                }
+        }
+    }
+
+    return -1;
+}
+
+// Reverse a gamma table
+cmsToneCurve* CMSEXPORT cmsReverseToneCurveEx(cmsInt32Number nResultSamples, const cmsToneCurve* InCurve)
+{
+    cmsToneCurve *out;
+    cmsFloat64Number a = 0, b = 0, y, x1, y1, x2, y2;
+    int i, j;
+    int Ascending;
+    
+    _cmsAssert(InCurve != NULL);
+
+    // Try to reverse it analytically whatever possible
+    if (InCurve ->nSegments == 1 && InCurve ->Segments[0].Type > 0 && InCurve -> Segments[0].Type <= 5) {
+
+        return cmsBuildParametricToneCurve(InCurve ->InterpParams->ContextID, 
+                                       -(InCurve -> Segments[0].Type), 
+                                       InCurve -> Segments[0].Params);
+    }
+
+    // Nope, reverse the table. 
+    out = cmsBuildTabulatedToneCurve16(InCurve ->InterpParams->ContextID, nResultSamples, NULL);
+    if (out == NULL)
+        return NULL;
+
+    // We want to know if this is an ascending or descending table
+    Ascending = !cmsIsToneCurveDescending(InCurve);
+
+    // Iterate across Y axis
+    for (i=0; i <  nResultSamples; i++) {
+
+        y = (cmsFloat64Number) i * 65535.0 / (nResultSamples - 1);
+
+        // Find interval in which y is within. 
+        j = GetInterval(y, InCurve->Table16, InCurve->InterpParams);
+        if (j >= 0) {
+
+
+            // Get limits of interval
+            x1 = InCurve ->Table16[j]; 
+            x2 = InCurve ->Table16[j+1];
+
+            y1 = (cmsFloat64Number) (j * 65535.0) / (InCurve ->nEntries - 1);
+            y2 = (cmsFloat64Number) ((j+1) * 65535.0 ) / (InCurve ->nEntries - 1);
+    
+            // If collapsed, then use any
+            if (x1 == x2) {
+
+                out ->Table16[i] = _cmsQuickSaturateWord(Ascending ? y2 : y1);
+                continue;
+
+            } else {
+
+                // Interpolate      
+                a = (y2 - y1) / (x2 - x1);
+                b = y2 - a * x2;
+            }
+        }
+           
+        out ->Table16[i] = _cmsQuickSaturateWord(a* y + b);
+    }
+
+
+    return out;
+}
+
+// Reverse a gamma table
+cmsToneCurve* CMSEXPORT cmsReverseToneCurve(const cmsToneCurve* InGamma)
+{
+    _cmsAssert(InGamma != NULL);
+
+    return cmsReverseToneCurveEx(4096, InGamma);
+}
+
+// From: Eilers, P.H.C. (1994) Smoothing and interpolation with finite
+// differences. in: Graphic Gems IV, Heckbert, P.S. (ed.), Academic press.
+//
+// Smoothing and interpolation with second differences.
+//
+//   Input:  weights (w), data (y): vector from 1 to m.
+//   Input:  smoothing parameter (lambda), length (m).
+//   Output: smoothed vector (z): vector from 1 to m.
+
+static
+cmsBool smooth2(cmsContext ContextID, cmsFloat32Number w[], cmsFloat32Number y[], cmsFloat32Number z[], cmsFloat32Number lambda, int m)
+{
+    int i, i1, i2;
+    cmsFloat32Number *c, *d, *e;
+    cmsBool st;
+
+
+    c = (cmsFloat32Number*) _cmsCalloc(ContextID, MAX_NODES_IN_CURVE, sizeof(cmsFloat32Number));
+    d = (cmsFloat32Number*) _cmsCalloc(ContextID, MAX_NODES_IN_CURVE, sizeof(cmsFloat32Number));
+    e = (cmsFloat32Number*) _cmsCalloc(ContextID, MAX_NODES_IN_CURVE, sizeof(cmsFloat32Number));
+    
+    if (c != NULL && d != NULL && e != NULL) {
+
+
+    d[1] = w[1] + lambda;
+    c[1] = -2 * lambda / d[1];
+    e[1] = lambda /d[1];
+    z[1] = w[1] * y[1];
+    d[2] = w[2] + 5 * lambda - d[1] * c[1] *  c[1];
+    c[2] = (-4 * lambda - d[1] * c[1] * e[1]) / d[2];
+    e[2] = lambda / d[2];
+    z[2] = w[2] * y[2] - c[1] * z[1];
+    
+    for (i = 3; i < m - 1; i++) {
+        i1 = i - 1; i2 = i - 2;
+        d[i]= w[i] + 6 * lambda - c[i1] * c[i1] * d[i1] - e[i2] * e[i2] * d[i2];
+        c[i] = (-4 * lambda -d[i1] * c[i1] * e[i1])/ d[i];
+        e[i] = lambda / d[i];
+        z[i] = w[i] * y[i] - c[i1] * z[i1] - e[i2] * z[i2];
+    }
+    
+    i1 = m - 2; i2 = m - 3;
+    
+    d[m - 1] = w[m - 1] + 5 * lambda -c[i1] * c[i1] * d[i1] - e[i2] * e[i2] * d[i2];
+    c[m - 1] = (-2 * lambda - d[i1] * c[i1] * e[i1]) / d[m - 1];
+    z[m - 1] = w[m - 1] * y[m - 1] - c[i1] * z[i1] - e[i2] * z[i2];
+    i1 = m - 1; i2 = m - 2;
+    
+    d[m] = w[m] + lambda - c[i1] * c[i1] * d[i1] - e[i2] * e[i2] * d[i2];
+    z[m] = (w[m] * y[m] - c[i1] * z[i1] - e[i2] * z[i2]) / d[m];
+    z[m - 1] = z[m - 1] / d[m - 1] - c[m - 1] * z[m];
+    
+    for (i = m - 2; 1<= i; i--)
+        z[i] = z[i] / d[i] - c[i] * z[i + 1] - e[i] * z[i + 2];
+
+      st = TRUE;
+    }
+    else st = FALSE;
+
+    if (c != NULL) _cmsFree(ContextID, c);
+    if (d != NULL) _cmsFree(ContextID, d);
+    if (e != NULL) _cmsFree(ContextID, e);
+
+    return st;
+}
+
+// Smooths a curve sampled at regular intervals. 
+cmsBool  CMSEXPORT cmsSmoothToneCurve(cmsToneCurve* Tab, cmsFloat64Number lambda)
+{
+    cmsFloat32Number w[MAX_NODES_IN_CURVE], y[MAX_NODES_IN_CURVE], z[MAX_NODES_IN_CURVE];
+    int i, nItems, Zeros, Poles;
+
+    if (Tab == NULL) return FALSE;
+
+    if (cmsIsToneCurveLinear(Tab)) return FALSE; // Nothing to do
+
+    nItems = Tab -> nEntries;
+
+    if (nItems >= MAX_NODES_IN_CURVE) {
+        cmsSignalError(Tab ->InterpParams->ContextID, cmsERROR_RANGE, "cmsSmoothToneCurve: too many points.");
+        return FALSE;
+    }
+
+    memset(w, 0, nItems * sizeof(cmsFloat32Number));
+    memset(y, 0, nItems * sizeof(cmsFloat32Number));
+    memset(z, 0, nItems * sizeof(cmsFloat32Number));
+
+    for (i=0; i < nItems; i++)
+    {
+        y[i+1] = (cmsFloat32Number) Tab -> Table16[i];
+        w[i+1] = 1.0;
+    }
+
+    if (!smooth2(Tab ->InterpParams->ContextID, w, y, z, (cmsFloat32Number) lambda, nItems)) return FALSE;
+
+    // Do some reality - checking...
+    Zeros = Poles = 0;
+    for (i=nItems; i > 1; --i) {
+
+        if (z[i] == 0.) Zeros++;
+        if (z[i] >= 65535.) Poles++;
+        if (z[i] < z[i-1]) return FALSE; // Non-Monotonic
+    }
+
+    if (Zeros > (nItems / 3)) return FALSE;  // Degenerated, mostly zeros
+    if (Poles > (nItems / 3)) return FALSE;  // Degenerated, mostly poles
+
+    // Seems ok
+    for (i=0; i < nItems; i++) {
+
+        // Clamp to cmsUInt16Number
+        Tab -> Table16[i] = _cmsQuickSaturateWord(z[i+1]);
+    }
+
+    return TRUE;
+}
+
+// Is a table linear? Do not use parametric since we cannot guarantee some weird parameters resulting
+// in a linear table. This way assures it is linear in 12 bits, which should be enought in most cases.
+cmsBool CMSEXPORT cmsIsToneCurveLinear(const cmsToneCurve* Curve)
+{
+    cmsUInt32Number i;
+    int diff;
+
+    _cmsAssert(Curve != NULL);
+
+    for (i=0; i < Curve ->nEntries; i++) {
+
+        diff = abs((int) Curve->Table16[i] - (int) _cmsQuantizeVal(i, Curve ->nEntries));
+        if (diff > 0x0f)
+            return FALSE;
+    }
+
+    return TRUE;
+}
+
+// Same, but for monotonicity
+cmsBool  CMSEXPORT cmsIsToneCurveMonotonic(const cmsToneCurve* t)
+{
+    int n;
+    int i, last;
+
+    _cmsAssert(t != NULL);
+
+    n    = t ->nEntries;
+    last = t ->Table16[n-1];
+
+    for (i = n-2; i >= 0; --i) {
+
+        if (t ->Table16[i] > last)
+
+            return FALSE;
+        else
+            last = t ->Table16[i];
+
+    }
+
+    return TRUE;
+}
+
+// Same, but for descending tables
+cmsBool  CMSEXPORT cmsIsToneCurveDescending(const cmsToneCurve* t)
+{
+    _cmsAssert(t != NULL);
+
+    return t ->Table16[0] > t ->Table16[t ->nEntries-1];
+}
+
+
+// Another info fn: is out gamma table multisegment?
+cmsBool  CMSEXPORT cmsIsToneCurveMultisegment(const cmsToneCurve* t)
+{
+    _cmsAssert(t != NULL);
+
+    return t -> nSegments > 1;
+}
+
+cmsInt32Number  CMSEXPORT cmsGetToneCurveParametricType(const cmsToneCurve* t)
+{
+    _cmsAssert(t != NULL);
+
+    if (t -> nSegments != 1) return 0;
+    return t ->Segments[0].Type;
+}
+
+// We need accuracy this time
+cmsFloat32Number CMSEXPORT cmsEvalToneCurveFloat(const cmsToneCurve* Curve, cmsFloat32Number v)
+{
+    _cmsAssert(Curve != NULL);
+
+    // Check for 16 bits table. If so, this is a limited-precision tone curve
+    if (Curve ->nSegments == 0) {
+
+        cmsUInt16Number In, Out;
+        
+        In = (cmsUInt16Number) _cmsQuickSaturateWord(v * 65535.0);
+        Out = cmsEvalToneCurve16(Curve, In);
+        
+        return (cmsFloat32Number) (Out / 65535.0);
+    }
+
+    return (cmsFloat32Number) EvalSegmentedFn(Curve, v);
+}
+
+// We need xput over here
+cmsUInt16Number CMSEXPORT cmsEvalToneCurve16(const cmsToneCurve* Curve, cmsUInt16Number v)
+{
+    cmsUInt16Number out;
+
+    _cmsAssert(Curve != NULL);
+
+    Curve ->InterpParams ->Interpolation.Lerp16(&v, &out, Curve ->InterpParams);
+    return out;
+}
+
+
+// Least squares fitting.
+// A mathematical procedure for finding the best-fitting curve to a given set of points by 
+// minimizing the sum of the squares of the offsets ("the residuals") of the points from the curve. 
+// The sum of the squares of the offsets is used instead of the offset absolute values because 
+// this allows the residuals to be treated as a continuous differentiable quantity. 
+//
+// y = f(x) = x ^ g
+//
+// R  = (yi - (xi^g))
+// R2 = (yi - (xi^g))2
+// SUM R2 = SUM (yi - (xi^g))2
+// 
+// dR2/dg = -2 SUM x^g log(x)(y - x^g)    
+// solving for dR2/dg = 0 
+// 
+// g = 1/n * SUM(log(y) / log(x)) 
+
+cmsFloat64Number CMSEXPORT cmsEstimateGamma(const cmsToneCurve* t, cmsFloat64Number Precision)
+{
+    cmsFloat64Number gamma, sum, sum2;
+    cmsFloat64Number n, x, y, Std;
+    cmsUInt32Number i;
+
+    _cmsAssert(t != NULL);
+
+    sum = sum2 = n = 0;
+
+    // Excluding endpoints   
+    for (i=1; i < (MAX_NODES_IN_CURVE-1); i++) {
+
+        x = (cmsFloat64Number) i / (MAX_NODES_IN_CURVE-1);
+        y = (cmsFloat64Number) cmsEvalToneCurveFloat(t, (cmsFloat32Number) x);
+
+        // Avoid 7% on lower part to prevent 
+        // artifacts due to linear ramps
+
+        if (y > 0. && y < 1. && x > 0.07) {
+
+            gamma = log(y) / log(x);
+            sum  += gamma;
+            sum2 += gamma * gamma;
+            n++;
+        }
+    }
+
+    // Take a look on SD to see if gamma isn't exponential at all
+    Std = sqrt((n * sum2 - sum * sum) / (n*(n-1)));
+
+    if (Std > Precision)
+        return -1.0;
+
+    return (sum / n);   // The mean
+}
diff --git a/thirdparty/liblcms2/src/cmsgmt.c b/thirdparty/liblcms2/src/cmsgmt.c
new file mode 100644
index 00000000..c2f3a9a5
--- /dev/null
+++ b/thirdparty/liblcms2/src/cmsgmt.c
@@ -0,0 +1,591 @@
+//---------------------------------------------------------------------------------
+//
+//  Little Color Management System
+//  Copyright (c) 1998-2010 Marti Maria Saguer
+//
+// Permission is hereby granted, free of charge, to any person obtaining 
+// a copy of this software and associated documentation files (the "Software"), 
+// to deal in the Software without restriction, including without limitation 
+// the rights to use, copy, modify, merge, publish, distribute, sublicense, 
+// and/or sell copies of the Software, and to permit persons to whom the Software 
+// is furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in 
+// all copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 
+// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO 
+// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE 
+// LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 
+// OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 
+// WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+//
+//---------------------------------------------------------------------------------
+//
+
+#include "lcms2_internal.h"
+
+
+// Auxiliar: append a Lab identity after the given sequence of profiles
+// and return the transform. Lab profile is closed, rest of profiles are kept open.
+cmsHTRANSFORM _cmsChain2Lab(cmsContext            ContextID,
+                            cmsUInt32Number        nProfiles,
+                            cmsUInt32Number        InputFormat,
+                            cmsUInt32Number        OutputFormat,
+                            const cmsUInt32Number  Intents[], 
+                            const cmsHPROFILE      hProfiles[], 
+                            const cmsBool          BPC[],
+                            const cmsFloat64Number AdaptationStates[],
+                            cmsUInt32Number        dwFlags)
+{
+    cmsHTRANSFORM xform;
+    cmsHPROFILE   hLab;
+    cmsHPROFILE   ProfileList[256];
+    cmsBool       BPCList[256];
+    cmsFloat64Number AdaptationList[256];
+    cmsUInt32Number IntentList[256];
+    cmsUInt32Number i;
+
+    // This is a rather big number and there is no need of dynamic memory
+    // since we are adding a profile, 254 + 1 = 255 and this is the limit
+    if (nProfiles > 254) return NULL;
+    
+    // The output space
+    hLab = cmsCreateLab4ProfileTHR(ContextID, NULL);
+    if (hLab == NULL) return NULL;
+
+    // Create a copy of parameters
+    for (i=0; i < nProfiles; i++) {
+
+        ProfileList[i]    = hProfiles[i];
+        BPCList[i]        = BPC[i];
+        AdaptationList[i] = AdaptationStates[i];
+        IntentList[i]     = Intents[i];
+    }
+
+    // Place Lab identity at chain's end.
+    ProfileList[nProfiles]    = hLab;
+    BPCList[nProfiles]        = 0;
+    AdaptationList[nProfiles] = 1.0;
+    IntentList[nProfiles]     = INTENT_RELATIVE_COLORIMETRIC;
+   
+    // Create the transform
+    xform = cmsCreateExtendedTransform(ContextID, nProfiles + 1, ProfileList, 
+                                       BPCList, 
+                                       IntentList, 
+                                       AdaptationList, 
+                                       NULL, 0, 
+                                       InputFormat, 
+                                       OutputFormat, 
+                                       dwFlags);
+    
+    cmsCloseProfile(hLab);
+
+    return xform;
+}
+
+
+// Compute K -> L* relationship. Flags may include black point compensation. In this case, 
+// the relationship is assumed from the profile with BPC to a black point zero.
+static
+cmsToneCurve* ComputeKToLstar(cmsContext            ContextID,
+                               cmsUInt32Number       nPoints,
+                               cmsUInt32Number       nProfiles,
+                               const cmsUInt32Number Intents[], 
+                               const cmsHPROFILE     hProfiles[], 
+                               const cmsBool         BPC[],
+                               const cmsFloat64Number AdaptationStates[],
+                               cmsUInt32Number dwFlags)
+{
+    cmsToneCurve* out = NULL;   
+    cmsUInt32Number i;
+    cmsHTRANSFORM xform;
+    cmsCIELab Lab;
+    cmsFloat32Number cmyk[4];
+    cmsFloat32Number* SampledPoints;
+    
+    xform = _cmsChain2Lab(ContextID, nProfiles, TYPE_CMYK_FLT, TYPE_Lab_DBL, Intents, hProfiles, BPC, AdaptationStates, dwFlags);
+    if (xform == NULL) return NULL;
+        
+    SampledPoints = (cmsFloat32Number*) _cmsCalloc(ContextID, nPoints, sizeof(cmsFloat32Number));
+    if (SampledPoints  == NULL) goto Error;
+        
+    for (i=0; i < nPoints; i++) {
+
+        cmyk[0] = 0;
+        cmyk[1] = 0;
+        cmyk[2] = 0;
+        cmyk[3] = (cmsFloat32Number) ((i * 100.0) / (nPoints-1));
+
+        cmsDoTransform(xform, cmyk, &Lab, 1);
+        SampledPoints[i]= (cmsFloat32Number) (1.0 - Lab.L / 100.0); // Negate K for easier operation
+    }
+        
+    out = cmsBuildTabulatedToneCurveFloat(ContextID, nPoints, SampledPoints);
+
+Error:
+
+    cmsDeleteTransform(xform);  
+    if (SampledPoints) _cmsFree(ContextID, SampledPoints);
+
+    return out;
+}
+
+
+// Compute Black tone curve on a CMYK -> CMYK transform. This is done by
+// using the proof direction on both profiles to find K->L* relationship
+// then joining both curves. dwFlags may include black point compensation.
+cmsToneCurve* _cmsBuildKToneCurve(cmsContext        ContextID,
+                                   cmsUInt32Number   nPoints,
+                                   cmsUInt32Number   nProfiles,
+                                   const cmsUInt32Number Intents[], 
+                                   const cmsHPROFILE hProfiles[], 
+                                   const cmsBool     BPC[],
+                                   const cmsFloat64Number AdaptationStates[],
+                                   cmsUInt32Number   dwFlags)
+{
+    cmsToneCurve *in, *out, *KTone;       
+ 
+    // Make sure CMYK -> CMYK
+    if (cmsGetColorSpace(hProfiles[0]) != cmsSigCmykData ||
+        cmsGetColorSpace(hProfiles[nProfiles-1])!= cmsSigCmykData) return NULL;
+
+    
+    // Make sure last is an output profile
+    if (cmsGetDeviceClass(hProfiles[nProfiles - 1]) != cmsSigOutputClass) return NULL;
+
+    // Create individual curves. BPC works also as each K to L* is
+    // computed as a BPC to zero black point in case of L*
+    in  = ComputeKToLstar(ContextID, nPoints, nProfiles - 1, Intents, hProfiles, BPC, AdaptationStates, dwFlags);
+    if (in == NULL) return NULL;
+    
+    out = ComputeKToLstar(ContextID, nPoints, 1,             
+                            Intents + (nProfiles - 1), 
+                            hProfiles + (nProfiles - 1), 
+                            BPC + (nProfiles - 1),  
+                            AdaptationStates + (nProfiles - 1), 
+                            dwFlags);
+    if (out == NULL) {
+        cmsFreeToneCurve(in);
+        return NULL;
+    }
+
+    // Build the relationship. This effectively limits the maximum accuracy to 16 bits, but 
+    // since this is used on black-preserving LUTs, we are not loosing  accuracy in any case 
+    KTone = cmsJoinToneCurve(ContextID, in, out, nPoints);
+
+    // Get rid of components
+    cmsFreeToneCurve(in); cmsFreeToneCurve(out);
+
+    // Something went wrong...
+    if (KTone == NULL) return NULL;
+
+    // Make sure it is monotonic    
+    if (!cmsIsToneCurveMonotonic(KTone)) {
+
+        cmsFreeToneCurve(KTone);
+        return NULL;
+    }
+    
+    return KTone;
+}
+
+
+// Gamut LUT Creation -----------------------------------------------------------------------------------------
+
+// Used by gamut & softproofing
+
+typedef struct {
+
+    cmsHTRANSFORM hInput;               // From whatever input color space. 16 bits to DBL
+    cmsHTRANSFORM hForward, hReverse;   // Transforms going from Lab to colorant and back
+    cmsFloat64Number Thereshold;        // The thereshold after which is considered out of gamut
+    
+    } GAMUTCHAIN;
+
+// This sampler does compute gamut boundaries by comparing original
+// values with a transform going back and forth. Values above ERR_THERESHOLD 
+// of maximum are considered out of gamut.
+
+#define ERR_THERESHOLD      5 
+
+
+static
+int GamutSampler(register const cmsUInt16Number In[], register cmsUInt16Number Out[], register void* Cargo)
+{
+    GAMUTCHAIN*  t = (GAMUTCHAIN* ) Cargo;          
+    cmsCIELab LabIn1, LabOut1;  
+    cmsCIELab LabIn2, LabOut2;      
+    cmsUInt16Number Proof[cmsMAXCHANNELS], Proof2[cmsMAXCHANNELS];
+    cmsFloat64Number dE1, dE2, ErrorRatio;
+
+    // Assume in-gamut by default.
+    dE1 = 0.;
+    dE2 = 0;
+    ErrorRatio = 1.0;
+
+    // Convert input to Lab
+    if (t -> hInput != NULL) 
+        cmsDoTransform(t -> hInput, In, &LabIn1, 1);
+
+    // converts from PCS to colorant. This always
+    // does return in-gamut values, 
+    cmsDoTransform(t -> hForward, &LabIn1, Proof, 1);
+
+    // Now, do the inverse, from colorant to PCS.
+    cmsDoTransform(t -> hReverse, Proof, &LabOut1, 1);
+
+    memmove(&LabIn2, &LabOut1, sizeof(cmsCIELab));
+
+    // Try again, but this time taking Check as input
+    cmsDoTransform(t -> hForward, &LabOut1, Proof2,  1);
+    cmsDoTransform(t -> hReverse, Proof2, &LabOut2, 1);
+    
+    // Take difference of direct value
+    dE1 = cmsDeltaE(&LabIn1, &LabOut1);        
+
+    // Take difference of converted value
+    dE2 = cmsDeltaE(&LabIn2, &LabOut2);                 
+
+
+    // if dE1 is small and dE2 is small, value is likely to be in gamut
+    if (dE1 < t->Thereshold && dE2 < t->Thereshold) 
+        Out[0] = 0;
+    else {
+
+        // if dE1 is small and dE2 is big, undefined. Assume in gamut
+        if (dE1 < t->Thereshold && dE2 > t->Thereshold)
+            Out[0] = 0;
+        else
+            // dE1 is big and dE2 is small, clearly out of gamut
+            if (dE1 > t->Thereshold && dE2 < t->Thereshold)
+                Out[0] = (cmsUInt16Number) _cmsQuickFloor((dE1 - t->Thereshold) + .5);
+            else  {
+
+                // dE1 is big and dE2 is also big, could be due to perceptual mapping
+                // so take error ratio
+                if (dE2 == 0.0)
+                    ErrorRatio = dE1;
+                else
+                    ErrorRatio = dE1 / dE2;
+
+                if (ErrorRatio > t->Thereshold) 
+                    Out[0] = (cmsUInt16Number)  _cmsQuickFloor((ErrorRatio - t->Thereshold) + .5);
+                else
+                    Out[0] = 0;
+            }
+    }
+
+
+    return TRUE;
+}
+
+// Does compute a gamut LUT going back and forth across pcs -> relativ. colorimetric intent -> pcs
+// the dE obtained is then annotated on the LUT. Values truely out of gamut are clipped to dE = 0xFFFE
+// and values changed are supposed to be handled by any gamut remapping, so, are out of gamut as well.
+//
+// **WARNING: This algorithm does assume that gamut remapping algorithms does NOT move in-gamut colors,
+// of course, many perceptual and saturation intents does not work in such way, but relativ. ones should.
+
+cmsPipeline* _cmsCreateGamutCheckPipeline(cmsContext ContextID,
+										  cmsHPROFILE hProfiles[], 
+										  cmsBool  BPC[], 
+										  cmsUInt32Number Intents[], 
+										  cmsFloat64Number AdaptationStates[],
+										  cmsUInt32Number nGamutPCSposition, 
+										  cmsHPROFILE hGamut)
+{
+	cmsHPROFILE hLab;
+	cmsPipeline* Gamut;
+	cmsStage* CLUT;
+	cmsUInt32Number dwFormat;
+	GAMUTCHAIN Chain;
+	int nChannels, nGridpoints;
+	cmsColorSpaceSignature ColorSpace;
+	cmsUInt32Number i;    
+	cmsHPROFILE ProfileList[256];
+	cmsBool     BPCList[256];
+	cmsFloat64Number AdaptationList[256];
+	cmsUInt32Number IntentList[256];
+
+	memset(&Chain, 0, sizeof(GAMUTCHAIN)); 
+
+
+	if (nGamutPCSposition <= 0 || nGamutPCSposition > 255) {
+		cmsSignalError(ContextID, cmsERROR_RANGE, "Wrong position of PCS. 1..255 expected, %d found.", nGamutPCSposition);
+		return NULL;
+	}
+
+	hLab = cmsCreateLab4ProfileTHR(ContextID, NULL);
+	if (hLab == NULL) return NULL;
+
+
+	// The figure of merit. On matrix-shaper profiles, should be almost zero as
+	// the conversion is pretty exact. On LUT based profiles, different resolutions
+	// of input and output CLUT may result in differences. 
+
+	if (cmsIsMatrixShaper(hGamut)) {
+
+		Chain.Thereshold = 1.0;
+	}
+	else {
+		Chain.Thereshold = ERR_THERESHOLD;
+	}
+
+
+	// Create a copy of parameters
+	for (i=0; i < nGamutPCSposition; i++) {
+		ProfileList[i]    = hProfiles[i];
+		BPCList[i]        = BPC[i];
+		AdaptationList[i] = AdaptationStates[i];
+		IntentList[i]     = Intents[i];
+	}
+
+	// Fill Lab identity
+	ProfileList[nGamutPCSposition] = hLab;
+	BPCList[nGamutPCSposition] = 0;
+	AdaptationList[nGamutPCSposition] = 1.0;
+	Intents[nGamutPCSposition] = INTENT_RELATIVE_COLORIMETRIC;
+
+
+	ColorSpace  = cmsGetColorSpace(hGamut);  
+
+	nChannels   = cmsChannelsOf(ColorSpace);     
+	nGridpoints = _cmsReasonableGridpointsByColorspace(ColorSpace, cmsFLAGS_HIGHRESPRECALC);
+	dwFormat    = (CHANNELS_SH(nChannels)|BYTES_SH(2));
+
+	// 16 bits to Lab double
+	Chain.hInput = cmsCreateExtendedTransform(ContextID, 
+		nGamutPCSposition + 1, 
+		ProfileList, 
+		BPCList, 
+		Intents, 
+		AdaptationList, 
+		NULL, 0, 
+		dwFormat, TYPE_Lab_DBL, 
+		cmsFLAGS_NOCACHE);
+
+
+	// Does create the forward step. Lab double to device
+	dwFormat    = (CHANNELS_SH(nChannels)|BYTES_SH(2));
+	Chain.hForward = cmsCreateTransformTHR(ContextID,
+		hLab, TYPE_Lab_DBL, 
+		hGamut, dwFormat, 
+		INTENT_RELATIVE_COLORIMETRIC,
+		cmsFLAGS_NOCACHE);
+
+	// Does create the backwards step
+	Chain.hReverse = cmsCreateTransformTHR(ContextID, hGamut, dwFormat, 
+		hLab, TYPE_Lab_DBL,                                      
+		INTENT_RELATIVE_COLORIMETRIC,
+		cmsFLAGS_NOCACHE);
+
+
+	// All ok?
+	if (Chain.hForward && Chain.hReverse) {
+
+		// Go on, try to compute gamut LUT from PCS. This consist on a single channel containing 
+		// dE when doing a transform back and forth on the colorimetric intent. 
+
+		Gamut = cmsPipelineAlloc(ContextID, 3, 1);
+
+		if (Gamut != NULL) {
+
+			CLUT = cmsStageAllocCLut16bit(ContextID, nGridpoints, nChannels, 1, NULL);
+			cmsPipelineInsertStage(Gamut, cmsAT_BEGIN, CLUT);
+
+			cmsStageSampleCLut16bit(CLUT, GamutSampler, (void*) &Chain, 0);
+		}
+	}
+	else 
+		Gamut = NULL;   // Didn't work...
+
+	// Free all needed stuff.
+	if (Chain.hInput)   cmsDeleteTransform(Chain.hInput);
+	if (Chain.hForward) cmsDeleteTransform(Chain.hForward);
+	if (Chain.hReverse) cmsDeleteTransform(Chain.hReverse);
+	if (hLab) cmsCloseProfile(hLab);
+
+	// And return computed hull
+	return Gamut;
+}
+
+// Total Area Coverage estimation ----------------------------------------------------------------
+
+typedef struct {
+    cmsUInt32Number  nOutputChans;
+    cmsHTRANSFORM    hRoundTrip;               
+    cmsFloat32Number MaxTAC;
+    cmsFloat32Number MaxInput[cmsMAXCHANNELS];
+
+} cmsTACestimator;
+
+
+// This callback just accounts the maximum ink dropped in the given node. It does not populate any
+// memory, as the destination table is NULL. Its only purpose it to know the global maximum.
+static
+int EstimateTAC(register const cmsUInt16Number In[], register cmsUInt16Number Out[], register void * Cargo)
+{
+    cmsTACestimator* bp = (cmsTACestimator*) Cargo;
+    cmsFloat32Number RoundTrip[cmsMAXCHANNELS];
+    cmsUInt32Number i;
+    cmsFloat32Number Sum;
+
+    
+    // Evaluate the xform
+    cmsDoTransform(bp->hRoundTrip, In, RoundTrip, 1);
+    
+    // All all amounts of ink
+    for (Sum=0, i=0; i < bp ->nOutputChans; i++) 
+            Sum += RoundTrip[i];
+
+    // If above maximum, keep track of input values
+    if (Sum > bp ->MaxTAC) {
+
+            bp ->MaxTAC = Sum;
+
+            for (i=0; i < bp ->nOutputChans; i++) {
+                bp ->MaxInput[i] = In[i];
+            }
+    }
+    
+    return TRUE;
+
+    cmsUNUSED_PARAMETER(Out);
+}
+
+
+// Detect Total area coverage of the profile
+cmsFloat64Number CMSEXPORT cmsDetectTAC(cmsHPROFILE hProfile)
+{
+    cmsTACestimator bp;
+    cmsUInt32Number dwFormatter;
+    cmsUInt32Number GridPoints[MAX_INPUT_DIMENSIONS];        
+    cmsHPROFILE hLab;
+    cmsContext ContextID = cmsGetProfileContextID(hProfile);
+
+    // TAC only works on output profiles
+    if (cmsGetDeviceClass(hProfile) != cmsSigOutputClass) {        
+        return 0;       
+    }
+
+    // Create a fake formatter for result
+    dwFormatter = cmsFormatterForColorspaceOfProfile(hProfile, 4, TRUE);
+
+    bp.nOutputChans = T_CHANNELS(dwFormatter);
+    bp.MaxTAC = 0;    // Initial TAC is 0
+
+    //  for safety
+    if (bp.nOutputChans >= cmsMAXCHANNELS) return 0;
+
+    hLab = cmsCreateLab4ProfileTHR(ContextID, NULL);
+    if (hLab == NULL) return 0;
+    // Setup a roundtrip on perceptual intent in output profile for TAC estimation
+    bp.hRoundTrip = cmsCreateTransformTHR(ContextID, hLab, TYPE_Lab_16, 
+                                          hProfile, dwFormatter, INTENT_PERCEPTUAL, cmsFLAGS_NOOPTIMIZE|cmsFLAGS_NOCACHE);
+
+    cmsCloseProfile(hLab);  
+    if (bp.hRoundTrip == NULL) return 0;
+
+    // For L* we only need black and white. For C* we need many points
+    GridPoints[0] = 6;
+    GridPoints[1] = 74;
+    GridPoints[2] = 74; 
+
+    
+	if (!cmsSliceSpace16(3, GridPoints, EstimateTAC, &bp)) {
+		bp.MaxTAC = 0;
+	}
+
+    cmsDeleteTransform(bp.hRoundTrip);
+
+    // Results in %
+    return bp.MaxTAC;
+}
+
+
+// Carefully,  clamp on CIELab space.
+
+cmsBool CMSEXPORT cmsDesaturateLab(cmsCIELab* Lab, 
+                                   double amax, double amin,
+                                   double bmax, double bmin)
+{
+
+    // Whole Luma surface to zero
+
+    if (Lab -> L < 0) {
+
+        Lab-> L = Lab->a = Lab-> b = 0.0;
+        return FALSE;
+    }
+
+    // Clamp white, DISCARD HIGHLIGHTS. This is done
+    // in such way because icc spec doesn't allow the
+    // use of L>100 as a highlight means.
+
+    if (Lab->L > 100)
+        Lab -> L = 100;
+
+    // Check out gamut prism, on a, b faces
+
+    if (Lab -> a < amin || Lab->a > amax||
+        Lab -> b < bmin || Lab->b > bmax) {
+
+            cmsCIELCh LCh;
+            double h, slope;
+
+            // Falls outside a, b limits. Transports to LCh space,
+            // and then do the clipping
+
+
+            if (Lab -> a == 0.0) { // Is hue exactly 90?
+
+                // atan will not work, so clamp here
+                Lab -> b = Lab->b < 0 ? bmin : bmax;
+                return TRUE;
+            }
+
+            cmsLab2LCh(&LCh, Lab);
+
+            slope = Lab -> b / Lab -> a;
+            h = LCh.h;
+
+            // There are 4 zones
+
+            if ((h >= 0. && h < 45.) ||
+                (h >= 315 && h <= 360.)) {
+
+                    // clip by amax
+                    Lab -> a = amax;
+                    Lab -> b = amax * slope;
+            }
+            else
+                if (h >= 45. && h < 135.)
+                {
+                    // clip by bmax
+                    Lab -> b = bmax;
+                    Lab -> a = bmax / slope;
+                }
+                else
+                    if (h >= 135. && h < 225.) {
+                        // clip by amin
+                        Lab -> a = amin;
+                        Lab -> b = amin * slope;
+
+                    }
+                    else
+                        if (h >= 225. && h < 315.) {
+                            // clip by bmin
+                            Lab -> b = bmin;
+                            Lab -> a = bmin / slope;
+                        }
+                        else  {
+                            cmsSignalError(0, cmsERROR_RANGE, "Invalid angle");
+                            return FALSE;
+                        }
+
+    }
+
+    return TRUE;
+}
diff --git a/thirdparty/liblcms2/src/cmsintrp.c b/thirdparty/liblcms2/src/cmsintrp.c
new file mode 100644
index 00000000..9aced860
--- /dev/null
+++ b/thirdparty/liblcms2/src/cmsintrp.c
@@ -0,0 +1,1463 @@
+//---------------------------------------------------------------------------------
+//
+//  Little Color Management System
+//  Copyright (c) 1998-2010 Marti Maria Saguer
+//
+// Permission is hereby granted, free of charge, to any person obtaining 
+// a copy of this software and associated documentation files (the "Software"), 
+// to deal in the Software without restriction, including without limitation 
+// the rights to use, copy, modify, merge, publish, distribute, sublicense, 
+// and/or sell copies of the Software, and to permit persons to whom the Software 
+// is furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in 
+// all copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 
+// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO 
+// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE 
+// LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 
+// OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 
+// WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+//
+//---------------------------------------------------------------------------------
+//
+
+#include "lcms2_internal.h"
+
+// This module incorporates several interpolation routines, for 1 to 8 channels on input and
+// up to 65535 channels on output. The user may change those by using the interpolation plug-in
+
+// Interpolation routines by default
+static cmsInterpFunction DefaultInterpolatorsFactory(cmsUInt32Number nInputChannels, cmsUInt32Number nOutputChannels, cmsUInt32Number dwFlags);
+
+// This is the default factory
+static cmsInterpFnFactory Interpolators = DefaultInterpolatorsFactory;
+
+
+// Main plug-in entry
+cmsBool  _cmsRegisterInterpPlugin(cmsPluginBase* Data)
+{
+    cmsPluginInterpolation* Plugin = (cmsPluginInterpolation*) Data;
+
+    if (Data == NULL) {
+    
+        Interpolators = DefaultInterpolatorsFactory;
+        return TRUE;
+    }
+
+    // Set replacement functions
+    Interpolators = Plugin ->InterpolatorsFactory;  
+    return TRUE;
+}
+
+
+// Set the interpolation method
+
+cmsBool _cmsSetInterpolationRoutine(cmsInterpParams* p)
+{
+    // Invoke factory, possibly in the Plug-in
+    p ->Interpolation = Interpolators(p -> nInputs, p ->nOutputs, p ->dwFlags);
+
+    // If unsupported by the plug-in, go for the LittleCMS default. 
+    // If happens only if an extern plug-in is being used
+    if (p ->Interpolation.Lerp16 == NULL)
+        p ->Interpolation = DefaultInterpolatorsFactory(p ->nInputs, p ->nOutputs, p ->dwFlags);
+
+    // Check for valid interpolator (we just check one member of the union)
+    if (p ->Interpolation.Lerp16 == NULL) {    
+            return FALSE;
+    }
+    return TRUE;
+}
+
+
+// This function precalculates as many parameters as possible to speed up the interpolation.
+cmsInterpParams* _cmsComputeInterpParamsEx(cmsContext ContextID,
+                                           const cmsUInt32Number nSamples[], 
+                                           int InputChan, int OutputChan, 
+                                           const void *Table,
+                                           cmsUInt32Number dwFlags)
+{        
+    cmsInterpParams* p;
+    int i;
+    
+    // Check for maximum inputs
+    if (InputChan > MAX_INPUT_DIMENSIONS) {
+             cmsSignalError(ContextID, cmsERROR_RANGE, "Too many input channels (%d channels, max=%d)", InputChan, MAX_INPUT_DIMENSIONS);            
+            return NULL;
+    }
+
+    // Creates an empty object
+    p = (cmsInterpParams*) _cmsMallocZero(ContextID, sizeof(cmsInterpParams));
+    if (p == NULL) return NULL;
+
+    // Keep original parameters
+    p -> dwFlags  = dwFlags;
+    p -> nInputs  = InputChan;
+    p -> nOutputs = OutputChan;
+    p ->Table     = Table;
+    p ->ContextID  = ContextID;
+
+    // Fill samples per input direction and domain (which is number of nodes minus one)
+    for (i=0; i < InputChan; i++) {
+
+        p -> nSamples[i] = nSamples[i];
+        p -> Domain[i]   = nSamples[i] - 1;     
+    }
+
+    // Compute factors to apply to each component to index the grid array
+    p -> opta[0] = p -> nOutputs;  
+    for (i=1; i < InputChan; i++)
+        p ->opta[i] = p ->opta[i-1] * nSamples[InputChan-i];
+
+
+    if (!_cmsSetInterpolationRoutine(p)) {
+         cmsSignalError(ContextID, cmsERROR_UNKNOWN_EXTENSION, "Unsupported interpolation (%d->%d channels)", InputChan, OutputChan);            
+        _cmsFree(ContextID, p);
+        return NULL;
+    }
+
+    // All seems ok
+    return p;
+}
+
+
+// This one is a wrapper on the anterior, but assuming all directions have same number of nodes
+cmsInterpParams* _cmsComputeInterpParams(cmsContext ContextID, int nSamples, int InputChan, int OutputChan, const void* Table, cmsUInt32Number dwFlags)
+{
+    int i;
+    cmsUInt32Number Samples[MAX_INPUT_DIMENSIONS];
+
+    // Fill the auxiliar array
+    for (i=0; i < MAX_INPUT_DIMENSIONS; i++) 
+        Samples[i] = nSamples;
+
+    // Call the extended function
+    return _cmsComputeInterpParamsEx(ContextID, Samples, InputChan, OutputChan, Table, dwFlags);
+}
+
+
+// Free all associated memory
+void _cmsFreeInterpParams(cmsInterpParams* p)
+{
+    if (p != NULL) _cmsFree(p ->ContextID, p);
+}
+
+
+// Inline fixed point interpolation 
+cmsINLINE cmsUInt16Number LinearInterp(cmsS15Fixed16Number a, cmsS15Fixed16Number l, cmsS15Fixed16Number h)
+{
+    cmsUInt32Number dif = (cmsUInt32Number) (h - l) * a + 0x8000;           
+    dif = (dif >> 16) + l;        
+    return (cmsUInt16Number) (dif);
+}
+
+
+//  Linear interpolation (Fixed-point optimized)
+static
+void LinLerp1D(register const cmsUInt16Number Value[], 
+               register cmsUInt16Number Output[], 
+               register const cmsInterpParams* p)
+{
+    cmsUInt16Number y1, y0; 
+    int cell0, rest;
+    int val3;
+    const cmsUInt16Number* LutTable = (cmsUInt16Number*) p ->Table;
+
+    // if last value... 
+    if (Value[0] == 0xffff) {
+
+        Output[0] = LutTable[p -> Domain[0]];
+        return;
+    }
+
+    val3 = p -> Domain[0] * Value[0];
+    val3 = _cmsToFixedDomain(val3);    // To fixed 15.16
+
+    cell0 = FIXED_TO_INT(val3);             // Cell is 16 MSB bits
+    rest  = FIXED_REST_TO_INT(val3);        // Rest is 16 LSB bits
+
+    y0 = LutTable[cell0];
+    y1 = LutTable[cell0+1];
+
+    
+    Output[0] = LinearInterp(rest, y0, y1);
+}
+
+
+// Floating-point version of 1D interpolation
+static
+void LinLerp1Dfloat(const cmsFloat32Number Value[], 
+                    cmsFloat32Number Output[],  
+                    const cmsInterpParams* p)
+{
+       cmsFloat32Number y1, y0;
+       cmsFloat32Number val2, rest;
+       int cell0, cell1;
+       const cmsFloat32Number* LutTable = (cmsFloat32Number*) p ->Table;
+
+       // if last value...
+       if (Value[0] == 1.0) {
+           Output[0] = LutTable[p -> Domain[0]];
+           return;
+       }
+
+       val2 = p -> Domain[0] * Value[0];
+
+       cell0 = (int) floor(val2);
+       cell1 = (int) ceil(val2);
+
+       // Rest is 16 LSB bits
+       rest = val2 - cell0;
+
+       y0 = LutTable[cell0] ;
+       y1 = LutTable[cell1] ;
+
+       Output[0] = y0 + (y1 - y0) * rest;     
+}
+
+
+
+// Eval gray LUT having only one input channel 
+static
+void Eval1Input(register const cmsUInt16Number Input[], 
+                register cmsUInt16Number Output[], 
+                register const cmsInterpParams* p16)
+{
+       cmsS15Fixed16Number fk;
+       cmsS15Fixed16Number k0, k1, rk, K0, K1;
+       int v;
+       cmsUInt32Number OutChan;
+       const cmsUInt16Number* LutTable = (cmsUInt16Number*) p16 -> Table;
+
+       v = Input[0] * p16 -> Domain[0];
+       fk = _cmsToFixedDomain(v);
+
+       k0 = FIXED_TO_INT(fk);
+       rk = (cmsUInt16Number) FIXED_REST_TO_INT(fk);
+
+       k1 = k0 + (Input[0] != 0xFFFFU ? 1 : 0);
+
+       K0 = p16 -> opta[0] * k0;
+       K1 = p16 -> opta[0] * k1;
+
+       for (OutChan=0; OutChan < p16->nOutputs; OutChan++) {
+
+           Output[OutChan] = LinearInterp(rk, LutTable[K0+OutChan], LutTable[K1+OutChan]);
+       }
+}
+
+
+
+// Eval gray LUT having only one input channel 
+static
+void Eval1InputFloat(const cmsFloat32Number Value[], 
+                     cmsFloat32Number Output[], 
+                     const cmsInterpParams* p)
+{
+    cmsFloat32Number y1, y0;
+    cmsFloat32Number val2, rest;
+    int cell0, cell1;
+    cmsUInt32Number OutChan;
+    const cmsFloat32Number* LutTable = (cmsFloat32Number*) p ->Table; 
+
+        // if last value...
+       if (Value[0] == 1.0) {
+           Output[0] = LutTable[p -> Domain[0]];
+           return;
+       }
+
+       val2 = p -> Domain[0] * Value[0];
+
+       cell0 = (int) floor(val2);
+       cell1 = (int) ceil(val2);
+
+       // Rest is 16 LSB bits
+       rest = val2 - cell0;
+       
+       cell0 *= p -> opta[0];
+       cell1 *= p -> opta[0];
+
+       for (OutChan=0; OutChan < p->nOutputs; OutChan++) {
+
+            y0 = LutTable[cell0 + OutChan] ;
+            y1 = LutTable[cell1 + OutChan] ;
+
+            Output[OutChan] = y0 + (y1 - y0) * rest;              
+       }
+}
+
+// Bilinear interpolation (16 bits) - cmsFloat32Number version
+static
+void BilinearInterpFloat(const cmsFloat32Number Input[], 
+                         cmsFloat32Number Output[],                          
+                         const cmsInterpParams* p)
+
+{
+#   define LERP(a,l,h)    (cmsFloat32Number) ((l)+(((h)-(l))*(a)))
+#   define DENS(i,j)      (LutTable[(i)+(j)+OutChan])
+
+    const cmsFloat32Number* LutTable = (cmsFloat32Number*) p ->Table; 
+    cmsFloat32Number      px, py;
+    int        x0, y0,
+               X0, Y0, X1, Y1;
+    int        TotalOut, OutChan;
+    cmsFloat32Number      fx, fy,
+        d00, d01, d10, d11,
+        dx0, dx1,
+        dxy;
+    
+    TotalOut   = p -> nOutputs;
+    px = Input[0] * p->Domain[0];
+    py = Input[1] * p->Domain[1];
+
+    x0 = (int) _cmsQuickFloor(px); fx = px - (cmsFloat32Number) x0;
+    y0 = (int) _cmsQuickFloor(py); fy = py - (cmsFloat32Number) y0;
+    
+    X0 = p -> opta[1] * x0;
+    X1 = X0 + (Input[0] >= 1.0 ? 0 : p->opta[1]);
+
+    Y0 = p -> opta[0] * y0;
+    Y1 = Y0 + (Input[1] >= 1.0 ? 0 : p->opta[0]);
+   
+    for (OutChan = 0; OutChan < TotalOut; OutChan++) {
+       
+        d00 = DENS(X0, Y0);
+        d01 = DENS(X0, Y1);
+        d10 = DENS(X1, Y0);
+        d11 = DENS(X1, Y1);
+
+        dx0 = LERP(fx, d00, d10);
+        dx1 = LERP(fx, d01, d11);
+
+        dxy = LERP(fy, dx0, dx1);
+
+        Output[OutChan] = dxy;
+    }
+
+
+#   undef LERP
+#   undef DENS
+}
+
+// Bilinear interpolation (16 bits) - optimized version
+static
+void BilinearInterp16(register const cmsUInt16Number Input[], 
+                      register cmsUInt16Number Output[],
+                      register const cmsInterpParams* p)
+
+{
+#define DENS(i,j) (LutTable[(i)+(j)+OutChan])
+#define LERP(a,l,h)     (cmsUInt16Number) (l + ROUND_FIXED_TO_INT(((h-l)*a)))
+
+           const cmsUInt16Number* LutTable = (cmsUInt16Number*) p ->Table; 
+           int        OutChan, TotalOut;
+           cmsS15Fixed16Number    fx, fy;
+  register int        rx, ry;
+           int        x0, y0;
+  register int        X0, X1, Y0, Y1;
+           int        d00, d01, d10, d11,
+                      dx0, dx1,
+                      dxy;
+
+    TotalOut   = p -> nOutputs;
+
+    fx = _cmsToFixedDomain((int) Input[0] * p -> Domain[0]);
+    x0  = FIXED_TO_INT(fx);
+    rx  = FIXED_REST_TO_INT(fx);    // Rest in 0..1.0 domain
+
+
+    fy = _cmsToFixedDomain((int) Input[1] * p -> Domain[1]);
+    y0  = FIXED_TO_INT(fy);
+    ry  = FIXED_REST_TO_INT(fy);
+
+
+    X0 = p -> opta[1] * x0;
+    X1 = X0 + (Input[0] == 0xFFFFU ? 0 : p->opta[1]);
+
+    Y0 = p -> opta[0] * y0;
+    Y1 = Y0 + (Input[1] == 0xFFFFU ? 0 : p->opta[0]);
+   
+    for (OutChan = 0; OutChan < TotalOut; OutChan++) {
+
+        d00 = DENS(X0, Y0);
+        d01 = DENS(X0, Y1);
+        d10 = DENS(X1, Y0);
+        d11 = DENS(X1, Y1);
+
+        dx0 = LERP(rx, d00, d10);
+        dx1 = LERP(rx, d01, d11);
+
+        dxy = LERP(ry, dx0, dx1);
+
+        Output[OutChan] = (cmsUInt16Number) dxy;
+    }
+
+
+#   undef LERP
+#   undef DENS
+}
+
+
+// Trilinear interpolation (16 bits) - cmsFloat32Number version
+static
+void TrilinearInterpFloat(const cmsFloat32Number Input[], 
+                          cmsFloat32Number Output[],                          
+                          const cmsInterpParams* p)
+
+{
+#   define LERP(a,l,h)      (cmsFloat32Number) ((l)+(((h)-(l))*(a)))
+#   define DENS(i,j,k)      (LutTable[(i)+(j)+(k)+OutChan])
+
+    const cmsFloat32Number* LutTable = (cmsFloat32Number*) p ->Table; 
+    cmsFloat32Number      px, py, pz;
+    int        x0, y0, z0,
+               X0, Y0, Z0, X1, Y1, Z1;
+    int        TotalOut, OutChan;
+    cmsFloat32Number      fx, fy, fz,
+        d000, d001, d010, d011,
+        d100, d101, d110, d111,
+        dx00, dx01, dx10, dx11,
+        dxy0, dxy1, dxyz;
+    
+    TotalOut   = p -> nOutputs;
+    
+    // We need some clipping here
+    px = Input[0];
+    py = Input[1];
+    pz = Input[2];
+
+    if (px < 0) px = 0; 
+    if (px > 1) px = 1;
+    if (py < 0) py = 0; 
+    if (py > 1) py = 1;
+    if (pz < 0) pz = 0; 
+    if (pz > 1) pz = 1;
+
+    px *= p->Domain[0];
+    py *= p->Domain[1];
+    pz *= p->Domain[2];
+
+    x0 = (int) _cmsQuickFloor(px); fx = px - (cmsFloat32Number) x0;
+    y0 = (int) _cmsQuickFloor(py); fy = py - (cmsFloat32Number) y0;
+    z0 = (int) _cmsQuickFloor(pz); fz = pz - (cmsFloat32Number) z0;
+    
+    X0 = p -> opta[2] * x0;
+    X1 = X0 + (Input[0] >= 1.0 ? 0 : p->opta[2]);
+
+    Y0 = p -> opta[1] * y0;
+    Y1 = Y0 + (Input[1] >= 1.0 ? 0 : p->opta[1]);
+   
+    Z0 = p -> opta[0] * z0;
+    Z1 = Z0 + (Input[2] >= 1.0 ? 0 : p->opta[0]);
+   
+    for (OutChan = 0; OutChan < TotalOut; OutChan++) {
+
+        d000 = DENS(X0, Y0, Z0);
+        d001 = DENS(X0, Y0, Z1);
+        d010 = DENS(X0, Y1, Z0);
+        d011 = DENS(X0, Y1, Z1);
+
+        d100 = DENS(X1, Y0, Z0);
+        d101 = DENS(X1, Y0, Z1);
+        d110 = DENS(X1, Y1, Z0);
+        d111 = DENS(X1, Y1, Z1);
+
+
+        dx00 = LERP(fx, d000, d100);
+        dx01 = LERP(fx, d001, d101);
+        dx10 = LERP(fx, d010, d110);
+        dx11 = LERP(fx, d011, d111);
+
+        dxy0 = LERP(fy, dx00, dx10);
+        dxy1 = LERP(fy, dx01, dx11);
+
+        dxyz = LERP(fz, dxy0, dxy1);
+
+        Output[OutChan] = dxyz;
+    }
+
+
+#   undef LERP
+#   undef DENS
+}
+
+// Trilinear interpolation (16 bits) - optimized version
+static
+void TrilinearInterp16(register const cmsUInt16Number Input[], 
+                       register cmsUInt16Number Output[],
+                       register const cmsInterpParams* p)
+
+{
+#define DENS(i,j,k) (LutTable[(i)+(j)+(k)+OutChan])
+#define LERP(a,l,h)     (cmsUInt16Number) (l + ROUND_FIXED_TO_INT(((h-l)*a)))
+
+           const cmsUInt16Number* LutTable = (cmsUInt16Number*) p ->Table; 
+           int        OutChan, TotalOut;
+           cmsS15Fixed16Number    fx, fy, fz;
+  register int        rx, ry, rz;
+           int        x0, y0, z0;
+  register int        X0, X1, Y0, Y1, Z0, Z1;
+           int        d000, d001, d010, d011,
+                      d100, d101, d110, d111,
+                      dx00, dx01, dx10, dx11,
+                      dxy0, dxy1, dxyz;
+
+    TotalOut   = p -> nOutputs;
+
+    fx = _cmsToFixedDomain((int) Input[0] * p -> Domain[0]);
+    x0  = FIXED_TO_INT(fx);
+    rx  = FIXED_REST_TO_INT(fx);    // Rest in 0..1.0 domain
+
+
+    fy = _cmsToFixedDomain((int) Input[1] * p -> Domain[1]);
+    y0  = FIXED_TO_INT(fy);
+    ry  = FIXED_REST_TO_INT(fy);
+
+    fz = _cmsToFixedDomain((int) Input[2] * p -> Domain[2]);
+    z0 = FIXED_TO_INT(fz);
+    rz = FIXED_REST_TO_INT(fz);
+
+
+    X0 = p -> opta[2] * x0;
+    X1 = X0 + (Input[0] == 0xFFFFU ? 0 : p->opta[2]);
+
+    Y0 = p -> opta[1] * y0;
+    Y1 = Y0 + (Input[1] == 0xFFFFU ? 0 : p->opta[1]);
+   
+    Z0 = p -> opta[0] * z0;
+    Z1 = Z0 + (Input[2] == 0xFFFFU ? 0 : p->opta[0]);
+
+    for (OutChan = 0; OutChan < TotalOut; OutChan++) {
+
+        d000 = DENS(X0, Y0, Z0);
+        d001 = DENS(X0, Y0, Z1);
+        d010 = DENS(X0, Y1, Z0);
+        d011 = DENS(X0, Y1, Z1);
+
+        d100 = DENS(X1, Y0, Z0);
+        d101 = DENS(X1, Y0, Z1);
+        d110 = DENS(X1, Y1, Z0);
+        d111 = DENS(X1, Y1, Z1);
+
+
+        dx00 = LERP(rx, d000, d100);
+        dx01 = LERP(rx, d001, d101);
+        dx10 = LERP(rx, d010, d110);
+        dx11 = LERP(rx, d011, d111);
+
+        dxy0 = LERP(ry, dx00, dx10);
+        dxy1 = LERP(ry, dx01, dx11);
+
+        dxyz = LERP(rz, dxy0, dxy1);
+
+        Output[OutChan] = (cmsUInt16Number) dxyz;
+    }
+
+
+#   undef LERP
+#   undef DENS
+}
+
+
+// Tetrahedral interpolation, using Sakamoto algorithm. 
+#define DENS(i,j,k) (LutTable[(i)+(j)+(k)+OutChan])
+static
+void TetrahedralInterpFloat(const cmsFloat32Number Input[], 
+                            cmsFloat32Number Output[],                          
+                            const cmsInterpParams* p)
+{
+    const cmsFloat32Number* LutTable = (cmsFloat32Number*) p -> Table; 
+    cmsFloat32Number     px, py, pz;
+    int        x0, y0, z0,
+               X0, Y0, Z0, X1, Y1, Z1;
+    cmsFloat32Number     rx, ry, rz;
+    cmsFloat32Number     c0, c1=0, c2=0, c3=0;
+    int                  OutChan, TotalOut;
+
+    TotalOut   = p -> nOutputs;
+
+    // We need some clipping here
+    px = Input[0];
+    py = Input[1];
+    pz = Input[2];
+
+    if (px < 0) px = 0; 
+    if (px > 1) px = 1;
+    if (py < 0) py = 0; 
+    if (py > 1) py = 1;
+    if (pz < 0) pz = 0; 
+    if (pz > 1) pz = 1;
+
+    px *= p->Domain[0];
+    py *= p->Domain[1];
+    pz *= p->Domain[2];
+
+    x0 = (int) _cmsQuickFloor(px); rx = (px - (cmsFloat32Number) x0);
+    y0 = (int) _cmsQuickFloor(py); ry = (py - (cmsFloat32Number) y0);
+    z0 = (int) _cmsQuickFloor(pz); rz = (pz - (cmsFloat32Number) z0);
+
+
+    X0 = p -> opta[2] * x0;
+    X1 = X0 + (Input[0] >= 1.0 ? 0 : p->opta[2]);
+
+    Y0 = p -> opta[1] * y0;
+    Y1 = Y0 + (Input[1] >= 1.0 ? 0 : p->opta[1]);
+   
+    Z0 = p -> opta[0] * z0;
+    Z1 = Z0 + (Input[2] >= 1.0 ? 0 : p->opta[0]);
+   
+    for (OutChan=0; OutChan < TotalOut; OutChan++) {
+
+       // These are the 6 Tetrahedral
+
+        c0 = DENS(X0, Y0, Z0);
+
+        if (rx >= ry && ry >= rz) {
+
+            c1 = DENS(X1, Y0, Z0) - c0;
+            c2 = DENS(X1, Y1, Z0) - DENS(X1, Y0, Z0);
+            c3 = DENS(X1, Y1, Z1) - DENS(X1, Y1, Z0);
+
+        }
+        else
+            if (rx >= rz && rz >= ry) {            
+
+                c1 = DENS(X1, Y0, Z0) - c0;
+                c2 = DENS(X1, Y1, Z1) - DENS(X1, Y0, Z1);
+                c3 = DENS(X1, Y0, Z1) - DENS(X1, Y0, Z0);
+
+            }
+            else
+                if (rz >= rx && rx >= ry) {
+
+                    c1 = DENS(X1, Y0, Z1) - DENS(X0, Y0, Z1);
+                    c2 = DENS(X1, Y1, Z1) - DENS(X1, Y0, Z1);
+                    c3 = DENS(X0, Y0, Z1) - c0;                            
+
+                }
+                else
+                    if (ry >= rx && rx >= rz) {
+
+                        c1 = DENS(X1, Y1, Z0) - DENS(X0, Y1, Z0);
+                        c2 = DENS(X0, Y1, Z0) - c0;
+                        c3 = DENS(X1, Y1, Z1) - DENS(X1, Y1, Z0);
+
+                    }
+                    else
+                        if (ry >= rz && rz >= rx) {
+
+                            c1 = DENS(X1, Y1, Z1) - DENS(X0, Y1, Z1);
+                            c2 = DENS(X0, Y1, Z0) - c0;
+                            c3 = DENS(X0, Y1, Z1) - DENS(X0, Y1, Z0);
+
+                        }
+                        else
+                            if (rz >= ry && ry >= rx) {             
+
+                                c1 = DENS(X1, Y1, Z1) - DENS(X0, Y1, Z1);
+                                c2 = DENS(X0, Y1, Z1) - DENS(X0, Y0, Z1);
+                                c3 = DENS(X0, Y0, Z1) - c0;
+
+                            }
+                            else  {
+                                c1 = c2 = c3 = 0;                               
+                            }
+
+       Output[OutChan] = c0 + c1 * rx + c2 * ry + c3 * rz;
+       }
+
+}
+
+#undef DENS
+
+
+
+#define DENS(i,j,k) (LutTable[(i)+(j)+(k)+OutChan])
+
+static
+void TetrahedralInterp16(register const cmsUInt16Number Input[],
+                         register cmsUInt16Number Output[],
+                         register const cmsInterpParams* p)
+{
+    const cmsUInt16Number* LutTable = (cmsUInt16Number*) p -> Table;
+    cmsS15Fixed16Number    fx, fy, fz;
+    cmsS15Fixed16Number    rx, ry, rz;
+    int                    x0, y0, z0;
+    cmsS15Fixed16Number    c0, c1, c2, c3, Rest;       
+    cmsUInt32Number        OutChan;
+    cmsS15Fixed16Number    X0, X1, Y0, Y1, Z0, Z1;
+    cmsUInt32Number        TotalOut = p -> nOutputs;
+    
+
+    fx  = _cmsToFixedDomain((int) Input[0] * p -> Domain[0]);
+    fy  = _cmsToFixedDomain((int) Input[1] * p -> Domain[1]);
+    fz  = _cmsToFixedDomain((int) Input[2] * p -> Domain[2]);
+
+    x0  = FIXED_TO_INT(fx);
+    y0  = FIXED_TO_INT(fy); 
+    z0  = FIXED_TO_INT(fz);
+
+    rx  = FIXED_REST_TO_INT(fx);   
+    ry  = FIXED_REST_TO_INT(fy);      
+    rz  = FIXED_REST_TO_INT(fz);
+
+    X0 = p -> opta[2] * x0;
+    X1 = X0 + (Input[0] == 0xFFFFU ? 0 : p->opta[2]);
+
+    Y0 = p -> opta[1] * y0;
+    Y1 = Y0 + (Input[1] == 0xFFFFU ? 0 : p->opta[1]);
+
+    Z0 = p -> opta[0] * z0;
+    Z1 = Z0 + (Input[2] == 0xFFFFU ? 0 : p->opta[0]);
+
+    // These are the 6 Tetrahedral
+    for (OutChan=0; OutChan < TotalOut; OutChan++) {
+
+        c0 = DENS(X0, Y0, Z0);
+
+        if (rx >= ry && ry >= rz) {
+
+            c1 = DENS(X1, Y0, Z0) - c0;
+            c2 = DENS(X1, Y1, Z0) - DENS(X1, Y0, Z0);
+            c3 = DENS(X1, Y1, Z1) - DENS(X1, Y1, Z0);
+
+        }
+        else
+            if (rx >= rz && rz >= ry) {            
+
+                c1 = DENS(X1, Y0, Z0) - c0;
+                c2 = DENS(X1, Y1, Z1) - DENS(X1, Y0, Z1);
+                c3 = DENS(X1, Y0, Z1) - DENS(X1, Y0, Z0);
+
+            }
+            else
+                if (rz >= rx && rx >= ry) {
+
+                    c1 = DENS(X1, Y0, Z1) - DENS(X0, Y0, Z1);
+                    c2 = DENS(X1, Y1, Z1) - DENS(X1, Y0, Z1);
+                    c3 = DENS(X0, Y0, Z1) - c0;                            
+
+                }
+                else
+                    if (ry >= rx && rx >= rz) {
+
+                        c1 = DENS(X1, Y1, Z0) - DENS(X0, Y1, Z0);
+                        c2 = DENS(X0, Y1, Z0) - c0;
+                        c3 = DENS(X1, Y1, Z1) - DENS(X1, Y1, Z0);
+
+                    }
+                    else
+                        if (ry >= rz && rz >= rx) {
+
+                            c1 = DENS(X1, Y1, Z1) - DENS(X0, Y1, Z1);
+                            c2 = DENS(X0, Y1, Z0) - c0;
+                            c3 = DENS(X0, Y1, Z1) - DENS(X0, Y1, Z0);
+
+                        }
+                        else
+                            if (rz >= ry && ry >= rx) {             
+
+                                c1 = DENS(X1, Y1, Z1) - DENS(X0, Y1, Z1);
+                                c2 = DENS(X0, Y1, Z1) - DENS(X0, Y0, Z1);
+                                c3 = DENS(X0, Y0, Z1) - c0;
+
+                            }
+                            else  {
+                                c1 = c2 = c3 = 0;                               
+                            }
+
+                            Rest = c1 * rx + c2 * ry + c3 * rz;                
+
+                            Output[OutChan] = (cmsUInt16Number) c0 + ROUND_FIXED_TO_INT(_cmsToFixedDomain(Rest));
+    }
+
+}
+#undef DENS
+
+
+#define DENS(i,j,k) (LutTable[(i)+(j)+(k)+OutChan])
+static
+void Eval4Inputs(register const cmsUInt16Number Input[], 
+                     register cmsUInt16Number Output[], 
+                     register const cmsInterpParams* p16)
+{                        
+    const cmsUInt16Number* LutTable = (cmsUInt16Number*) p16 -> Table; 
+    cmsS15Fixed16Number fk;
+    cmsS15Fixed16Number k0, rk;
+    int K0, K1;
+    cmsS15Fixed16Number    fx, fy, fz;
+    cmsS15Fixed16Number    rx, ry, rz;
+    int                    x0, y0, z0;
+    cmsS15Fixed16Number    X0, X1, Y0, Y1, Z0, Z1;
+    cmsUInt32Number i;
+    cmsS15Fixed16Number    c0, c1, c2, c3, Rest;       
+    cmsUInt32Number        OutChan;
+    cmsUInt16Number        Tmp1[MAX_STAGE_CHANNELS], Tmp2[MAX_STAGE_CHANNELS];
+
+
+    fk  = _cmsToFixedDomain((int) Input[0] * p16 -> Domain[0]);
+    fx  = _cmsToFixedDomain((int) Input[1] * p16 -> Domain[1]);
+    fy  = _cmsToFixedDomain((int) Input[2] * p16 -> Domain[2]);
+    fz  = _cmsToFixedDomain((int) Input[3] * p16 -> Domain[3]);
+
+    k0  = FIXED_TO_INT(fk);     
+    x0  = FIXED_TO_INT(fx);
+    y0  = FIXED_TO_INT(fy); 
+    z0  = FIXED_TO_INT(fz);
+
+    rk  = FIXED_REST_TO_INT(fk);
+    rx  = FIXED_REST_TO_INT(fx);   
+    ry  = FIXED_REST_TO_INT(fy);      
+    rz  = FIXED_REST_TO_INT(fz);
+
+    K0 = p16 -> opta[3] * k0;
+    K1 = K0 + (Input[0] == 0xFFFFU ? 0 : p16->opta[3]);
+
+    X0 = p16 -> opta[2] * x0;
+    X1 = X0 + (Input[1] == 0xFFFFU ? 0 : p16->opta[2]);
+
+    Y0 = p16 -> opta[1] * y0;
+    Y1 = Y0 + (Input[2] == 0xFFFFU ? 0 : p16->opta[1]);
+
+    Z0 = p16 -> opta[0] * z0;
+    Z1 = Z0 + (Input[3] == 0xFFFFU ? 0 : p16->opta[0]);
+
+    LutTable = (cmsUInt16Number*) p16 -> Table;
+    LutTable += K0;
+
+    for (OutChan=0; OutChan < p16 -> nOutputs; OutChan++) {
+
+        c0 = DENS(X0, Y0, Z0);
+
+        if (rx >= ry && ry >= rz) {
+
+            c1 = DENS(X1, Y0, Z0) - c0;
+            c2 = DENS(X1, Y1, Z0) - DENS(X1, Y0, Z0);
+            c3 = DENS(X1, Y1, Z1) - DENS(X1, Y1, Z0);
+
+        }
+        else
+            if (rx >= rz && rz >= ry) {            
+
+                c1 = DENS(X1, Y0, Z0) - c0;
+                c2 = DENS(X1, Y1, Z1) - DENS(X1, Y0, Z1);
+                c3 = DENS(X1, Y0, Z1) - DENS(X1, Y0, Z0);
+
+            }
+            else
+                if (rz >= rx && rx >= ry) {
+
+                    c1 = DENS(X1, Y0, Z1) - DENS(X0, Y0, Z1);
+                    c2 = DENS(X1, Y1, Z1) - DENS(X1, Y0, Z1);
+                    c3 = DENS(X0, Y0, Z1) - c0;                            
+
+                }
+                else
+                    if (ry >= rx && rx >= rz) {
+
+                        c1 = DENS(X1, Y1, Z0) - DENS(X0, Y1, Z0);
+                        c2 = DENS(X0, Y1, Z0) - c0;
+                        c3 = DENS(X1, Y1, Z1) - DENS(X1, Y1, Z0);
+
+                    }
+                    else
+                        if (ry >= rz && rz >= rx) {
+
+                            c1 = DENS(X1, Y1, Z1) - DENS(X0, Y1, Z1);
+                            c2 = DENS(X0, Y1, Z0) - c0;
+                            c3 = DENS(X0, Y1, Z1) - DENS(X0, Y1, Z0);
+
+                        }
+                        else
+                            if (rz >= ry && ry >= rx) {             
+
+                                c1 = DENS(X1, Y1, Z1) - DENS(X0, Y1, Z1);
+                                c2 = DENS(X0, Y1, Z1) - DENS(X0, Y0, Z1);
+                                c3 = DENS(X0, Y0, Z1) - c0;
+
+                            }
+                            else  {
+                                c1 = c2 = c3 = 0;                               
+                            }
+
+                            Rest = c1 * rx + c2 * ry + c3 * rz;                
+
+                            Tmp1[OutChan] = (cmsUInt16Number) c0 + ROUND_FIXED_TO_INT(_cmsToFixedDomain(Rest));
+    }
+
+
+    LutTable = (cmsUInt16Number*) p16 -> Table;
+    LutTable += K1;
+
+    for (OutChan=0; OutChan < p16 -> nOutputs; OutChan++) {
+
+        c0 = DENS(X0, Y0, Z0);
+
+        if (rx >= ry && ry >= rz) {
+
+            c1 = DENS(X1, Y0, Z0) - c0;
+            c2 = DENS(X1, Y1, Z0) - DENS(X1, Y0, Z0);
+            c3 = DENS(X1, Y1, Z1) - DENS(X1, Y1, Z0);
+
+        }
+        else
+            if (rx >= rz && rz >= ry) {            
+
+                c1 = DENS(X1, Y0, Z0) - c0;
+                c2 = DENS(X1, Y1, Z1) - DENS(X1, Y0, Z1);
+                c3 = DENS(X1, Y0, Z1) - DENS(X1, Y0, Z0);
+
+            }
+            else
+                if (rz >= rx && rx >= ry) {
+
+                    c1 = DENS(X1, Y0, Z1) - DENS(X0, Y0, Z1);
+                    c2 = DENS(X1, Y1, Z1) - DENS(X1, Y0, Z1);
+                    c3 = DENS(X0, Y0, Z1) - c0;                            
+
+                }
+                else
+                    if (ry >= rx && rx >= rz) {
+
+                        c1 = DENS(X1, Y1, Z0) - DENS(X0, Y1, Z0);
+                        c2 = DENS(X0, Y1, Z0) - c0;
+                        c3 = DENS(X1, Y1, Z1) - DENS(X1, Y1, Z0);
+
+                    }
+                    else
+                        if (ry >= rz && rz >= rx) {
+
+                            c1 = DENS(X1, Y1, Z1) - DENS(X0, Y1, Z1);
+                            c2 = DENS(X0, Y1, Z0) - c0;
+                            c3 = DENS(X0, Y1, Z1) - DENS(X0, Y1, Z0);
+
+                        }
+                        else
+                            if (rz >= ry && ry >= rx) {             
+
+                                c1 = DENS(X1, Y1, Z1) - DENS(X0, Y1, Z1);
+                                c2 = DENS(X0, Y1, Z1) - DENS(X0, Y0, Z1);
+                                c3 = DENS(X0, Y0, Z1) - c0;
+
+                            }
+                            else  {
+                                c1 = c2 = c3 = 0;                               
+                            }
+
+                            Rest = c1 * rx + c2 * ry + c3 * rz;                
+
+                            Tmp2[OutChan] = (cmsUInt16Number) c0 + ROUND_FIXED_TO_INT(_cmsToFixedDomain(Rest));
+    }
+
+
+
+    for (i=0; i < p16 -> nOutputs; i++) {
+        Output[i] = LinearInterp(rk, Tmp1[i], Tmp2[i]);              
+    }
+}
+#undef DENS
+
+
+// For more that 3 inputs (i.e., CMYK)
+// evaluate two 3-dimensional interpolations and then linearly interpolate between them.
+
+
+static
+void Eval4InputsFloat(const cmsFloat32Number Input[], 
+                      cmsFloat32Number Output[], 
+                      const cmsInterpParams* p)
+{                             
+       const cmsFloat32Number* LutTable = (cmsFloat32Number*) p -> Table; 
+       cmsFloat32Number rest;
+       cmsFloat32Number pk;
+       int k0, K0, K1;
+       const cmsFloat32Number* T;
+       cmsUInt32Number i;
+       cmsFloat32Number Tmp1[MAX_STAGE_CHANNELS], Tmp2[MAX_STAGE_CHANNELS];
+       cmsInterpParams p1;
+
+       
+       pk = Input[0] * p->Domain[0];    
+       k0 = _cmsQuickFloor(pk); 
+       rest = pk - (cmsFloat32Number) k0;
+
+       K0 = p -> opta[3] * k0;
+       K1 = K0 + (Input[0] >= 1.0 ? 0 : p->opta[3]);
+
+       p1 = *p;
+       memmove(&p1.Domain[0], &p ->Domain[1], 3*sizeof(cmsUInt32Number));
+       
+       T = LutTable + K0;
+       p1.Table = T;
+
+       TetrahedralInterpFloat(Input + 1,  Tmp1, &p1);
+
+       T = LutTable + K1;
+       p1.Table = T;
+       TetrahedralInterpFloat(Input + 1,  Tmp2, &p1);
+
+       for (i=0; i < p -> nOutputs; i++)
+       {
+              cmsFloat32Number y0 = Tmp1[i];
+              cmsFloat32Number y1 = Tmp2[i];
+
+              Output[i] = y0 + (y1 - y0) * rest;               
+       }
+}
+
+
+static
+void Eval5Inputs(register const cmsUInt16Number Input[], 
+                 register cmsUInt16Number Output[], 
+
+                 register const cmsInterpParams* p16)
+{       
+       const cmsUInt16Number* LutTable = (cmsUInt16Number*) p16 -> Table;  
+       cmsS15Fixed16Number fk;
+       cmsS15Fixed16Number k0, rk;
+       int K0, K1;
+       const cmsUInt16Number* T;
+       cmsUInt32Number i;
+       cmsUInt16Number Tmp1[MAX_STAGE_CHANNELS], Tmp2[MAX_STAGE_CHANNELS];
+       cmsInterpParams p1;
+
+       
+       fk = _cmsToFixedDomain((cmsS15Fixed16Number) Input[0] * p16 -> Domain[0]);
+       k0 = FIXED_TO_INT(fk);
+       rk = FIXED_REST_TO_INT(fk);
+
+       K0 = p16 -> opta[4] * k0;
+       K1 = p16 -> opta[4] * (k0 + (Input[0] != 0xFFFFU ? 1 : 0));
+
+       p1 = *p16;
+       memmove(&p1.Domain[0], &p16 ->Domain[1], 4*sizeof(cmsUInt32Number));
+
+       T = LutTable + K0;
+       p1.Table = T;
+
+       Eval4Inputs(Input + 1, Tmp1, &p1);
+
+       T = LutTable + K1;
+       p1.Table = T;
+
+       Eval4Inputs(Input + 1, Tmp2, &p1);
+
+       for (i=0; i < p16 -> nOutputs; i++) {
+
+              Output[i] = LinearInterp(rk, Tmp1[i], Tmp2[i]);                         
+       }
+
+}
+
+
+static
+void Eval5InputsFloat(const cmsFloat32Number Input[], 
+                      cmsFloat32Number Output[], 
+                      const cmsInterpParams* p)
+{                             
+       const cmsFloat32Number* LutTable = (cmsFloat32Number*) p -> Table; 
+       cmsFloat32Number rest;
+       cmsFloat32Number pk;
+       int k0, K0, K1;
+       const cmsFloat32Number* T;
+       cmsUInt32Number i;
+       cmsFloat32Number Tmp1[MAX_STAGE_CHANNELS], Tmp2[MAX_STAGE_CHANNELS];
+       cmsInterpParams p1;
+
+       pk = Input[0] * p->Domain[0];    
+       k0 = _cmsQuickFloor(pk); 
+       rest = pk - (cmsFloat32Number) k0;
+
+       K0 = p -> opta[4] * k0;
+       K1 = K0 + (Input[0] >= 1.0 ? 0 : p->opta[4]);
+
+       p1 = *p;
+       memmove(&p1.Domain[0], &p ->Domain[1], 4*sizeof(cmsUInt32Number));
+
+       T = LutTable + K0;
+       p1.Table = T;
+       
+       Eval4InputsFloat(Input + 1,  Tmp1, &p1);
+      
+       T = LutTable + K1;
+       p1.Table = T;
+
+       Eval4InputsFloat(Input + 1,  Tmp2, &p1);
+      
+       for (i=0; i < p -> nOutputs; i++) {
+
+              cmsFloat32Number y0 = Tmp1[i];
+              cmsFloat32Number y1 = Tmp2[i];
+
+              Output[i] = y0 + (y1 - y0) * rest;             
+       }
+}
+
+
+
+static
+void Eval6Inputs(register const cmsUInt16Number Input[], 
+                 register cmsUInt16Number Output[], 
+                 register const cmsInterpParams* p16)
+{       
+       const cmsUInt16Number* LutTable = (cmsUInt16Number*) p16 -> Table; 
+       cmsS15Fixed16Number fk;
+       cmsS15Fixed16Number k0, rk;
+       int K0, K1;
+       const cmsUInt16Number* T;
+       cmsUInt32Number i;
+       cmsUInt16Number Tmp1[MAX_STAGE_CHANNELS], Tmp2[MAX_STAGE_CHANNELS];
+       cmsInterpParams p1;
+       
+       fk = _cmsToFixedDomain((cmsS15Fixed16Number) Input[0] * p16 -> Domain[0]);
+       k0 = FIXED_TO_INT(fk);
+       rk = FIXED_REST_TO_INT(fk);
+
+       K0 = p16 -> opta[5] * k0;
+       K1 = p16 -> opta[5] * (k0 + (Input[0] != 0xFFFFU ? 1 : 0));
+
+       p1 = *p16;
+       memmove(&p1.Domain[0], &p16 ->Domain[1], 5*sizeof(cmsUInt32Number));
+
+       T = LutTable + K0;
+       p1.Table = T;
+
+       Eval5Inputs(Input + 1, Tmp1, &p1);
+
+       T = LutTable + K1;
+       p1.Table = T;
+
+       Eval5Inputs(Input + 1, Tmp2, &p1);
+
+       for (i=0; i < p16 -> nOutputs; i++) {
+
+              Output[i] = LinearInterp(rk, Tmp1[i], Tmp2[i]);
+       }
+
+}
+
+
+static
+void Eval6InputsFloat(const cmsFloat32Number Input[], 
+                      cmsFloat32Number Output[], 
+                      const cmsInterpParams* p)
+{                             
+       const cmsFloat32Number* LutTable = (cmsFloat32Number*) p -> Table; 
+       cmsFloat32Number rest;
+       cmsFloat32Number pk;
+       int k0, K0, K1;
+       const cmsFloat32Number* T;
+       cmsUInt32Number i;
+       cmsFloat32Number Tmp1[MAX_STAGE_CHANNELS], Tmp2[MAX_STAGE_CHANNELS];
+       cmsInterpParams p1;
+
+       pk = Input[0] * p->Domain[0];    
+       k0 = _cmsQuickFloor(pk); 
+       rest = pk - (cmsFloat32Number) k0;
+
+       K0 = p -> opta[5] * k0;
+       K1 = K0 + (Input[0] >= 1.0 ? 0 : p->opta[5]);
+
+       p1 = *p;
+       memmove(&p1.Domain[0], &p ->Domain[1], 5*sizeof(cmsUInt32Number));
+       
+       T = LutTable + K0;
+       p1.Table = T;
+  
+       Eval5InputsFloat(Input + 1,  Tmp1, &p1);
+      
+       T = LutTable + K1;
+       p1.Table = T;
+
+       Eval5InputsFloat(Input + 1,  Tmp2, &p1);
+      
+       for (i=0; i < p -> nOutputs; i++) {
+
+              cmsFloat32Number y0 = Tmp1[i];
+              cmsFloat32Number y1 = Tmp2[i];
+
+              Output[i] = y0 + (y1 - y0) * rest;                   
+       }
+}
+
+
+static
+void Eval7Inputs(register const cmsUInt16Number Input[], 
+                 register cmsUInt16Number Output[], 
+                 register const cmsInterpParams* p16)
+{       
+       const cmsUInt16Number* LutTable = (cmsUInt16Number*) p16 -> Table; 
+       cmsS15Fixed16Number fk;
+       cmsS15Fixed16Number k0, rk;
+       int K0, K1;
+       const cmsUInt16Number* T;
+       cmsUInt32Number i;
+       cmsUInt16Number Tmp1[MAX_STAGE_CHANNELS], Tmp2[MAX_STAGE_CHANNELS];
+       cmsInterpParams p1;
+
+       
+       fk = _cmsToFixedDomain((cmsS15Fixed16Number) Input[0] * p16 -> Domain[0]);
+       k0 = FIXED_TO_INT(fk);
+       rk = FIXED_REST_TO_INT(fk);
+
+       K0 = p16 -> opta[6] * k0;
+       K1 = p16 -> opta[6] * (k0 + (Input[0] != 0xFFFFU ? 1 : 0));
+
+       p1 = *p16;
+       memmove(&p1.Domain[0], &p16 ->Domain[1], 5*sizeof(cmsUInt32Number));
+       
+       T = LutTable + K0;
+       p1.Table = T;
+
+       Eval6Inputs(Input + 1, Tmp1, &p1);
+
+       T = LutTable + K1;
+       p1.Table = T;
+
+       Eval6Inputs(Input + 1, Tmp2, &p1);
+
+       for (i=0; i < p16 -> nOutputs; i++) {
+              Output[i] = LinearInterp(rk, Tmp1[i], Tmp2[i]);
+       }
+}
+
+
+static
+void Eval7InputsFloat(const cmsFloat32Number Input[], 
+                      cmsFloat32Number Output[], 
+                      const cmsInterpParams* p)
+{                             
+       const cmsFloat32Number* LutTable = (cmsFloat32Number*) p -> Table; 
+       cmsFloat32Number rest;
+       cmsFloat32Number pk;
+       int k0, K0, K1;
+       const cmsFloat32Number* T;
+       cmsUInt32Number i;
+       cmsFloat32Number Tmp1[MAX_STAGE_CHANNELS], Tmp2[MAX_STAGE_CHANNELS];
+       cmsInterpParams p1;
+
+       pk = Input[0] * p->Domain[0];    
+       k0 = _cmsQuickFloor(pk); 
+       rest = pk - (cmsFloat32Number) k0;
+
+       K0 = p -> opta[6] * k0;
+       K1 = K0 + (Input[0] >= 1.0 ? 0 : p->opta[6]);
+
+       p1 = *p;
+       memmove(&p1.Domain[0], &p ->Domain[1], 6*sizeof(cmsUInt32Number));
+       
+       T = LutTable + K0;
+       p1.Table = T;
+
+       Eval6InputsFloat(Input + 1,  Tmp1, &p1);
+      
+       T = LutTable + K1;
+       p1.Table = T;
+       
+       Eval6InputsFloat(Input + 1,  Tmp2, &p1);
+      
+       
+       for (i=0; i < p -> nOutputs; i++) {
+
+              cmsFloat32Number y0 = Tmp1[i];
+              cmsFloat32Number y1 = Tmp2[i];
+
+              Output[i] = y0 + (y1 - y0) * rest;     
+              
+       }
+}
+
+static
+void Eval8Inputs(register const cmsUInt16Number Input[], 
+                 register cmsUInt16Number Output[], 
+                 register const cmsInterpParams* p16)
+{       
+       const cmsUInt16Number* LutTable = (cmsUInt16Number*) p16 -> Table; 
+       cmsS15Fixed16Number fk;
+       cmsS15Fixed16Number k0, rk;
+       int K0, K1;
+       const cmsUInt16Number* T;
+       cmsUInt32Number i;
+       cmsUInt16Number Tmp1[MAX_STAGE_CHANNELS], Tmp2[MAX_STAGE_CHANNELS];
+       cmsInterpParams p1;
+       
+       fk = _cmsToFixedDomain((cmsS15Fixed16Number) Input[0] * p16 -> Domain[0]);
+       k0 = FIXED_TO_INT(fk);
+       rk = FIXED_REST_TO_INT(fk);
+
+       K0 = p16 -> opta[7] * k0;
+       K1 = p16 -> opta[7] * (k0 + (Input[0] != 0xFFFFU ? 1 : 0));
+
+       p1 = *p16;
+       memmove(&p1.Domain[0], &p16 ->Domain[1], 7*sizeof(cmsUInt32Number));
+       
+       T = LutTable + K0;
+       p1.Table = T;
+
+       Eval7Inputs(Input + 1, Tmp1, &p1);
+
+       T = LutTable + K1;
+       p1.Table = T;
+       Eval7Inputs(Input + 1, Tmp2, &p1);
+
+       for (i=0; i < p16 -> nOutputs; i++) {
+              Output[i] = LinearInterp(rk, Tmp1[i], Tmp2[i]);
+       }
+}
+
+
+
+static
+void Eval8InputsFloat(const cmsFloat32Number Input[], 
+                      cmsFloat32Number Output[], 
+                      const cmsInterpParams* p)
+{                             
+       const cmsFloat32Number* LutTable = (cmsFloat32Number*) p -> Table; 
+       cmsFloat32Number rest;
+       cmsFloat32Number pk;
+       int k0, K0, K1;
+       const cmsFloat32Number* T;
+       cmsUInt32Number i;
+       cmsFloat32Number Tmp1[MAX_STAGE_CHANNELS], Tmp2[MAX_STAGE_CHANNELS];
+       cmsInterpParams p1;
+       
+       pk = Input[0] * p->Domain[0];    
+       k0 = _cmsQuickFloor(pk); 
+       rest = pk - (cmsFloat32Number) k0;
+
+       K0 = p -> opta[7] * k0;
+       K1 = K0 + (Input[0] >= 1.0 ? 0 : p->opta[7]);
+
+       p1 = *p;
+       memmove(&p1.Domain[0], &p ->Domain[1], 7*sizeof(cmsUInt32Number));
+
+       T = LutTable + K0;
+       p1.Table = T;
+
+       Eval7InputsFloat(Input + 1,  Tmp1, &p1);
+      
+       T = LutTable + K1;
+       p1.Table = T;
+
+       Eval7InputsFloat(Input + 1,  Tmp2, &p1);
+      
+
+       for (i=0; i < p -> nOutputs; i++) {
+
+              cmsFloat32Number y0 = Tmp1[i];
+              cmsFloat32Number y1 = Tmp2[i];
+
+              Output[i] = y0 + (y1 - y0) * rest;                   
+       }
+}
+
+// The default factory
+static 
+cmsInterpFunction DefaultInterpolatorsFactory(cmsUInt32Number nInputChannels, cmsUInt32Number nOutputChannels, cmsUInt32Number dwFlags)
+{
+
+    cmsInterpFunction Interpolation;
+    cmsBool  IsFloat     = (dwFlags & CMS_LERP_FLAGS_FLOAT);
+    cmsBool  IsTrilinear = (dwFlags & CMS_LERP_FLAGS_TRILINEAR);
+
+    memset(&Interpolation, 0, sizeof(Interpolation));
+
+    // Safety check
+    if (nInputChannels >= 4 && nOutputChannels >= MAX_STAGE_CHANNELS) 
+        return Interpolation;
+
+    switch (nInputChannels) {
+
+           case 1: // Gray LUT / linear
+
+               if (nOutputChannels == 1) {
+
+                   if (IsFloat) 
+                       Interpolation.LerpFloat = LinLerp1Dfloat;
+                   else
+                       Interpolation.Lerp16 = LinLerp1D;
+
+               }
+               else {
+
+                   if (IsFloat)
+                       Interpolation.LerpFloat = Eval1InputFloat;               
+                   else
+                       Interpolation.Lerp16 = Eval1Input;               
+               }
+               break;
+
+           case 2: // Duotone
+               if (IsFloat) 
+                      Interpolation.LerpFloat =  BilinearInterpFloat;
+               else
+                      Interpolation.Lerp16    =  BilinearInterp16;
+               break;
+
+           case 3:  // RGB et al   
+
+               if (IsTrilinear) {
+
+                   if (IsFloat) 
+                       Interpolation.LerpFloat = TrilinearInterpFloat;
+                   else
+                       Interpolation.Lerp16 = TrilinearInterp16;
+               }
+               else {
+
+                   if (IsFloat) 
+                       Interpolation.LerpFloat = TetrahedralInterpFloat;
+                   else {
+
+                       Interpolation.Lerp16 = TetrahedralInterp16;
+                   }
+               }
+               break;
+
+           case 4:  // CMYK lut             
+
+               if (IsFloat) 
+                   Interpolation.LerpFloat =  Eval4InputsFloat;
+               else
+                   Interpolation.Lerp16    =  Eval4Inputs;
+               break;
+
+           case 5: // 5 Inks
+               if (IsFloat) 
+                   Interpolation.LerpFloat =  Eval5InputsFloat;
+               else
+                   Interpolation.Lerp16    =  Eval5Inputs;
+               break;              
+
+           case 6: // 6 Inks
+               if (IsFloat) 
+                   Interpolation.LerpFloat =  Eval6InputsFloat;
+               else
+                   Interpolation.Lerp16    =  Eval6Inputs;
+               break;
+
+           case 7: // 7 inks
+               if (IsFloat) 
+                   Interpolation.LerpFloat =  Eval7InputsFloat;
+               else
+                   Interpolation.Lerp16    =  Eval7Inputs;
+               break;
+
+           case 8: // 8 inks
+               if (IsFloat) 
+                   Interpolation.LerpFloat =  Eval8InputsFloat;
+               else
+                   Interpolation.Lerp16    =  Eval8Inputs;
+               break;
+
+               break;
+
+           default:
+               Interpolation.Lerp16 = NULL;
+    }
+
+    return Interpolation;
+}
diff --git a/thirdparty/liblcms2/src/cmsio0.c b/thirdparty/liblcms2/src/cmsio0.c
new file mode 100644
index 00000000..589ea6a3
--- /dev/null
+++ b/thirdparty/liblcms2/src/cmsio0.c
@@ -0,0 +1,1720 @@
+//---------------------------------------------------------------------------------
+//
+//  Little Color Management System
+//  Copyright (c) 1998-2010 Marti Maria Saguer
+//
+// Permission is hereby granted, free of charge, to any person obtaining 
+// a copy of this software and associated documentation files (the "Software"), 
+// to deal in the Software without restriction, including without limitation 
+// the rights to use, copy, modify, merge, publish, distribute, sublicense, 
+// and/or sell copies of the Software, and to permit persons to whom the Software 
+// is furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in 
+// all copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 
+// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO 
+// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE 
+// LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 
+// OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 
+// WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+//
+//---------------------------------------------------------------------------------
+//
+
+#include "lcms2_internal.h"
+
+// Generic I/O, tag dictionary management, profile struct
+
+// IOhandlers are abstractions used by littleCMS to read from whatever file, stream,
+// memory block or any storage. Each IOhandler provides implementations for read,
+// write, seek and tell functions. LittleCMS code deals with IO across those objects.
+// In this way, is easier to add support for new storage media.
+
+// NULL stream, for taking care of used space -------------------------------------
+
+// NULL IOhandler basically does nothing but keep track on how many bytes have been
+// written. This is handy when creating profiles, where the file size is needed in the
+// header. Then, whole profile is serialized across NULL IOhandler and a second pass 
+// writes the bytes to the pertinent IOhandler.
+
+typedef struct {        
+    cmsUInt32Number Pointer;         // Points to current location
+} FILENULL;
+
+static
+cmsUInt32Number NULLRead(cmsIOHANDLER* iohandler, void *Buffer, cmsUInt32Number size, cmsUInt32Number count)
+{   
+    FILENULL* ResData = (FILENULL*) iohandler ->stream;
+    
+    cmsUInt32Number len = size * count;
+    ResData -> Pointer += len;
+    return count;
+
+    cmsUNUSED_PARAMETER(Buffer);
+}
+
+static
+cmsBool  NULLSeek(cmsIOHANDLER* iohandler, cmsUInt32Number offset)
+{
+    FILENULL* ResData = (FILENULL*) iohandler ->stream;
+
+    ResData ->Pointer = offset; 
+    return TRUE; 
+}
+
+static
+cmsUInt32Number NULLTell(cmsIOHANDLER* iohandler)
+{
+    FILENULL* ResData = (FILENULL*) iohandler ->stream;
+    return ResData -> Pointer;     
+}
+
+static
+cmsBool  NULLWrite(cmsIOHANDLER* iohandler, cmsUInt32Number size, const void *Ptr)
+{
+    FILENULL* ResData = (FILENULL*) iohandler ->stream;
+
+    ResData ->Pointer += size;
+    if (ResData ->Pointer > iohandler->UsedSpace)
+        iohandler->UsedSpace = ResData ->Pointer;
+
+    return TRUE;
+
+    cmsUNUSED_PARAMETER(Ptr);
+}
+
+static
+cmsBool  NULLClose(cmsIOHANDLER* iohandler)
+{   
+    FILENULL* ResData = (FILENULL*) iohandler ->stream;
+    
+    _cmsFree(iohandler ->ContextID, ResData);
+    _cmsFree(iohandler ->ContextID, iohandler);
+    return TRUE;
+}
+
+// The NULL IOhandler creator
+cmsIOHANDLER*  CMSEXPORT cmsOpenIOhandlerFromNULL(cmsContext ContextID)
+{
+    struct _cms_io_handler* iohandler = NULL;
+    FILENULL* fm = NULL;
+        
+    iohandler = (struct _cms_io_handler*) _cmsMallocZero(ContextID, sizeof(struct _cms_io_handler));
+    if (iohandler == NULL) return NULL;
+
+    fm = (FILENULL*) _cmsMallocZero(ContextID, sizeof(FILENULL));
+    if (fm == NULL) goto Error;
+    
+    fm ->Pointer = 0;
+
+    iohandler ->ContextID = ContextID;
+    iohandler ->stream  = (void*) fm;   
+    iohandler ->UsedSpace = 0;
+    iohandler ->ReportedSize = 0;
+    iohandler ->PhysicalFile[0] = 0;
+
+    iohandler ->Read    = NULLRead;
+    iohandler ->Seek    = NULLSeek;
+    iohandler ->Close   = NULLClose;
+    iohandler ->Tell    = NULLTell;
+    iohandler ->Write   = NULLWrite;
+
+    return iohandler;
+
+Error:
+    if (fm) _cmsFree(ContextID, fm);
+    if (iohandler) _cmsFree(ContextID, iohandler);
+    return NULL;
+
+}
+
+
+// Memory-based stream --------------------------------------------------------------
+
+// Those functions implements an iohandler which takes a block of memory as storage medium.
+
+typedef struct {
+    cmsUInt8Number* Block;    // Points to allocated memory
+    cmsUInt32Number Size;     // Size of allocated memory
+    cmsUInt32Number Pointer;  // Points to current location
+    int FreeBlockOnClose;     // As title
+
+} FILEMEM;
+
+static
+cmsUInt32Number MemoryRead(struct _cms_io_handler* iohandler, void *Buffer, cmsUInt32Number size, cmsUInt32Number count)
+{   
+    FILEMEM* ResData = (FILEMEM*) iohandler ->stream;
+    cmsUInt8Number* Ptr;
+    cmsUInt32Number len = size * count;
+
+    if (ResData -> Pointer + len > ResData -> Size){
+
+        len = (ResData -> Size - ResData -> Pointer);
+        cmsSignalError(iohandler ->ContextID, cmsERROR_READ, "Read from memory error. Got %d bytes, block should be of %d bytes", len, count * size);
+        return 0;
+    }
+
+    Ptr  = ResData -> Block;
+    Ptr += ResData -> Pointer;
+    memmove(Buffer, Ptr, len);        
+    ResData -> Pointer += len;
+
+    return count;
+}
+
+// SEEK_CUR is assumed
+static
+cmsBool  MemorySeek(struct _cms_io_handler* iohandler, cmsUInt32Number offset)
+{
+    FILEMEM* ResData = (FILEMEM*) iohandler ->stream;
+
+    if (offset > ResData ->Size) {
+        cmsSignalError(iohandler ->ContextID, cmsERROR_SEEK,  "Too few data; probably corrupted profile");        
+        return FALSE;
+    }
+
+    ResData ->Pointer = offset; 
+    return TRUE; 
+}
+
+// Tell for memory
+static
+cmsUInt32Number MemoryTell(struct _cms_io_handler* iohandler)
+{
+    FILEMEM* ResData = (FILEMEM*) iohandler ->stream;
+
+	if (ResData == NULL) return 0;
+    return ResData -> Pointer;     
+}
+
+
+// Writes data to memory, also keeps used space for further reference. 
+static
+cmsBool  MemoryWrite(struct _cms_io_handler* iohandler, cmsUInt32Number size, const void *Ptr)
+{
+    FILEMEM* ResData = (FILEMEM*) iohandler ->stream;
+
+	if (ResData == NULL) return FALSE; // Housekeeping
+
+    if (size == 0) return TRUE;     // Write zero bytes is ok, but does nothing
+
+    memmove(ResData ->Block + ResData ->Pointer, Ptr, size);
+    ResData ->Pointer += size;
+
+    if (ResData ->Pointer > iohandler->UsedSpace)
+        iohandler->UsedSpace = ResData ->Pointer;
+
+    
+    iohandler->UsedSpace += size;    
+
+    return TRUE;
+}
+
+
+static
+cmsBool  MemoryClose(struct _cms_io_handler* iohandler)
+{   
+    FILEMEM* ResData = (FILEMEM*) iohandler ->stream;
+
+    if (ResData ->FreeBlockOnClose) {
+
+        if (ResData ->Block) _cmsFree(iohandler ->ContextID, ResData ->Block);
+    }
+    
+    _cmsFree(iohandler ->ContextID, ResData);
+    _cmsFree(iohandler ->ContextID, iohandler);
+
+    return TRUE;
+}
+
+// Create a iohandler for memory block. AccessMode=='r' assumes the iohandler is going to read, and makes
+// a copy of the memory block for letting user to free the memory after invoking open profile. In write 
+// mode ("w"), Buffere points to the begin of memory block to be written.
+cmsIOHANDLER* CMSEXPORT cmsOpenIOhandlerFromMem(cmsContext ContextID, void *Buffer, cmsUInt32Number size, const char* AccessMode)
+{
+    cmsIOHANDLER* iohandler = NULL;
+    FILEMEM* fm = NULL;
+
+	_cmsAssert(AccessMode != NULL);
+
+    iohandler = (cmsIOHANDLER*) _cmsMallocZero(ContextID, sizeof(cmsIOHANDLER));
+    if (iohandler == NULL) return NULL;
+
+    switch (*AccessMode) {
+
+    case 'r':
+        fm = (FILEMEM*) _cmsMallocZero(ContextID, sizeof(FILEMEM));
+        if (fm == NULL) goto Error;
+
+        if (Buffer == NULL) {
+            cmsSignalError(ContextID, cmsERROR_READ, "Couldn't read profile from NULL pointer");
+            goto Error;
+        }
+
+        fm ->Block = (cmsUInt8Number*) _cmsMalloc(ContextID, size);
+        if (fm ->Block == NULL) {
+
+            _cmsFree(ContextID, fm);
+            _cmsFree(ContextID, iohandler);
+            cmsSignalError(ContextID, cmsERROR_READ, "Couldn't allocate %ld bytes for profile", size);
+            return NULL;
+        }
+
+
+        memmove(fm->Block, Buffer, size);
+        fm ->FreeBlockOnClose = TRUE;
+        fm ->Size    = size;
+        fm ->Pointer = 0;
+        iohandler -> ReportedSize = size;
+        break;
+
+    case 'w': 
+        fm = (FILEMEM*) _cmsMallocZero(ContextID, sizeof(FILEMEM));
+        if (fm == NULL) goto Error;
+
+        fm ->Block = (cmsUInt8Number*) Buffer;
+        fm ->FreeBlockOnClose = FALSE;
+        fm ->Size    = size;
+        fm ->Pointer = 0;     
+        iohandler -> ReportedSize = 0;
+        break;
+
+    default:
+        cmsSignalError(ContextID, cmsERROR_UNKNOWN_EXTENSION, "Unknown access mode '%c'", *AccessMode);
+        return NULL;
+    }
+
+    iohandler ->ContextID = ContextID;
+    iohandler ->stream  = (void*) fm;   
+    iohandler ->UsedSpace = 0;
+    iohandler ->PhysicalFile[0] = 0;
+
+    iohandler ->Read    = MemoryRead;
+    iohandler ->Seek    = MemorySeek;
+    iohandler ->Close   = MemoryClose;
+    iohandler ->Tell    = MemoryTell;
+    iohandler ->Write   = MemoryWrite;
+
+    return iohandler;
+
+Error:
+    if (fm) _cmsFree(ContextID, fm);
+    if (iohandler) _cmsFree(ContextID, iohandler);
+    return NULL;
+}
+
+// File-based stream -------------------------------------------------------
+
+// Read count elements of size bytes each. Return number of elements read
+static
+cmsUInt32Number FileRead(cmsIOHANDLER* iohandler, void *Buffer, cmsUInt32Number size, cmsUInt32Number count)
+{
+    cmsUInt32Number nReaded = (cmsUInt32Number) fread(Buffer, size, count, (FILE*) iohandler->stream);
+
+    if (nReaded != count) {
+            cmsSignalError(iohandler ->ContextID, cmsERROR_FILE, "Read error. Got %d bytes, block should be of %d bytes", nReaded * size, count * size);
+            return 0;
+    }
+
+    return nReaded;
+}
+
+// Postion file pointer in the file
+static
+cmsBool  FileSeek(cmsIOHANDLER* iohandler, cmsUInt32Number offset)
+{
+    if (fseek((FILE*) iohandler ->stream, (long) offset, SEEK_SET) != 0) {
+
+       cmsSignalError(iohandler ->ContextID, cmsERROR_FILE, "Seek error; probably corrupted file");
+       return FALSE;
+    }
+
+    return TRUE;
+}
+
+// Returns file pointer position
+static
+cmsUInt32Number FileTell(cmsIOHANDLER* iohandler)
+{       
+    return ftell((FILE*)iohandler ->stream);
+}
+
+// Writes data to stream, also keeps used space for further reference. Returns TRUE on success, FALSE on error
+static
+cmsBool  FileWrite(cmsIOHANDLER* iohandler, cmsUInt32Number size, const void* Buffer)
+{
+       if (size == 0) return TRUE;  // We allow to write 0 bytes, but nothing is written
+
+       iohandler->UsedSpace += size;
+       return (fwrite(Buffer, size, 1, (FILE*) iohandler->stream) == 1);
+}
+
+// Closes the file
+static
+cmsBool  FileClose(cmsIOHANDLER* iohandler)
+{   
+    if (fclose((FILE*) iohandler ->stream) != 0) return FALSE;
+    _cmsFree(iohandler ->ContextID, iohandler);
+    return TRUE;
+}
+
+// Create a iohandler for disk based files. if FileName is NULL, then 'stream' member is also set
+// to NULL and no real writting is performed. This only happens in writting access mode
+cmsIOHANDLER* CMSEXPORT cmsOpenIOhandlerFromFile(cmsContext ContextID, const char* FileName, const char* AccessMode)
+{
+    cmsIOHANDLER* iohandler = NULL;
+    FILE* fm = NULL;
+
+    iohandler = (cmsIOHANDLER*) _cmsMallocZero(ContextID, sizeof(cmsIOHANDLER));
+    if (iohandler == NULL) return NULL;
+
+    switch (*AccessMode) {
+
+    case 'r':
+        fm = fopen(FileName, "rb");
+        if (fm == NULL) {
+            _cmsFree(ContextID, iohandler);
+             cmsSignalError(ContextID, cmsERROR_FILE, "File '%s' not found", FileName);
+            return NULL;
+        }
+        iohandler -> ReportedSize = cmsfilelength(fm);
+        break;
+
+    case 'w':
+        fm = fopen(FileName, "wb");
+        if (fm == NULL) {
+            _cmsFree(ContextID, iohandler);
+             cmsSignalError(ContextID, cmsERROR_FILE, "Couldn't create '%s'", FileName);
+            return NULL;
+        }
+        iohandler -> ReportedSize = 0;
+        break;
+
+    default:
+        _cmsFree(ContextID, iohandler);
+         cmsSignalError(ContextID, cmsERROR_FILE, "Unknown access mode '%c'", *AccessMode);
+        return NULL;
+    }
+
+    iohandler ->ContextID = ContextID;
+    iohandler ->stream = (void*) fm;
+    iohandler ->UsedSpace = 0;
+
+    // Keep track of the original file
+    if (FileName != NULL)  {
+
+        strncpy(iohandler -> PhysicalFile, FileName, sizeof(iohandler -> PhysicalFile)-1);
+        iohandler -> PhysicalFile[sizeof(iohandler -> PhysicalFile)-1] = 0;
+    }
+
+    iohandler ->Read    = FileRead;
+    iohandler ->Seek    = FileSeek;
+    iohandler ->Close   = FileClose;
+    iohandler ->Tell    = FileTell;
+    iohandler ->Write   = FileWrite;
+
+    return iohandler;
+}
+
+// Create a iohandler for stream based files 
+cmsIOHANDLER* CMSEXPORT cmsOpenIOhandlerFromStream(cmsContext ContextID, FILE* Stream)
+{
+    cmsIOHANDLER* iohandler = NULL;
+    
+    iohandler = (cmsIOHANDLER*) _cmsMallocZero(ContextID, sizeof(cmsIOHANDLER));
+    if (iohandler == NULL) return NULL;
+
+    iohandler -> ContextID = ContextID;
+    iohandler -> stream = (void*) Stream;
+    iohandler -> UsedSpace = 0;
+    iohandler -> ReportedSize = cmsfilelength(Stream);
+    iohandler -> PhysicalFile[0] = 0;
+    
+    iohandler ->Read    = FileRead;
+    iohandler ->Seek    = FileSeek;
+    iohandler ->Close   = FileClose;
+    iohandler ->Tell    = FileTell;
+    iohandler ->Write   = FileWrite;
+
+    return iohandler;
+}
+
+
+
+// Close an open IO handler
+cmsBool CMSEXPORT cmsCloseIOhandler(cmsIOHANDLER* io)
+{
+    return io -> Close(io);
+}
+
+// -------------------------------------------------------------------------------------------------------
+
+// Creates an empty structure holding all required parameters
+cmsHPROFILE CMSEXPORT cmsCreateProfilePlaceholder(cmsContext ContextID)
+{
+    time_t now = time(NULL);
+    _cmsICCPROFILE* Icc = (_cmsICCPROFILE*) _cmsMallocZero(ContextID, sizeof(_cmsICCPROFILE));
+    if (Icc == NULL) return NULL;
+
+    Icc ->ContextID = ContextID;
+
+    // Set it to empty
+    Icc -> TagCount   = 0;
+
+    // Set default version 
+    Icc ->Version =  0x02100000;
+
+    // Set creation date/time
+    memmove(&Icc ->Created, gmtime(&now), sizeof(Icc ->Created));
+
+    // Return the handle
+    return (cmsHPROFILE) Icc;
+}
+
+cmsContext CMSEXPORT cmsGetProfileContextID(cmsHPROFILE hProfile)
+{
+     _cmsICCPROFILE* Icc = (_cmsICCPROFILE*) hProfile;  
+
+    if (Icc == NULL) return NULL;
+    return Icc -> ContextID;
+}
+
+
+// Return the number of tags
+cmsInt32Number CMSEXPORT cmsGetTagCount(cmsHPROFILE hProfile)
+{
+    _cmsICCPROFILE* Icc = (_cmsICCPROFILE*) hProfile;  
+	if (Icc == NULL) return -1;
+
+    return  Icc->TagCount;
+}
+
+// Return the tag signature of a given tag number
+cmsTagSignature CMSEXPORT cmsGetTagSignature(cmsHPROFILE hProfile, cmsUInt32Number n)
+{
+    _cmsICCPROFILE* Icc = (_cmsICCPROFILE*) hProfile;    
+
+    if (n > Icc->TagCount) return (cmsTagSignature) 0;  // Mark as not available     
+    if (n >= MAX_TABLE_TAG) return (cmsTagSignature) 0; // As double check
+
+    return Icc ->TagNames[n];
+}
+
+
+static
+int SearchOneTag(_cmsICCPROFILE* Profile, cmsTagSignature sig)
+{
+	cmsUInt32Number i;
+
+	for (i=0; i < Profile -> TagCount; i++) {
+
+		if (sig == Profile -> TagNames[i])
+			return i;
+	}
+
+	return -1;
+}
+
+// Search for a specific tag in tag dictionary. Returns position or -1 if tag not found.
+// If followlinks is turned on, then the position of the linked tag is returned
+int _cmsSearchTag(_cmsICCPROFILE* Icc, cmsTagSignature sig, cmsBool lFollowLinks)
+{
+	int n;
+	cmsTagSignature LinkedSig;
+
+	do {
+
+		// Search for given tag in ICC profile directory
+		n = SearchOneTag(Icc, sig);
+		if (n < 0) 
+			return -1;        // Not found
+
+		if (!lFollowLinks) 
+			return n;         // Found, don't follow links
+
+		// Is this a linked tag?
+		LinkedSig = Icc ->TagLinked[n];
+
+		// Yes, follow link
+		if (LinkedSig != (cmsTagSignature) 0) {
+			sig = LinkedSig;
+		}
+
+	} while (LinkedSig != (cmsTagSignature) 0);
+
+	return n;
+}
+
+
+// Create a new tag entry
+
+static
+cmsBool _cmsNewTag(_cmsICCPROFILE* Icc, cmsTagSignature sig, int* NewPos)
+{
+	int i;
+
+	// Search for the tag
+    i = _cmsSearchTag(Icc, sig, FALSE);
+
+    // Now let's do it easy. If the tag has been already written, that's an error
+    if (i >= 0) {
+        cmsSignalError(Icc ->ContextID, cmsERROR_ALREADY_DEFINED, "Tag '%x' already exists", sig);
+        return FALSE;
+    }
+    else  {
+
+        // New one
+        
+        if (Icc -> TagCount >= MAX_TABLE_TAG) {
+            cmsSignalError(Icc ->ContextID, cmsERROR_RANGE, "Too many tags (%d)", MAX_TABLE_TAG);
+            return FALSE;
+        }
+
+		*NewPos = Icc ->TagCount;
+        Icc -> TagCount++;
+    }
+
+	return TRUE;
+}
+
+
+// Check existance
+cmsBool CMSEXPORT cmsIsTag(cmsHPROFILE hProfile, cmsTagSignature sig)
+{
+       _cmsICCPROFILE*  Icc = (_cmsICCPROFILE*) (void*) hProfile;
+       return _cmsSearchTag(Icc, sig, FALSE) >= 0;
+}
+
+
+// Read profile header and validate it
+cmsBool _cmsReadHeader(_cmsICCPROFILE* Icc)
+{
+    cmsTagEntry Tag;
+    cmsICCHeader Header;
+    cmsUInt32Number i, j;
+    cmsUInt32Number HeaderSize;
+    cmsIOHANDLER* io = Icc ->IOhandler;
+    cmsUInt32Number TagCount;
+
+
+    // Read the header
+    if (io -> Read(io, &Header, sizeof(cmsICCHeader), 1) != 1) {        
+        return FALSE; 
+    }
+
+    // Validate file as an ICC profile
+    if (_cmsAdjustEndianess32(Header.magic) != cmsMagicNumber) {
+        cmsSignalError(Icc ->ContextID, cmsERROR_BAD_SIGNATURE, "not an ICC profile, invalid signature");
+        return FALSE; 
+    }
+
+    // Adjust endianess of the used parameters
+    Icc -> DeviceClass     = (cmsProfileClassSignature) _cmsAdjustEndianess32(Header.deviceClass);
+    Icc -> ColorSpace      = (cmsColorSpaceSignature)   _cmsAdjustEndianess32(Header.colorSpace);
+    Icc -> PCS             = (cmsColorSpaceSignature)   _cmsAdjustEndianess32(Header.pcs);
+    Icc -> RenderingIntent = _cmsAdjustEndianess32(Header.renderingIntent);
+    Icc -> flags           = _cmsAdjustEndianess32(Header.flags);
+    Icc -> manufacturer    = _cmsAdjustEndianess32(Header.manufacturer);
+    Icc -> model           = _cmsAdjustEndianess32(Header.model);
+    _cmsAdjustEndianess64(&Icc -> attributes, Header.attributes);
+    Icc -> Version         = _cmsAdjustEndianess32(Header.version);
+
+    // Get size as reported in header
+    HeaderSize = _cmsAdjustEndianess32(Header.size);
+
+    // Make sure HeaderSize is lower than profile size
+    if (HeaderSize >= Icc ->IOhandler ->ReportedSize)
+            HeaderSize = Icc ->IOhandler ->ReportedSize;
+
+   
+    // Get creation date/time
+    _cmsDecodeDateTimeNumber(&Header.date, &Icc ->Created);
+
+    // The profile ID are 32 raw bytes
+    memmove(Icc ->ProfileID.ID32, Header.profileID.ID32, 16);
+
+
+    // Read tag directory
+    if (!_cmsReadUInt32Number(io, &TagCount)) return FALSE;                          
+    if (TagCount > MAX_TABLE_TAG) {
+
+        cmsSignalError(Icc ->ContextID, cmsERROR_RANGE, "Too many tags (%d)", TagCount);
+        return FALSE;
+    }
+
+
+    // Read tag directory
+    Icc -> TagCount = 0;
+    for (i=0; i < TagCount; i++) {
+
+        if (!_cmsReadUInt32Number(io, (cmsUInt32Number *) &Tag.sig)) return FALSE;
+        if (!_cmsReadUInt32Number(io, &Tag.offset)) return FALSE;
+        if (!_cmsReadUInt32Number(io, &Tag.size)) return FALSE;
+
+        // Perform some sanity check. Offset + size should fall inside file.
+        if (Tag.offset + Tag.size > HeaderSize ||
+            Tag.offset + Tag.size < Tag.offset)       
+                  continue;
+
+        Icc -> TagNames[Icc ->TagCount]   = Tag.sig;
+        Icc -> TagOffsets[Icc ->TagCount] = Tag.offset;
+        Icc -> TagSizes[Icc ->TagCount]   = Tag.size;
+
+       // Search for links
+        for (j=0; j < Icc ->TagCount; j++) {
+
+            if ((Icc ->TagOffsets[j] == Tag.offset) &&
+                (Icc ->TagSizes[j]   == Tag.size)) {
+
+                Icc ->TagLinked[Icc ->TagCount] = Icc ->TagNames[j];   
+            }
+
+        }
+
+        Icc ->TagCount++;
+    }
+  
+    return TRUE;
+}
+
+// Saves profile header
+cmsBool _cmsWriteHeader(_cmsICCPROFILE* Icc, cmsUInt32Number UsedSpace)
+{
+    cmsICCHeader Header;    
+    cmsUInt32Number i;
+    cmsTagEntry Tag;       
+    cmsInt32Number Count = 0;
+
+    Header.size        = _cmsAdjustEndianess32(UsedSpace);
+    Header.cmmId       = _cmsAdjustEndianess32(lcmsSignature);
+    Header.version     = _cmsAdjustEndianess32(Icc ->Version);
+
+    Header.deviceClass = (cmsProfileClassSignature) _cmsAdjustEndianess32(Icc -> DeviceClass);
+    Header.colorSpace  = (cmsColorSpaceSignature) _cmsAdjustEndianess32(Icc -> ColorSpace);
+    Header.pcs         = (cmsColorSpaceSignature) _cmsAdjustEndianess32(Icc -> PCS);
+
+    //   NOTE: in v4 Timestamp must be in UTC rather than in local time
+    _cmsEncodeDateTimeNumber(&Header.date, &Icc ->Created);
+
+    Header.magic       = _cmsAdjustEndianess32(cmsMagicNumber);
+
+#ifdef CMS_IS_WINDOWS_
+    Header.platform    = (cmsPlatformSignature) _cmsAdjustEndianess32(cmsSigMicrosoft);  
+#else
+    Header.platform    = (cmsPlatformSignature) _cmsAdjustEndianess32(cmsSigMacintosh);  
+#endif
+
+    Header.flags        = _cmsAdjustEndianess32(Icc -> flags);
+    Header.manufacturer = _cmsAdjustEndianess32(Icc -> manufacturer);
+    Header.model        = _cmsAdjustEndianess32(Icc -> model);
+
+    _cmsAdjustEndianess64(&Header.attributes, Icc -> attributes);              
+    
+    // Rendering intent in the header (for embedded profiles)
+    Header.renderingIntent = _cmsAdjustEndianess32(Icc -> RenderingIntent);
+
+    // Illuminant is always D50
+    Header.illuminant.X = _cmsAdjustEndianess32(_cmsDoubleTo15Fixed16(cmsD50_XYZ()->X));
+    Header.illuminant.Y = _cmsAdjustEndianess32(_cmsDoubleTo15Fixed16(cmsD50_XYZ()->Y));
+    Header.illuminant.Z = _cmsAdjustEndianess32(_cmsDoubleTo15Fixed16(cmsD50_XYZ()->Z));
+
+    // Created by LittleCMS (that's me!)
+    Header.creator      = _cmsAdjustEndianess32(lcmsSignature);
+
+    memset(&Header.reserved, 0, sizeof(Header.reserved));
+
+    // Set profile ID. Endianess is always big endian   
+    memmove(&Header.profileID, &Icc ->ProfileID, 16);
+
+    // Dump the header
+    if (!Icc -> IOhandler->Write(Icc->IOhandler, sizeof(cmsICCHeader), &Header)) return FALSE;
+
+    // Saves Tag directory
+
+    // Get true count
+    for (i=0;  i < Icc -> TagCount; i++) {
+        if (Icc ->TagNames[i] != 0)
+            Count++;
+    }
+
+    // Store number of tags
+    if (!_cmsWriteUInt32Number(Icc ->IOhandler, Count)) return FALSE;   
+
+    for (i=0; i < Icc -> TagCount; i++) {
+
+        if (Icc ->TagNames[i] == 0) continue;   // It is just a placeholder
+
+        Tag.sig    = (cmsTagSignature) _cmsAdjustEndianess32((cmsInt32Number) Icc -> TagNames[i]);
+        Tag.offset = _cmsAdjustEndianess32((cmsInt32Number) Icc -> TagOffsets[i]);
+        Tag.size   = _cmsAdjustEndianess32((cmsInt32Number) Icc -> TagSizes[i]);
+
+        if (!Icc ->IOhandler -> Write(Icc-> IOhandler, sizeof(cmsTagEntry), &Tag)) return FALSE;
+    }
+
+    return TRUE;
+}
+
+// ----------------------------------------------------------------------- Set/Get several struct members
+
+
+cmsUInt32Number CMSEXPORT cmsGetHeaderRenderingIntent(cmsHPROFILE hProfile)
+{
+    _cmsICCPROFILE*  Icc = (_cmsICCPROFILE*) hProfile;
+    return Icc -> RenderingIntent;
+}
+
+void CMSEXPORT cmsSetHeaderRenderingIntent(cmsHPROFILE hProfile, cmsUInt32Number RenderingIntent)
+{
+    _cmsICCPROFILE*  Icc = (_cmsICCPROFILE*) hProfile;
+    Icc -> RenderingIntent = RenderingIntent;
+}
+
+cmsUInt32Number CMSEXPORT cmsGetHeaderFlags(cmsHPROFILE hProfile)
+{
+    _cmsICCPROFILE*  Icc = (_cmsICCPROFILE*) hProfile;
+    return (cmsUInt32Number) Icc -> flags;
+}
+
+void CMSEXPORT cmsSetHeaderFlags(cmsHPROFILE hProfile, cmsUInt32Number Flags)
+{
+    _cmsICCPROFILE*  Icc = (_cmsICCPROFILE*) hProfile;
+    Icc -> flags = (cmsUInt32Number) Flags;
+}
+
+cmsUInt32Number CMSEXPORT cmsGetHeaderManufacturer(cmsHPROFILE hProfile)
+{
+    _cmsICCPROFILE*  Icc = (_cmsICCPROFILE*) hProfile;
+    return (cmsUInt32Number) Icc ->manufacturer;
+}
+
+void CMSEXPORT cmsSetHeaderManufacturer(cmsHPROFILE hProfile, cmsUInt32Number manufacturer)
+{
+    _cmsICCPROFILE*  Icc = (_cmsICCPROFILE*) hProfile;
+    Icc -> manufacturer = (cmsUInt32Number) manufacturer;
+}
+
+cmsUInt32Number CMSEXPORT cmsGetHeaderModel(cmsHPROFILE hProfile)
+{
+    _cmsICCPROFILE*  Icc = (_cmsICCPROFILE*) hProfile;
+    return (cmsUInt32Number) Icc ->model;
+}
+
+void CMSEXPORT cmsSetHeaderModel(cmsHPROFILE hProfile, cmsUInt32Number model)
+{
+    _cmsICCPROFILE*  Icc = (_cmsICCPROFILE*) hProfile;
+    Icc -> manufacturer = (cmsUInt32Number) model;
+}
+
+
+void CMSEXPORT cmsGetHeaderAttributes(cmsHPROFILE hProfile, cmsUInt64Number* Flags)
+{
+    _cmsICCPROFILE*  Icc = (_cmsICCPROFILE*) hProfile;
+    memmove(Flags, &Icc -> attributes, sizeof(cmsUInt64Number));
+}
+
+void CMSEXPORT cmsSetHeaderAttributes(cmsHPROFILE hProfile, cmsUInt64Number Flags)
+{
+    _cmsICCPROFILE*  Icc = (_cmsICCPROFILE*) hProfile;
+    memmove(&Icc -> attributes, &Flags, sizeof(cmsUInt64Number));
+}
+
+void CMSEXPORT cmsGetHeaderProfileID(cmsHPROFILE hProfile, cmsUInt8Number* ProfileID)
+{
+    _cmsICCPROFILE*  Icc = (_cmsICCPROFILE*) hProfile;    
+    memmove(ProfileID, Icc ->ProfileID.ID8, 16);    
+}
+
+void CMSEXPORT cmsSetHeaderProfileID(cmsHPROFILE hProfile, cmsUInt8Number* ProfileID)
+{
+    _cmsICCPROFILE*  Icc = (_cmsICCPROFILE*) hProfile;
+    memmove(&Icc -> ProfileID, ProfileID, 16);
+}
+
+cmsBool  CMSEXPORT cmsGetHeaderCreationDateTime(cmsHPROFILE hProfile, struct tm *Dest)
+{
+    _cmsICCPROFILE*  Icc = (_cmsICCPROFILE*) hProfile;    
+    memmove(Dest, &Icc ->Created, sizeof(struct tm));    
+    return TRUE;
+}
+
+cmsColorSpaceSignature CMSEXPORT cmsGetPCS(cmsHPROFILE hProfile)
+{
+    _cmsICCPROFILE*  Icc = (_cmsICCPROFILE*) hProfile;
+    return Icc -> PCS;
+}
+
+void CMSEXPORT cmsSetPCS(cmsHPROFILE hProfile, cmsColorSpaceSignature pcs)
+{
+    _cmsICCPROFILE*  Icc = (_cmsICCPROFILE*) hProfile;
+    Icc -> PCS = pcs;
+}
+
+cmsColorSpaceSignature CMSEXPORT cmsGetColorSpace(cmsHPROFILE hProfile)
+{
+    _cmsICCPROFILE*  Icc = (_cmsICCPROFILE*) hProfile;
+    return Icc -> ColorSpace;
+}
+
+void CMSEXPORT cmsSetColorSpace(cmsHPROFILE hProfile, cmsColorSpaceSignature sig)
+{
+    _cmsICCPROFILE*  Icc = (_cmsICCPROFILE*) hProfile;
+    Icc -> ColorSpace = sig;
+}
+
+cmsProfileClassSignature CMSEXPORT cmsGetDeviceClass(cmsHPROFILE hProfile)
+{
+    _cmsICCPROFILE*  Icc = (_cmsICCPROFILE*) hProfile;
+    return Icc -> DeviceClass;
+}
+
+void CMSEXPORT cmsSetDeviceClass(cmsHPROFILE hProfile, cmsProfileClassSignature sig)
+{
+    _cmsICCPROFILE*  Icc = (_cmsICCPROFILE*) hProfile;
+    Icc -> DeviceClass = sig;
+}
+
+cmsUInt32Number CMSEXPORT cmsGetEncodedICCversion(cmsHPROFILE hProfile)
+{
+    _cmsICCPROFILE*  Icc = (_cmsICCPROFILE*) hProfile;
+    return Icc -> Version;
+}
+
+void CMSEXPORT cmsSetEncodedICCversion(cmsHPROFILE hProfile, cmsUInt32Number Version)
+{
+    _cmsICCPROFILE*  Icc = (_cmsICCPROFILE*) hProfile;
+    Icc -> Version = Version;
+}
+
+// Get an hexadecimal number with same digits as v
+static
+cmsUInt32Number BaseToBase(cmsUInt32Number in, int BaseIn, int BaseOut)
+{
+    char Buff[100];
+    int i, len;
+    cmsUInt32Number out;
+    
+    for (len=0; in > 0 && len < 100; len++) {
+    
+        Buff[len] = (char) (in % BaseIn);
+        in /= BaseIn;
+    }
+
+    for (i=len-1, out=0; i >= 0; --i) {
+        out = out * BaseOut + Buff[i];
+    }
+
+    return out;
+}
+
+void  CMSEXPORT cmsSetProfileVersion(cmsHPROFILE hProfile, cmsFloat64Number Version)
+{
+    _cmsICCPROFILE*  Icc = (_cmsICCPROFILE*) hProfile;
+
+    // 4.2 -> 0x4200000
+
+    Icc -> Version = BaseToBase((cmsUInt32Number) floor(Version * 100.0), 10, 16) << 16;
+}
+
+cmsFloat64Number CMSEXPORT cmsGetProfileVersion(cmsHPROFILE hProfile)
+{
+    _cmsICCPROFILE*  Icc = (_cmsICCPROFILE*) hProfile;
+    cmsUInt32Number n = Icc -> Version >> 16;
+
+    return BaseToBase(n, 16, 10) / 100.0;
+}
+// --------------------------------------------------------------------------------------------------------------
+
+
+// Create profile from IOhandler
+cmsHPROFILE CMSEXPORT cmsOpenProfileFromIOhandlerTHR(cmsContext ContextID, cmsIOHANDLER* io)
+{
+    _cmsICCPROFILE* NewIcc;
+    cmsHPROFILE hEmpty = cmsCreateProfilePlaceholder(ContextID);
+
+    if (hEmpty == NULL) return NULL;
+
+    NewIcc = (_cmsICCPROFILE*) hEmpty;
+
+    NewIcc ->IOhandler = io;
+    if (!_cmsReadHeader(NewIcc)) goto Error;    
+    return hEmpty;
+
+Error:
+    cmsCloseProfile(hEmpty);
+    return NULL;
+}
+
+// Create profile from disk file
+cmsHPROFILE CMSEXPORT cmsOpenProfileFromFileTHR(cmsContext ContextID, const char *lpFileName, const char *sAccess)
+{
+    _cmsICCPROFILE* NewIcc;
+    cmsHPROFILE hEmpty = cmsCreateProfilePlaceholder(ContextID);
+
+    if (hEmpty == NULL) return NULL;
+
+    NewIcc = (_cmsICCPROFILE*) hEmpty;
+
+    NewIcc ->IOhandler = cmsOpenIOhandlerFromFile(ContextID, lpFileName, sAccess);
+    if (NewIcc ->IOhandler == NULL) goto Error;
+
+    if (*sAccess == 'W' || *sAccess == 'w') {
+
+        NewIcc -> IsWrite = TRUE;
+        
+        return hEmpty;
+    }
+
+    if (!_cmsReadHeader(NewIcc)) goto Error;
+    return hEmpty;
+
+Error:
+    cmsCloseProfile(hEmpty);
+    return NULL;
+}
+
+
+cmsHPROFILE CMSEXPORT cmsOpenProfileFromFile(const char *ICCProfile, const char *sAccess)
+{
+    return cmsOpenProfileFromFileTHR(NULL, ICCProfile, sAccess);
+}
+
+
+cmsHPROFILE  CMSEXPORT cmsOpenProfileFromStreamTHR(cmsContext ContextID, FILE* ICCProfile, const char *sAccess)
+{
+    _cmsICCPROFILE* NewIcc;
+    cmsHPROFILE hEmpty = cmsCreateProfilePlaceholder(ContextID);
+
+    if (hEmpty == NULL) return NULL;
+
+    NewIcc = (_cmsICCPROFILE*) hEmpty;
+
+    NewIcc ->IOhandler = cmsOpenIOhandlerFromStream(ContextID, ICCProfile);
+    if (NewIcc ->IOhandler == NULL) goto Error;
+
+    if (*sAccess == 'w') {
+
+        NewIcc -> IsWrite = TRUE;       
+        return hEmpty;
+    }
+
+    if (!_cmsReadHeader(NewIcc)) goto Error;    
+    return hEmpty;
+
+Error:
+    cmsCloseProfile(hEmpty);
+    return NULL;
+
+}
+
+cmsHPROFILE  CMSEXPORT cmsOpenProfileFromStream(FILE* ICCProfile, const char *sAccess)
+{
+    return cmsOpenProfileFromStreamTHR(NULL, ICCProfile, sAccess);
+}
+
+
+// Open from memory block
+cmsHPROFILE CMSEXPORT cmsOpenProfileFromMemTHR(cmsContext ContextID, const void* MemPtr, cmsUInt32Number dwSize)
+{
+    _cmsICCPROFILE* NewIcc;
+    cmsHPROFILE hEmpty;
+
+    hEmpty = cmsCreateProfilePlaceholder(ContextID);
+    if (hEmpty == NULL) return NULL;
+
+    NewIcc = (_cmsICCPROFILE*) hEmpty;
+
+	// Ok, in this case const void* is casted to void* just because open IO handler 
+	// shares read and writting modes. Don't abuse this feature!
+    NewIcc ->IOhandler = cmsOpenIOhandlerFromMem(ContextID, (void*) MemPtr, dwSize, "r");
+    if (NewIcc ->IOhandler == NULL) goto Error;
+
+    if (!_cmsReadHeader(NewIcc)) goto Error;
+
+    return hEmpty;
+
+Error:
+    cmsCloseProfile(hEmpty);
+    return NULL;
+}
+
+cmsHPROFILE CMSEXPORT cmsOpenProfileFromMem(const void* MemPtr, cmsUInt32Number dwSize)
+{
+    return cmsOpenProfileFromMemTHR(NULL, MemPtr, dwSize);
+}
+
+
+
+// Dump tag contents. If the profile is being modified, untouched tags are copied from FileOrig
+static
+cmsBool SaveTags(_cmsICCPROFILE* Icc, _cmsICCPROFILE* FileOrig)
+{
+    cmsUInt8Number* Data;
+    cmsUInt32Number i;
+    cmsUInt32Number Begin;
+    cmsIOHANDLER* io = Icc ->IOhandler;
+    cmsTagDescriptor* TagDescriptor;
+    cmsTagTypeSignature TypeBase;
+    cmsTagTypeHandler* TypeHandler;
+
+
+    for (i=0; i < Icc -> TagCount; i++) {
+
+       
+        if (Icc ->TagNames[i] == 0) continue;
+
+        // Linked tags are not written
+        if (Icc ->TagLinked[i] != (cmsTagSignature) 0) continue;
+
+        Icc -> TagOffsets[i] = Begin = io ->UsedSpace;
+
+        Data = (cmsUInt8Number*)  Icc -> TagPtrs[i];
+
+        if (!Data) {
+
+            // Reach here if we are copying a tag from a disk-based ICC profile which has not been modified by user. 
+            // In this case a blind copy of the block data is performed
+            if (FileOrig != NULL && Icc -> TagOffsets[i]) {
+
+                cmsUInt32Number TagSize   = FileOrig -> TagSizes[i];
+                cmsUInt32Number TagOffset = FileOrig -> TagOffsets[i];
+                void* Mem;
+
+                if (!FileOrig ->IOhandler->Seek(FileOrig ->IOhandler, TagOffset)) return FALSE;
+
+                Mem = _cmsMalloc(Icc ->ContextID, TagSize);                  
+                if (Mem == NULL) return FALSE;
+
+                if (FileOrig ->IOhandler->Read(FileOrig->IOhandler, Mem, TagSize, 1) != 1) return FALSE;
+                if (!io ->Write(io, TagSize, Mem)) return FALSE;
+                _cmsFree(Icc ->ContextID, Mem);
+
+                Icc -> TagSizes[i] = (io ->UsedSpace - Begin);
+                
+
+                // Align to 32 bit boundary.
+                if (! _cmsWriteAlignment(io)) 
+                    return FALSE;                   
+            }
+
+            continue;
+        }
+
+
+        // Should this tag be saved as RAW? If so, tagsizes should be specified in advance (no further cooking is done)
+        if (Icc ->TagSaveAsRaw[i]) {
+
+            if (io -> Write(io, Icc ->TagSizes[i], Data) != 1) return FALSE;
+        }
+        else {
+
+            // Search for support on this tag
+            TagDescriptor = _cmsGetTagDescriptor(Icc -> TagNames[i]);
+            if (TagDescriptor == NULL) continue;                        // Unsupported, ignore it
+
+            TypeHandler = Icc ->TagTypeHandlers[i];
+
+            if (TypeHandler == NULL) {
+                cmsSignalError(Icc ->ContextID, cmsERROR_INTERNAL, "(Internal) no handler for tag %x", Icc -> TagNames[i]);
+                continue;
+            }
+
+            TypeBase    = TypeHandler ->Signature;
+            if (!_cmsWriteTypeBase(io, TypeBase)) 
+                return FALSE;
+
+            TypeHandler ->ContextID  = Icc ->ContextID;
+            TypeHandler ->ICCVersion = Icc ->Version;
+            if (!TypeHandler ->WritePtr(TypeHandler, io, Data, TagDescriptor ->ElemCount)) {
+
+				char String[5];
+
+				 _cmsTagSignature2String(String, (cmsTagSignature) TypeBase);
+                cmsSignalError(Icc ->ContextID, cmsERROR_WRITE, "Couldn't write type '%s'", String);
+                return FALSE;   
+            }
+        }
+
+
+        Icc -> TagSizes[i] = (io ->UsedSpace - Begin);      
+
+        // Align to 32 bit boundary.
+        if (! _cmsWriteAlignment(io)) 
+            return FALSE;                   
+    }
+
+
+    return TRUE;
+}
+
+
+// Fill the offset and size fields for all linked tags
+static
+cmsBool SetLinks( _cmsICCPROFILE* Icc)
+{
+    cmsUInt32Number i;
+
+    for (i=0; i < Icc -> TagCount; i++) {
+
+        cmsTagSignature lnk = Icc ->TagLinked[i];
+        if (lnk != (cmsTagSignature) 0) {
+
+            int j = _cmsSearchTag(Icc, lnk, FALSE);
+            if (j >= 0) {
+
+                Icc ->TagOffsets[i] = Icc ->TagOffsets[j];
+                Icc ->TagSizes[i]   = Icc ->TagSizes[j];
+            }
+
+        }
+    }
+
+    return TRUE;
+}
+
+// Low-level save to IOHANDLER. It returns the number of bytes used to
+// store the profile, or zero on error. io may be NULL and in this case
+// no data is written--only sizes are calculated
+cmsUInt32Number CMSEXPORT cmsSaveProfileToIOhandler(cmsHPROFILE hProfile, cmsIOHANDLER* io)
+{
+    _cmsICCPROFILE* Icc = (_cmsICCPROFILE*) hProfile;
+    _cmsICCPROFILE Keep;              
+    cmsIOHANDLER* PrevIO;
+    cmsUInt32Number UsedSpace;
+    cmsContext ContextID;
+
+    memmove(&Keep, Icc, sizeof(_cmsICCPROFILE));
+
+    ContextID = cmsGetProfileContextID(hProfile);
+    PrevIO = Icc ->IOhandler = cmsOpenIOhandlerFromNULL(ContextID);
+    if (PrevIO == NULL) return 0;
+
+    // Pass #1 does compute offsets
+
+    if (!_cmsWriteHeader(Icc, 0)) return 0;
+    if (!SaveTags(Icc, &Keep)) return 0;
+
+    UsedSpace = PrevIO ->UsedSpace;
+
+    // Pass #2 does save to iohandler
+
+    if (io != NULL) {
+        Icc ->IOhandler = io;     
+        if (!SetLinks(Icc)) goto CleanUp;
+        if (!_cmsWriteHeader(Icc, UsedSpace)) goto CleanUp;
+        if (!SaveTags(Icc, &Keep)) goto CleanUp;
+    }
+
+    memmove(Icc, &Keep, sizeof(_cmsICCPROFILE));
+    if (!cmsCloseIOhandler(PrevIO)) return 0;
+
+    return UsedSpace;
+
+
+CleanUp:     
+    cmsCloseIOhandler(PrevIO);      
+    memmove(Icc, &Keep, sizeof(_cmsICCPROFILE));
+    return 0;
+}
+
+
+// Low-level save to disk. 
+cmsBool  CMSEXPORT cmsSaveProfileToFile(cmsHPROFILE hProfile, const char* FileName)
+{   
+    cmsContext ContextID = cmsGetProfileContextID(hProfile);
+    cmsIOHANDLER* io = cmsOpenIOhandlerFromFile(ContextID, FileName, "w");
+    cmsBool rc;
+    
+    if (io == NULL) return FALSE;
+
+    rc = (cmsSaveProfileToIOhandler(hProfile, io) != 0);
+    rc &= cmsCloseIOhandler(io);
+
+    if (rc == FALSE) {          // remove() is C99 per 7.19.4.1
+            remove(FileName);   // We have to IGNORE return value in this case
+    }
+    return rc;
+}
+
+// Same as anterior, but for streams
+cmsBool CMSEXPORT cmsSaveProfileToStream(cmsHPROFILE hProfile, FILE* Stream)
+{
+    cmsBool rc;
+    cmsContext ContextID = cmsGetProfileContextID(hProfile);
+    cmsIOHANDLER* io = cmsOpenIOhandlerFromStream(ContextID, Stream);
+    
+    if (io == NULL) return FALSE;
+
+    rc = (cmsSaveProfileToIOhandler(hProfile, io) != 0);
+    rc &= cmsCloseIOhandler(io);
+
+    return rc;
+}
+
+
+// Same as anterior, but for memory blocks. In this case, a NULL as MemPtr means calculate needed space only
+cmsBool CMSEXPORT cmsSaveProfileToMem(cmsHPROFILE hProfile, void *MemPtr, cmsUInt32Number* BytesNeeded)
+{
+    cmsBool rc;
+    cmsIOHANDLER* io;
+    cmsContext ContextID = cmsGetProfileContextID(hProfile);
+    
+    // Should we just calculate the needed space?
+    if (MemPtr == NULL) {   
+
+           *BytesNeeded =  cmsSaveProfileToIOhandler(hProfile, NULL);
+            return TRUE;
+    }
+
+    // That is a real write operation
+    io =  cmsOpenIOhandlerFromMem(ContextID, MemPtr, *BytesNeeded, "w");
+    if (io == NULL) return FALSE;
+
+    rc = (cmsSaveProfileToIOhandler(hProfile, io) != 0);
+    rc &= cmsCloseIOhandler(io);
+
+    return rc;
+}
+
+
+
+// Closes a profile freeing any involved resources
+cmsBool  CMSEXPORT cmsCloseProfile(cmsHPROFILE hProfile)
+{
+    _cmsICCPROFILE* Icc = (_cmsICCPROFILE*) hProfile;
+    cmsBool  rc = TRUE;
+    cmsUInt32Number i;         
+
+    if (!Icc) return FALSE;
+
+    // Was open in write mode?   
+    if (Icc ->IsWrite) {
+
+        Icc ->IsWrite = FALSE;      // Assure no further writting
+        rc &= cmsSaveProfileToFile(hProfile, Icc ->IOhandler->PhysicalFile);        
+    }
+
+    for (i=0; i < Icc -> TagCount; i++) {
+
+        if (Icc -> TagPtrs[i]) {
+
+            cmsTagTypeHandler* TypeHandler = Icc ->TagTypeHandlers[i];
+
+            if (TypeHandler != NULL) {
+
+                TypeHandler ->ContextID = Icc ->ContextID;              // As an additional parameters
+                TypeHandler ->ICCVersion = Icc ->Version;
+                TypeHandler ->FreePtr(TypeHandler, Icc -> TagPtrs[i]);       
+            }
+            else
+                _cmsFree(Icc ->ContextID, Icc ->TagPtrs[i]);
+        }
+    }
+
+    if (Icc ->IOhandler != NULL) {      
+        rc &= cmsCloseIOhandler(Icc->IOhandler);   
+    }       
+
+    _cmsFree(Icc ->ContextID, Icc);   // Free placeholder memory
+
+    return rc;
+}
+
+
+// -------------------------------------------------------------------------------------------------------------------
+
+
+// Returns TRUE if a given tag is supported by a plug-in
+static
+cmsBool IsTypeSupported(cmsTagDescriptor* TagDescriptor, cmsTagTypeSignature Type)
+{
+    cmsUInt32Number i, nMaxTypes;
+
+    nMaxTypes = TagDescriptor->nSupportedTypes;
+    if (nMaxTypes >= MAX_TYPES_IN_LCMS_PLUGIN)
+        nMaxTypes = MAX_TYPES_IN_LCMS_PLUGIN;
+    
+    for (i=0; i < nMaxTypes; i++) {
+        if (Type == TagDescriptor ->SupportedTypes[i]) return TRUE; 
+    }
+
+    return FALSE;
+}
+
+
+// That's the main read function
+void* CMSEXPORT cmsReadTag(cmsHPROFILE hProfile, cmsTagSignature sig)
+{
+    _cmsICCPROFILE* Icc = (_cmsICCPROFILE*) hProfile; 
+    cmsIOHANDLER* io = Icc ->IOhandler;
+    cmsTagTypeHandler* TypeHandler;
+    cmsTagDescriptor*  TagDescriptor;
+    cmsTagTypeSignature BaseType;
+    cmsUInt32Number Offset, TagSize;
+    cmsUInt32Number ElemCount;
+    int n;
+
+	n = _cmsSearchTag(Icc, sig, TRUE);
+	if (n < 0) return NULL;                 // Not found, return NULL
+
+
+
+	// If the element is already in memory, return the pointer
+	if (Icc -> TagPtrs[n]) {
+
+		if (Icc ->TagSaveAsRaw[n]) return NULL;  // We don't support read raw tags as cooked
+		return Icc -> TagPtrs[n];
+	}
+
+	// We need to read it. Get the offset and size to the file
+    Offset    = Icc -> TagOffsets[n];
+    TagSize   = Icc -> TagSizes[n]; 
+    
+    // Seek to its location
+    if (!io -> Seek(io, Offset))
+            return NULL;
+
+    // Search for support on this tag
+    TagDescriptor = _cmsGetTagDescriptor(sig);
+    if (TagDescriptor == NULL) return NULL;     // Unsupported. 
+
+    // if supported, get type and check if in list
+    BaseType = _cmsReadTypeBase(io);
+    if (BaseType == 0) return NULL;
+
+    if (!IsTypeSupported(TagDescriptor, BaseType)) return NULL;
+
+    TagSize  -= 8;                      // Alredy read by the type base logic
+
+    // Get type handler
+    TypeHandler = _cmsGetTagTypeHandler(BaseType);
+    if (TypeHandler == NULL) return NULL;
+
+
+    // Read the tag
+    Icc -> TagTypeHandlers[n] = TypeHandler;
+
+    TypeHandler ->ContextID = Icc ->ContextID;
+    TypeHandler ->ICCVersion = Icc ->Version;
+    Icc -> TagPtrs[n] = TypeHandler ->ReadPtr(TypeHandler, io, &ElemCount, TagSize);
+
+    // The tag type is supported, but something wrong happend and we cannot read the tag.
+    // let know the user about this (although it is just a warning)
+    if (Icc -> TagPtrs[n] == NULL) {
+
+        char String[5];
+
+        _cmsTagSignature2String(String, sig);
+        cmsSignalError(Icc ->ContextID, cmsERROR_CORRUPTION_DETECTED, "Corrupted tag '%s'", String);
+        return NULL;
+    }
+
+    // This is a weird error that may be a symptom of something more serious, the number of
+    // stored item is actually less than the number of required elements. 
+    if (ElemCount < TagDescriptor ->ElemCount) {
+
+        char String[5];
+
+        _cmsTagSignature2String(String, sig);
+        cmsSignalError(Icc ->ContextID, cmsERROR_CORRUPTION_DETECTED, "'%s' Inconsistent number of items: expected %d, got %d", 
+                                                             String, TagDescriptor ->ElemCount, ElemCount);
+    }
+
+    
+    // Return the data
+    return Icc -> TagPtrs[n];
+}
+
+
+// Get true type of data
+cmsTagTypeSignature _cmsGetTagTrueType(cmsHPROFILE hProfile, cmsTagSignature sig)
+{
+	_cmsICCPROFILE* Icc = (_cmsICCPROFILE*) hProfile; 
+	cmsTagTypeHandler* TypeHandler;	
+	int n;
+
+	// Search for given tag in ICC profile directory
+	n = _cmsSearchTag(Icc, sig, TRUE);
+	if (n < 0) return (cmsTagTypeSignature) 0;                // Not found, return NULL
+
+	// Get the handler. The true type is there
+	TypeHandler =  Icc -> TagTypeHandlers[n];
+	return TypeHandler ->Signature;
+}
+
+
+// Write a single tag. This just keeps track of the tak into a list of "to be written". If the tag is already
+// in that list, the previous version is deleted.
+cmsBool CMSEXPORT cmsWriteTag(cmsHPROFILE hProfile, cmsTagSignature sig, const void* data)
+{
+    _cmsICCPROFILE* Icc = (_cmsICCPROFILE*) hProfile;  
+    cmsTagTypeHandler* TypeHandler = NULL;
+    cmsTagDescriptor* TagDescriptor = NULL;
+    cmsTagTypeSignature Type;
+    int i;
+    cmsFloat64Number Version;
+    char TypeString[5], SigString[5];
+
+
+    if (data == NULL) {
+
+         cmsSignalError(cmsGetProfileContextID(hProfile), cmsERROR_NULL, "couldn't wite NULL to tag");
+         return FALSE;
+    }
+
+    i = _cmsSearchTag(Icc, sig, FALSE);
+    if (i >=0) {
+
+        if (Icc -> TagPtrs[i] != NULL) {
+
+            // Already exists. Free previous version
+            if (Icc ->TagSaveAsRaw[i]) {
+                _cmsFree(Icc ->ContextID, Icc ->TagPtrs[i]);
+            }
+            else {
+                TypeHandler = Icc ->TagTypeHandlers[i];
+
+                if (TypeHandler != NULL) {
+
+                    TypeHandler ->ContextID = Icc ->ContextID;              // As an additional parameter
+                    TypeHandler ->ICCVersion = Icc ->Version;
+                TypeHandler->FreePtr(TypeHandler, Icc -> TagPtrs[i]);       
+            }
+        }
+    }
+    }
+    else  {
+        // New one
+        i = Icc -> TagCount;
+
+        if (i >= MAX_TABLE_TAG) {
+            cmsSignalError(Icc ->ContextID, cmsERROR_RANGE, "Too many tags (%d)", MAX_TABLE_TAG);
+            return FALSE;
+        }
+
+        Icc -> TagCount++;
+    }
+
+    // This is not raw
+    Icc ->TagSaveAsRaw[i] = FALSE;
+
+    // This is not a link
+    Icc ->TagLinked[i] = (cmsTagSignature) 0;
+
+    // Get information about the TAG. 
+    TagDescriptor = _cmsGetTagDescriptor(sig);
+    if (TagDescriptor == NULL){
+         cmsSignalError(Icc ->ContextID, cmsERROR_UNKNOWN_EXTENSION, "Unsupported tag '%x'", sig);
+        return FALSE;
+    }
+
+    
+    // Now we need to know which type to use. It depends on the version. 
+    Version = cmsGetProfileVersion(hProfile);
+
+    if (TagDescriptor ->DecideType != NULL) {
+
+        // Let the tag descriptor to decide the type base on depending on
+        // the data. This is useful for example on parametric curves, where 
+        // curves specified by a table cannot be saved as parametric and needs
+        // to be revented to single v2-curves, even on v4 profiles.
+
+        Type = TagDescriptor ->DecideType(Version, data);
+    }
+    else {
+
+
+        Type = TagDescriptor ->SupportedTypes[0];
+    }
+
+    // Does the tag support this type?
+    if (!IsTypeSupported(TagDescriptor, Type)) {
+               
+        _cmsTagSignature2String(TypeString, (cmsTagSignature) Type);
+        _cmsTagSignature2String(SigString,  sig);
+
+        cmsSignalError(Icc ->ContextID, cmsERROR_UNKNOWN_EXTENSION, "Unsupported type '%s' for tag '%s'", TypeString, SigString);
+        return FALSE;
+    }
+
+    // Does we have a handler for this type?
+    TypeHandler =  _cmsGetTagTypeHandler(Type);
+    if (TypeHandler == NULL) {
+
+        _cmsTagSignature2String(TypeString, (cmsTagSignature) Type);
+        _cmsTagSignature2String(SigString,  sig);
+
+        cmsSignalError(Icc ->ContextID, cmsERROR_UNKNOWN_EXTENSION, "Unsupported type '%s' for tag '%s'", TypeString, SigString);
+        return FALSE;           // Should never happen
+    }
+
+
+    // Fill fields on icc structure
+    Icc ->TagTypeHandlers[i]  = TypeHandler;
+    Icc ->TagNames[i]         = sig;
+    Icc ->TagSizes[i]         = 0;
+    Icc ->TagOffsets[i]       = 0;
+
+    TypeHandler ->ContextID  = Icc ->ContextID;
+    TypeHandler ->ICCVersion = Icc ->Version;
+    Icc ->TagPtrs[i]          = TypeHandler ->DupPtr(TypeHandler, data, TagDescriptor ->ElemCount); 
+
+    if (Icc ->TagPtrs[i] == NULL)  {
+
+        _cmsTagSignature2String(TypeString, (cmsTagSignature) Type);
+        _cmsTagSignature2String(SigString,  sig);
+        cmsSignalError(Icc ->ContextID, cmsERROR_CORRUPTION_DETECTED, "Malformed struct in type '%s' for tag '%s'", TypeString, SigString);
+        
+        return FALSE;
+    }
+    
+    return TRUE;
+}
+
+// Read and write raw data. The only way those function would work and keep consistence with normal read and write 
+// is to do an additional step of serialization. That means, readRaw would issue a normal read and then convert the obtained
+// data to raw bytes by using the "write" serialization logic. And vice-versa. I know this may end in situations where
+// raw data written does not exactly correspond with the raw data proposed to cmsWriteRaw data, but this approach allows
+// to write a tag as raw data and the read it as handled.
+
+cmsInt32Number CMSEXPORT cmsReadRawTag(cmsHPROFILE hProfile, cmsTagSignature sig, void* data, cmsUInt32Number BufferSize)
+{
+    _cmsICCPROFILE* Icc = (_cmsICCPROFILE*) hProfile;  
+    void *Object; 
+    int i;
+    cmsIOHANDLER* MemIO;
+    cmsTagTypeHandler* TypeHandler = NULL;
+    cmsTagDescriptor* TagDescriptor = NULL;
+    cmsUInt32Number rc;
+    cmsUInt32Number Offset, TagSize;
+
+	// Search for given tag in ICC profile directory
+	i = _cmsSearchTag(Icc, sig, TRUE);
+	if (i < 0) return 0;                 // Not found, return 0
+
+	// It is already read?
+    if (Icc -> TagPtrs[i] == NULL) {
+
+        // No yet, get original position
+        Offset   = Icc ->TagOffsets[i];
+        TagSize  = Icc ->TagSizes[i];
+
+        
+        // read the data directly, don't keep copy
+		if (data != NULL) {
+
+			if (BufferSize < TagSize)
+                 TagSize = BufferSize;
+
+            if (!Icc ->IOhandler ->Seek(Icc ->IOhandler, Offset)) return 0;
+            if (!Icc ->IOhandler ->Read(Icc ->IOhandler, data, 1, TagSize)) return 0;
+		}
+
+        return Icc ->TagSizes[i];
+    }
+
+    // The data has been already read, or written. But wait!, maybe the user choosed to save as
+    // raw data. In this case, return the raw data directly
+    if (Icc ->TagSaveAsRaw[i]) {
+               
+		if (data != NULL)  {
+
+			 TagSize  = Icc ->TagSizes[i];
+			if (BufferSize < TagSize)
+                       TagSize = BufferSize;
+
+            memmove(data, Icc ->TagPtrs[i], TagSize);
+		}
+
+        return Icc ->TagSizes[i];
+    }
+
+    // Already readed, or previously set by cmsWriteTag(). We need to serialize that 
+    // data to raw in order to maintain consistency.
+    Object = cmsReadTag(hProfile, sig);
+    if (Object == NULL) return 0;
+
+    // Now we need to serialize to a memory block: just use a memory iohandler
+
+	if (data == NULL) {
+		  MemIO = cmsOpenIOhandlerFromNULL(cmsGetProfileContextID(hProfile));
+	} else{
+          MemIO = cmsOpenIOhandlerFromMem(cmsGetProfileContextID(hProfile), data, BufferSize, "w");
+	}
+    if (MemIO == NULL) return 0;
+
+    // Obtain type handling for the tag
+    TypeHandler = Icc ->TagTypeHandlers[i];
+    TagDescriptor = _cmsGetTagDescriptor(sig);
+
+    // Serialize
+    TypeHandler ->ContextID  = Icc ->ContextID;
+    TypeHandler ->ICCVersion = Icc ->Version;
+    if (!TypeHandler ->WritePtr(TypeHandler, MemIO, Object, TagDescriptor ->ElemCount)) return 0;
+
+    // Get Size and close
+    rc = MemIO ->Tell(MemIO);
+    cmsCloseIOhandler(MemIO);      // Ignore return code this time
+
+    return rc;
+}
+
+// Similar to the anterior. This function allows to write directly to the ICC profile any data, without
+// checking anything. As a rule, mixing Raw with cooked doesn't work, so writting a tag as raw and then reading 
+// it as cooked without serializing does result into an error. If that is wha you want, you will need to dump
+// the profile to memry or disk and then reopen it.
+cmsBool CMSEXPORT cmsWriteRawTag(cmsHPROFILE hProfile, cmsTagSignature sig, const void* data, cmsUInt32Number Size)
+{
+    _cmsICCPROFILE* Icc = (_cmsICCPROFILE*) hProfile;  
+    int i;
+
+	if (!_cmsNewTag(Icc, sig, &i)) return FALSE;
+
+    // Mark the tag as being written as RAW
+    Icc ->TagSaveAsRaw[i] = TRUE;
+    Icc ->TagNames[i]     = sig;
+    Icc ->TagLinked[i]    = (cmsTagSignature) 0;
+
+    // Keep a copy of the block
+    Icc ->TagPtrs[i]  = _cmsDupMem(Icc ->ContextID, data, Size);
+    Icc ->TagSizes[i] = Size;
+
+    return TRUE;
+}
+
+// Using this function you can collapse several tag entries to the same block in the profile
+cmsBool CMSEXPORT cmsLinkTag(cmsHPROFILE hProfile, cmsTagSignature sig, cmsTagSignature dest)
+{
+     _cmsICCPROFILE* Icc = (_cmsICCPROFILE*) hProfile;  
+    int i;
+
+	if (!_cmsNewTag(Icc, sig, &i)) return FALSE;
+
+    // Keep necessary information
+    Icc ->TagSaveAsRaw[i] = FALSE;
+    Icc ->TagNames[i]     = sig;
+    Icc ->TagLinked[i]    = dest;
+    
+    Icc ->TagPtrs[i]    = NULL;
+    Icc ->TagSizes[i]   = 0;
+    Icc ->TagOffsets[i] = 0;
+
+    return TRUE;
+}
+
+
+// Returns the tag linked to sig, in the case two tags are sharing same resource
+cmsTagSignature  CMSEXPORT cmsTagLinkedTo(cmsHPROFILE hProfile, cmsTagSignature sig)
+{
+    _cmsICCPROFILE* Icc = (_cmsICCPROFILE*) hProfile;  
+    int i;
+
+    // Search for given tag in ICC profile directory
+	i = _cmsSearchTag(Icc, sig, FALSE);
+	if (i < 0) return (cmsTagSignature) 0;                 // Not found, return 0
+
+    return Icc -> TagLinked[i];
+}
diff --git a/thirdparty/liblcms2/src/cmsio1.c b/thirdparty/liblcms2/src/cmsio1.c
new file mode 100644
index 00000000..c93eaa80
--- /dev/null
+++ b/thirdparty/liblcms2/src/cmsio1.c
@@ -0,0 +1,760 @@
+//---------------------------------------------------------------------------------
+//
+//  Little Color Management System
+//  Copyright (c) 1998-2010 Marti Maria Saguer
+//
+// Permission is hereby granted, free of charge, to any person obtaining 
+// a copy of this software and associated documentation files (the "Software"), 
+// to deal in the Software without restriction, including without limitation 
+// the rights to use, copy, modify, merge, publish, distribute, sublicense, 
+// and/or sell copies of the Software, and to permit persons to whom the Software 
+// is furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in 
+// all copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 
+// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO 
+// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE 
+// LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 
+// OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 
+// WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+//
+//---------------------------------------------------------------------------------
+//
+
+#include "lcms2_internal.h"
+
+// Read tags using low-level functions, provides necessary glue code to adapt versions, etc.
+
+// LUT tags
+static const cmsTagSignature Device2PCS16[]   =  {cmsSigAToB0Tag,     // Perceptual
+                                                  cmsSigAToB1Tag,     // Relative colorimetric
+                                                  cmsSigAToB2Tag,     // Saturation
+                                                  cmsSigAToB1Tag };   // Absolute colorimetric
+
+static const cmsTagSignature Device2PCSFloat[] = {cmsSigDToB0Tag,     // Perceptual
+                                                  cmsSigDToB1Tag,     // Relative colorimetric
+                                                  cmsSigDToB2Tag,     // Saturation
+                                                  cmsSigDToB3Tag };   // Absolute colorimetric
+
+static const cmsTagSignature PCS2Device16[]    = {cmsSigBToA0Tag,     // Perceptual
+                                                  cmsSigBToA1Tag,     // Relative colorimetric
+                                                  cmsSigBToA2Tag,     // Saturation
+                                                  cmsSigBToA1Tag };   // Absolute colorimetric
+
+static const cmsTagSignature PCS2DeviceFloat[] = {cmsSigBToD0Tag,     // Perceptual
+                                                  cmsSigBToD1Tag,     // Relative colorimetric
+                                                  cmsSigBToD2Tag,     // Saturation
+                                                  cmsSigBToD3Tag };   // Absolute colorimetric
+
+
+// Factors to convert from 1.15 fixed point to 0..1.0 range and vice-versa
+#define InpAdj   (1.0/MAX_ENCODEABLE_XYZ)     // (65536.0/(65535.0*2.0))
+#define OutpAdj  (MAX_ENCODEABLE_XYZ)         // ((2.0*65535.0)/65536.0)
+
+// Several resources for gray conversions.
+static const cmsFloat64Number GrayInputMatrix[] = { (InpAdj*cmsD50X),  (InpAdj*cmsD50Y),  (InpAdj*cmsD50Z) };
+static const cmsFloat64Number OneToThreeInputMatrix[] = { 1, 1, 1 };   
+static const cmsFloat64Number PickYMatrix[] = { 0, (OutpAdj*cmsD50Y), 0 };   
+static const cmsFloat64Number PickLstarMatrix[] = { 1, 0, 0 };   
+
+// Get a media white point fixing some issues found in certain old profiles
+cmsBool  _cmsReadMediaWhitePoint(cmsCIEXYZ* Dest, cmsHPROFILE hProfile)
+{
+    cmsCIEXYZ* Tag;
+
+    _cmsAssert(Dest != NULL);
+
+    Tag = (cmsCIEXYZ*) cmsReadTag(hProfile, cmsSigMediaWhitePointTag);   
+
+    // If no wp, take D50
+    if (Tag == NULL) {
+        *Dest = *cmsD50_XYZ();
+        return TRUE;
+    }
+
+    // V2 display profiles should give D50
+    if (cmsGetEncodedICCversion(hProfile) < 0x4000000) {
+
+        if (cmsGetDeviceClass(hProfile) == cmsSigDisplayClass) {
+            *Dest = *cmsD50_XYZ();
+            return TRUE;            
+        }
+    }
+
+    // All seems ok
+    *Dest = *Tag;
+    return TRUE;
+}
+
+
+// Chromatic adaptation matrix. Fix some issues as well
+cmsBool  _cmsReadCHAD(cmsMAT3* Dest, cmsHPROFILE hProfile)
+{
+    cmsMAT3* Tag;
+
+    _cmsAssert(Dest != NULL);
+
+    Tag = (cmsMAT3*) cmsReadTag(hProfile, cmsSigChromaticAdaptationTag);
+
+    if (Tag != NULL) {
+
+        *Dest = *Tag;
+        return TRUE;
+    }
+
+    // No CHAD available, default it to identity
+    _cmsMAT3identity(Dest);
+
+    // V2 display profiles should give D50
+    if (cmsGetEncodedICCversion(hProfile) < 0x4000000) {
+
+        if (cmsGetDeviceClass(hProfile) == cmsSigDisplayClass) {
+
+            cmsCIEXYZ* White = (cmsCIEXYZ*) cmsReadTag(hProfile, cmsSigMediaWhitePointTag);   
+
+            if (White == NULL) {
+
+                _cmsMAT3identity(Dest);
+                return TRUE;
+            }
+
+            return _cmsAdaptationMatrix(Dest, NULL, cmsD50_XYZ(), White);
+        }
+    }
+
+    return TRUE;
+}
+
+
+// Auxiliar, read colorants as a MAT3 structure. Used by any function that needs a matrix-shaper
+static
+cmsBool ReadICCMatrixRGB2XYZ(cmsMAT3* r, cmsHPROFILE hProfile)
+{
+    cmsCIEXYZ *PtrRed, *PtrGreen, *PtrBlue;
+
+    _cmsAssert(r != NULL);
+
+    PtrRed   = (cmsCIEXYZ *) cmsReadTag(hProfile, cmsSigRedColorantTag);
+    PtrGreen = (cmsCIEXYZ *) cmsReadTag(hProfile, cmsSigGreenColorantTag);
+    PtrBlue  = (cmsCIEXYZ *) cmsReadTag(hProfile, cmsSigBlueColorantTag);
+
+    if (PtrRed == NULL || PtrGreen == NULL || PtrBlue == NULL) 
+        return FALSE;
+
+    _cmsVEC3init(&r -> v[0], PtrRed -> X, PtrGreen -> X,  PtrBlue -> X);
+    _cmsVEC3init(&r -> v[1], PtrRed -> Y, PtrGreen -> Y,  PtrBlue -> Y);
+    _cmsVEC3init(&r -> v[2], PtrRed -> Z, PtrGreen -> Z,  PtrBlue -> Z);
+
+    return TRUE;
+}
+
+
+// Gray input pipeline
+static
+cmsPipeline* BuildGrayInputMatrixPipeline(cmsHPROFILE hProfile)
+{
+    cmsToneCurve *GrayTRC;
+    cmsPipeline* Lut;
+    cmsContext ContextID = cmsGetProfileContextID(hProfile);
+    
+    GrayTRC = (cmsToneCurve *) cmsReadTag(hProfile, cmsSigGrayTRCTag);
+    if (GrayTRC == NULL) return NULL;
+
+    Lut = cmsPipelineAlloc(ContextID, 1, 3);
+    if (Lut == NULL) return NULL;
+
+    if (cmsGetPCS(hProfile) == cmsSigLabData) {
+
+        // In this case we implement the profile as an  identity matrix plus 3 tone curves
+        cmsUInt16Number Zero[2] = { 0x8080, 0x8080 };
+        cmsToneCurve* EmptyTab;
+        cmsToneCurve* LabCurves[3];
+        
+        EmptyTab = cmsBuildTabulatedToneCurve16(ContextID, 2, Zero); 
+
+        if (EmptyTab == NULL) {
+
+                 cmsPipelineFree(Lut);
+                 return NULL;
+        }
+
+        LabCurves[0] = GrayTRC;
+        LabCurves[1] = EmptyTab;
+        LabCurves[2] = EmptyTab;
+
+        cmsPipelineInsertStage(Lut, cmsAT_END, cmsStageAllocMatrix(ContextID, 3,  1, OneToThreeInputMatrix, NULL));
+        cmsPipelineInsertStage(Lut, cmsAT_END, cmsStageAllocToneCurves(ContextID, 3, LabCurves));
+        
+        cmsFreeToneCurve(EmptyTab);
+
+    }
+    else  {
+       cmsPipelineInsertStage(Lut, cmsAT_END, cmsStageAllocToneCurves(ContextID, 1, &GrayTRC));
+       cmsPipelineInsertStage(Lut, cmsAT_END, cmsStageAllocMatrix(ContextID, 3,  1, GrayInputMatrix, NULL));
+    }
+  
+    return Lut;
+}
+
+// RGB Matrix shaper
+static
+cmsPipeline* BuildRGBInputMatrixShaper(cmsHPROFILE hProfile)
+{
+    cmsPipeline* Lut;
+    cmsMAT3 Mat;
+    cmsToneCurve *Shapes[3];
+    cmsContext ContextID = cmsGetProfileContextID(hProfile);
+    int i, j;
+
+    if (!ReadICCMatrixRGB2XYZ(&Mat, hProfile)) return NULL;
+
+    // XYZ PCS in encoded in 1.15 format, and the matrix output comes in 0..0xffff range, so
+    // we need to adjust the output by a factor of (0x10000/0xffff) to put data in 
+    // a 1.16 range, and then a >> 1 to obtain 1.15. The total factor is (65536.0)/(65535.0*2)
+    
+    for (i=0; i < 3; i++)
+        for (j=0; j < 3; j++)
+            Mat.v[i].n[j] *= InpAdj;
+    
+
+    Shapes[0] = (cmsToneCurve *) cmsReadTag(hProfile, cmsSigRedTRCTag);        
+    Shapes[1] = (cmsToneCurve *) cmsReadTag(hProfile, cmsSigGreenTRCTag);
+    Shapes[2] = (cmsToneCurve *) cmsReadTag(hProfile, cmsSigBlueTRCTag);
+
+    if (!Shapes[0] || !Shapes[1] || !Shapes[2])
+        return NULL;
+
+    Lut = cmsPipelineAlloc(ContextID, 3, 3);
+    if (Lut != NULL) {
+
+        cmsPipelineInsertStage(Lut, cmsAT_END, cmsStageAllocToneCurves(ContextID, 3, Shapes));
+        cmsPipelineInsertStage(Lut, cmsAT_END, cmsStageAllocMatrix(ContextID, 3, 3, (cmsFloat64Number*) &Mat, NULL));
+    }
+
+    return Lut;
+}
+
+// Read and create a BRAND NEW MPE LUT from a given profile. All stuff dependent of version, etc
+// is adjusted here in order to create a LUT that takes care of all those details
+cmsPipeline* _cmsReadInputLUT(cmsHPROFILE hProfile, int Intent)
+{
+    cmsTagTypeSignature OriginalType;
+    cmsTagSignature tag16    = Device2PCS16[Intent];
+    cmsTagSignature tagFloat = Device2PCSFloat[Intent];
+    cmsContext ContextID = cmsGetProfileContextID(hProfile);
+
+    if (cmsIsTag(hProfile, tagFloat)) {  // Float tag takes precedence
+
+        // Floating point LUT are always V4, so no adjustment is required
+        return cmsPipelineDup((cmsPipeline*) cmsReadTag(hProfile, tagFloat));
+    }
+
+    // Revert to perceptual if no tag is found
+    if (!cmsIsTag(hProfile, tag16)) {
+        tag16 = Device2PCS16[0];
+    }
+
+    if (cmsIsTag(hProfile, tag16)) { // Is there any LUT-Based table?
+
+        // Check profile version and LUT type. Do the necessary adjustments if needed
+
+        // First read the tag
+        cmsPipeline* Lut = (cmsPipeline*) cmsReadTag(hProfile, tag16);
+        if (Lut == NULL) return NULL;
+
+        // After reading it, we have now info about the original type
+        OriginalType =  _cmsGetTagTrueType(hProfile, tag16);
+
+        // The profile owns the Lut, so we need to copy it
+        Lut = cmsPipelineDup(Lut);
+
+        // We need to adjust data only for Lab16 on output
+        if (OriginalType != cmsSigLut16Type || cmsGetPCS(hProfile) != cmsSigLabData) 
+            return Lut;
+
+        // Add a matrix for conversion V2 to V4 Lab PCS
+        cmsPipelineInsertStage(Lut, cmsAT_END, _cmsStageAllocLabV2ToV4(ContextID));
+        return Lut;
+    }   
+
+    // Lut was not found, try to create a matrix-shaper
+
+    // Check if this is a grayscale profile.
+    if (cmsGetColorSpace(hProfile) == cmsSigGrayData) {
+
+        // if so, build appropiate conversion tables. 
+        // The tables are the PCS iluminant, scaled across GrayTRC
+        return BuildGrayInputMatrixPipeline(hProfile);              
+    }
+
+    // Not gray, create a normal matrix-shaper 
+    return BuildRGBInputMatrixShaper(hProfile);
+}
+
+// ---------------------------------------------------------------------------------------------------------------
+
+// Gray output pipeline. 
+// XYZ -> Gray or Lab -> Gray. Since we only know the GrayTRC, we need to do some assumptions. Gray component will be
+// given by Y on XYZ PCS and by L* on Lab PCS, Both across inverse TRC curve.
+// The complete pipeline on XYZ is Matrix[3:1] -> Tone curve and in Lab Matrix[3:1] -> Tone Curve as well.
+
+static
+cmsPipeline* BuildGrayOutputPipeline(cmsHPROFILE hProfile)
+{
+    cmsToneCurve *GrayTRC, *RevGrayTRC;
+    cmsPipeline* Lut;
+    cmsContext ContextID = cmsGetProfileContextID(hProfile);
+
+    GrayTRC = (cmsToneCurve *) cmsReadTag(hProfile, cmsSigGrayTRCTag);       
+    if (GrayTRC == NULL) return NULL;
+
+    RevGrayTRC = cmsReverseToneCurve(GrayTRC);
+    if (RevGrayTRC == NULL) return NULL;
+
+    Lut = cmsPipelineAlloc(ContextID, 3, 1);
+    if (Lut == NULL) {
+        cmsFreeToneCurve(RevGrayTRC);
+        return NULL;
+    }
+
+    if (cmsGetPCS(hProfile) == cmsSigLabData) {
+
+        cmsPipelineInsertStage(Lut, cmsAT_END, cmsStageAllocMatrix(ContextID, 1,  3, PickLstarMatrix, NULL));
+    }
+    else  {
+        cmsPipelineInsertStage(Lut, cmsAT_END, cmsStageAllocMatrix(ContextID, 1,  3, PickYMatrix, NULL));
+    }
+
+    cmsPipelineInsertStage(Lut, cmsAT_END, cmsStageAllocToneCurves(ContextID, 1, &RevGrayTRC));
+    cmsFreeToneCurve(RevGrayTRC);
+
+    return Lut;
+}
+
+
+
+
+static
+cmsPipeline* BuildRGBOutputMatrixShaper(cmsHPROFILE hProfile)
+{
+    cmsPipeline* Lut;
+    cmsToneCurve *Shapes[3], *InvShapes[3];
+    cmsMAT3 Mat, Inv;
+    int i, j;
+    cmsContext ContextID = cmsGetProfileContextID(hProfile);
+
+    if (!ReadICCMatrixRGB2XYZ(&Mat, hProfile))
+        return NULL;
+
+    if (!_cmsMAT3inverse(&Mat, &Inv))
+        return NULL;
+
+    // XYZ PCS in encoded in 1.15 format, and the matrix input should come in 0..0xffff range, so
+    // we need to adjust the input by a << 1 to obtain a 1.16 fixed and then by a factor of 
+    // (0xffff/0x10000) to put data in 0..0xffff range. Total factor is (2.0*65535.0)/65536.0;
+
+    for (i=0; i < 3; i++)
+        for (j=0; j < 3; j++)
+            Inv.v[i].n[j] *= OutpAdj;
+
+    Shapes[0] = (cmsToneCurve *) cmsReadTag(hProfile, cmsSigRedTRCTag);        
+    Shapes[1] = (cmsToneCurve *) cmsReadTag(hProfile, cmsSigGreenTRCTag);
+    Shapes[2] = (cmsToneCurve *) cmsReadTag(hProfile, cmsSigBlueTRCTag);
+
+    if (!Shapes[0] || !Shapes[1] || !Shapes[2])
+        return NULL;
+
+    InvShapes[0] = cmsReverseToneCurve(Shapes[0]);
+    InvShapes[1] = cmsReverseToneCurve(Shapes[1]);
+    InvShapes[2] = cmsReverseToneCurve(Shapes[2]);
+
+    if (!InvShapes[0] || !InvShapes[1] || !InvShapes[2]) {      
+        return NULL;
+    }
+
+    Lut = cmsPipelineAlloc(ContextID, 3, 3);
+    if (Lut != NULL) {
+
+        cmsPipelineInsertStage(Lut, cmsAT_END, cmsStageAllocMatrix(ContextID, 3, 3, (cmsFloat64Number*) &Inv, NULL));
+        cmsPipelineInsertStage(Lut, cmsAT_END, cmsStageAllocToneCurves(ContextID, 3, InvShapes));
+    }
+
+    cmsFreeToneCurveTriple(InvShapes);
+    return Lut;
+}
+
+
+// Change CLUT interpolation to trilinear
+static
+void ChangeInterpolationToTrilinear(cmsPipeline* Lut)
+{
+    cmsStage* Stage;
+
+    for (Stage = cmsPipelineGetPtrToFirstStage(Lut);
+        Stage != NULL;
+        Stage = cmsStageNext(Stage)) {
+
+            if (cmsStageType(Stage) == cmsSigCLutElemType) {
+
+                _cmsStageCLutData* CLUT = (_cmsStageCLutData*) Stage ->Data;
+
+                CLUT ->Params->dwFlags |= CMS_LERP_FLAGS_TRILINEAR;
+                _cmsSetInterpolationRoutine(CLUT ->Params);
+            }
+    }
+}
+
+// Create an output MPE LUT from agiven profile. Version mismatches are handled here
+cmsPipeline* _cmsReadOutputLUT(cmsHPROFILE hProfile, int Intent)
+{
+    cmsTagTypeSignature OriginalType;
+    cmsTagSignature tag16    = PCS2Device16[Intent];
+    cmsTagSignature tagFloat = PCS2DeviceFloat[Intent];
+    cmsContext ContextID     = cmsGetProfileContextID(hProfile);
+
+    if (cmsIsTag(hProfile, tagFloat)) {  // Float tag takes precedence
+
+        // Floating point LUT are always V4, so no adjustment is required
+        return cmsPipelineDup((cmsPipeline*) cmsReadTag(hProfile, tagFloat));
+    }
+
+    // Revert to perceptual if no tag is found
+    if (!cmsIsTag(hProfile, tag16)) {
+        tag16 = PCS2Device16[0];
+    }
+
+    if (cmsIsTag(hProfile, tag16)) { // Is there any LUT-Based table?
+
+        // Check profile version and LUT type. Do the necessary adjustments if needed
+
+        // First read the tag
+        cmsPipeline* Lut = (cmsPipeline*) cmsReadTag(hProfile, tag16);
+        if (Lut == NULL) return NULL;
+
+        // After reading it, we have info about the original type
+        OriginalType =  _cmsGetTagTrueType(hProfile, tag16);
+
+        // The profile owns the Lut, so we need to copy it
+        Lut = cmsPipelineDup(Lut);
+        if (Lut == NULL) return NULL;
+
+        // Now it is time for a controversial stuff. I found that for 3D LUTS using 
+        // Lab used as indexer space,  trilinear interpolation should be used         
+        if (cmsGetPCS(hProfile) == cmsSigLabData)
+                             ChangeInterpolationToTrilinear(Lut);       
+
+        // We need to adjust data only for Lab and Lut16 type
+        if (OriginalType != cmsSigLut16Type || cmsGetPCS(hProfile) != cmsSigLabData) 
+            return Lut;
+
+        // Add a matrix for conversion V4 to V2 Lab PCS
+        cmsPipelineInsertStage(Lut, cmsAT_BEGIN, _cmsStageAllocLabV4ToV2(ContextID));
+        return Lut;
+    }   
+
+    // Lut not found, try to create a matrix-shaper
+
+    // Check if this is a grayscale profile.
+     if (cmsGetColorSpace(hProfile) == cmsSigGrayData) {
+
+              // if so, build appropiate conversion tables. 
+              // The tables are the PCS iluminant, scaled across GrayTRC
+              return BuildGrayOutputPipeline(hProfile);              
+    }
+
+    // Not gray, create a normal matrix-shaper 
+    return BuildRGBOutputMatrixShaper(hProfile);
+}
+
+// ---------------------------------------------------------------------------------------------------------------
+
+// This one includes abstract profiles as well. Matrix-shaper cannot be obtained on that device class. The 
+// tag name here may default to AToB0
+cmsPipeline* _cmsReadDevicelinkLUT(cmsHPROFILE hProfile, int Intent)
+{
+    cmsPipeline* Lut;
+    cmsTagTypeSignature OriginalType;
+    cmsTagSignature tag16    = Device2PCS16[Intent];
+    cmsTagSignature tagFloat = Device2PCSFloat[Intent];
+    cmsContext ContextID = cmsGetProfileContextID(hProfile);
+
+    if (cmsIsTag(hProfile, tagFloat)) {  // Float tag takes precedence
+
+        // Floating point LUT are always V4, no adjustment is required
+        return cmsPipelineDup((cmsPipeline*) cmsReadTag(hProfile, tagFloat));
+    }
+
+    tagFloat = Device2PCSFloat[0];
+    if (cmsIsTag(hProfile, tagFloat)) {  
+        
+        return cmsPipelineDup((cmsPipeline*) cmsReadTag(hProfile, tagFloat));
+    }
+
+    if (!cmsIsTag(hProfile, tag16)) {  // Is there any LUT-Based table?
+        
+        tag16    = Device2PCS16[0];
+        if (!cmsIsTag(hProfile, tag16)) return NULL;        
+    }
+
+    // Check profile version and LUT type. Do the necessary adjustments if needed
+
+    // Read the tag
+    Lut = (cmsPipeline*) cmsReadTag(hProfile, tag16);
+    if (Lut == NULL) return NULL;
+
+    // The profile owns the Lut, so we need to copy it
+    Lut = cmsPipelineDup(Lut);
+    if (Lut == NULL) return NULL;
+
+     // Now it is time for a controversial stuff. I found that for 3D LUTS using 
+     // Lab used as indexer space,  trilinear interpolation should be used         
+    if (cmsGetColorSpace(hProfile) == cmsSigLabData)
+                        ChangeInterpolationToTrilinear(Lut);    
+
+    // After reading it, we have info about the original type
+    OriginalType =  _cmsGetTagTrueType(hProfile, tag16);
+
+    // We need to adjust data for Lab16 on output
+    if (OriginalType != cmsSigLut16Type) return Lut;
+        
+    // Here it is possible to get Lab on both sides
+
+    if (cmsGetPCS(hProfile) == cmsSigLabData) {
+            cmsPipelineInsertStage(Lut, cmsAT_BEGIN, _cmsStageAllocLabV4ToV2(ContextID));
+    }
+
+    if (cmsGetColorSpace(hProfile) == cmsSigLabData) {
+            cmsPipelineInsertStage(Lut, cmsAT_END, _cmsStageAllocLabV2ToV4(ContextID));
+    }
+
+    return Lut;
+
+
+}
+
+// ---------------------------------------------------------------------------------------------------------------
+
+// Returns TRUE if the profile is implemented as matrix-shaper
+cmsBool  CMSEXPORT cmsIsMatrixShaper(cmsHPROFILE hProfile)
+{    
+    switch (cmsGetColorSpace(hProfile)) {
+
+    case cmsSigGrayData:
+        
+        return cmsIsTag(hProfile, cmsSigGrayTRCTag);
+
+    case cmsSigRgbData:
+
+        return (cmsIsTag(hProfile, cmsSigRedColorantTag) &&
+                cmsIsTag(hProfile, cmsSigGreenColorantTag) &&
+                cmsIsTag(hProfile, cmsSigBlueColorantTag) &&
+                cmsIsTag(hProfile, cmsSigRedTRCTag) &&
+                cmsIsTag(hProfile, cmsSigGreenTRCTag) &&
+                cmsIsTag(hProfile, cmsSigBlueTRCTag));
+
+    default:
+
+        return FALSE;
+    }
+}
+
+// Returns TRUE if the intent is implemented as CLUT
+cmsBool  CMSEXPORT cmsIsCLUT(cmsHPROFILE hProfile, cmsUInt32Number Intent, cmsUInt32Number UsedDirection)
+{    
+    const cmsTagSignature* TagTable;
+
+    // For devicelinks, the supported intent is that one stated in the header
+    if (cmsGetDeviceClass(hProfile) == cmsSigLinkClass) {
+            return (cmsGetHeaderRenderingIntent(hProfile) == Intent);
+    }
+
+    switch (UsedDirection) {
+
+       case LCMS_USED_AS_INPUT: TagTable = Device2PCS16; break;
+       case LCMS_USED_AS_OUTPUT:TagTable = PCS2Device16; break; 
+
+       // For proofing, we need rel. colorimetric in output. Let's do some recursion
+       case LCMS_USED_AS_PROOF: 
+           return cmsIsIntentSupported(hProfile, Intent, LCMS_USED_AS_INPUT) &&
+                  cmsIsIntentSupported(hProfile, INTENT_RELATIVE_COLORIMETRIC, LCMS_USED_AS_OUTPUT);
+
+       default:
+           cmsSignalError(cmsGetProfileContextID(hProfile), cmsERROR_RANGE, "Unexpected direction (%d)", UsedDirection);
+           return FALSE;
+    }
+
+    return cmsIsTag(hProfile, TagTable[Intent]);
+
+}
+
+
+// Return info about supported intents
+cmsBool  CMSEXPORT cmsIsIntentSupported(cmsHPROFILE hProfile,
+                                        cmsUInt32Number Intent, cmsUInt32Number UsedDirection)
+{
+
+    if (cmsIsCLUT(hProfile, Intent, UsedDirection)) return TRUE;
+
+    // Is there any matrix-shaper? If so, the intent is supported. This is a bit odd, since V2 matrix shaper
+    // does not fully support relative colorimetric because they cannot deal with non-zero black points, but
+    // many profiles claims that, and this is certainly not true for V4 profiles. Lets answer "yes" no matter
+    // the accuracy would be less than optimal in rel.col and v2 case.
+
+    return cmsIsMatrixShaper(hProfile);
+}
+
+
+// ---------------------------------------------------------------------------------------------------------------
+
+// Read both, profile sequence description and profile sequence id if present. Then combine both to
+// create qa unique structure holding both. Shame on ICC to store things in such complicated way.
+
+cmsSEQ* _cmsReadProfileSequence(cmsHPROFILE hProfile)
+{
+    cmsSEQ* ProfileSeq;
+    cmsSEQ* ProfileId;
+    cmsSEQ* NewSeq;
+    cmsUInt32Number i;
+
+    // Take profile sequence description first
+    ProfileSeq = (cmsSEQ*) cmsReadTag(hProfile, cmsSigProfileSequenceDescTag);
+    
+    // Take profile sequence ID
+    ProfileId  = (cmsSEQ*) cmsReadTag(hProfile, cmsSigProfileSequenceIdTag);
+
+    if (ProfileSeq == NULL && ProfileId == NULL) return NULL;
+
+    if (ProfileSeq == NULL) return cmsDupProfileSequenceDescription(ProfileId);
+    if (ProfileId  == NULL) return cmsDupProfileSequenceDescription(ProfileSeq);
+
+    // We have to mix both together. For that they must agree 
+    if (ProfileSeq ->n != ProfileId ->n) return cmsDupProfileSequenceDescription(ProfileSeq);
+
+    NewSeq = cmsDupProfileSequenceDescription(ProfileSeq);
+    
+    // Ok, proceed to the mixing
+    for (i=0; i < ProfileSeq ->n; i++) {
+    
+        memmove(&NewSeq ->seq[i].ProfileID, &ProfileId ->seq[i].ProfileID, sizeof(cmsProfileID));
+        NewSeq ->seq[i].Description = cmsMLUdup(ProfileId ->seq[i].Description);
+    }
+
+    return NewSeq;
+}
+
+// Dump the contents of profile sequence in both tags (if v4 available)
+cmsBool _cmsWriteProfileSequence(cmsHPROFILE hProfile, const cmsSEQ* seq)
+{
+    if (!cmsWriteTag(hProfile, cmsSigProfileSequenceDescTag, seq)) return FALSE;
+
+    if (cmsGetProfileVersion(hProfile) >= 4.0) {
+
+            if (!cmsWriteTag(hProfile, cmsSigProfileSequenceIdTag, seq)) return FALSE;
+    }
+
+    return TRUE;
+}
+
+
+// Auxiliar, read and duplicate a MLU if found.
+static
+cmsMLU* GetMLUFromProfile(cmsHPROFILE h, cmsTagSignature sig)
+{
+    cmsMLU* mlu = (cmsMLU*) cmsReadTag(h, sig);
+    if (mlu == NULL) return NULL;
+
+    return cmsMLUdup(mlu);
+}
+
+// Create a sequence description out of an array of profiles
+cmsSEQ* _cmsCompileProfileSequence(cmsContext ContextID, cmsUInt32Number nProfiles, cmsHPROFILE hProfiles[])
+{
+    cmsUInt32Number i;
+    cmsSEQ* seq = cmsAllocProfileSequenceDescription(ContextID, nProfiles);
+
+    if (seq == NULL) return NULL;
+
+    for (i=0; i < nProfiles; i++) {
+
+        cmsPSEQDESC* ps = &seq ->seq[i];
+        cmsHPROFILE h = hProfiles[i];
+        cmsTechnologySignature* techpt;
+        
+        cmsGetHeaderAttributes(h, &ps ->attributes);
+        cmsGetHeaderProfileID(h, ps ->ProfileID.ID8);       
+        ps ->deviceMfg   = cmsGetHeaderManufacturer(h);
+        ps ->deviceModel = cmsGetHeaderModel(h);
+        
+        techpt = (cmsTechnologySignature*) cmsReadTag(h, cmsSigTechnologyTag);
+        if (techpt == NULL)
+            ps ->technology   =  (cmsTechnologySignature) 0;
+        else
+            ps ->technology   = *techpt;
+                
+        ps ->Manufacturer = GetMLUFromProfile(h,  cmsSigDeviceMfgDescTag);
+        ps ->Model        = GetMLUFromProfile(h,  cmsSigDeviceModelDescTag);        
+        ps ->Description  = GetMLUFromProfile(h, cmsSigProfileDescriptionTag);
+        
+    }
+
+    return seq;
+}
+
+// -------------------------------------------------------------------------------------------------------------------
+
+
+static
+const cmsMLU* GetInfo(cmsHPROFILE hProfile, cmsInfoType Info)
+{
+    cmsTagSignature sig;
+
+    switch (Info) {
+
+    case cmsInfoDescription:
+        sig = cmsSigProfileDescriptionTag; 
+        break;
+
+    case cmsInfoManufacturer:
+        sig = cmsSigDeviceMfgDescTag;
+        break;
+
+    case cmsInfoModel:
+        sig = cmsSigDeviceModelDescTag;
+         break;
+
+    case cmsInfoCopyright:
+        sig = cmsSigCopyrightTag;
+        break;
+
+    default: return NULL;
+    }
+
+
+    return (cmsMLU*) cmsReadTag(hProfile, sig);
+}
+
+
+
+cmsUInt32Number CMSEXPORT cmsGetProfileInfo(cmsHPROFILE hProfile, cmsInfoType Info, 
+                                            const char LanguageCode[3], const char CountryCode[3], 
+                                            wchar_t* Buffer, cmsUInt32Number BufferSize)
+{
+    const cmsMLU* mlu = GetInfo(hProfile, Info);
+    if (mlu == NULL) return 0;
+
+    return cmsMLUgetWide(mlu, LanguageCode, CountryCode, Buffer, BufferSize);
+}
+
+
+cmsUInt32Number  CMSEXPORT cmsGetProfileInfoASCII(cmsHPROFILE hProfile, cmsInfoType Info, 
+                                                          const char LanguageCode[3], const char CountryCode[3], 
+                                                          char* Buffer, cmsUInt32Number BufferSize)
+{
+    const cmsMLU* mlu = GetInfo(hProfile, Info);
+    if (mlu == NULL) return 0;
+
+    return cmsMLUgetASCII(mlu, LanguageCode, CountryCode, Buffer, BufferSize);
+}
diff --git a/thirdparty/liblcms2/src/cmslut.c b/thirdparty/liblcms2/src/cmslut.c
new file mode 100644
index 00000000..d0fe9c80
--- /dev/null
+++ b/thirdparty/liblcms2/src/cmslut.c
@@ -0,0 +1,1665 @@
+//---------------------------------------------------------------------------------
+//
+//  Little Color Management System
+//  Copyright (c) 1998-2010 Marti Maria Saguer
+//
+// Permission is hereby granted, free of charge, to any person obtaining 
+// a copy of this software and associated documentation files (the "Software"), 
+// to deal in the Software without restriction, including without limitation 
+// the rights to use, copy, modify, merge, publish, distribute, sublicense, 
+// and/or sell copies of the Software, and to permit persons to whom the Software 
+// is furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in 
+// all copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 
+// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO 
+// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE 
+// LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 
+// OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 
+// WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+//
+//---------------------------------------------------------------------------------
+//
+
+#include "lcms2_internal.h"
+
+
+// Allocates an empty multi profile element
+cmsStage* CMSEXPORT _cmsStageAllocPlaceholder(cmsContext ContextID, 
+                                cmsStageSignature Type,
+                                cmsUInt32Number InputChannels, 
+                                cmsUInt32Number OutputChannels,
+                                _cmsStageEvalFn     EvalPtr,           
+                                _cmsStageDupElemFn  DupElemPtr,        
+                                _cmsStageFreeElemFn FreePtr,           
+                                void*             Data)             
+{
+    cmsStage* ph = (cmsStage*) _cmsMallocZero(ContextID, sizeof(cmsStage));
+
+    if (ph == NULL) return NULL;
+    
+    
+    ph ->ContextID = ContextID;
+
+    ph ->Type       = Type;
+    ph ->Implements = Type;   // By default, no clue on what is implementing
+
+    ph ->InputChannels  = InputChannels;
+    ph ->OutputChannels = OutputChannels;
+    ph ->EvalPtr        = EvalPtr;      
+    ph ->DupElemPtr     = DupElemPtr;   
+    ph ->FreePtr        = FreePtr;               
+    ph ->Data           = Data;
+
+    return ph; 
+}
+
+
+static
+void EvaluateIdentity(const cmsFloat32Number In[], 
+                            cmsFloat32Number Out[], 
+                      const cmsStage *mpe)
+{
+    memmove(Out, In, mpe ->InputChannels * sizeof(cmsFloat32Number));
+}
+
+
+cmsStage* CMSEXPORT cmsStageAllocIdentity(cmsContext ContextID, cmsUInt32Number nChannels)
+{
+    return _cmsStageAllocPlaceholder(ContextID, 
+                                   cmsSigIdentityElemType, 
+                                   nChannels, nChannels,
+                                   EvaluateIdentity,                            
+                                   NULL,
+                                   NULL,
+                                   NULL);
+ }
+
+// Conversion functions. From floating point to 16 bits
+static
+void FromFloatTo16(const cmsFloat32Number In[], cmsUInt16Number Out[], cmsUInt32Number n)
+{
+    cmsUInt32Number i;
+
+    for (i=0; i < n; i++) {     
+        Out[i] = _cmsQuickSaturateWord(In[i] * 65535.0);        
+    }
+}
+
+// From 16 bits to floating point
+static
+void From16ToFloat(const cmsUInt16Number In[], cmsFloat32Number Out[], cmsUInt32Number n)
+{
+    cmsUInt32Number i;
+
+    for (i=0; i < n; i++) {
+        Out[i] = (cmsFloat32Number) In[i] / 65535.0F;
+    }
+}
+
+
+// This function is quite useful to analyze the structure of a LUT and retrieve the MPE elements
+// that conform the LUT. It should be called with the LUT, the number of expected elements and
+// then a list of expected types followed with a list of cmsFloat64Number pointers to MPE elements. If
+// the function founds a match with current pipeline, it fills the pointers and returns TRUE
+// if not, returns FALSE without touching anything. Setting pointers to NULL does bypass 
+// the storage process.
+cmsBool  CMSEXPORT cmsPipelineCheckAndRetreiveStages(const cmsPipeline* Lut, cmsUInt32Number n, ...)
+{
+    va_list args;
+    cmsUInt32Number i;
+    cmsStage* mpe;
+    cmsStageSignature Type;
+    void** ElemPtr;
+
+    // Make sure same number of elements
+    if (cmsPipelineStageCount(Lut) != n) return FALSE;
+
+    va_start(args, n);
+
+    // Iterate across asked types
+    mpe = Lut ->Elements;
+    for (i=0; i < n; i++) {
+
+        // Get asked type
+        Type  = va_arg(args, cmsStageSignature);      
+        if (mpe ->Type != Type) {
+
+            va_end(args);       // Mismatch. We are done.
+            return FALSE;       
+        }
+        mpe = mpe ->Next;
+    }
+
+    // Found a combination, fill pointers if not NULL
+    mpe = Lut ->Elements;
+    for (i=0; i < n; i++) {
+
+        ElemPtr = va_arg(args, void**);     
+        if (ElemPtr != NULL) 
+            *ElemPtr = mpe;
+
+        mpe = mpe ->Next;
+    }
+
+    va_end(args);   
+    return TRUE;
+}
+
+// Below there are implementations for several types of elements. Each type may be implemented by a
+// evaluation function, a duplication function, a function to free resources and a constructor.
+
+// *************************************************************************************************
+// Type cmsSigCurveSetElemType (curves)
+// *************************************************************************************************
+
+cmsToneCurve** _cmsStageGetPtrToCurveSet(const cmsStage* mpe)
+{
+    _cmsStageToneCurvesData* Data = (_cmsStageToneCurvesData*) mpe ->Data;
+
+    return Data ->TheCurves;
+}
+
+static
+void EvaluateCurves(const cmsFloat32Number In[], 
+                    cmsFloat32Number Out[], 
+                    const cmsStage *mpe)
+{
+    _cmsStageToneCurvesData* Data;
+    cmsUInt32Number i;
+
+    _cmsAssert(mpe != NULL);
+
+    Data = (_cmsStageToneCurvesData*) mpe ->Data;
+    if (Data == NULL) return;
+
+    if (Data ->TheCurves == NULL) return;
+
+    for (i=0; i < Data ->nCurves; i++) {
+        Out[i] = cmsEvalToneCurveFloat(Data ->TheCurves[i], In[i]);
+    }
+}
+
+static
+void CurveSetElemTypeFree(cmsStage* mpe)
+{
+    _cmsStageToneCurvesData* Data;
+    cmsUInt32Number i;
+
+    _cmsAssert(mpe != NULL);
+
+    Data = (_cmsStageToneCurvesData*) mpe ->Data;
+    if (Data == NULL) return;
+
+    if (Data ->TheCurves != NULL) {
+        for (i=0; i < Data ->nCurves; i++) {
+            if (Data ->TheCurves[i] != NULL) 
+                cmsFreeToneCurve(Data ->TheCurves[i]);
+        }
+    }
+    _cmsFree(mpe ->ContextID, Data ->TheCurves);
+    _cmsFree(mpe ->ContextID, Data);
+}
+
+
+static
+void* CurveSetDup(cmsStage* mpe)
+{
+    _cmsStageToneCurvesData* Data = (_cmsStageToneCurvesData*) mpe ->Data;
+    _cmsStageToneCurvesData* NewElem;
+    cmsUInt32Number i;
+
+    NewElem = (_cmsStageToneCurvesData*) _cmsMallocZero(mpe ->ContextID, sizeof(_cmsStageToneCurvesData));
+    if (NewElem == NULL) return NULL;
+
+    NewElem ->nCurves   = Data ->nCurves;
+    NewElem ->TheCurves = (cmsToneCurve**) _cmsCalloc(mpe ->ContextID, NewElem ->nCurves, sizeof(cmsToneCurve*));
+
+    if (NewElem ->TheCurves == NULL) goto Error;
+
+    for (i=0; i < NewElem ->nCurves; i++) {
+
+        // Duplicate each curve. It may fail.
+        NewElem ->TheCurves[i] = cmsDupToneCurve(Data ->TheCurves[i]);
+        if (NewElem ->TheCurves[i] == NULL) goto Error;
+
+
+    }
+    return (void*) NewElem;
+
+Error:
+
+    if (NewElem ->TheCurves != NULL) { 
+        for (i=0; i < NewElem ->nCurves; i++) {
+            if (NewElem ->TheCurves[i])
+                cmsFreeToneCurve(Data ->TheCurves[i]);
+        }
+    }
+    _cmsFree(mpe ->ContextID, Data ->TheCurves);
+    _cmsFree(mpe ->ContextID, NewElem);
+    return NULL;
+}
+
+
+// Curves == NULL forces identity curves
+cmsStage* CMSEXPORT cmsStageAllocToneCurves(cmsContext ContextID, cmsUInt32Number nChannels, cmsToneCurve* const Curves[])
+{
+    cmsUInt32Number i;
+    _cmsStageToneCurvesData* NewElem;
+    cmsStage* NewMPE;
+    
+    
+    NewMPE = _cmsStageAllocPlaceholder(ContextID, cmsSigCurveSetElemType, nChannels, nChannels,
+                                     EvaluateCurves, CurveSetDup, CurveSetElemTypeFree, NULL );
+    if (NewMPE == NULL) return NULL;
+
+    NewElem = (_cmsStageToneCurvesData*) _cmsMalloc(ContextID, sizeof(_cmsStageToneCurvesData));
+    if (NewElem == NULL) {
+        cmsStageFree(NewMPE); 
+        return NULL;
+    }
+
+    NewMPE ->Data  = (void*) NewElem;
+
+    NewElem ->nCurves   = nChannels;
+    NewElem ->TheCurves = (cmsToneCurve**) _cmsCalloc(ContextID, nChannels, sizeof(cmsToneCurve*));
+    if (NewElem ->TheCurves == NULL) {
+        cmsStageFree(NewMPE); 
+        return NULL;
+    }
+
+    for (i=0; i < nChannels; i++) {
+
+        if (Curves == NULL) {
+            NewElem ->TheCurves[i] = cmsBuildGamma(ContextID, 1.0);
+        }
+        else {
+            NewElem ->TheCurves[i] = cmsDupToneCurve(Curves[i]);
+        }
+
+        if (NewElem ->TheCurves[i] == NULL) {
+            cmsStageFree(NewMPE); 
+            return NULL;
+        }
+    }
+
+    return NewMPE;
+}
+
+
+// Create a bunch of identity curves
+cmsStage* _cmsStageAllocIdentityCurves(cmsContext ContextID, int nChannels)
+{
+    cmsStage* mpe = cmsStageAllocToneCurves(ContextID, nChannels, NULL);
+    
+    if (mpe == NULL) return NULL;
+    mpe ->Implements = cmsSigIdentityElemType;
+    return mpe;
+}
+
+
+// *************************************************************************************************
+// Type cmsSigMatrixElemType (Matrices)
+// *************************************************************************************************
+
+
+// Special care should be taken here because precision loss. A temporary cmsFloat64Number buffer is being used
+static
+void EvaluateMatrix(const cmsFloat32Number In[], 
+                    cmsFloat32Number Out[], 
+                    const cmsStage *mpe)
+{
+    cmsUInt32Number i, j;
+    _cmsStageMatrixData* Data = (_cmsStageMatrixData*) mpe ->Data;
+    cmsFloat64Number Tmp;
+
+    // Input is already in 0..1.0 notation
+    for (i=0; i < mpe ->OutputChannels; i++) {
+
+        Tmp = 0;
+        for (j=0; j < mpe->InputChannels; j++) {
+            Tmp += In[j] * Data->Double[i*mpe->InputChannels + j];
+        }
+
+        if (Data ->Offset != NULL) 
+            Tmp += Data->Offset[i];
+
+        Out[i] = (cmsFloat32Number) Tmp;     
+    }
+
+
+    // Output in 0..1.0 domain
+}
+
+
+// Duplicate a yet-existing matrix element
+static
+void* MatrixElemDup(cmsStage* mpe)
+{
+    _cmsStageMatrixData* Data = (_cmsStageMatrixData*) mpe ->Data;
+    _cmsStageMatrixData* NewElem;
+    cmsUInt32Number sz;
+    
+    NewElem = (_cmsStageMatrixData*) _cmsMallocZero(mpe ->ContextID, sizeof(_cmsStageMatrixData));
+    if (NewElem == NULL) return NULL;
+
+    sz = mpe ->InputChannels * mpe ->OutputChannels;
+
+    NewElem ->Double = (cmsFloat64Number*) _cmsDupMem(mpe ->ContextID, Data ->Double, sz * sizeof(cmsFloat64Number)) ;
+
+    if (Data ->Offset)
+        NewElem ->Offset = (cmsFloat64Number*) _cmsDupMem(mpe ->ContextID, 
+                                                Data ->Offset, mpe -> OutputChannels * sizeof(cmsFloat64Number)) ;
+
+    return (void*) NewElem;
+}
+
+
+static
+void MatrixElemTypeFree(cmsStage* mpe)
+{
+    _cmsStageMatrixData* Data = (_cmsStageMatrixData*) mpe ->Data;
+    if (Data ->Double)
+        _cmsFree(mpe ->ContextID, Data ->Double);
+
+    if (Data ->Offset)
+        _cmsFree(mpe ->ContextID, Data ->Offset);
+
+    _cmsFree(mpe ->ContextID, mpe ->Data);
+}
+
+
+
+cmsStage*  CMSEXPORT cmsStageAllocMatrix(cmsContext ContextID, cmsUInt32Number Rows, cmsUInt32Number Cols, 
+                                     const cmsFloat64Number* Matrix, const cmsFloat64Number* Offset)
+{
+    cmsUInt32Number i, n;
+    _cmsStageMatrixData* NewElem;
+    cmsStage* NewMPE;
+
+    n = Rows * Cols;
+
+    // Check for overflow
+    if (n == 0) return NULL;
+    if (n >= UINT_MAX / Cols) return NULL;
+    if (n >= UINT_MAX / Rows) return NULL;
+    if (n < Rows || n < Cols) return NULL;
+
+    NewMPE = _cmsStageAllocPlaceholder(ContextID, cmsSigMatrixElemType, Cols, Rows,
+                                     EvaluateMatrix, MatrixElemDup, MatrixElemTypeFree, NULL );
+    if (NewMPE == NULL) return NULL;
+
+   
+    NewElem = (_cmsStageMatrixData*) _cmsMallocZero(ContextID, sizeof(_cmsStageMatrixData));
+    if (NewElem == NULL) return NULL;
+  
+
+    NewElem ->Double = (cmsFloat64Number*) _cmsCalloc(ContextID, n, sizeof(cmsFloat64Number));
+
+    if (NewElem->Double == NULL) {
+        MatrixElemTypeFree(NewMPE);
+        return NULL;
+    }
+
+    for (i=0; i < n; i++) {
+        NewElem ->Double[i] = Matrix[i];
+    }
+
+
+    if (Offset != NULL) {
+    
+        NewElem ->Offset = (cmsFloat64Number*) _cmsCalloc(ContextID, Cols, sizeof(cmsFloat64Number));
+        if (NewElem->Offset == NULL) {
+           MatrixElemTypeFree(NewMPE);
+           return NULL;
+        }
+
+        for (i=0; i < Cols; i++) {
+                NewElem ->Offset[i] = Offset[i];
+        }
+
+    }
+    
+    NewMPE ->Data  = (void*) NewElem;
+    return NewMPE;
+}
+
+
+// *************************************************************************************************
+// Type cmsSigCLutElemType
+// *************************************************************************************************
+
+
+// Evaluate in true floating point
+static
+void EvaluateCLUTfloat(const cmsFloat32Number In[], cmsFloat32Number Out[], const cmsStage *mpe)
+{
+    _cmsStageCLutData* Data = (_cmsStageCLutData*) mpe ->Data;  
+
+    Data -> Params ->Interpolation.LerpFloat(In, Out, Data->Params);
+}
+
+
+// Convert to 16 bits, evaluate, and back to floating point
+static
+void EvaluateCLUTfloatIn16(const cmsFloat32Number In[], cmsFloat32Number Out[], const cmsStage *mpe)
+{
+    _cmsStageCLutData* Data = (_cmsStageCLutData*) mpe ->Data;
+    cmsUInt16Number In16[MAX_STAGE_CHANNELS], Out16[MAX_STAGE_CHANNELS];
+    
+    _cmsAssert(mpe ->InputChannels  <= MAX_STAGE_CHANNELS);
+    _cmsAssert(mpe ->OutputChannels <= MAX_STAGE_CHANNELS);
+
+    FromFloatTo16(In, In16, mpe ->InputChannels);   
+    Data -> Params ->Interpolation.Lerp16(In16, Out16, Data->Params);
+    From16ToFloat(Out16, Out,  mpe ->OutputChannels);
+}
+
+
+// Given an hypercube of b dimensions, with Dims[] number of nodes by dimension, calculate the total amount of nodes
+static
+cmsUInt32Number CubeSize(const cmsUInt32Number Dims[], cmsUInt32Number b)
+{
+    cmsUInt32Number rv, dim;
+
+    _cmsAssert(Dims != NULL);
+
+    for (rv = 1; b > 0; b--) {
+
+        dim = Dims[b-1];
+        if (dim == 0) return 0;  // Error
+
+        rv *= dim;
+
+        // Check for overflow
+        if (rv > UINT_MAX / dim) return 0;
+    }
+
+    return rv;
+}
+
+static
+void* CLUTElemDup(cmsStage* mpe)
+{
+    _cmsStageCLutData* Data = (_cmsStageCLutData*) mpe ->Data;
+    _cmsStageCLutData* NewElem;
+    
+    
+    NewElem = (_cmsStageCLutData*) _cmsMallocZero(mpe ->ContextID, sizeof(_cmsStageCLutData));
+    if (NewElem == NULL) return NULL;
+
+    NewElem ->nEntries       = Data ->nEntries;
+    NewElem ->HasFloatValues = Data ->HasFloatValues;
+
+    if (Data ->Tab.T) {
+
+        if (Data ->HasFloatValues)
+            NewElem ->Tab.TFloat = (cmsFloat32Number*) _cmsDupMem(mpe ->ContextID, Data ->Tab.TFloat, Data ->nEntries * sizeof (cmsFloat32Number));
+        else
+            NewElem ->Tab.T = (cmsUInt16Number*) _cmsDupMem(mpe ->ContextID, Data ->Tab.T, Data ->nEntries * sizeof (cmsUInt16Number));
+    }
+    
+    NewElem ->Params   = _cmsComputeInterpParamsEx(mpe ->ContextID,
+                                                   Data ->Params ->nSamples, 
+                                                   Data ->Params ->nInputs,
+                                                   Data ->Params ->nOutputs, 
+                                                   NewElem ->Tab.T,
+                                                   Data ->Params ->dwFlags);
+
+    return (void*) NewElem;
+}
+
+
+static
+void CLutElemTypeFree(cmsStage* mpe)
+{
+
+    _cmsStageCLutData* Data = (_cmsStageCLutData*) mpe ->Data;
+    
+    // Already empty
+    if (Data == NULL) return;
+
+    // This works for both types
+    if (Data -> Tab.T)
+        _cmsFree(mpe ->ContextID, Data -> Tab.T);
+
+    _cmsFreeInterpParams(Data ->Params);    
+    _cmsFree(mpe ->ContextID, mpe ->Data);
+}
+
+
+// Allocates a 16-bit multidimensional CLUT. This is evaluated at 16-bit precision. Table may have different
+// granularity on each dimension.
+cmsStage* CMSEXPORT cmsStageAllocCLut16bitGranular(cmsContext ContextID, 
+                                         const cmsUInt32Number clutPoints[], 
+                                         cmsUInt32Number inputChan, 
+                                         cmsUInt32Number outputChan, 
+                                         const cmsUInt16Number* Table)
+{
+    cmsUInt32Number i, n;
+    _cmsStageCLutData* NewElem;
+    cmsStage* NewMPE;
+    
+    NewMPE = _cmsStageAllocPlaceholder(ContextID, cmsSigCLutElemType, inputChan, outputChan,
+                                     EvaluateCLUTfloatIn16, CLUTElemDup, CLutElemTypeFree, NULL );
+
+    if (NewMPE == NULL) return NULL;
+
+    NewElem = (_cmsStageCLutData*) _cmsMalloc(ContextID, sizeof(_cmsStageCLutData));
+    if (NewElem == NULL) {
+        cmsStageFree(NewMPE);
+        return NULL;
+    }
+
+    NewMPE ->Data  = (void*) NewElem;
+
+    NewElem -> nEntries = n = outputChan * CubeSize(clutPoints, inputChan);
+    NewElem -> HasFloatValues = FALSE;
+
+    if (n == 0) {
+        cmsStageFree(NewMPE);
+        return NULL;
+    }
+
+
+    NewElem ->Tab.T  = (cmsUInt16Number*) _cmsCalloc(ContextID, n, sizeof(cmsUInt16Number));
+    if (NewElem ->Tab.T == NULL) {
+        cmsStageFree(NewMPE);
+        return NULL;
+    }
+
+    if (Table != NULL) {
+        for (i=0; i < n; i++) {
+            NewElem ->Tab.T[i] = Table[i];
+        }
+    }
+
+    NewElem ->Params = _cmsComputeInterpParamsEx(ContextID, clutPoints, inputChan, outputChan, NewElem ->Tab.T, CMS_LERP_FLAGS_16BITS);
+    if (NewElem ->Params == NULL) {
+        cmsStageFree(NewMPE);
+        return NULL;
+    }
+
+    return NewMPE;
+}
+
+cmsStage* CMSEXPORT cmsStageAllocCLut16bit(cmsContext ContextID, 
+                                    cmsUInt32Number nGridPoints, 
+                                    cmsUInt32Number inputChan, 
+                                    cmsUInt32Number outputChan, 
+                                    const cmsUInt16Number* Table)
+{
+    cmsUInt32Number Dimensions[MAX_INPUT_DIMENSIONS];
+    int i;
+
+   // Our resulting LUT would be same gridpoints on all dimensions
+    for (i=0; i < MAX_INPUT_DIMENSIONS; i++)
+        Dimensions[i] = nGridPoints;
+
+
+    return cmsStageAllocCLut16bitGranular(ContextID, Dimensions, inputChan, outputChan, Table);
+}
+
+
+cmsStage* CMSEXPORT cmsStageAllocCLutFloat(cmsContext ContextID, 
+                                       cmsUInt32Number nGridPoints, 
+                                       cmsUInt32Number inputChan, 
+                                       cmsUInt32Number outputChan, 
+                                       const cmsFloat32Number* Table)
+{
+   cmsUInt32Number Dimensions[MAX_INPUT_DIMENSIONS];
+   int i;
+
+    // Our resulting LUT would be same gridpoints on all dimensions
+    for (i=0; i < MAX_INPUT_DIMENSIONS; i++)
+        Dimensions[i] = nGridPoints;
+
+    return cmsStageAllocCLutFloatGranular(ContextID, Dimensions, inputChan, outputChan, Table);
+}
+
+
+
+cmsStage* CMSEXPORT cmsStageAllocCLutFloatGranular(cmsContext ContextID, const cmsUInt32Number clutPoints[], cmsUInt32Number inputChan, cmsUInt32Number outputChan, const cmsFloat32Number* Table)
+{
+    cmsUInt32Number i, n;
+    _cmsStageCLutData* NewElem;
+    cmsStage* NewMPE;
+    
+    _cmsAssert(clutPoints != NULL);
+
+    NewMPE = _cmsStageAllocPlaceholder(ContextID, cmsSigCLutElemType, inputChan, outputChan,
+                                             EvaluateCLUTfloat, CLUTElemDup, CLutElemTypeFree, NULL);
+    if (NewMPE == NULL) return NULL;
+
+  
+    NewElem = (_cmsStageCLutData*) _cmsMalloc(ContextID, sizeof(_cmsStageCLutData));
+    if (NewElem == NULL) {
+        cmsStageFree(NewMPE);
+        return NULL;
+    }
+
+    NewMPE ->Data  = (void*) NewElem;
+
+    // There is a potential integer overflow on conputing n and nEntries.
+    NewElem -> nEntries = n = outputChan * CubeSize( clutPoints, inputChan);
+    NewElem -> HasFloatValues = TRUE;
+
+    if (n == 0) {
+        cmsStageFree(NewMPE);
+        return NULL;
+    }
+
+    NewElem ->Tab.TFloat  = (cmsFloat32Number*) _cmsCalloc(ContextID, n, sizeof(cmsFloat32Number));
+    if (NewElem ->Tab.TFloat == NULL) {
+        cmsStageFree(NewMPE);
+        return NULL;
+    }
+
+    if (Table != NULL) {
+        for (i=0; i < n; i++) {
+            NewElem ->Tab.TFloat[i] = Table[i];
+        }
+    }
+
+
+    
+    NewElem ->Params = _cmsComputeInterpParamsEx(ContextID, clutPoints,  inputChan, outputChan, NewElem ->Tab.TFloat, CMS_LERP_FLAGS_FLOAT);
+    if (NewElem ->Params == NULL) {
+        cmsStageFree(NewMPE);
+        return NULL;
+    }
+
+
+
+    return NewMPE;
+}
+
+
+static
+int IdentitySampler(register const cmsUInt16Number In[], register cmsUInt16Number Out[], register void * Cargo)
+{
+    int nChan = *(int*) Cargo;
+    int i;
+
+    for (i=0; i < nChan; i++) 
+        Out[i] = In[i];
+
+    return 1;
+}
+
+// Creates an MPE that just copies input to output
+cmsStage* _cmsStageAllocIdentityCLut(cmsContext ContextID, int nChan)
+{
+    cmsUInt32Number Dimensions[MAX_INPUT_DIMENSIONS];
+    cmsStage* mpe ;
+    int i;
+        
+    for (i=0; i < MAX_INPUT_DIMENSIONS; i++)
+        Dimensions[i] = 2;
+
+    mpe = cmsStageAllocCLut16bitGranular(ContextID, Dimensions, nChan, nChan, NULL);
+    if (mpe == NULL) return NULL;
+    
+    if (!cmsStageSampleCLut16bit(mpe, IdentitySampler, &nChan, 0)) {
+        cmsStageFree(mpe);
+        return NULL;
+    }
+
+    mpe ->Implements = cmsSigIdentityElemType;
+    return mpe;
+}
+
+
+
+// Quantize a value 0 <= i < MaxSamples to 0..0xffff
+cmsUInt16Number _cmsQuantizeVal(cmsFloat64Number i, int MaxSamples)
+{
+    cmsFloat64Number x;
+
+    x = ((cmsFloat64Number) i * 65535.) / (cmsFloat64Number) (MaxSamples - 1);
+    return _cmsQuickSaturateWord(x);
+}
+
+
+// This routine does a sweep on whole input space, and calls its callback
+// function on knots. returns TRUE if all ok, FALSE otherwise.
+cmsBool CMSEXPORT cmsStageSampleCLut16bit(cmsStage* mpe, cmsSAMPLER16 Sampler, void * Cargo, cmsUInt32Number dwFlags)
+{
+    int i, t, nTotalPoints, index, rest;
+    int nInputs, nOutputs;  
+    cmsUInt32Number* nSamples;
+    cmsUInt16Number In[cmsMAXCHANNELS], Out[MAX_STAGE_CHANNELS];
+    _cmsStageCLutData* clut = (_cmsStageCLutData*) mpe->Data; 
+
+
+    nSamples = clut->Params ->nSamples;
+    nInputs  = clut->Params ->nInputs;
+    nOutputs = clut->Params ->nOutputs;
+
+    if (nInputs >= cmsMAXCHANNELS) return FALSE;
+    if (nOutputs >= MAX_STAGE_CHANNELS) return FALSE;
+
+    nTotalPoints = CubeSize(nSamples, nInputs);
+    if (nTotalPoints == 0) return FALSE;
+
+    index = 0;
+    for (i = 0; i < nTotalPoints; i++) {
+
+        rest = i;
+        for (t = nInputs-1; t >=0; --t) {
+
+            cmsUInt32Number  Colorant = rest % nSamples[t];
+
+            rest /= nSamples[t];
+
+            In[t] = _cmsQuantizeVal(Colorant, nSamples[t]);                         
+        }
+
+        if (clut ->Tab.T != NULL) {
+            for (t=0; t < nOutputs; t++)
+                Out[t] = clut->Tab.T[index + t];
+        }
+
+        if (!Sampler(In, Out, Cargo))
+            return FALSE;
+
+        if (!(dwFlags & SAMPLER_INSPECT)) {
+
+            if (clut ->Tab.T != NULL) {
+                for (t=0; t < nOutputs; t++)
+                    clut->Tab.T[index + t] = Out[t];
+            }
+        }
+
+        index += nOutputs;
+    }
+
+    return TRUE;
+}
+
+// Same as anterior, but for floting point
+cmsBool CMSEXPORT cmsStageSampleCLutFloat(cmsStage* mpe, cmsSAMPLERFLOAT Sampler, void * Cargo, cmsUInt32Number dwFlags)
+{
+    int i, t, nTotalPoints, index, rest;
+    int nInputs, nOutputs;
+    cmsUInt32Number* nSamples;
+    cmsFloat32Number In[cmsMAXCHANNELS], Out[MAX_STAGE_CHANNELS];   
+    _cmsStageCLutData* clut = (_cmsStageCLutData*) mpe->Data; 
+
+    nSamples = clut->Params ->nSamples;
+    nInputs  = clut->Params ->nInputs;
+    nOutputs = clut->Params ->nOutputs;
+
+    if (nInputs >= cmsMAXCHANNELS) return FALSE;
+    if (nOutputs >= MAX_STAGE_CHANNELS) return FALSE;
+
+    nTotalPoints = CubeSize(nSamples, nInputs);
+    if (nTotalPoints == 0) return FALSE;
+
+    index = 0;
+    for (i = 0; i < nTotalPoints; i++) {
+
+        rest = i;
+        for (t = nInputs-1; t >=0; --t) {
+         
+            cmsUInt32Number  Colorant = rest % nSamples[t];
+
+            rest /= nSamples[t];
+
+            In[t] =  (cmsFloat32Number) (_cmsQuantizeVal(Colorant, nSamples[t]) / 65535.0);                         
+        }
+
+        if (clut ->Tab.TFloat != NULL) {
+            for (t=0; t < nOutputs; t++)
+                Out[t] = clut->Tab.TFloat[index + t];
+        }
+
+        if (!Sampler(In, Out, Cargo))
+            return FALSE;
+
+        if (!(dwFlags & SAMPLER_INSPECT)) {
+
+            if (clut ->Tab.TFloat != NULL) {
+                for (t=0; t < nOutputs; t++)
+                    clut->Tab.TFloat[index + t] = Out[t];
+            }
+        }
+
+        index += nOutputs;
+    }
+
+    return TRUE;
+}
+
+
+
+// This routine does a sweep on whole input space, and calls its callback
+// function on knots. returns TRUE if all ok, FALSE otherwise.
+cmsBool CMSEXPORT cmsSliceSpace16(cmsUInt32Number nInputs, const cmsUInt32Number clutPoints[],
+                                         cmsSAMPLER16 Sampler, void * Cargo)
+{
+    int i, t, nTotalPoints, rest;
+    cmsUInt16Number In[cmsMAXCHANNELS];
+
+    if (nInputs >= cmsMAXCHANNELS) return FALSE;
+
+    nTotalPoints = CubeSize(clutPoints, nInputs);
+    if (nTotalPoints == 0) return FALSE;
+
+    for (i = 0; i < nTotalPoints; i++) {
+
+        rest = i;
+        for (t = nInputs-1; t >=0; --t) {
+
+            cmsUInt32Number  Colorant = rest % clutPoints[t];
+
+            rest /= clutPoints[t];
+            In[t] = _cmsQuantizeVal(Colorant, clutPoints[t]);   
+
+        }
+
+        if (!Sampler(In, NULL, Cargo))
+            return FALSE;
+    }
+
+    return TRUE;
+}
+
+cmsInt32Number CMSEXPORT cmsSliceSpaceFloat(cmsUInt32Number nInputs, const cmsUInt32Number clutPoints[],
+                                            cmsSAMPLERFLOAT Sampler, void * Cargo)
+{
+    int i, t, nTotalPoints, rest;
+    cmsFloat32Number In[cmsMAXCHANNELS];
+
+    if (nInputs >= cmsMAXCHANNELS) return FALSE;
+
+    nTotalPoints = CubeSize(clutPoints, nInputs);
+    if (nTotalPoints == 0) return FALSE;
+
+    for (i = 0; i < nTotalPoints; i++) {
+
+        rest = i;
+        for (t = nInputs-1; t >=0; --t) {
+
+            cmsUInt32Number  Colorant = rest % clutPoints[t];
+
+            rest /= clutPoints[t];
+            In[t] =  (cmsFloat32Number) (_cmsQuantizeVal(Colorant, clutPoints[t]) / 65535.0);    
+
+        }
+
+        if (!Sampler(In, NULL, Cargo))
+            return FALSE;
+    }
+
+    return TRUE;
+}
+
+// ********************************************************************************
+// Type cmsSigLab2XYZElemType
+// ********************************************************************************
+
+
+static
+void EvaluateLab2XYZ(const cmsFloat32Number In[], 
+                     cmsFloat32Number Out[], 
+                     const cmsStage *mpe)
+{
+    cmsCIELab Lab;
+    cmsCIEXYZ XYZ;
+    const cmsFloat64Number XYZadj = MAX_ENCODEABLE_XYZ;
+
+    // V4 rules
+    Lab.L = In[0] * 100.0;        
+    Lab.a = In[1] * 255.0 - 128.0;
+    Lab.b = In[2] * 255.0 - 128.0;
+
+    cmsLab2XYZ(NULL, &XYZ, &Lab);
+
+    // From XYZ, range 0..19997 to 0..1.0, note that 1.99997 comes from 0xffff 
+    // encoded as 1.15 fixed point, so 1 + (32767.0 / 32768.0)
+
+    Out[0] = (cmsFloat32Number) ((cmsFloat64Number) XYZ.X / XYZadj); 
+    Out[1] = (cmsFloat32Number) ((cmsFloat64Number) XYZ.Y / XYZadj); 
+    Out[2] = (cmsFloat32Number) ((cmsFloat64Number) XYZ.Z / XYZadj); 
+    return;
+
+    cmsUNUSED_PARAMETER(mpe);
+}
+
+
+// No dup or free routines needed, as the structure has no pointers in it.
+cmsStage* _cmsStageAllocLab2XYZ(cmsContext ContextID)
+{
+    return _cmsStageAllocPlaceholder(ContextID, cmsSigLab2XYZElemType, 3, 3, EvaluateLab2XYZ, NULL, NULL, NULL);
+}
+
+// ********************************************************************************
+
+// v2 L=100 is supposed to be placed on 0xFF00. There is no reasonable 
+// number of gridpoints that would make exact match. However, a prelinearization
+// of 258 entries, would map 0xFF00 exactly on entry 257, and this is good to avoid scum dot. 
+// Almost all what we need but unfortunately, the rest of entries should be scaled by
+// (255*257/256) and this is not exact.
+
+cmsStage* _cmsStageAllocLabV2ToV4curves(cmsContext ContextID)
+{
+    cmsStage* mpe;
+    cmsToneCurve* LabTable[3];
+    int i, j;
+
+    LabTable[0] = cmsBuildTabulatedToneCurve16(ContextID, 258, NULL);
+    LabTable[1] = cmsBuildTabulatedToneCurve16(ContextID, 258, NULL);
+    LabTable[2] = cmsBuildTabulatedToneCurve16(ContextID, 258, NULL);
+
+    for (j=0; j < 3; j++) {
+    
+        if (LabTable[j] == NULL) {
+            cmsFreeToneCurveTriple(LabTable);
+            return NULL;
+        }
+        
+        // We need to map * (0xffff / 0xff00), thats same as (257 / 256) 
+        // So we can use 258-entry tables to do the trick (i / 257) * (255 * 257) * (257 / 256);
+        for (i=0; i < 257; i++)  {
+
+            LabTable[j]->Table16[i] = (cmsUInt16Number) ((i * 0xffff + 0x80) >> 8);
+        }
+
+        LabTable[j] ->Table16[257] = 0xffff;
+    }
+
+    mpe = cmsStageAllocToneCurves(ContextID, 3, LabTable);
+    cmsFreeToneCurveTriple(LabTable);
+
+    mpe ->Implements = cmsSigLabV2toV4;
+    return mpe;
+}
+
+// ********************************************************************************
+
+// Matrix-based conversion, which is more accurate, but slower and cannot properly be saved in devicelink profiles
+cmsStage* _cmsStageAllocLabV2ToV4(cmsContext ContextID)
+{
+    static const cmsFloat64Number V2ToV4[] = { 65535.0/65280.0, 0, 0,
+                                     0, 65535.0/65280.0, 0, 
+                                     0, 0, 65535.0/65280.0 
+                                     };
+
+    cmsStage *mpe = cmsStageAllocMatrix(ContextID, 3, 3, V2ToV4, NULL);
+
+    if (mpe == NULL) return mpe;
+    mpe ->Implements = cmsSigLabV2toV4;
+    return mpe;
+}
+
+
+// Reverse direction
+cmsStage* _cmsStageAllocLabV4ToV2(cmsContext ContextID)
+{
+    static const cmsFloat64Number V4ToV2[] = { 65280.0/65535.0, 0, 0,
+                                     0, 65280.0/65535.0, 0, 
+                                     0, 0, 65280.0/65535.0 
+                                     };
+
+     cmsStage *mpe = cmsStageAllocMatrix(ContextID, 3, 3, V4ToV2, NULL);
+
+    if (mpe == NULL) return mpe;
+    mpe ->Implements = cmsSigLabV4toV2;
+    return mpe;
+}
+
+
+// ********************************************************************************
+// Type cmsSigXYZ2LabElemType
+// ********************************************************************************
+
+static
+void EvaluateXYZ2Lab(const cmsFloat32Number In[], cmsFloat32Number Out[], const cmsStage *mpe)
+{
+    cmsCIELab Lab;
+    cmsCIEXYZ XYZ;
+    const cmsFloat64Number XYZadj = MAX_ENCODEABLE_XYZ; 
+
+    // From 0..1.0 to XYZ
+
+    XYZ.X = In[0] * XYZadj; 
+    XYZ.Y = In[1] * XYZadj; 
+    XYZ.Z = In[2] * XYZadj;
+
+    cmsXYZ2Lab(NULL, &Lab, &XYZ);
+    
+    // From V4 Lab to 0..1.0
+
+    Out[0] = (cmsFloat32Number) (Lab.L / 100.0); 
+    Out[1] = (cmsFloat32Number) ((Lab.a + 128.0) / 255.0); 
+    Out[2] = (cmsFloat32Number) ((Lab.b + 128.0) / 255.0);
+    return;
+
+    cmsUNUSED_PARAMETER(mpe);
+}
+
+cmsStage* _cmsStageAllocXYZ2Lab(cmsContext ContextID)
+{    
+    return _cmsStageAllocPlaceholder(ContextID, cmsSigXYZ2LabElemType, 3, 3, EvaluateXYZ2Lab, NULL, NULL, NULL);
+ 
+}
+
+// ********************************************************************************
+
+// For v4, S-Shaped curves are placed in a/b axis to increase resolution near gray
+
+cmsStage* _cmsStageAllocLabPrelin(cmsContext ContextID)
+{
+    cmsToneCurve* LabTable[3];
+    cmsFloat64Number Params[1] =  {2.4} ;
+
+    LabTable[0] = cmsBuildGamma(ContextID, 1.0);
+    LabTable[1] = cmsBuildParametricToneCurve(ContextID, 108, Params);
+    LabTable[2] = cmsBuildParametricToneCurve(ContextID, 108, Params);
+
+    return cmsStageAllocToneCurves(ContextID, 3, LabTable);
+}
+
+
+// Free a single MPE 
+void CMSEXPORT cmsStageFree(cmsStage* mpe)
+{
+    if (mpe ->FreePtr) 
+        mpe ->FreePtr(mpe);
+
+    _cmsFree(mpe ->ContextID, mpe);
+}
+
+
+cmsUInt32Number  CMSEXPORT cmsStageInputChannels(const cmsStage* mpe)
+{
+    return mpe ->InputChannels;
+}
+
+cmsUInt32Number  CMSEXPORT cmsStageOutputChannels(const cmsStage* mpe)
+{
+    return mpe ->OutputChannels;
+}
+
+cmsStageSignature CMSEXPORT cmsStageType(const cmsStage* mpe)
+{
+    return mpe -> Type;
+}
+
+void* CMSEXPORT cmsStageData(const cmsStage* mpe)
+{
+    return mpe -> Data;
+}
+
+cmsStage*  CMSEXPORT cmsStageNext(const cmsStage* mpe)
+{
+    return mpe -> Next;
+}
+
+
+// Duplicates an MPE
+cmsStage* CMSEXPORT cmsStageDup(cmsStage* mpe)
+{
+    cmsStage* NewMPE;
+    
+    if (mpe == NULL) return NULL;
+    NewMPE = _cmsStageAllocPlaceholder(mpe ->ContextID, 
+                                     mpe ->Type, 
+                                     mpe ->InputChannels, 
+                                     mpe ->OutputChannels,
+                                     mpe ->EvalPtr,
+                                     mpe ->DupElemPtr,
+                                     mpe ->FreePtr,
+                                     NULL);
+    if (NewMPE == NULL) return NULL;
+  
+    NewMPE ->Implements     = mpe ->Implements;
+ 
+    if (mpe ->DupElemPtr) 
+        NewMPE ->Data       = mpe ->DupElemPtr(mpe);
+    else
+        NewMPE ->Data       = NULL;
+
+    return NewMPE;
+}
+
+
+// ***********************************************************************************************************
+
+// This function sets up the channel count
+
+static
+void BlessLUT(cmsPipeline* lut)
+{
+    // We can set the input/ouput channels only if we have elements. 
+    if (lut ->Elements != NULL) {
+
+        cmsStage *First, *Last;
+
+        First  = cmsPipelineGetPtrToFirstStage(lut);
+        Last   = cmsPipelineGetPtrToLastStage(lut);
+
+        if (First != NULL)lut ->InputChannels = First ->InputChannels;
+        if (Last != NULL) lut ->OutputChannels = Last ->OutputChannels;
+    }
+}
+
+
+// Default to evaluate the LUT on 16 bit-basis. Precision is retained. 
+static
+void _LUTeval16(register const cmsUInt16Number In[], register cmsUInt16Number Out[],  register const void* D)
+{
+    cmsPipeline* lut = (cmsPipeline*) D;
+    cmsStage *mpe;   
+    cmsFloat32Number Storage[2][MAX_STAGE_CHANNELS];
+    int Phase = 0, NextPhase;
+    
+    From16ToFloat(In, &Storage[Phase][0], lut ->InputChannels);
+
+    for (mpe = lut ->Elements; 
+         mpe != NULL; 
+         mpe = mpe ->Next) {
+
+             NextPhase = Phase ^ 1;            
+             mpe ->EvalPtr(&Storage[Phase][0], &Storage[NextPhase][0], mpe);
+             Phase = NextPhase;
+    }
+    
+
+    FromFloatTo16(&Storage[Phase][0], Out, lut ->OutputChannels);
+}
+
+
+
+// Does evaluate the LUT on cmsFloat32Number-basis. 
+static
+void _LUTevalFloat(register const cmsFloat32Number In[], register cmsFloat32Number Out[], const void* D)
+{
+    cmsPipeline* lut = (cmsPipeline*) D;
+    cmsStage *mpe;   
+    cmsFloat32Number Storage[2][MAX_STAGE_CHANNELS];
+    int Phase = 0, NextPhase;
+    
+    memmove(&Storage[Phase][0], In, lut ->InputChannels  * sizeof(cmsFloat32Number));
+
+    for (mpe = lut ->Elements; 
+         mpe != NULL; 
+         mpe = mpe ->Next) {
+
+              NextPhase = Phase ^ 1;
+              mpe ->EvalPtr(&Storage[Phase][0], &Storage[NextPhase][0], mpe);
+              Phase = NextPhase;          
+    }
+
+    memmove(Out, &Storage[Phase][0], lut ->OutputChannels * sizeof(cmsFloat32Number));
+}
+
+
+
+
+// LUT Creation & Destruction
+
+cmsPipeline* CMSEXPORT cmsPipelineAlloc(cmsContext ContextID, cmsUInt32Number InputChannels, cmsUInt32Number OutputChannels)
+{
+       cmsPipeline* NewLUT;
+
+       if (InputChannels >= cmsMAXCHANNELS ||
+           OutputChannels >= cmsMAXCHANNELS) return NULL;
+
+       NewLUT = (cmsPipeline*) _cmsMallocZero(ContextID, sizeof(cmsPipeline));
+       if (NewLUT == NULL) return NULL;
+
+
+       NewLUT -> InputChannels  = InputChannels;
+       NewLUT -> OutputChannels = OutputChannels;
+
+       NewLUT ->Eval16Fn    = _LUTeval16;
+       NewLUT ->EvalFloatFn = _LUTevalFloat;
+       NewLUT ->DupDataFn   = NULL;
+       NewLUT ->FreeDataFn  = NULL;
+       NewLUT ->Data        = NewLUT;     
+       NewLUT ->ContextID   = ContextID;
+
+       BlessLUT(NewLUT);
+
+       return NewLUT;
+}
+
+
+cmsUInt32Number CMSEXPORT cmsPipelineInputChannels(const cmsPipeline* lut)
+{
+    return lut ->InputChannels;
+}
+
+cmsUInt32Number CMSEXPORT cmsPipelineOutputChannels(const cmsPipeline* lut)
+{
+    return lut ->OutputChannels;
+}
+
+// Free a profile elements LUT
+void CMSEXPORT cmsPipelineFree(cmsPipeline* lut)
+{
+    cmsStage *mpe, *Next;   
+
+    if (lut == NULL) return;
+
+    for (mpe = lut ->Elements; 
+        mpe != NULL; 
+        mpe = Next) {
+
+            Next = mpe ->Next;
+            cmsStageFree(mpe);            
+    }
+
+    if (lut ->FreeDataFn) lut ->FreeDataFn(lut ->ContextID, lut ->Data);
+
+    _cmsFree(lut ->ContextID, lut);
+}
+
+
+// Default to evaluate the LUT on 16 bit-basis. 
+void CMSEXPORT cmsPipelineEval16(const cmsUInt16Number In[], cmsUInt16Number Out[],  const cmsPipeline* lut)
+{
+    lut ->Eval16Fn(In, Out, lut->Data);
+}
+
+
+// Does evaluate the LUT on cmsFloat32Number-basis. 
+void CMSEXPORT cmsPipelineEvalFloat(const cmsFloat32Number In[], cmsFloat32Number Out[], const cmsPipeline* lut)
+{
+    lut ->EvalFloatFn(In, Out, lut);
+}
+
+
+
+// Duplicates a LUT
+cmsPipeline* CMSEXPORT cmsPipelineDup(const cmsPipeline* lut)
+{
+    cmsPipeline* NewLUT;
+    cmsStage *NewMPE, *Anterior = NULL, *mpe;   
+    cmsBool  First = TRUE;
+
+    if (lut == NULL) return NULL;
+
+    NewLUT = cmsPipelineAlloc(lut ->ContextID, lut ->InputChannels, lut ->OutputChannels);  
+    for (mpe = lut ->Elements; 
+         mpe != NULL; 
+         mpe = mpe ->Next) {
+
+             NewMPE = cmsStageDup(mpe);
+
+             if (NewMPE == NULL) {
+                 cmsPipelineFree(NewLUT);
+                 return NULL;
+             }
+    
+             if (First) {
+                 NewLUT ->Elements = NewMPE;
+                 First = FALSE;
+             }
+             else {
+                Anterior ->Next = NewMPE;               
+             }
+
+            Anterior = NewMPE;
+    }
+
+    NewLUT ->DupDataFn  = lut ->DupDataFn;
+    NewLUT ->FreeDataFn = lut ->FreeDataFn;
+
+    if (NewLUT ->DupDataFn != NULL) 
+        NewLUT ->Data = NewLUT ->DupDataFn(lut ->ContextID, lut->Data);
+
+
+    NewLUT ->SaveAs8Bits    = lut ->SaveAs8Bits;
+
+    BlessLUT(NewLUT);
+    return NewLUT;
+}
+
+
+void CMSEXPORT cmsPipelineInsertStage(cmsPipeline* lut, cmsStageLoc loc, cmsStage* mpe)
+{
+    cmsStage* Anterior = NULL, *pt;
+
+    _cmsAssert(lut != NULL);
+    _cmsAssert(mpe != NULL);
+
+    switch (loc) {
+
+        case cmsAT_BEGIN:
+            mpe ->Next = lut ->Elements;
+            lut ->Elements = mpe;
+            break;
+
+        case cmsAT_END:
+
+            if (lut ->Elements == NULL) 
+                lut ->Elements = mpe;
+            else {
+            
+                for (pt = lut ->Elements;
+                     pt != NULL;
+                     pt = pt -> Next) Anterior = pt;
+
+                Anterior ->Next = mpe;
+                mpe ->Next = NULL;
+            }
+            break;
+        default:;
+    }
+
+    BlessLUT(lut);
+}
+
+// Unlink an element and return the pointer to it
+void CMSEXPORT cmsPipelineUnlinkStage(cmsPipeline* lut, cmsStageLoc loc, cmsStage** mpe)
+{
+    cmsStage *Anterior, *pt, *Last;
+    cmsStage *Unlinked = NULL;
+
+    
+    // If empty LUT, there is nothing to remove
+    if (lut ->Elements == NULL) {
+        if (mpe) *mpe = NULL;
+        return;
+    }
+
+    // On depending on the strategy...
+    switch (loc) {
+
+        case cmsAT_BEGIN: 
+            {
+                cmsStage* elem = lut ->Elements;
+             
+                lut ->Elements = elem -> Next;
+                elem ->Next = NULL;
+                Unlinked = elem;
+                           
+            }
+            break;
+
+        case cmsAT_END:
+            Anterior = Last = NULL;
+            for (pt = lut ->Elements;
+                pt != NULL;
+                pt = pt -> Next) {
+                    Anterior = Last;
+                    Last = pt;              
+            }
+
+            Unlinked = Last;  // Next already points to NULL
+
+            // Truncate the chain
+            if (Anterior) 
+                Anterior ->Next = NULL;
+            else 
+                lut ->Elements = NULL;
+            break;
+        default:;
+    }
+
+    if (mpe) 
+        *mpe = Unlinked;
+    else
+        cmsStageFree(Unlinked);
+
+    BlessLUT(lut);
+}
+
+
+// Concatenate two LUT into a new single one
+cmsBool  CMSEXPORT cmsPipelineCat(cmsPipeline* l1, const cmsPipeline* l2)
+{
+    cmsStage* mpe, *NewMPE;
+
+    // If both LUTS does not have elements, we need to inherit 
+    // the number of channels
+    if (l1 ->Elements == NULL && l2 ->Elements == NULL) {
+        l1 ->InputChannels  = l2 ->InputChannels;
+        l1 ->OutputChannels = l2 ->OutputChannels;
+    }
+
+    // Cat second
+    for (mpe = l2 ->Elements; 
+         mpe != NULL; 
+         mpe = mpe ->Next) {
+
+            // We have to dup each element
+             NewMPE = cmsStageDup(mpe);
+
+             if (NewMPE == NULL) {
+                 return FALSE;
+             }
+    
+             cmsPipelineInsertStage(l1, cmsAT_END, NewMPE);
+    }
+
+  BlessLUT(l1);
+  return TRUE;
+}
+
+
+cmsBool CMSEXPORT cmsPipelineSetSaveAs8bitsFlag(cmsPipeline* lut, cmsBool On)
+{
+    cmsBool Anterior = lut ->SaveAs8Bits;
+
+    lut ->SaveAs8Bits = On;
+    return Anterior;
+}
+
+
+cmsStage* CMSEXPORT cmsPipelineGetPtrToFirstStage(const cmsPipeline* lut)
+{
+    return lut ->Elements;
+}
+
+cmsStage* CMSEXPORT cmsPipelineGetPtrToLastStage(const cmsPipeline* lut)
+{
+    cmsStage *mpe, *Anterior = NULL;
+
+    for (mpe = lut ->Elements; mpe != NULL; mpe = mpe ->Next)
+        Anterior = mpe;
+
+    return Anterior;
+}
+
+cmsUInt32Number CMSEXPORT cmsPipelineStageCount(const cmsPipeline* lut)
+{
+    cmsStage *mpe;
+    cmsUInt32Number n;
+
+    for (n=0, mpe = lut ->Elements; mpe != NULL; mpe = mpe ->Next)
+            n++;
+
+    return n;
+}
+
+// This function may be used to set the optional evalueator and a block of private data. If private data is being used, an optional
+// duplicator and free functions should also be specified in order to duplicate the LUT construct. Use NULL to inhibit such functionality.
+void CMSEXPORT _cmsPipelineSetOptimizationParameters(cmsPipeline* Lut, 
+                                        _cmsOPTeval16Fn Eval16, 
+                                        void* PrivateData, 
+                                        _cmsOPTfreeDataFn FreePrivateDataFn, 
+                                        _cmsOPTdupDataFn DupPrivateDataFn)
+{
+
+    Lut ->Eval16Fn = Eval16;
+    Lut ->DupDataFn = DupPrivateDataFn;
+    Lut ->FreeDataFn = FreePrivateDataFn;   
+    Lut ->Data = PrivateData;
+}
+
+
+// ----------------------------------------------------------- Reverse interpolation
+// Here's how it goes. The derivative Df(x) of the function f is the linear 
+// transformation that best approximates f near the point x. It can be represented 
+// by a matrix A whose entries are the partial derivatives of the components of f 
+// with respect to all the coordinates. This is know as the Jacobian
+//
+// The best linear approximation to f is given by the matrix equation: 
+// 
+// y-y0 = A (x-x0) 
+// 
+// So, if x0 is a good "guess" for the zero of f, then solving for the zero of this 
+// linear approximation will give a "better guess" for the zero of f. Thus let y=0, 
+// and since y0=f(x0) one can solve the above equation for x. This leads to the 
+// Newton's method formula: 
+//
+// xn+1 = xn - A-1 f(xn) 
+// 
+// where xn+1 denotes the (n+1)-st guess, obtained from the n-th guess xn in the 
+// fashion described above. Iterating this will give better and better approximations 
+// if you have a "good enough" initial guess. 
+
+
+#define JACOBIAN_EPSILON            0.001f
+#define INVERSION_MAX_ITERATIONS    30
+
+// Increment with reflexion on boundary
+static 
+void IncDelta(cmsFloat32Number *Val)
+{
+    if (*Val < (1.0 - JACOBIAN_EPSILON)) 
+
+        *Val += JACOBIAN_EPSILON;
+    
+    else 
+        *Val -= JACOBIAN_EPSILON;
+    
+}
+
+
+
+// Euclidean distance between two vectors of n elements each one
+static
+cmsFloat32Number EuclideanDistance(cmsFloat32Number a[], cmsFloat32Number b[], int n)
+{
+    cmsFloat32Number sum = 0;
+    int i;
+
+    for (i=0; i < n; i++) {
+        cmsFloat32Number dif = b[i] - a[i];
+        sum +=  dif * dif;
+    }
+
+    return sqrtf(sum);
+}
+
+
+// Evaluate a LUT in reverse direction. It only searches on 3->3 LUT. Uses Newton method
+//
+// x1 <- x - [J(x)]^-1 * f(x)
+// 
+// lut: The LUT on where to do the search
+// Target: LabK, 3 values of Lab plus destination K which is fixed
+// Result: The obtained CMYK
+// Hint:   Location where begin the search
+
+cmsBool CMSEXPORT cmsPipelineEvalReverseFloat(cmsFloat32Number Target[],
+                                              cmsFloat32Number Result[],
+                                              cmsFloat32Number Hint[],
+                                              const cmsPipeline* lut)
+{
+    cmsUInt32Number  i, j;
+    cmsFloat64Number  error, LastError = 1E20;    
+    cmsFloat32Number  fx[4], x[4], xd[4], fxd[4];
+    cmsVEC3 tmp, tmp2;
+    cmsMAT3 Jacobian;    
+    cmsFloat64Number LastResult[4];
+
+
+    // Only 3->3 and 4->3 are supported
+    if (lut ->InputChannels != 3 && lut ->InputChannels != 4) return FALSE;
+    if (lut ->OutputChannels != 3) return FALSE;
+
+    // Mark result of -1
+    LastResult[0] = LastResult[1] = LastResult[2] = -1.0f;
+
+    // Take the hint as starting point if specified
+    if (Hint == NULL) {
+
+        // Begin at any point, we choose 1/3 of CMY axis
+        x[0] = x[1] = x[2] = 0.3f;
+    }
+    else {
+
+        // Only copy 3 channels from hint...
+        for (j=0; j < 3; j++) 
+            x[j] = Hint[j];         
+    }
+    
+    // If Lut is 4-dimensions, then grab target[3], which is fixed
+    if (lut ->InputChannels == 4) {
+        x[3] = Target[3];
+    }
+    else x[3] = 0; // To keep lint happy
+
+
+    // Iterate    
+    for (i = 0; i < INVERSION_MAX_ITERATIONS; i++) {
+
+        // Get beginning fx
+        cmsPipelineEvalFloat(x, fx, lut);
+
+        // Compute error
+        error = EuclideanDistance(fx, Target, 3);
+
+        // If not convergent, return last safe value
+        if (error >= LastError) 
+            break;
+
+        // Keep latest values
+        LastError     = error;
+        for (j=0; j < lut ->InputChannels; j++) 
+                Result[j] = x[j]; 
+
+        // Found an exact match?
+        if (error <= 0) 
+            break;
+
+        // Obtain slope (the Jacobian)              
+        for (j = 0; j < 3; j++) {
+
+            xd[0] = x[0];
+            xd[1] = x[1];
+            xd[2] = x[2];
+            xd[3] = x[3];  // Keep fixed channel
+
+            IncDelta(&xd[j]);
+
+            cmsPipelineEvalFloat(xd, fxd, lut);
+
+            Jacobian.v[0].n[j] = ((fxd[0] - fx[0]) / JACOBIAN_EPSILON);
+            Jacobian.v[1].n[j] = ((fxd[1] - fx[1]) / JACOBIAN_EPSILON);
+            Jacobian.v[2].n[j] = ((fxd[2] - fx[2]) / JACOBIAN_EPSILON);        
+        }
+
+        // Solve system
+        tmp2.n[0] = fx[0] - Target[0];
+        tmp2.n[1] = fx[1] - Target[1];
+        tmp2.n[2] = fx[2] - Target[2];
+
+        if (!_cmsMAT3solve(&tmp, &Jacobian, &tmp2))
+            return FALSE;
+
+        // Move our guess
+        x[0] -= (cmsFloat32Number) tmp.n[0];
+        x[1] -= (cmsFloat32Number) tmp.n[1];
+        x[2] -= (cmsFloat32Number) tmp.n[2];
+
+        // Some clipping....
+        for (j=0; j < 3; j++) {
+            if (x[j] < 0) x[j] = 0;
+            else
+                if (x[j] > 1.0) x[j] = 1.0;
+        }
+    }
+
+    return TRUE;
+}
+
diff --git a/thirdparty/liblcms2/src/cmsmd5.c b/thirdparty/liblcms2/src/cmsmd5.c
new file mode 100644
index 00000000..ecf3d907
--- /dev/null
+++ b/thirdparty/liblcms2/src/cmsmd5.c
@@ -0,0 +1,317 @@
+//---------------------------------------------------------------------------------
+//
+//  Little Color Management System
+//  Copyright (c) 1998-2010 Marti Maria Saguer
+//
+// Permission is hereby granted, free of charge, to any person obtaining 
+// a copy of this software and associated documentation files (the "Software"), 
+// to deal in the Software without restriction, including without limitation 
+// the rights to use, copy, modify, merge, publish, distribute, sublicense, 
+// and/or sell copies of the Software, and to permit persons to whom the Software 
+// is furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in 
+// all copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 
+// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO 
+// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE 
+// LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 
+// OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 
+// WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+//
+//---------------------------------------------------------------------------------
+
+
+#include "lcms2_internal.h"
+
+#ifdef CMS_USE_BIG_ENDIAN
+
+static
+void byteReverse(cmsUInt8Number * buf, cmsUInt32Number longs)
+{
+    do {
+
+        cmsUInt32Number t = _cmsAdjustEndianess32(*(cmsUInt32Number *) buf);
+        *(cmsUInt32Number *) buf = t;
+        buf += sizeof(cmsUInt32Number);
+
+    } while (--longs);
+
+}
+
+#else
+#define byteReverse(buf, len)   
+#endif
+
+
+typedef struct {
+
+    cmsUInt32Number buf[4];
+    cmsUInt32Number bits[2];
+    cmsUInt8Number in[64];
+    cmsContext ContextID;
+
+} _cmsMD5;
+
+#define F1(x, y, z) (z ^ (x & (y ^ z)))
+#define F2(x, y, z) F1(z, x, y)
+#define F3(x, y, z) (x ^ y ^ z)
+#define F4(x, y, z) (y ^ (x | ~z))
+
+#define STEP(f, w, x, y, z, data, s) \
+    ( w += f(x, y, z) + data,  w = w<<s | w>>(32-s),  w += x )
+
+
+static
+void MD5_Transform(cmsUInt32Number buf[4], cmsUInt32Number in[16])
+    
+{
+    register cmsUInt32Number a, b, c, d;
+
+    a = buf[0];
+    b = buf[1];
+    c = buf[2];
+    d = buf[3];
+
+    STEP(F1, a, b, c, d, in[0] + 0xd76aa478, 7);
+    STEP(F1, d, a, b, c, in[1] + 0xe8c7b756, 12);
+    STEP(F1, c, d, a, b, in[2] + 0x242070db, 17);
+    STEP(F1, b, c, d, a, in[3] + 0xc1bdceee, 22);
+    STEP(F1, a, b, c, d, in[4] + 0xf57c0faf, 7);
+    STEP(F1, d, a, b, c, in[5] + 0x4787c62a, 12);
+    STEP(F1, c, d, a, b, in[6] + 0xa8304613, 17);
+    STEP(F1, b, c, d, a, in[7] + 0xfd469501, 22);
+    STEP(F1, a, b, c, d, in[8] + 0x698098d8, 7);
+    STEP(F1, d, a, b, c, in[9] + 0x8b44f7af, 12);
+    STEP(F1, c, d, a, b, in[10] + 0xffff5bb1, 17);
+    STEP(F1, b, c, d, a, in[11] + 0x895cd7be, 22);
+    STEP(F1, a, b, c, d, in[12] + 0x6b901122, 7);
+    STEP(F1, d, a, b, c, in[13] + 0xfd987193, 12);
+    STEP(F1, c, d, a, b, in[14] + 0xa679438e, 17);
+    STEP(F1, b, c, d, a, in[15] + 0x49b40821, 22);
+
+    STEP(F2, a, b, c, d, in[1] + 0xf61e2562, 5);
+    STEP(F2, d, a, b, c, in[6] + 0xc040b340, 9);
+    STEP(F2, c, d, a, b, in[11] + 0x265e5a51, 14);
+    STEP(F2, b, c, d, a, in[0] + 0xe9b6c7aa, 20);
+    STEP(F2, a, b, c, d, in[5] + 0xd62f105d, 5);
+    STEP(F2, d, a, b, c, in[10] + 0x02441453, 9);
+    STEP(F2, c, d, a, b, in[15] + 0xd8a1e681, 14);
+    STEP(F2, b, c, d, a, in[4] + 0xe7d3fbc8, 20);
+    STEP(F2, a, b, c, d, in[9] + 0x21e1cde6, 5);
+    STEP(F2, d, a, b, c, in[14] + 0xc33707d6, 9);
+    STEP(F2, c, d, a, b, in[3] + 0xf4d50d87, 14);
+    STEP(F2, b, c, d, a, in[8] + 0x455a14ed, 20);
+    STEP(F2, a, b, c, d, in[13] + 0xa9e3e905, 5);
+    STEP(F2, d, a, b, c, in[2] + 0xfcefa3f8, 9);
+    STEP(F2, c, d, a, b, in[7] + 0x676f02d9, 14);
+    STEP(F2, b, c, d, a, in[12] + 0x8d2a4c8a, 20);
+
+    STEP(F3, a, b, c, d, in[5] + 0xfffa3942, 4);
+    STEP(F3, d, a, b, c, in[8] + 0x8771f681, 11);
+    STEP(F3, c, d, a, b, in[11] + 0x6d9d6122, 16);
+    STEP(F3, b, c, d, a, in[14] + 0xfde5380c, 23);
+    STEP(F3, a, b, c, d, in[1] + 0xa4beea44, 4);
+    STEP(F3, d, a, b, c, in[4] + 0x4bdecfa9, 11);
+    STEP(F3, c, d, a, b, in[7] + 0xf6bb4b60, 16);
+    STEP(F3, b, c, d, a, in[10] + 0xbebfbc70, 23);
+    STEP(F3, a, b, c, d, in[13] + 0x289b7ec6, 4);
+    STEP(F3, d, a, b, c, in[0] + 0xeaa127fa, 11);
+    STEP(F3, c, d, a, b, in[3] + 0xd4ef3085, 16);
+    STEP(F3, b, c, d, a, in[6] + 0x04881d05, 23);
+    STEP(F3, a, b, c, d, in[9] + 0xd9d4d039, 4);
+    STEP(F3, d, a, b, c, in[12] + 0xe6db99e5, 11);
+    STEP(F3, c, d, a, b, in[15] + 0x1fa27cf8, 16);
+    STEP(F3, b, c, d, a, in[2] + 0xc4ac5665, 23);
+
+    STEP(F4, a, b, c, d, in[0] + 0xf4292244, 6);
+    STEP(F4, d, a, b, c, in[7] + 0x432aff97, 10);
+    STEP(F4, c, d, a, b, in[14] + 0xab9423a7, 15);
+    STEP(F4, b, c, d, a, in[5] + 0xfc93a039, 21);
+    STEP(F4, a, b, c, d, in[12] + 0x655b59c3, 6);
+    STEP(F4, d, a, b, c, in[3] + 0x8f0ccc92, 10);
+    STEP(F4, c, d, a, b, in[10] + 0xffeff47d, 15);
+    STEP(F4, b, c, d, a, in[1] + 0x85845dd1, 21);
+    STEP(F4, a, b, c, d, in[8] + 0x6fa87e4f, 6);
+    STEP(F4, d, a, b, c, in[15] + 0xfe2ce6e0, 10);
+    STEP(F4, c, d, a, b, in[6] + 0xa3014314, 15);
+    STEP(F4, b, c, d, a, in[13] + 0x4e0811a1, 21);
+    STEP(F4, a, b, c, d, in[4] + 0xf7537e82, 6);
+    STEP(F4, d, a, b, c, in[11] + 0xbd3af235, 10);
+    STEP(F4, c, d, a, b, in[2] + 0x2ad7d2bb, 15);
+    STEP(F4, b, c, d, a, in[9] + 0xeb86d391, 21);
+
+    buf[0] += a;
+    buf[1] += b;
+    buf[2] += c;
+    buf[3] += d;
+}
+
+
+// Create a MD5 object
+static
+cmsHANDLE  MD5alloc(cmsContext ContextID)
+{
+    _cmsMD5* ctx = (_cmsMD5*) _cmsMallocZero(ContextID, sizeof(_cmsMD5));
+    if (ctx == NULL) return NULL;
+
+    ctx ->ContextID = ContextID;
+
+    ctx->buf[0] = 0x67452301;
+    ctx->buf[1] = 0xefcdab89;
+    ctx->buf[2] = 0x98badcfe;
+    ctx->buf[3] = 0x10325476;
+
+    ctx->bits[0] = 0;
+    ctx->bits[1] = 0;
+
+    return (cmsHANDLE) ctx;
+}
+
+
+static
+void MD5add(cmsHANDLE Handle, cmsUInt8Number* buf, cmsUInt32Number len)
+{
+    _cmsMD5* ctx = (_cmsMD5*) Handle;
+    cmsUInt32Number t;
+    
+    t = ctx->bits[0];
+    if ((ctx->bits[0] = t + (len << 3)) < t)
+        ctx->bits[1]++; 
+
+    ctx->bits[1] += len >> 29;
+
+    t = (t >> 3) & 0x3f;    
+
+    if (t) {
+
+        cmsUInt8Number *p = (cmsUInt8Number *) ctx->in + t;
+
+        t = 64 - t;
+        if (len < t) {
+            memmove(p, buf, len);
+            return;
+        }
+
+        memmove(p, buf, t);
+        byteReverse(ctx->in, 16);
+
+        MD5_Transform(ctx->buf, (cmsUInt32Number *) ctx->in);
+        buf += t;
+        len -= t;
+    }
+
+    while (len >= 64) {
+        memmove(ctx->in, buf, 64);
+        byteReverse(ctx->in, 16);
+        MD5_Transform(ctx->buf, (cmsUInt32Number *) ctx->in);
+        buf += 64;
+        len -= 64;
+    }
+
+    memmove(ctx->in, buf, len);
+}
+
+// Destroy the object and return the checksum
+static
+void MD5finish(cmsProfileID* ProfileID,  cmsHANDLE Handle)
+{
+    _cmsMD5* ctx = (_cmsMD5*) Handle;
+    cmsUInt32Number count;
+    cmsUInt8Number *p;
+
+    count = (ctx->bits[0] >> 3) & 0x3F;
+
+    p = ctx->in + count;
+    *p++ = 0x80;
+
+    count = 64 - 1 - count;
+
+    if (count < 8) {
+
+        memset(p, 0, count);
+        byteReverse(ctx->in, 16);
+        MD5_Transform(ctx->buf, (cmsUInt32Number *) ctx->in);
+
+        memset(ctx->in, 0, 56);
+    } else {
+        memset(p, 0, count - 8);
+    }
+    byteReverse(ctx->in, 14);
+
+    ((cmsUInt32Number *) ctx->in)[14] = ctx->bits[0];
+    ((cmsUInt32Number *) ctx->in)[15] = ctx->bits[1];
+
+    MD5_Transform(ctx->buf, (cmsUInt32Number *) ctx->in);
+
+    byteReverse((cmsUInt8Number *) ctx->buf, 4);
+    memmove(ProfileID ->ID8, ctx->buf, 16);
+
+    _cmsFree(ctx ->ContextID, ctx);
+}
+
+
+
+// Assuming io points to an ICC profile, compute and store MD5 checksum
+// In the header, rendering intentent, attributes and ID should be set to zero
+// before computing MD5 checksum (per 6.1.13 in ICC spec)
+
+cmsBool CMSEXPORT cmsMD5computeID(cmsHPROFILE hProfile)
+{
+    cmsContext   ContextID;
+    cmsUInt32Number BytesNeeded;
+    cmsUInt8Number* Mem = NULL;
+    cmsHANDLE  MD5 = NULL;
+    _cmsICCPROFILE* Icc = (_cmsICCPROFILE*) hProfile;
+    _cmsICCPROFILE Keep;              
+
+	_cmsAssert(hProfile != NULL);
+
+    ContextID = cmsGetProfileContextID(hProfile);
+
+    // Save a copy of the profile header
+    memmove(&Keep, Icc, sizeof(_cmsICCPROFILE));
+    
+    // Set RI, attributes and ID
+    memset(&Icc ->attributes, 0, sizeof(Icc ->attributes));
+    Icc ->RenderingIntent = 0;
+    memset(&Icc ->ProfileID, 0, sizeof(Icc ->ProfileID));
+
+    // Compute needed storage
+    if (!cmsSaveProfileToMem(hProfile, NULL, &BytesNeeded)) goto Error;
+
+    // Allocate memory
+    Mem = (cmsUInt8Number*) _cmsMalloc(ContextID, BytesNeeded);
+    if (Mem == NULL) goto Error;
+
+    // Save to temporary storage
+    if (!cmsSaveProfileToMem(hProfile, Mem, &BytesNeeded)) goto Error;
+     
+    // Create MD5 object
+    MD5 = MD5alloc(ContextID);
+    if (MD5 == NULL) goto Error;
+
+    // Add all bytes
+    MD5add(MD5, Mem, BytesNeeded);
+
+    // Temp storage is no longer needed
+    _cmsFree(ContextID, Mem);
+    
+    // Restore header
+    memmove(Icc, &Keep, sizeof(_cmsICCPROFILE));
+
+    // And store the ID
+    MD5finish(&Icc ->ProfileID,  MD5);
+    return TRUE;
+
+Error:
+
+    // Free resources as something went wrong
+    if (MD5 != NULL) _cmsFree(ContextID, MD5);
+    if (Mem != NULL) _cmsFree(ContextID, Mem);
+    memmove(Icc, &Keep, sizeof(_cmsICCPROFILE));
+    return FALSE;
+}
+
diff --git a/thirdparty/liblcms2/src/cmsmtrx.c b/thirdparty/liblcms2/src/cmsmtrx.c
new file mode 100644
index 00000000..84035c93
--- /dev/null
+++ b/thirdparty/liblcms2/src/cmsmtrx.c
@@ -0,0 +1,176 @@
+//---------------------------------------------------------------------------------
+//
+//  Little Color Management System
+//  Copyright (c) 1998-2010 Marti Maria Saguer
+//
+// Permission is hereby granted, free of charge, to any person obtaining 
+// a copy of this software and associated documentation files (the "Software"), 
+// to deal in the Software without restriction, including without limitation 
+// the rights to use, copy, modify, merge, publish, distribute, sublicense, 
+// and/or sell copies of the Software, and to permit persons to whom the Software 
+// is furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in 
+// all copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 
+// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO 
+// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE 
+// LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 
+// OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 
+// WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+//
+//---------------------------------------------------------------------------------
+//
+
+#include "lcms2_internal.h"
+
+
+#define DSWAP(x, y)     {cmsFloat64Number tmp = (x); (x)=(y); (y)=tmp;}
+
+
+// Initiate a vector 
+void CMSEXPORT _cmsVEC3init(cmsVEC3* r, cmsFloat64Number x, cmsFloat64Number y, cmsFloat64Number z)
+{
+    r -> n[VX] = x;
+    r -> n[VY] = y;
+    r -> n[VZ] = z;
+}
+
+// Vector substraction
+void CMSEXPORT _cmsVEC3minus(cmsVEC3* r, const cmsVEC3* a, const cmsVEC3* b)
+{
+  r -> n[VX] = a -> n[VX] - b -> n[VX];
+  r -> n[VY] = a -> n[VY] - b -> n[VY];
+  r -> n[VZ] = a -> n[VZ] - b -> n[VZ];
+}
+
+// Vector cross product
+void CMSEXPORT _cmsVEC3cross(cmsVEC3* r, const cmsVEC3* u, const cmsVEC3* v)
+{
+    r ->n[VX] = u->n[VY] * v->n[VZ] - v->n[VY] * u->n[VZ];
+    r ->n[VY] = u->n[VZ] * v->n[VX] - v->n[VZ] * u->n[VX];
+    r ->n[VZ] = u->n[VX] * v->n[VY] - v->n[VX] * u->n[VY];
+}
+
+// Vector dot product
+cmsFloat64Number CMSEXPORT _cmsVEC3dot(const cmsVEC3* u, const cmsVEC3* v)
+{
+    return u->n[VX] * v->n[VX] + u->n[VY] * v->n[VY] + u->n[VZ] * v->n[VZ];
+}
+
+// Euclidean length 
+cmsFloat64Number CMSEXPORT _cmsVEC3length(const cmsVEC3* a)
+{
+    return sqrt(a ->n[VX] * a ->n[VX] +
+                a ->n[VY] * a ->n[VY] +
+                a ->n[VZ] * a ->n[VZ]);
+}
+
+// Euclidean distance
+cmsFloat64Number CMSEXPORT _cmsVEC3distance(const cmsVEC3* a, const cmsVEC3* b)
+{
+    cmsFloat64Number d1 = a ->n[VX] - b ->n[VX];
+    cmsFloat64Number d2 = a ->n[VY] - b ->n[VY];
+    cmsFloat64Number d3 = a ->n[VZ] - b ->n[VZ];
+
+    return sqrt(d1*d1 + d2*d2 + d3*d3);
+}
+
+
+
+// 3x3 Identity
+void CMSEXPORT _cmsMAT3identity(cmsMAT3* a)
+{
+    _cmsVEC3init(&a-> v[0], 1.0, 0.0, 0.0);
+    _cmsVEC3init(&a-> v[1], 0.0, 1.0, 0.0);
+    _cmsVEC3init(&a-> v[2], 0.0, 0.0, 1.0);
+}
+
+static
+cmsBool CloseEnough(cmsFloat64Number a, cmsFloat64Number b)
+{
+    return fabs(b - a) < (1.0 / 65535.0);
+}
+
+
+cmsBool CMSEXPORT _cmsMAT3isIdentity(const cmsMAT3* a)
+{
+	cmsMAT3 Identity;
+	int i, j;
+
+	_cmsMAT3identity(&Identity);
+
+	for (i=0; i < 3; i++)
+		for (j=0; j < 3; j++)
+			if (!CloseEnough(a ->v[i].n[j], Identity.v[i].n[j])) return FALSE;
+
+	return TRUE;
+}
+
+
+// Multiply two matrices
+void CMSEXPORT _cmsMAT3per(cmsMAT3* r, const cmsMAT3* a, const cmsMAT3* b)
+{
+#define ROWCOL(i, j) \
+    a->v[i].n[0]*b->v[0].n[j] + a->v[i].n[1]*b->v[1].n[j] + a->v[i].n[2]*b->v[2].n[j]
+
+    _cmsVEC3init(&r-> v[0], ROWCOL(0,0), ROWCOL(0,1), ROWCOL(0,2));
+    _cmsVEC3init(&r-> v[1], ROWCOL(1,0), ROWCOL(1,1), ROWCOL(1,2));
+    _cmsVEC3init(&r-> v[2], ROWCOL(2,0), ROWCOL(2,1), ROWCOL(2,2));
+
+#undef ROWCOL //(i, j)
+}
+
+
+
+// Inverse of a matrix b = a^(-1)
+cmsBool  CMSEXPORT _cmsMAT3inverse(const cmsMAT3* a, cmsMAT3* b)
+{
+   cmsFloat64Number det, c0, c1, c2;
+
+   c0 =  a -> v[1].n[1]*a -> v[2].n[2] - a -> v[1].n[2]*a -> v[2].n[1];
+   c1 = -a -> v[1].n[0]*a -> v[2].n[2] + a -> v[1].n[2]*a -> v[2].n[0];
+   c2 =  a -> v[1].n[0]*a -> v[2].n[1] - a -> v[1].n[1]*a -> v[2].n[0];
+
+   det = a -> v[0].n[0]*c0 + a -> v[0].n[1]*c1 + a -> v[0].n[2]*c2;
+
+   if (fabs(det) < MATRIX_DET_TOLERANCE) return FALSE;  // singular matrix; can't invert
+
+   b -> v[0].n[0] = c0/det;
+   b -> v[0].n[1] = (a -> v[0].n[2]*a -> v[2].n[1] - a -> v[0].n[1]*a -> v[2].n[2])/det;
+   b -> v[0].n[2] = (a -> v[0].n[1]*a -> v[1].n[2] - a -> v[0].n[2]*a -> v[1].n[1])/det;
+   b -> v[1].n[0] = c1/det;
+   b -> v[1].n[1] = (a -> v[0].n[0]*a -> v[2].n[2] - a -> v[0].n[2]*a -> v[2].n[0])/det;
+   b -> v[1].n[2] = (a -> v[0].n[2]*a -> v[1].n[0] - a -> v[0].n[0]*a -> v[1].n[2])/det;
+   b -> v[2].n[0] = c2/det;
+   b -> v[2].n[1] = (a -> v[0].n[1]*a -> v[2].n[0] - a -> v[0].n[0]*a -> v[2].n[1])/det;
+   b -> v[2].n[2] = (a -> v[0].n[0]*a -> v[1].n[1] - a -> v[0].n[1]*a -> v[1].n[0])/det;
+
+   return TRUE;
+}
+
+
+// Solve a system in the form Ax = b
+cmsBool  CMSEXPORT _cmsMAT3solve(cmsVEC3* x, cmsMAT3* a, cmsVEC3* b)
+{
+    cmsMAT3 m, a_1;
+
+    memmove(&m, a, sizeof(cmsMAT3));
+
+    if (!_cmsMAT3inverse(&m, &a_1)) return FALSE;  // Singular matrix
+
+    _cmsMAT3eval(x, &a_1, b);
+    return TRUE;
+}
+
+// Evaluate a vector across a matrix
+void CMSEXPORT _cmsMAT3eval(cmsVEC3* r, const cmsMAT3* a, const cmsVEC3* v)
+{
+    r->n[VX] = a->v[0].n[VX]*v->n[VX] + a->v[0].n[VY]*v->n[VY] + a->v[0].n[VZ]*v->n[VZ];
+    r->n[VY] = a->v[1].n[VX]*v->n[VX] + a->v[1].n[VY]*v->n[VY] + a->v[1].n[VZ]*v->n[VZ];
+    r->n[VZ] = a->v[2].n[VX]*v->n[VX] + a->v[2].n[VY]*v->n[VY] + a->v[2].n[VZ]*v->n[VZ];
+}
+
+
diff --git a/thirdparty/liblcms2/src/cmsnamed.c b/thirdparty/liblcms2/src/cmsnamed.c
new file mode 100644
index 00000000..d1a86b6e
--- /dev/null
+++ b/thirdparty/liblcms2/src/cmsnamed.c
@@ -0,0 +1,750 @@
+//---------------------------------------------------------------------------------
+//
+//  Little Color Management System
+//  Copyright (c) 1998-2010 Marti Maria Saguer
+//
+// Permission is hereby granted, free of charge, to any person obtaining 
+// a copy of this software and associated documentation files (the "Software"), 
+// to deal in the Software without restriction, including without limitation 
+// the rights to use, copy, modify, merge, publish, distribute, sublicense, 
+// and/or sell copies of the Software, and to permit persons to whom the Software 
+// is furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in 
+// all copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 
+// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO 
+// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE 
+// LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 
+// OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 
+// WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+//
+//---------------------------------------------------------------------------------
+//
+
+#include "lcms2_internal.h"
+
+// Multilocalized unicode objects. That is an attempt to encapsulate i18n.
+
+
+// Allocates an empty multi localizad unicode object
+cmsMLU* CMSEXPORT cmsMLUalloc(cmsContext ContextID, cmsUInt32Number nItems)
+{
+	cmsMLU* mlu;
+
+	// nItems should be positive if given
+	if (nItems <= 0) nItems = 2;
+
+	// Create the container
+	mlu = (cmsMLU*) _cmsMallocZero(ContextID, sizeof(cmsMLU));
+	if (mlu == NULL) return NULL;
+
+	mlu ->ContextID = ContextID;
+
+	// Create entry array
+	mlu ->Entries = (_cmsMLUentry*) _cmsCalloc(ContextID, nItems, sizeof(_cmsMLUentry));
+	if (mlu ->Entries == NULL) {
+		_cmsFree(ContextID, mlu);
+		return NULL;
+	}
+
+	// Ok, keep indexes up to date
+	mlu ->AllocatedEntries    = nItems;
+	mlu ->UsedEntries         = 0;
+
+	return mlu;
+}
+
+
+// Grows a mempool table for a MLU. Each time this function is called, mempool size is multiplied times two. 
+static
+cmsBool GrowMLUpool(cmsMLU* mlu)
+{
+	cmsUInt32Number size;
+	void *NewPtr;
+
+	// Sanity check
+	if (mlu == NULL) return FALSE;
+
+	if (mlu ->PoolSize == 0)
+		size = 256;
+	else 
+		size = mlu ->PoolSize * 2;
+
+	// Check for overflow
+	if (size < mlu ->PoolSize) return FALSE;
+
+	// Reallocate the pool
+	NewPtr = _cmsRealloc(mlu ->ContextID, mlu ->MemPool, size);
+	if (NewPtr == NULL) return FALSE;
+
+
+	mlu ->MemPool  = NewPtr;
+	mlu ->PoolSize = size;
+
+	return TRUE;
+}
+
+
+// Grows a ntry table for a MLU. Each time this function is called, table size is multiplied times two. 
+static
+cmsBool GrowMLUtable(cmsMLU* mlu)
+{
+    int AllocatedEntries;
+    _cmsMLUentry *NewPtr;
+	
+	// Sanity check
+	if (mlu == NULL) return FALSE;
+
+    AllocatedEntries = mlu ->AllocatedEntries * 2;
+
+	// Check for overflow
+	if (AllocatedEntries / 2 != mlu ->AllocatedEntries) return FALSE;
+
+	// Reallocate the memory
+    NewPtr = (_cmsMLUentry*)_cmsRealloc(mlu ->ContextID, mlu ->Entries, AllocatedEntries*sizeof(_cmsMLUentry));
+    if (NewPtr == NULL) return FALSE;
+    
+    mlu ->Entries          = NewPtr;
+    mlu ->AllocatedEntries = AllocatedEntries;
+
+    return TRUE;
+}
+
+
+// Search for a specific entry in the structure. Language and Country are used. 
+static
+int SearchMLUEntry(cmsMLU* mlu, cmsUInt16Number LanguageCode, cmsUInt16Number CountryCode)
+{
+    int i;
+	
+	// Sanity check
+	if (mlu == NULL) return -1;
+
+	// Iterate whole table
+    for (i=0; i < mlu ->UsedEntries; i++) {
+
+        if (mlu ->Entries[i].Country  == CountryCode && 
+            mlu ->Entries[i].Language == LanguageCode) return i;
+    }
+
+	// Not found
+    return -1;
+}
+
+// Add a block of characters to the intended MLU. Language and country are specified. 
+// Only one entry for Language/country pair is allowed.
+static
+cmsBool AddMLUBlock(cmsMLU* mlu, cmsUInt32Number size, const wchar_t *Block, 
+                     cmsUInt16Number LanguageCode, cmsUInt16Number CountryCode)
+{
+    cmsUInt32Number Offset;
+    cmsUInt8Number* Ptr;
+
+	// Sanity check
+	if (mlu == NULL) return FALSE;
+
+    // Is there any room available?
+    if (mlu ->UsedEntries >= mlu ->AllocatedEntries) {
+        if (!GrowMLUtable(mlu)) return FALSE;
+    }
+
+    // Only one ASCII string
+    if (SearchMLUEntry(mlu, LanguageCode, CountryCode) >= 0) return FALSE;  // Only one  is allowed!
+
+    // Check for size
+	while ((mlu ->PoolSize - mlu ->PoolUsed) < size) {
+
+            if (!GrowMLUpool(mlu)) return FALSE;
+	}
+
+    Offset = mlu ->PoolUsed;
+    
+	Ptr = (cmsUInt8Number*) mlu ->MemPool;
+	if (Ptr == NULL) return FALSE;
+
+	// Set the entry
+    memmove(Ptr + Offset, Block, size);
+    mlu ->PoolUsed += size;
+    
+    mlu ->Entries[mlu ->UsedEntries].StrW     = Offset;
+    mlu ->Entries[mlu ->UsedEntries].Len      = size;
+    mlu ->Entries[mlu ->UsedEntries].Country  = CountryCode;
+    mlu ->Entries[mlu ->UsedEntries].Language = LanguageCode;
+    mlu ->UsedEntries++;
+
+    return TRUE;
+}
+
+
+// Add an ASCII entry. 
+cmsBool CMSEXPORT cmsMLUsetASCII(cmsMLU* mlu, const char LanguageCode[3], const char CountryCode[3], const char* ASCIIString)
+{
+    cmsUInt32Number i, len = (cmsUInt32Number) strlen(ASCIIString)+1;
+    wchar_t* WStr;
+    cmsBool  rc;
+    cmsUInt16Number Lang  = _cmsAdjustEndianess16(*(cmsUInt16Number*) LanguageCode);
+    cmsUInt16Number Cntry = _cmsAdjustEndianess16(*(cmsUInt16Number*) CountryCode);
+
+    if (mlu == NULL) return FALSE;
+
+    WStr = (wchar_t*) _cmsCalloc(mlu ->ContextID, len,  sizeof(wchar_t));
+    if (WStr == NULL) return FALSE;
+
+    for (i=0; i < len; i++)
+        WStr[i] = (wchar_t) ASCIIString[i];
+    
+    rc = AddMLUBlock(mlu, len  * sizeof(wchar_t), WStr, Lang, Cntry);
+
+    _cmsFree(mlu ->ContextID, WStr);
+    return rc;
+    
+}
+
+// We don't need any wcs support library
+static 
+cmsUInt32Number mywcslen(const wchar_t *s)
+{
+    const wchar_t *p;
+
+	p = s;
+    while (*p)
+        p++;
+
+    return (cmsUInt32Number)(p - s);
+}
+
+
+// Add a wide entry
+cmsBool  CMSEXPORT cmsMLUsetWide(cmsMLU* mlu, const char Language[3], const char Country[3], const wchar_t* WideString)
+{
+    cmsUInt16Number Lang  = _cmsAdjustEndianess16(*(cmsUInt16Number*) Language);
+    cmsUInt16Number Cntry = _cmsAdjustEndianess16(*(cmsUInt16Number*) Country);
+    cmsUInt32Number len;
+    
+    if (mlu == NULL) return FALSE;
+	if (WideString == NULL) return FALSE;
+
+    len = (cmsUInt32Number) (mywcslen(WideString) + 1) * sizeof(wchar_t);
+    return AddMLUBlock(mlu, len, WideString, Lang, Cntry);
+}
+
+// Duplicating a MLU is as easy as copying all members
+cmsMLU* CMSEXPORT cmsMLUdup(const cmsMLU* mlu)
+{
+	cmsMLU* NewMlu = NULL;
+
+	// Duplicating a NULL obtains a NULL
+	if (mlu == NULL) return NULL;
+
+	NewMlu = cmsMLUalloc(mlu ->ContextID, mlu ->UsedEntries);
+	if (NewMlu == NULL) return NULL;
+
+	// Should never happen
+	if (NewMlu ->AllocatedEntries < mlu ->UsedEntries)
+		goto Error;
+
+	// Sanitize...
+	if (NewMlu ->Entries == NULL || mlu ->Entries == NULL)  goto Error;
+
+	memmove(NewMlu ->Entries, mlu ->Entries, mlu ->UsedEntries * sizeof(_cmsMLUentry));
+	NewMlu ->UsedEntries = mlu ->UsedEntries;
+
+	// The MLU may be empty
+	if (mlu ->PoolUsed == 0) {
+		NewMlu ->MemPool = NULL;
+	}
+	else {
+		// It is not empty
+		NewMlu ->MemPool = _cmsMalloc(mlu ->ContextID, mlu ->PoolUsed);
+		if (NewMlu ->MemPool == NULL) goto Error;
+	}
+
+	NewMlu ->PoolSize = mlu ->PoolUsed;
+
+	if (NewMlu ->MemPool == NULL || mlu ->MemPool == NULL) goto Error;
+
+	memmove(NewMlu ->MemPool, mlu->MemPool, mlu ->PoolUsed);
+	NewMlu ->PoolUsed = mlu ->PoolUsed;
+
+	return NewMlu;
+
+Error:
+
+	if (NewMlu != NULL) cmsMLUfree(NewMlu);
+	return NULL;
+}
+
+// Free any used memory
+void CMSEXPORT cmsMLUfree(cmsMLU* mlu)
+{
+	if (mlu) {
+
+		if (mlu -> Entries) _cmsFree(mlu ->ContextID, mlu->Entries);
+		if (mlu -> MemPool) _cmsFree(mlu ->ContextID, mlu->MemPool);
+
+		_cmsFree(mlu ->ContextID, mlu);
+	}
+}
+
+
+// The algorithm first searches for an exact match of country and language, if not found it uses 
+// the Language. If none is found, first entry is used instead.
+static
+const wchar_t* _cmsMLUgetWide(const cmsMLU* mlu, 
+							  cmsUInt32Number *len, 
+							  cmsUInt16Number LanguageCode, cmsUInt16Number CountryCode,
+							  cmsUInt16Number* UsedLanguageCode, cmsUInt16Number* UsedCountryCode)
+{
+    int i;
+    int Best = -1;
+	_cmsMLUentry* v;
+
+    if (mlu == NULL) return NULL;
+
+    if (mlu -> AllocatedEntries <= 0) return NULL;
+
+    for (i=0; i < mlu ->UsedEntries; i++) {
+
+        v = mlu ->Entries + i;
+
+        if (v -> Language == LanguageCode) {
+
+            if (Best == -1) Best = i;
+
+            if (v -> Country == CountryCode) {
+
+				if (UsedLanguageCode != NULL) *UsedLanguageCode = v ->Language;
+				if (UsedCountryCode  != NULL) *UsedCountryCode = v ->Country;
+
+                if (len != NULL) *len = v ->Len;
+
+                return (wchar_t*) ((cmsUInt8Number*) mlu ->MemPool + v -> StrW);        // Found exact match                       
+            }
+        }
+    }
+
+    // No string found. Return First one
+    if (Best == -1)
+        Best = 0;
+
+	 v = mlu ->Entries + Best;
+
+	 if (UsedLanguageCode != NULL) *UsedLanguageCode = v ->Language;
+	if (UsedCountryCode  != NULL) *UsedCountryCode = v ->Country;
+
+    if (len != NULL) *len   = v ->Len;
+
+	return(wchar_t*) ((cmsUInt8Number*) mlu ->MemPool + v ->StrW);
+}
+
+
+// Obtain an ASCII representation of the wide string. Setting buffer to NULL returns the len
+cmsUInt32Number CMSEXPORT cmsMLUgetASCII(const cmsMLU* mlu, 
+									   const char LanguageCode[3], const char CountryCode[3], 
+									   char* Buffer, cmsUInt32Number BufferSize)
+{
+    const wchar_t *Wide;
+    cmsUInt32Number  StrLen = 0;
+    cmsUInt32Number ASCIIlen, i;
+
+	cmsUInt16Number Lang  = _cmsAdjustEndianess16(*(cmsUInt16Number*) LanguageCode);
+    cmsUInt16Number Cntry = _cmsAdjustEndianess16(*(cmsUInt16Number*) CountryCode);
+
+	// Sanitize
+    if (mlu == NULL) return 0;
+
+    // Get WideChar
+    Wide = _cmsMLUgetWide(mlu, &StrLen, Lang, Cntry, NULL, NULL);
+    if (Wide == NULL) return 0;
+
+    ASCIIlen = StrLen / sizeof(wchar_t);
+
+    // Maybe we want only to know the len?
+    if (Buffer == NULL) return ASCIIlen + 1; // Note the zero at the end
+
+    // No buffer size means no data
+    if (BufferSize <= 0) return 0;
+
+    // Some clipping may be required
+    if (BufferSize < ASCIIlen + 1)
+        ASCIIlen = BufferSize - 1;
+
+    // Precess each character
+    for (i=0; i < ASCIIlen; i++) { 
+
+        if (Wide[i] == 0)
+            Buffer[i] = 0;
+        else
+            Buffer[i] = (char) Wide[i];
+    }
+
+	// We put a termination "\0"
+    Buffer[ASCIIlen] = 0;
+    return ASCIIlen + 1;
+}
+
+// Obtain a wide representation of the MLU, on depending on current locale settings 
+cmsUInt32Number CMSEXPORT cmsMLUgetWide(const cmsMLU* mlu, 
+									  const char LanguageCode[3], const char CountryCode[3], 
+									  wchar_t* Buffer, cmsUInt32Number BufferSize)
+{
+    const wchar_t *Wide;
+    cmsUInt32Number  StrLen = 0;
+
+	cmsUInt16Number Lang  = _cmsAdjustEndianess16(*(cmsUInt16Number*) LanguageCode);
+    cmsUInt16Number Cntry = _cmsAdjustEndianess16(*(cmsUInt16Number*) CountryCode);
+
+	// Sanitize
+    if (mlu == NULL) return 0;
+
+    Wide = _cmsMLUgetWide(mlu, &StrLen, Lang, Cntry, NULL, NULL);
+    if (Wide == NULL) return 0;
+    
+    // Maybe we want only to know the len?
+    if (Buffer == NULL) return StrLen + sizeof(wchar_t);
+
+  // No buffer size means no data
+    if (BufferSize <= 0) return 0;
+
+    // Some clipping may be required
+    if (BufferSize < StrLen + sizeof(wchar_t))
+        StrLen = BufferSize - + sizeof(wchar_t);
+
+    memmove(Buffer, Wide, StrLen);
+	Buffer[StrLen / sizeof(wchar_t)] = 0;
+
+    return StrLen + sizeof(wchar_t);
+}
+
+
+// Get also the language and country
+CMSAPI cmsBool CMSEXPORT cmsMLUgetTranslation(const cmsMLU* mlu,   
+		                    		          const char LanguageCode[3], const char CountryCode[3], 
+										      char ObtainedLanguage[3], char ObtainedCountry[3])
+{
+	const wchar_t *Wide;
+ 
+	cmsUInt16Number Lang  = _cmsAdjustEndianess16(*(cmsUInt16Number*) LanguageCode);
+    cmsUInt16Number Cntry = _cmsAdjustEndianess16(*(cmsUInt16Number*) CountryCode);
+    cmsUInt16Number ObtLang, ObtCode; 
+
+	// Sanitize
+    if (mlu == NULL) return FALSE;
+
+    Wide = _cmsMLUgetWide(mlu, NULL, Lang, Cntry, &ObtLang, &ObtCode);
+	if (Wide == NULL) return FALSE;
+    
+	// Get used language and code
+    *(cmsUInt16Number *)ObtainedLanguage = _cmsAdjustEndianess16(ObtLang);
+	*(cmsUInt16Number *)ObtainedCountry  = _cmsAdjustEndianess16(ObtCode);	
+
+	ObtainedLanguage[2] = ObtainedCountry[2] = 0;    	
+	return TRUE;
+}
+
+
+// Named color lists --------------------------------------------------------------------------------------------
+
+// Grow the list to keep at least NumElements
+static
+cmsBool  GrowNamedColorList(cmsNAMEDCOLORLIST* v)
+{           
+    cmsUInt32Number size;
+    _cmsNAMEDCOLOR * NewPtr;
+
+    if (v == NULL) return FALSE;
+
+    if (v ->Allocated == 0)
+        size = 64;   // Initial guess
+    else
+        size = v ->Allocated * 2;
+
+    // Keep a maximum color lists can grow, 100K entries seems reasonable
+    if (size > 1024*100) return FALSE;
+
+    NewPtr = (_cmsNAMEDCOLOR*) _cmsRealloc(v ->ContextID, v ->List, size * sizeof(_cmsNAMEDCOLOR));
+    if (NewPtr == NULL) 
+        return FALSE;
+        
+    v ->List      = NewPtr;
+    v ->Allocated = size;
+    return TRUE;
+}
+
+// Allocate a list for n elements
+cmsNAMEDCOLORLIST* CMSEXPORT cmsAllocNamedColorList(cmsContext ContextID, cmsUInt32Number n, cmsUInt32Number ColorantCount, const char* Prefix, const char* Suffix)
+{
+    cmsNAMEDCOLORLIST* v = (cmsNAMEDCOLORLIST*) _cmsMallocZero(ContextID, sizeof(cmsNAMEDCOLORLIST));
+   
+    if (v == NULL) return NULL;
+    
+    v ->List      = NULL;
+    v ->nColors   = 0;
+    v ->ContextID  = ContextID;
+
+    while (v -> Allocated < n)
+        GrowNamedColorList(v);
+
+    strncpy(v ->Prefix, Prefix, sizeof(v ->Prefix));
+    strncpy(v ->Suffix, Suffix, sizeof(v ->Suffix));
+    v -> ColorantCount = ColorantCount;
+
+    return v;
+}
+
+// Free a list
+void CMSEXPORT cmsFreeNamedColorList(cmsNAMEDCOLORLIST* v)
+{               
+    if (v ->List) _cmsFree(v ->ContextID, v ->List);
+    if (v) _cmsFree(v ->ContextID, v);
+}   
+
+cmsNAMEDCOLORLIST* CMSEXPORT cmsDupNamedColorList(const cmsNAMEDCOLORLIST* v)
+{
+    cmsNAMEDCOLORLIST* NewNC;
+        
+    if (v == NULL) return NULL;
+
+    NewNC= cmsAllocNamedColorList(v ->ContextID, v -> nColors, v ->ColorantCount, v ->Prefix, v ->Suffix);
+    if (NewNC == NULL) return NULL;
+
+    // For really large tables we need this
+    while (NewNC ->Allocated < v ->Allocated)
+        GrowNamedColorList(NewNC);
+
+    memmove(NewNC ->Prefix, v ->Prefix, sizeof(v ->Prefix));
+    memmove(NewNC ->Suffix, v ->Suffix, sizeof(v ->Suffix));
+    NewNC ->ColorantCount = v ->ColorantCount;
+    memmove(NewNC->List, v ->List, v->nColors * sizeof(_cmsNAMEDCOLOR));
+    NewNC ->nColors = v ->nColors;
+    return NewNC;
+}
+
+
+// Append a color to a list. List pointer may change if reallocated
+cmsBool  CMSEXPORT cmsAppendNamedColor(cmsNAMEDCOLORLIST* NamedColorList, 
+                                       const char* Name, 
+                                       cmsUInt16Number PCS[3], cmsUInt16Number Colorant[cmsMAXCHANNELS])
+{    
+    cmsUInt32Number i;
+
+    if (NamedColorList == NULL) return FALSE;
+
+    if (NamedColorList ->nColors + 1 > NamedColorList ->Allocated) {
+        if (!GrowNamedColorList(NamedColorList)) return FALSE;
+    }
+
+    for (i=0; i < NamedColorList ->ColorantCount; i++)
+        NamedColorList ->List[NamedColorList ->nColors].DeviceColorant[i] = Colorant == NULL? 0 : Colorant[i];
+
+    for (i=0; i < 3; i++)
+        NamedColorList ->List[NamedColorList ->nColors].PCS[i] = PCS == NULL ? 0 : PCS[i];
+
+    if (Name != NULL)
+        strncpy(NamedColorList ->List[NamedColorList ->nColors].Name, Name, 
+                    sizeof(NamedColorList ->List[NamedColorList ->nColors].Name));
+    else
+        NamedColorList ->List[NamedColorList ->nColors].Name[0] = 0;
+
+
+    NamedColorList ->nColors++;
+    return TRUE;
+}
+
+// Returns number of elements 
+cmsUInt32Number CMSEXPORT cmsNamedColorCount(const cmsNAMEDCOLORLIST* NamedColorList)
+{    
+     if (NamedColorList == NULL) return 0;
+     return NamedColorList ->nColors;
+}
+
+// Info aboout a given color
+cmsBool  CMSEXPORT cmsNamedColorInfo(const cmsNAMEDCOLORLIST* NamedColorList, cmsUInt32Number nColor, 
+                                     char* Name, 
+                                     char* Prefix, 
+                                     char* Suffix,
+                                     cmsUInt16Number* PCS, 
+                                     cmsUInt16Number* Colorant)
+{    
+    if (NamedColorList == NULL) return FALSE;
+
+    if (nColor >= cmsNamedColorCount(NamedColorList)) return FALSE;
+
+    if (Name) strcpy(Name, NamedColorList->List[nColor].Name);
+    if (Prefix) strcpy(Prefix, NamedColorList->Prefix);
+    if (Suffix) strcpy(Suffix, NamedColorList->Suffix);
+    if (PCS) 
+        memmove(PCS, NamedColorList ->List[nColor].PCS, 3*sizeof(cmsUInt16Number));
+
+    if (Colorant) 
+        memmove(Colorant, NamedColorList ->List[nColor].DeviceColorant, 
+                                sizeof(cmsUInt16Number) * NamedColorList ->ColorantCount);
+
+
+    return TRUE;
+}
+
+// Search for a given color name (no prefix or suffix)
+cmsInt32Number CMSEXPORT cmsNamedColorIndex(const cmsNAMEDCOLORLIST* NamedColorList, const char* Name)
+{    
+    int i, n;
+
+    if (NamedColorList == NULL) return -1;
+    n = cmsNamedColorCount(NamedColorList);
+    for (i=0; i < n; i++) {
+        if (cmsstrcasecmp(Name,  NamedColorList->List[i].Name) == 0)
+            return i;
+    }
+
+    return -1;
+}
+
+// MPE support -----------------------------------------------------------------------------------------------------------------
+
+static
+void FreeNamedColorList(cmsStage* mpe)
+{
+    cmsNAMEDCOLORLIST* List = (cmsNAMEDCOLORLIST*) mpe ->Data;
+    cmsFreeNamedColorList(List);
+}
+
+static
+void* DupNamedColorList(cmsStage* mpe)
+{
+    cmsNAMEDCOLORLIST* List = (cmsNAMEDCOLORLIST*) mpe ->Data;
+    return cmsDupNamedColorList(List);
+}
+
+static
+void EvalNamedColor(const cmsFloat32Number In[], cmsFloat32Number Out[], const cmsStage *mpe)
+{
+    cmsNAMEDCOLORLIST* NamedColorList = (cmsNAMEDCOLORLIST*) mpe ->Data;
+    cmsUInt16Number index = (cmsUInt16Number) _cmsQuickSaturateWord(In[0] * 65535.0);
+    cmsUInt32Number j;
+
+    if (index >= NamedColorList-> nColors) {
+        cmsSignalError(NamedColorList ->ContextID, cmsERROR_RANGE, "Color %d out of range; ignored", index);
+    }
+    else {
+        for (j=0; j < NamedColorList ->ColorantCount; j++) 
+            Out[j] = (cmsFloat32Number) (NamedColorList->List[index].DeviceColorant[j] / 65535.0);      
+    }
+}
+
+
+// Named color lookup element
+cmsStage* _cmsStageAllocNamedColor(cmsNAMEDCOLORLIST* NamedColorList)
+{
+    return _cmsStageAllocPlaceholder(NamedColorList ->ContextID, 
+		                           cmsSigNamedColorElemType, 
+								   1, 3,
+								   EvalNamedColor,
+								   DupNamedColorList,
+								   FreeNamedColorList,
+								   cmsDupNamedColorList(NamedColorList));
+  
+}
+
+
+// Retrieve the named color list from a transform. Should be first element in the LUT
+cmsNAMEDCOLORLIST* CMSEXPORT cmsGetNamedColorList(cmsHTRANSFORM xform)
+{
+    _cmsTRANSFORM* v = (_cmsTRANSFORM*) xform;
+    cmsStage* mpe  = v ->Lut->Elements;
+
+    if (mpe ->Type != cmsSigNamedColorElemType) return NULL;
+    return (cmsNAMEDCOLORLIST*) mpe ->Data;
+}
+
+
+// Profile sequence description routines -------------------------------------------------------------------------------------
+
+cmsSEQ* CMSEXPORT cmsAllocProfileSequenceDescription(cmsContext ContextID, cmsUInt32Number n)
+{
+    cmsSEQ* Seq;
+    cmsUInt32Number i;
+
+    if (n == 0) return NULL;
+
+    // In a absolutely arbitrary way, I hereby decide to allow a maxim of 255 profiles linked
+    // in a devicelink. It makes not sense anyway and may be used for exploits, so let's close the door!
+    if (n > 255) return NULL;
+
+    Seq = (cmsSEQ*) _cmsMallocZero(ContextID, sizeof(cmsSEQ));
+    if (Seq == NULL) return NULL;   
+    
+    Seq -> ContextID = ContextID;
+    Seq -> seq      = (cmsPSEQDESC*) _cmsCalloc(ContextID, n, sizeof(cmsPSEQDESC));
+    Seq -> n        = n;
+
+    
+    for (i=0; i < n; i++) {
+        Seq -> seq[i].Manufacturer = NULL;
+        Seq -> seq[i].Model        = NULL;
+        Seq -> seq[i].Description  = NULL;
+    }
+    
+    return Seq;
+}
+
+void CMSEXPORT cmsFreeProfileSequenceDescription(cmsSEQ* pseq)
+{
+    cmsUInt32Number i;
+
+    for (i=0; i < pseq ->n; i++) {
+        if (pseq ->seq[i].Manufacturer != NULL) cmsMLUfree(pseq ->seq[i].Manufacturer);
+        if (pseq ->seq[i].Model != NULL) cmsMLUfree(pseq ->seq[i].Model);
+        if (pseq ->seq[i].Description != NULL) cmsMLUfree(pseq ->seq[i].Description);
+    }
+
+    if (pseq ->seq != NULL) _cmsFree(pseq ->ContextID, pseq ->seq);
+    _cmsFree(pseq -> ContextID, pseq);
+}
+
+cmsSEQ* CMSEXPORT cmsDupProfileSequenceDescription(const cmsSEQ* pseq)
+{
+    cmsSEQ *NewSeq;
+    cmsUInt32Number i;
+
+    if (pseq == NULL)
+        return NULL;
+
+    NewSeq = (cmsSEQ*) _cmsMalloc(pseq -> ContextID, sizeof(cmsSEQ));
+    if (NewSeq == NULL) return NULL;
+
+    
+    NewSeq -> seq      = (cmsPSEQDESC*) _cmsCalloc(pseq ->ContextID, pseq ->n, sizeof(cmsPSEQDESC));
+    if (NewSeq ->seq == NULL) goto Error;
+    
+    NewSeq -> ContextID = pseq ->ContextID;
+    NewSeq -> n        = pseq ->n;
+
+    for (i=0; i < pseq->n; i++) {
+
+        memmove(&NewSeq ->seq[i].attributes, &pseq ->seq[i].attributes, sizeof(cmsUInt64Number));
+
+        NewSeq ->seq[i].deviceMfg   = pseq ->seq[i].deviceMfg;
+        NewSeq ->seq[i].deviceModel = pseq ->seq[i].deviceModel;
+        memmove(&NewSeq ->seq[i].ProfileID, &pseq ->seq[i].ProfileID, sizeof(cmsProfileID));
+        NewSeq ->seq[i].technology  = pseq ->seq[i].technology;
+
+        NewSeq ->seq[i].Manufacturer = cmsMLUdup(pseq ->seq[i].Manufacturer);
+        NewSeq ->seq[i].Model        = cmsMLUdup(pseq ->seq[i].Model);
+        NewSeq ->seq[i].Description  = cmsMLUdup(pseq ->seq[i].Description);
+    
+    }
+
+    return NewSeq;
+
+Error:
+
+    cmsFreeProfileSequenceDescription(NewSeq);
+    return NULL;
+}
+
+
+
+
diff --git a/thirdparty/liblcms2/src/cmsopt.c b/thirdparty/liblcms2/src/cmsopt.c
new file mode 100644
index 00000000..b1ce98e3
--- /dev/null
+++ b/thirdparty/liblcms2/src/cmsopt.c
@@ -0,0 +1,1631 @@
+
+//---------------------------------------------------------------------------------
+//
+//  Little Color Management System
+//  Copyright (c) 1998-2010 Marti Maria Saguer
+//
+// Permission is hereby granted, free of charge, to any person obtaining 
+// a copy of this software and associated documentation files (the "Software"), 
+// to deal in the Software without restriction, including without limitation 
+// the rights to use, copy, modify, merge, publish, distribute, sublicense, 
+// and/or sell copies of the Software, and to permit persons to whom the Software 
+// is furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in 
+// all copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 
+// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO 
+// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE 
+// LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 
+// OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 
+// WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+//
+//---------------------------------------------------------------------------------
+//
+
+#include "lcms2_internal.h"
+
+
+//----------------------------------------------------------------------------------
+
+// Optimization for 8 bits, Shaper-CLUT (3 inputs only)
+typedef struct {
+
+    cmsContext ContextID;
+
+    const cmsInterpParams* p;   // Tetrahedrical interpolation parameters. This is a not-owned pointer.
+
+    cmsUInt16Number rx[256], ry[256], rz[256];
+    cmsUInt32Number X0[256], Y0[256], Z0[256];  // Precomputed nodes and offsets for 8-bit input data
+    
+
+} Prelin8Data;
+
+
+// Generic optimization for 16 bits Shaper-CLUT-Shaper (any inputs)
+typedef struct {
+
+    cmsContext ContextID;
+
+    // Number of channels
+    int nInputs;
+    int nOutputs;
+    
+    // Since there is no limitation of the output number of channels, this buffer holding the connexion CLUT-shaper
+    // has to be dynamically allocated. This is not the case of first step shaper-CLUT, which is limited to max inputs
+    cmsUInt16Number* StageDEF;
+
+    _cmsInterpFn16 EvalCurveIn16[MAX_INPUT_DIMENSIONS];       // The maximum number of input channels is known in advance 
+    cmsInterpParams*  ParamsCurveIn16[MAX_INPUT_DIMENSIONS];  
+    
+    _cmsInterpFn16 EvalCLUT;            // The evaluator for 3D grid
+    const cmsInterpParams* CLUTparams;  // (not-owned pointer)
+
+    
+    _cmsInterpFn16* EvalCurveOut16;       // Points to an array of curve evaluators in 16 bits (not-owned pointer)
+    cmsInterpParams**  ParamsCurveOut16;  // Points to an array of references to interpolation params (not-owned pointer)
+    
+
+} Prelin16Data;
+
+
+// Optimization for matrix-shaper in 8 bits. Numbers are operated in n.14 signed, tables are stored in 1.14 fixed 
+
+typedef cmsInt32Number cmsS1Fixed14Number;   // Note that this may hold more than 16 bits!
+
+#define DOUBLE_TO_1FIXED14(x) ((cmsS1Fixed14Number) floor((x) * 16384.0 + 0.5))
+
+typedef struct {
+ 
+    cmsContext ContextID;
+
+    cmsS1Fixed14Number Shaper1R[256];  // from 0..255 to 1.14  (0.0...1.0)
+    cmsS1Fixed14Number Shaper1G[256];
+    cmsS1Fixed14Number Shaper1B[256];
+
+    cmsS1Fixed14Number Mat[3][3];     // n.14 to n.14 (needs a saturation after that)
+    cmsS1Fixed14Number Off[3];
+
+    cmsUInt16Number Shaper2R[16385];    // 1.14 to 0..255 
+    cmsUInt16Number Shaper2G[16385];
+    cmsUInt16Number Shaper2B[16385];
+
+} MatShaper8Data;
+
+// Curves, optimization is shared between 8 and 16 bits
+typedef struct {
+
+    cmsContext ContextID;
+
+    int nCurves;                  // Number of curves
+    int nElements;                // Elements in curves
+    cmsUInt16Number** Curves;     // Points to a dynamically  allocated array
+
+} Curves16Data;
+
+
+// Simple optimizations ----------------------------------------------------------------------------------------------------------
+
+
+// Remove an element in linked chain
+static
+void _RemoveElement(cmsStage** head)
+{
+    cmsStage* mpe = *head;
+    cmsStage* next = mpe ->Next;
+    *head = next;
+    cmsStageFree(mpe);
+}
+
+// Remove all identities in chain. Note that pt actually is a double pointer to the element that holds the pointer. 
+static
+cmsBool _Remove1Op(cmsPipeline* Lut, cmsStageSignature UnaryOp)
+{   
+    cmsStage** pt = &Lut ->Elements;
+    cmsBool AnyOpt = FALSE;
+
+    while (*pt != NULL) {
+
+        if ((*pt) ->Implements == UnaryOp) {
+            _RemoveElement(pt);
+            AnyOpt = TRUE;
+        }
+        else  
+            pt = &((*pt) -> Next);
+    }
+
+    return AnyOpt;
+}
+
+// Same, but only if two adjacent elements are found
+static
+cmsBool _Remove2Op(cmsPipeline* Lut, cmsStageSignature Op1, cmsStageSignature Op2)
+{   
+    cmsStage** pt1;
+    cmsStage** pt2;
+    cmsBool AnyOpt = FALSE;
+
+    pt1 = &Lut ->Elements;
+    if (*pt1 == NULL) return AnyOpt;
+    
+    while (*pt1 != NULL) {
+
+        pt2 = &((*pt1) -> Next);
+        if (*pt2 == NULL) return AnyOpt;
+
+        if ((*pt1) ->Implements == Op1 && (*pt2) ->Implements == Op2) {
+            _RemoveElement(pt2);
+            _RemoveElement(pt1);
+            AnyOpt = TRUE;
+        }
+        else  
+            pt1 = &((*pt1) -> Next);            
+    }
+
+    return AnyOpt;
+}
+
+// Preoptimize just gets rif of no-ops coming paired. Conversion from v2 to v4 followed 
+// by a v4 to v2 and vice-versa. The elements are then discarded.
+static
+cmsBool PreOptimize(cmsPipeline* Lut)
+{    
+    cmsBool AnyOpt = FALSE, Opt;
+
+    AnyOpt = FALSE;
+
+    do {
+
+        Opt = FALSE;
+
+        // Remove all identities
+        Opt |= _Remove1Op(Lut, cmsSigIdentityElemType);
+
+        // Remove XYZ2Lab followed by Lab2XYZ
+        Opt |= _Remove2Op(Lut, cmsSigXYZ2LabElemType, cmsSigLab2XYZElemType);
+
+        // Remove Lab2XYZ followed by XYZ2Lab
+        Opt |= _Remove2Op(Lut, cmsSigLab2XYZElemType, cmsSigXYZ2LabElemType);
+
+        // Remove V4 to V2 followed by V2 to V4
+        Opt |= _Remove2Op(Lut, cmsSigLabV4toV2, cmsSigLabV2toV4);
+
+        // Remove V2 to V4 followed by V4 to V2
+        Opt |= _Remove2Op(Lut, cmsSigLabV2toV4, cmsSigLabV4toV2);
+
+        if (Opt) AnyOpt = TRUE;
+
+    } while (Opt);
+
+    return AnyOpt;
+}
+
+static
+void Eval16nop1D(register const cmsUInt16Number Input[],
+                 register cmsUInt16Number Output[],                               
+                 register const struct _cms_interp_struc* p)
+{
+    Output[0] = Input[0];
+
+    cmsUNUSED_PARAMETER(p);  
+}
+
+static
+void PrelinEval16(register const cmsUInt16Number Input[],
+                  register cmsUInt16Number Output[],
+                  register const void* D)
+{
+    Prelin16Data* p16 = (Prelin16Data*) D;
+    cmsUInt16Number  StageABC[MAX_INPUT_DIMENSIONS];
+    int i;
+
+    for (i=0; i < p16 ->nInputs; i++) {
+    
+        p16 ->EvalCurveIn16[i](&Input[i], &StageABC[i], p16 ->ParamsCurveIn16[i]);
+    }
+
+    p16 ->EvalCLUT(StageABC, p16 ->StageDEF, p16 ->CLUTparams);
+
+    for (i=0; i < p16 ->nOutputs; i++) {
+    
+        p16 ->EvalCurveOut16[i](&p16->StageDEF[i], &Output[i], p16 ->ParamsCurveOut16[i]);
+    }
+}
+
+
+static
+void PrelinOpt16free(cmsContext ContextID, void* ptr)
+{
+    Prelin16Data* p16 = (Prelin16Data*) ptr;
+
+    _cmsFree(ContextID, p16 ->StageDEF);
+    _cmsFree(ContextID, p16 ->EvalCurveOut16);
+    _cmsFree(ContextID, p16 ->ParamsCurveOut16);
+
+    _cmsFree(ContextID, p16);
+}
+
+static
+void* Prelin16dup(cmsContext ContextID, const void* ptr)
+{   
+    Prelin16Data* p16 = (Prelin16Data*) ptr;
+    Prelin16Data* Duped = _cmsDupMem(ContextID, p16, sizeof(Prelin16Data));
+
+    if (Duped == NULL) return NULL;
+
+    Duped ->StageDEF         = _cmsCalloc(ContextID, p16 ->nOutputs, sizeof(cmsUInt16Number)); 
+    Duped ->EvalCurveOut16   = _cmsDupMem(ContextID, p16 ->EvalCurveOut16, p16 ->nOutputs * sizeof(_cmsInterpFn16));
+    Duped ->ParamsCurveOut16 = _cmsDupMem(ContextID, p16 ->ParamsCurveOut16, p16 ->nOutputs * sizeof(cmsInterpParams* ));
+
+    return Duped;
+}
+
+
+static
+Prelin16Data* PrelinOpt16alloc(cmsContext ContextID, 
+                               const cmsInterpParams* ColorMap, 
+                               int nInputs, cmsToneCurve** In, 
+                               int nOutputs, cmsToneCurve** Out )
+{
+    int i;
+    Prelin16Data* p16 = (Prelin16Data*) _cmsMallocZero(ContextID, sizeof(Prelin16Data));
+    if (p16 == NULL) return NULL;
+
+    p16 ->nInputs = nInputs;
+    p16 -> nOutputs = nOutputs;
+
+    
+    for (i=0; i < nInputs; i++) {
+
+        if (In == NULL) {
+            p16 -> ParamsCurveIn16[i] = NULL;
+            p16 -> EvalCurveIn16[i] = Eval16nop1D;
+
+        }
+        else {
+            p16 -> ParamsCurveIn16[i] = In[i] ->InterpParams;
+            p16 -> EvalCurveIn16[i] = p16 ->ParamsCurveIn16[i]->Interpolation.Lerp16;
+        }
+    }
+
+    p16 ->CLUTparams = ColorMap;
+    p16 ->EvalCLUT   = ColorMap ->Interpolation.Lerp16;
+
+
+    p16 -> StageDEF = _cmsCalloc(ContextID, p16 ->nOutputs, sizeof(cmsUInt16Number)); 
+    p16 -> EvalCurveOut16 = (_cmsInterpFn16*) _cmsCalloc(ContextID, nOutputs, sizeof(_cmsInterpFn16));
+    p16 -> ParamsCurveOut16 = (cmsInterpParams**) _cmsCalloc(ContextID, nOutputs, sizeof(cmsInterpParams* ));
+
+    for (i=0; i < nOutputs; i++) {
+
+        if (Out == NULL) {
+            p16 ->ParamsCurveOut16[i] = NULL;
+            p16 -> EvalCurveOut16[i] = Eval16nop1D;
+        }
+        else {
+
+            p16 ->ParamsCurveOut16[i] = Out[i] ->InterpParams;
+            p16 -> EvalCurveOut16[i] = p16 ->ParamsCurveOut16[i]->Interpolation.Lerp16;
+        }
+    }
+
+    return p16;
+}
+
+
+
+// Resampling ---------------------------------------------------------------------------------
+
+#define PRELINEARIZATION_POINTS 4096
+
+// Sampler implemented by another LUT. This is a clean way to precalculate the devicelink 3D CLUT for 
+// almost any transform. We use floating point precision and then convert from floating point to 16 bits.
+static
+int XFormSampler16(register const cmsUInt16Number In[], register cmsUInt16Number Out[], register void* Cargo)
+{
+    cmsPipeline* Lut = (cmsPipeline*) Cargo;
+    cmsFloat32Number InFloat[cmsMAXCHANNELS], OutFloat[cmsMAXCHANNELS];
+    cmsUInt32Number i;
+
+    _cmsAssert(Lut -> InputChannels < cmsMAXCHANNELS);
+    _cmsAssert(Lut -> OutputChannels < cmsMAXCHANNELS);
+
+    // From 16 bit to floating point
+    for (i=0; i < Lut ->InputChannels; i++) 
+        InFloat[i] = (cmsFloat32Number) (In[i] / 65535.0);
+
+    // Evaluate in floating point
+    cmsPipelineEvalFloat(InFloat, OutFloat, Lut);
+
+    // Back to 16 bits representation
+    for (i=0; i < Lut ->OutputChannels; i++) 
+        Out[i] = _cmsQuickSaturateWord(OutFloat[i] * 65535.0);
+
+    // Always succeed
+    return TRUE;
+}
+
+// Try to see if the curves of a given MPE are linear
+static
+cmsBool AllCurvesAreLinear(cmsStage* mpe)
+{
+    cmsToneCurve** Curves; 
+    cmsUInt32Number i, n;
+
+    Curves = _cmsStageGetPtrToCurveSet(mpe);
+    if (Curves == NULL) return FALSE;  
+
+    n = cmsStageOutputChannels(mpe);
+
+    for (i=0; i < n; i++) {
+        if (!cmsIsToneCurveLinear(Curves[i])) return FALSE;
+    }
+
+    return TRUE;
+}
+
+// This function replaces a specific node placed in "At" by the "Value" numbers. Its purpose
+// is to fix scum dot on broken profiles/transforms. Works on 1, 3 and 4 channels
+static
+cmsBool  PatchLUT(cmsStage* CLUT, cmsUInt16Number At[], cmsUInt16Number Value[],
+                  int nChannelsOut, int nChannelsIn)
+{
+    _cmsStageCLutData* Grid = (_cmsStageCLutData*) CLUT ->Data;
+    cmsInterpParams* p16  = Grid ->Params;
+    cmsFloat64Number px, py, pz, pw;
+    int        x0, y0, z0, w0;
+    int        i, index;
+
+    if (CLUT -> Type != cmsSigCLutElemType) {
+        cmsSignalError(CLUT->ContextID, cmsERROR_INTERNAL, "(internal) Attempt to PatchLUT on non-lut MPE");
+        return FALSE;
+    }
+
+    px = ((cmsFloat64Number) At[0] * (p16->Domain[0])) / 65535.0;
+    py = ((cmsFloat64Number) At[1] * (p16->Domain[1])) / 65535.0;
+    pz = ((cmsFloat64Number) At[2] * (p16->Domain[2])) / 65535.0;
+    pw = ((cmsFloat64Number) At[3] * (p16->Domain[3])) / 65535.0;
+
+    x0 = (int) floor(px);
+    y0 = (int) floor(py);
+    z0 = (int) floor(pz);
+    w0 = (int) floor(pw);
+
+    if (nChannelsIn == 4) {
+
+        if (((px - x0) != 0) ||
+            ((py - y0) != 0) ||
+            ((pz - z0) != 0) ||
+            ((pw - w0) != 0)) return FALSE; // Not on exact node
+
+        index = p16 -> opta[3] * x0 +
+            p16 -> opta[2] * y0 +
+            p16 -> opta[1] * z0 +
+            p16 -> opta[0] * w0;
+    }
+    else 
+        if (nChannelsIn == 3) {
+
+            if (((px - x0) != 0) ||
+                ((py - y0) != 0) ||
+                ((pz - z0) != 0)) return FALSE;  // Not on exact node
+
+            index = p16 -> opta[2] * x0 +
+                p16 -> opta[1] * y0 +
+                p16 -> opta[0] * z0;
+        }
+        else 
+            if (nChannelsIn == 1) {
+
+                if (((px - x0) != 0)) return FALSE; // Not on exact node
+
+                index = p16 -> opta[0] * x0;    
+            }
+            else {
+                cmsSignalError(CLUT->ContextID, cmsERROR_INTERNAL, "(internal) %d Channels are not supported on PatchLUT", nChannelsIn);
+                return FALSE;
+            }
+
+            for (i=0; i < nChannelsOut; i++)
+                Grid -> Tab.T[index + i] = Value[i];
+
+            return TRUE;
+}
+
+// Auxiliar, to see if two values are equal.
+static
+cmsBool WhitesAreEqual(int n, cmsUInt16Number White1[], cmsUInt16Number White2[] ) 
+{
+    int i;
+
+    for (i=0; i < n; i++) {
+        if (White1[i] != White2[i]) return FALSE;
+    }
+    return TRUE;
+}
+
+
+// Locate the node for the white point and fix it to pure white in order to avoid scum dot.
+static
+cmsBool FixWhiteMisalignment(cmsPipeline* Lut, cmsColorSpaceSignature EntryColorSpace, cmsColorSpaceSignature ExitColorSpace)
+{
+    cmsUInt16Number *WhitePointIn, *WhitePointOut;
+    cmsUInt16Number  WhiteIn[cmsMAXCHANNELS], WhiteOut[cmsMAXCHANNELS], ObtainedOut[cmsMAXCHANNELS];
+    cmsUInt32Number i, nOuts, nIns;
+    cmsStage *PreLin = NULL, *CLUT = NULL, *PostLin = NULL;
+    
+    if (!_cmsEndPointsBySpace(EntryColorSpace,
+        &WhitePointIn, NULL, &nIns)) return FALSE;
+
+    if (!_cmsEndPointsBySpace(ExitColorSpace,
+        &WhitePointOut, NULL, &nOuts)) return FALSE;
+
+    // It needs to be fixed?
+
+    cmsPipelineEval16(WhitePointIn, ObtainedOut, Lut);
+
+    if (WhitesAreEqual(nOuts, WhitePointOut, ObtainedOut)) return TRUE; // whites already match 
+    
+    // Check if the LUT comes as Prelin, CLUT or Postlin. We allow all combinations
+    if (!cmsPipelineCheckAndRetreiveStages(Lut, 3, cmsSigCurveSetElemType, cmsSigCLutElemType, cmsSigCurveSetElemType, &PreLin, &CLUT, &PostLin))
+        if (!cmsPipelineCheckAndRetreiveStages(Lut, 2, cmsSigCurveSetElemType, cmsSigCLutElemType, &PreLin, &CLUT))
+            if (!cmsPipelineCheckAndRetreiveStages(Lut, 2, cmsSigCLutElemType, cmsSigCurveSetElemType, &CLUT, &PostLin))
+                if (!cmsPipelineCheckAndRetreiveStages(Lut, 1, cmsSigCLutElemType, &CLUT))
+                    return FALSE;
+
+    // We need to interpolate white points of both, pre and post curves
+    if (PreLin) {
+
+        cmsToneCurve** Curves = _cmsStageGetPtrToCurveSet(PreLin);
+
+        for (i=0; i < nIns; i++) {    
+            WhiteIn[i] = cmsEvalToneCurve16(Curves[i], WhitePointIn[i]);
+        }
+    }
+    else {
+        for (i=0; i < nIns; i++) 
+            WhiteIn[i] = WhitePointIn[i]; 
+    }
+
+    // If any post-linearization, we need to find how is represented white before the curve, do
+    // a reverse interpolation in this case.
+    if (PostLin) {
+        
+        cmsToneCurve** Curves = _cmsStageGetPtrToCurveSet(PostLin);
+        
+        for (i=0; i < nOuts; i++) {
+        
+            cmsToneCurve* InversePostLin = cmsReverseToneCurve(Curves[i]);
+            WhiteOut[i] = cmsEvalToneCurve16(InversePostLin, WhitePointOut[i]);
+            cmsFreeToneCurve(InversePostLin);
+        }
+    }
+    else {
+        for (i=0; i < nOuts; i++) 
+            WhiteOut[i] = WhitePointOut[i]; 
+    }
+
+    // Ok, proceed with patching. May fail and we don't care if it fails
+    PatchLUT(CLUT, WhiteIn, WhiteOut, nOuts, nIns);
+
+    return TRUE;
+}
+
+// -----------------------------------------------------------------------------------------------------------------------------------------------
+// This function creates simple LUT from complex ones. The generated LUT has an optional set of 
+// prelinearization curves, a CLUT of nGridPoints and optional postlinearization tables. 
+// These curves have to exist in the original LUT in order to be used in the simplified output. 
+// Caller may also use the flags to allow this feature.
+// LUTS with all curves will be simplified to a single curve. Parametric curves are lost.
+// This function should be used on 16-bits LUTS only, as floating point losses precision when simplified
+// -----------------------------------------------------------------------------------------------------------------------------------------------
+
+static
+cmsBool OptimizeByResampling(cmsPipeline** Lut, cmsUInt32Number Intent, cmsUInt32Number* InputFormat, cmsUInt32Number* OutputFormat, cmsUInt32Number* dwFlags)                          
+{
+    cmsPipeline* Src;
+    cmsPipeline* Dest;   
+    cmsStage* CLUT;      
+    cmsStage *KeepPreLin = NULL, *KeepPostLin = NULL;
+    int nGridPoints;    
+    cmsColorSpaceSignature ColorSpace, OutputColorSpace;
+    cmsStage *NewPreLin = NULL;
+    cmsStage *NewPostLin = NULL;
+    _cmsStageCLutData* DataCLUT;
+    cmsToneCurve** DataSetIn;
+    cmsToneCurve** DataSetOut;
+    Prelin16Data* p16;
+
+
+    // This is a loosy optimization! does not apply in floating-point cases
+    if (_cmsFormatterIsFloat(*InputFormat) || _cmsFormatterIsFloat(*OutputFormat)) return FALSE;
+
+    ColorSpace       = _cmsICCcolorSpace(T_COLORSPACE(*InputFormat));
+    OutputColorSpace = _cmsICCcolorSpace(T_COLORSPACE(*OutputFormat));
+    nGridPoints      = _cmsReasonableGridpointsByColorspace(ColorSpace, *dwFlags);
+
+    // For empty LUTs, 2 points are enough
+    if (cmsPipelineStageCount(*Lut) == 0)
+        nGridPoints = 2;
+
+    Src = *Lut;
+
+    // Allocate an empty LUT    
+    Dest =  cmsPipelineAlloc(Src ->ContextID, Src ->InputChannels, Src ->OutputChannels);
+    if (!Dest) return FALSE;
+
+    // Prelinearization tables are kept unless indicated by flags
+    if (*dwFlags & cmsFLAGS_CLUT_PRE_LINEARIZATION) {
+
+        // Get a pointer to the prelinearization element
+        cmsStage* PreLin = cmsPipelineGetPtrToFirstStage(Src);
+
+        // Check if suitable
+        if (PreLin ->Type == cmsSigCurveSetElemType) {
+
+            // Maybe this is a linear tram, so we can avoid the whole stuff
+            if (!AllCurvesAreLinear(PreLin)) {
+
+                // All seems ok, proceed.    
+                NewPreLin = cmsStageDup(PreLin);
+                cmsPipelineInsertStage(Dest, cmsAT_BEGIN, NewPreLin);
+
+                // Remove prelinearization. Since we have duplicated the curve
+                // in destination LUT, the sampling shoud be applied after this stage.
+                cmsPipelineUnlinkStage(Src, cmsAT_BEGIN, &KeepPreLin);
+            }
+        }
+    }
+
+    // Allocate the CLUT
+    CLUT = cmsStageAllocCLut16bit(Src ->ContextID, nGridPoints, Src ->InputChannels, Src->OutputChannels, NULL);
+    if (CLUT == NULL) return FALSE;
+
+    // Add the CLUT to the destination LUT
+    cmsPipelineInsertStage(Dest, cmsAT_END, CLUT);
+
+    // Postlinearization tables are kept unless indicated by flags
+    if (*dwFlags & cmsFLAGS_CLUT_POST_LINEARIZATION) {
+
+        // Get a pointer to the postlinearization if present
+        cmsStage* PostLin = cmsPipelineGetPtrToLastStage(Src);
+
+        // Check if suitable
+        if (cmsStageType(PostLin) == cmsSigCurveSetElemType) {
+
+            // Maybe this is a linear tram, so we can avoid the whole stuff
+            if (!AllCurvesAreLinear(PostLin)) {
+
+                // All seems ok, proceed.       
+                NewPostLin = cmsStageDup(PostLin);
+                cmsPipelineInsertStage(Dest, cmsAT_END, NewPostLin);
+
+                // In destination LUT, the sampling shoud be applied after this stage.
+                cmsPipelineUnlinkStage(Src, cmsAT_END, &KeepPostLin);
+            }
+        }
+    }
+
+    // Now its time to do the sampling. We have to ignore pre/post linearization 
+    // The source LUT whithout pre/post curves is passed as parameter.
+    if (!cmsStageSampleCLut16bit(CLUT, XFormSampler16, (void*) Src, 0)) {
+
+        // Ops, something went wrong, Restore stages
+        if (KeepPreLin != NULL)  cmsPipelineInsertStage(Src, cmsAT_BEGIN, KeepPreLin);
+        if (KeepPostLin != NULL) cmsPipelineInsertStage(Src, cmsAT_END,   KeepPostLin);
+        cmsPipelineFree(Dest);
+        return FALSE;
+    }      
+
+    // Done. 
+
+    if (KeepPreLin != NULL) cmsStageFree(KeepPreLin);
+    if (KeepPostLin != NULL) cmsStageFree(KeepPostLin);
+    cmsPipelineFree(Src);
+
+    DataCLUT = (_cmsStageCLutData*) CLUT ->Data;
+
+    if (NewPreLin == NULL) DataSetIn = NULL;
+    else DataSetIn = ((_cmsStageToneCurvesData*) NewPreLin ->Data) ->TheCurves;
+
+    if (NewPostLin == NULL) DataSetOut = NULL;
+    else  DataSetOut = ((_cmsStageToneCurvesData*) NewPostLin ->Data) ->TheCurves;
+
+
+    if (DataSetIn == NULL && DataSetOut == NULL) {
+
+        _cmsPipelineSetOptimizationParameters(Dest, (_cmsOPTeval16Fn) DataCLUT->Params->Interpolation.Lerp16, DataCLUT->Params, NULL, NULL);
+    }
+    else {
+
+        p16 = PrelinOpt16alloc(Dest ->ContextID, 
+                               DataCLUT ->Params, 
+                               Dest ->InputChannels,
+                               DataSetIn,
+                               Dest ->OutputChannels,
+                               DataSetOut);
+
+
+        _cmsPipelineSetOptimizationParameters(Dest, PrelinEval16, (void*) p16, PrelinOpt16free, Prelin16dup);
+    }
+
+
+    // Don't fix white on absolute colorimetric
+    if (Intent == INTENT_ABSOLUTE_COLORIMETRIC)
+        *dwFlags |= cmsFLAGS_NOWHITEONWHITEFIXUP;
+
+    if (!(*dwFlags & cmsFLAGS_NOWHITEONWHITEFIXUP)) {
+
+        FixWhiteMisalignment(Dest, ColorSpace, OutputColorSpace);
+    }
+
+    *Lut = Dest;
+    return TRUE;
+
+    cmsUNUSED_PARAMETER(Intent);    
+}
+
+
+// -----------------------------------------------------------------------------------------------------------------------------------------------
+// Fixes the gamma balancing of transform. This is described in my paper "Prelinearization Stages on 
+// Color-Management Application-Specific Integrated Circuits (ASICs)" presented at NIP24. It only works 
+// for RGB transforms. See the paper for more details
+// -----------------------------------------------------------------------------------------------------------------------------------------------
+
+
+// Normalize endpoints by slope limiting max and min. This assures endpoints as well.
+// Descending curves are handled as well.
+static
+void SlopeLimiting(cmsToneCurve* g)
+{
+    int BeginVal, EndVal;
+    int AtBegin = (int) floor((cmsFloat64Number) g ->nEntries * 0.02 + 0.5);   // Cutoff at 2%
+    int AtEnd   = g ->nEntries - AtBegin - 1;                                  // And 98%
+    cmsFloat64Number Val, Slope, beta;
+    int i;
+
+    if (cmsIsToneCurveDescending(g)) {
+        BeginVal = 0xffff; EndVal = 0;
+    }
+    else {
+        BeginVal = 0; EndVal = 0xffff;
+    }
+
+    // Compute slope and offset for begin of curve
+    Val   = g ->Table16[AtBegin];
+    Slope = (Val - BeginVal) / AtBegin;
+    beta  = Val - Slope * AtBegin;
+
+    for (i=0; i < AtBegin; i++)
+        g ->Table16[i] = _cmsQuickSaturateWord(i * Slope + beta);
+
+    // Compute slope and offset for the end
+    Val   = g ->Table16[AtEnd];
+    Slope = (EndVal - Val) / AtBegin;   // AtBegin holds the X interval, which is same in both cases
+    beta  = Val - Slope * AtEnd;
+
+    for (i = AtEnd; i < (int) g ->nEntries; i++) 
+        g ->Table16[i] = _cmsQuickSaturateWord(i * Slope + beta);
+}
+
+
+// Precomputes tables for 8-bit on input devicelink. 
+static
+Prelin8Data* PrelinOpt8alloc(cmsContext ContextID, const cmsInterpParams* p, cmsToneCurve* G[3])
+{
+    int i;
+    cmsUInt16Number Input[3];
+    cmsS15Fixed16Number v1, v2, v3;
+    Prelin8Data* p8;
+
+    p8 = _cmsMallocZero(ContextID, sizeof(Prelin8Data));
+    if (p8 == NULL) return NULL;
+    
+    // Since this only works for 8 bit input, values comes always as x * 257, 
+    // we can safely take msb byte (x << 8 + x)
+
+    for (i=0; i < 256; i++) {
+
+        if (G != NULL) {
+
+            // Get 16-bit representation
+            Input[0] = cmsEvalToneCurve16(G[0], FROM_8_TO_16(i));
+            Input[1] = cmsEvalToneCurve16(G[1], FROM_8_TO_16(i));
+            Input[2] = cmsEvalToneCurve16(G[2], FROM_8_TO_16(i));
+        }
+        else {
+            Input[0] = FROM_8_TO_16(i);
+            Input[1] = FROM_8_TO_16(i);
+            Input[2] = FROM_8_TO_16(i);
+        }
+
+
+        // Move to 0..1.0 in fixed domain
+        v1 = _cmsToFixedDomain(Input[0] * p -> Domain[0]);
+        v2 = _cmsToFixedDomain(Input[1] * p -> Domain[1]);
+        v3 = _cmsToFixedDomain(Input[2] * p -> Domain[2]);
+
+        // Store the precalculated table of nodes
+        p8 ->X0[i] = (p->opta[2] * FIXED_TO_INT(v1));
+        p8 ->Y0[i] = (p->opta[1] * FIXED_TO_INT(v2));
+        p8 ->Z0[i] = (p->opta[0] * FIXED_TO_INT(v3));
+
+        // Store the precalculated table of offsets
+        p8 ->rx[i] = (cmsUInt16Number) FIXED_REST_TO_INT(v1);
+        p8 ->ry[i] = (cmsUInt16Number) FIXED_REST_TO_INT(v2);
+        p8 ->rz[i] = (cmsUInt16Number) FIXED_REST_TO_INT(v3);
+    }
+
+    p8 ->ContextID = ContextID;
+    p8 ->p = p;
+
+    return p8;
+}
+
+static
+void Prelin8free(cmsContext ContextID, void* ptr)
+{   
+    _cmsFree(ContextID, ptr);
+}
+
+static
+void* Prelin8dup(cmsContext ContextID, const void* ptr)
+{   
+    return _cmsDupMem(ContextID, ptr, sizeof(Prelin8Data));
+}
+
+
+
+// A optimized interpolation for 8-bit input.
+#define DENS(i,j,k) (LutTable[(i)+(j)+(k)+OutChan])
+static
+void PrelinEval8(register const cmsUInt16Number Input[],
+                  register cmsUInt16Number Output[],
+                  register const void* D)
+{
+    
+    cmsUInt8Number         r, g, b;
+    cmsS15Fixed16Number    rx, ry, rz;            
+    cmsS15Fixed16Number    c0, c1, c2, c3, Rest;       
+    int        OutChan;
+    register   cmsS15Fixed16Number    X0, X1, Y0, Y1, Z0, Z1;
+    Prelin8Data* p8 = (Prelin8Data*) D;
+    register const cmsInterpParams* p = p8 ->p;
+    int                    TotalOut = p -> nOutputs;
+    const cmsUInt16Number* LutTable = p -> Table;
+    
+    r = Input[0] >> 8;
+    g = Input[1] >> 8;
+    b = Input[2] >> 8;
+
+    X0 = X1 = p8->X0[r];
+    Y0 = Y1 = p8->Y0[g];
+    Z0 = Z1 = p8->Z0[b];
+
+    rx = p8 ->rx[r];
+    ry = p8 ->ry[g];
+    rz = p8 ->rz[b];
+
+    X1 = X0 + ((rx == 0) ? 0 : p ->opta[2]);
+    Y1 = Y0 + ((ry == 0) ? 0 : p ->opta[1]);
+    Z1 = Z0 + ((rz == 0) ? 0 : p ->opta[0]);
+
+  
+    // These are the 6 Tetrahedral
+    for (OutChan=0; OutChan < TotalOut; OutChan++) {
+
+        c0 = DENS(X0, Y0, Z0);
+
+        if (rx >= ry && ry >= rz)
+        {
+            c1 = DENS(X1, Y0, Z0) - c0;
+            c2 = DENS(X1, Y1, Z0) - DENS(X1, Y0, Z0);
+            c3 = DENS(X1, Y1, Z1) - DENS(X1, Y1, Z0);                
+        }
+        else
+            if (rx >= rz && rz >= ry)
+            {            
+                c1 = DENS(X1, Y0, Z0) - c0;
+                c2 = DENS(X1, Y1, Z1) - DENS(X1, Y0, Z1);
+                c3 = DENS(X1, Y0, Z1) - DENS(X1, Y0, Z0);
+            }
+            else
+                if (rz >= rx && rx >= ry)
+                {
+                    c1 = DENS(X1, Y0, Z1) - DENS(X0, Y0, Z1);
+                    c2 = DENS(X1, Y1, Z1) - DENS(X1, Y0, Z1);
+                    c3 = DENS(X0, Y0, Z1) - c0;                            
+                }
+                else
+                    if (ry >= rx && rx >= rz)
+                    {
+                        c1 = DENS(X1, Y1, Z0) - DENS(X0, Y1, Z0);
+                        c2 = DENS(X0, Y1, Z0) - c0;
+                        c3 = DENS(X1, Y1, Z1) - DENS(X1, Y1, Z0);          
+                    }
+                    else
+                        if (ry >= rz && rz >= rx)
+                        {
+                            c1 = DENS(X1, Y1, Z1) - DENS(X0, Y1, Z1);
+                            c2 = DENS(X0, Y1, Z0) - c0;
+                            c3 = DENS(X0, Y1, Z1) - DENS(X0, Y1, Z0);                
+                        }
+                        else
+                            if (rz >= ry && ry >= rx)
+                            {             
+                                c1 = DENS(X1, Y1, Z1) - DENS(X0, Y1, Z1);
+                                c2 = DENS(X0, Y1, Z1) - DENS(X0, Y0, Z1);
+                                c3 = DENS(X0, Y0, Z1) - c0;              
+                            }
+                            else  {
+                                c1 = c2 = c3 = 0;              
+                            }
+
+
+                            Rest = c1 * rx + c2 * ry + c3 * rz;
+
+                            Output[OutChan] = (cmsUInt16Number)c0 + ROUND_FIXED_TO_INT(_cmsToFixedDomain(Rest));
+                            
+    }
+}
+
+#undef DENS
+
+// --------------------------------------------------------------------------------------------------------------
+// We need xput over here
+
+static
+cmsBool OptimizeByComputingLinearization(cmsPipeline** Lut, cmsUInt32Number Intent, cmsUInt32Number* InputFormat, cmsUInt32Number* OutputFormat, cmsUInt32Number* dwFlags)      
+{
+    cmsPipeline* OriginalLut;
+    int nGridPoints;
+    cmsToneCurve *Trans[cmsMAXCHANNELS], *TransReverse[cmsMAXCHANNELS];
+    cmsUInt32Number t, i;  
+    cmsFloat32Number v, In[cmsMAXCHANNELS], Out[cmsMAXCHANNELS];
+    cmsBool lIsSuitable, lIsLinear;
+    cmsPipeline* OptimizedLUT = NULL, *LutPlusCurves = NULL;    
+    cmsStage* OptimizedCLUTmpe;
+    cmsColorSpaceSignature ColorSpace, OutputColorSpace;
+    cmsStage* OptimizedPrelinMpe;
+    cmsToneCurve**   OptimizedPrelinCurves;
+    _cmsStageCLutData*     OptimizedPrelinCLUT;
+
+
+    // This is a loosy optimization! does not apply in floating-point cases
+    if (_cmsFormatterIsFloat(*InputFormat) || _cmsFormatterIsFloat(*OutputFormat)) return FALSE;
+
+    // Only on RGB
+    if (T_COLORSPACE(*InputFormat)  != PT_RGB) return FALSE;
+    if (T_COLORSPACE(*OutputFormat) != PT_RGB) return FALSE;
+
+
+    // On 16 bits, user has to specify the feature
+    if (!_cmsFormatterIs8bit(*InputFormat)) {
+        if (!(*dwFlags & cmsFLAGS_CLUT_PRE_LINEARIZATION)) return FALSE;
+    }
+
+    OriginalLut = *Lut;
+    ColorSpace       = _cmsICCcolorSpace(T_COLORSPACE(*InputFormat));
+    OutputColorSpace = _cmsICCcolorSpace(T_COLORSPACE(*OutputFormat));
+    nGridPoints      = _cmsReasonableGridpointsByColorspace(ColorSpace, *dwFlags);
+
+    // Empty gamma containers
+    memset(Trans, 0, sizeof(Trans));
+    memset(TransReverse, 0, sizeof(TransReverse));
+
+    for (t = 0; t < OriginalLut ->InputChannels; t++) {
+        Trans[t] = cmsBuildTabulatedToneCurve16(OriginalLut ->ContextID, PRELINEARIZATION_POINTS, NULL);
+        if (Trans[t] == NULL) goto Error;
+    }
+
+    // Populate the curves
+    for (i=0; i < PRELINEARIZATION_POINTS; i++) {
+
+        v = (cmsFloat32Number) ((cmsFloat64Number) i / (PRELINEARIZATION_POINTS - 1));
+
+        // Feed input with a gray ramp
+        for (t=0; t < OriginalLut ->InputChannels; t++)
+            In[t] = v;
+
+        // Evaluate the gray value
+        cmsPipelineEvalFloat(In, Out, OriginalLut);
+
+        // Store result in curve
+        for (t=0; t < OriginalLut ->InputChannels; t++)
+            Trans[t] ->Table16[i] = _cmsQuickSaturateWord(Out[t] * 65535.0);
+    }
+
+    // Slope-limit the obtained curves
+    for (t = 0; t < OriginalLut ->InputChannels; t++) 
+        SlopeLimiting(Trans[t]);
+
+    // Check for validity
+    lIsSuitable = TRUE;
+    lIsLinear   = TRUE;
+    for (t=0; (lIsSuitable && (t < OriginalLut ->InputChannels)); t++) {
+
+        // Exclude if already linear
+        if (!cmsIsToneCurveLinear(Trans[t]))
+            lIsLinear = FALSE;
+
+        // Exclude if non-monotonic
+        if (!cmsIsToneCurveMonotonic(Trans[t]))
+            lIsSuitable = FALSE;                             
+    }
+
+    // If it is not suitable, just quit
+    if (!lIsSuitable) goto Error;
+
+    // Invert curves if possible
+    for (t = 0; t < OriginalLut ->InputChannels; t++) {
+        TransReverse[t] = cmsReverseToneCurveEx(PRELINEARIZATION_POINTS, Trans[t]);
+        if (TransReverse[t] == NULL) goto Error;
+    }
+
+    // Now inset the reversed curves at the begin of transform
+    LutPlusCurves = cmsPipelineDup(OriginalLut);
+    if (LutPlusCurves == NULL) goto Error;
+
+    cmsPipelineInsertStage(LutPlusCurves, cmsAT_BEGIN, cmsStageAllocToneCurves(OriginalLut ->ContextID, OriginalLut ->InputChannels, TransReverse));
+
+    // Create the result LUT
+    OptimizedLUT = cmsPipelineAlloc(OriginalLut ->ContextID, OriginalLut ->InputChannels, OriginalLut ->OutputChannels);
+    if (OptimizedLUT == NULL) goto Error;
+
+    OptimizedPrelinMpe = cmsStageAllocToneCurves(OriginalLut ->ContextID, OriginalLut ->InputChannels, Trans);
+
+    // Create and insert the curves at the beginning    
+    cmsPipelineInsertStage(OptimizedLUT, cmsAT_BEGIN, OptimizedPrelinMpe);
+
+    // Allocate the CLUT for result
+    OptimizedCLUTmpe = cmsStageAllocCLut16bit(OriginalLut ->ContextID, nGridPoints, OriginalLut ->InputChannels, OriginalLut ->OutputChannels, NULL);
+
+    // Add the CLUT to the destination LUT
+    cmsPipelineInsertStage(OptimizedLUT, cmsAT_END, OptimizedCLUTmpe);
+
+    // Resample the LUT
+    if (!cmsStageSampleCLut16bit(OptimizedCLUTmpe, XFormSampler16, (void*) LutPlusCurves, 0)) goto Error;
+
+    // Free resources
+    for (t = 0; t < OriginalLut ->InputChannels; t++) {
+
+        if (Trans[t]) cmsFreeToneCurve(Trans[t]);
+        if (TransReverse[t]) cmsFreeToneCurve(TransReverse[t]);
+    }
+
+    cmsPipelineFree(LutPlusCurves);
+
+
+    OptimizedPrelinCurves = _cmsStageGetPtrToCurveSet(OptimizedPrelinMpe);
+    OptimizedPrelinCLUT   = (_cmsStageCLutData*) OptimizedCLUTmpe ->Data;
+
+    // Set the evaluator if 8-bit
+    if (_cmsFormatterIs8bit(*InputFormat)) {
+
+        Prelin8Data* p8 = PrelinOpt8alloc(OptimizedLUT ->ContextID, 
+                                                OptimizedPrelinCLUT ->Params, 
+                                                OptimizedPrelinCurves);
+        if (p8 == NULL) return FALSE;
+
+        _cmsPipelineSetOptimizationParameters(OptimizedLUT, PrelinEval8, (void*) p8, Prelin8free, Prelin8dup);
+
+    } 
+    else
+    {
+        Prelin16Data* p16 = PrelinOpt16alloc(OptimizedLUT ->ContextID, 
+            OptimizedPrelinCLUT ->Params, 
+            3, OptimizedPrelinCurves, 3, NULL);
+        if (p16 == NULL) return FALSE;
+
+        _cmsPipelineSetOptimizationParameters(OptimizedLUT, PrelinEval16, (void*) p16, PrelinOpt16free, Prelin16dup);
+
+    }
+
+    // Don't fix white on absolute colorimetric
+    if (Intent == INTENT_ABSOLUTE_COLORIMETRIC)
+        *dwFlags |= cmsFLAGS_NOWHITEONWHITEFIXUP;
+
+    if (!(*dwFlags & cmsFLAGS_NOWHITEONWHITEFIXUP)) {
+
+        if (!FixWhiteMisalignment(OptimizedLUT, ColorSpace, OutputColorSpace)) {
+
+            return FALSE;
+        }
+    }
+
+    // And return the obtained LUT
+
+    cmsPipelineFree(OriginalLut);
+    *Lut = OptimizedLUT;
+    return TRUE;
+
+Error:
+
+    for (t = 0; t < OriginalLut ->InputChannels; t++) {
+
+        if (Trans[t]) cmsFreeToneCurve(Trans[t]);
+        if (TransReverse[t]) cmsFreeToneCurve(TransReverse[t]);
+    }
+
+    if (LutPlusCurves != NULL) cmsPipelineFree(LutPlusCurves);   
+    if (OptimizedLUT != NULL) cmsPipelineFree(OptimizedLUT);
+
+    return FALSE;    
+
+    cmsUNUSED_PARAMETER(Intent);
+}
+
+
+// Curves optimizer ------------------------------------------------------------------------------------------------------------------
+
+static
+void CurvesFree(cmsContext ContextID, void* ptr)
+{   
+     Curves16Data* Data = (Curves16Data*) ptr;
+     int i;
+
+     for (i=0; i < Data -> nCurves; i++) {
+     
+         _cmsFree(ContextID, Data ->Curves[i]);
+     }
+
+     _cmsFree(ContextID, Data ->Curves);
+     _cmsFree(ContextID, ptr);
+}
+
+static
+void* CurvesDup(cmsContext ContextID, const void* ptr)
+{   
+    Curves16Data* Data = _cmsDupMem(ContextID, ptr, sizeof(Curves16Data));
+    int i;
+
+    if (Data == NULL) return NULL;
+
+    Data ->Curves = _cmsDupMem(ContextID, Data ->Curves, Data ->nCurves * sizeof(cmsUInt16Number*));
+
+    for (i=0; i < Data -> nCurves; i++) {
+        Data ->Curves[i] = _cmsDupMem(ContextID, Data ->Curves[i], Data -> nElements * sizeof(cmsUInt16Number));
+    }
+
+    return (void*) Data;
+}
+
+// Precomputes tables for 8-bit on input devicelink. 
+static
+Curves16Data* CurvesAlloc(cmsContext ContextID, int nCurves, int nElements, cmsToneCurve** G)
+{
+    int i, j;
+    Curves16Data* c16;
+
+    c16 = _cmsMallocZero(ContextID, sizeof(Curves16Data));
+    if (c16 == NULL) return NULL;
+
+    c16 ->nCurves = nCurves;
+    c16 ->nElements = nElements;
+
+    c16 ->Curves = _cmsCalloc(ContextID, nCurves, sizeof(cmsUInt16Number*));
+    if (c16 ->Curves == NULL) return NULL;
+
+    for (i=0; i < nCurves; i++) {
+
+        c16->Curves[i] = _cmsCalloc(ContextID, nElements, sizeof(cmsUInt16Number));
+
+        if (nElements == 256) {
+
+            for (j=0; j < nElements; j++) {
+
+                c16 ->Curves[i][j] = cmsEvalToneCurve16(G[i], FROM_8_TO_16(j));             
+            }
+        }
+        else {
+
+            for (j=0; j < nElements; j++) {
+                c16 ->Curves[i][j] = cmsEvalToneCurve16(G[i], (cmsUInt16Number) j);             
+            }
+        }
+    }
+
+    return c16;
+}
+
+static
+void FastEvaluateCurves8(register const cmsUInt16Number In[], 
+                          register cmsUInt16Number Out[], 
+                          register const void* D)
+{   
+    Curves16Data* Data = (Curves16Data*) D;
+    cmsUInt8Number x;
+    int i;
+    
+    for (i=0; i < Data ->nCurves; i++) {
+
+         x = (In[i] >> 8);
+         Out[i] = Data -> Curves[i][x];
+    }
+}
+
+    
+static
+void FastEvaluateCurves16(register const cmsUInt16Number In[], 
+                          register cmsUInt16Number Out[], 
+                          register const void* D)
+{   
+    Curves16Data* Data = (Curves16Data*) D;
+    int i;
+    
+    for (i=0; i < Data ->nCurves; i++) {
+         Out[i] = Data -> Curves[i][In[i]];
+    }
+}
+
+
+static
+void FastIdentity16(register const cmsUInt16Number In[], 
+                    register cmsUInt16Number Out[], 
+                    register const void* D)
+{
+    cmsPipeline* Lut = (cmsPipeline*) D;
+    cmsUInt32Number i;
+
+    for (i=0; i < Lut ->InputChannels; i++) {
+         Out[i] = In[i];   
+    }
+}
+
+
+// If the target LUT holds only curves, the optimization procedure is to join all those
+// curves together. That only works on curves and does not work on matrices.
+static
+cmsBool OptimizeByJoiningCurves(cmsPipeline** Lut, cmsUInt32Number Intent, cmsUInt32Number* InputFormat, cmsUInt32Number* OutputFormat, cmsUInt32Number* dwFlags)
+{
+    cmsToneCurve** GammaTables = NULL; 
+    cmsFloat32Number InFloat[cmsMAXCHANNELS], OutFloat[cmsMAXCHANNELS];
+    cmsUInt32Number i, j;
+    cmsPipeline* Src = *Lut;
+    cmsPipeline* Dest = NULL;
+    cmsStage* mpe;
+    cmsStage* ObtainedCurves = NULL;
+
+
+    // This is a loosy optimization! does not apply in floating-point cases
+    if (_cmsFormatterIsFloat(*InputFormat) || _cmsFormatterIsFloat(*OutputFormat)) return FALSE;
+
+    //  Only curves in this LUT?
+    for (mpe = cmsPipelineGetPtrToFirstStage(Src);
+         mpe != NULL;
+         mpe = cmsStageNext(mpe)) {
+            if (cmsStageType(mpe) != cmsSigCurveSetElemType) return FALSE;
+    }
+
+    // Allocate an empty LUT 
+    Dest =  cmsPipelineAlloc(Src ->ContextID, Src ->InputChannels, Src ->OutputChannels);
+    if (Dest == NULL) return FALSE;
+
+    // Create target curves
+    GammaTables = (cmsToneCurve**) _cmsCalloc(Src ->ContextID, Src ->InputChannels, sizeof(cmsToneCurve*));
+    if (GammaTables == NULL) goto Error;
+
+    for (i=0; i < Src ->InputChannels; i++) {
+        GammaTables[i] = cmsBuildTabulatedToneCurve16(Src ->ContextID, PRELINEARIZATION_POINTS, NULL);
+        if (GammaTables[i] == NULL) goto Error;
+    }
+
+    // Compute 16 bit result by using floating point
+    for (i=0; i < PRELINEARIZATION_POINTS; i++) {
+
+        for (j=0; j < Src ->InputChannels; j++) 
+            InFloat[j] = (cmsFloat32Number) ((cmsFloat64Number) i / (PRELINEARIZATION_POINTS - 1));
+
+        cmsPipelineEvalFloat(InFloat, OutFloat, Src);
+
+        for (j=0; j < Src ->InputChannels; j++)
+            GammaTables[j] -> Table16[i] = _cmsQuickSaturateWord(OutFloat[j] * 65535.0);
+    }
+
+    ObtainedCurves = cmsStageAllocToneCurves(Src ->ContextID, Src ->InputChannels, GammaTables);
+    if (ObtainedCurves == NULL) goto Error;
+
+    for (i=0; i < Src ->InputChannels; i++) {
+        cmsFreeToneCurve(GammaTables[i]);
+        GammaTables[i] = NULL;
+    }
+
+    if (GammaTables != NULL) _cmsFree(Src ->ContextID, GammaTables);
+
+    // Maybe the curves are linear at the end
+    if (!AllCurvesAreLinear(ObtainedCurves)) {
+
+        cmsPipelineInsertStage(Dest, cmsAT_BEGIN, ObtainedCurves); 
+
+        // If the curves are to be applied in 8 bits, we can save memory
+        if (_cmsFormatterIs8bit(*InputFormat)) {
+
+            _cmsStageToneCurvesData* Data = (_cmsStageToneCurvesData*) ObtainedCurves ->Data;
+             Curves16Data* c16 = CurvesAlloc(Dest ->ContextID, Data ->nCurves, 256, Data ->TheCurves);
+
+             *dwFlags |= cmsFLAGS_NOCACHE;
+            _cmsPipelineSetOptimizationParameters(Dest, FastEvaluateCurves8, c16, CurvesFree, CurvesDup);
+
+        }
+        else {
+
+            _cmsStageToneCurvesData* Data = (_cmsStageToneCurvesData*) cmsStageData(ObtainedCurves);
+             Curves16Data* c16 = CurvesAlloc(Dest ->ContextID, Data ->nCurves, 65536, Data ->TheCurves);
+
+             *dwFlags |= cmsFLAGS_NOCACHE;
+            _cmsPipelineSetOptimizationParameters(Dest, FastEvaluateCurves16, c16, CurvesFree, CurvesDup);          
+        }
+    }
+    else {
+
+        // LUT optimizes to nothing. Set the identity LUT
+        cmsStageFree(ObtainedCurves);
+
+        cmsPipelineInsertStage(Dest, cmsAT_BEGIN, cmsStageAllocIdentity(Dest ->ContextID, Src ->InputChannels));
+
+        *dwFlags |= cmsFLAGS_NOCACHE;
+        _cmsPipelineSetOptimizationParameters(Dest, FastIdentity16, (void*) Dest, NULL, NULL);
+    }
+
+    // We are done.
+    cmsPipelineFree(Src);
+    *Lut = Dest;
+    return TRUE;
+
+Error:
+
+    if (ObtainedCurves != NULL) cmsStageFree(ObtainedCurves);
+    if (GammaTables != NULL) {
+        for (i=0; i < Src ->InputChannels; i++) {
+            if (GammaTables[i] != NULL) cmsFreeToneCurve(GammaTables[i]);
+        }
+
+        _cmsFree(Src ->ContextID, GammaTables);
+    }
+
+    if (Dest != NULL) cmsPipelineFree(Dest);
+    return FALSE;
+
+    cmsUNUSED_PARAMETER(Intent);
+    cmsUNUSED_PARAMETER(InputFormat);
+    cmsUNUSED_PARAMETER(OutputFormat);
+    cmsUNUSED_PARAMETER(dwFlags);
+}
+
+// -------------------------------------------------------------------------------------------------------------------------------------
+// LUT is Shaper - Matrix - Matrix - Shaper, which is very frequent when combining two matrix-shaper profiles
+
+
+static
+void  FreeMatShaper(cmsContext ContextID, void* Data)
+{
+    if (Data != NULL) _cmsFree(ContextID, Data);
+}
+
+static
+void* DupMatShaper(cmsContext ContextID, const void* Data)
+{
+    return _cmsDupMem(ContextID, Data, sizeof(MatShaper8Data));
+}
+
+
+// A fast matrix-shaper evaluator for 8 bits. This is a bit ticky since I'm using 1.14 signed fixed point 
+// to accomplish some performance. Actually it takes 256x3 16 bits tables and 16385 x 3 tables of 8 bits, 
+// in total about 50K, and the performance boost is huge!
+static
+void MatShaperEval16(register const cmsUInt16Number In[], 
+                     register cmsUInt16Number Out[], 
+                     register const void* D)
+{    
+    MatShaper8Data* p = (MatShaper8Data*) D;
+    cmsS1Fixed14Number l1, l2, l3, r, g, b;
+    cmsUInt32Number ri, gi, bi;
+
+    // In this case (and only in this case!) we can use this simplification since 
+    // In[] is assured to come from a 8 bit number. (a << 8 | a)
+    ri = In[0] & 0xFF;
+    gi = In[1] & 0xFF;
+    bi = In[2] & 0xFF;
+    
+    // Across first shaper, which also converts to 1.14 fixed point
+    r = p->Shaper1R[ri];
+    g = p->Shaper1G[gi];
+    b = p->Shaper1B[bi];
+        
+    // Evaluate the matrix in 1.14 fixed point
+    l1 =  (p->Mat[0][0] * r + p->Mat[0][1] * g + p->Mat[0][2] * b + p->Off[0] + 0x2000) >> 14;
+    l2 =  (p->Mat[1][0] * r + p->Mat[1][1] * g + p->Mat[1][2] * b + p->Off[1] + 0x2000) >> 14;
+    l3 =  (p->Mat[2][0] * r + p->Mat[2][1] * g + p->Mat[2][2] * b + p->Off[2] + 0x2000) >> 14;
+    
+    // Now we have to clip to 0..1.0 range 
+    ri = (l1 < 0) ? 0 : ((l1 > 16384) ? 16384 : l1);               
+    gi = (l2 < 0) ? 0 : ((l2 > 16384) ? 16384 : l2);               
+    bi = (l3 < 0) ? 0 : ((l3 > 16384) ? 16384 : l3);               
+         
+    // And across second shaper, 
+    Out[0] = p->Shaper2R[ri];
+    Out[1] = p->Shaper2G[gi];
+    Out[2] = p->Shaper2B[bi];
+   
+}
+
+// This table converts from 8 bits to 1.14 after applying the curve
+static
+void FillFirstShaper(cmsS1Fixed14Number* Table, cmsToneCurve* Curve)
+{
+    int i;
+    cmsFloat32Number R, y;
+
+    for (i=0; i < 256; i++) {
+        
+        R   = (cmsFloat32Number) (i / 255.0);
+        y   = cmsEvalToneCurveFloat(Curve, R);        
+
+        Table[i] = DOUBLE_TO_1FIXED14(y);
+    }
+}
+
+// This table converts form 1.14 (being 0x4000 the last entry) to 8 bits after applying the curve
+static
+void FillSecondShaper(cmsUInt16Number* Table, cmsToneCurve* Curve, cmsBool Is8BitsOutput)
+{
+    int i;
+    cmsFloat32Number R, Val;
+
+    for (i=0; i < 16385; i++) {
+
+        R   = (cmsFloat32Number) (i / 16384.0);
+        Val = cmsEvalToneCurveFloat(Curve, R);    // Val comes 0..1.0
+        
+        if (Is8BitsOutput) {
+
+            // If 8 bits output, we can optimize further by computing the / 257 part.
+            // first we compute the resulting byte and then we store the byte times
+            // 257. This quantization allows to round very quick by doing a >> 8, but
+            // since the low byte is always equal to msb, we can do a & 0xff and this works!
+            cmsUInt16Number w = _cmsQuickSaturateWord(Val * 65535.0 + 0.5);        
+            cmsUInt8Number  b = FROM_16_TO_8(w);
+
+            Table[i] = FROM_8_TO_16(b);
+        }
+        else Table[i]  = _cmsQuickSaturateWord(Val * 65535.0 + 0.5);        
+    }
+}
+
+// Compute the matrix-shaper structure
+static
+cmsBool SetMatShaper(cmsPipeline* Dest, cmsToneCurve* Curve1[3], cmsMAT3* Mat, cmsVEC3* Off, cmsToneCurve* Curve2[3], cmsUInt32Number* OutputFormat)
+{
+    MatShaper8Data* p;
+    int i, j;
+    cmsBool Is8Bits = _cmsFormatterIs8bit(*OutputFormat);
+
+    // Allocate a big chuck of memory to store precomputed tables
+    p = (MatShaper8Data*) _cmsMalloc(Dest ->ContextID, sizeof(MatShaper8Data));
+    if (p == NULL) return FALSE;
+
+    p -> ContextID = Dest -> ContextID;
+
+    // Precompute tables
+    FillFirstShaper(p ->Shaper1R, Curve1[0]);
+    FillFirstShaper(p ->Shaper1G, Curve1[1]);
+    FillFirstShaper(p ->Shaper1B, Curve1[2]);
+
+    FillSecondShaper(p ->Shaper2R, Curve2[0], Is8Bits);
+    FillSecondShaper(p ->Shaper2G, Curve2[1], Is8Bits);
+    FillSecondShaper(p ->Shaper2B, Curve2[2], Is8Bits);
+
+    // Convert matrix to nFixed14. Note that those values may take more than 16 bits as
+    for (i=0; i < 3; i++) {
+        for (j=0; j < 3; j++) {         
+            p ->Mat[i][j] = DOUBLE_TO_1FIXED14(Mat->v[i].n[j]);
+        }
+    }
+    
+    for (i=0; i < 3; i++) {
+
+        if (Off == NULL) {          
+            p ->Off[i] = 0;
+        }
+        else {      
+            p ->Off[i] = DOUBLE_TO_1FIXED14(Off->n[i]);
+        }
+    }
+
+    // Mark as optimized for faster formatter
+    if (Is8Bits)
+        *OutputFormat |= OPTIMIZED_SH(1);
+
+    // Fill function pointers    
+    _cmsPipelineSetOptimizationParameters(Dest, MatShaperEval16, (void*) p, FreeMatShaper, DupMatShaper);
+    return TRUE;
+}
+
+//  8 bits on input allows matrix-shaper boot up to 25 Mpixels per second on RGB. That's fast!
+// TODO: Allow a third matrix for abs. colorimetric
+static
+cmsBool OptimizeMatrixShaper(cmsPipeline** Lut, cmsUInt32Number Intent, cmsUInt32Number* InputFormat, cmsUInt32Number* OutputFormat, cmsUInt32Number* dwFlags)
+{
+    cmsStage* Curve1, *Curve2;
+    cmsStage* Matrix1, *Matrix2;
+    _cmsStageMatrixData* Data1;
+    _cmsStageMatrixData* Data2;
+    cmsMAT3 res;
+    cmsBool IdentityMat;
+    cmsPipeline* Dest, *Src;
+   
+    // Only works on RGB to RGB
+    if (T_CHANNELS(*InputFormat) != 3 || T_CHANNELS(*OutputFormat) != 3) return FALSE;
+
+    // Only works on 8 bit input
+    if (!_cmsFormatterIs8bit(*InputFormat)) return FALSE;
+
+    // Seems suitable, proceed
+    Src = *Lut;
+
+    // Check for shaper-matrix-matrix-shaper structure, that is what this optimizer stands for
+    if (!cmsPipelineCheckAndRetreiveStages(Src, 4, 
+        cmsSigCurveSetElemType, cmsSigMatrixElemType, cmsSigMatrixElemType, cmsSigCurveSetElemType, 
+        &Curve1, &Matrix1, &Matrix2, &Curve2)) return FALSE;
+
+    // Get both matrices
+    Data1 = (_cmsStageMatrixData*) cmsStageData(Matrix1);
+    Data2 = (_cmsStageMatrixData*) cmsStageData(Matrix2);
+
+    // Input offset should be zero
+    if (Data1 ->Offset != NULL) return FALSE;
+
+    // Multiply both matrices to get the result
+    _cmsMAT3per(&res, (cmsMAT3*) Data2 ->Double, (cmsMAT3*) Data1 ->Double);
+
+    // Now the result is in res + Data2 -> Offset. Maybe is a plain identity?
+    IdentityMat = FALSE;
+    if (_cmsMAT3isIdentity(&res) && Data2 ->Offset == NULL) {
+
+        // We can get rid of full matrix
+        IdentityMat = TRUE;
+    }
+
+      // Allocate an empty LUT 
+    Dest =  cmsPipelineAlloc(Src ->ContextID, Src ->InputChannels, Src ->OutputChannels);
+    if (!Dest) return FALSE;
+
+    // Assamble the new LUT
+    cmsPipelineInsertStage(Dest, cmsAT_BEGIN, cmsStageDup(Curve1));
+    if (!IdentityMat) 
+        cmsPipelineInsertStage(Dest, cmsAT_END, cmsStageAllocMatrix(Dest ->ContextID, 3, 3, (const cmsFloat64Number*) &res, Data2 ->Offset));
+    cmsPipelineInsertStage(Dest, cmsAT_END, cmsStageDup(Curve2));
+
+    // If identity on matrix, we can further optimize the curves, so call the join curves routine
+    if (IdentityMat) {
+
+        OptimizeByJoiningCurves(&Dest, Intent, InputFormat, OutputFormat, dwFlags);     
+    }
+    else {
+        _cmsStageToneCurvesData* mpeC1 = (_cmsStageToneCurvesData*) cmsStageData(Curve1);
+        _cmsStageToneCurvesData* mpeC2 = (_cmsStageToneCurvesData*) cmsStageData(Curve2);
+                
+        // In this particular optimization, caché does not help as it takes more time to deal with 
+        // the caché that with the pixel handling
+        *dwFlags |= cmsFLAGS_NOCACHE;
+
+        // Setup the optimizarion routines
+        SetMatShaper(Dest, mpeC1 ->TheCurves, &res, (cmsVEC3*) Data2 ->Offset, mpeC2->TheCurves, OutputFormat);
+    }
+
+    cmsPipelineFree(Src);
+    *Lut = Dest;
+    return TRUE;
+}
+
+
+// -------------------------------------------------------------------------------------------------------------------------------------
+// Optimization plug-ins
+
+// List of optimizations
+typedef struct _cmsOptimizationCollection_st {
+    
+    _cmsOPToptimizeFn  OptimizePtr;
+    
+    struct _cmsOptimizationCollection_st *Next;
+
+} _cmsOptimizationCollection;
+
+
+// The built-in list. We currently implement 4 types of optimizations. Joining of curves, matrix-shaper, linearization and resampling
+static _cmsOptimizationCollection DefaultOptimization[] = {
+
+    { OptimizeByJoiningCurves,            &DefaultOptimization[1] },
+    { OptimizeMatrixShaper,               &DefaultOptimization[2] },
+    { OptimizeByComputingLinearization,   &DefaultOptimization[3] },
+    { OptimizeByResampling,               NULL }
+};
+
+// The linked list head
+static _cmsOptimizationCollection* OptimizationCollection = DefaultOptimization;
+
+// Register new ways to optimize
+cmsBool  _cmsRegisterOptimizationPlugin(cmsPluginBase* Data)
+{
+    cmsPluginOptimization* Plugin = (cmsPluginOptimization*) Data;
+    _cmsOptimizationCollection* fl;
+    
+    if (Data == NULL) {
+
+        OptimizationCollection = DefaultOptimization; 
+        return TRUE;
+    }
+    
+    // Optimizer callback is required
+    if (Plugin ->OptimizePtr == NULL) return FALSE;
+
+    fl = (_cmsOptimizationCollection*) _cmsPluginMalloc(sizeof(_cmsOptimizationCollection));
+    if (fl == NULL) return FALSE;
+
+    // Copy the parameters
+    fl ->OptimizePtr = Plugin ->OptimizePtr;
+        
+    // Keep linked list
+    fl ->Next = OptimizationCollection;
+    OptimizationCollection = fl;
+
+    // All is ok
+    return TRUE;
+}
+
+// The entry point for LUT optimization
+cmsBool _cmsOptimizePipeline(cmsPipeline**    PtrLut,                                               
+                             int              Intent,
+                             cmsUInt32Number* InputFormat, 
+                             cmsUInt32Number* OutputFormat,
+                             cmsUInt32Number* dwFlags)
+{    
+    _cmsOptimizationCollection* Opts;
+    cmsBool AnySuccess = FALSE;
+   
+    // A CLUT is being asked, so force this specific optimization
+    if (*dwFlags & cmsFLAGS_FORCE_CLUT) {
+    
+        PreOptimize(*PtrLut);
+        return OptimizeByResampling(PtrLut, Intent, InputFormat, OutputFormat, dwFlags);
+    }
+
+    // Anything to optimize?
+    if ((*PtrLut) ->Elements == NULL) {
+        _cmsPipelineSetOptimizationParameters(*PtrLut, FastIdentity16, (void*) *PtrLut, NULL, NULL);
+        return TRUE;        
+    }
+
+    // Try to get rid of identities and trivial conversions.
+    AnySuccess = PreOptimize(*PtrLut);
+
+    // After removal do we end with an identity?
+    if ((*PtrLut) ->Elements == NULL) {
+        _cmsPipelineSetOptimizationParameters(*PtrLut, FastIdentity16, (void*) *PtrLut, NULL, NULL);
+        return TRUE;
+    }
+
+    // Do not optimize, keep all precision
+    if (*dwFlags & cmsFLAGS_NOOPTIMIZE)
+        return FALSE;
+    
+    // Try built-in optimizations and plug-in
+    for (Opts = OptimizationCollection;
+         Opts != NULL;
+         Opts = Opts ->Next) {
+            
+            // If one schema succeeded, we are done
+            if (Opts ->OptimizePtr(PtrLut, Intent, InputFormat, OutputFormat, dwFlags)) {
+                
+                return TRUE;    // Optimized!
+            }
+    }
+    
+    // Only simple optimizations succeeded
+    return AnySuccess;
+}
+
+
+
diff --git a/thirdparty/liblcms2/src/cmspack.c b/thirdparty/liblcms2/src/cmspack.c
new file mode 100644
index 00000000..196afa94
--- /dev/null
+++ b/thirdparty/liblcms2/src/cmspack.c
@@ -0,0 +1,2558 @@
+//---------------------------------------------------------------------------------
+//
+//  Little Color Management System
+//  Copyright (c) 1998-2010 Marti Maria Saguer
+//
+// Permission is hereby granted, free of charge, to any person obtaining 
+// a copy of this software and associated documentation files (the "Software"), 
+// to deal in the Software without restriction, including without limitation 
+// the rights to use, copy, modify, merge, publish, distribute, sublicense, 
+// and/or sell copies of the Software, and to permit persons to whom the Software 
+// is furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in 
+// all copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 
+// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO 
+// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE 
+// LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 
+// OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 
+// WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+//
+//---------------------------------------------------------------------------------
+//
+
+#include "lcms2_internal.h"
+
+// This module handles all formats supported by lcms. There are two flavors, 16 bits and 
+// floating point. Floating point is supported only in a subset, those formats holding
+// cmsFloat32Number (4 bytes per component) and double (marked as 0 bytes per component as special 
+// case)
+
+// ---------------------------------------------------------------------------
+
+
+// This macro return words stored as big endian
+#define CHANGE_ENDIAN(w)    (cmsUInt16Number) ((cmsUInt16Number) ((w)<<8)|((w)>>8))
+
+// These macros handles reversing (negative)
+#define REVERSE_FLAVOR_8(x)     ((cmsUInt8Number) (0xff-(x)))
+#define REVERSE_FLAVOR_16(x)    ((cmsUInt16Number)(0xffff-(x)))
+
+// * 0xffff / 0xff00 = (255 * 257) / (255 * 256) = 257 / 256
+cmsINLINE cmsUInt16Number FomLabV2ToLabV4(cmsUInt16Number x) 
+{
+    int a;
+
+    a = (x << 8 | x) >> 8;  // * 257 / 256
+    if ( a > 0xffff) return 0xffff;
+    return (cmsUInt16Number) a;
+}
+
+// * 0xf00 / 0xffff = * 256 / 257
+cmsINLINE cmsUInt16Number FomLabV4ToLabV2(cmsUInt16Number x) 
+{
+    return (cmsUInt16Number) (((x << 8) + 0x80) / 257);
+}
+
+
+typedef struct {
+    cmsUInt32Number Type;
+    cmsUInt32Number Mask;
+    cmsFormatter16 Frm;
+
+} cmsFormatters16;
+
+typedef struct {
+    cmsUInt32Number    Type;
+    cmsUInt32Number    Mask;
+    cmsFormatterFloat Frm;
+
+} cmsFormattersFloat;
+
+#define ANYSPACE        COLORSPACE_SH(31)
+#define ANYCHANNELS     CHANNELS_SH(15)
+#define ANYEXTRA        EXTRA_SH(7)
+#define ANYPLANAR       PLANAR_SH(1)
+#define ANYENDIAN       ENDIAN16_SH(1)
+#define ANYSWAP         DOSWAP_SH(1)
+#define ANYSWAPFIRST    SWAPFIRST_SH(1) 
+#define ANYFLAVOR       FLAVOR_SH(1)
+
+
+// Supress waning about info never being used
+
+#ifdef _MSC_VER
+#pragma warning(disable : 4100)
+#endif
+
+// Unpacking routines (16 bits) ---------------------------------------------------------------------------------------- 
+
+// Does almost everything but is slow
+static
+cmsUInt8Number* UnrollChunkyBytes(register _cmsTRANSFORM* info, 
+                                  register cmsUInt16Number wIn[], 
+                                  register cmsUInt8Number* accum,
+                                  register cmsUInt32Number Stride)
+{
+    int nChan      = T_CHANNELS(info -> InputFormat);
+    int DoSwap     = T_DOSWAP(info ->InputFormat);
+    int Reverse    = T_FLAVOR(info ->InputFormat);
+    int SwapFirst  = T_SWAPFIRST(info -> InputFormat);
+    int Extra      = T_EXTRA(info -> InputFormat);
+    int ExtraFirst = DoSwap && !SwapFirst;
+    cmsUInt16Number v;
+    int i;
+
+    if (ExtraFirst) {
+        accum += Extra;
+    }
+
+    for (i=0; i < nChan; i++) {
+        int index = DoSwap ? (nChan - i - 1) : i;
+
+        v = FROM_8_TO_16(*accum); 
+        v = Reverse ? REVERSE_FLAVOR_16(v) : v;
+        wIn[index] = v;
+        accum++;              
+    }
+
+    if (!ExtraFirst) {
+        accum += Extra;
+    }
+
+    if (Extra == 0 && SwapFirst) {
+        cmsUInt16Number tmp = wIn[0];
+
+        memmove(&wIn[0], &wIn[1], (nChan-1) * sizeof(cmsUInt16Number));
+        wIn[nChan-1] = tmp;
+    }
+
+    return accum;
+}
+
+// Extra channels are just ignored because come in the next planes
+static
+cmsUInt8Number* UnrollPlanarBytes(register _cmsTRANSFORM* info, 
+                                  register cmsUInt16Number wIn[], 
+                                  register cmsUInt8Number* accum,
+                                  register cmsUInt32Number Stride)
+{
+    int nChan = T_CHANNELS(info -> InputFormat);
+    int DoSwap= T_DOSWAP(info ->InputFormat);
+    int Reverse= T_FLAVOR(info ->InputFormat);
+    int i;
+    cmsUInt8Number* Init = accum;
+
+    if (DoSwap) {
+        accum += T_EXTRA(info -> InputFormat) * Stride;
+    }
+
+    for (i=0; i < nChan; i++) {
+
+        int index = DoSwap ? (nChan - i - 1) : i;
+        cmsUInt16Number v = FROM_8_TO_16(*accum); 
+
+        wIn[index] = Reverse ? REVERSE_FLAVOR_16(v) : v;
+        accum += Stride;
+    }
+
+    return (Init + 1);
+}
+
+// Special cases, provided for performance
+static
+cmsUInt8Number* Unroll4Bytes(register _cmsTRANSFORM* info, 
+                             register cmsUInt16Number wIn[], 
+                             register cmsUInt8Number* accum,
+                             register cmsUInt32Number Stride)
+{
+    wIn[0] = FROM_8_TO_16(*accum); accum++; // C
+    wIn[1] = FROM_8_TO_16(*accum); accum++; // M
+    wIn[2] = FROM_8_TO_16(*accum); accum++; // Y
+    wIn[3] = FROM_8_TO_16(*accum); accum++; // K
+
+    return accum;
+}
+
+static
+cmsUInt8Number* Unroll4BytesReverse(register _cmsTRANSFORM* info, 
+                                    register cmsUInt16Number wIn[], 
+                                    register cmsUInt8Number* accum,
+                                    register cmsUInt32Number Stride)
+{
+    wIn[0] = FROM_8_TO_16(REVERSE_FLAVOR_8(*accum)); accum++; // C
+    wIn[1] = FROM_8_TO_16(REVERSE_FLAVOR_8(*accum)); accum++; // M
+    wIn[2] = FROM_8_TO_16(REVERSE_FLAVOR_8(*accum)); accum++; // Y
+    wIn[3] = FROM_8_TO_16(REVERSE_FLAVOR_8(*accum)); accum++; // K
+
+    return accum;
+}
+
+static
+cmsUInt8Number* Unroll4BytesSwapFirst(register _cmsTRANSFORM* info, 
+                                      register cmsUInt16Number wIn[], 
+                                      register cmsUInt8Number* accum,
+                                      register cmsUInt32Number Stride)
+{
+    wIn[3] = FROM_8_TO_16(*accum); accum++; // K
+    wIn[0] = FROM_8_TO_16(*accum); accum++; // C
+    wIn[1] = FROM_8_TO_16(*accum); accum++; // M
+    wIn[2] = FROM_8_TO_16(*accum); accum++; // Y
+
+    return accum;
+}
+
+// KYMC
+static
+cmsUInt8Number* Unroll4BytesSwap(register _cmsTRANSFORM* info, 
+                                 register cmsUInt16Number wIn[], 
+                                 register cmsUInt8Number* accum,
+                                 register cmsUInt32Number Stride)
+{
+    wIn[3] = FROM_8_TO_16(*accum); accum++;  // K
+    wIn[2] = FROM_8_TO_16(*accum); accum++;  // Y
+    wIn[1] = FROM_8_TO_16(*accum); accum++;  // M
+    wIn[0] = FROM_8_TO_16(*accum); accum++;  // C
+
+    return accum;
+}
+
+static
+cmsUInt8Number* Unroll4BytesSwapSwapFirst(register _cmsTRANSFORM* info, 
+                                          register cmsUInt16Number wIn[], 
+                                          register cmsUInt8Number* accum,
+                                          register cmsUInt32Number Stride)
+{
+    wIn[2] = FROM_8_TO_16(*accum); accum++;  // K
+    wIn[1] = FROM_8_TO_16(*accum); accum++;  // Y
+    wIn[0] = FROM_8_TO_16(*accum); accum++;  // M
+    wIn[3] = FROM_8_TO_16(*accum); accum++;  // C
+
+    return accum;
+}
+
+static
+cmsUInt8Number* Unroll3Bytes(register _cmsTRANSFORM* info, 
+                             register cmsUInt16Number wIn[], 
+                             register cmsUInt8Number* accum,
+                             register cmsUInt32Number Stride)
+{
+    wIn[0] = FROM_8_TO_16(*accum); accum++;     // R
+    wIn[1] = FROM_8_TO_16(*accum); accum++;     // G
+    wIn[2] = FROM_8_TO_16(*accum); accum++;     // B
+
+    return accum;
+}
+
+static
+cmsUInt8Number* Unroll3BytesSkip1Swap(register _cmsTRANSFORM* info, 
+                                      register cmsUInt16Number wIn[], 
+                                      register cmsUInt8Number* accum,
+                                      register cmsUInt32Number Stride)
+{
+    accum++; // A
+    wIn[2] = FROM_8_TO_16(*accum); accum++; // B
+    wIn[1] = FROM_8_TO_16(*accum); accum++; // G
+    wIn[0] = FROM_8_TO_16(*accum); accum++; // R
+
+    return accum;
+}
+
+static
+cmsUInt8Number* Unroll3BytesSkip1SwapFirst(register _cmsTRANSFORM* info, 
+                                           register cmsUInt16Number wIn[], 
+                                           register cmsUInt8Number* accum,
+                                           register cmsUInt32Number Stride)
+{
+    accum++; // A
+    wIn[0] = FROM_8_TO_16(*accum); accum++; // R
+    wIn[1] = FROM_8_TO_16(*accum); accum++; // G
+    wIn[2] = FROM_8_TO_16(*accum); accum++; // B
+
+    return accum;
+}
+
+
+// BRG
+static
+cmsUInt8Number* Unroll3BytesSwap(register _cmsTRANSFORM* info, 
+                                 register cmsUInt16Number wIn[], 
+                                 register cmsUInt8Number* accum,
+                                 register cmsUInt32Number Stride)
+{
+    wIn[2] = FROM_8_TO_16(*accum); accum++;     // B
+    wIn[1] = FROM_8_TO_16(*accum); accum++;     // G
+    wIn[0] = FROM_8_TO_16(*accum); accum++;     // R
+
+    return accum;
+}
+
+static
+cmsUInt8Number* UnrollLabV2_8(register _cmsTRANSFORM* info, 
+                              register cmsUInt16Number wIn[], 
+                              register cmsUInt8Number* accum,
+                              register cmsUInt32Number Stride)
+{
+    wIn[0] = FomLabV2ToLabV4(FROM_8_TO_16(*accum)); accum++;     // L
+    wIn[1] = FomLabV2ToLabV4(FROM_8_TO_16(*accum)); accum++;     // a
+    wIn[2] = FomLabV2ToLabV4(FROM_8_TO_16(*accum)); accum++;     // b
+
+    return accum;
+}
+
+static
+cmsUInt8Number* UnrollALabV2_8(register _cmsTRANSFORM* info, 
+                               register cmsUInt16Number wIn[], 
+                               register cmsUInt8Number* accum,
+                               register cmsUInt32Number Stride)
+{
+    accum++;  // A
+    wIn[0] = FomLabV2ToLabV4(FROM_8_TO_16(*accum)); accum++;     // L
+    wIn[1] = FomLabV2ToLabV4(FROM_8_TO_16(*accum)); accum++;     // a
+    wIn[2] = FomLabV2ToLabV4(FROM_8_TO_16(*accum)); accum++;     // b
+
+    return accum;
+}
+
+static
+cmsUInt8Number* UnrollLabV2_16(register _cmsTRANSFORM* info, 
+                               register cmsUInt16Number wIn[], 
+                               register cmsUInt8Number* accum,
+                               register cmsUInt32Number Stride)
+{
+    wIn[0] = FomLabV2ToLabV4(*(cmsUInt16Number*) accum); accum += 2;     // L
+    wIn[1] = FomLabV2ToLabV4(*(cmsUInt16Number*) accum); accum += 2;     // a
+    wIn[2] = FomLabV2ToLabV4(*(cmsUInt16Number*) accum); accum += 2;     // b
+
+    return accum;
+}
+
+// for duplex
+static
+cmsUInt8Number* Unroll2Bytes(register _cmsTRANSFORM* info, 
+                             register cmsUInt16Number wIn[], 
+                             register cmsUInt8Number* accum,
+                             register cmsUInt32Number Stride)
+{
+    wIn[0] = FROM_8_TO_16(*accum); accum++;     // ch1
+    wIn[1] = FROM_8_TO_16(*accum); accum++;     // ch2
+    return accum;
+}
+
+
+
+
+// Monochrome duplicates L into RGB for null-transforms
+static
+cmsUInt8Number* Unroll1Byte(register _cmsTRANSFORM* info, 
+                                     register cmsUInt16Number wIn[], 
+                                     register cmsUInt8Number* accum,
+                                     register cmsUInt32Number Stride)
+{
+    wIn[0] = wIn[1] = wIn[2] = FROM_8_TO_16(*accum); accum++;     // L       
+    return accum;
+}
+
+
+static
+cmsUInt8Number* Unroll1ByteSkip1(register _cmsTRANSFORM* info, 
+                            register cmsUInt16Number wIn[], 
+                            register cmsUInt8Number* accum,
+                            register cmsUInt32Number Stride)
+{
+    wIn[0] = wIn[1] = wIn[2] = FROM_8_TO_16(*accum); accum++;     // L
+    accum += 1;
+    return accum;
+}
+
+static
+cmsUInt8Number* Unroll1ByteSkip2(register _cmsTRANSFORM* info, 
+                                 register cmsUInt16Number wIn[], 
+                                 register cmsUInt8Number* accum,
+                                 register cmsUInt32Number Stride)
+{
+    wIn[0] = wIn[1] = wIn[2] = FROM_8_TO_16(*accum); accum++;     // L
+    accum += 2;
+    return accum;
+}
+
+static
+cmsUInt8Number* Unroll1ByteReversed(register _cmsTRANSFORM* info, 
+                                    register cmsUInt16Number wIn[], 
+                                    register cmsUInt8Number* accum,
+                                    register cmsUInt32Number Stride)
+{
+    wIn[0] = wIn[1] = wIn[2] = REVERSE_FLAVOR_16(FROM_8_TO_16(*accum)); accum++;     // L
+    return accum;
+}
+
+
+static
+cmsUInt8Number* UnrollAnyWords(register _cmsTRANSFORM* info, 
+                               register cmsUInt16Number wIn[], 
+                               register cmsUInt8Number* accum,
+                               register cmsUInt32Number Stride)
+{
+    int nChan       = T_CHANNELS(info -> InputFormat);
+    int SwapEndian  = T_ENDIAN16(info -> InputFormat);
+    int DoSwap      = T_DOSWAP(info ->InputFormat);
+    int Reverse     = T_FLAVOR(info ->InputFormat);
+    int SwapFirst   = T_SWAPFIRST(info -> InputFormat);
+    int Extra       = T_EXTRA(info -> InputFormat);
+    int ExtraFirst  = DoSwap && !SwapFirst;
+    int i;
+
+    if (ExtraFirst) {
+        accum += Extra * sizeof(cmsUInt16Number);
+    }
+
+    for (i=0; i < nChan; i++) {
+
+        int index = DoSwap ? (nChan - i - 1) : i;
+        cmsUInt16Number v = *(cmsUInt16Number*) accum; 
+
+        if (SwapEndian)
+            v = CHANGE_ENDIAN(v);
+
+        wIn[index] = Reverse ? REVERSE_FLAVOR_16(v) : v;
+
+        accum += sizeof(cmsUInt16Number);              
+    }
+
+    if (!ExtraFirst) {
+        accum += Extra * sizeof(cmsUInt16Number);
+    }
+
+    if (Extra == 0 && SwapFirst) {
+
+        cmsUInt16Number tmp = wIn[0];
+
+        memmove(&wIn[0], &wIn[1], (nChan-1) * sizeof(cmsUInt16Number));
+        wIn[nChan-1] = tmp;
+    }
+
+    return accum;
+}
+
+static
+cmsUInt8Number* UnrollPlanarWords(register _cmsTRANSFORM* info, 
+                                  register cmsUInt16Number wIn[], 
+                                  register cmsUInt8Number* accum,
+                                  register cmsUInt32Number Stride)
+{
+    int nChan = T_CHANNELS(info -> InputFormat);
+    int DoSwap= T_DOSWAP(info ->InputFormat);
+    int Reverse= T_FLAVOR(info ->InputFormat);
+    int SwapEndian = T_ENDIAN16(info -> InputFormat);
+    int i;
+    cmsUInt8Number* Init = accum;
+
+    if (DoSwap) {
+        accum += T_EXTRA(info -> InputFormat) * Stride * sizeof(cmsUInt16Number);
+    }
+
+    for (i=0; i < nChan; i++) {
+
+        int index = DoSwap ? (nChan - i - 1) : i;
+        cmsUInt16Number v = *(cmsUInt16Number*) accum; 
+
+        if (SwapEndian)
+            v = CHANGE_ENDIAN(v);
+
+        wIn[index] = Reverse ? REVERSE_FLAVOR_16(v) : v;
+
+        accum +=  Stride * sizeof(cmsUInt16Number);           
+    }
+
+    return (Init + sizeof(cmsUInt16Number));
+}
+
+
+static
+cmsUInt8Number* Unroll4Words(register _cmsTRANSFORM* info, 
+                             register cmsUInt16Number wIn[], 
+                             register cmsUInt8Number* accum,
+                             register cmsUInt32Number Stride)
+{
+    wIn[0] = *(cmsUInt16Number*) accum; accum+= 2; // C
+    wIn[1] = *(cmsUInt16Number*) accum; accum+= 2; // M
+    wIn[2] = *(cmsUInt16Number*) accum; accum+= 2; // Y
+    wIn[3] = *(cmsUInt16Number*) accum; accum+= 2; // K
+
+    return accum;
+}
+
+static
+cmsUInt8Number* Unroll4WordsReverse(register _cmsTRANSFORM* info, 
+                                    register cmsUInt16Number wIn[], 
+                                    register cmsUInt8Number* accum,
+                                    register cmsUInt32Number Stride)
+{
+    wIn[0] = REVERSE_FLAVOR_16(*(cmsUInt16Number*) accum); accum+= 2; // C
+    wIn[1] = REVERSE_FLAVOR_16(*(cmsUInt16Number*) accum); accum+= 2; // M
+    wIn[2] = REVERSE_FLAVOR_16(*(cmsUInt16Number*) accum); accum+= 2; // Y
+    wIn[3] = REVERSE_FLAVOR_16(*(cmsUInt16Number*) accum); accum+= 2; // K
+
+    return accum;
+}
+
+static
+cmsUInt8Number* Unroll4WordsSwapFirst(register _cmsTRANSFORM* info, 
+                                      register cmsUInt16Number wIn[], 
+                                      register cmsUInt8Number* accum,
+                                      register cmsUInt32Number Stride)
+{
+    wIn[3] = *(cmsUInt16Number*) accum; accum+= 2; // K
+    wIn[0] = *(cmsUInt16Number*) accum; accum+= 2; // C
+    wIn[1] = *(cmsUInt16Number*) accum; accum+= 2; // M
+    wIn[2] = *(cmsUInt16Number*) accum; accum+= 2; // Y
+
+    return accum;
+}
+
+// KYMC
+static
+cmsUInt8Number* Unroll4WordsSwap(register _cmsTRANSFORM* info, 
+                                 register cmsUInt16Number wIn[], 
+                                 register cmsUInt8Number* accum,
+                                 register cmsUInt32Number Stride)
+{
+    wIn[3] = *(cmsUInt16Number*) accum; accum+= 2; // K
+    wIn[2] = *(cmsUInt16Number*) accum; accum+= 2; // Y
+    wIn[1] = *(cmsUInt16Number*) accum; accum+= 2; // M
+    wIn[0] = *(cmsUInt16Number*) accum; accum+= 2; // C
+
+    return accum;
+}
+
+static
+cmsUInt8Number* Unroll4WordsSwapSwapFirst(register _cmsTRANSFORM* info, 
+                                          register cmsUInt16Number wIn[], 
+                                          register cmsUInt8Number* accum,
+                                          register cmsUInt32Number Stride)
+{
+    wIn[2] = *(cmsUInt16Number*) accum; accum+= 2; // K
+    wIn[1] = *(cmsUInt16Number*) accum; accum+= 2; // Y
+    wIn[0] = *(cmsUInt16Number*) accum; accum+= 2; // M
+    wIn[3] = *(cmsUInt16Number*) accum; accum+= 2; // C
+
+    return accum;
+}
+
+static
+cmsUInt8Number* Unroll3Words(register _cmsTRANSFORM* info, 
+                             register cmsUInt16Number wIn[], 
+                             register cmsUInt8Number* accum,
+                             register cmsUInt32Number Stride)
+{
+    wIn[0] = *(cmsUInt16Number*) accum; accum+= 2;  // C R
+    wIn[1] = *(cmsUInt16Number*) accum; accum+= 2;  // M G
+    wIn[2] = *(cmsUInt16Number*) accum; accum+= 2;  // Y B
+    return accum;
+}
+
+static
+cmsUInt8Number* Unroll3WordsSwap(register _cmsTRANSFORM* info, 
+                                 register cmsUInt16Number wIn[], 
+                                 register cmsUInt8Number* accum,
+                                 register cmsUInt32Number Stride)
+{
+    wIn[2] = *(cmsUInt16Number*) accum; accum+= 2;  // C R
+    wIn[1] = *(cmsUInt16Number*) accum; accum+= 2;  // M G
+    wIn[0] = *(cmsUInt16Number*) accum; accum+= 2;  // Y B
+    return accum;
+}
+
+static
+cmsUInt8Number* Unroll3WordsSkip1Swap(register _cmsTRANSFORM* info, 
+                                      register cmsUInt16Number wIn[], 
+                                      register cmsUInt8Number* accum,
+                                      register cmsUInt32Number Stride)
+{
+    accum += 2; // A
+    wIn[2] = *(cmsUInt16Number*) accum; accum += 2; // R
+    wIn[1] = *(cmsUInt16Number*) accum; accum += 2; // G
+    wIn[0] = *(cmsUInt16Number*) accum; accum += 2; // B
+
+    return accum;
+}
+
+static
+cmsUInt8Number* Unroll3WordsSkip1SwapFirst(register _cmsTRANSFORM* info, 
+                                           register cmsUInt16Number wIn[], 
+                                           register cmsUInt8Number* accum,
+                                           register cmsUInt32Number Stride)
+{
+    accum += 2; // A
+    wIn[0] = *(cmsUInt16Number*) accum; accum += 2; // R
+    wIn[1] = *(cmsUInt16Number*) accum; accum += 2; // G
+    wIn[2] = *(cmsUInt16Number*) accum; accum += 2; // B
+
+    return accum;
+}
+
+static
+cmsUInt8Number* Unroll1Word(register _cmsTRANSFORM* info, 
+                            register cmsUInt16Number wIn[], 
+                            register cmsUInt8Number* accum,
+                            register cmsUInt32Number Stride)
+{
+    wIn[0] = wIn[1] = wIn[2] = *(cmsUInt16Number*) accum; accum+= 2;   // L
+    return accum;
+}
+
+static
+cmsUInt8Number* Unroll1WordReversed(register _cmsTRANSFORM* info, 
+                                    register cmsUInt16Number wIn[], 
+                                    register cmsUInt8Number* accum,
+                                    register cmsUInt32Number Stride)
+{
+    wIn[0] = wIn[1] = wIn[2] = REVERSE_FLAVOR_16(*(cmsUInt16Number*) accum); accum+= 2;
+    return accum;
+}
+
+static
+cmsUInt8Number* Unroll1WordSkip3(register _cmsTRANSFORM* info, 
+                                 register cmsUInt16Number wIn[], 
+                                 register cmsUInt8Number* accum,
+                                 register cmsUInt32Number Stride)
+{
+    wIn[0] = wIn[1] = wIn[2] = *(cmsUInt16Number*) accum; 
+
+    accum += 8;
+    return accum;
+}
+
+static
+cmsUInt8Number* Unroll2Words(register _cmsTRANSFORM* info, 
+                             register cmsUInt16Number wIn[], 
+                             register cmsUInt8Number* accum,
+                             register cmsUInt32Number Stride)
+{
+    wIn[0] = *(cmsUInt16Number*) accum; accum += 2;    // ch1
+    wIn[1] = *(cmsUInt16Number*) accum; accum += 2;    // ch2
+
+    return accum;
+}
+
+
+// This is a conversion of Lab double to 16 bits
+static
+cmsUInt8Number* UnrollLabDoubleTo16(register _cmsTRANSFORM* info, 
+                                    register cmsUInt16Number wIn[], 
+                                    register cmsUInt8Number* accum,
+                                    register cmsUInt32Number  Stride)
+{       
+    if (T_PLANAR(info -> InputFormat)) {
+
+        cmsFloat64Number* Pt = (cmsFloat64Number*) accum;
+
+        cmsCIELab Lab;
+
+        Lab.L = Pt[0];
+        Lab.a = Pt[Stride];
+        Lab.b = Pt[Stride*2];
+
+        cmsFloat2LabEncoded(wIn, &Lab);
+        return accum + sizeof(cmsFloat64Number);
+    }
+    else {
+
+        cmsFloat2LabEncoded(wIn, (cmsCIELab*) accum);
+        accum += sizeof(cmsCIELab) + T_EXTRA(info ->InputFormat) * sizeof(cmsFloat64Number);
+        return accum;
+    }
+}
+
+// This is a conversion of XYZ double to 16 bits
+static
+cmsUInt8Number* UnrollXYZDoubleTo16(register _cmsTRANSFORM* info, 
+                                    register cmsUInt16Number wIn[], 
+                                    register cmsUInt8Number* accum,
+                                    register cmsUInt32Number Stride)
+{   
+    if (T_PLANAR(info -> InputFormat)) {
+
+        cmsFloat64Number* Pt = (cmsFloat64Number*) accum;
+        cmsCIEXYZ XYZ;
+
+        XYZ.X = Pt[0];
+        XYZ.Y = Pt[Stride];
+        XYZ.Z = Pt[Stride*2];
+        cmsFloat2XYZEncoded(wIn, &XYZ);
+
+        return accum + sizeof(cmsFloat64Number);
+
+    }
+
+    else {
+        cmsFloat2XYZEncoded(wIn, (cmsCIEXYZ*) accum);
+        accum += sizeof(cmsCIEXYZ) + T_EXTRA(info ->InputFormat) * sizeof(cmsFloat64Number);
+
+        return accum;
+    }
+}
+
+// Check if space is marked as ink
+cmsINLINE cmsBool IsInkSpace(cmsUInt32Number Type)
+{
+    switch (T_COLORSPACE(Type)) {
+
+     case PT_CMY:       
+     case PT_CMYK:      
+     case PT_MCH5:      
+     case PT_MCH6:
+     case PT_MCH7:     
+     case PT_MCH8:     
+     case PT_MCH9:     
+     case PT_MCH10:    
+     case PT_MCH11:    
+     case PT_MCH12:    
+     case PT_MCH13:    
+     case PT_MCH14:    
+     case PT_MCH15: return TRUE;
+
+     default: return FALSE;
+    }
+}
+
+// Inks does come in percentage, remaining cases are between 0..1.0, again to 16 bits
+static
+cmsUInt8Number* UnrollDoubleTo16(register _cmsTRANSFORM* info, 
+                                 register cmsUInt16Number wIn[], 
+                                 register cmsUInt8Number* accum,
+                                 register cmsUInt32Number Stride)
+{
+    cmsFloat64Number* Inks = (cmsFloat64Number*) accum;
+    int nChan  = T_CHANNELS(info -> InputFormat);
+    int Planar = T_PLANAR(info -> InputFormat);
+    int i;
+    cmsFloat64Number v;
+    cmsFloat64Number maximum = IsInkSpace(info ->InputFormat) ? 655.35 : 65535.0;
+
+    for (i=0; i < nChan; i++) {
+
+        if (Planar)
+
+            v = Inks[i * Stride];
+        else
+            v = Inks[i];
+
+        wIn[i] = _cmsQuickSaturateWord(v * maximum);
+    }
+
+    if (T_PLANAR(info -> InputFormat))
+        return accum + sizeof(cmsFloat64Number);
+    else
+        return accum + (nChan + T_EXTRA(info ->InputFormat)) * sizeof(cmsFloat64Number);
+}
+
+static
+cmsUInt8Number* UnrollFloatTo16(register _cmsTRANSFORM* info, 
+                                register cmsUInt16Number wIn[], 
+                                register cmsUInt8Number* accum,
+                                register cmsUInt32Number Stride)
+{
+    cmsFloat32Number* Inks = (cmsFloat32Number*) accum;
+    int nChan  = T_CHANNELS(info -> InputFormat);
+    int Planar = T_PLANAR(info -> InputFormat);
+    int i;
+    cmsFloat32Number v;
+    cmsFloat64Number maximum = IsInkSpace(info ->InputFormat) ? 655.35 : 65535.0;
+
+    for (i=0; i < nChan; i++) {
+
+        if (Planar)
+
+            v = Inks[i * Stride];
+        else
+            v = Inks[i];
+
+        wIn[i] = _cmsQuickSaturateWord(v * maximum);
+    }
+
+    if (T_PLANAR(info -> InputFormat))
+        return accum + sizeof(cmsFloat32Number);
+    else
+        return accum + (nChan + T_EXTRA(info ->InputFormat)) * sizeof(cmsFloat32Number);
+}
+
+
+// For 1 channel, we need to duplicate data (it comes in 0..1.0 range)
+static
+cmsUInt8Number* UnrollDouble1Chan(register _cmsTRANSFORM* info, 
+                                  register cmsUInt16Number wIn[], 
+                                  register cmsUInt8Number* accum,
+                                  register cmsUInt32Number Stride)
+{
+    cmsFloat64Number* Inks = (cmsFloat64Number*) accum;
+
+    wIn[0] = wIn[1] = wIn[2] = _cmsQuickSaturateWord(Inks[0] * 65535.0);
+
+    return accum + sizeof(cmsFloat64Number);    
+}
+
+//-------------------------------------------------------------------------------------------------------------------
+
+// True float transformation. 
+
+// For anything going from cmsFloat32Number 
+static
+cmsUInt8Number* UnrollFloatsToFloat(_cmsTRANSFORM* info, 
+                                    cmsFloat32Number wIn[], 
+                                    cmsUInt8Number* accum,
+                                    cmsUInt32Number Stride)
+{
+    cmsFloat32Number* Inks = (cmsFloat32Number*) accum;
+    int nChan  = T_CHANNELS(info -> InputFormat);
+    int Planar = T_PLANAR(info -> InputFormat);
+    int i;
+    cmsFloat64Number maximum = IsInkSpace(info ->InputFormat) ? 100.0 : 1.0;
+
+
+    for (i=0; i <  nChan; i++) {
+
+        if (Planar)
+            wIn[i] = (cmsFloat32Number) (Inks[i * Stride] / maximum);
+        else
+            wIn[i] = (cmsFloat32Number) (Inks[i] / maximum);                            
+    }
+
+    if (T_PLANAR(info -> InputFormat))
+        return accum + sizeof(cmsFloat32Number);
+    else
+        return accum + (nChan + T_EXTRA(info ->InputFormat)) * sizeof(cmsFloat32Number);
+}
+
+// For anything going from double
+static
+cmsUInt8Number* UnrollDoublesToFloat(_cmsTRANSFORM* info, 
+                                     cmsFloat32Number wIn[], 
+                                     cmsUInt8Number* accum,
+                                     cmsUInt32Number Stride)
+{
+    cmsFloat64Number* Inks = (cmsFloat64Number*) accum;
+    int nChan  = T_CHANNELS(info -> InputFormat);
+    int Planar = T_PLANAR(info -> InputFormat);
+    int i;
+    cmsFloat64Number maximum = IsInkSpace(info ->InputFormat) ? 100.0 : 1.0;
+
+    for (i=0; i <  nChan; i++) {
+
+        if (Planar)
+            wIn[i] = (cmsFloat32Number) (Inks[i * Stride] / maximum);
+        else
+            wIn[i] = (cmsFloat32Number) (Inks[i] / maximum);                            
+    }
+
+    if (T_PLANAR(info -> InputFormat))
+        return accum + sizeof(cmsFloat64Number);
+    else
+        return accum + (nChan + T_EXTRA(info ->InputFormat)) * sizeof(cmsFloat64Number);
+}
+
+
+// From Lab double to cmsFloat32Number
+static
+cmsUInt8Number* UnrollLabDoubleToFloat(_cmsTRANSFORM* info,  
+                                       cmsFloat32Number wIn[], 
+                                       cmsUInt8Number* accum,
+                                       cmsUInt32Number Stride)
+{  
+    cmsFloat64Number* Pt = (cmsFloat64Number*) accum;
+
+    if (T_PLANAR(info -> InputFormat)) {
+
+        wIn[0] = (cmsFloat32Number) (Pt[0] / 100.0);                            // from 0..100 to 0..1 
+        wIn[1] = (cmsFloat32Number) ((Pt[Stride] + 128) / 255.0);    // form -128..+127 to 0..1
+        wIn[2] = (cmsFloat32Number) ((Pt[Stride*2] + 128) / 255.0);
+
+        return accum + sizeof(cmsFloat64Number);
+    }
+    else {
+
+        wIn[0] = (cmsFloat32Number) (Pt[0] / 100.0);             // from 0..100 to 0..1 
+        wIn[1] = (cmsFloat32Number) ((Pt[1] + 128) / 255.0);    // form -128..+127 to 0..1
+        wIn[2] = (cmsFloat32Number) ((Pt[2] + 128) / 255.0);
+
+        accum += sizeof(cmsFloat64Number)*(3 + T_EXTRA(info ->InputFormat));
+        return accum;
+    }
+}
+
+// From Lab double to cmsFloat32Number
+static
+cmsUInt8Number* UnrollLabFloatToFloat(_cmsTRANSFORM* info, 
+                                      cmsFloat32Number wIn[], 
+                                      cmsUInt8Number* accum,
+                                      cmsUInt32Number Stride)
+{  
+    cmsFloat32Number* Pt = (cmsFloat32Number*) accum;
+
+    if (T_PLANAR(info -> InputFormat)) {
+
+        wIn[0] = (cmsFloat32Number) (Pt[0] / 100.0);                            // from 0..100 to 0..1 
+        wIn[1] = (cmsFloat32Number) ((Pt[Stride] + 128) / 255.0);    // form -128..+127 to 0..1
+        wIn[2] = (cmsFloat32Number) ((Pt[Stride*2] + 128) / 255.0);
+
+        return accum + sizeof(cmsFloat32Number);
+    }
+    else {
+
+        wIn[0] = (cmsFloat32Number) (Pt[0] / 100.0);             // from 0..100 to 0..1 
+        wIn[1] = (cmsFloat32Number) ((Pt[1] + 128) / 255.0);    // form -128..+127 to 0..1
+        wIn[2] = (cmsFloat32Number) ((Pt[2] + 128) / 255.0);
+
+        accum += sizeof(cmsFloat32Number)*(3 + T_EXTRA(info ->InputFormat));
+        return accum;
+    }
+}
+
+
+// 1.15 fixed point, that means maximum value is MAX_ENCODEABLE_XYZ (0xFFFF)
+static
+cmsUInt8Number* UnrollXYZDoubleToFloat(_cmsTRANSFORM* info,  
+                                       cmsFloat32Number wIn[], 
+                                       cmsUInt8Number* accum,
+                                       cmsUInt32Number Stride)
+{  
+    cmsFloat64Number* Pt = (cmsFloat64Number*) accum;
+
+    if (T_PLANAR(info -> InputFormat)) {
+
+        wIn[0] = (cmsFloat32Number) (Pt[0] / MAX_ENCODEABLE_XYZ);                   
+        wIn[1] = (cmsFloat32Number) (Pt[Stride] / MAX_ENCODEABLE_XYZ);    
+        wIn[2] = (cmsFloat32Number) (Pt[Stride*2] / MAX_ENCODEABLE_XYZ);
+
+        return accum + sizeof(cmsFloat64Number);
+    }
+    else {
+
+        wIn[0] = (cmsFloat32Number) (Pt[0] / MAX_ENCODEABLE_XYZ);            
+        wIn[1] = (cmsFloat32Number) (Pt[1] / MAX_ENCODEABLE_XYZ);    
+        wIn[2] = (cmsFloat32Number) (Pt[2] / MAX_ENCODEABLE_XYZ);
+
+        accum += sizeof(cmsFloat64Number)*(3 + T_EXTRA(info ->InputFormat));
+        return accum;
+    }
+}
+
+static
+cmsUInt8Number* UnrollXYZFloatToFloat(_cmsTRANSFORM* info,  
+                                      cmsFloat32Number wIn[], 
+                                      cmsUInt8Number* accum,
+                                      cmsUInt32Number Stride)
+{  
+    cmsFloat32Number* Pt = (cmsFloat32Number*) accum;
+
+    if (T_PLANAR(info -> InputFormat)) {
+
+        wIn[0] = (cmsFloat32Number) (Pt[0] / MAX_ENCODEABLE_XYZ);                   
+        wIn[1] = (cmsFloat32Number) (Pt[Stride] / MAX_ENCODEABLE_XYZ);    
+        wIn[2] = (cmsFloat32Number) (Pt[Stride*2] / MAX_ENCODEABLE_XYZ);
+
+        return accum + sizeof(cmsFloat32Number);
+    }
+    else {
+
+        wIn[0] = (cmsFloat32Number) (Pt[0] / MAX_ENCODEABLE_XYZ);            
+        wIn[1] = (cmsFloat32Number) (Pt[1] / MAX_ENCODEABLE_XYZ);    
+        wIn[2] = (cmsFloat32Number) (Pt[2] / MAX_ENCODEABLE_XYZ);
+
+        accum += sizeof(cmsFloat32Number)*(3 + T_EXTRA(info ->InputFormat));
+        return accum;
+    }
+}
+
+// Packing routines -----------------------------------------------------------------------------------------------------------
+
+
+// Generic chunky for byte
+
+static
+cmsUInt8Number* PackAnyBytes(register _cmsTRANSFORM* info, 
+                             register cmsUInt16Number wOut[], 
+                             register cmsUInt8Number* output,
+                             register cmsUInt32Number Stride)
+{
+    int nChan      = T_CHANNELS(info -> OutputFormat);
+    int DoSwap     = T_DOSWAP(info ->OutputFormat);
+    int Reverse    = T_FLAVOR(info ->OutputFormat);
+    int Extra      = T_EXTRA(info -> OutputFormat);
+    int SwapFirst  = T_SWAPFIRST(info -> OutputFormat);
+    int ExtraFirst = DoSwap && !SwapFirst;
+    cmsUInt8Number* swap1;
+    cmsUInt8Number v = 0;
+    int i;
+
+    swap1 = output;
+
+    if (ExtraFirst) {
+        output += Extra;
+    }
+
+    for (i=0; i < nChan; i++) {
+
+        int index = DoSwap ? (nChan - i - 1) : i;
+
+        v = FROM_16_TO_8(wOut[index]);
+
+        if (Reverse)
+            v = REVERSE_FLAVOR_8(v);
+
+        *output++ = v;
+    }
+
+    if (!ExtraFirst) {
+        output += Extra;
+    }
+
+    if (Extra == 0 && SwapFirst) {
+
+        memmove(swap1 + 1, swap1, nChan-1);
+        *swap1 = v;
+    }
+
+
+    return output;
+}
+
+
+
+static
+cmsUInt8Number* PackAnyWords(register _cmsTRANSFORM* info, 
+                             register cmsUInt16Number wOut[], 
+                             register cmsUInt8Number* output,
+                             register cmsUInt32Number Stride)
+{
+    int nChan      = T_CHANNELS(info -> OutputFormat);
+    int SwapEndian = T_ENDIAN16(info -> InputFormat);
+    int DoSwap     = T_DOSWAP(info ->OutputFormat);
+    int Reverse    = T_FLAVOR(info ->OutputFormat);
+    int Extra      = T_EXTRA(info -> OutputFormat);
+    int SwapFirst  = T_SWAPFIRST(info -> OutputFormat);
+    int ExtraFirst = DoSwap && !SwapFirst;
+    cmsUInt16Number* swap1;
+    cmsUInt16Number v = 0;
+    int i;
+
+    swap1 = (cmsUInt16Number*) output;
+
+    if (ExtraFirst) {
+        output += Extra * sizeof(cmsUInt16Number);
+    }
+
+    for (i=0; i < nChan; i++) {
+
+        int index = DoSwap ? (nChan - i - 1) : i;
+
+        v = wOut[index];
+
+        if (SwapEndian)
+            v = CHANGE_ENDIAN(v);
+
+        if (Reverse)
+            v = REVERSE_FLAVOR_16(v);
+
+        *(cmsUInt16Number*) output = v;
+
+        output += sizeof(cmsUInt16Number);
+    }
+
+    if (!ExtraFirst) {
+        output += Extra * sizeof(cmsUInt16Number);
+    }
+
+    if (Extra == 0 && SwapFirst) {
+
+        memmove(swap1 + 1, swap1, (nChan-1)* sizeof(cmsUInt16Number));
+        *swap1 = v;
+    }
+
+
+    return output;
+}
+
+
+static
+cmsUInt8Number* PackPlanarBytes(register _cmsTRANSFORM* info, 
+                                register cmsUInt16Number wOut[], 
+                                register cmsUInt8Number* output,
+                                register cmsUInt32Number Stride)
+{
+    int nChan = T_CHANNELS(info -> OutputFormat);
+    int DoSwap = T_DOSWAP(info ->OutputFormat);
+    int Reverse= T_FLAVOR(info ->OutputFormat);
+    int i;
+    cmsUInt8Number* Init = output;
+
+    for (i=0; i < nChan; i++) {
+
+        int index = DoSwap ? (nChan - i - 1) : i;
+        cmsUInt8Number v = FROM_16_TO_8(wOut[index]);
+
+        *(cmsUInt8Number*)  output = (cmsUInt8Number) (Reverse ? REVERSE_FLAVOR_8(v) : v);
+        output += Stride;
+    }
+
+    return (Init + 1);
+}
+
+
+static
+cmsUInt8Number* PackPlanarWords(register _cmsTRANSFORM* info, 
+                                register cmsUInt16Number wOut[], 
+                                register cmsUInt8Number* output,
+                                register cmsUInt32Number Stride)
+{
+    int nChan = T_CHANNELS(info -> OutputFormat);
+    int DoSwap = T_DOSWAP(info ->OutputFormat);
+    int Reverse= T_FLAVOR(info ->OutputFormat);
+    int SwapEndian = T_ENDIAN16(info -> OutputFormat);
+    int i;
+    cmsUInt8Number* Init = output;
+    cmsUInt16Number v;
+
+    if (DoSwap) {
+        output += T_EXTRA(info -> OutputFormat) * Stride * sizeof(cmsUInt16Number);
+    }
+
+    for (i=0; i < nChan; i++) {
+
+        int index = DoSwap ? (nChan - i - 1) : i;
+
+        v = wOut[index];
+
+        if (SwapEndian)
+            v = CHANGE_ENDIAN(v);
+
+        if (Reverse) 
+            v =  REVERSE_FLAVOR_16(v);
+
+        *(cmsUInt16Number*) output = v;
+        output += (Stride * sizeof(cmsUInt16Number));
+    }
+
+    return (Init + sizeof(cmsUInt16Number));
+}
+
+// CMYKcm (unrolled for speed)
+
+static
+cmsUInt8Number* Pack6Bytes(register _cmsTRANSFORM* info, 
+                           register cmsUInt16Number wOut[], 
+                           register cmsUInt8Number* output,
+                           register cmsUInt32Number Stride)
+{
+    *output++ = FROM_16_TO_8(wOut[0]);
+    *output++ = FROM_16_TO_8(wOut[1]);
+    *output++ = FROM_16_TO_8(wOut[2]);
+    *output++ = FROM_16_TO_8(wOut[3]);
+    *output++ = FROM_16_TO_8(wOut[4]);
+    *output++ = FROM_16_TO_8(wOut[5]);
+
+    return output;
+}
+
+// KCMYcm
+
+static
+cmsUInt8Number* Pack6BytesSwap(register _cmsTRANSFORM* info, 
+                               register cmsUInt16Number wOut[], 
+                               register cmsUInt8Number* output,
+                               register cmsUInt32Number Stride)
+{
+    *output++ = FROM_16_TO_8(wOut[5]);
+    *output++ = FROM_16_TO_8(wOut[4]);
+    *output++ = FROM_16_TO_8(wOut[3]);
+    *output++ = FROM_16_TO_8(wOut[2]);
+    *output++ = FROM_16_TO_8(wOut[1]);
+    *output++ = FROM_16_TO_8(wOut[0]);
+
+    return output;
+}
+
+// CMYKcm
+static
+cmsUInt8Number* Pack6Words(register _cmsTRANSFORM* info, 
+                           register cmsUInt16Number wOut[], 
+                           register cmsUInt8Number* output,
+                           register cmsUInt32Number Stride)
+{
+    *(cmsUInt16Number*) output = wOut[0];
+    output+= 2;
+    *(cmsUInt16Number*) output = wOut[1];
+    output+= 2;
+    *(cmsUInt16Number*) output = wOut[2];
+    output+= 2;
+    *(cmsUInt16Number*) output = wOut[3];
+    output+= 2;
+    *(cmsUInt16Number*) output = wOut[4];
+    output+= 2;
+    *(cmsUInt16Number*) output = wOut[5];
+    output+= 2;
+
+    return output;
+}
+
+// KCMYcm
+static
+cmsUInt8Number* Pack6WordsSwap(register _cmsTRANSFORM* info, 
+                               register cmsUInt16Number wOut[], 
+                               register cmsUInt8Number* output,
+                               register cmsUInt32Number Stride)
+{
+    *(cmsUInt16Number*) output = wOut[5];
+    output+= 2;
+    *(cmsUInt16Number*) output = wOut[4];
+    output+= 2;
+    *(cmsUInt16Number*) output = wOut[3];
+    output+= 2;
+    *(cmsUInt16Number*) output = wOut[2];
+    output+= 2;
+    *(cmsUInt16Number*) output = wOut[1];
+    output+= 2;
+    *(cmsUInt16Number*) output = wOut[0];
+    output+= 2;
+
+    return output;
+}
+
+
+static
+cmsUInt8Number* Pack4Bytes(register _cmsTRANSFORM* info, 
+                           register cmsUInt16Number wOut[], 
+                           register cmsUInt8Number* output,
+                           register cmsUInt32Number Stride)
+{
+    *output++ = FROM_16_TO_8(wOut[0]);
+    *output++ = FROM_16_TO_8(wOut[1]);
+    *output++ = FROM_16_TO_8(wOut[2]);
+    *output++ = FROM_16_TO_8(wOut[3]);
+
+    return output;
+}
+
+static
+cmsUInt8Number* Pack4BytesReverse(register _cmsTRANSFORM* info, 
+                                  register cmsUInt16Number wOut[], 
+                                  register cmsUInt8Number* output,
+                                  register cmsUInt32Number Stride)
+{
+    *output++ = REVERSE_FLAVOR_8(FROM_16_TO_8(wOut[0]));
+    *output++ = REVERSE_FLAVOR_8(FROM_16_TO_8(wOut[1]));
+    *output++ = REVERSE_FLAVOR_8(FROM_16_TO_8(wOut[2]));
+    *output++ = REVERSE_FLAVOR_8(FROM_16_TO_8(wOut[3]));
+
+    return output;
+}
+
+
+static
+cmsUInt8Number* Pack4BytesSwapFirst(register _cmsTRANSFORM* info, 
+                                    register cmsUInt16Number wOut[], 
+                                    register cmsUInt8Number* output,
+                                    register cmsUInt32Number Stride)
+{
+    *output++ = FROM_16_TO_8(wOut[3]);
+    *output++ = FROM_16_TO_8(wOut[0]);
+    *output++ = FROM_16_TO_8(wOut[1]);
+    *output++ = FROM_16_TO_8(wOut[2]);
+
+    return output;
+}
+
+// ABGR
+static
+cmsUInt8Number* Pack4BytesSwap(register _cmsTRANSFORM* info, 
+                               register cmsUInt16Number wOut[], 
+                               register cmsUInt8Number* output,
+                               register cmsUInt32Number Stride)
+{
+    *output++ = FROM_16_TO_8(wOut[3]);
+    *output++ = FROM_16_TO_8(wOut[2]);
+    *output++ = FROM_16_TO_8(wOut[1]);
+    *output++ = FROM_16_TO_8(wOut[0]);
+
+    return output;
+}
+
+static
+cmsUInt8Number* Pack4BytesSwapSwapFirst(register _cmsTRANSFORM* info, 
+                                        register cmsUInt16Number wOut[], 
+                                        register cmsUInt8Number* output,
+                                        register cmsUInt32Number Stride)
+{
+    *output++ = FROM_16_TO_8(wOut[2]);
+    *output++ = FROM_16_TO_8(wOut[1]);
+    *output++ = FROM_16_TO_8(wOut[0]);
+    *output++ = FROM_16_TO_8(wOut[3]);
+
+    return output;
+}
+
+static
+cmsUInt8Number* Pack4Words(register _cmsTRANSFORM* info, 
+                           register cmsUInt16Number wOut[], 
+                           register cmsUInt8Number* output,
+                           register cmsUInt32Number Stride)
+{
+    *(cmsUInt16Number*) output = wOut[0];
+    output+= 2;
+    *(cmsUInt16Number*) output = wOut[1];
+    output+= 2;
+    *(cmsUInt16Number*) output = wOut[2];
+    output+= 2;
+    *(cmsUInt16Number*) output = wOut[3];
+    output+= 2;
+
+    return output;
+}
+
+static
+cmsUInt8Number* Pack4WordsReverse(register _cmsTRANSFORM* info, 
+                                  register cmsUInt16Number wOut[], 
+                                  register cmsUInt8Number* output,
+                                  register cmsUInt32Number Stride)
+{
+    *(cmsUInt16Number*) output = REVERSE_FLAVOR_16(wOut[0]);
+    output+= 2;
+    *(cmsUInt16Number*) output = REVERSE_FLAVOR_16(wOut[1]);
+    output+= 2;
+    *(cmsUInt16Number*) output = REVERSE_FLAVOR_16(wOut[2]);
+    output+= 2;
+    *(cmsUInt16Number*) output = REVERSE_FLAVOR_16(wOut[3]);
+    output+= 2;
+
+    return output;
+}
+
+// ABGR
+static
+cmsUInt8Number* Pack4WordsSwap(register _cmsTRANSFORM* info, 
+                               register cmsUInt16Number wOut[], 
+                               register cmsUInt8Number* output,
+                               register cmsUInt32Number Stride)
+{
+    *(cmsUInt16Number*) output = wOut[3];
+    output+= 2;
+    *(cmsUInt16Number*) output = wOut[2];
+    output+= 2;
+    *(cmsUInt16Number*) output = wOut[1];
+    output+= 2;
+    *(cmsUInt16Number*) output = wOut[0];
+    output+= 2;
+
+    return output;
+}
+
+// CMYK
+static
+cmsUInt8Number* Pack4WordsBigEndian(register _cmsTRANSFORM* info, 
+                                    register cmsUInt16Number wOut[], 
+                                    register cmsUInt8Number* output,
+                                    register cmsUInt32Number Stride)
+{
+    *(cmsUInt16Number*) output = CHANGE_ENDIAN(wOut[0]);
+    output+= 2;
+    *(cmsUInt16Number*) output = CHANGE_ENDIAN(wOut[1]);
+    output+= 2;
+    *(cmsUInt16Number*) output = CHANGE_ENDIAN(wOut[2]);
+    output+= 2;
+    *(cmsUInt16Number*) output = CHANGE_ENDIAN(wOut[3]);
+    output+= 2;
+
+    return output;
+}
+
+
+static
+cmsUInt8Number* PackLabV2_8(register _cmsTRANSFORM* info, 
+                            register cmsUInt16Number wOut[], 
+                            register cmsUInt8Number* output,
+                            register cmsUInt32Number Stride)
+{
+    *output++ = FROM_16_TO_8(FomLabV4ToLabV2(wOut[0]));
+    *output++ = FROM_16_TO_8(FomLabV4ToLabV2(wOut[1]));
+    *output++ = FROM_16_TO_8(FomLabV4ToLabV2(wOut[2]));
+
+    return output;
+}
+
+static
+cmsUInt8Number* PackALabV2_8(register _cmsTRANSFORM* info, 
+                             register cmsUInt16Number wOut[], 
+                             register cmsUInt8Number* output,
+                             register cmsUInt32Number Stride)
+{
+    output++;
+    *output++ = FROM_16_TO_8(FomLabV4ToLabV2(wOut[0]));
+    *output++ = FROM_16_TO_8(FomLabV4ToLabV2(wOut[1]));
+    *output++ = FROM_16_TO_8(FomLabV4ToLabV2(wOut[2]));
+
+    return output;
+}
+
+static
+cmsUInt8Number* PackLabV2_16(register _cmsTRANSFORM* info, 
+                             register cmsUInt16Number wOut[], 
+                             register cmsUInt8Number* output,
+                             register cmsUInt32Number Stride)
+{
+    *(cmsUInt16Number*) output = FomLabV4ToLabV2(wOut[0]);
+    output += 2;
+    *(cmsUInt16Number*) output = FomLabV4ToLabV2(wOut[1]);
+    output += 2;
+    *(cmsUInt16Number*) output = FomLabV4ToLabV2(wOut[2]);
+    output += 2;
+
+    return output;
+}
+
+static
+cmsUInt8Number* Pack3Bytes(register _cmsTRANSFORM* info, 
+                           register cmsUInt16Number wOut[], 
+                           register cmsUInt8Number* output,
+                           register cmsUInt32Number Stride)
+{
+    *output++ = FROM_16_TO_8(wOut[0]);
+    *output++ = FROM_16_TO_8(wOut[1]);
+    *output++ = FROM_16_TO_8(wOut[2]);
+
+    return output;
+}
+
+static
+cmsUInt8Number* Pack3BytesOptimized(register _cmsTRANSFORM* info, 
+                                    register cmsUInt16Number wOut[], 
+                                    register cmsUInt8Number* output,
+                                    register cmsUInt32Number Stride)
+{
+    *output++ = (wOut[0] & 0xFF);
+    *output++ = (wOut[1] & 0xFF);
+    *output++ = (wOut[2] & 0xFF);
+
+    return output;
+}
+
+static
+cmsUInt8Number* Pack3BytesSwap(register _cmsTRANSFORM* info, 
+                               register cmsUInt16Number wOut[], 
+                               register cmsUInt8Number* output,
+                               register cmsUInt32Number Stride)
+{
+    *output++ = FROM_16_TO_8(wOut[2]);
+    *output++ = FROM_16_TO_8(wOut[1]);
+    *output++ = FROM_16_TO_8(wOut[0]);
+
+    return output;
+}
+
+static
+cmsUInt8Number* Pack3BytesSwapOptimized(register _cmsTRANSFORM* info, 
+                                        register cmsUInt16Number wOut[], 
+                                        register cmsUInt8Number* output,
+                                        register cmsUInt32Number Stride)
+{
+    *output++ = (wOut[2] & 0xFF);
+    *output++ = (wOut[1] & 0xFF);
+    *output++ = (wOut[0] & 0xFF);
+
+    return output;
+}
+
+
+static
+cmsUInt8Number* Pack3Words(register _cmsTRANSFORM* info, 
+                           register cmsUInt16Number wOut[], 
+                           register cmsUInt8Number* output,
+                           register cmsUInt32Number Stride)
+{
+    *(cmsUInt16Number*) output = wOut[0];
+    output+= 2;
+    *(cmsUInt16Number*) output = wOut[1];
+    output+= 2;
+    *(cmsUInt16Number*) output = wOut[2];
+    output+= 2;
+
+    return output;
+}
+
+static
+cmsUInt8Number* Pack3WordsSwap(register _cmsTRANSFORM* info, 
+                               register cmsUInt16Number wOut[], 
+                               register cmsUInt8Number* output,
+                               register cmsUInt32Number Stride)
+{
+    *(cmsUInt16Number*) output = wOut[2];
+    output+= 2;
+    *(cmsUInt16Number*) output = wOut[1];
+    output+= 2;
+    *(cmsUInt16Number*) output = wOut[0];
+    output+= 2;
+
+    return output;
+}
+
+static
+cmsUInt8Number* Pack3WordsBigEndian(register _cmsTRANSFORM* info, 
+                                    register cmsUInt16Number wOut[], 
+                                    register cmsUInt8Number* output,
+                                    register cmsUInt32Number Stride)
+{
+    *(cmsUInt16Number*) output = CHANGE_ENDIAN(wOut[0]);
+    output+= 2;
+    *(cmsUInt16Number*) output = CHANGE_ENDIAN(wOut[1]);
+    output+= 2;
+    *(cmsUInt16Number*) output = CHANGE_ENDIAN(wOut[2]);
+    output+= 2;
+
+    return output;
+}
+
+static
+cmsUInt8Number* Pack3BytesAndSkip1(register _cmsTRANSFORM* Info, 
+                                   register cmsUInt16Number wOut[], 
+                                   register cmsUInt8Number* output,
+                                   register cmsUInt32Number Stride)
+{
+    *output++ = FROM_16_TO_8(wOut[0]);
+    *output++ = FROM_16_TO_8(wOut[1]);
+    *output++ = FROM_16_TO_8(wOut[2]);
+    output++;
+
+    return output;
+}
+
+static
+cmsUInt8Number* Pack3BytesAndSkip1Optimized(register _cmsTRANSFORM* Info, 
+                                            register cmsUInt16Number wOut[], 
+                                            register cmsUInt8Number* output,
+                                            register cmsUInt32Number Stride)
+{
+    *output++ = (wOut[0] & 0xFF);
+    *output++ = (wOut[1] & 0xFF);
+    *output++ = (wOut[2] & 0xFF);
+    output++;
+
+    return output;
+}
+
+
+static
+cmsUInt8Number* Pack3BytesAndSkip1SwapFirst(register _cmsTRANSFORM* Info, 
+                                            register cmsUInt16Number wOut[], 
+                                            register cmsUInt8Number* output,
+                                            register cmsUInt32Number Stride)
+{
+    output++;
+    *output++ = FROM_16_TO_8(wOut[0]);
+    *output++ = FROM_16_TO_8(wOut[1]);
+    *output++ = FROM_16_TO_8(wOut[2]);
+
+    return output;
+}
+
+static
+cmsUInt8Number* Pack3BytesAndSkip1SwapFirstOptimized(register _cmsTRANSFORM* Info, 
+                                                     register cmsUInt16Number wOut[], 
+                                                     register cmsUInt8Number* output,
+                                                     register cmsUInt32Number Stride)
+{
+    output++;
+    *output++ = (wOut[0] & 0xFF);
+    *output++ = (wOut[1] & 0xFF);
+    *output++ = (wOut[2] & 0xFF);
+
+    return output;
+}
+
+static
+cmsUInt8Number* Pack3BytesAndSkip1Swap(register _cmsTRANSFORM* Info, 
+                                       register cmsUInt16Number wOut[], 
+                                       register cmsUInt8Number* output,
+                                       register cmsUInt32Number Stride)
+{
+    output++;
+    *output++ = FROM_16_TO_8(wOut[2]);
+    *output++ = FROM_16_TO_8(wOut[1]);
+    *output++ = FROM_16_TO_8(wOut[0]);
+
+    return output;
+}
+
+static
+cmsUInt8Number* Pack3BytesAndSkip1SwapOptimized(register _cmsTRANSFORM* Info, 
+                                                register cmsUInt16Number wOut[], 
+                                                register cmsUInt8Number* output,
+                                                register cmsUInt32Number Stride)
+{
+    output++;
+    *output++ = (wOut[2] & 0xFF);
+    *output++ = (wOut[1] & 0xFF);
+    *output++ = (wOut[0] & 0xFF);
+
+    return output;
+}
+
+
+static
+cmsUInt8Number* Pack3BytesAndSkip1SwapSwapFirst(register _cmsTRANSFORM* Info, 
+                                                register cmsUInt16Number wOut[], 
+                                                register cmsUInt8Number* output,
+                                                register cmsUInt32Number Stride)
+{       
+    *output++ = FROM_16_TO_8(wOut[2]);
+    *output++ = FROM_16_TO_8(wOut[1]);
+    *output++ = FROM_16_TO_8(wOut[0]);
+    output++;
+
+    return output;
+}
+
+static
+cmsUInt8Number* Pack3BytesAndSkip1SwapSwapFirstOptimized(register _cmsTRANSFORM* Info, 
+                                                         register cmsUInt16Number wOut[], 
+                                                         register cmsUInt8Number* output,
+                                                         register cmsUInt32Number Stride)
+{       
+    *output++ = (wOut[2] & 0xFF);
+    *output++ = (wOut[1] & 0xFF);
+    *output++ = (wOut[0] & 0xFF);
+    output++;
+
+    return output;
+}
+
+static
+cmsUInt8Number* Pack3WordsAndSkip1(register _cmsTRANSFORM* Info, 
+                                   register cmsUInt16Number wOut[], 
+                                   register cmsUInt8Number* output,
+                                   register cmsUInt32Number Stride)
+{
+    *(cmsUInt16Number*) output = wOut[0];
+    output+= 2;
+    *(cmsUInt16Number*) output = wOut[1];
+    output+= 2;
+    *(cmsUInt16Number*) output = wOut[2];
+    output+= 2;
+    output+= 2;
+
+    return output;
+}
+
+static
+cmsUInt8Number* Pack3WordsAndSkip1Swap(register _cmsTRANSFORM* Info, 
+                                       register cmsUInt16Number wOut[], 
+                                       register cmsUInt8Number* output,
+                                       register cmsUInt32Number Stride)
+{
+    output+= 2;
+    *(cmsUInt16Number*) output = wOut[2];
+    output+= 2;
+    *(cmsUInt16Number*) output = wOut[1];
+    output+= 2;
+    *(cmsUInt16Number*) output = wOut[0];
+    output+= 2;
+
+    return output;
+}
+
+
+static
+cmsUInt8Number* Pack3WordsAndSkip1SwapFirst(register _cmsTRANSFORM* Info, 
+                                            register cmsUInt16Number wOut[], 
+                                            register cmsUInt8Number* output,
+                                            register cmsUInt32Number Stride)
+{   
+    output+= 2;
+    *(cmsUInt16Number*) output = wOut[0];
+    output+= 2;
+    *(cmsUInt16Number*) output = wOut[1];
+    output+= 2;
+    *(cmsUInt16Number*) output = wOut[2];
+    output+= 2;
+
+    return output;
+}
+
+
+static
+cmsUInt8Number* Pack3WordsAndSkip1SwapSwapFirst(register _cmsTRANSFORM* Info, 
+                                                register cmsUInt16Number wOut[], 
+                                                register cmsUInt8Number* output,
+                                                register cmsUInt32Number Stride)
+{      
+    *(cmsUInt16Number*) output = wOut[2];
+    output+= 2;
+    *(cmsUInt16Number*) output = wOut[1];
+    output+= 2;
+    *(cmsUInt16Number*) output = wOut[0];
+    output+= 2;
+    output+= 2;
+
+    return output;
+}
+
+
+
+static
+cmsUInt8Number* Pack1Byte(register _cmsTRANSFORM* Info, 
+                          register cmsUInt16Number wOut[], 
+                          register cmsUInt8Number* output,
+                          register cmsUInt32Number Stride)
+{
+    *output++ = FROM_16_TO_8(wOut[0]);
+    return output;
+}
+
+
+static
+cmsUInt8Number* Pack1ByteReversed(register _cmsTRANSFORM* Info, 
+                                  register cmsUInt16Number wOut[], 
+                                  register cmsUInt8Number* output,
+                                  register cmsUInt32Number Stride)
+{
+    *output++ = FROM_16_TO_8(REVERSE_FLAVOR_16(wOut[0]));
+    return output;
+}
+
+
+static
+cmsUInt8Number* Pack1ByteSkip1(register _cmsTRANSFORM* Info, 
+                               register cmsUInt16Number wOut[], 
+                               register cmsUInt8Number* output,
+                               register cmsUInt32Number Stride)
+{
+    *output++ = FROM_16_TO_8(wOut[0]);
+    output++;
+    return output;
+}
+
+
+static
+cmsUInt8Number* Pack1ByteSkip1SwapFirst(register _cmsTRANSFORM* Info, 
+                                        register cmsUInt16Number wOut[], 
+                                        register cmsUInt8Number* output,
+                                        register cmsUInt32Number Stride)
+{
+    output++;
+    *output++ = FROM_16_TO_8(wOut[0]);
+
+    return output;
+}
+
+static
+cmsUInt8Number* Pack1Word(register _cmsTRANSFORM* Info, 
+                          register cmsUInt16Number wOut[], 
+                          register cmsUInt8Number* output,
+                          register cmsUInt32Number Stride)
+{
+    *(cmsUInt16Number*) output = wOut[0];
+    output+= 2;
+
+    return output;
+}
+
+
+static
+cmsUInt8Number* Pack1WordReversed(register _cmsTRANSFORM* Info, 
+                                  register cmsUInt16Number wOut[], 
+                                  register cmsUInt8Number* output,
+                                  register cmsUInt32Number Stride)
+{
+    *(cmsUInt16Number*) output = REVERSE_FLAVOR_16(wOut[0]);
+    output+= 2;
+
+    return output;
+}
+
+static
+cmsUInt8Number* Pack1WordBigEndian(register _cmsTRANSFORM* Info, 
+                                   register cmsUInt16Number wOut[], 
+                                   register cmsUInt8Number* output,
+                                   register cmsUInt32Number Stride)
+{
+    *(cmsUInt16Number*) output = CHANGE_ENDIAN(wOut[0]);
+    output+= 2;
+
+    return output;
+}
+
+
+static
+cmsUInt8Number* Pack1WordSkip1(register _cmsTRANSFORM* Info, 
+                               register cmsUInt16Number wOut[], 
+                               register cmsUInt8Number* output,
+                               register cmsUInt32Number Stride)
+{
+    *(cmsUInt16Number*) output = wOut[0];
+    output+= 4;
+
+    return output;
+}
+
+static
+cmsUInt8Number* Pack1WordSkip1SwapFirst(register _cmsTRANSFORM* Info, 
+                                        register cmsUInt16Number wOut[], 
+                                        register cmsUInt8Number* output,
+                                        register cmsUInt32Number Stride)
+{
+    output += 2; 
+    *(cmsUInt16Number*) output = wOut[0];
+    output+= 2;
+
+    return output;
+}
+
+
+// Unencoded Float values -- don't try optimize speed
+static
+cmsUInt8Number* PackLabDoubleFrom16(register _cmsTRANSFORM* Info, 
+                                    register cmsUInt16Number wOut[], 
+                                    register cmsUInt8Number* output, 
+                                    register cmsUInt32Number Stride)
+{
+
+    if (T_PLANAR(Info -> OutputFormat)) {
+
+        cmsCIELab  Lab;
+        cmsFloat64Number* Out = (cmsFloat64Number*) output;
+        cmsLabEncoded2Float(&Lab, wOut);
+
+        Out[0]        = Lab.L;
+        Out[Stride]   = Lab.a;
+        Out[Stride*2] = Lab.b;
+
+        return output + sizeof(cmsFloat64Number);
+    }
+    else {
+
+        cmsLabEncoded2Float((cmsCIELab*) output, wOut);
+        return output + (sizeof(cmsCIELab) + T_EXTRA(Info ->OutputFormat) * sizeof(cmsFloat64Number));            
+    }
+
+}
+
+static
+cmsUInt8Number* PackXYZDoubleFrom16(register _cmsTRANSFORM* Info, 
+                                    register cmsUInt16Number wOut[], 
+                                    register cmsUInt8Number* output,
+                                    register cmsUInt32Number Stride)
+{
+    if (T_PLANAR(Info -> OutputFormat)) {
+
+        cmsCIEXYZ XYZ;
+        cmsFloat64Number* Out = (cmsFloat64Number*) output;
+        cmsXYZEncoded2Float(&XYZ, wOut);
+
+        Out[0]                  = XYZ.X;
+        Out[Stride]   = XYZ.Y;
+        Out[Stride*2] = XYZ.Z;
+
+        return output + sizeof(cmsFloat64Number);
+
+    }
+    else {
+
+        cmsXYZEncoded2Float((cmsCIEXYZ*) output, wOut);
+
+        return output + (sizeof(cmsCIEXYZ) + T_EXTRA(Info ->OutputFormat) * sizeof(cmsFloat64Number));
+    }
+}
+
+static
+cmsUInt8Number* PackDoubleFrom16(register _cmsTRANSFORM* Info, 
+                                 register cmsUInt16Number wOut[], 
+                                 register cmsUInt8Number* output,
+                                 register cmsUInt32Number Stride)
+{
+    cmsFloat64Number* Inks = (cmsFloat64Number*) output;
+    int nChan = T_CHANNELS(Info -> OutputFormat);
+    int i;
+    cmsFloat64Number maximum = IsInkSpace(Info ->InputFormat) ? 655.35 : 65535.0;
+
+    if (T_PLANAR(Info -> OutputFormat)) {
+
+        for (i=0; i <  nChan; i++) {
+
+            Inks[i*Stride] = wOut[i] / maximum;
+        }
+
+        return output + sizeof(cmsFloat64Number);
+    } 
+    else {
+
+        for (i=0; i <  nChan; i++) {
+
+            Inks[i] = wOut[i] /  maximum;
+        }
+
+
+        return output + (nChan + T_EXTRA(Info ->OutputFormat)) * sizeof(cmsFloat64Number);
+    }
+
+}
+
+static
+cmsUInt8Number* PackFloatFrom16(register _cmsTRANSFORM* Info, 
+                                register cmsUInt16Number wOut[], 
+                                register cmsUInt8Number* output,
+                                register cmsUInt32Number Stride)
+{
+    cmsFloat32Number* Inks = (cmsFloat32Number*) output;
+    int nChan = T_CHANNELS(Info -> OutputFormat);
+    int i;
+    cmsFloat64Number maximum = IsInkSpace(Info ->OutputFormat) ? 655.35 : 65535.0;
+
+    if (T_PLANAR(Info -> OutputFormat)) {
+
+        for (i=0; i <  nChan; i++) {
+
+            Inks[i*Stride] = (cmsFloat32Number) (wOut[i] / maximum);
+        }
+
+        return output + sizeof(cmsFloat32Number);
+    } 
+    else {
+
+        for (i=0; i <  nChan; i++) {
+
+            Inks[i] = (cmsFloat32Number) (wOut[i] /  maximum);
+        }
+
+
+        return output + (nChan + T_EXTRA(Info ->OutputFormat)) * sizeof(cmsFloat32Number);
+    }
+
+}
+
+
+// --------------------------------------------------------------------------------------------------------
+
+static
+cmsUInt8Number* PackChunkyFloatsFromFloat(_cmsTRANSFORM* info, 
+                                          cmsFloat32Number wOut[], 
+                                          cmsUInt8Number* output,
+                                          cmsUInt32Number Stride)
+{
+    int nChan      = T_CHANNELS(info -> OutputFormat);
+    int DoSwap     = T_DOSWAP(info ->OutputFormat);
+    int Reverse    = T_FLAVOR(info ->OutputFormat);
+    int Extra      = T_EXTRA(info -> OutputFormat);
+    int SwapFirst  = T_SWAPFIRST(info -> OutputFormat);
+    int ExtraFirst = DoSwap && !SwapFirst;
+    cmsFloat64Number maximum = IsInkSpace(info ->OutputFormat) ? 100.0 : 1.0;
+    cmsFloat32Number* swap1;
+    cmsFloat64Number v = 0;
+    int i;
+
+    swap1 = (cmsFloat32Number*) output;
+
+    if (ExtraFirst) {
+        output += Extra * sizeof(cmsFloat32Number);
+    }
+
+    for (i=0; i < nChan; i++) {
+
+        int index = DoSwap ? (nChan - i - 1) : i;
+
+        v = wOut[index] * maximum;
+
+        if (Reverse)
+            v = maximum - v;
+
+        *(cmsFloat32Number*) output = (cmsFloat32Number) v;
+
+        output += sizeof(cmsFloat32Number);
+    }
+
+    if (!ExtraFirst) {
+        output += Extra * sizeof(cmsFloat32Number);
+    }
+
+    if (Extra == 0 && SwapFirst) {
+
+        memmove(swap1 + 1, swap1, (nChan-1)* sizeof(cmsFloat32Number));
+        *swap1 = (cmsFloat32Number) v;
+    }
+
+
+    return output;
+}
+
+static
+cmsUInt8Number* PackPlanarFloatsFromFloat(_cmsTRANSFORM* info, 
+                                          cmsFloat32Number wOut[], 
+                                          cmsUInt8Number* output,
+                                          cmsUInt32Number Stride)
+{
+    int nChan = T_CHANNELS(info -> OutputFormat);
+    int DoSwap = T_DOSWAP(info ->OutputFormat);
+    int Reverse= T_FLAVOR(info ->OutputFormat);
+    int i;
+    cmsUInt8Number* Init = output;
+    cmsFloat64Number maximum = IsInkSpace(info ->OutputFormat) ? 100.0 : 1.0;
+    cmsFloat64Number v;
+
+    if (DoSwap) {
+        output += T_EXTRA(info -> OutputFormat) * Stride * sizeof(cmsFloat32Number);
+    }
+
+    for (i=0; i < nChan; i++) {
+
+        int index = DoSwap ? (nChan - i - 1) : i;
+
+        v = wOut[index] * maximum;
+
+        if (Reverse) 
+            v =  maximum - v;
+
+        *(cmsFloat32Number*) output = (cmsFloat32Number) v;
+        output += (Stride * sizeof(cmsFloat32Number));
+    }
+
+    return (Init + sizeof(cmsFloat32Number));
+}
+
+
+static
+cmsUInt8Number* PackChunkyDoublesFromFloat(_cmsTRANSFORM* info, 
+                                           cmsFloat32Number wOut[], 
+                                           cmsUInt8Number* output,
+                                           cmsUInt32Number Stride)
+{
+    int nChan      = T_CHANNELS(info -> OutputFormat);
+    int DoSwap     = T_DOSWAP(info ->OutputFormat);
+    int Reverse    = T_FLAVOR(info ->OutputFormat);
+    int Extra      = T_EXTRA(info -> OutputFormat);
+    int SwapFirst  = T_SWAPFIRST(info -> OutputFormat);
+    int ExtraFirst = DoSwap && !SwapFirst;
+    cmsFloat64Number* swap1;
+    cmsFloat64Number maximum = IsInkSpace(info ->OutputFormat) ? 100.0 : 1.0;
+    cmsFloat64Number v = 0;
+    int i;
+
+    swap1 = (cmsFloat64Number*) output;
+
+    if (ExtraFirst) {
+        output += Extra * sizeof(cmsFloat64Number);
+    }
+
+    for (i=0; i < nChan; i++) {
+
+        int index = DoSwap ? (nChan - i - 1) : i;
+
+        v = (cmsFloat64Number) wOut[index] * maximum;
+
+        if (Reverse)
+            v = maximum - v;
+
+        *(cmsFloat64Number*) output = v;
+
+        output += sizeof(cmsFloat64Number);
+    }
+
+    if (!ExtraFirst) {
+        output += Extra * sizeof(cmsFloat64Number);
+    }
+
+    if (Extra == 0 && SwapFirst) {
+
+        memmove(swap1 + 1, swap1, (nChan-1)* sizeof(cmsFloat64Number));
+        *swap1 = v;
+    }
+
+
+    return output;
+}
+
+static
+cmsUInt8Number* PackPlanarDoublesFromFloat(_cmsTRANSFORM* info, 
+                                           cmsFloat32Number wOut[], 
+                                           cmsUInt8Number* output,
+                                           cmsUInt32Number Stride)
+{
+    int nChan = T_CHANNELS(info -> OutputFormat);
+    int DoSwap = T_DOSWAP(info ->OutputFormat);
+    int Reverse= T_FLAVOR(info ->OutputFormat);
+    int i;
+    cmsUInt8Number* Init = output;
+    cmsFloat64Number maximum = IsInkSpace(info ->OutputFormat) ? 100.0 : 1.0;
+    cmsFloat64Number v;
+
+    if (DoSwap) {
+        output += T_EXTRA(info -> OutputFormat) * Stride * sizeof(cmsFloat64Number);
+    }
+
+    for (i=0; i < nChan; i++) {
+
+        int index = DoSwap ? (nChan - i - 1) : i;
+
+        v = (cmsFloat64Number) wOut[index] * maximum;
+
+        if (Reverse) 
+            v =  maximum - v;
+
+        *(cmsFloat64Number*) output = v;
+        output += (Stride * sizeof(cmsFloat64Number));
+    }
+
+    return (Init + sizeof(cmsFloat64Number));
+}
+
+
+
+
+static
+cmsUInt8Number* PackLabFloatFromFloat(_cmsTRANSFORM* Info, 
+                                      cmsFloat32Number wOut[], 
+                                      cmsUInt8Number* output,
+                                      cmsUInt32Number Stride)
+{        
+    cmsFloat32Number* Out = (cmsFloat32Number*) output;
+
+    if (T_PLANAR(Info -> OutputFormat)) {
+
+        Out[0]        = (cmsFloat32Number) (wOut[0] * 100.0);
+        Out[Stride]   = (cmsFloat32Number) (wOut[1] * 255.0 - 128.0);
+        Out[Stride*2] = (cmsFloat32Number) (wOut[2] * 255.0 - 128.0);
+
+        return output + sizeof(cmsFloat32Number);
+    }
+    else {
+
+        Out[0] = (cmsFloat32Number) (wOut[0] * 100.0);
+        Out[1] = (cmsFloat32Number) (wOut[1] * 255.0 - 128.0);
+        Out[2] = (cmsFloat32Number) (wOut[2] * 255.0 - 128.0);
+
+        return output + (sizeof(cmsFloat32Number)*3 + T_EXTRA(Info ->OutputFormat) * sizeof(cmsFloat32Number));            
+    }
+
+}
+
+static
+cmsUInt8Number* PackLabDoubleFromFloat(_cmsTRANSFORM* Info, 
+                                       cmsFloat32Number wOut[], 
+                                       cmsUInt8Number* output,
+                                       cmsUInt32Number Stride)
+{        
+    cmsFloat64Number* Out = (cmsFloat64Number*) output;
+
+    if (T_PLANAR(Info -> OutputFormat)) {
+
+        Out[0]        = (cmsFloat64Number) (wOut[0] * 100.0);
+        Out[Stride]   = (cmsFloat64Number) (wOut[1] * 255.0 - 128.0);
+        Out[Stride*2] = (cmsFloat64Number) (wOut[2] * 255.0 - 128.0);
+
+        return output + sizeof(cmsFloat64Number);
+    }
+    else {
+
+        Out[0] = (cmsFloat64Number) (wOut[0] * 100.0);
+        Out[1] = (cmsFloat64Number) (wOut[1] * 255.0 - 128.0);
+        Out[2] = (cmsFloat64Number) (wOut[2] * 255.0 - 128.0);
+
+        return output + (sizeof(cmsFloat64Number)*3 + T_EXTRA(Info ->OutputFormat) * sizeof(cmsFloat64Number));            
+    }
+
+}
+
+
+// From 0..1 range to 0..MAX_ENCODEABLE_XYZ
+static
+cmsUInt8Number* PackXYZFloatFromFloat(_cmsTRANSFORM* Info, 
+                                      cmsFloat32Number wOut[], 
+                                      cmsUInt8Number* output,
+                                      cmsUInt32Number Stride)
+{        
+    cmsFloat32Number* Out = (cmsFloat32Number*) output;
+
+    if (T_PLANAR(Info -> OutputFormat)) {
+
+        Out[0]        = (cmsFloat32Number) (wOut[0] * MAX_ENCODEABLE_XYZ);
+        Out[Stride]   = (cmsFloat32Number) (wOut[1] * MAX_ENCODEABLE_XYZ);
+        Out[Stride*2] = (cmsFloat32Number) (wOut[2] * MAX_ENCODEABLE_XYZ);
+
+        return output + sizeof(cmsFloat32Number);
+    }
+    else {
+
+        Out[0] = (cmsFloat32Number) (wOut[0] * MAX_ENCODEABLE_XYZ);
+        Out[1] = (cmsFloat32Number) (wOut[1] * MAX_ENCODEABLE_XYZ);
+        Out[2] = (cmsFloat32Number) (wOut[2] * MAX_ENCODEABLE_XYZ);
+
+        return output + (sizeof(cmsFloat32Number)*3 + T_EXTRA(Info ->OutputFormat) * sizeof(cmsFloat32Number));            
+    }
+
+}
+
+
+// Same, but convert to double
+static
+cmsUInt8Number* PackXYZDoubleFromFloat(_cmsTRANSFORM* Info, 
+                                       cmsFloat32Number wOut[], 
+                                       cmsUInt8Number* output,
+                                       cmsUInt32Number Stride)
+{        
+    cmsFloat64Number* Out = (cmsFloat64Number*) output;
+
+    if (T_PLANAR(Info -> OutputFormat)) {
+
+        Out[0]        = (cmsFloat64Number) (wOut[0] * MAX_ENCODEABLE_XYZ);
+        Out[Stride]   = (cmsFloat64Number) (wOut[1] * MAX_ENCODEABLE_XYZ);
+        Out[Stride*2] = (cmsFloat64Number) (wOut[2] * MAX_ENCODEABLE_XYZ);
+
+        return output + sizeof(cmsFloat64Number);
+    }
+    else {
+
+        Out[0] = (cmsFloat64Number) (wOut[0] * MAX_ENCODEABLE_XYZ);
+        Out[1] = (cmsFloat64Number) (wOut[1] * MAX_ENCODEABLE_XYZ);
+        Out[2] = (cmsFloat64Number) (wOut[2] * MAX_ENCODEABLE_XYZ);
+
+        return output + (sizeof(cmsFloat64Number)*3 + T_EXTRA(Info ->OutputFormat) * sizeof(cmsFloat64Number));            
+    }
+
+}
+
+
+// ----------------------------------------------------------------------------------------------------------------
+
+
+static cmsFormatters16 InputFormatters16[] = {
+
+    //    Type                                          Mask                  Function
+    //  ----------------------------   ------------------------------------  ----------------------------
+    { TYPE_Lab_DBL,                                 ANYPLANAR|ANYEXTRA,   UnrollLabDoubleTo16},
+    { TYPE_XYZ_DBL,                                 ANYPLANAR|ANYEXTRA,   UnrollXYZDoubleTo16},
+    { TYPE_GRAY_DBL,                                                 0,   UnrollDouble1Chan},
+    { FLOAT_SH(1)|BYTES_SH(0), ANYCHANNELS|ANYPLANAR|ANYEXTRA|ANYSPACE,   UnrollDoubleTo16},
+    { FLOAT_SH(1)|BYTES_SH(4), ANYCHANNELS|ANYPLANAR|ANYEXTRA|ANYSPACE,   UnrollFloatTo16},
+
+
+    { CHANNELS_SH(1)|BYTES_SH(1),                              ANYSPACE,  Unroll1Byte}, 
+    { CHANNELS_SH(1)|BYTES_SH(1)|EXTRA_SH(1),                  ANYSPACE,  Unroll1ByteSkip1},
+    { CHANNELS_SH(1)|BYTES_SH(1)|EXTRA_SH(2),                  ANYSPACE,  Unroll1ByteSkip2},
+    { CHANNELS_SH(1)|BYTES_SH(1)|FLAVOR_SH(1),                 ANYSPACE,  Unroll1ByteReversed},
+    { COLORSPACE_SH(PT_MCH2)|CHANNELS_SH(2)|BYTES_SH(1),              0,  Unroll2Bytes},
+
+    { TYPE_LabV2_8,                                                   0,  UnrollLabV2_8 },
+    { TYPE_ALabV2_8,                                                  0,  UnrollALabV2_8 },
+    { TYPE_LabV2_16,                                                  0,  UnrollLabV2_16 },
+
+    { CHANNELS_SH(3)|BYTES_SH(1),                              ANYSPACE,  Unroll3Bytes},
+    { CHANNELS_SH(3)|BYTES_SH(1)|DOSWAP_SH(1),                 ANYSPACE,  Unroll3BytesSwap},
+    { CHANNELS_SH(3)|EXTRA_SH(1)|BYTES_SH(1)|DOSWAP_SH(1),     ANYSPACE,  Unroll3BytesSkip1Swap},
+    { CHANNELS_SH(3)|EXTRA_SH(1)|BYTES_SH(1)|SWAPFIRST_SH(1),  ANYSPACE,  Unroll3BytesSkip1SwapFirst},
+
+    { CHANNELS_SH(4)|BYTES_SH(1),                              ANYSPACE,  Unroll4Bytes},
+    { CHANNELS_SH(4)|BYTES_SH(1)|FLAVOR_SH(1),                 ANYSPACE,  Unroll4BytesReverse},
+    { CHANNELS_SH(4)|BYTES_SH(1)|SWAPFIRST_SH(1),              ANYSPACE,  Unroll4BytesSwapFirst}, 
+    { CHANNELS_SH(4)|BYTES_SH(1)|DOSWAP_SH(1),                 ANYSPACE,  Unroll4BytesSwap}, 
+    { CHANNELS_SH(4)|BYTES_SH(1)|DOSWAP_SH(1)|SWAPFIRST_SH(1), ANYSPACE,  Unroll4BytesSwapSwapFirst}, 
+
+    { BYTES_SH(1)|PLANAR_SH(1),    ANYFLAVOR|ANYSWAP|ANYEXTRA|ANYCHANNELS|ANYSPACE, UnrollPlanarBytes},
+    { BYTES_SH(1),    ANYFLAVOR|ANYSWAPFIRST|ANYSWAP|ANYEXTRA|ANYCHANNELS|ANYSPACE, UnrollChunkyBytes},     
+
+
+    { CHANNELS_SH(1)|BYTES_SH(2),                              ANYSPACE,  Unroll1Word},
+    { CHANNELS_SH(1)|BYTES_SH(2)|FLAVOR_SH(1),                 ANYSPACE,  Unroll1WordReversed},
+    { CHANNELS_SH(1)|BYTES_SH(2)|EXTRA_SH(3),                  ANYSPACE,  Unroll1WordSkip3},
+
+    { CHANNELS_SH(2)|BYTES_SH(2),                              ANYSPACE,  Unroll2Words},
+    { CHANNELS_SH(3)|BYTES_SH(2),                              ANYSPACE,  Unroll3Words},
+    { CHANNELS_SH(4)|BYTES_SH(2),                              ANYSPACE,  Unroll4Words},
+
+    { CHANNELS_SH(3)|BYTES_SH(2)|DOSWAP_SH(1),                 ANYSPACE,  Unroll3WordsSwap},
+    { CHANNELS_SH(3)|BYTES_SH(2)|EXTRA_SH(1)|SWAPFIRST_SH(1),  ANYSPACE,  Unroll3WordsSkip1SwapFirst},
+    { CHANNELS_SH(3)|BYTES_SH(2)|EXTRA_SH(1)|DOSWAP_SH(1),     ANYSPACE,  Unroll3WordsSkip1Swap},
+    { CHANNELS_SH(4)|BYTES_SH(2)|FLAVOR_SH(1),                 ANYSPACE,  Unroll4WordsReverse},
+    { CHANNELS_SH(4)|BYTES_SH(2)|SWAPFIRST_SH(1),              ANYSPACE,  Unroll4WordsSwapFirst}, 
+    { CHANNELS_SH(4)|BYTES_SH(2)|DOSWAP_SH(1),                 ANYSPACE,  Unroll4WordsSwap}, 
+    { CHANNELS_SH(4)|BYTES_SH(2)|DOSWAP_SH(1)|SWAPFIRST_SH(1), ANYSPACE,  Unroll4WordsSwapSwapFirst}, 
+
+
+    { BYTES_SH(2)|PLANAR_SH(1),  ANYFLAVOR|ANYSWAP|ANYENDIAN|ANYEXTRA|ANYCHANNELS|ANYSPACE,  UnrollPlanarWords },
+    { BYTES_SH(2),  ANYFLAVOR|ANYSWAPFIRST|ANYSWAP|ANYENDIAN|ANYEXTRA|ANYCHANNELS|ANYSPACE,  UnrollAnyWords}, 
+};
+
+
+
+static cmsFormattersFloat InputFormattersFloat[] = {
+
+    //    Type                                          Mask                  Function
+    //  ----------------------------   ------------------------------------  ----------------------------
+    {     TYPE_Lab_DBL,                                ANYPLANAR|ANYEXTRA,   UnrollLabDoubleToFloat},
+    {     TYPE_Lab_FLT,                                ANYPLANAR|ANYEXTRA,   UnrollLabFloatToFloat},
+    {     TYPE_XYZ_DBL,                                ANYPLANAR|ANYEXTRA,   UnrollXYZDoubleToFloat},
+    {     TYPE_XYZ_FLT,                                ANYPLANAR|ANYEXTRA,   UnrollXYZFloatToFloat},
+
+    {     FLOAT_SH(1)|BYTES_SH(4), ANYPLANAR|ANYEXTRA|ANYCHANNELS|ANYSPACE,  UnrollFloatsToFloat},
+    {     FLOAT_SH(1)|BYTES_SH(0), ANYPLANAR|ANYEXTRA|ANYCHANNELS|ANYSPACE,  UnrollDoublesToFloat},
+};
+
+
+// Bit fields set to one in the mask are not compared
+static
+cmsFormatter _cmsGetStockInputFormatter(cmsUInt32Number dwInput, cmsUInt32Number dwFlags)
+{
+    cmsUInt32Number i;
+    cmsFormatter fr;
+
+
+    if (!(dwFlags & CMS_PACK_FLAGS_FLOAT)) {
+
+        for (i=0; i < sizeof(InputFormatters16) / sizeof(cmsFormatters16); i++) {
+            cmsFormatters16* f = InputFormatters16 + i;
+
+            if ((dwInput & ~f ->Mask) == f ->Type) {
+                fr.Fmt16 = f ->Frm;
+                return fr;
+            }
+        }
+    }
+    else {
+        for (i=0; i < sizeof(InputFormattersFloat) / sizeof(cmsFormattersFloat); i++) {
+            cmsFormattersFloat* f = InputFormattersFloat + i;
+
+            if ((dwInput & ~f ->Mask) == f ->Type) {
+                fr.FmtFloat = f ->Frm;
+                return fr;
+            }
+        }
+    }
+
+    fr.Fmt16 = NULL;
+    return fr;
+}
+
+static cmsFormatters16 OutputFormatters16[] = {
+    //    Type                                          Mask                  Function
+    //  ----------------------------   ------------------------------------  ----------------------------
+
+    { TYPE_Lab_DBL,                                      ANYPLANAR|ANYEXTRA,  PackLabDoubleFrom16},
+    { TYPE_XYZ_DBL,                                      ANYPLANAR|ANYEXTRA,  PackXYZDoubleFrom16},
+    { FLOAT_SH(1)|BYTES_SH(0),      ANYCHANNELS|ANYPLANAR|ANYEXTRA|ANYSPACE,  PackDoubleFrom16},
+    { FLOAT_SH(1)|BYTES_SH(4),      ANYCHANNELS|ANYPLANAR|ANYEXTRA|ANYSPACE,  PackFloatFrom16},
+
+    { CHANNELS_SH(1)|BYTES_SH(1),                                  ANYSPACE,  Pack1Byte},   
+    { CHANNELS_SH(1)|BYTES_SH(1)|EXTRA_SH(1),                      ANYSPACE,  Pack1ByteSkip1},
+    { CHANNELS_SH(1)|BYTES_SH(1)|EXTRA_SH(1)|SWAPFIRST_SH(1),      ANYSPACE,  Pack1ByteSkip1SwapFirst},
+
+    { CHANNELS_SH(1)|BYTES_SH(1)|FLAVOR_SH(1),                     ANYSPACE,  Pack1ByteReversed},
+
+    { TYPE_LabV2_8,                                                       0,  PackLabV2_8 },
+    { TYPE_ALabV2_8,                                                      0,  PackALabV2_8 },
+    { TYPE_LabV2_16,                                                      0,  PackLabV2_16 },
+
+    { CHANNELS_SH(3)|BYTES_SH(1)|OPTIMIZED_SH(1),                  ANYSPACE,  Pack3BytesOptimized},
+    { CHANNELS_SH(3)|BYTES_SH(1)|EXTRA_SH(1)|OPTIMIZED_SH(1),      ANYSPACE,  Pack3BytesAndSkip1Optimized},
+    { CHANNELS_SH(3)|BYTES_SH(1)|EXTRA_SH(1)|SWAPFIRST_SH(1)|OPTIMIZED_SH(1),
+                                                                   ANYSPACE,  Pack3BytesAndSkip1SwapFirstOptimized},
+    { CHANNELS_SH(3)|BYTES_SH(1)|EXTRA_SH(1)|DOSWAP_SH(1)|SWAPFIRST_SH(1)|OPTIMIZED_SH(1),  
+                                                                   ANYSPACE,  Pack3BytesAndSkip1SwapSwapFirstOptimized},
+    { CHANNELS_SH(3)|BYTES_SH(1)|DOSWAP_SH(1)|EXTRA_SH(1)|OPTIMIZED_SH(1),         
+                                                                   ANYSPACE,  Pack3BytesAndSkip1SwapOptimized},
+    { CHANNELS_SH(3)|BYTES_SH(1)|DOSWAP_SH(1)|OPTIMIZED_SH(1),     ANYSPACE,  Pack3BytesSwapOptimized},
+
+
+
+    { CHANNELS_SH(3)|BYTES_SH(1),                                  ANYSPACE,  Pack3Bytes},
+    { CHANNELS_SH(3)|BYTES_SH(1)|EXTRA_SH(1),                      ANYSPACE,  Pack3BytesAndSkip1},
+    { CHANNELS_SH(3)|BYTES_SH(1)|EXTRA_SH(1)|SWAPFIRST_SH(1),      ANYSPACE,  Pack3BytesAndSkip1SwapFirst},
+    { CHANNELS_SH(3)|BYTES_SH(1)|EXTRA_SH(1)|DOSWAP_SH(1)|SWAPFIRST_SH(1),  
+                                                                   ANYSPACE,  Pack3BytesAndSkip1SwapSwapFirst},
+    { CHANNELS_SH(3)|BYTES_SH(1)|DOSWAP_SH(1)|EXTRA_SH(1),         ANYSPACE,  Pack3BytesAndSkip1Swap},
+    { CHANNELS_SH(3)|BYTES_SH(1)|DOSWAP_SH(1),                     ANYSPACE,  Pack3BytesSwap},
+    { CHANNELS_SH(6)|BYTES_SH(1),                                  ANYSPACE,  Pack6Bytes},
+    { CHANNELS_SH(6)|BYTES_SH(1)|DOSWAP_SH(1),                     ANYSPACE,  Pack6BytesSwap},
+    { CHANNELS_SH(4)|BYTES_SH(1),                                  ANYSPACE,  Pack4Bytes},
+    { CHANNELS_SH(4)|BYTES_SH(1)|FLAVOR_SH(1),                     ANYSPACE,  Pack4BytesReverse},
+    { CHANNELS_SH(4)|BYTES_SH(1)|SWAPFIRST_SH(1),                  ANYSPACE,  Pack4BytesSwapFirst}, 
+    { CHANNELS_SH(4)|BYTES_SH(1)|DOSWAP_SH(1),                     ANYSPACE,  Pack4BytesSwap}, 
+    { CHANNELS_SH(4)|BYTES_SH(1)|DOSWAP_SH(1)|SWAPFIRST_SH(1),     ANYSPACE,  Pack4BytesSwapSwapFirst}, 
+
+    { BYTES_SH(1),                 ANYFLAVOR|ANYSWAPFIRST|ANYSWAP|ANYEXTRA|ANYCHANNELS|ANYSPACE, PackAnyBytes},     
+    { BYTES_SH(1)|PLANAR_SH(1),    ANYFLAVOR|ANYSWAP|ANYEXTRA|ANYCHANNELS|ANYSPACE, PackPlanarBytes},   
+
+    { CHANNELS_SH(1)|BYTES_SH(2),                                  ANYSPACE,  Pack1Word},
+    { CHANNELS_SH(1)|BYTES_SH(2)|EXTRA_SH(1),                      ANYSPACE,  Pack1WordSkip1},
+    { CHANNELS_SH(1)|BYTES_SH(2)|EXTRA_SH(1)|SWAPFIRST_SH(1),      ANYSPACE,  Pack1WordSkip1SwapFirst},
+    { CHANNELS_SH(1)|BYTES_SH(2)|FLAVOR_SH(1),                     ANYSPACE,  Pack1WordReversed},
+    { CHANNELS_SH(1)|BYTES_SH(2)|ENDIAN16_SH(1),                   ANYSPACE,  Pack1WordBigEndian},
+    { CHANNELS_SH(3)|BYTES_SH(2),                                  ANYSPACE,  Pack3Words},
+    { CHANNELS_SH(3)|BYTES_SH(2)|DOSWAP_SH(1),                     ANYSPACE,  Pack3WordsSwap},
+    { CHANNELS_SH(3)|BYTES_SH(2)|ENDIAN16_SH(1),                   ANYSPACE,  Pack3WordsBigEndian},
+    { CHANNELS_SH(3)|BYTES_SH(2)|EXTRA_SH(1),                      ANYSPACE,  Pack3WordsAndSkip1},
+    { CHANNELS_SH(3)|BYTES_SH(2)|EXTRA_SH(1)|DOSWAP_SH(1),         ANYSPACE,  Pack3WordsAndSkip1Swap},
+    { CHANNELS_SH(3)|BYTES_SH(2)|EXTRA_SH(1)|SWAPFIRST_SH(1),      ANYSPACE,  Pack3WordsAndSkip1SwapFirst},
+
+    { CHANNELS_SH(3)|BYTES_SH(2)|EXTRA_SH(1)|DOSWAP_SH(1)|SWAPFIRST_SH(1),             
+                                                                   ANYSPACE,  Pack3WordsAndSkip1SwapSwapFirst},
+
+    { CHANNELS_SH(4)|BYTES_SH(2),                                  ANYSPACE,  Pack4Words},
+    { CHANNELS_SH(4)|BYTES_SH(2)|FLAVOR_SH(1),                     ANYSPACE,  Pack4WordsReverse},
+    { CHANNELS_SH(4)|BYTES_SH(2)|DOSWAP_SH(1),                     ANYSPACE,  Pack4WordsSwap}, 
+    { CHANNELS_SH(4)|BYTES_SH(2)|ENDIAN16_SH(1),                   ANYSPACE,  Pack4WordsBigEndian},
+
+    { CHANNELS_SH(6)|BYTES_SH(2),                                  ANYSPACE,  Pack6Words},
+    { CHANNELS_SH(6)|BYTES_SH(2)|DOSWAP_SH(1),                     ANYSPACE,  Pack6WordsSwap},
+
+    { BYTES_SH(2)|PLANAR_SH(1),     ANYFLAVOR|ANYENDIAN|ANYSWAP|ANYEXTRA|ANYCHANNELS|ANYSPACE, PackPlanarWords}, 
+    { BYTES_SH(2),                  ANYFLAVOR|ANYSWAPFIRST|ANYSWAP|ANYENDIAN|ANYEXTRA|ANYCHANNELS|ANYSPACE, PackAnyWords}
+
+};
+
+
+static cmsFormattersFloat OutputFormattersFloat[] = {
+    //    Type                                          Mask                                 Function
+    //  ----------------------------   ---------------------------------------------------  ----------------------------
+    {     TYPE_Lab_FLT,                                                ANYPLANAR|ANYEXTRA,   PackLabFloatFromFloat},
+    {     TYPE_XYZ_FLT,                                                ANYPLANAR|ANYEXTRA,   PackXYZFloatFromFloat},
+    {     TYPE_Lab_DBL,                                                ANYPLANAR|ANYEXTRA,   PackLabDoubleFromFloat},
+    {     TYPE_XYZ_DBL,                                                ANYPLANAR|ANYEXTRA,   PackXYZDoubleFromFloat},
+    {     FLOAT_SH(1)|BYTES_SH(4), 
+                             ANYFLAVOR|ANYSWAPFIRST|ANYSWAP|ANYEXTRA|ANYCHANNELS|ANYSPACE,   PackChunkyFloatsFromFloat }, 
+    {     FLOAT_SH(1)|BYTES_SH(4)|PLANAR_SH(1),             ANYEXTRA|ANYCHANNELS|ANYSPACE,   PackPlanarFloatsFromFloat},
+    {     FLOAT_SH(1)|BYTES_SH(0),
+                             ANYFLAVOR|ANYSWAPFIRST|ANYSWAP|ANYEXTRA|ANYCHANNELS|ANYSPACE,   PackChunkyDoublesFromFloat }, 
+    {     FLOAT_SH(1)|BYTES_SH(0)|PLANAR_SH(1),             ANYEXTRA|ANYCHANNELS|ANYSPACE,   PackPlanarDoublesFromFloat},
+
+
+};
+
+
+// Bit fields set to one in the mask are not compared
+cmsFormatter _cmsGetStockOutputFormatter(cmsUInt32Number dwInput, cmsUInt32Number dwFlags)
+{
+    cmsUInt32Number i;
+    cmsFormatter fr;
+
+
+    if (dwFlags & CMS_PACK_FLAGS_FLOAT) {
+
+        for (i=0; i < sizeof(OutputFormattersFloat) / sizeof(cmsFormattersFloat); i++) {
+            cmsFormattersFloat* f = OutputFormattersFloat + i;
+
+            if ((dwInput & ~f ->Mask) == f ->Type) {
+                fr.FmtFloat = f ->Frm;
+                return fr;
+            }
+        }
+
+    }
+    else {
+
+        for (i=0; i < sizeof(OutputFormatters16) / sizeof(cmsFormatters16); i++) {
+            cmsFormatters16* f = OutputFormatters16 + i;
+
+            if ((dwInput & ~f ->Mask) == f ->Type) {
+                fr.Fmt16 = f ->Frm;
+                return fr;
+            }
+        }
+    }
+
+    fr.Fmt16 = NULL;
+    return fr;
+}
+
+
+typedef struct _cms_formatters_factory_list {
+
+    cmsFormatterFactory Factory;
+    struct _cms_formatters_factory_list *Next;
+
+} cmsFormattersFactoryList;
+
+static cmsFormattersFactoryList* FactoryList = NULL;
+
+
+// Formatters management
+cmsBool  _cmsRegisterFormattersPlugin(cmsPluginBase* Data)
+{
+    cmsPluginFormatters* Plugin = (cmsPluginFormatters*) Data;
+    cmsFormattersFactoryList* fl ;
+
+    // Reset
+    if (Data == NULL) {
+
+          FactoryList = NULL;
+          return TRUE;
+    }
+    
+    fl = (cmsFormattersFactoryList*) _cmsPluginMalloc(sizeof(cmsFormattersFactoryList));
+    if (fl == NULL) return FALSE;
+
+    fl ->Factory    = Plugin ->FormattersFactory;
+
+    fl ->Next = FactoryList;
+    FactoryList = fl;
+
+    return TRUE;
+}
+
+cmsFormatter _cmsGetFormatter(cmsUInt32Number Type,         // Specific type, i.e. TYPE_RGB_8
+                             cmsFormatterDirection Dir, 
+                             cmsUInt32Number dwFlags)       // Float or 16 bits
+{
+    cmsFormattersFactoryList* f;
+
+    for (f = FactoryList; f != NULL; f = f ->Next) {
+
+        cmsFormatter fn = f ->Factory(Type, Dir, dwFlags);
+        if (fn.Fmt16 != NULL) return fn;
+    }
+
+    // Revert to default
+    if (Dir == cmsFormatterInput) 
+        return _cmsGetStockInputFormatter(Type, dwFlags);
+    else 
+        return _cmsGetStockOutputFormatter(Type, dwFlags);
+}
+
+
+// Return whatever given formatter refers to float values
+cmsBool  _cmsFormatterIsFloat(cmsUInt32Number Type)
+{
+    return T_FLOAT(Type) ? TRUE : FALSE;
+}
+
+// Return whatever given formatter refers to 8 bits
+cmsBool  _cmsFormatterIs8bit(cmsUInt32Number Type)
+{
+    int Bytes = T_BYTES(Type);
+
+    return (Bytes == 1);
+}
+
+// Build a suitable formatter for the colorspace of this profile
+cmsUInt32Number CMSEXPORT cmsFormatterForColorspaceOfProfile(cmsHPROFILE hProfile, cmsUInt32Number nBytes, cmsBool lIsFloat)
+{
+ 
+    cmsColorSpaceSignature ColorSpace      = cmsGetColorSpace(hProfile);
+    cmsUInt32Number        ColorSpaceBits  = _cmsLCMScolorSpace(ColorSpace);
+    cmsUInt32Number        nOutputChans    = cmsChannelsOf(ColorSpace);
+    cmsUInt32Number        Float           = lIsFloat ? 1 : 0;
+
+    // Create a fake formatter for result
+    return FLOAT_SH(Float) | COLORSPACE_SH(ColorSpaceBits) | BYTES_SH(nBytes) | CHANNELS_SH(nOutputChans);
+}
+
+// Build a suitable formatter for the colorspace of this profile
+cmsUInt32Number CMSEXPORT cmsFormatterForPCSOfProfile(cmsHPROFILE hProfile, cmsUInt32Number nBytes, cmsBool lIsFloat)
+{
+ 
+    cmsColorSpaceSignature ColorSpace      = cmsGetPCS(hProfile);
+    int                    ColorSpaceBits  = _cmsLCMScolorSpace(ColorSpace);
+    cmsUInt32Number        nOutputChans    = cmsChannelsOf(ColorSpace);
+    cmsUInt32Number        Float           = lIsFloat ? 1 : 0;
+
+    // Create a fake formatter for result
+    return FLOAT_SH(Float) | COLORSPACE_SH(ColorSpaceBits) | BYTES_SH(nBytes) | CHANNELS_SH(nOutputChans);
+}
+
diff --git a/thirdparty/liblcms2/src/cmspcs.c b/thirdparty/liblcms2/src/cmspcs.c
new file mode 100644
index 00000000..8dd1c22c
--- /dev/null
+++ b/thirdparty/liblcms2/src/cmspcs.c
@@ -0,0 +1,926 @@
+//---------------------------------------------------------------------------------
+//
+//  Little Color Management System
+//  Copyright (c) 1998-2010 Marti Maria Saguer
+//
+// Permission is hereby granted, free of charge, to any person obtaining 
+// a copy of this software and associated documentation files (the "Software"), 
+// to deal in the Software without restriction, including without limitation 
+// the rights to use, copy, modify, merge, publish, distribute, sublicense, 
+// and/or sell copies of the Software, and to permit persons to whom the Software 
+// is furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in 
+// all copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 
+// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO 
+// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE 
+// LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 
+// OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 
+// WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+//
+//---------------------------------------------------------------------------------
+//
+
+#include "lcms2_internal.h"
+
+//      inter PCS conversions XYZ <-> CIE L* a* b*
+/*
+
+
+       CIE 15:2004 CIELab is defined as:
+
+       L* = 116*f(Y/Yn) - 16                     0 <= L* <= 100
+       a* = 500*[f(X/Xn) - f(Y/Yn)]
+       b* = 200*[f(Y/Yn) - f(Z/Zn)]
+
+       and
+
+              f(t) = t^(1/3)                     1 >= t >  (24/116)^3
+                     (841/108)*t + (16/116)      0 <= t <= (24/116)^3
+
+
+       Reverse transform is:
+
+       X = Xn*[a* / 500 + (L* + 16) / 116] ^ 3   if (X/Xn) > (24/116)
+         = Xn*(a* / 500 + L* / 116) / 7.787      if (X/Xn) <= (24/116)
+
+
+
+       PCS in Lab2 is encoded as:
+
+              8 bit Lab PCS:
+
+                     L*      0..100 into a 0..ff byte.
+                     a*      t + 128 range is -128.0  +127.0
+                     b*
+
+             16 bit Lab PCS:
+
+                     L*     0..100  into a 0..ff00 word.
+                     a*     t + 128  range is  -128.0  +127.9961
+                     b*
+
+
+
+Interchange Space   Component     Actual Range        Encoded Range
+CIE XYZ             X             0 -> 1.99997        0x0000 -> 0xffff
+CIE XYZ             Y             0 -> 1.99997        0x0000 -> 0xffff
+CIE XYZ             Z             0 -> 1.99997        0x0000 -> 0xffff
+
+Version 2,3
+-----------
+
+CIELAB (16 bit)     L*            0 -> 100.0          0x0000 -> 0xff00
+CIELAB (16 bit)     a*            -128.0 -> +127.996  0x0000 -> 0x8000 -> 0xffff
+CIELAB (16 bit)     b*            -128.0 -> +127.996  0x0000 -> 0x8000 -> 0xffff
+
+
+Version 4
+---------
+
+CIELAB (16 bit)     L*            0 -> 100.0          0x0000 -> 0xffff
+CIELAB (16 bit)     a*            -128.0 -> +127      0x0000 -> 0x8080 -> 0xffff
+CIELAB (16 bit)     b*            -128.0 -> +127      0x0000 -> 0x8080 -> 0xffff
+
+*/
+
+// Conversions
+void CMSEXPORT cmsXYZ2xyY(cmsCIExyY* Dest, const cmsCIEXYZ* Source)
+{
+    cmsFloat64Number ISum;
+
+    ISum = 1./(Source -> X + Source -> Y + Source -> Z);
+
+    Dest -> x = (Source -> X) * ISum;
+    Dest -> y = (Source -> Y) * ISum;
+    Dest -> Y = Source -> Y;
+}
+
+void CMSEXPORT cmsxyY2XYZ(cmsCIEXYZ* Dest, const cmsCIExyY* Source)
+{
+    Dest -> X = (Source -> x / Source -> y) * Source -> Y;
+    Dest -> Y = Source -> Y;
+    Dest -> Z = ((1 - Source -> x - Source -> y) / Source -> y) * Source -> Y;
+}
+
+static
+cmsFloat64Number f(cmsFloat64Number t)
+{
+    const cmsFloat64Number Limit = (24.0/116.0) * (24.0/116.0) * (24.0/116.0);
+
+    if (t <= Limit)
+        return (841.0/108.0) * t + (16.0/116.0);
+    else
+        return pow(t, 1.0/3.0); 
+}
+
+static
+cmsFloat64Number f_1(cmsFloat64Number t)
+{
+    const cmsFloat64Number Limit = (24.0/116.0);
+
+    if (t <= Limit) {
+        return (108.0/841.0) * (t - (16.0/116.0));
+    }
+
+    return t * t * t;
+}
+
+
+// Standard XYZ to Lab. it can handle negative XZY numbers in some cases
+void CMSEXPORT cmsXYZ2Lab(const cmsCIEXYZ* WhitePoint, cmsCIELab* Lab, const cmsCIEXYZ* xyz)
+{
+    cmsFloat64Number fx, fy, fz;
+
+    if (WhitePoint == NULL) 
+        WhitePoint = cmsD50_XYZ();
+
+    fx = f(xyz->X / WhitePoint->X);
+    fy = f(xyz->Y / WhitePoint->Y);
+    fz = f(xyz->Z / WhitePoint->Z);
+
+    Lab->L = 116.0*fy - 16.0;
+    Lab->a = 500.0*(fx - fy);
+    Lab->b = 200.0*(fy - fz);
+}
+
+
+// Standard XYZ to Lab. It can return negative XYZ in some cases
+void CMSEXPORT cmsLab2XYZ(const cmsCIEXYZ* WhitePoint, cmsCIEXYZ* xyz,  const cmsCIELab* Lab)
+{
+    cmsFloat64Number x, y, z;
+
+    if (WhitePoint == NULL) 
+        WhitePoint = cmsD50_XYZ();
+
+    y = (Lab-> L + 16.0) / 116.0;
+    x = y + 0.002 * Lab -> a;
+    z = y - 0.005 * Lab -> b;
+
+    xyz -> X = f_1(x) * WhitePoint -> X;
+    xyz -> Y = f_1(y) * WhitePoint -> Y;
+    xyz -> Z = f_1(z) * WhitePoint -> Z;
+
+}
+
+static
+cmsFloat64Number L2float2(cmsUInt16Number v)
+{
+	return (cmsFloat64Number) v / 652.800;
+}
+
+// the a/b part
+static
+cmsFloat64Number ab2float2(cmsUInt16Number v)
+{   
+	return ((cmsFloat64Number) v / 256.0) - 128.0;
+}
+
+static
+cmsUInt16Number L2Fix2(cmsFloat64Number L)
+{
+	return _cmsQuickSaturateWord(L *  652.8);
+}
+
+static
+cmsUInt16Number ab2Fix2(cmsFloat64Number ab)
+{
+	return _cmsQuickSaturateWord((ab + 128.0) * 256.0);
+}
+
+
+static
+cmsFloat64Number L2float4(cmsUInt16Number v)
+{
+    return (cmsFloat64Number) v / 655.35;
+}
+
+// the a/b part
+static
+cmsFloat64Number ab2float4(cmsUInt16Number v)
+{
+    return ((cmsFloat64Number) v / 257.0) - 128.0;
+}
+
+
+void CMSEXPORT cmsLabEncoded2FloatV2(cmsCIELab* Lab, const cmsUInt16Number wLab[3])
+{
+        Lab->L = L2float2(wLab[0]);
+        Lab->a = ab2float2(wLab[1]);
+        Lab->b = ab2float2(wLab[2]);
+}
+
+
+void CMSEXPORT cmsLabEncoded2Float(cmsCIELab* Lab, const cmsUInt16Number wLab[3])
+{
+        Lab->L = L2float4(wLab[0]);
+        Lab->a = ab2float4(wLab[1]);
+        Lab->b = ab2float4(wLab[2]);
+}
+
+static
+cmsFloat64Number Clamp_L_doubleV2(cmsFloat64Number L)
+{
+    const cmsFloat64Number L_max = (cmsFloat64Number) (0xFFFF * 100.0) / 0xFF00;
+
+    if (L < 0) L = 0;
+    if (L > L_max) L = L_max;
+
+    return L;
+}
+
+
+static
+cmsFloat64Number Clamp_ab_doubleV2(cmsFloat64Number ab)
+{
+    if (ab < MIN_ENCODEABLE_ab2) ab = MIN_ENCODEABLE_ab2;
+    if (ab > MAX_ENCODEABLE_ab2) ab = MAX_ENCODEABLE_ab2;
+
+    return ab;
+}
+
+void CMSEXPORT cmsFloat2LabEncodedV2(cmsUInt16Number wLab[3], const cmsCIELab* fLab)
+{
+    cmsCIELab Lab;
+
+    Lab.L = Clamp_L_doubleV2(fLab ->L);
+    Lab.a = Clamp_ab_doubleV2(fLab ->a);
+    Lab.b = Clamp_ab_doubleV2(fLab ->b);
+                                                
+    wLab[0] = L2Fix2(Lab.L);
+    wLab[1] = ab2Fix2(Lab.a);
+    wLab[2] = ab2Fix2(Lab.b);
+}
+
+
+static
+cmsFloat64Number Clamp_L_doubleV4(cmsFloat64Number L)
+{
+    if (L < 0) L = 0;
+    if (L > 100.0) L = 100.0;
+
+    return L;
+}
+
+static
+cmsFloat64Number Clamp_ab_doubleV4(cmsFloat64Number ab)
+{
+    if (ab < MIN_ENCODEABLE_ab4) ab = MIN_ENCODEABLE_ab4;
+    if (ab > MAX_ENCODEABLE_ab4) ab = MAX_ENCODEABLE_ab4;
+
+    return ab;
+}
+
+static 
+cmsUInt16Number L2Fix4(cmsFloat64Number L)
+{
+    return _cmsQuickSaturateWord(L *  655.35);
+}
+
+static
+cmsUInt16Number ab2Fix4(cmsFloat64Number ab)
+{
+    return _cmsQuickSaturateWord((ab + 128.0) * 257.0);
+}
+
+void CMSEXPORT cmsFloat2LabEncoded(cmsUInt16Number wLab[3], const cmsCIELab* fLab)
+{
+    cmsCIELab Lab;
+    
+    Lab.L = Clamp_L_doubleV4(fLab ->L);
+    Lab.a = Clamp_ab_doubleV4(fLab ->a);
+    Lab.b = Clamp_ab_doubleV4(fLab ->b);
+                                
+    wLab[0] = L2Fix4(Lab.L);
+    wLab[1] = ab2Fix4(Lab.a);
+    wLab[2] = ab2Fix4(Lab.b);
+}
+
+// Auxiliar: convert to Radians
+static
+cmsFloat64Number RADIANS(cmsFloat64Number deg)
+{
+    return (deg * M_PI) / 180.;
+}
+
+
+// Auxiliar: atan2 but operating in degrees and returning 0 if a==b==0
+static
+cmsFloat64Number atan2deg(cmsFloat64Number a, cmsFloat64Number b)
+{
+   cmsFloat64Number h;
+
+   if (a == 0 && b == 0)
+            h   = 0;
+    else
+            h = atan2(a, b);
+   
+    h *= (180. / M_PI);
+    
+    while (h > 360.)         
+        h -= 360.;
+    
+    while ( h < 0)
+        h += 360.;
+    
+    return h;
+}
+
+
+// Auxiliar: Square
+static
+cmsFloat64Number Sqr(cmsFloat64Number v)
+{
+    return v *  v; 
+}
+// From cylindrical coordinates. No check is performed, then negative values are allowed
+void CMSEXPORT cmsLab2LCh(cmsCIELCh* LCh, const cmsCIELab* Lab)
+{
+    LCh -> L = Lab -> L;
+    LCh -> C = pow(Sqr(Lab ->a) + Sqr(Lab ->b), 0.5);
+    LCh -> h = atan2deg(Lab ->b, Lab ->a);
+}
+
+
+// To cylindrical coordinates. No check is performed, then negative values are allowed
+void CMSEXPORT cmsLCh2Lab(cmsCIELab* Lab, const cmsCIELCh* LCh)
+{
+    cmsFloat64Number h = (LCh -> h * M_PI) / 180.0;
+
+    Lab -> L = LCh -> L;
+    Lab -> a = LCh -> C * cos(h);
+    Lab -> b = LCh -> C * sin(h);          
+}
+
+// In XYZ All 3 components are encoded using 1.15 fixed point
+static
+cmsUInt16Number XYZ2Fix(cmsFloat64Number d)
+{     
+    return _cmsQuickSaturateWord(d * 32768.0);
+}
+
+void CMSEXPORT cmsFloat2XYZEncoded(cmsUInt16Number XYZ[3], const cmsCIEXYZ* fXYZ)
+{
+    cmsCIEXYZ xyz;
+
+    xyz.X = fXYZ -> X;
+    xyz.Y = fXYZ -> Y;
+    xyz.Z = fXYZ -> Z;
+
+    // Clamp to encodeable values.     
+    if (xyz.Y <= 0) {
+
+        xyz.X = 0;
+        xyz.Y = 0;
+        xyz.Z = 0;
+    }
+
+    if (xyz.X > MAX_ENCODEABLE_XYZ)            
+        xyz.X = MAX_ENCODEABLE_XYZ;
+
+    if (xyz.X < 0)
+        xyz.X = 0;
+
+    if (xyz.Y > MAX_ENCODEABLE_XYZ)            
+        xyz.Y = MAX_ENCODEABLE_XYZ;
+
+    if (xyz.Y < 0)
+        xyz.Y = 0;
+
+    if (xyz.Z > MAX_ENCODEABLE_XYZ)            
+        xyz.Z = MAX_ENCODEABLE_XYZ;
+
+    if (xyz.Z < 0)
+        xyz.Z = 0;
+
+
+    XYZ[0] = XYZ2Fix(xyz.X);
+    XYZ[1] = XYZ2Fix(xyz.Y);
+    XYZ[2] = XYZ2Fix(xyz.Z);        
+}
+
+
+//  To convert from Fixed 1.15 point to cmsFloat64Number
+static
+cmsFloat64Number XYZ2float(cmsUInt16Number v)
+{
+    cmsS15Fixed16Number fix32;
+
+    // From 1.15 to 15.16
+    fix32 = v << 1;
+
+    // From fixed 15.16 to cmsFloat64Number
+    return _cms15Fixed16toDouble(fix32);
+}
+
+
+void CMSEXPORT cmsXYZEncoded2Float(cmsCIEXYZ* fXYZ, const cmsUInt16Number XYZ[3])
+{
+    fXYZ -> X = XYZ2float(XYZ[0]);
+    fXYZ -> Y = XYZ2float(XYZ[1]);
+    fXYZ -> Z = XYZ2float(XYZ[2]);
+}       
+
+
+// Returns dE on two Lab values
+cmsFloat64Number CMSEXPORT cmsDeltaE(const cmsCIELab* Lab1, const cmsCIELab* Lab2)
+{
+    cmsFloat64Number dL, da, db;
+
+    dL = fabs(Lab1 -> L - Lab2 -> L);
+    da = fabs(Lab1 -> a - Lab2 -> a);
+    db = fabs(Lab1 -> b - Lab2 -> b);
+
+    return pow(Sqr(dL) + Sqr(da) + Sqr(db), 0.5);
+}
+
+
+// Return the CIE94 Delta E 
+cmsFloat64Number CMSEXPORT cmsCIE94DeltaE(const cmsCIELab* Lab1, const cmsCIELab* Lab2)
+{
+    cmsCIELCh LCh1, LCh2;
+    cmsFloat64Number dE, dL, dC, dh, dhsq;
+    cmsFloat64Number c12, sc, sh;
+
+    dL = fabs(Lab1 ->L - Lab2 ->L);
+
+    cmsLab2LCh(&LCh1, Lab1);
+    cmsLab2LCh(&LCh2, Lab2);
+
+    dC  = fabs(LCh1.C - LCh2.C);
+    dE  = cmsDeltaE(Lab1, Lab2);
+    
+    dhsq = Sqr(dE) - Sqr(dL) - Sqr(dC);
+    if (dhsq < 0)
+        dh = 0;
+    else
+        dh = pow(dhsq, 0.5);
+
+    c12 = sqrt(LCh1.C * LCh2.C);
+
+    sc = 1.0 + (0.048 * c12);
+    sh = 1.0 + (0.014 * c12);
+            
+    return sqrt(Sqr(dL)  + Sqr(dC) / Sqr(sc) + Sqr(dh) / Sqr(sh));
+}
+
+
+// Auxiliary
+static
+cmsFloat64Number ComputeLBFD(const cmsCIELab* Lab)
+{
+  cmsFloat64Number yt;
+
+  if (Lab->L > 7.996969)
+        yt = (Sqr((Lab->L+16)/116)*((Lab->L+16)/116))*100;
+  else
+        yt = 100 * (Lab->L / 903.3);
+
+  return (54.6 * (M_LOG10E * (log(yt + 1.5))) - 9.6);
+}
+
+
+
+// bfd - gets BFD(1:1) difference between Lab1, Lab2
+cmsFloat64Number CMSEXPORT cmsBFDdeltaE(const cmsCIELab* Lab1, const cmsCIELab* Lab2)
+{
+    cmsFloat64Number lbfd1,lbfd2,AveC,Aveh,dE,deltaL,
+        deltaC,deltah,dc,t,g,dh,rh,rc,rt,bfd;
+    cmsCIELCh LCh1, LCh2;
+
+
+    lbfd1 = ComputeLBFD(Lab1);
+    lbfd2 = ComputeLBFD(Lab2);
+    deltaL = lbfd2 - lbfd1;
+
+    cmsLab2LCh(&LCh1, Lab1);
+    cmsLab2LCh(&LCh2, Lab2);
+
+    deltaC = LCh2.C - LCh1.C;
+    AveC = (LCh1.C+LCh2.C)/2;
+    Aveh = (LCh1.h+LCh2.h)/2;
+
+    dE = cmsDeltaE(Lab1, Lab2);
+
+    if (Sqr(dE)>(Sqr(Lab2->L-Lab1->L)+Sqr(deltaC)))
+        deltah = sqrt(Sqr(dE)-Sqr(Lab2->L-Lab1->L)-Sqr(deltaC));
+    else
+        deltah =0;
+
+
+    dc   = 0.035 * AveC / (1 + 0.00365 * AveC)+0.521;
+    g    = sqrt(Sqr(Sqr(AveC))/(Sqr(Sqr(AveC))+14000));
+    t    = 0.627+(0.055*cos((Aveh-254)/(180/M_PI))-       
+           0.040*cos((2*Aveh-136)/(180/M_PI))+
+           0.070*cos((3*Aveh-31)/(180/M_PI))+
+           0.049*cos((4*Aveh+114)/(180/M_PI))-
+           0.015*cos((5*Aveh-103)/(180/M_PI)));
+
+    dh    = dc*(g*t+1-g);
+    rh    = -0.260*cos((Aveh-308)/(180/M_PI))-
+           0.379*cos((2*Aveh-160)/(180/M_PI))-
+           0.636*cos((3*Aveh+254)/(180/M_PI))+
+           0.226*cos((4*Aveh+140)/(180/M_PI))-
+           0.194*cos((5*Aveh+280)/(180/M_PI));
+
+    rc = sqrt((AveC*AveC*AveC*AveC*AveC*AveC)/((AveC*AveC*AveC*AveC*AveC*AveC)+70000000));
+    rt = rh*rc;
+
+    bfd = sqrt(Sqr(deltaL)+Sqr(deltaC/dc)+Sqr(deltah/dh)+(rt*(deltaC/dc)*(deltah/dh)));
+
+    return bfd;
+}
+
+
+//  cmc - CMC(l:c) difference between Lab1, Lab2
+cmsFloat64Number CMSEXPORT cmsCMCdeltaE(const cmsCIELab* Lab1, const cmsCIELab* Lab2, cmsFloat64Number l, cmsFloat64Number c)
+{
+  cmsFloat64Number dE,dL,dC,dh,sl,sc,sh,t,f,cmc;
+  cmsCIELCh LCh1, LCh2;
+
+  if (Lab1 ->L == 0 && Lab2 ->L == 0) return 0;
+
+  cmsLab2LCh(&LCh1, Lab1);
+  cmsLab2LCh(&LCh2, Lab2);
+
+  
+  dL = Lab2->L-Lab1->L;
+  dC = LCh2.C-LCh1.C;
+
+  dE = cmsDeltaE(Lab1, Lab2);
+
+  if (Sqr(dE)>(Sqr(dL)+Sqr(dC))) 
+            dh = sqrt(Sqr(dE)-Sqr(dL)-Sqr(dC));
+  else
+            dh =0;
+
+  if ((LCh1.h > 164) && (LCh1.h < 345)) 
+      t = 0.56 + fabs(0.2 * cos(((LCh1.h + 168)/(180/M_PI))));
+  else 
+      t = 0.36 + fabs(0.4 * cos(((LCh1.h + 35 )/(180/M_PI))));
+
+   sc  = 0.0638   * LCh1.C / (1 + 0.0131  * LCh1.C) + 0.638;
+   sl  = 0.040975 * Lab1->L /(1 + 0.01765 * Lab1->L);
+   
+   if (Lab1->L<16)
+         sl = 0.511; 
+
+   f   = sqrt((LCh1.C * LCh1.C * LCh1.C * LCh1.C)/((LCh1.C * LCh1.C * LCh1.C * LCh1.C)+1900));
+   sh  = sc*(t*f+1-f);
+   cmc = sqrt(Sqr(dL/(l*sl))+Sqr(dC/(c*sc))+Sqr(dh/sh));
+
+   return cmc;
+}
+
+// dE2000 The weightings KL, KC and KH can be modified to reflect the relative 
+// importance of lightness, chroma and hue in different industrial applications
+cmsFloat64Number CMSEXPORT cmsCIE2000DeltaE(const cmsCIELab* Lab1, const cmsCIELab* Lab2,
+                                  cmsFloat64Number Kl, cmsFloat64Number Kc, cmsFloat64Number Kh)
+{
+    cmsFloat64Number L1  = Lab1->L;
+    cmsFloat64Number a1  = Lab1->a;
+    cmsFloat64Number b1  = Lab1->b;
+    cmsFloat64Number C   = sqrt( Sqr(a1) + Sqr(b1) );
+
+    cmsFloat64Number Ls = Lab2 ->L;
+    cmsFloat64Number as = Lab2 ->a;
+    cmsFloat64Number bs = Lab2 ->b;
+    cmsFloat64Number Cs = sqrt( Sqr(as) + Sqr(bs) );
+
+    cmsFloat64Number G = 0.5 * ( 1 - sqrt(pow((C + Cs) / 2 , 7.0) / (pow((C + Cs) / 2, 7.0) + pow(25.0, 7.0) ) ));
+
+    cmsFloat64Number a_p = (1 + G ) * a1;
+    cmsFloat64Number b_p = b1;
+    cmsFloat64Number C_p = sqrt( Sqr(a_p) + Sqr(b_p));
+    cmsFloat64Number h_p = atan2deg(b_p, a_p); 
+                
+
+    cmsFloat64Number a_ps = (1 + G) * as;
+    cmsFloat64Number b_ps = bs;
+    cmsFloat64Number C_ps = sqrt(Sqr(a_ps) + Sqr(b_ps));
+    cmsFloat64Number h_ps = atan2deg(b_ps, a_ps);
+              
+    cmsFloat64Number meanC_p =(C_p + C_ps) / 2;
+
+    cmsFloat64Number hps_plus_hp  = h_ps + h_p;
+    cmsFloat64Number hps_minus_hp = h_ps - h_p;
+
+    cmsFloat64Number meanh_p = fabs(hps_minus_hp) <= 180.000001 ? (hps_plus_hp)/2 : 
+                            (hps_plus_hp) < 360 ? (hps_plus_hp + 360)/2 : 
+                                                 (hps_plus_hp - 360)/2;
+
+    cmsFloat64Number delta_h = (hps_minus_hp) <= -180.000001 ?  (hps_minus_hp + 360) :
+                            (hps_minus_hp) > 180 ? (hps_minus_hp - 360) : 
+                                                    (hps_minus_hp);
+    cmsFloat64Number delta_L = (Ls - L1);
+    cmsFloat64Number delta_C = (C_ps - C_p );
+
+
+    cmsFloat64Number delta_H =2 * sqrt(C_ps*C_p) * sin(RADIANS(delta_h) / 2);
+
+    cmsFloat64Number T = 1 - 0.17 * cos(RADIANS(meanh_p-30)) 
+                 + 0.24 * cos(RADIANS(2*meanh_p))  
+                 + 0.32 * cos(RADIANS(3*meanh_p + 6)) 
+                 - 0.2  * cos(RADIANS(4*meanh_p - 63));
+
+    cmsFloat64Number Sl = 1 + (0.015 * Sqr((Ls + L1) /2- 50) )/ sqrt(20 + Sqr( (Ls+L1)/2 - 50) );
+
+    cmsFloat64Number Sc = 1 + 0.045 * (C_p + C_ps)/2;
+    cmsFloat64Number Sh = 1 + 0.015 * ((C_ps + C_p)/2) * T;
+
+    cmsFloat64Number delta_ro = 30 * exp( -Sqr(((meanh_p - 275 ) / 25)));
+
+    cmsFloat64Number Rc = 2 * sqrt(( pow(meanC_p, 7.0) )/( pow(meanC_p, 7.0) + pow(25.0, 7.0)));
+
+    cmsFloat64Number Rt = -sin(2 * RADIANS(delta_ro)) * Rc;
+
+    cmsFloat64Number deltaE00 = sqrt( Sqr(delta_L /(Sl * Kl)) + 
+                            Sqr(delta_C/(Sc * Kc))  + 
+                            Sqr(delta_H/(Sh * Kh))  + 
+                            Rt*(delta_C/(Sc * Kc)) * (delta_H / (Sh * Kh)));
+
+    return deltaE00;
+}
+
+// This function returns a number of gridpoints to be used as LUT table. It assumes same number
+// of gripdpoints in all dimensions. Flags may override the choice.
+int _cmsReasonableGridpointsByColorspace(cmsColorSpaceSignature Colorspace, cmsUInt32Number dwFlags)
+{
+    int nChannels;
+
+    // Already specified?
+    if (dwFlags & 0x00FF0000) {
+            // Yes, grab'em
+            return (dwFlags >> 16) & 0xFF;
+    }
+
+    nChannels = cmsChannelsOf(Colorspace);
+
+    // HighResPrecalc is maximum resolution
+    if (dwFlags & cmsFLAGS_HIGHRESPRECALC) {
+
+        if (nChannels > 4) 
+                return 7;       // 7 for Hifi
+
+        if (nChannels == 4)     // 23 for CMYK
+                return 23;
+    
+        return 49;      // 49 for RGB and others        
+    }
+
+
+    // LowResPrecal is lower resolution
+    if (dwFlags & cmsFLAGS_LOWRESPRECALC) {
+        
+        if (nChannels > 4) 
+                return 6;       // 6 for more than 4 channels
+
+        if (nChannels == 1) 
+                return 33;      // For monochrome
+
+        return 17;              // 17 for remaining
+    }
+
+    // Default values
+    if (nChannels > 4) 
+                return 7;       // 7 for Hifi
+
+    if (nChannels == 4)
+                return 17;      // 17 for CMYK
+
+    return 33;                  // 33 for RGB   
+}
+
+
+cmsBool  _cmsEndPointsBySpace(cmsColorSpaceSignature Space, 
+                             cmsUInt16Number **White, 
+                             cmsUInt16Number **Black,
+                             cmsUInt32Number *nOutputs)
+{
+       // Only most common spaces
+
+       static cmsUInt16Number RGBblack[4]  = { 0, 0, 0 };
+       static cmsUInt16Number RGBwhite[4]  = { 0xffff, 0xffff, 0xffff };
+       static cmsUInt16Number CMYKblack[4] = { 0xffff, 0xffff, 0xffff, 0xffff };   // 400% of ink
+       static cmsUInt16Number CMYKwhite[4] = { 0, 0, 0, 0 };
+       static cmsUInt16Number LABblack[4]  = { 0, 0x8080, 0x8080 };               // V4 Lab encoding
+       static cmsUInt16Number LABwhite[4]  = { 0xFFFF, 0x8080, 0x8080 };
+       static cmsUInt16Number CMYblack[4]  = { 0xffff, 0xffff, 0xffff };
+       static cmsUInt16Number CMYwhite[4]  = { 0, 0, 0 };
+       static cmsUInt16Number Grayblack[4] = { 0 };
+       static cmsUInt16Number GrayWhite[4] = { 0xffff };
+
+       switch (Space) {
+
+       case cmsSigGrayData: if (White)    *White = GrayWhite;
+                           if (Black)    *Black = Grayblack;
+                           if (nOutputs) *nOutputs = 1;
+                           return TRUE;
+                        
+       case cmsSigRgbData:  if (White)    *White = RGBwhite;
+                           if (Black)    *Black = RGBblack;
+                           if (nOutputs) *nOutputs = 3;
+                           return TRUE;
+
+       case cmsSigLabData:  if (White)    *White = LABwhite;
+                           if (Black)    *Black = LABblack;
+                           if (nOutputs) *nOutputs = 3;
+                           return TRUE;
+
+       case cmsSigCmykData: if (White)    *White = CMYKwhite;
+                           if (Black)    *Black = CMYKblack;
+                           if (nOutputs) *nOutputs = 4;
+                           return TRUE;
+
+       case cmsSigCmyData:  if (White)    *White = CMYwhite;
+                           if (Black)    *Black = CMYblack;
+                           if (nOutputs) *nOutputs = 3;
+                           return TRUE;
+
+       default:;
+       }
+
+  return FALSE;
+}
+
+   
+
+// Several utilities -------------------------------------------------------
+
+// Translate from our colorspace to ICC representation
+
+cmsColorSpaceSignature CMSEXPORT _cmsICCcolorSpace(int OurNotation)
+{
+       switch (OurNotation) {
+
+       case 1:
+       case PT_GRAY: return cmsSigGrayData;
+
+       case 2:
+       case PT_RGB:  return cmsSigRgbData;
+
+       case PT_CMY:  return cmsSigCmyData;
+       case PT_CMYK: return cmsSigCmykData;
+       case PT_YCbCr:return cmsSigYCbCrData;
+       case PT_YUV:  return cmsSigLuvData;
+       case PT_XYZ:  return cmsSigXYZData;
+
+       case PT_LabV2:
+       case PT_Lab:  return cmsSigLabData;
+
+       case PT_YUVK: return cmsSigLuvKData;
+       case PT_HSV:  return cmsSigHsvData;
+       case PT_HLS:  return cmsSigHlsData;
+       case PT_Yxy:  return cmsSigYxyData;
+
+       case PT_MCH1: return cmsSigMCH1Data;
+       case PT_MCH2: return cmsSigMCH2Data;
+       case PT_MCH3: return cmsSigMCH3Data;
+       case PT_MCH4: return cmsSigMCH4Data;
+       case PT_MCH5: return cmsSigMCH5Data;
+       case PT_MCH6: return cmsSigMCH6Data;
+       case PT_MCH7: return cmsSigMCH7Data;
+       case PT_MCH8: return cmsSigMCH8Data;
+
+       case PT_MCH9:  return cmsSigMCH9Data;
+       case PT_MCH10: return cmsSigMCHAData;
+       case PT_MCH11: return cmsSigMCHBData;
+       case PT_MCH12: return cmsSigMCHCData;
+       case PT_MCH13: return cmsSigMCHDData;
+       case PT_MCH14: return cmsSigMCHEData;
+       case PT_MCH15: return cmsSigMCHFData;
+
+       default:  return (cmsColorSpaceSignature) (-1);
+       }
+}
+
+
+int CMSEXPORT _cmsLCMScolorSpace(cmsColorSpaceSignature ProfileSpace)
+{    
+    switch (ProfileSpace) {
+        
+    case cmsSigGrayData: return  PT_GRAY;
+    case cmsSigRgbData:  return  PT_RGB;
+    case cmsSigCmyData:  return  PT_CMY;
+    case cmsSigCmykData: return  PT_CMYK;
+    case cmsSigYCbCrData:return  PT_YCbCr;
+    case cmsSigLuvData:  return  PT_YUV;
+    case cmsSigXYZData:  return  PT_XYZ;
+    case cmsSigLabData:  return  PT_Lab;
+    case cmsSigLuvKData: return  PT_YUVK;
+    case cmsSigHsvData:  return  PT_HSV;
+    case cmsSigHlsData:  return  PT_HLS;
+    case cmsSigYxyData:  return  PT_Yxy;
+    
+    case cmsSig1colorData:
+    case cmsSigMCH1Data: return PT_MCH1;
+    
+    case cmsSig2colorData:
+    case cmsSigMCH2Data: return PT_MCH2;
+    
+    case cmsSig3colorData:
+    case cmsSigMCH3Data: return PT_MCH3;
+    
+    case cmsSig4colorData:
+    case cmsSigMCH4Data: return PT_MCH4;
+    
+    case cmsSig5colorData:
+    case cmsSigMCH5Data: return PT_MCH5;
+    
+    case cmsSig6colorData:
+    case cmsSigMCH6Data: return PT_MCH6;
+        
+    case cmsSigMCH7Data:
+    case cmsSig7colorData:return PT_MCH7;
+        
+    case cmsSigMCH8Data:
+    case cmsSig8colorData:return PT_MCH8;
+        
+    case cmsSigMCH9Data:
+    case cmsSig9colorData:return PT_MCH9;
+        
+    case cmsSigMCHAData:
+    case cmsSig10colorData:return PT_MCH10;
+        
+    case cmsSigMCHBData:
+    case cmsSig11colorData:return PT_MCH11;
+        
+    case cmsSigMCHCData:
+    case cmsSig12colorData:return PT_MCH12;
+        
+    case cmsSigMCHDData:
+    case cmsSig13colorData:return PT_MCH13;
+        
+    case cmsSigMCHEData:
+    case cmsSig14colorData:return PT_MCH14;
+        
+    case cmsSigMCHFData:
+    case cmsSig15colorData:return PT_MCH15;
+                
+    default:  return (cmsColorSpaceSignature) (-1);
+    }
+}
+
+
+cmsUInt32Number CMSEXPORT cmsChannelsOf(cmsColorSpaceSignature ColorSpace)
+{
+    switch (ColorSpace) {
+
+    case cmsSigGrayData: return 1;
+
+    case cmsSig2colorData:  return 2;
+        
+    case cmsSigXYZData:
+    case cmsSigLabData:
+    case cmsSigLuvData:
+    case cmsSigYCbCrData:
+    case cmsSigYxyData:
+    case cmsSigRgbData:   
+    case cmsSigHsvData:
+    case cmsSigHlsData:
+    case cmsSigCmyData: 
+    case cmsSig3colorData:  return 3;
+                   
+    case cmsSigLuvKData:
+    case cmsSigCmykData:
+    case cmsSig4colorData:  return 4;
+
+    case cmsSigMCH5Data:
+    case cmsSig5colorData:  return 5;  
+
+    case cmsSigMCH6Data:   
+    case cmsSig6colorData:  return 6;
+        
+    case cmsSigMCH7Data:
+    case cmsSig7colorData:  return  7;
+
+    case cmsSigMCH8Data:
+    case cmsSig8colorData:  return  8;
+
+    case cmsSigMCH9Data:
+    case cmsSig9colorData:  return  9;
+
+    case cmsSigMCHAData:
+    case cmsSig10colorData: return 10;
+
+    case cmsSigMCHBData:
+    case cmsSig11colorData: return 11;
+    
+    case cmsSigMCHCData:
+    case cmsSig12colorData: return 12;
+
+    case cmsSigMCHDData:
+    case cmsSig13colorData: return 13;
+
+    case cmsSigMCHEData:
+    case cmsSig14colorData: return 14;
+
+    case cmsSigMCHFData:
+    case cmsSig15colorData: return 15;
+
+    default: return 3;
+    }
+}
diff --git a/thirdparty/liblcms2/src/cmsplugin.c b/thirdparty/liblcms2/src/cmsplugin.c
new file mode 100644
index 00000000..1fa5ff4e
--- /dev/null
+++ b/thirdparty/liblcms2/src/cmsplugin.c
@@ -0,0 +1,612 @@
+//---------------------------------------------------------------------------------
+//
+//  Little Color Management System
+//  Copyright (c) 1998-2010 Marti Maria Saguer
+//
+// Permission is hereby granted, free of charge, to any person obtaining 
+// a copy of this software and associated documentation files (the "Software"), 
+// to deal in the Software without restriction, including without limitation 
+// the rights to use, copy, modify, merge, publish, distribute, sublicense, 
+// and/or sell copies of the Software, and to permit persons to whom the Software 
+// is furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in 
+// all copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 
+// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO 
+// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE 
+// LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 
+// OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 
+// WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+//
+//---------------------------------------------------------------------------------
+//
+
+#include "lcms2_internal.h"
+
+
+// ----------------------------------------------------------------------------------
+// Encoding & Decoding support functions
+// ----------------------------------------------------------------------------------
+
+//      Little-Endian to Big-Endian
+
+// Adjust a word value after being readed/ before being written from/to an ICC profile
+cmsUInt16Number CMSEXPORT  _cmsAdjustEndianess16(cmsUInt16Number Word)
+{
+#ifndef CMS_USE_BIG_ENDIAN
+
+    cmsUInt8Number* pByte = (cmsUInt8Number*) &Word;
+    cmsUInt8Number tmp;
+
+    tmp = pByte[0];
+    pByte[0] = pByte[1];
+    pByte[1] = tmp;
+#endif      
+
+    return Word;
+}
+
+
+// Transports to properly encoded values - note that icc profiles does use big endian notation.
+
+// 1 2 3 4
+// 4 3 2 1
+
+cmsUInt32Number CMSEXPORT  _cmsAdjustEndianess32(cmsUInt32Number DWord)
+{
+#ifndef CMS_USE_BIG_ENDIAN
+
+    cmsUInt8Number* pByte = (cmsUInt8Number*) &DWord;
+    cmsUInt8Number temp1;
+    cmsUInt8Number temp2;
+
+    temp1 = *pByte++;
+    temp2 = *pByte++;
+    *(pByte-1) = *pByte;
+    *pByte++ = temp2;
+    *(pByte-3) = *pByte;
+    *pByte = temp1;
+#endif
+    return DWord;
+}
+
+// 1 2 3 4 5 6 7 8
+// 8 7 6 5 4 3 2 1
+
+void CMSEXPORT  _cmsAdjustEndianess64(cmsUInt64Number* Result, cmsUInt64Number QWord)
+{
+    
+#ifndef CMS_USE_BIG_ENDIAN
+    
+    cmsUInt8Number* pIn  = (cmsUInt8Number*) &QWord;
+    cmsUInt8Number* pOut = (cmsUInt8Number*) Result;
+
+    _cmsAssert(Result != NULL);
+
+    pOut[7] = pIn[0];
+    pOut[6] = pIn[1];
+    pOut[5] = pIn[2];
+    pOut[4] = pIn[3];
+    pOut[3] = pIn[4];
+    pOut[2] = pIn[5];   
+    pOut[1] = pIn[6];
+    pOut[0] = pIn[7];
+
+#else
+
+    _cmsAssert(Result != NULL);
+
+    *Result = QWord;
+#endif
+}
+
+// Auxiliar -- read 8, 16 and 32-bit numbers
+cmsBool CMSEXPORT  _cmsReadUInt8Number(cmsIOHANDLER* io, cmsUInt8Number* n)
+{
+    cmsUInt8Number tmp;
+
+    _cmsAssert(io != NULL);
+
+    if (io -> Read(io, &tmp, sizeof(cmsUInt8Number), 1) != 1) 
+            return FALSE;   
+
+    if (n != NULL) *n = tmp;
+    return TRUE;
+}
+
+cmsBool CMSEXPORT  _cmsReadUInt16Number(cmsIOHANDLER* io, cmsUInt16Number* n)
+{
+    cmsUInt16Number tmp;
+
+    _cmsAssert(io != NULL);
+
+    if (io -> Read(io, &tmp, sizeof(cmsUInt16Number), 1) != 1) 
+            return FALSE;   
+
+    if (n != NULL) *n = _cmsAdjustEndianess16(tmp);
+    return TRUE;
+}
+
+cmsBool CMSEXPORT  _cmsReadUInt16Array(cmsIOHANDLER* io, cmsUInt32Number n, cmsUInt16Number* Array)
+{
+    cmsUInt32Number i;
+
+    _cmsAssert(io != NULL);
+
+    for (i=0; i < n; i++) {
+
+        if (Array != NULL) {
+            if (!_cmsReadUInt16Number(io, Array + i)) return FALSE;
+        }
+        else {
+            if (!_cmsReadUInt16Number(io, NULL)) return FALSE;
+        }
+
+    }
+    return TRUE;
+}
+
+cmsBool CMSEXPORT  _cmsReadUInt32Number(cmsIOHANDLER* io, cmsUInt32Number* n)
+{
+    cmsUInt32Number tmp;
+
+    _cmsAssert(io != NULL);
+
+    if (io -> Read(io, &tmp, sizeof(cmsUInt32Number), 1) != 1) 
+            return FALSE;   
+
+    if (n != NULL) *n = _cmsAdjustEndianess32(tmp);
+    return TRUE;
+}
+
+cmsBool CMSEXPORT  _cmsReadFloat32Number(cmsIOHANDLER* io, cmsFloat32Number* n)
+{
+    cmsUInt32Number tmp;
+
+    _cmsAssert(io != NULL);
+
+    if (io -> Read(io, &tmp, sizeof(cmsFloat32Number), 1) != 1) 
+            return FALSE;   
+
+    if (n != NULL) {
+
+        tmp = _cmsAdjustEndianess32(tmp);
+        *n = *(cmsFloat32Number*) &tmp;
+    }
+    return TRUE;
+}
+
+
+cmsBool CMSEXPORT   _cmsReadUInt64Number(cmsIOHANDLER* io, cmsUInt64Number* n)
+{
+    cmsUInt64Number tmp;
+
+    _cmsAssert(io != NULL);
+
+    if (io -> Read(io, &tmp, sizeof(cmsUInt64Number), 1) != 1) 
+            return FALSE;   
+
+    if (n != NULL) _cmsAdjustEndianess64(n, tmp);
+    return TRUE;
+}
+
+
+cmsBool CMSEXPORT  _cmsRead15Fixed16Number(cmsIOHANDLER* io, cmsFloat64Number* n)
+{
+    cmsUInt32Number tmp;
+
+    _cmsAssert(io != NULL);
+
+    if (io -> Read(io, &tmp, sizeof(cmsUInt32Number), 1) != 1) 
+            return FALSE;   
+
+    if (n != NULL) {
+        *n = _cms15Fixed16toDouble(_cmsAdjustEndianess32(tmp));
+    }
+
+    return TRUE;
+}
+
+
+// Jun-21-2000: Some profiles (those that comes with W2K) comes
+// with the media white (media black?) x 100. Add a sanity check
+
+static
+void NormalizeXYZ(cmsCIEXYZ* Dest)
+{
+    while (Dest -> X > 2. &&
+           Dest -> Y > 2. &&
+           Dest -> Z > 2.) {
+
+               Dest -> X /= 10.;
+               Dest -> Y /= 10.;
+               Dest -> Z /= 10.;
+       }
+}
+
+cmsBool CMSEXPORT  _cmsReadXYZNumber(cmsIOHANDLER* io, cmsCIEXYZ* XYZ)
+{
+    cmsEncodedXYZNumber xyz;
+
+    _cmsAssert(io != NULL);
+
+    if (io ->Read(io, &xyz, sizeof(cmsEncodedXYZNumber), 1) != 1) return FALSE;
+
+    if (XYZ != NULL) {
+
+        XYZ->X = _cms15Fixed16toDouble(_cmsAdjustEndianess32(xyz.X));
+        XYZ->Y = _cms15Fixed16toDouble(_cmsAdjustEndianess32(xyz.Y));
+        XYZ->Z = _cms15Fixed16toDouble(_cmsAdjustEndianess32(xyz.Z));
+
+        NormalizeXYZ(XYZ);
+    }
+    return TRUE;
+}
+
+cmsBool CMSEXPORT  _cmsWriteUInt8Number(cmsIOHANDLER* io, cmsUInt8Number n)
+{
+    _cmsAssert(io != NULL);
+
+    if (io -> Write(io, sizeof(cmsUInt8Number), &n) != 1) 
+            return FALSE;   
+    
+    return TRUE;
+}
+
+cmsBool CMSEXPORT  _cmsWriteUInt16Number(cmsIOHANDLER* io, cmsUInt16Number n)
+{
+    cmsUInt16Number tmp;
+
+    _cmsAssert(io != NULL);
+
+    tmp = _cmsAdjustEndianess16(n);
+    if (io -> Write(io, sizeof(cmsUInt16Number), &tmp) != 1) 
+            return FALSE;   
+    
+    return TRUE;
+}
+
+cmsBool CMSEXPORT  _cmsWriteUInt16Array(cmsIOHANDLER* io, cmsUInt32Number n, const cmsUInt16Number* Array)
+{
+    cmsUInt32Number i;
+
+    _cmsAssert(io != NULL);
+    _cmsAssert(Array != NULL);
+
+    for (i=0; i < n; i++) {
+        if (!_cmsWriteUInt16Number(io, Array[i])) return FALSE;
+    }
+
+    return TRUE;
+}
+
+cmsBool CMSEXPORT  _cmsWriteUInt32Number(cmsIOHANDLER* io, cmsUInt32Number n)
+{
+    cmsUInt32Number tmp;
+
+    _cmsAssert(io != NULL);
+
+    tmp = _cmsAdjustEndianess32(n);
+    if (io -> Write(io, sizeof(cmsUInt32Number), &tmp) != 1) 
+            return FALSE;   
+    
+    return TRUE;
+}
+
+
+cmsBool CMSEXPORT  _cmsWriteFloat32Number(cmsIOHANDLER* io, cmsFloat32Number n)
+{
+    cmsUInt32Number tmp;
+
+    _cmsAssert(io != NULL);
+
+    tmp = *(cmsUInt32Number*) &n;
+    tmp = _cmsAdjustEndianess32(tmp);
+    if (io -> Write(io, sizeof(cmsUInt32Number), &tmp) != 1) 
+            return FALSE;   
+    
+    return TRUE;
+}
+
+cmsBool CMSEXPORT  _cmsWriteUInt64Number(cmsIOHANDLER* io, cmsUInt64Number n)
+{
+    cmsUInt64Number tmp;
+
+    _cmsAssert(io != NULL);
+
+    _cmsAdjustEndianess64(&tmp, n);
+    if (io -> Write(io, sizeof(cmsUInt64Number), &tmp) != 1) 
+            return FALSE;   
+    
+    return TRUE;
+}
+
+cmsBool CMSEXPORT  _cmsWrite15Fixed16Number(cmsIOHANDLER* io, cmsFloat64Number n)
+{
+    cmsUInt32Number tmp;
+
+    _cmsAssert(io != NULL);
+
+    tmp = _cmsAdjustEndianess32(_cmsDoubleTo15Fixed16(n));
+    if (io -> Write(io, sizeof(cmsUInt32Number), &tmp) != 1) 
+            return FALSE;   
+    
+    return TRUE;
+}
+
+cmsBool CMSEXPORT  _cmsWriteXYZNumber(cmsIOHANDLER* io, const cmsCIEXYZ* XYZ)
+{
+    cmsEncodedXYZNumber xyz;
+  
+    _cmsAssert(io != NULL);
+    _cmsAssert(XYZ != NULL);
+
+    xyz.X = _cmsAdjustEndianess32(_cmsDoubleTo15Fixed16(XYZ->X));
+    xyz.Y = _cmsAdjustEndianess32(_cmsDoubleTo15Fixed16(XYZ->Y));
+    xyz.Z = _cmsAdjustEndianess32(_cmsDoubleTo15Fixed16(XYZ->Z));
+
+    return io -> Write(io,  sizeof(cmsEncodedXYZNumber), &xyz);
+}
+
+// from Fixed point 8.8 to double
+cmsFloat64Number CMSEXPORT _cms8Fixed8toDouble(cmsUInt16Number fixed8)
+{
+       cmsUInt8Number  msb, lsb;
+
+       lsb = (cmsUInt8Number) (fixed8 & 0xff);
+       msb = (cmsUInt8Number) (((cmsUInt16Number) fixed8 >> 8) & 0xff);
+
+       return (cmsFloat64Number) ((cmsFloat64Number) msb + ((cmsFloat64Number) lsb / 256.0));
+}
+
+cmsUInt16Number CMSEXPORT _cmsDoubleTo8Fixed8(cmsFloat64Number val)
+{
+    cmsS15Fixed16Number GammaFixed32 = _cmsDoubleTo15Fixed16(val);
+    return  (cmsUInt16Number) ((GammaFixed32 >> 8) & 0xFFFF);       
+}
+
+// from Fixed point 15.16 to double
+cmsFloat64Number CMSEXPORT _cms15Fixed16toDouble(cmsS15Fixed16Number fix32)
+{
+    cmsFloat64Number floater, sign, mid;
+    int Whole, FracPart;
+
+    sign  = (fix32 < 0 ? -1 : 1);
+    fix32 = abs(fix32);
+
+    Whole     = (cmsUInt16Number)(fix32 >> 16) & 0xffff;
+    FracPart  = (cmsUInt16Number)(fix32 & 0xffff);
+
+    mid     = (cmsFloat64Number) FracPart / 65536.0;
+    floater = (cmsFloat64Number) Whole + mid;
+
+    return sign * floater;
+}
+
+// from double to Fixed point 15.16 
+cmsS15Fixed16Number CMSEXPORT _cmsDoubleTo15Fixed16(cmsFloat64Number v)
+{
+    return ((cmsS15Fixed16Number) floor((v)*65536.0 + 0.5));
+}
+
+// Date/Time functions 
+
+void CMSEXPORT _cmsDecodeDateTimeNumber(const cmsDateTimeNumber *Source, struct tm *Dest)
+{
+
+    _cmsAssert(Dest != NULL);
+    _cmsAssert(Source != NULL);
+
+    Dest->tm_sec   = _cmsAdjustEndianess16(Source->seconds);
+    Dest->tm_min   = _cmsAdjustEndianess16(Source->minutes);
+    Dest->tm_hour  = _cmsAdjustEndianess16(Source->hours);
+    Dest->tm_mday  = _cmsAdjustEndianess16(Source->day);
+    Dest->tm_mon   = _cmsAdjustEndianess16(Source->month) - 1;
+    Dest->tm_year  = _cmsAdjustEndianess16(Source->year) - 1900;
+    Dest->tm_wday  = -1;
+    Dest->tm_yday  = -1;
+    Dest->tm_isdst = 0;
+}
+
+void CMSEXPORT _cmsEncodeDateTimeNumber(cmsDateTimeNumber *Dest, const struct tm *Source)
+{
+    _cmsAssert(Dest != NULL);
+    _cmsAssert(Source != NULL);
+
+    Dest->seconds = _cmsAdjustEndianess16((cmsUInt16Number) Source->tm_sec);
+    Dest->minutes = _cmsAdjustEndianess16((cmsUInt16Number) Source->tm_min);
+    Dest->hours   = _cmsAdjustEndianess16((cmsUInt16Number) Source->tm_hour);
+    Dest->day     = _cmsAdjustEndianess16((cmsUInt16Number) Source->tm_mday);
+    Dest->month   = _cmsAdjustEndianess16((cmsUInt16Number) (Source->tm_mon + 1));
+    Dest->year    = _cmsAdjustEndianess16((cmsUInt16Number) (Source->tm_year + 1900));
+}
+
+// Read base and return type base
+cmsTagTypeSignature CMSEXPORT _cmsReadTypeBase(cmsIOHANDLER* io)
+{
+    _cmsTagBase Base;
+
+    _cmsAssert(io != NULL);
+
+    if (io -> Read(io, &Base, sizeof(_cmsTagBase), 1) != 1) 
+        return (cmsTagTypeSignature) 0;
+
+    return (cmsTagTypeSignature) _cmsAdjustEndianess32(Base.sig);
+}
+
+// Setup base marker
+cmsBool  CMSEXPORT _cmsWriteTypeBase(cmsIOHANDLER* io, cmsTagTypeSignature sig)
+{
+    _cmsTagBase  Base;
+
+    _cmsAssert(io != NULL);
+
+    Base.sig = (cmsTagTypeSignature) _cmsAdjustEndianess32(sig);
+    memset(&Base.reserved, 0, sizeof(Base.reserved));
+    return io -> Write(io, sizeof(_cmsTagBase), &Base);
+}
+
+cmsBool CMSEXPORT _cmsReadAlignment(cmsIOHANDLER* io)
+{
+    cmsUInt8Number  Buffer[4];
+    cmsUInt32Number NextAligned, At;
+    cmsUInt32Number BytesToNextAlignedPos;
+    
+    _cmsAssert(io != NULL);
+
+    At = io -> Tell(io);
+    NextAligned = _cmsALIGNLONG(At);
+    BytesToNextAlignedPos = NextAligned - At;
+    if (BytesToNextAlignedPos == 0) return TRUE;
+    if (BytesToNextAlignedPos > 4)  return FALSE;
+
+    return (io ->Read(io, Buffer, BytesToNextAlignedPos, 1) == 1);
+}
+
+cmsBool CMSEXPORT _cmsWriteAlignment(cmsIOHANDLER* io)
+{
+    cmsUInt8Number  Buffer[4];
+    cmsUInt32Number NextAligned, At;
+    cmsUInt32Number BytesToNextAlignedPos;
+
+    _cmsAssert(io != NULL);
+
+    At = io -> Tell(io);
+    NextAligned = _cmsALIGNLONG(At);
+    BytesToNextAlignedPos = NextAligned - At;
+    if (BytesToNextAlignedPos == 0) return TRUE;
+    if (BytesToNextAlignedPos > 4)  return FALSE;
+
+    memset(Buffer, 0, BytesToNextAlignedPos);
+    return io -> Write(io, BytesToNextAlignedPos, Buffer);
+}
+
+
+// To deal with text streams. 2K at most
+cmsBool CMSEXPORT _cmsIOPrintf(cmsIOHANDLER* io, const char* frm, ...)
+{
+    va_list args;
+    int len;
+    cmsUInt8Number Buffer[2048];
+    cmsBool rc;
+
+    _cmsAssert(io != NULL);
+    _cmsAssert(frm != NULL);
+
+    va_start(args, frm);
+
+    len = vsnprintf((char*) Buffer, 2047, frm, args);
+    if (len < 0) return FALSE;   // Truncated, which is a fatal error for us
+
+    rc = io ->Write(io, len, Buffer);
+    
+    va_end(args);
+
+    return rc;
+}
+
+
+// Plugin memory management -------------------------------------------------------------------------------------------------
+
+static _cmsSubAllocator* PluginPool = NULL;
+
+// Specialized malloc for plug-ins, that is freed upon exit.
+void* _cmsPluginMalloc(cmsUInt32Number size)
+{
+    if (PluginPool == NULL)
+        PluginPool = _cmsCreateSubAlloc(0, 4*1024);
+
+    return _cmsSubAlloc(PluginPool, size);
+}
+
+
+// Main plug-in dispatcher
+cmsBool CMSEXPORT cmsPlugin(void* Plug_in)
+{
+    cmsPluginBase* Plugin;
+
+    for (Plugin = (cmsPluginBase*) Plug_in; 
+         Plugin != NULL; 
+         Plugin = Plugin -> Next) {
+
+            if (Plugin -> Magic != cmsPluginMagicNumber) {
+                cmsSignalError(0, cmsERROR_UNKNOWN_EXTENSION, "Unrecognized plugin");
+                return FALSE;
+            }
+
+			if (Plugin ->ExpectedVersion > LCMS_VERSION) {
+				cmsSignalError(0, cmsERROR_UNKNOWN_EXTENSION, "plugin needs Little CMS %d, current  version is %d", 
+					Plugin ->ExpectedVersion, LCMS_VERSION);
+				return FALSE;
+			}
+
+            switch (Plugin -> Type) {
+
+                case cmsPluginMemHandlerSig:
+                    if (!_cmsRegisterMemHandlerPlugin(Plugin)) return FALSE;
+                    break;
+
+                case cmsPluginInterpolationSig:
+                    if (!_cmsRegisterInterpPlugin(Plugin)) return FALSE;
+                    break;
+                
+                case cmsPluginTagTypeSig:
+                    if (!_cmsRegisterTagTypePlugin(Plugin)) return FALSE;
+                    break;
+            
+                case cmsPluginTagSig:
+                    if (!_cmsRegisterTagPlugin(Plugin)) return FALSE;
+                    break;
+
+                case cmsPluginFormattersSig:
+                    if (!_cmsRegisterFormattersPlugin(Plugin)) return FALSE;
+                    break;
+
+                case cmsPluginRenderingIntentSig:
+                    if (!_cmsRegisterRenderingIntentPlugin(Plugin)) return FALSE;
+                    break;
+
+                case cmsPluginParametricCurveSig:
+                    if (!_cmsRegisterParametricCurvesPlugin(Plugin)) return FALSE;
+                    break;
+
+                case cmsPluginMultiProcessElementSig:
+                    if (!_cmsRegisterMultiProcessElementPlugin(Plugin)) return FALSE;
+                    break;
+
+                case cmsPluginOptimizationSig:
+                    if (!_cmsRegisterOptimizationPlugin(Plugin)) return FALSE;
+                    break;
+
+                default:
+                    cmsSignalError(0, cmsERROR_UNKNOWN_EXTENSION, "Unrecognized plugin type '%X'", Plugin -> Type);
+                    return FALSE;
+            }            
+    }
+
+    // Keep a reference to the plug-in
+    return TRUE;
+}
+
+
+// Revert all plug-ins to default
+void CMSEXPORT cmsUnregisterPlugins(void)
+{
+    _cmsRegisterMemHandlerPlugin(NULL);
+    _cmsRegisterInterpPlugin(NULL);
+    _cmsRegisterTagTypePlugin(NULL);
+    _cmsRegisterTagPlugin(NULL);
+    _cmsRegisterFormattersPlugin(NULL);
+    _cmsRegisterRenderingIntentPlugin(NULL);
+    _cmsRegisterParametricCurvesPlugin(NULL);
+    _cmsRegisterMultiProcessElementPlugin(NULL);
+    _cmsRegisterOptimizationPlugin(NULL);
+
+    if (PluginPool != NULL)
+        _cmsSubAllocDestroy(PluginPool);
+
+    PluginPool = NULL;
+}
diff --git a/thirdparty/liblcms2/src/cmsps2.c b/thirdparty/liblcms2/src/cmsps2.c
new file mode 100644
index 00000000..b41f58ff
--- /dev/null
+++ b/thirdparty/liblcms2/src/cmsps2.c
@@ -0,0 +1,1595 @@
+//---------------------------------------------------------------------------------
+//
+//  Little Color Management System
+//  Copyright (c) 1998-2008 Marti Maria Saguer
+//
+// Permission is hereby granted, free of charge, to any person obtaining 
+// a copy of this software and associated documentation files (the "Software"), 
+// to deal in the Software without restriction, including without limitation 
+// the rights to use, copy, modify, merge, publish, distribute, sublicense, 
+// and/or sell copies of the Software, and to permit persons to whom the Software 
+// is furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in 
+// all copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 
+// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO 
+// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE 
+// LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 
+// OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 
+// WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+//
+//---------------------------------------------------------------------------------
+//
+
+#include "lcms2_internal.h"
+
+// PostScript ColorRenderingDictionary and ColorSpaceArray															   
+
+
+#define MAXPSCOLS   60      // Columns on tables
+
+/*
+    Implementation
+    --------------
+
+  PostScript does use XYZ as its internal PCS. But since PostScript 
+  interpolation tables are limited to 8 bits, I use Lab as a way to 
+  improve the accuracy, favoring perceptual results. So, for the creation 
+  of each CRD, CSA the profiles are converted to Lab via a device 
+  link between  profile -> Lab or Lab -> profile. The PS code necessary to
+  convert Lab <-> XYZ is also included.
+
+
+
+  Color Space Arrays (CSA) 
+  ==================================================================================
+
+  In order to obtain precision, code chooses between three ways to implement
+  the device -> XYZ transform. These cases identifies monochrome profiles (often
+  implemented as a set of curves), matrix-shaper and Pipeline-based.
+
+  Monochrome 
+  -----------
+
+  This is implemented as /CIEBasedA CSA. The prelinearization curve is 
+  placed into /DecodeA section, and matrix equals to D50. Since here is
+  no interpolation tables, I do the conversion directly to XYZ
+
+  NOTE: CLUT-based monochrome profiles are NOT supported. So, cmsFLAGS_MATRIXINPUT
+  flag is forced on such profiles.
+
+    [ /CIEBasedA
+      <<            
+            /DecodeA { transfer function } bind
+            /MatrixA [D50]  
+            /RangeLMN [ 0.0 cmsD50X 0.0 cmsD50Y 0.0 cmsD50Z ]
+            /WhitePoint [D50]
+            /BlackPoint [BP]
+            /RenderingIntent (intent)
+      >>
+    ] 
+
+   On simpler profiles, the PCS is already XYZ, so no conversion is required.
+
+ 
+   Matrix-shaper based
+   -------------------
+
+   This is implemented both with /CIEBasedABC or /CIEBasedDEF on dependig
+   of profile implementation. Since here there are no interpolation tables, I do 
+   the conversion directly to XYZ
+
+
+
+    [ /CIEBasedABC
+            <<
+                /DecodeABC [ {transfer1} {transfer2} {transfer3} ]
+                /MatrixABC [Matrix]
+                /RangeLMN [ 0.0 cmsD50X 0.0 cmsD50Y 0.0 cmsD50Z ]
+                /DecodeLMN [ { / 2} dup dup ]
+                /WhitePoint [D50]
+                /BlackPoint [BP]
+                /RenderingIntent (intent)
+            >>
+    ] 
+
+
+    CLUT based
+    ----------
+
+     Lab is used in such cases.
+
+    [ /CIEBasedDEF
+            <<
+            /DecodeDEF [ <prelinearization> ]
+            /Table [ p p p [<...>]]
+            /RangeABC [ 0 1 0 1 0 1]
+            /DecodeABC[ <postlinearization> ]
+            /RangeLMN [ -0.236 1.254 0 1 -0.635 1.640 ] 
+               % -128/500 1+127/500 0 1  -127/200 1+128/200 
+            /MatrixABC [ 1 1 1 1 0 0 0 0 -1]
+            /WhitePoint [D50]
+            /BlackPoint [BP]
+            /RenderingIntent (intent)
+    ] 
+
+
+  Color Rendering Dictionaries (CRD)
+  ==================================
+  These are always implemented as CLUT, and always are using Lab. Since CRD are expected to
+  be used as resources, the code adds the definition as well.
+
+  <<
+    /ColorRenderingType 1
+    /WhitePoint [ D50 ]
+    /BlackPoint [BP]
+    /MatrixPQR [ Bradford ]
+    /RangePQR [-0.125 1.375 -0.125 1.375 -0.125 1.375 ]
+    /TransformPQR [            
+    {4 index 3 get div 2 index 3 get mul exch pop exch pop exch pop exch pop } bind
+    {4 index 4 get div 2 index 4 get mul exch pop exch pop exch pop exch pop } bind
+    {4 index 5 get div 2 index 5 get mul exch pop exch pop exch pop exch pop } bind
+    ]
+    /MatrixABC <...>
+    /EncodeABC <...>
+    /RangeABC  <.. used for  XYZ -> Lab>
+    /EncodeLMN
+    /RenderTable [ p p p [<...>]]   
+    
+    /RenderingIntent (Perceptual)
+  >> 
+  /Current exch /ColorRendering defineresource pop
+
+
+  The following stages are used to convert from XYZ to Lab
+  --------------------------------------------------------  
+
+  Input is given at LMN stage on X, Y, Z
+
+  Encode LMN gives us f(X/Xn), f(Y/Yn), f(Z/Zn)
+
+  /EncodeLMN [
+
+    { 0.964200  div dup 0.008856 le {7.787 mul 16 116 div add}{1 3 div exp} ifelse } bind
+    { 1.000000  div dup 0.008856 le {7.787 mul 16 116 div add}{1 3 div exp} ifelse } bind
+    { 0.824900  div dup 0.008856 le {7.787 mul 16 116 div add}{1 3 div exp} ifelse } bind
+
+    ]
+    
+      
+  MatrixABC is used to compute f(Y/Yn), f(X/Xn) - f(Y/Yn), f(Y/Yn) - f(Z/Zn)
+
+  | 0  1  0|
+  | 1 -1  0|
+  | 0  1 -1|
+
+  /MatrixABC [ 0 1 0 1 -1 1 0 0 -1 ]
+
+ EncodeABC finally gives Lab values.
+
+  /EncodeABC [
+    { 116 mul  16 sub 100 div  } bind
+    { 500 mul 128 add 255 div  } bind
+    { 200 mul 128 add 255 div  } bind
+    ]   
+    
+  The following stages are used to convert Lab to XYZ
+  ----------------------------------------------------
+
+    /RangeABC [ 0 1 0 1 0 1]
+    /DecodeABC [ { 100 mul 16 add 116 div } bind
+                 { 255 mul 128 sub 500 div } bind
+                 { 255 mul 128 sub 200 div } bind 
+               ]
+    
+    /MatrixABC [ 1 1 1 1 0 0 0 0 -1]
+    /DecodeLMN [
+                {dup 6 29 div ge {dup dup mul mul} {4 29 div sub 108 841 div mul} ifelse 0.964200 mul} bind
+                {dup 6 29 div ge {dup dup mul mul} {4 29 div sub 108 841 div mul} ifelse } bind
+                {dup 6 29 div ge {dup dup mul mul} {4 29 div sub 108 841 div mul} ifelse 0.824900 mul} bind
+                ]
+
+
+*/
+
+/*
+
+ PostScript algorithms discussion.
+ =========================================================================================================
+
+  1D interpolation algorithm 
+
+
+  1D interpolation (float)
+  ------------------------
+    
+    val2 = Domain * Value;
+
+    cell0 = (int) floor(val2);
+    cell1 = (int) ceil(val2);
+
+    rest = val2 - cell0;
+
+    y0 = LutTable[cell0] ;
+    y1 = LutTable[cell1] ;
+
+    y = y0 + (y1 - y0) * rest;
+
+
+  
+  PostScript code                   Stack
+  ================================================
+
+  {                                 % v
+    <check 0..1.0>
+    [array]                         % v tab
+    dup                             % v tab tab
+    length 1 sub                    % v tab dom
+
+    3 -1 roll                       % tab dom v
+
+    mul                             % tab val2
+    dup                             % tab val2 val2
+    dup                             % tab val2 val2 val2
+    floor cvi                       % tab val2 val2 cell0
+    exch                            % tab val2 cell0 val2
+    ceiling cvi                     % tab val2 cell0 cell1
+
+    3 index                         % tab val2 cell0 cell1 tab  
+    exch                            % tab val2 cell0 tab cell1
+    get                             % tab val2 cell0 y1
+
+    4 -1 roll                       % val2 cell0 y1 tab
+    3 -1 roll                       % val2 y1 tab cell0 
+    get                             % val2 y1 y0 
+
+    dup                             % val2 y1 y0 y0
+    3 1 roll                        % val2 y0 y1 y0 
+
+    sub                             % val2 y0 (y1-y0)
+    3 -1 roll                       % y0 (y1-y0) val2
+    dup                             % y0 (y1-y0) val2 val2
+    floor cvi                       % y0 (y1-y0) val2 floor(val2) 
+    sub                             % y0 (y1-y0) rest
+    mul                             % y0 t1
+    add                             % y
+    65535 div                       % result
+
+  } bind
+
+
+*/
+
+
+// This struct holds the memory block currently being write
+typedef struct {
+	_cmsStageCLutData* Pipeline;
+	cmsIOHANDLER* m;
+
+	int FirstComponent;
+	int SecondComponent;
+	
+	const char* PreMaj;
+	const char* PostMaj;
+	const char* PreMin;
+	const char* PostMin;
+
+	int  FixWhite;    // Force mapping of pure white 
+
+	cmsColorSpaceSignature  ColorSpace;  // ColorSpace of profile
+
+
+} cmsPsSamplerCargo;
+
+static int _cmsPSActualColumn = 0;
+
+
+// Convert to byte
+static
+cmsUInt8Number Word2Byte(cmsUInt16Number w)
+{
+    return (cmsUInt8Number) floor((cmsFloat64Number) w / 257.0 + 0.5);
+}
+
+
+// Convert to byte (using ICC2 notation)
+/*
+static
+cmsUInt8Number L2Byte(cmsUInt16Number w)
+{    
+	int ww = w + 0x0080;
+
+	if (ww > 0xFFFF) return 0xFF;
+
+    return (cmsUInt8Number) ((cmsUInt16Number) (ww >> 8) & 0xFF);
+}
+*/
+
+// Write a cooked byte
+
+static
+void WriteByte(cmsIOHANDLER* m, cmsUInt8Number b)
+{
+	_cmsIOPrintf(m, "%02x", b);	
+	_cmsPSActualColumn += 2;
+
+	if (_cmsPSActualColumn > MAXPSCOLS) {
+
+		_cmsIOPrintf(m, "\n");
+		_cmsPSActualColumn = 0;
+	}	
+}
+
+// ----------------------------------------------------------------- PostScript generation
+
+
+// Removes offending Carriage returns
+static 
+char* RemoveCR(const char* txt)
+{
+    static char Buffer[2048];
+    char* pt;
+
+    strncpy(Buffer, txt, 2047);
+    Buffer[2047] = 0;
+    for (pt = Buffer; *pt; pt++)
+            if (*pt == '\n' || *pt == '\r') *pt = ' ';
+
+    return Buffer;
+
+}
+
+static
+void EmitHeader(cmsIOHANDLER* m, const char* Title, cmsHPROFILE hProfile)
+{
+    time_t timer;
+	cmsMLU *Description, *Copyright;
+	char DescASCII[256], CopyrightASCII[256];
+    
+    time(&timer);
+	
+	Description = (cmsMLU*) cmsReadTag(hProfile, cmsSigProfileDescriptionTag);
+	Copyright   = (cmsMLU*) cmsReadTag(hProfile, cmsSigCopyrightTag);
+
+	DescASCII[0] = DescASCII[255] = 0;
+    CopyrightASCII[0] = CopyrightASCII[255] = 0;
+
+	if (Description != NULL) cmsMLUgetASCII(Description,  cmsNoLanguage, cmsNoCountry, DescASCII,       255);
+	if (Copyright != NULL)   cmsMLUgetASCII(Copyright,    cmsNoLanguage, cmsNoCountry, CopyrightASCII,  255);
+
+    _cmsIOPrintf(m, "%%!PS-Adobe-3.0\n");
+    _cmsIOPrintf(m, "%%\n");
+    _cmsIOPrintf(m, "%% %s\n", Title);
+    _cmsIOPrintf(m, "%% Source: %s\n", RemoveCR(DescASCII));
+    _cmsIOPrintf(m, "%%         %s\n", RemoveCR(CopyrightASCII));
+    _cmsIOPrintf(m, "%% Created: %s", ctime(&timer)); // ctime appends a \n!!!
+    _cmsIOPrintf(m, "%%\n");
+    _cmsIOPrintf(m, "%%%%BeginResource\n");
+
+}
+
+
+// Emits White & Black point. White point is always D50, Black point is the device 
+// Black point adapted to D50. 
+
+static
+void EmitWhiteBlackD50(cmsIOHANDLER* m, cmsCIEXYZ* BlackPoint)
+{
+
+    _cmsIOPrintf(m, "/BlackPoint [%f %f %f]\n", BlackPoint -> X,
+                                          BlackPoint -> Y,
+                                          BlackPoint -> Z);
+
+    _cmsIOPrintf(m, "/WhitePoint [%f %f %f]\n", cmsD50_XYZ()->X, 
+                                          cmsD50_XYZ()->Y,
+                                          cmsD50_XYZ()->Z);
+}
+
+
+static
+void EmitRangeCheck(cmsIOHANDLER* m)
+{
+    _cmsIOPrintf(m, "dup 0.0 lt { pop 0.0 } if "
+                    "dup 1.0 gt { pop 1.0 } if ");
+
+}
+
+// Does write the intent
+
+static
+void EmitIntent(cmsIOHANDLER* m, int RenderingIntent)
+{
+    const char *intent;
+
+    switch (RenderingIntent) {
+
+        case INTENT_PERCEPTUAL:            intent = "Perceptual"; break;
+        case INTENT_RELATIVE_COLORIMETRIC: intent = "RelativeColorimetric"; break;
+        case INTENT_ABSOLUTE_COLORIMETRIC: intent = "AbsoluteColorimetric"; break;
+        case INTENT_SATURATION:            intent = "Saturation"; break;
+
+        default: intent = "Undefined"; break;
+    }
+
+    _cmsIOPrintf(m, "/RenderingIntent (%s)\n", intent );    
+}
+
+//
+//  Convert L* to Y
+//
+//      Y = Yn*[ (L* + 16) / 116] ^ 3   if (L*) >= 6 / 29
+//        = Yn*( L* / 116) / 7.787      if (L*) < 6 / 29
+//
+
+/*
+static
+void EmitL2Y(cmsIOHANDLER* m)
+{
+    _cmsIOPrintf(m, 
+            "{ " 
+                "100 mul 16 add 116 div "               // (L * 100 + 16) / 116
+                 "dup 6 29 div ge "                     // >= 6 / 29 ?          
+                 "{ dup dup mul mul } "                 // yes, ^3 and done
+                 "{ 4 29 div sub 108 841 div mul } "    // no, slope limiting
+            "ifelse } bind "); 
+}
+*/
+
+
+// Lab -> XYZ, see the discussion above
+
+static
+void EmitLab2XYZ(cmsIOHANDLER* m)
+{
+    _cmsIOPrintf(m, "/RangeABC [ 0 1 0 1 0 1]\n");
+    _cmsIOPrintf(m, "/DecodeABC [\n");
+    _cmsIOPrintf(m, "{100 mul  16 add 116 div } bind\n");
+    _cmsIOPrintf(m, "{255 mul 128 sub 500 div } bind\n");
+    _cmsIOPrintf(m, "{255 mul 128 sub 200 div } bind\n");
+    _cmsIOPrintf(m, "]\n");
+    _cmsIOPrintf(m, "/MatrixABC [ 1 1 1 1 0 0 0 0 -1]\n");
+	_cmsIOPrintf(m, "/RangeLMN [ -0.236 1.254 0 1 -0.635 1.640 ]\n"); 
+    _cmsIOPrintf(m, "/DecodeLMN [\n");
+    _cmsIOPrintf(m, "{dup 6 29 div ge {dup dup mul mul} {4 29 div sub 108 841 div mul} ifelse 0.964200 mul} bind\n");
+    _cmsIOPrintf(m, "{dup 6 29 div ge {dup dup mul mul} {4 29 div sub 108 841 div mul} ifelse } bind\n");
+    _cmsIOPrintf(m, "{dup 6 29 div ge {dup dup mul mul} {4 29 div sub 108 841 div mul} ifelse 0.824900 mul} bind\n");
+    _cmsIOPrintf(m, "]\n");
+}
+
+
+
+// Outputs a table of words. It does use 16 bits
+
+static
+void Emit1Gamma(cmsIOHANDLER* m, cmsToneCurve* Table)
+{
+    cmsUInt32Number i;
+    cmsFloat64Number gamma;
+
+
+    if (Table ->nEntries <= 0) return;  // Empty table
+
+    // Suppress whole if identity
+    if (cmsIsToneCurveLinear(Table)) return;
+
+    // Check if is really an exponential. If so, emit "exp"
+	gamma = cmsEstimateGamma(Table, 0.001);
+     if (gamma > 0) {
+            _cmsIOPrintf(m, "{ %g exp } bind ", gamma);
+            return;
+     }
+
+    _cmsIOPrintf(m, "{ ");
+
+    // Bounds check
+    EmitRangeCheck(m);
+    
+    // Emit intepolation code
+
+    // PostScript code                      Stack
+    // ===============                      ========================
+                                            // v
+    _cmsIOPrintf(m, " [");
+
+    for (i=0; i < Table->nEntries; i++) {
+		_cmsIOPrintf(m, "%d ", Table->Table16[i]);
+    }
+
+    _cmsIOPrintf(m, "] ");                        // v tab
+
+    _cmsIOPrintf(m, "dup ");                      // v tab tab        
+    _cmsIOPrintf(m, "length 1 sub ");             // v tab dom
+    _cmsIOPrintf(m, "3 -1 roll ");                // tab dom v
+    _cmsIOPrintf(m, "mul ");                      // tab val2
+    _cmsIOPrintf(m, "dup ");                      // tab val2 val2
+    _cmsIOPrintf(m, "dup ");                      // tab val2 val2 val2
+    _cmsIOPrintf(m, "floor cvi ");                // tab val2 val2 cell0
+    _cmsIOPrintf(m, "exch ");                     // tab val2 cell0 val2
+    _cmsIOPrintf(m, "ceiling cvi ");              // tab val2 cell0 cell1
+    _cmsIOPrintf(m, "3 index ");                  // tab val2 cell0 cell1 tab 
+    _cmsIOPrintf(m, "exch ");                     // tab val2 cell0 tab cell1
+    _cmsIOPrintf(m, "get ");                      // tab val2 cell0 y1
+    _cmsIOPrintf(m, "4 -1 roll ");                // val2 cell0 y1 tab
+    _cmsIOPrintf(m, "3 -1 roll ");                // val2 y1 tab cell0 
+    _cmsIOPrintf(m, "get ");                      // val2 y1 y0 
+    _cmsIOPrintf(m, "dup ");                      // val2 y1 y0 y0
+    _cmsIOPrintf(m, "3 1 roll ");                 // val2 y0 y1 y0 
+    _cmsIOPrintf(m, "sub ");                      // val2 y0 (y1-y0)
+    _cmsIOPrintf(m, "3 -1 roll ");                // y0 (y1-y0) val2
+    _cmsIOPrintf(m, "dup ");                      // y0 (y1-y0) val2 val2
+    _cmsIOPrintf(m, "floor cvi ");                // y0 (y1-y0) val2 floor(val2) 
+    _cmsIOPrintf(m, "sub ");                      // y0 (y1-y0) rest
+    _cmsIOPrintf(m, "mul ");                      // y0 t1
+    _cmsIOPrintf(m, "add ");                      // y
+    _cmsIOPrintf(m, "65535 div ");                // result
+
+    _cmsIOPrintf(m, " } bind ");
+}
+
+
+// Compare gamma table
+
+static
+cmsBool GammaTableEquals(cmsUInt16Number* g1, cmsUInt16Number* g2, int nEntries)
+{    
+    return memcmp(g1, g2, nEntries* sizeof(cmsUInt16Number)) == 0;
+}
+
+
+// Does write a set of gamma curves
+
+static
+void EmitNGamma(cmsIOHANDLER* m, int n, cmsToneCurve* g[])                  
+{
+    int i;
+    
+    for( i=0; i < n; i++ )
+    {                
+		if (i > 0 && GammaTableEquals(g[i-1]->Table16, g[i]->Table16, g[i]->nEntries)) {
+
+            _cmsIOPrintf(m, "dup ");
+        }
+        else {    
+			Emit1Gamma(m, g[i]);
+        }
+    }
+    
+}
+
+
+
+
+
+// Following code dumps a LUT onto memory stream
+        
+
+// This is the sampler. Intended to work in SAMPLER_INSPECT mode,
+// that is, the callback will be called for each knot with
+//
+//          In[]  The grid location coordinates, normalized to 0..ffff
+//          Out[] The Pipeline values, normalized to 0..ffff
+//
+//  Returning a value other than 0 does terminate the sampling process
+//
+//  Each row contains Pipeline values for all but first component. So, I 
+//  detect row changing by keeping a copy of last value of first 
+//  component. -1 is used to mark begining of whole block.
+
+static
+int OutputValueSampler(register const cmsUInt16Number In[], register cmsUInt16Number Out[], register void* Cargo)
+{
+    cmsPsSamplerCargo* sc = (cmsPsSamplerCargo*) Cargo;
+    cmsUInt32Number i;
+
+
+    if (sc -> FixWhite) {
+
+        if (In[0] == 0xFFFF) {  // Only in L* = 100, ab = [-8..8]
+
+            if ((In[1] >= 0x7800 && In[1] <= 0x8800) &&
+                (In[2] >= 0x7800 && In[2] <= 0x8800)) {
+
+                cmsUInt16Number* Black;
+                cmsUInt16Number* White;
+                cmsUInt32Number nOutputs;
+
+                if (!_cmsEndPointsBySpace(sc ->ColorSpace, &White, &Black, &nOutputs))
+                        return 0;
+
+                for (i=0; i < nOutputs; i++)
+                        Out[i] = White[i];
+            }
+
+             
+        }
+    }
+
+
+    // Hadle the parenthesis on rows
+
+    if (In[0] != sc ->FirstComponent) {
+            
+            if (sc ->FirstComponent != -1) {
+
+                    _cmsIOPrintf(sc ->m, sc ->PostMin);
+                    sc ->SecondComponent = -1;
+                    _cmsIOPrintf(sc ->m, sc ->PostMaj);           
+            }
+
+            // Begin block  
+            _cmsPSActualColumn = 0;
+                    
+            _cmsIOPrintf(sc ->m, sc ->PreMaj);            
+            sc ->FirstComponent = In[0]; 
+    }
+
+
+      if (In[1] != sc ->SecondComponent) {
+            
+            if (sc ->SecondComponent != -1) {
+
+                    _cmsIOPrintf(sc ->m, sc ->PostMin);           
+            }
+                    
+            _cmsIOPrintf(sc ->m, sc ->PreMin);            
+            sc ->SecondComponent = In[1]; 
+    }
+
+	  // Dump table. 
+
+	  for (i=0; i < sc -> Pipeline ->Params->nOutputs; i++) {
+
+		  cmsUInt16Number wWordOut = Out[i];
+          cmsUInt8Number wByteOut;           // Value as byte
+		 
+
+		  // We always deal with Lab4
+		  
+		  wByteOut = Word2Byte(wWordOut);
+		  WriteByte(sc -> m, wByteOut);
+	  }
+
+	  return 1;
+}
+
+// Writes a Pipeline on memstream. Could be 8 or 16 bits based
+
+static
+void WriteCLUT(cmsIOHANDLER* m, cmsStage* mpe, const char* PreMaj, 
+                                             const char* PostMaj,
+                                             const char* PreMin,
+                                             const char* PostMin,                                             
+                                             int FixWhite,
+                                             cmsColorSpaceSignature ColorSpace)
+{
+    cmsUInt32Number i;
+    cmsPsSamplerCargo sc;
+
+    sc.FirstComponent = -1;
+    sc.SecondComponent = -1;
+	sc.Pipeline = (_cmsStageCLutData *) mpe ->Data;
+    sc.m   = m;    
+    sc.PreMaj = PreMaj;
+    sc.PostMaj= PostMaj;
+
+    sc.PreMin   = PreMin;
+    sc.PostMin  = PostMin;    
+    sc.FixWhite = FixWhite;
+    sc.ColorSpace = ColorSpace;
+
+    _cmsIOPrintf(m, "[");
+
+	for (i=0; i < sc.Pipeline->Params->nInputs; i++)
+		_cmsIOPrintf(m, " %d ", sc.Pipeline->Params->nSamples[i]);
+
+    _cmsIOPrintf(m, " [\n");
+
+    cmsStageSampleCLut16bit(mpe, OutputValueSampler, (void*) &sc, SAMPLER_INSPECT);
+    
+    _cmsIOPrintf(m, PostMin);
+    _cmsIOPrintf(m, PostMaj);
+    _cmsIOPrintf(m, "] ");
+
+}
+
+
+// Dumps CIEBasedA Color Space Array
+
+static
+int EmitCIEBasedA(cmsIOHANDLER* m, cmsToneCurve* Curve, cmsCIEXYZ* BlackPoint)
+{
+
+	_cmsIOPrintf(m, "[ /CIEBasedA\n");
+	_cmsIOPrintf(m, "  <<\n");
+
+	_cmsIOPrintf(m, "/DecodeA ");
+
+	Emit1Gamma(m, Curve);
+
+	_cmsIOPrintf(m, " \n");
+
+	_cmsIOPrintf(m, "/MatrixA [ 0.9642 1.0000 0.8249 ]\n");
+	_cmsIOPrintf(m, "/RangeLMN [ 0.0 0.9642 0.0 1.0000 0.0 0.8249 ]\n");
+
+	EmitWhiteBlackD50(m, BlackPoint);
+	EmitIntent(m, INTENT_PERCEPTUAL);
+
+	_cmsIOPrintf(m, ">>\n");        
+	_cmsIOPrintf(m, "]\n");
+
+	return 1;
+}
+
+
+// Dumps CIEBasedABC Color Space Array
+
+static
+int EmitCIEBasedABC(cmsIOHANDLER* m, cmsFloat64Number* Matrix, cmsToneCurve** CurveSet, cmsCIEXYZ* BlackPoint)
+{
+	int i;
+	
+	_cmsIOPrintf(m, "[ /CIEBasedABC\n");
+	_cmsIOPrintf(m, "<<\n");
+	_cmsIOPrintf(m, "/DecodeABC [ ");
+
+	EmitNGamma(m, 3, CurveSet);
+
+	_cmsIOPrintf(m, "]\n");
+
+	_cmsIOPrintf(m, "/MatrixABC [ " );
+
+	for( i=0; i < 3; i++ ) {
+
+        _cmsIOPrintf(m, "%.6f %.6f %.6f ", Matrix[i + 3*0],
+                                           Matrix[i + 3*1],
+                                           Matrix[i + 3*2]);      
+	}
+
+
+	_cmsIOPrintf(m, "]\n");
+
+	_cmsIOPrintf(m, "/RangeLMN [ 0.0 0.9642 0.0 1.0000 0.0 0.8249 ]\n");
+
+	EmitWhiteBlackD50(m, BlackPoint);
+	EmitIntent(m, INTENT_PERCEPTUAL);
+
+	_cmsIOPrintf(m, ">>\n");
+	_cmsIOPrintf(m, "]\n");
+
+
+	return 1;
+}
+
+
+static
+int EmitCIEBasedDEF(cmsIOHANDLER* m, cmsPipeline* Pipeline, int Intent, cmsCIEXYZ* BlackPoint)
+{
+    const char* PreMaj;
+    const char* PostMaj;
+    const char* PreMin, *PostMin;
+	cmsStage* mpe;
+
+	mpe = Pipeline ->Elements;
+
+
+	switch (cmsStageInputChannels(mpe)) {
+    case 3:
+
+            _cmsIOPrintf(m, "[ /CIEBasedDEF\n");
+            PreMaj ="<"; 
+            PostMaj= ">\n";
+            PreMin = PostMin = "";
+            break;
+    case 4:
+            _cmsIOPrintf(m, "[ /CIEBasedDEFG\n");
+            PreMaj = "[";
+            PostMaj = "]\n";
+            PreMin = "<";
+            PostMin = ">\n";
+            break;
+    default:
+            return 0;
+
+    }
+
+    _cmsIOPrintf(m, "<<\n");
+
+	if (cmsStageType(mpe) == cmsSigCurveSetElemType) {
+    		
+        _cmsIOPrintf(m, "/DecodeDEF [ ");
+		EmitNGamma(m, cmsStageOutputChannels(mpe), _cmsStageGetPtrToCurveSet(mpe));
+        _cmsIOPrintf(m, "]\n");
+
+		mpe = mpe ->Next;
+    }
+
+
+
+	if (cmsStageType(mpe) == cmsSigCLutElemType) {
+
+            _cmsIOPrintf(m, "/Table ");    
+            WriteCLUT(m, mpe, PreMaj, PostMaj, PreMin, PostMin, FALSE, (cmsColorSpaceSignature) 0);
+            _cmsIOPrintf(m, "]\n");
+    }
+       
+    EmitLab2XYZ(m);
+    EmitWhiteBlackD50(m, BlackPoint);
+    EmitIntent(m, Intent);
+
+    _cmsIOPrintf(m, "   >>\n");       
+    _cmsIOPrintf(m, "]\n");
+    
+
+    return 1;
+}
+
+// Generates a curve from a gray profile
+
+static
+cmsToneCurve* ExtractGray2Y(cmsContext ContextID, cmsHPROFILE hProfile, int Intent)
+{
+    cmsToneCurve* Out = cmsBuildTabulatedToneCurve16(ContextID, 256, NULL);
+    cmsHPROFILE hXYZ  = cmsCreateXYZProfile();
+    cmsHTRANSFORM xform = cmsCreateTransformTHR(ContextID, hProfile, TYPE_GRAY_8, hXYZ, TYPE_XYZ_DBL, Intent, cmsFLAGS_NOOPTIMIZE);
+    int i;
+
+    for (i=0; i < 256; i++) {
+        
+      cmsUInt8Number Gray = (cmsUInt8Number) i;
+      cmsCIEXYZ XYZ;
+      
+        cmsDoTransform(xform, &Gray, &XYZ, 1);
+        
+		Out ->Table16[i] =_cmsQuickSaturateWord(XYZ.Y * 65535.0);
+    }
+
+    cmsDeleteTransform(xform);
+    cmsCloseProfile(hXYZ);
+    return Out;
+}
+
+
+
+// Because PostScript has only 8 bits in /Table, we should use
+// a more perceptually uniform space... I do choose Lab.
+
+static
+int WriteInputLUT(cmsIOHANDLER* m, cmsHPROFILE hProfile, int Intent, cmsUInt32Number dwFlags)
+{
+    cmsHPROFILE hLab;
+    cmsHTRANSFORM xform;
+    cmsUInt32Number nChannels;
+    cmsUInt32Number InputFormat;
+    int rc;
+    cmsHPROFILE Profiles[2];
+    cmsCIEXYZ BlackPointAdaptedToD50;
+
+    // Does create a device-link based transform. 
+    // The DeviceLink is next dumped as working CSA.
+    
+    InputFormat = cmsFormatterForColorspaceOfProfile(hProfile, 2, FALSE);
+    nChannels   = T_CHANNELS(InputFormat);
+
+	
+	cmsDetectBlackPoint(&BlackPointAdaptedToD50, hProfile, Intent, 0);
+
+ 	// Adjust output to Lab4 
+    hLab = cmsCreateLab4ProfileTHR(m ->ContextID, NULL);
+
+	Profiles[0] = hProfile;
+	Profiles[1] = hLab;
+
+	xform = cmsCreateMultiprofileTransform(Profiles, 2,  InputFormat, TYPE_Lab_DBL, Intent, 0);
+	cmsCloseProfile(hLab);
+	
+	if (xform == NULL) {
+
+		cmsSignalError(m ->ContextID, cmsERROR_COLORSPACE_CHECK, "Cannot create transform Profile -> Lab");
+		return 0;
+	}
+    
+    // Only 1, 3 and 4 channels are allowed
+
+    switch (nChannels) {
+
+    case 1: {            
+		    cmsToneCurve* Gray2Y = ExtractGray2Y(m ->ContextID, hProfile, Intent);
+            EmitCIEBasedA(m, Gray2Y, &BlackPointAdaptedToD50);            
+            cmsFreeToneCurve(Gray2Y);            
+            }
+            break;
+
+    case 3: 
+    case 4: {
+		    cmsUInt32Number OutFrm = TYPE_Lab_16;
+            cmsPipeline* DeviceLink;
+            _cmsTRANSFORM* v = (_cmsTRANSFORM*) xform;
+
+			DeviceLink = cmsPipelineDup(v ->Lut);
+			if (DeviceLink == NULL) return 0;
+
+			dwFlags |= cmsFLAGS_FORCE_CLUT;
+			_cmsOptimizePipeline(&DeviceLink, Intent, &InputFormat, &OutFrm, &dwFlags);
+            
+            rc = EmitCIEBasedDEF(m, DeviceLink, Intent, &BlackPointAdaptedToD50);
+            cmsPipelineFree(DeviceLink);            
+            }
+            break;
+
+    default:
+
+		cmsSignalError(m ->ContextID, cmsERROR_COLORSPACE_CHECK, "Only 3, 4 channels supported for CSA. This profile has %d channels.", nChannels);
+        return 0;
+    }
+    
+
+    cmsDeleteTransform(xform);
+    
+    return 1;
+}
+
+static
+cmsFloat64Number* GetPtrToMatrix(const cmsStage* mpe)
+{
+    _cmsStageMatrixData* Data = (_cmsStageMatrixData*) mpe ->Data;
+
+    return Data -> Double;
+}
+
+
+// Does create CSA based on matrix-shaper. Allowed types are gray and RGB based
+
+static
+int WriteInputMatrixShaper(cmsIOHANDLER* m, cmsHPROFILE hProfile, cmsStage* Matrix, cmsStage* Shaper)
+{
+    cmsColorSpaceSignature ColorSpace;    
+    int rc;
+    cmsCIEXYZ BlackPointAdaptedToD50;
+
+    ColorSpace = cmsGetColorSpace(hProfile);
+    
+    cmsDetectBlackPoint(&BlackPointAdaptedToD50, hProfile, INTENT_RELATIVE_COLORIMETRIC, 0);
+
+    if (ColorSpace == cmsSigGrayData) {
+            
+		    cmsToneCurve** ShaperCurve = _cmsStageGetPtrToCurveSet(Shaper);
+            rc = EmitCIEBasedA(m, ShaperCurve[0], &BlackPointAdaptedToD50);
+        
+    }
+    else
+        if (ColorSpace == cmsSigRgbData) {
+        			
+            cmsMAT3 Mat;
+            int i, j;
+
+            memmove(&Mat, GetPtrToMatrix(Matrix), sizeof(Mat));
+
+            for (i=0; i < 3; i++)
+                for (j=0; j < 3; j++)
+                    Mat.v[i].n[j] *= MAX_ENCODEABLE_XYZ;
+
+            rc = EmitCIEBasedABC(m,  (cmsFloat64Number *) &Mat, 
+			                        _cmsStageGetPtrToCurveSet(Shaper), 
+									&BlackPointAdaptedToD50);      
+        }
+        else  {
+
+			cmsSignalError(m ->ContextID, cmsERROR_COLORSPACE_CHECK, "Profile is not suitable for CSA. Unsupported colorspace.");
+            return 0;
+        }
+    
+    return rc;
+}
+
+
+
+// Creates a PostScript color list from a named profile data. 
+// This is a HP extension, and it works in Lab instead of XYZ
+
+static
+int WriteNamedColorCSA(cmsIOHANDLER* m, cmsHPROFILE hNamedColor, int Intent)
+{
+    cmsHTRANSFORM xform;
+    cmsHPROFILE   hLab;
+    int i, nColors;
+    char ColorName[32];
+    cmsNAMEDCOLORLIST* NamedColorList;
+
+	hLab  = cmsCreateLab4ProfileTHR(m ->ContextID, NULL);
+    xform = cmsCreateTransform(hNamedColor, TYPE_NAMED_COLOR_INDEX, hLab, TYPE_Lab_DBL, Intent, 0);
+    if (xform == NULL) return 0;
+
+	NamedColorList = cmsGetNamedColorList(xform);
+    if (NamedColorList == NULL) return 0;
+
+    _cmsIOPrintf(m, "<<\n");
+    _cmsIOPrintf(m, "(colorlistcomment) (%s)\n", "Named color CSA");
+    _cmsIOPrintf(m, "(Prefix) [ (Pantone ) (PANTONE ) ]\n");
+    _cmsIOPrintf(m, "(Suffix) [ ( CV) ( CVC) ( C) ]\n");
+
+    nColors   = cmsNamedColorCount(NamedColorList);
+
+
+    for (i=0; i < nColors; i++) {
+        
+        cmsUInt16Number In[1];
+        cmsCIELab Lab;
+
+        In[0] = (cmsUInt16Number) i;
+
+        if (!cmsNamedColorInfo(NamedColorList, i, ColorName, NULL, NULL, NULL, NULL))
+                continue;
+
+        cmsDoTransform(xform, In, &Lab, 1);     
+        _cmsIOPrintf(m, "  (%s) [ %.3f %.3f %.3f ]\n", ColorName, Lab.L, Lab.a, Lab.b);
+    }
+
+
+        
+    _cmsIOPrintf(m, ">>\n");
+
+    cmsDeleteTransform(xform);
+    cmsCloseProfile(hLab);
+    return 1;
+}
+
+
+// Does create a Color Space Array on XYZ colorspace for PostScript usage
+static
+cmsUInt32Number GenerateCSA(cmsContext ContextID, 
+						    cmsHPROFILE hProfile, 
+						    cmsUInt32Number Intent, 
+						    cmsUInt32Number dwFlags, 
+						    cmsIOHANDLER* mem)
+{	
+	cmsUInt32Number dwBytesUsed;
+	cmsPipeline* lut = NULL;
+	cmsStage* Matrix, *Shaper;
+
+
+	// Is a named color profile?
+	if (cmsGetDeviceClass(hProfile) == cmsSigNamedColorClass) {
+
+		if (!WriteNamedColorCSA(mem, hProfile, Intent)) goto Error;
+	}
+	else {
+
+
+		// Any profile class are allowed (including devicelink), but
+		// output (PCS) colorspace must be XYZ or Lab
+		cmsColorSpaceSignature ColorSpace = cmsGetPCS(hProfile);
+
+		if (ColorSpace != cmsSigXYZData &&
+			ColorSpace != cmsSigLabData) {
+
+				cmsSignalError(ContextID, cmsERROR_COLORSPACE_CHECK, "Invalid output color space");
+				goto Error;
+		}
+
+
+		// Read the lut with all necessary conversion stages
+		lut = _cmsReadInputLUT(hProfile, Intent);
+		if (lut == NULL) goto Error;
+
+
+		// Tone curves + matrix can be implemented without any LUT
+		if (cmsPipelineCheckAndRetreiveStages(lut, 2, cmsSigCurveSetElemType, cmsSigMatrixElemType, &Shaper, &Matrix)) {
+
+			if (!WriteInputMatrixShaper(mem, hProfile, Matrix, Shaper)) goto Error;
+
+		}
+		else {
+		   // We need a LUT for the rest
+		   if (!WriteInputLUT(mem, hProfile, Intent, dwFlags)) goto Error;
+		}
+	}
+
+
+	// Done, keep memory usage
+	dwBytesUsed = mem ->UsedSpace;
+
+	// Get rid of LUT
+	if (lut != NULL) cmsPipelineFree(lut);
+
+	// Finally, return used byte count
+	return dwBytesUsed;
+
+Error:
+	if (lut != NULL) cmsPipelineFree(lut);
+	return 0;	
+}
+
+// ------------------------------------------------------ Color Rendering Dictionary (CRD)
+
+
+
+/*
+
+  Black point compensation plus chromatic adaptation:
+
+  Step 1 - Chromatic adaptation
+  =============================
+
+          WPout
+    X = ------- PQR
+          Wpin
+
+  Step 2 - Black point compensation
+  =================================
+
+          (WPout - BPout)*X - WPout*(BPin - BPout)
+    out = --------------------------------------- 
+                        WPout - BPin
+
+
+  Algorithm discussion
+  ====================
+      
+  TransformPQR(WPin, BPin, WPout, BPout, PQR)
+
+  Wpin,etc= { Xws Yws Zws Pws Qws Rws }
+
+
+  Algorithm             Stack 0...n
+  ===========================================================
+                        PQR BPout WPout BPin WPin
+  4 index 3 get         WPin PQR BPout WPout BPin WPin  
+  div                   (PQR/WPin) BPout WPout BPin WPin   
+  2 index 3 get         WPout (PQR/WPin) BPout WPout BPin WPin   
+  mult                  WPout*(PQR/WPin) BPout WPout BPin WPin   
+  
+  2 index 3 get         WPout WPout*(PQR/WPin) BPout WPout BPin WPin   
+  2 index 3 get         BPout WPout WPout*(PQR/WPin) BPout WPout BPin WPin     
+  sub                   (WPout-BPout) WPout*(PQR/WPin) BPout WPout BPin WPin    
+  mult                  (WPout-BPout)* WPout*(PQR/WPin) BPout WPout BPin WPin   
+          
+  2 index 3 get         WPout (BPout-WPout)* WPout*(PQR/WPin) BPout WPout BPin WPin     
+  4 index 3 get         BPin WPout (BPout-WPout)* WPout*(PQR/WPin) BPout WPout BPin WPin     
+  3 index 3 get         BPout BPin WPout (BPout-WPout)* WPout*(PQR/WPin) BPout WPout BPin WPin
+  
+  sub                   (BPin-BPout) WPout (BPout-WPout)* WPout*(PQR/WPin) BPout WPout BPin WPin     
+  mult                  (BPin-BPout)*WPout (BPout-WPout)* WPout*(PQR/WPin) BPout WPout BPin WPin     
+  sub                   (BPout-WPout)* WPout*(PQR/WPin)-(BPin-BPout)*WPout BPout WPout BPin WPin     
+
+  3 index 3 get         BPin (BPout-WPout)* WPout*(PQR/WPin)-(BPin-BPout)*WPout BPout WPout BPin WPin     
+  3 index 3 get         WPout BPin (BPout-WPout)* WPout*(PQR/WPin)-(BPin-BPout)*WPout BPout WPout BPin WPin     
+  exch
+  sub                   (WPout-BPin) (BPout-WPout)* WPout*(PQR/WPin)-(BPin-BPout)*WPout BPout WPout BPin WPin       
+  div                
+  
+  exch pop 
+  exch pop
+  exch pop
+  exch pop
+
+*/
+
+
+static
+void EmitPQRStage(cmsIOHANDLER* m, cmsHPROFILE hProfile, int DoBPC, int lIsAbsolute)
+{
+
+   
+        if (lIsAbsolute) {
+
+            // For absolute colorimetric intent, encode back to relative 
+			// and generate a relative Pipeline
+
+			// Relative encoding is obtained across XYZpcs*(D50/WhitePoint)
+
+			cmsCIEXYZ White;
+
+			_cmsReadMediaWhitePoint(&White, hProfile);
+
+			_cmsIOPrintf(m,"/MatrixPQR [1 0 0 0 1 0 0 0 1 ]\n");
+            _cmsIOPrintf(m,"/RangePQR [ -0.5 2 -0.5 2 -0.5 2 ]\n");
+
+            _cmsIOPrintf(m, "%% Absolute colorimetric -- encode to relative to maximize LUT usage\n"
+                      "/TransformPQR [\n"
+                      "{0.9642 mul %g div exch pop exch pop exch pop exch pop} bind\n"
+                      "{1.0000 mul %g div exch pop exch pop exch pop exch pop} bind\n"
+                      "{0.8249 mul %g div exch pop exch pop exch pop exch pop} bind\n]\n", 
+					  White.X, White.Y, White.Z);
+            return;
+        }
+
+
+        _cmsIOPrintf(m,"%% Bradford Cone Space\n"
+                 "/MatrixPQR [0.8951 -0.7502 0.0389 0.2664 1.7135 -0.0685 -0.1614 0.0367 1.0296 ] \n");
+
+        _cmsIOPrintf(m, "/RangePQR [ -0.5 2 -0.5 2 -0.5 2 ]\n");
+
+
+        // No BPC
+
+        if (!DoBPC) {
+
+            _cmsIOPrintf(m, "%% VonKries-like transform in Bradford Cone Space\n"
+                      "/TransformPQR [\n"
+                      "{exch pop exch 3 get mul exch pop exch 3 get div} bind\n"
+                      "{exch pop exch 4 get mul exch pop exch 4 get div} bind\n"
+                      "{exch pop exch 5 get mul exch pop exch 5 get div} bind\n]\n"); 
+        } else {
+
+            // BPC
+
+            _cmsIOPrintf(m, "%% VonKries-like transform in Bradford Cone Space plus BPC\n"
+                      "/TransformPQR [\n");
+                  
+            _cmsIOPrintf(m, "{4 index 3 get div 2 index 3 get mul "
+                    "2 index 3 get 2 index 3 get sub mul "                          
+                    "2 index 3 get 4 index 3 get 3 index 3 get sub mul sub "
+                    "3 index 3 get 3 index 3 get exch sub div "
+                    "exch pop exch pop exch pop exch pop } bind\n");
+
+            _cmsIOPrintf(m, "{4 index 4 get div 2 index 4 get mul "
+                    "2 index 4 get 2 index 4 get sub mul "
+                    "2 index 4 get 4 index 4 get 3 index 4 get sub mul sub "
+                    "3 index 4 get 3 index 4 get exch sub div "
+                    "exch pop exch pop exch pop exch pop } bind\n");
+
+            _cmsIOPrintf(m, "{4 index 5 get div 2 index 5 get mul "
+                    "2 index 5 get 2 index 5 get sub mul "
+                    "2 index 5 get 4 index 5 get 3 index 5 get sub mul sub "
+                    "3 index 5 get 3 index 5 get exch sub div "
+                    "exch pop exch pop exch pop exch pop } bind\n]\n");
+
+        }
+          
+        
+}
+
+
+static
+void EmitXYZ2Lab(cmsIOHANDLER* m)
+{
+    _cmsIOPrintf(m, "/RangeLMN [ -0.635 2.0 0 2 -0.635 2.0 ]\n"); 
+    _cmsIOPrintf(m, "/EncodeLMN [\n");
+    _cmsIOPrintf(m, "{ 0.964200  div dup 0.008856 le {7.787 mul 16 116 div add}{1 3 div exp} ifelse } bind\n");
+    _cmsIOPrintf(m, "{ 1.000000  div dup 0.008856 le {7.787 mul 16 116 div add}{1 3 div exp} ifelse } bind\n");
+    _cmsIOPrintf(m, "{ 0.824900  div dup 0.008856 le {7.787 mul 16 116 div add}{1 3 div exp} ifelse } bind\n");
+    _cmsIOPrintf(m, "]\n");
+    _cmsIOPrintf(m, "/MatrixABC [ 0 1 0 1 -1 1 0 0 -1 ]\n");
+    _cmsIOPrintf(m, "/EncodeABC [\n");
+    
+        
+    _cmsIOPrintf(m, "{ 116 mul  16 sub 100 div  } bind\n");
+    _cmsIOPrintf(m, "{ 500 mul 128 add 256 div  } bind\n");
+    _cmsIOPrintf(m, "{ 200 mul 128 add 256 div  } bind\n");
+    
+    
+    _cmsIOPrintf(m, "]\n");
+    
+
+}
+
+// Due to impedance mismatch between XYZ and almost all RGB and CMYK spaces
+// I choose to dump LUTS in Lab instead of XYZ. There is still a lot of wasted
+// space on 3D CLUT, but since space seems not to be a problem here, 33 points
+// would give a reasonable accurancy. Note also that CRD tables must operate in 
+// 8 bits.
+
+static
+int WriteOutputLUT(cmsIOHANDLER* m, cmsHPROFILE hProfile, int Intent, cmsUInt32Number dwFlags)
+{
+    cmsHPROFILE hLab;
+    cmsHTRANSFORM xform;
+    int i, nChannels;
+    cmsUInt32Number OutputFormat;
+    _cmsTRANSFORM* v;
+    cmsPipeline* DeviceLink;
+    cmsHPROFILE Profiles[3];
+    cmsCIEXYZ BlackPointAdaptedToD50;
+    cmsBool lDoBPC = (dwFlags & cmsFLAGS_BLACKPOINTCOMPENSATION);
+    cmsBool lFixWhite = !(dwFlags & cmsFLAGS_NOWHITEONWHITEFIXUP);
+	cmsUInt32Number InFrm = TYPE_Lab_16;
+	int RelativeEncodingIntent;
+	cmsColorSpaceSignature ColorSpace;
+    
+    
+	hLab = cmsCreateLab4ProfileTHR(m ->ContextID, NULL);
+	if (hLab == NULL) return 0;
+
+    OutputFormat = cmsFormatterForColorspaceOfProfile(hProfile, 2, FALSE);
+	nChannels    = T_CHANNELS(OutputFormat);
+
+	ColorSpace = cmsGetColorSpace(hProfile);
+
+	// For absolute colorimetric, the LUT is encoded as relative in order to preserve precision.
+
+    RelativeEncodingIntent = Intent;
+	if (RelativeEncodingIntent == INTENT_ABSOLUTE_COLORIMETRIC)
+		RelativeEncodingIntent = INTENT_RELATIVE_COLORIMETRIC;
+
+
+	// Use V4 Lab always
+	Profiles[0] = hLab;
+	Profiles[1] = hProfile;
+
+	xform = cmsCreateMultiprofileTransformTHR(m ->ContextID, 
+		                                      Profiles, 2, TYPE_Lab_DBL, 
+		                                      OutputFormat, RelativeEncodingIntent, 0);
+	cmsCloseProfile(hLab);
+
+    if (xform == NULL) {
+                        
+		cmsSignalError(m ->ContextID, cmsERROR_COLORSPACE_CHECK, "Cannot create transform Lab -> Profile in CRD creation");
+        return 0;
+    }
+
+    // Get a copy of the internal devicelink
+    v = (_cmsTRANSFORM*) xform;
+    DeviceLink = cmsPipelineDup(v ->Lut);
+	if (DeviceLink == NULL) return 0;
+   
+  
+	// We need a CLUT
+	dwFlags |= cmsFLAGS_FORCE_CLUT;
+	_cmsOptimizePipeline(&DeviceLink, RelativeEncodingIntent, &InFrm, &OutputFormat, &dwFlags);
+  
+    _cmsIOPrintf(m, "<<\n");
+    _cmsIOPrintf(m, "/ColorRenderingType 1\n");
+
+
+    cmsDetectBlackPoint(&BlackPointAdaptedToD50, hProfile, Intent, 0);
+
+    // Emit headers, etc.
+    EmitWhiteBlackD50(m, &BlackPointAdaptedToD50);
+    EmitPQRStage(m, hProfile, lDoBPC, Intent == INTENT_ABSOLUTE_COLORIMETRIC);
+    EmitXYZ2Lab(m);
+        
+   
+    // FIXUP: map Lab (100, 0, 0) to perfect white, because the particular encoding for Lab 
+    // does map a=b=0 not falling into any specific node. Since range a,b goes -128..127, 
+    // zero is slightly moved towards right, so assure next node (in L=100 slice) is mapped to
+    // zero. This would sacrifice a bit of highlights, but failure to do so would cause
+    // scum dot. Ouch.
+    
+    if (Intent == INTENT_ABSOLUTE_COLORIMETRIC)
+            lFixWhite = FALSE;
+
+    _cmsIOPrintf(m, "/RenderTable ");
+    
+	
+    WriteCLUT(m, cmsPipelineGetPtrToFirstStage(DeviceLink), "<", ">\n", "", "", lFixWhite, ColorSpace);
+    
+    _cmsIOPrintf(m, " %d {} bind ", nChannels);
+
+    for (i=1; i < nChannels; i++)
+            _cmsIOPrintf(m, "dup ");
+
+    _cmsIOPrintf(m, "]\n");
+
+        
+    EmitIntent(m, Intent);
+
+    _cmsIOPrintf(m, ">>\n");
+
+    if (!(dwFlags & cmsFLAGS_NODEFAULTRESOURCEDEF)) {
+
+        _cmsIOPrintf(m, "/Current exch /ColorRendering defineresource pop\n");
+    }
+
+    cmsPipelineFree(DeviceLink);
+    cmsDeleteTransform(xform);
+ 
+    return 1;   
+}
+
+
+// Builds a ASCII string containing colorant list in 0..1.0 range
+static
+void BuildColorantList(char *Colorant, int nColorant, cmsUInt16Number Out[])
+{
+    char Buff[32];
+    int j;
+
+    Colorant[0] = 0;
+    if (nColorant > cmsMAXCHANNELS)
+        nColorant = cmsMAXCHANNELS;
+
+    for (j=0; j < nColorant; j++) {
+
+                sprintf(Buff, "%.3f", Out[j] / 65535.0);
+                strcat(Colorant, Buff);
+                if (j < nColorant -1) 
+                        strcat(Colorant, " ");
+
+        }       
+}
+
+
+// Creates a PostScript color list from a named profile data. 
+// This is a HP extension.
+
+static
+int WriteNamedColorCRD(cmsIOHANDLER* m, cmsHPROFILE hNamedColor, int Intent, cmsUInt32Number dwFlags)
+{
+    cmsHTRANSFORM xform;    
+    int i, nColors, nColorant;
+    cmsUInt32Number OutputFormat;
+    char ColorName[32];
+    char Colorant[128];
+	cmsNAMEDCOLORLIST* NamedColorList;
+
+	
+    OutputFormat = cmsFormatterForColorspaceOfProfile(hNamedColor, 2, FALSE);
+	nColorant    = T_CHANNELS(OutputFormat);
+
+	
+    xform = cmsCreateTransform(hNamedColor, TYPE_NAMED_COLOR_INDEX, NULL, OutputFormat, Intent, dwFlags);
+    if (xform == NULL) return 0;
+
+
+	NamedColorList = cmsGetNamedColorList(xform);
+	if (NamedColorList == NULL) return 0;
+
+    _cmsIOPrintf(m, "<<\n");
+    _cmsIOPrintf(m, "(colorlistcomment) (%s) \n", "Named profile");
+    _cmsIOPrintf(m, "(Prefix) [ (Pantone ) (PANTONE ) ]\n");
+    _cmsIOPrintf(m, "(Suffix) [ ( CV) ( CVC) ( C) ]\n");
+
+    nColors   = cmsNamedColorCount(NamedColorList);
+    
+    for (i=0; i < nColors; i++) {
+        
+        cmsUInt16Number In[1];
+        cmsUInt16Number Out[cmsMAXCHANNELS];
+
+        In[0] = (cmsUInt16Number) i;
+
+        if (!cmsNamedColorInfo(NamedColorList, i, ColorName, NULL, NULL, NULL, NULL))
+                continue;
+
+        cmsDoTransform(xform, In, Out, 1);      
+        BuildColorantList(Colorant, nColorant, Out);
+        _cmsIOPrintf(m, "  (%s) [ %s ]\n", ColorName, Colorant);
+    }
+
+    _cmsIOPrintf(m, "   >>");
+
+    if (!(dwFlags & cmsFLAGS_NODEFAULTRESOURCEDEF)) {
+
+    _cmsIOPrintf(m, " /Current exch /HPSpotTable defineresource pop\n");
+    }
+
+    cmsDeleteTransform(xform);  
+    return 1;
+}
+
+
+
+// This one does create a Color Rendering Dictionary. 
+// CRD are always LUT-Based, no matter if profile is
+// implemented as matrix-shaper.
+
+static
+cmsUInt32Number  GenerateCRD(cmsContext ContextID,
+							 cmsHPROFILE hProfile, 
+							 cmsUInt32Number Intent, cmsUInt32Number dwFlags,
+							 cmsIOHANDLER* mem)
+{    
+	cmsUInt32Number dwBytesUsed;
+
+	if (!(dwFlags & cmsFLAGS_NODEFAULTRESOURCEDEF)) {
+
+		EmitHeader(mem, "Color Rendering Dictionary (CRD)", hProfile);
+	}
+
+
+	// Is a named color profile?
+	if (cmsGetDeviceClass(hProfile) == cmsSigNamedColorClass) {
+
+		if (!WriteNamedColorCRD(mem, hProfile, Intent, dwFlags)) {
+			return 0;
+		}
+	}
+	else {
+
+		// CRD are always implemented as LUT 
+
+		if (!WriteOutputLUT(mem, hProfile, Intent, dwFlags)) {
+			return 0;
+		}
+	}
+
+	if (!(dwFlags & cmsFLAGS_NODEFAULTRESOURCEDEF)) {
+
+		_cmsIOPrintf(mem, "%%%%EndResource\n");
+		_cmsIOPrintf(mem, "\n%% CRD End\n");
+	}
+
+	// Done, keep memory usage
+	dwBytesUsed = mem ->UsedSpace;
+
+	// Finally, return used byte count
+	return dwBytesUsed;
+
+	cmsUNUSED_PARAMETER(ContextID);
+}
+
+
+
+
+cmsUInt32Number CMSEXPORT cmsGetPostScriptColorResource(cmsContext ContextID, 															   
+															   cmsPSResourceType Type,
+															   cmsHPROFILE hProfile, 
+															   cmsUInt32Number Intent, 
+															   cmsUInt32Number dwFlags, 
+															   cmsIOHANDLER* io)
+{
+	cmsUInt32Number  rc; 
+
+
+	switch (Type) {
+
+		case cmsPS_RESOURCE_CSA:
+			rc = GenerateCSA(ContextID, hProfile, Intent, dwFlags, io);
+			break;
+			
+		default:
+		case cmsPS_RESOURCE_CRD:
+			rc = GenerateCRD(ContextID, hProfile, Intent, dwFlags, io);
+			break;
+	}
+
+	return rc;
+}
+
+
+
+cmsUInt32Number CMSEXPORT cmsGetPostScriptCRD(cmsContext ContextID,
+	                          cmsHPROFILE hProfile, 
+                              cmsUInt32Number Intent, cmsUInt32Number dwFlags,
+                              void* Buffer, cmsUInt32Number dwBufferLen)
+{
+	cmsIOHANDLER* mem;
+    cmsUInt32Number dwBytesUsed;
+
+    // Set up the serialization engine
+	if (Buffer == NULL)
+		mem = cmsOpenIOhandlerFromNULL(ContextID);
+	else
+        mem = cmsOpenIOhandlerFromMem(ContextID, Buffer, dwBufferLen, "w");
+
+    if (!mem) return 0;
+
+	dwBytesUsed =  cmsGetPostScriptColorResource(ContextID, cmsPS_RESOURCE_CRD, hProfile, Intent, dwFlags, mem);
+															   
+    // Get rid of memory stream
+	cmsCloseIOhandler(mem);
+
+	return dwBytesUsed;
+}
+
+
+
+// Does create a Color Space Array on XYZ colorspace for PostScript usage
+cmsUInt32Number CMSEXPORT cmsGetPostScriptCSA(cmsContext ContextID, 
+											  cmsHPROFILE hProfile, 
+											  cmsUInt32Number Intent, 
+											  cmsUInt32Number dwFlags, 
+											  void* Buffer, 
+											  cmsUInt32Number dwBufferLen)  
+{
+	cmsIOHANDLER* mem;
+    cmsUInt32Number dwBytesUsed;
+    
+	if (Buffer == NULL)
+		mem = cmsOpenIOhandlerFromNULL(ContextID);
+	else
+        mem = cmsOpenIOhandlerFromMem(ContextID, Buffer, dwBufferLen, "w");
+
+    if (!mem) return 0;
+
+	dwBytesUsed =  cmsGetPostScriptColorResource(ContextID, cmsPS_RESOURCE_CSA, hProfile, Intent, dwFlags, mem);
+															   
+    // Get rid of memory stream
+	cmsCloseIOhandler(mem);
+
+	return dwBytesUsed;
+
+}
diff --git a/thirdparty/liblcms2/src/cmssamp.c b/thirdparty/liblcms2/src/cmssamp.c
new file mode 100644
index 00000000..090d96de
--- /dev/null
+++ b/thirdparty/liblcms2/src/cmssamp.c
@@ -0,0 +1,266 @@
+//---------------------------------------------------------------------------------
+//
+//  Little Color Management System
+//  Copyright (c) 1998-2010 Marti Maria Saguer
+//
+// Permission is hereby granted, free of charge, to any person obtaining 
+// a copy of this software and associated documentation files (the "Software"), 
+// to deal in the Software without restriction, including without limitation 
+// the rights to use, copy, modify, merge, publish, distribute, sublicense, 
+// and/or sell copies of the Software, and to permit persons to whom the Software 
+// is furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in 
+// all copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 
+// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO 
+// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE 
+// LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 
+// OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 
+// WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+//
+//---------------------------------------------------------------------------------
+//
+
+#include "lcms2_internal.h"
+
+
+
+// This file contains routines for resampling and LUT optimization, black point detection
+// and black preservation. 
+
+// Black point detection -------------------------------------------------------------------------
+
+
+// PCS -> PCS round trip transform, always uses relative intent on the device -> pcs 
+static
+cmsHTRANSFORM CreateRoundtripXForm(cmsHPROFILE hProfile, cmsUInt32Number nIntent)
+{
+    cmsHPROFILE hLab = cmsCreateLab4Profile(NULL);
+    cmsHTRANSFORM xform;
+    cmsBool BPC[4] = { FALSE, FALSE, FALSE, FALSE };
+    cmsFloat64Number States[4] = { 1.0, 1.0, 1.0, 1.0 };
+    cmsHPROFILE hProfiles[4];
+    cmsUInt32Number Intents[4];
+    cmsContext ContextID = cmsGetProfileContextID(hProfile);
+
+    hProfiles[0] = hLab; hProfiles[1] = hProfile; hProfiles[2] = hProfile; hProfiles[3] = hLab;
+    Intents[0]   = INTENT_RELATIVE_COLORIMETRIC; Intents[1] = nIntent; Intents[2] = INTENT_RELATIVE_COLORIMETRIC; Intents[3] = INTENT_RELATIVE_COLORIMETRIC;
+
+    xform =  cmsCreateExtendedTransform(ContextID, 4, hProfiles, BPC, Intents, 
+        States, NULL, 0, TYPE_Lab_DBL, TYPE_Lab_DBL, cmsFLAGS_NOCACHE|cmsFLAGS_NOOPTIMIZE);
+
+    cmsCloseProfile(hLab);
+    return xform;
+}
+
+// Use darker colorants to obtain black point. This works in the relative colorimetric intent and
+// assumes more ink results in darker colors. No ink limit is assumed.
+static
+cmsBool  BlackPointAsDarkerColorant(cmsHPROFILE    hInput,                               
+                                    cmsUInt32Number Intent,
+                                    cmsCIEXYZ* BlackPoint,
+                                    cmsUInt32Number dwFlags)
+{
+    cmsUInt16Number *Black;
+    cmsHTRANSFORM xform;
+    cmsColorSpaceSignature Space;
+    cmsUInt32Number nChannels;
+    cmsUInt32Number dwFormat; 
+    cmsHPROFILE hLab;
+    cmsCIELab  Lab;
+    cmsCIEXYZ  BlackXYZ;        
+    cmsContext ContextID = cmsGetProfileContextID(hInput);
+    
+    // If the profile does not support input direction, assume Black point 0    
+    if (!cmsIsIntentSupported(hInput, Intent, LCMS_USED_AS_INPUT)) {
+
+        BlackPoint -> X = BlackPoint ->Y = BlackPoint -> Z = 0.0;
+        return FALSE;
+    }
+    
+    // Create a formatter which has n channels and floating point
+    dwFormat = cmsFormatterForColorspaceOfProfile(hInput, 2, FALSE);
+
+   // Try to get black by using black colorant    
+    Space = cmsGetColorSpace(hInput);
+
+    // This function returns darker colorant in 16 bits for several spaces
+    if (!_cmsEndPointsBySpace(Space, NULL, &Black, &nChannels)) {
+        
+        BlackPoint -> X = BlackPoint ->Y = BlackPoint -> Z = 0.0;
+        return FALSE;
+    }
+
+    if (nChannels != T_CHANNELS(dwFormat)) {
+       BlackPoint -> X = BlackPoint ->Y = BlackPoint -> Z = 0.0;
+       return FALSE;
+    }
+
+    // Lab will be used as the output space, but lab2 will avoid recursion
+    hLab = cmsCreateLab2ProfileTHR(ContextID, NULL);
+    if (hLab == NULL) {
+       BlackPoint -> X = BlackPoint ->Y = BlackPoint -> Z = 0.0;
+       return FALSE;    
+    }
+
+    // Create the transform
+    xform = cmsCreateTransformTHR(ContextID, hInput, dwFormat,
+                                hLab, TYPE_Lab_DBL, Intent, cmsFLAGS_NOOPTIMIZE|cmsFLAGS_NOCACHE);
+    cmsCloseProfile(hLab);
+    
+    if (xform == NULL) {
+        // Something went wrong. Get rid of open resources and return zero as black
+        
+        BlackPoint -> X = BlackPoint ->Y = BlackPoint -> Z = 0.0;
+        return FALSE;
+    }
+    
+    // Convert black to Lab
+    cmsDoTransform(xform, Black, &Lab, 1);
+
+    // Force it to be neutral, clip to max. L* of 50
+    Lab.a = Lab.b = 0;
+    if (Lab.L > 50) Lab.L = 50;
+
+    // Free the resources    
+    cmsDeleteTransform(xform);
+    
+    // Convert from Lab (which is now clipped) to XYZ.
+    cmsLab2XYZ(NULL, &BlackXYZ, &Lab);
+    
+    if (BlackPoint != NULL)
+        *BlackPoint = BlackXYZ;
+      
+    return TRUE;
+
+    cmsUNUSED_PARAMETER(dwFlags);
+}
+
+// Get a black point of output CMYK profile, discounting any ink-limiting embedded 
+// in the profile. For doing that, we use perceptual intent in input direction:
+// Lab (0, 0, 0) -> [Perceptual] Profile -> CMYK -> [Rel. colorimetric] Profile -> Lab
+static
+cmsBool BlackPointUsingPerceptualBlack(cmsCIEXYZ* BlackPoint, cmsHPROFILE hProfile)
+                                   
+{    
+    cmsHTRANSFORM hRoundTrip;    
+    cmsCIELab LabIn, LabOut;
+    cmsCIEXYZ  BlackXYZ;        
+ 
+     // Is the intent supported by the profile?
+    if (!cmsIsIntentSupported(hProfile, INTENT_PERCEPTUAL, LCMS_USED_AS_INPUT)) {
+
+        BlackPoint -> X = BlackPoint ->Y = BlackPoint -> Z = 0.0;
+        return TRUE;
+    }
+        
+    hRoundTrip = CreateRoundtripXForm(hProfile, INTENT_PERCEPTUAL);
+    if (hRoundTrip == NULL) {
+        BlackPoint -> X = BlackPoint ->Y = BlackPoint -> Z = 0.0;
+        return FALSE;
+    }
+
+    LabIn.L = LabIn.a = LabIn.b = 0;
+    cmsDoTransform(hRoundTrip, &LabIn, &LabOut, 1);
+
+    // Clip Lab to reasonable limits
+    if (LabOut.L > 50) LabOut.L = 50;
+    LabOut.a = LabOut.b = 0;
+
+    cmsDeleteTransform(hRoundTrip);
+  
+    // Convert it to XYZ
+    cmsLab2XYZ(NULL, &BlackXYZ, &LabOut);   
+    
+    if (BlackPoint != NULL)
+        *BlackPoint = BlackXYZ;
+
+    return TRUE;
+}
+
+// This function shouldn't exist at all -- there is such quantity of broken
+// profiles on black point tag, that we must somehow fix chromaticity to 
+// avoid huge tint when doing Black point compensation. This function does
+// just that. There is a special flag for using black point tag, but turned 
+// off by default because it is bogus on most profiles. The detection algorithm 
+// involves to turn BP to neutral and to use only L component.  
+
+cmsBool CMSEXPORT cmsDetectBlackPoint(cmsCIEXYZ* BlackPoint, cmsHPROFILE hProfile, cmsUInt32Number Intent, cmsUInt32Number dwFlags)
+{    
+
+    // Zero for black point
+    if (cmsGetDeviceClass(hProfile) == cmsSigLinkClass) {
+
+        BlackPoint -> X = BlackPoint ->Y = BlackPoint -> Z = 0.0;
+        return FALSE;       
+    }
+
+    // v4 + perceptual & saturation intents does have its own black point, and it is 
+    // well specified enough to use it. Black point tag is deprecated in V4.
+
+    if ((cmsGetEncodedICCversion(hProfile) >= 0x4000000) &&     
+        (Intent == INTENT_PERCEPTUAL || Intent == INTENT_SATURATION)) {
+
+            // Matrix shaper share MRC & perceptual intents
+            if (cmsIsMatrixShaper(hProfile)) 
+                return BlackPointAsDarkerColorant(hProfile, INTENT_RELATIVE_COLORIMETRIC, BlackPoint, 0);
+
+            // Get Perceptual black out of v4 profiles. That is fixed for perceptual & saturation intents
+            BlackPoint -> X = cmsPERCEPTUAL_BLACK_X; 
+            BlackPoint -> Y = cmsPERCEPTUAL_BLACK_Y;
+            BlackPoint -> Z = cmsPERCEPTUAL_BLACK_Z;
+
+            return TRUE;
+    }
+
+
+#ifdef CMS_USE_PROFILE_BLACK_POINT_TAG
+
+    // v2, v4 rel/abs colorimetric
+    if (cmsIsTag(hProfile, cmsSigMediaBlackPointTag) && 
+        Intent == INTENT_RELATIVE_COLORIMETRIC) {
+
+            cmsCIEXYZ *BlackPtr, BlackXYZ, UntrustedBlackPoint, TrustedBlackPoint, MediaWhite;
+            cmsCIELab Lab;
+
+            // If black point is specified, then use it, 
+
+            BlackPtr = cmsReadTag(hProfile, cmsSigMediaBlackPointTag);
+            if (BlackPtr != NULL) {
+
+                BlackXYZ = *BlackPtr;
+                _cmsReadMediaWhitePoint(&MediaWhite, hProfile);
+
+                // Black point is absolute XYZ, so adapt to D50 to get PCS value
+                cmsAdaptToIlluminant(&UntrustedBlackPoint, &MediaWhite, cmsD50_XYZ(), &BlackXYZ);
+
+                // Force a=b=0 to get rid of any chroma
+                cmsXYZ2Lab(NULL, &Lab, &UntrustedBlackPoint);
+                Lab.a = Lab.b = 0;
+                if (Lab.L > 50) Lab.L = 50; // Clip to L* <= 50
+                cmsLab2XYZ(NULL, &TrustedBlackPoint, &Lab);
+
+                if (BlackPoint != NULL)
+                    *BlackPoint = TrustedBlackPoint;
+                
+                return TRUE;
+            }
+    }
+#endif
+
+    // That is about v2 profiles. 
+
+    // If output profile, discount ink-limiting and that's all
+    if (Intent == INTENT_RELATIVE_COLORIMETRIC && 
+        (cmsGetDeviceClass(hProfile) == cmsSigOutputClass) &&
+        (cmsGetColorSpace(hProfile)  == cmsSigCmykData))
+        return BlackPointUsingPerceptualBlack(BlackPoint, hProfile);
+
+    // Nope, compute BP using current intent.
+    return BlackPointAsDarkerColorant(hProfile, Intent, BlackPoint, dwFlags);
+}
+
+
diff --git a/thirdparty/liblcms2/src/cmssm.c b/thirdparty/liblcms2/src/cmssm.c
new file mode 100644
index 00000000..e5a6b0d5
--- /dev/null
+++ b/thirdparty/liblcms2/src/cmssm.c
@@ -0,0 +1,734 @@
+//---------------------------------------------------------------------------------
+//
+//  Little Color Management System
+//  Copyright (c) 1998-2010 Marti Maria Saguer
+//
+// Permission is hereby granted, free of charge, to any person obtaining 
+// a copy of this software and associated documentation files (the "Software"), 
+// to deal in the Software without restriction, including without limitation 
+// the rights to use, copy, modify, merge, publish, distribute, sublicense, 
+// and/or sell copies of the Software, and to permit persons to whom the Software 
+// is furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in 
+// all copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 
+// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO 
+// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE 
+// LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 
+// OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 
+// WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+//
+//---------------------------------------------------------------------------------
+//
+
+#include "lcms2_internal.h"
+
+
+// ------------------------------------------------------------------------
+
+// Gamut boundary description by using Jan Morovic's Segment maxima method
+// Many thanks to Jan for allowing me to use his algorithm.
+
+// r = C*
+// alpha = Hab
+// theta = L*
+
+#define SECTORS 16      // number of divisions in alpha and theta 
+
+// Spherical coordinates
+typedef struct {
+
+    cmsFloat64Number r;
+    cmsFloat64Number alpha;
+    cmsFloat64Number theta;
+
+} cmsSpherical;
+
+typedef  enum {          
+        GP_EMPTY,
+        GP_SPECIFIED,
+        GP_MODELED
+
+    } GDBPointType;
+
+
+typedef struct {
+
+    GDBPointType Type;
+    cmsSpherical p;         // Keep also alpha & theta of maximum
+
+} cmsGDBPoint;
+
+
+typedef struct {
+
+    cmsContext ContextID;
+    cmsGDBPoint Gamut[SECTORS][SECTORS];
+
+} cmsGDB;
+
+
+// A line using the parametric form
+// P = a + t*u
+typedef struct {
+
+    cmsVEC3 a;
+    cmsVEC3 u;
+
+} cmsLine;
+
+
+// A plane using the parametric form
+// Q = b + r*v + s*w
+typedef struct {
+
+    cmsVEC3 b;
+    cmsVEC3 v;
+    cmsVEC3 w;
+
+} cmsPlane;
+
+
+
+// --------------------------------------------------------------------------------------------
+
+// ATAN2() which always returns degree positive numbers
+
+static
+cmsFloat64Number _cmsAtan2(cmsFloat64Number y, cmsFloat64Number x)
+{
+    cmsFloat64Number a;
+    
+    // Deal with undefined case
+    if (x == 0.0 && y == 0.0) return 0;
+
+    a = (atan2(y, x) * 180.0) / M_PI;
+
+    while (a < 0) {
+        a += 360;
+    }
+
+    return a;
+}
+
+// Convert to spherical coordinates
+static
+void ToSpherical(cmsSpherical* sp, const cmsVEC3* v)
+{
+
+    cmsFloat64Number L, a, b;
+    
+    L = v ->n[VX];
+    a = v ->n[VY];
+    b = v ->n[VZ];
+    
+    sp ->r = sqrt( L*L + a*a + b*b );
+
+   if (sp ->r == 0) {
+        sp ->alpha = sp ->theta = 0;
+        return;
+    }
+	    
+    sp ->alpha = _cmsAtan2(a, b);   
+    sp ->theta = _cmsAtan2(sqrt(a*a + b*b), L);  
+}
+
+
+// Convert to cartesian from spherical
+static
+void ToCartesian(cmsVEC3* v, const cmsSpherical* sp)
+{
+    cmsFloat64Number sin_alpha;
+    cmsFloat64Number cos_alpha;
+    cmsFloat64Number sin_theta;
+    cmsFloat64Number cos_theta;
+    cmsFloat64Number L, a, b;
+
+    sin_alpha = sin((M_PI * sp ->alpha) / 180.0);
+    cos_alpha = cos((M_PI * sp ->alpha) / 180.0);
+    sin_theta = sin((M_PI * sp ->theta) / 180.0);
+    cos_theta = cos((M_PI * sp ->theta) / 180.0);
+
+    a = sp ->r * sin_theta * sin_alpha;
+    b = sp ->r * sin_theta * cos_alpha; 
+    L = sp ->r * cos_theta;
+
+    v ->n[VX] = L;
+    v ->n[VY] = a;
+    v ->n[VZ] = b;
+}
+
+
+// Quantize sector of a spherical coordinate. Saturate 360, 180 to last sector
+// The limits are the centers of each sector, so
+static
+void QuantizeToSector(const cmsSpherical* sp, int* alpha, int* theta)
+{   
+    *alpha = (int) floor(((sp->alpha * (SECTORS)) / 360.0) );            
+    *theta = (int) floor(((sp->theta * (SECTORS)) / 180.0) );    
+
+	if (*alpha >= SECTORS)
+		*alpha = SECTORS-1;
+	if (*theta >= SECTORS)
+		*theta = SECTORS-1;
+}
+
+
+// Line determined by 2 points
+static
+void LineOf2Points(cmsLine* line, cmsVEC3* a, cmsVEC3* b)
+{
+
+    _cmsVEC3init(&line ->a, a ->n[VX], a ->n[VY], a ->n[VZ]);
+    _cmsVEC3init(&line ->u, b ->n[VX] - a ->n[VX], 
+                            b ->n[VY] - a ->n[VY], 
+                            b ->n[VZ] - a ->n[VZ]);     
+}
+
+
+// Evaluate parametric line 
+static
+void GetPointOfLine(cmsVEC3* p, const cmsLine* line, cmsFloat64Number t)
+{
+    p ->n[VX] = line ->a.n[VX] + t * line->u.n[VX];
+    p ->n[VY] = line ->a.n[VY] + t * line->u.n[VY];
+    p ->n[VZ] = line ->a.n[VZ] + t * line->u.n[VZ];     
+}
+
+
+
+/*
+    Closest point in sector line1 to sector line2 (both are defined as 0 <=t <= 1)
+    http://softsurfer.com/Archive/algorithm_0106/algorithm_0106.htm
+
+    Copyright 2001, softSurfer (www.softsurfer.com)
+    This code may be freely used and modified for any purpose
+    providing that this copyright notice is included with it.
+    SoftSurfer makes no warranty for this code, and cannot be held
+    liable for any real or imagined damage resulting from its use.
+    Users of this code must verify correctness for their application.
+
+*/
+
+static
+cmsBool ClosestLineToLine(cmsVEC3* r, const cmsLine* line1, const cmsLine* line2)
+{
+    cmsFloat64Number a, b, c, d, e, D;
+    cmsFloat64Number sc, sN, sD; 
+    cmsFloat64Number tc, tN, tD;
+    cmsVEC3 w0;
+
+    _cmsVEC3minus(&w0, &line1 ->a, &line2 ->a);
+
+    a  = _cmsVEC3dot(&line1 ->u, &line1 ->u);
+    b  = _cmsVEC3dot(&line1 ->u, &line2 ->u);
+    c  = _cmsVEC3dot(&line2 ->u, &line2 ->u);
+    d  = _cmsVEC3dot(&line1 ->u, &w0);
+    e  = _cmsVEC3dot(&line2 ->u, &w0);
+
+    D  = a*c - b * b;      // Denominator
+    sD = tD = D;           // default sD = D >= 0
+
+    if (D <  MATRIX_DET_TOLERANCE) {   // the lines are almost parallel
+
+        sN = 0.0;        // force using point P0 on segment S1
+        sD = 1.0;        // to prevent possible division by 0.0 later
+        tN = e;
+        tD = c;
+    }
+    else {                // get the closest points on the infinite lines
+
+        sN = (b*e - c*d);
+        tN = (a*e - b*d);
+
+        if (sN < 0.0) {       // sc < 0 => the s=0 edge is visible
+
+            sN = 0.0;
+            tN = e;
+            tD = c;
+        }
+        else if (sN > sD) {   // sc > 1 => the s=1 edge is visible
+            sN = sD;
+            tN = e + b;
+            tD = c;
+        }
+    }
+
+    if (tN < 0.0) {           // tc < 0 => the t=0 edge is visible
+
+        tN = 0.0;
+        // recompute sc for this edge
+        if (-d < 0.0)
+            sN = 0.0;
+        else if (-d > a)
+            sN = sD;
+        else {
+            sN = -d;
+            sD = a;
+        }
+    }
+    else if (tN > tD) {      // tc > 1 => the t=1 edge is visible
+        
+        tN = tD;
+        
+        // recompute sc for this edge
+        if ((-d + b) < 0.0)
+            sN = 0;
+        else if ((-d + b) > a)
+            sN = sD;
+        else {
+            sN = (-d + b);
+            sD = a;
+        }
+    }
+    // finally do the division to get sc and tc
+    sc = (fabs(sN) < MATRIX_DET_TOLERANCE ? 0.0 : sN / sD);
+    tc = (fabs(tN) < MATRIX_DET_TOLERANCE ? 0.0 : tN / tD);
+
+    GetPointOfLine(r, line1, sc);
+    return TRUE;
+}
+
+
+
+// ------------------------------------------------------------------ Wrapper
+
+
+// Allocate & free structure
+cmsHANDLE  CMSEXPORT cmsGBDAlloc(cmsContext ContextID)
+{
+    cmsGDB* gbd = (cmsGDB*) _cmsMallocZero(ContextID, sizeof(cmsGDB));
+    if (gbd == NULL) return NULL;
+    
+    gbd -> ContextID = ContextID;
+
+    return (cmsHANDLE) gbd;
+}
+
+
+void CMSEXPORT cmsGBDFree(cmsHANDLE hGBD)
+{   
+    cmsGDB* gbd = (cmsGDB*) hGBD;
+    if (hGBD != NULL) 
+        _cmsFree(gbd->ContextID, (void*) gbd);
+}
+
+
+// Auxiliar to retrieve a pointer to the segmentr containing the Lab value
+static
+cmsGDBPoint* GetPoint(cmsGDB* gbd, const cmsCIELab* Lab, cmsSpherical* sp)
+{
+    cmsVEC3 v;  
+    int alpha, theta;
+
+    // Housekeeping
+    _cmsAssert(gbd != NULL);
+	_cmsAssert(Lab != NULL);
+	_cmsAssert(sp != NULL);
+
+    // Center L* by substracting half of its domain, that's 50 
+    _cmsVEC3init(&v, Lab ->L - 50.0, Lab ->a, Lab ->b);
+	
+    // Convert to spherical coordinates
+    ToSpherical(sp, &v);
+    
+    if (sp ->r < 0 || sp ->alpha < 0 || sp->theta < 0) {
+         cmsSignalError(gbd ->ContextID, cmsERROR_RANGE, "spherical value out of range");
+         return NULL;
+    }
+
+    // On which sector it falls?
+    QuantizeToSector(sp, &alpha, &theta);
+  	
+    if (alpha < 0 || theta < 0 || alpha >= SECTORS || theta >= SECTORS) {
+         cmsSignalError(gbd ->ContextID, cmsERROR_RANGE, " quadrant out of range");
+         return NULL;
+    }
+
+    // Get pointer to the sector
+    return &gbd ->Gamut[theta][alpha];
+}
+
+// Add a point to gamut descriptor. Point to add is in Lab color space. 
+// GBD is centered on a=b=0 and L*=50
+cmsBool CMSEXPORT cmsGDBAddPoint(cmsHANDLE hGBD, const cmsCIELab* Lab)
+{
+    cmsGDB* gbd = (cmsGDB*) hGBD;
+    cmsGDBPoint* ptr;
+    cmsSpherical sp;
+
+
+    // Get pointer to the sector
+    ptr = GetPoint(gbd, Lab, &sp);
+    if (ptr == NULL) return FALSE;
+
+    // If no samples at this sector, add it
+    if (ptr ->Type == GP_EMPTY) {
+
+        ptr -> Type = GP_SPECIFIED;
+        ptr -> p    = sp;
+    }
+    else {
+
+
+        // Substitute only if radius is greater
+        if (sp.r > ptr -> p.r) {
+
+                ptr -> Type = GP_SPECIFIED;
+                ptr -> p    = sp;
+        }
+    }
+
+    return TRUE;
+}
+
+// Check if a given point falls inside gamut
+cmsBool CMSEXPORT cmsGDBCheckPoint(cmsHANDLE hGBD, const cmsCIELab* Lab)
+{
+    cmsGDB* gbd = (cmsGDB*) hGBD;   
+    cmsGDBPoint* ptr;
+    cmsSpherical sp;
+
+    // Get pointer to the sector
+    ptr = GetPoint(gbd, Lab, &sp);
+    if (ptr == NULL) return FALSE;
+
+    // If no samples at this sector, return no data
+    if (ptr ->Type == GP_EMPTY) return FALSE;
+
+    // In gamut only if radius is greater
+
+    return (sp.r <= ptr -> p.r);
+}
+
+// -----------------------------------------------------------------------------------------------------------------------
+
+// Find near sectors. The list of sectors found is returned on Close[]. 
+// The function returns the number of sectors as well.
+
+// 24   9  10  11  12
+// 23   8   1   2  13
+// 22   7   *   3  14
+// 21   6   5   4  15
+// 20  19  18  17  16
+//
+// Those are the relative movements
+// {-2,-2}, {-1, -2}, {0, -2}, {+1, -2}, {+2,  -2}, 
+// {-2,-1}, {-1, -1}, {0, -1}, {+1, -1}, {+2,  -1},
+// {-2, 0}, {-1,  0}, {0,  0}, {+1,  0}, {+2,   0},
+// {-2,+1}, {-1, +1}, {0, +1}, {+1,  +1}, {+2,  +1},
+// {-2,+2}, {-1, +2}, {0, +2}, {+1,  +2}, {+2,  +2}};
+
+
+static 
+const struct _spiral { 
+    
+    int AdvX, AdvY;
+    
+    } Spiral[] = { {0,  -1}, {+1, -1}, {+1,  0}, {+1, +1}, {0,  +1}, {-1, +1}, 
+                   {-1,  0}, {-1, -1}, {-1, -2}, {0,  -2}, {+1, -2}, {+2, -2}, 
+                   {+2, -1}, {+2,  0}, {+2, +1}, {+2, +2}, {+1, +2}, {0,  +2}, 
+                   {-1, +2}, {-2, +2}, {-2, +1}, {-2, 0},  {-2, -1}, {-2, -2} };
+
+#define NSTEPS (sizeof(Spiral) / sizeof(struct _spiral))
+
+static
+int FindNearSectors(cmsGDB* gbd, int alpha, int theta, cmsGDBPoint* Close[])
+{
+    int nSectors = 0;
+    int i, a, t;
+    cmsGDBPoint* pt;
+
+    for (i=0; i < NSTEPS; i++) {
+
+        a = alpha + Spiral[i].AdvX;
+        t = theta + Spiral[i].AdvY;
+
+        // Cycle at the end
+        a %= SECTORS;
+        t %= SECTORS;
+
+        // Cycle at the begin
+        if (a < 0) a = SECTORS + a;
+        if (t < 0) t = SECTORS + t;   
+
+        pt = &gbd ->Gamut[t][a];
+        
+        if (pt -> Type != GP_EMPTY) {
+
+            Close[nSectors++] = pt;
+        }                           
+    }
+
+    return nSectors;
+}
+
+
+// Interpolate a missing sector. Method identifies whatever this is top, bottom or mid
+static
+cmsBool InterpolateMissingSector(cmsGDB* gbd, int alpha, int theta)
+{   
+    cmsSpherical sp;
+    cmsVEC3 Lab;
+    cmsVEC3 Centre;
+    cmsLine ray;
+    int nCloseSectors;
+    cmsGDBPoint* Close[NSTEPS]; 
+    cmsSpherical closel, templ;
+    cmsLine edge;
+    int k, m;
+    
+    // Is that point already specified?
+    if (gbd ->Gamut[theta][alpha].Type != GP_EMPTY) return TRUE;
+
+    // Fill close points
+    nCloseSectors = FindNearSectors(gbd, alpha, theta, Close);
+
+
+    // Find a central point on the sector 
+    sp.alpha = (cmsFloat64Number) ((alpha + 0.5) * 360.0) / (SECTORS);
+    sp.theta = (cmsFloat64Number) ((theta + 0.5) * 180.0) / (SECTORS);
+    sp.r     = 50.0; 
+
+    // Convert to Cartesian
+    ToCartesian(&Lab, &sp);
+
+    // Create a ray line from centre to this point
+    _cmsVEC3init(&Centre, 50.0, 0, 0);
+    LineOf2Points(&ray, &Lab, &Centre);
+
+    // For all close sectors
+    closel.r = 0.0;
+    closel.alpha = 0;
+    closel.theta = 0;
+
+    for (k=0; k < nCloseSectors; k++) {
+
+        for(m = k+1; m < nCloseSectors; m++) {
+
+            cmsVEC3 temp, a1, a2;
+        
+            // A line from sector to sector
+            ToCartesian(&a1, &Close[k]->p);
+            ToCartesian(&a2, &Close[m]->p);
+
+            LineOf2Points(&edge, &a1, &a2);
+
+            // Find a line  
+            ClosestLineToLine(&temp, &ray, &edge);
+
+            // Convert to spherical
+            ToSpherical(&templ, &temp);
+            
+
+            if ( templ.r > closel.r && 
+                 templ.theta >= (theta*180.0/SECTORS) && 
+                 templ.theta <= ((theta+1)*180.0/SECTORS) &&
+                 templ.alpha >= (alpha*360.0/SECTORS) &&
+                 templ.alpha <= ((alpha+1)*360.0/SECTORS)) {
+
+                closel = templ;         
+            }       
+        }
+    }
+
+    gbd ->Gamut[theta][alpha].p = closel;
+    gbd ->Gamut[theta][alpha].Type = GP_MODELED;
+
+    return TRUE;
+
+}
+
+
+// Interpolate missing parts. The algorithm fist computes slices at
+// theta=0 and theta=Max.
+cmsBool CMSEXPORT cmsGDBCompute(cmsHANDLE hGBD, cmsUInt32Number dwFlags)
+{
+    int alpha, theta;
+    cmsGDB* gbd = (cmsGDB*) hGBD;
+
+    _cmsAssert(hGBD != NULL);
+
+    // Interpolate black
+    for (alpha = 0; alpha <= SECTORS; alpha++) {
+
+        if (!InterpolateMissingSector(gbd, alpha, 0)) return FALSE;
+    }
+
+    // Interpolate white
+    for (alpha = 0; alpha <= SECTORS; alpha++) {
+
+        if (!InterpolateMissingSector(gbd, alpha, SECTORS-1)) return FALSE;
+    }
+
+
+    // Interpolate Mid
+    for (theta = 1; theta < SECTORS; theta++) {
+        for (alpha = 0; alpha <= SECTORS; alpha++) {
+
+            if (!InterpolateMissingSector(gbd, alpha, theta)) return FALSE;
+        }
+    }
+
+    // Done
+    return TRUE;
+
+    cmsUNUSED_PARAMETER(dwFlags);
+}
+
+
+
+
+// --------------------------------------------------------------------------------------------------------
+
+// Great for debug, but not suitable for real use
+
+#if 0
+cmsBool cmsGBDdumpVRML(cmsHANDLE hGBD, const char* fname)
+{
+    FILE* fp;
+    int   i, j;
+    cmsGDB* gbd = (cmsGDB*) hGBD;
+    cmsGDBPoint* pt;
+
+    fp = fopen (fname, "wt");
+    if (fp == NULL)
+        return FALSE;
+
+    fprintf (fp, "#VRML V2.0 utf8\n");
+
+    // set the viewing orientation and distance 
+    fprintf (fp, "DEF CamTest Group {\n");
+    fprintf (fp, "\tchildren [\n"); 
+    fprintf (fp, "\t\tDEF Cameras Group {\n"); 
+    fprintf (fp, "\t\t\tchildren [\n"); 
+    fprintf (fp, "\t\t\t\tDEF DefaultView Viewpoint {\n"); 
+    fprintf (fp, "\t\t\t\t\tposition 0 0 340\n"); 
+    fprintf (fp, "\t\t\t\t\torientation 0 0 1 0\n"); 
+    fprintf (fp, "\t\t\t\t\tdescription \"default view\"\n"); 
+    fprintf (fp, "\t\t\t\t}\n"); 
+    fprintf (fp, "\t\t\t]\n"); 
+    fprintf (fp, "\t\t},\n"); 
+    fprintf (fp, "\t]\n"); 
+    fprintf (fp, "}\n"); 
+
+    // Output the background stuff 
+    fprintf (fp, "Background {\n");
+    fprintf (fp, "\tskyColor [\n");
+    fprintf (fp, "\t\t.5 .5 .5\n");
+    fprintf (fp, "\t]\n");
+    fprintf (fp, "}\n");
+
+    // Output the shape stuff 
+    fprintf (fp, "Transform {\n");
+    fprintf (fp, "\tscale .3 .3 .3\n");
+    fprintf (fp, "\tchildren [\n");
+
+    // Draw the axes as a shape: 
+    fprintf (fp, "\t\tShape {\n");
+    fprintf (fp, "\t\t\tappearance Appearance {\n");
+    fprintf (fp, "\t\t\t\tmaterial Material {\n");
+    fprintf (fp, "\t\t\t\t\tdiffuseColor 0 0.8 0\n");
+    fprintf (fp, "\t\t\t\t\temissiveColor 1.0 1.0 1.0\n");
+    fprintf (fp, "\t\t\t\t\tshininess 0.8\n");
+    fprintf (fp, "\t\t\t\t}\n");
+    fprintf (fp, "\t\t\t}\n");
+    fprintf (fp, "\t\t\tgeometry IndexedLineSet {\n");
+    fprintf (fp, "\t\t\t\tcoord Coordinate {\n");
+    fprintf (fp, "\t\t\t\t\tpoint [\n");
+    fprintf (fp, "\t\t\t\t\t0.0 0.0 0.0,\n");
+    fprintf (fp, "\t\t\t\t\t%f 0.0 0.0,\n",  255.0);
+    fprintf (fp, "\t\t\t\t\t0.0 %f 0.0,\n",  255.0);
+    fprintf (fp, "\t\t\t\t\t0.0 0.0 %f]\n",  255.0);
+    fprintf (fp, "\t\t\t\t}\n");
+    fprintf (fp, "\t\t\t\tcoordIndex [\n");
+    fprintf (fp, "\t\t\t\t\t0, 1, -1\n");
+    fprintf (fp, "\t\t\t\t\t0, 2, -1\n");
+    fprintf (fp, "\t\t\t\t\t0, 3, -1]\n");
+    fprintf (fp, "\t\t\t}\n");
+    fprintf (fp, "\t\t}\n");
+
+    
+    fprintf (fp, "\t\tShape {\n");
+    fprintf (fp, "\t\t\tappearance Appearance {\n");
+    fprintf (fp, "\t\t\t\tmaterial Material {\n");
+    fprintf (fp, "\t\t\t\t\tdiffuseColor 0 0.8 0\n");
+    fprintf (fp, "\t\t\t\t\temissiveColor 1 1 1\n");
+    fprintf (fp, "\t\t\t\t\tshininess 0.8\n");
+    fprintf (fp, "\t\t\t\t}\n");
+    fprintf (fp, "\t\t\t}\n");
+    fprintf (fp, "\t\t\tgeometry PointSet {\n");
+    
+    // fill in the points here 
+    fprintf (fp, "\t\t\t\tcoord Coordinate {\n");
+    fprintf (fp, "\t\t\t\t\tpoint [\n");
+
+    // We need to transverse all gamut hull.
+    for (i=0; i < SECTORS; i++)
+        for (j=0; j < SECTORS; j++) {
+
+            cmsVEC3 v;
+
+            pt = &gbd ->Gamut[i][j];
+            ToCartesian(&v, &pt ->p);
+
+            fprintf (fp, "\t\t\t\t\t%g %g %g", v.n[0]+50, v.n[1], v.n[2]);
+
+            if ((j == SECTORS - 1) && (i == SECTORS - 1)) 
+                fprintf (fp, "]\n");
+            else
+                fprintf (fp, ",\n");
+
+        }
+
+        fprintf (fp, "\t\t\t\t}\n");
+
+
+
+    // fill in the face colors 
+    fprintf (fp, "\t\t\t\tcolor Color {\n");
+    fprintf (fp, "\t\t\t\t\tcolor [\n");
+
+    for (i=0; i < SECTORS; i++)
+        for (j=0; j < SECTORS; j++) {
+
+           cmsVEC3 v;
+   
+            pt = &gbd ->Gamut[i][j];
+
+		        
+            ToCartesian(&v, &pt ->p);
+
+
+        if (pt ->Type == GP_EMPTY) 
+            fprintf (fp, "\t\t\t\t\t%g %g %g", 0.0, 0.0, 0.0);
+        else
+            if (pt ->Type == GP_MODELED)
+                fprintf (fp, "\t\t\t\t\t%g %g %g", 1.0, .5, .5);
+			else {
+                fprintf (fp, "\t\t\t\t\t%g %g %g", 1.0, 1.0, 1.0);
+
+			}
+
+        if ((j == SECTORS - 1) && (i == SECTORS - 1)) 
+                fprintf (fp, "]\n");
+            else
+                fprintf (fp, ",\n");
+    }
+    fprintf (fp, "\t\t\t}\n");
+    
+
+    fprintf (fp, "\t\t\t}\n");
+    fprintf (fp, "\t\t}\n");
+    fprintf (fp, "\t]\n");
+    fprintf (fp, "}\n");
+
+    fclose (fp);
+
+    return TRUE;
+}
+#endif
+
diff --git a/thirdparty/liblcms2/src/cmstypes.c b/thirdparty/liblcms2/src/cmstypes.c
new file mode 100644
index 00000000..28f06d5d
--- /dev/null
+++ b/thirdparty/liblcms2/src/cmstypes.c
@@ -0,0 +1,4955 @@
+//---------------------------------------------------------------------------------
+//
+//  Little Color Management System
+//  Copyright (c) 1998-2010 Marti Maria Saguer
+//
+// Permission is hereby granted, free of charge, to any person obtaining 
+// a copy of this software and associated documentation files (the "Software"), 
+// to deal in the Software without restriction, including without limitation 
+// the rights to use, copy, modify, merge, publish, distribute, sublicense, 
+// and/or sell copies of the Software, and to permit persons to whom the Software 
+// is furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in 
+// all copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 
+// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO 
+// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE 
+// LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 
+// OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 
+// WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+//
+//---------------------------------------------------------------------------------
+//
+
+#include "lcms2_internal.h"
+
+// Tag Serialization  ----------------------------------------------------------------------------- 
+// This file implements every single tag and tag type as described in the ICC spec. Some types
+// have been deprecated, like ncl and Data. There is no implementation for those types as there
+// are no profiles holding them. The programmer can also extend this list by defining his own types
+// by using the appropiate plug-in. There are three types of plug ins regarding that. First type
+// allows to define new tags using any existing type. Next plug-in type allows to define new types
+// and the third one is very specific: allows to extend the number of elements in the multiprofile
+// elements special type. 
+//--------------------------------------------------------------------------------------------------
+
+// Some broken types
+#define cmsCorbisBrokenXYZtype    ((cmsTagTypeSignature) 0x17A505B8)
+#define cmsMonacoBrokenCurveType  ((cmsTagTypeSignature) 0x9478ee00)
+
+// This is the linked list that keeps track of the defined types
+typedef struct _cmsTagTypeLinkedList_st {
+
+    cmsTagTypeHandler Handler;
+    struct _cmsTagTypeLinkedList_st* Next;
+
+} _cmsTagTypeLinkedList;
+
+// Some macros to define callbacks.
+#define READ_FN(x)  Type_##x##_Read
+#define WRITE_FN(x) Type_##x##_Write
+#define FREE_FN(x)  Type_##x##_Free
+#define DUP_FN(x)   Type_##x##_Dup
+
+// Helper macro to define a handler. Callbacks do have a fixed naming convention.
+#define TYPE_HANDLER(t, x)  { (t), READ_FN(x), WRITE_FN(x), DUP_FN(x), FREE_FN(x) }
+
+// Helper macro to define a MPE handler. Callbacks do have a fixed naming convention
+#define TYPE_MPE_HANDLER(t, x)  { (t), READ_FN(x), WRITE_FN(x), GenericMPEdup, GenericMPEfree }
+
+// Register a new type handler. This routine is shared between normal types and MPE
+static
+cmsBool RegisterTypesPlugin(cmsPluginBase* Data, _cmsTagTypeLinkedList* LinkedList, cmsUInt32Number DefaultListCount)
+{
+    cmsPluginTagType* Plugin = (cmsPluginTagType*) Data;
+    _cmsTagTypeLinkedList *pt, *Anterior = NULL;
+
+    // Calling the function with NULL as plug-in would unregister the plug in.
+    if (Data == NULL) {
+
+        LinkedList[DefaultListCount-1].Next = NULL;
+        return TRUE;
+    }
+
+    pt = Anterior = LinkedList; 
+    while (pt != NULL) {
+
+        if (Plugin->Handler.Signature == pt -> Handler.Signature) {
+            pt ->Handler = Plugin ->Handler;    // Replace old behaviour. 
+            // Note that since no memory is allocated, unregister does not
+            // reset this action. 
+            return TRUE;
+        }   
+
+        Anterior = pt;          
+        pt = pt ->Next;
+    }
+
+    // Registering happens in plug-in memory pool
+    pt = (_cmsTagTypeLinkedList*) _cmsPluginMalloc(sizeof(_cmsTagTypeLinkedList));
+    if (pt == NULL) return FALSE;
+
+    pt ->Handler   = Plugin ->Handler;  
+    pt ->Next      = NULL;
+
+    if (Anterior) 
+        Anterior -> Next = pt;
+
+    return TRUE;
+}
+
+// Return handler for a given type or NULL if not found. Shared between normal types and MPE
+static
+cmsTagTypeHandler* GetHandler(cmsTagTypeSignature sig, _cmsTagTypeLinkedList* LinkedList)
+{
+    _cmsTagTypeLinkedList* pt;
+
+    for (pt = LinkedList; 
+         pt != NULL;
+         pt = pt ->Next) {
+
+            if (sig == pt -> Handler.Signature) return &pt ->Handler;
+    }
+
+    return NULL;
+}
+
+
+// Auxiliar to convert UTF-32 to UTF-16 in some cases
+static
+cmsBool _cmsWriteWCharArray(cmsIOHANDLER* io, cmsUInt32Number n, const wchar_t* Array)
+{
+    cmsUInt32Number i;
+
+    _cmsAssert(io != NULL);
+    _cmsAssert(Array != NULL);
+
+    for (i=0; i < n; i++) {
+        if (!_cmsWriteUInt16Number(io, (cmsUInt16Number) Array[i])) return FALSE;
+    }
+
+    return TRUE;
+}
+
+static
+cmsBool _cmsReadWCharArray(cmsIOHANDLER* io, cmsUInt32Number n, wchar_t* Array)
+{
+    cmsUInt32Number i;
+    cmsUInt16Number tmp;
+
+    _cmsAssert(io != NULL);
+
+    for (i=0; i < n; i++) {
+
+        if (Array != NULL) {
+
+            if (!_cmsReadUInt16Number(io, &tmp)) return FALSE;
+            Array[i] = (wchar_t) tmp;
+        }
+        else {
+            if (!_cmsReadUInt16Number(io, NULL)) return FALSE;
+        }
+
+    }
+    return TRUE;
+}
+
+// To deal with position tables
+typedef cmsBool (* PositionTableEntryFn)(struct _cms_typehandler_struct* self, 
+                                             cmsIOHANDLER* io,
+                                             void* Cargo,
+                                             cmsUInt32Number n, 
+                                             cmsUInt32Number SizeOfTag);
+
+// Helper function to deal with position tables as decribed in several addendums to ICC spec 4.2
+// A table of n elements is written, where first comes n records containing offsets and sizes and
+// then a block containing the data itself. This allows to reuse same data in more than one entry
+static
+cmsBool ReadPositionTable(struct _cms_typehandler_struct* self, 
+                              cmsIOHANDLER* io, 
+                              cmsUInt32Number Count, 
+                              cmsUInt32Number BaseOffset, 
+                              void *Cargo,
+                              PositionTableEntryFn ElementFn)
+{
+    cmsUInt32Number i;
+    cmsUInt32Number *ElementOffsets = NULL, *ElementSizes = NULL;
+
+    // Let's take the offsets to each element
+    ElementOffsets = (cmsUInt32Number *) _cmsCalloc(io ->ContextID, Count, sizeof(cmsUInt32Number *));
+    if (ElementOffsets == NULL) goto Error;
+
+    ElementSizes = (cmsUInt32Number *) _cmsCalloc(io ->ContextID, Count, sizeof(cmsUInt32Number *));
+    if (ElementSizes == NULL) goto Error;
+
+    for (i=0; i < Count; i++) {
+
+        if (!_cmsReadUInt32Number(io, &ElementOffsets[i])) goto Error;
+        if (!_cmsReadUInt32Number(io, &ElementSizes[i])) goto Error;
+
+        ElementOffsets[i] += BaseOffset;
+    }
+
+    // Seek to each element and read it
+    for (i=0; i < Count; i++) {
+
+        if (!io -> Seek(io, ElementOffsets[i])) goto Error;
+
+        // This is the reader callback
+        if (!ElementFn(self, io, Cargo, i, ElementSizes[i])) goto Error;
+    }
+
+    // Success
+    if (ElementOffsets != NULL) _cmsFree(io ->ContextID, ElementOffsets);
+    if (ElementSizes != NULL) _cmsFree(io ->ContextID, ElementSizes);    
+    return TRUE;
+
+Error:    
+    if (ElementOffsets != NULL) _cmsFree(io ->ContextID, ElementOffsets);
+    if (ElementSizes != NULL) _cmsFree(io ->ContextID, ElementSizes);
+    return FALSE;
+}
+
+// Same as anterior, but for write position tables
+static
+cmsBool WritePositionTable(struct _cms_typehandler_struct* self, 
+                               cmsIOHANDLER* io,
+                               cmsUInt32Number SizeOfTag,
+                               cmsUInt32Number Count, 
+                               cmsUInt32Number BaseOffset, 
+                               void *Cargo,
+                               PositionTableEntryFn ElementFn)
+{
+    cmsUInt32Number i;
+    cmsUInt32Number DirectoryPos, CurrentPos, Before;
+    cmsUInt32Number *ElementOffsets = NULL, *ElementSizes = NULL;
+
+     // Create table
+    ElementOffsets = (cmsUInt32Number *) _cmsCalloc(io ->ContextID, Count, sizeof(cmsUInt32Number *));
+    if (ElementOffsets == NULL) goto Error;
+
+    ElementSizes = (cmsUInt32Number *) _cmsCalloc(io ->ContextID, Count, sizeof(cmsUInt32Number *));
+    if (ElementSizes == NULL) goto Error;
+  
+    // Keep starting position of curve offsets
+    DirectoryPos = io ->Tell(io);
+
+    // Write a fake directory to be filled latter on
+    for (i=0; i < Count; i++) {
+
+        if (!_cmsWriteUInt32Number(io, 0)) goto Error;  // Offset 
+        if (!_cmsWriteUInt32Number(io, 0)) goto Error;  // size
+    }
+
+    // Write each element. Keep track of the size as well.
+    for (i=0; i < Count; i++) {
+
+        Before = io ->Tell(io);
+        ElementOffsets[i] = Before - BaseOffset;                
+
+        // Callback to write...
+        if (!ElementFn(self, io, Cargo, i, SizeOfTag)) goto Error;
+
+        // Now the size
+        ElementSizes[i] = io ->Tell(io) - Before;       
+    }
+
+    // Write the directory
+    CurrentPos = io ->Tell(io);
+    if (!io ->Seek(io, DirectoryPos)) goto Error;
+
+    for (i=0; i <  Count; i++) {
+        if (!_cmsWriteUInt32Number(io, ElementOffsets[i])) goto Error;  
+        if (!_cmsWriteUInt32Number(io, ElementSizes[i])) goto Error;  
+    }
+
+    if (!io ->Seek(io, CurrentPos)) goto Error;
+
+    if (ElementOffsets != NULL) _cmsFree(io ->ContextID, ElementOffsets);
+    if (ElementSizes != NULL) _cmsFree(io ->ContextID, ElementSizes);
+    return TRUE;
+
+Error:
+    if (ElementOffsets != NULL) _cmsFree(io ->ContextID, ElementOffsets);
+    if (ElementSizes != NULL) _cmsFree(io ->ContextID, ElementSizes);
+    return FALSE;
+}
+
+
+// ********************************************************************************
+// Type XYZ. Only one value is allowed
+// ********************************************************************************
+
+//The XYZType contains an array of three encoded values for the XYZ tristimulus
+//values. Tristimulus values must be non-negative. The signed encoding allows for 
+//implementation optimizations by minimizing the number of fixed formats.
+
+
+static
+void *Type_XYZ_Read(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsUInt32Number* nItems, cmsUInt32Number SizeOfTag)
+{
+    cmsCIEXYZ* xyz;
+
+    *nItems = 0;
+    xyz = (cmsCIEXYZ*) _cmsMallocZero(self ->ContextID, sizeof(cmsCIEXYZ));
+    if (xyz == NULL) return NULL;
+
+    if (!_cmsReadXYZNumber(io, xyz)) {
+        _cmsFree(self ->ContextID, xyz);
+        return NULL;
+    }
+
+    *nItems = 1;
+    return (void*) xyz;
+
+    cmsUNUSED_PARAMETER(SizeOfTag);
+}
+
+static
+cmsBool  Type_XYZ_Write(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, void* Ptr, cmsUInt32Number nItems)
+{
+    return _cmsWriteXYZNumber(io, (cmsCIEXYZ*) Ptr);
+
+    cmsUNUSED_PARAMETER(nItems);
+    cmsUNUSED_PARAMETER(self);
+}
+
+static
+void* Type_XYZ_Dup(struct _cms_typehandler_struct* self, const void *Ptr, cmsUInt32Number n)
+{
+    return _cmsDupMem(self ->ContextID, Ptr, sizeof(cmsCIEXYZ));
+
+    cmsUNUSED_PARAMETER(n);
+}
+
+static
+void Type_XYZ_Free(struct _cms_typehandler_struct* self, void *Ptr)
+{
+    _cmsFree(self ->ContextID, Ptr);
+}
+
+
+static
+cmsTagTypeSignature DecideXYZtype(cmsFloat64Number ICCVersion, const void *Data)
+{
+    return cmsSigXYZType;
+
+    cmsUNUSED_PARAMETER(ICCVersion);
+    cmsUNUSED_PARAMETER(Data);
+}
+
+
+// ********************************************************************************
+// Type chromaticity. Only one value is allowed
+// ********************************************************************************
+// The chromaticity tag type provides basic chromaticity data and type of 
+// phosphors or colorants of a monitor to applications and utilities.
+
+static
+void *Type_Chromaticity_Read(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsUInt32Number* nItems, cmsUInt32Number SizeOfTag)
+{
+    cmsCIExyYTRIPLE* chrm;
+    cmsUInt16Number nChans, Table;
+
+    *nItems = 0;
+    chrm =  (cmsCIExyYTRIPLE*) _cmsMallocZero(self ->ContextID, sizeof(cmsCIExyYTRIPLE));
+    if (chrm == NULL) return NULL;
+
+    if (!_cmsReadUInt16Number(io, &nChans)) goto Error;
+
+    // Let's recover from a bug introduced in early versions of lcms1
+    if (nChans == 0 && SizeOfTag == 32) {
+
+        if (!_cmsReadUInt16Number(io, NULL)) goto Error;
+        if (!_cmsReadUInt16Number(io, &nChans)) goto Error;
+    }
+
+    if (nChans != 3) goto Error;
+
+    if (!_cmsReadUInt16Number(io, &Table)) goto Error;
+
+    if (!_cmsRead15Fixed16Number(io, &chrm ->Red.x)) goto Error;
+    if (!_cmsRead15Fixed16Number(io, &chrm ->Red.y)) goto Error;
+
+    chrm ->Red.Y = 1.0;
+
+    if (!_cmsRead15Fixed16Number(io, &chrm ->Green.x)) goto Error;
+    if (!_cmsRead15Fixed16Number(io, &chrm ->Green.y)) goto Error;
+
+    chrm ->Green.Y = 1.0;
+
+    if (!_cmsRead15Fixed16Number(io, &chrm ->Blue.x)) goto Error;
+    if (!_cmsRead15Fixed16Number(io, &chrm ->Blue.y)) goto Error;
+
+    chrm ->Blue.Y = 1.0;
+
+    *nItems = 1;
+    return (void*) chrm;
+
+Error:
+    _cmsFree(self ->ContextID, (void*) chrm);
+    return NULL;
+
+    cmsUNUSED_PARAMETER(SizeOfTag);
+}
+
+static
+cmsBool  SaveOneChromaticity(cmsFloat64Number x, cmsFloat64Number y, cmsIOHANDLER* io)
+{
+    if (!_cmsWriteUInt32Number(io, _cmsDoubleTo15Fixed16(x))) return FALSE;     
+    if (!_cmsWriteUInt32Number(io, _cmsDoubleTo15Fixed16(y))) return FALSE;     
+
+    return TRUE;
+}
+
+static
+cmsBool  Type_Chromaticity_Write(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, void* Ptr, cmsUInt32Number nItems)
+{
+    cmsCIExyYTRIPLE* chrm = (cmsCIExyYTRIPLE*) Ptr;
+
+    if (!_cmsWriteUInt16Number(io, 3)) return FALSE;        // nChannels
+    if (!_cmsWriteUInt16Number(io, 0)) return FALSE;        // Table
+
+    if (!SaveOneChromaticity(chrm -> Red.x,   chrm -> Red.y, io)) return FALSE;
+    if (!SaveOneChromaticity(chrm -> Green.x, chrm -> Green.y, io)) return FALSE;
+    if (!SaveOneChromaticity(chrm -> Blue.x,  chrm -> Blue.y, io)) return FALSE;
+
+    return TRUE;
+
+    cmsUNUSED_PARAMETER(nItems);
+    cmsUNUSED_PARAMETER(self);  
+}
+
+static
+void* Type_Chromaticity_Dup(struct _cms_typehandler_struct* self, const void *Ptr, cmsUInt32Number n)
+{
+    return _cmsDupMem(self ->ContextID, Ptr, sizeof(cmsCIExyYTRIPLE));
+    cmsUNUSED_PARAMETER(n);
+}
+
+static
+void Type_Chromaticity_Free(struct _cms_typehandler_struct* self, void* Ptr)
+{
+    _cmsFree(self ->ContextID, Ptr);
+}
+
+
+// ********************************************************************************
+// Type cmsSigColorantOrderType
+// ********************************************************************************
+
+// This is an optional tag which specifies the laydown order in which colorants will 
+// be printed on an n-colorant device. The laydown order may be the same as the 
+// channel generation order listed in the colorantTableTag or the channel order of a 
+// colour space such as CMYK, in which case this tag is not needed. When this is not 
+// the case (for example, ink-towers sometimes use the order KCMY), this tag may be 
+// used to specify the laydown order of the colorants. 
+
+
+static
+void *Type_ColorantOrderType_Read(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsUInt32Number* nItems, cmsUInt32Number SizeOfTag)
+{
+    cmsUInt8Number* ColorantOrder;
+    cmsUInt32Number Count;
+
+    *nItems = 0;
+    if (!_cmsReadUInt32Number(io, &Count)) return NULL;
+    if (Count > cmsMAXCHANNELS) return NULL;
+
+    ColorantOrder = (cmsUInt8Number*) _cmsCalloc(self ->ContextID, cmsMAXCHANNELS, sizeof(cmsUInt8Number));
+    if (ColorantOrder == NULL) return NULL;
+
+    // We use FF as end marker
+    memset(ColorantOrder, 0xFF, cmsMAXCHANNELS * sizeof(cmsUInt8Number));
+
+    if (io ->Read(io, ColorantOrder, sizeof(cmsUInt8Number), Count) != Count) {
+
+        _cmsFree(self ->ContextID, (void*) ColorantOrder);
+        return NULL;
+    }
+
+    *nItems = 1;
+    return (void*) ColorantOrder;
+
+    cmsUNUSED_PARAMETER(SizeOfTag);
+}
+
+static
+cmsBool Type_ColorantOrderType_Write(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, void* Ptr, cmsUInt32Number nItems)
+{
+    cmsUInt8Number*  ColorantOrder = (cmsUInt8Number*) Ptr; 
+    cmsUInt32Number i, sz, Count;
+
+    // Get the length
+    for (Count=i=0; i < cmsMAXCHANNELS; i++) {
+        if (ColorantOrder[i] != 0xFF) Count++;
+    }
+
+    if (!_cmsWriteUInt32Number(io, Count)) return FALSE;    
+
+    sz = Count * sizeof(cmsUInt8Number);
+    if (!io -> Write(io, sz, ColorantOrder)) return FALSE;          
+
+    return TRUE;
+
+    cmsUNUSED_PARAMETER(nItems);
+    cmsUNUSED_PARAMETER(self);
+}
+
+static
+void* Type_ColorantOrderType_Dup(struct _cms_typehandler_struct* self, const void *Ptr, cmsUInt32Number n)
+{
+    return _cmsDupMem(self ->ContextID, Ptr, cmsMAXCHANNELS * sizeof(cmsUInt8Number));
+
+    cmsUNUSED_PARAMETER(n);
+}
+
+
+static
+void Type_ColorantOrderType_Free(struct _cms_typehandler_struct* self, void* Ptr)
+{
+    _cmsFree(self ->ContextID, Ptr);
+}
+
+// ********************************************************************************
+// Type cmsSigS15Fixed16ArrayType
+// ********************************************************************************
+// This type represents an array of generic 4-byte/32-bit fixed point quantity. 
+// The number of values is determined from the size of the tag.
+
+static
+void *Type_S15Fixed16_Read(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsUInt32Number* nItems, cmsUInt32Number SizeOfTag)
+{
+    cmsFloat64Number*  array_double;
+    cmsUInt32Number i, n;
+
+    *nItems = 0;
+    n = SizeOfTag / sizeof(cmsUInt32Number);
+    array_double = (cmsFloat64Number*) _cmsCalloc(self ->ContextID, n, sizeof(cmsFloat64Number));
+    if (array_double == NULL) return NULL;
+
+    for (i=0; i < n; i++) {
+
+        if (!_cmsRead15Fixed16Number(io, &array_double[i])) {
+
+            _cmsFree(self ->ContextID, array_double);
+            return NULL;
+        }       
+    }
+
+    *nItems = n;
+    return (void*) array_double;
+}
+
+static
+cmsBool Type_S15Fixed16_Write(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, void* Ptr, cmsUInt32Number nItems)
+{
+    cmsFloat64Number* Value = (cmsFloat64Number*) Ptr;
+    cmsUInt32Number i;
+
+    for (i=0; i < nItems; i++) {
+
+        if (!_cmsWrite15Fixed16Number(io, Value[i])) return FALSE;  
+    }
+
+    return TRUE;
+
+    cmsUNUSED_PARAMETER(self);
+}
+
+static
+void* Type_S15Fixed16_Dup(struct _cms_typehandler_struct* self, const void *Ptr, cmsUInt32Number n)
+{
+    return _cmsDupMem(self ->ContextID, Ptr, n * sizeof(cmsFloat64Number));
+}
+
+
+static
+void Type_S15Fixed16_Free(struct _cms_typehandler_struct* self, void* Ptr)
+{
+    _cmsFree(self ->ContextID, Ptr);
+}
+
+// ********************************************************************************
+// Type cmsSigU16Fixed16ArrayType
+// ********************************************************************************
+// This type represents an array of generic 4-byte/32-bit quantity. 
+// The number of values is determined from the size of the tag.
+
+
+static
+void *Type_U16Fixed16_Read(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsUInt32Number* nItems, cmsUInt32Number SizeOfTag)
+{
+    cmsFloat64Number*  array_double;
+    cmsUInt32Number v;
+    cmsUInt32Number i, n;
+
+    *nItems = 0;
+    n = SizeOfTag / sizeof(cmsUInt32Number);
+    array_double = (cmsFloat64Number*) _cmsCalloc(self ->ContextID, n, sizeof(cmsFloat64Number));
+    if (array_double == NULL) return NULL;
+
+    for (i=0; i < n; i++) {
+
+        if (!_cmsReadUInt32Number(io, &v)) {
+            _cmsFree(self ->ContextID, (void*) array_double);
+            return NULL;
+        }
+
+        // Convert to cmsFloat64Number
+        array_double[i] =  (cmsFloat64Number) (v / 65536.0);
+    }
+
+    *nItems = n;
+    return (void*) array_double;
+}
+
+static
+cmsBool Type_U16Fixed16_Write(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, void* Ptr, cmsUInt32Number nItems)
+{
+    cmsFloat64Number* Value = (cmsFloat64Number*) Ptr;
+    cmsUInt32Number i;
+
+    for (i=0; i < nItems; i++) {
+
+        cmsUInt32Number v = (cmsUInt32Number) floor(Value[i]*65536.0 + 0.5);
+
+        if (!_cmsWriteUInt32Number(io, v)) return FALSE;    
+    }
+
+    return TRUE;
+
+    cmsUNUSED_PARAMETER(self);
+}
+
+
+static
+void* Type_U16Fixed16_Dup(struct _cms_typehandler_struct* self, const void *Ptr, cmsUInt32Number n)
+{
+    return _cmsDupMem(self ->ContextID, Ptr, n * sizeof(cmsFloat64Number));
+}
+
+static
+void Type_U16Fixed16_Free(struct _cms_typehandler_struct* self, void* Ptr)
+{
+    _cmsFree(self ->ContextID, Ptr);
+}
+
+// ********************************************************************************
+// Type cmsSigSignatureType
+// ********************************************************************************
+//
+// The signatureType contains a four-byte sequence, Sequences of less than four 
+// characters are padded at the end with spaces, 20h. 
+// Typically this type is used for registered tags that can be displayed on many 
+// development systems as a sequence of four characters.
+
+static
+void *Type_Signature_Read(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsUInt32Number* nItems, cmsUInt32Number SizeOfTag)
+{
+    cmsSignature* SigPtr = (cmsSignature*) _cmsMalloc(self ->ContextID, sizeof(cmsSignature));
+    if (SigPtr == NULL) return NULL;
+
+     if (!_cmsReadUInt32Number(io, SigPtr)) return NULL;
+     *nItems = 1;
+
+     return SigPtr;
+
+     cmsUNUSED_PARAMETER(SizeOfTag);
+}
+
+static
+cmsBool  Type_Signature_Write(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, void* Ptr, cmsUInt32Number nItems)
+{
+    cmsSignature* SigPtr = (cmsSignature*) Ptr; 
+    
+    return _cmsWriteUInt32Number(io, *SigPtr);
+
+    cmsUNUSED_PARAMETER(nItems);
+    cmsUNUSED_PARAMETER(self);
+}
+
+static
+void* Type_Signature_Dup(struct _cms_typehandler_struct* self, const void *Ptr, cmsUInt32Number n)
+{
+    return _cmsDupMem(self ->ContextID, Ptr, n * sizeof(cmsSignature));
+}
+
+static
+void Type_Signature_Free(struct _cms_typehandler_struct* self, void* Ptr)
+{
+    _cmsFree(self ->ContextID, Ptr);
+}
+
+
+// ********************************************************************************
+// Type cmsSigTextType
+// ********************************************************************************
+//
+// The textType is a simple text structure that contains a 7-bit ASCII text string. 
+// The length of the string is obtained by subtracting 8 from the element size portion 
+// of the tag itself. This string must be terminated with a 00h byte.
+
+static
+void *Type_Text_Read(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsUInt32Number* nItems, cmsUInt32Number SizeOfTag)
+{
+    char* Text = NULL;
+    cmsMLU* mlu = NULL;
+
+    // Create a container
+    mlu = cmsMLUalloc(self ->ContextID, 1);
+    if (mlu == NULL) return NULL;
+
+    *nItems = 0;
+
+    // We need to store the "\0" at the end, so +1
+    if (SizeOfTag == UINT_MAX) goto Error;
+
+    Text = (char*) _cmsMalloc(self ->ContextID, SizeOfTag + 1);
+    if (Text == NULL) goto Error;
+
+    if (io -> Read(io, Text, sizeof(char), SizeOfTag) != SizeOfTag) goto Error;        
+
+    // Make sure text is properly ended
+    Text[SizeOfTag] = 0;
+    *nItems = 1;
+    
+    // Keep the result
+    if (!cmsMLUsetASCII(mlu, cmsNoLanguage, cmsNoCountry, Text)) goto Error;
+
+    _cmsFree(self ->ContextID, Text);
+    return (void*) mlu;
+
+Error:
+    if (mlu != NULL) 
+        cmsMLUfree(mlu);
+    if (Text != NULL)
+        _cmsFree(self ->ContextID, Text);
+
+    return NULL;
+}
+
+// The conversion implies to choose a language. So, we choose the actual language.
+static
+cmsBool Type_Text_Write(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, void* Ptr, cmsUInt32Number nItems)
+{
+    cmsMLU* mlu = (cmsMLU*) Ptr; 
+    cmsUInt32Number size;
+    cmsBool  rc;
+    char* Text;
+   
+    // Get the size of the string. Note there is an extra "\0" at the end
+    size = cmsMLUgetASCII(mlu, cmsNoLanguage, cmsNoCountry, NULL, 0);
+    if (size == 0) return FALSE;       // Cannot be zero!
+
+    // Create memory
+    Text = (char*) _cmsMalloc(self ->ContextID, size);
+    cmsMLUgetASCII(mlu, cmsNoLanguage, cmsNoCountry, Text, size);
+
+    // Write it, including separator
+    rc = io ->Write(io, size, Text);
+
+    _cmsFree(self ->ContextID, Text);
+    return rc;
+
+    cmsUNUSED_PARAMETER(nItems);
+}
+
+static
+void* Type_Text_Dup(struct _cms_typehandler_struct* self, const void *Ptr, cmsUInt32Number n)
+{
+    return (void*) cmsMLUdup((cmsMLU*) Ptr);
+
+    cmsUNUSED_PARAMETER(n);
+    cmsUNUSED_PARAMETER(self);
+}
+
+
+static
+void Type_Text_Free(struct _cms_typehandler_struct* self, void* Ptr)
+{
+    cmsMLU* mlu = (cmsMLU*) Ptr;
+    cmsMLUfree(mlu);
+    return;
+
+    cmsUNUSED_PARAMETER(self);
+}
+
+static
+cmsTagTypeSignature DecideTextType(cmsFloat64Number ICCVersion, const void *Data)
+{
+    if (ICCVersion >= 4.0) 
+        return cmsSigMultiLocalizedUnicodeType;
+        
+    return cmsSigTextType;
+    
+    cmsUNUSED_PARAMETER(Data);
+}
+
+
+// ********************************************************************************
+// Type cmsSigDataType
+// ********************************************************************************
+
+// General purpose data type
+static
+void *Type_Data_Read(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsUInt32Number* nItems, cmsUInt32Number SizeOfTag)
+{
+    cmsICCData* BinData;
+    cmsUInt32Number LenOfData;
+
+    *nItems = 0;
+    
+    if (SizeOfTag < sizeof(cmsUInt32Number)) return NULL;
+
+    LenOfData = SizeOfTag - sizeof(cmsUInt32Number);
+    if (LenOfData > INT_MAX) return NULL;
+
+    BinData = (cmsICCData*) _cmsMalloc(self ->ContextID, sizeof(cmsICCData) + LenOfData - 1);
+    if (BinData == NULL) return NULL;
+
+    BinData ->len = LenOfData;
+    if (!_cmsReadUInt32Number(io, &BinData->flag)) {
+        _cmsFree(self ->ContextID, BinData);
+        return NULL;
+    }
+
+    if (io -> Read(io, BinData ->data, sizeof(cmsUInt8Number), LenOfData) != LenOfData) {
+       
+        _cmsFree(self ->ContextID, BinData);
+        return NULL;
+    }
+
+    *nItems = 1;
+  
+    return (void*) BinData;
+}
+
+
+static
+cmsBool Type_Data_Write(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, void* Ptr, cmsUInt32Number nItems)
+{
+   cmsICCData* BinData = (cmsICCData*) Ptr;
+    
+   if (!_cmsWriteUInt32Number(io, BinData ->flag)) return FALSE;
+
+   return io ->Write(io, BinData ->len, BinData ->data);
+
+   cmsUNUSED_PARAMETER(nItems);
+   cmsUNUSED_PARAMETER(self);
+}
+
+
+static
+void* Type_Data_Dup(struct _cms_typehandler_struct* self, const void *Ptr, cmsUInt32Number n)
+{
+    cmsICCData* BinData = (cmsICCData*) Ptr;
+
+    return _cmsDupMem(self ->ContextID, Ptr, sizeof(cmsICCData) + BinData ->len - 1);
+
+    cmsUNUSED_PARAMETER(n);
+}
+
+static
+void Type_Data_Free(struct _cms_typehandler_struct* self, void* Ptr)
+{
+    _cmsFree(self ->ContextID, Ptr);
+}
+
+// ********************************************************************************
+// Type cmsSigTextDescriptionType
+// ********************************************************************************
+
+static
+void *Type_Text_Description_Read(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsUInt32Number* nItems, cmsUInt32Number SizeOfTag)
+{
+    char* Text = NULL;
+    cmsMLU* mlu = NULL;
+    cmsUInt32Number  AsciiCount;
+    cmsUInt32Number  i, UnicodeCode, UnicodeCount;
+    cmsUInt16Number  ScriptCodeCode, Dummy;
+    cmsUInt8Number   ScriptCodeCount;
+
+    *nItems = 0;
+
+    //  One dword should be there
+    if (SizeOfTag < sizeof(cmsUInt32Number)) return NULL;
+
+    // Read len of ASCII
+    if (!_cmsReadUInt32Number(io, &AsciiCount)) return NULL;
+    SizeOfTag -= sizeof(cmsUInt32Number);
+
+    // Check for size
+    if (SizeOfTag < AsciiCount) return NULL; 
+
+    // All seems Ok, allocate the container
+    mlu = cmsMLUalloc(self ->ContextID, 1);
+    if (mlu == NULL) return NULL;
+
+    // As many memory as size of tag
+    Text = (char*) _cmsMalloc(self ->ContextID, AsciiCount + 1);
+    if (Text == NULL) goto Error;
+
+    // Read it
+    if (io ->Read(io, Text, sizeof(char), AsciiCount) != AsciiCount) goto Error;
+    SizeOfTag -= AsciiCount;
+
+    // Make sure there is a terminator
+    Text[AsciiCount] = 0;
+
+    // Set the MLU entry. From here we can be tolerant to wrong types
+    if (!cmsMLUsetASCII(mlu, cmsNoLanguage, cmsNoCountry, Text)) goto Error;
+    _cmsFree(self ->ContextID, (void*) Text);
+    Text = NULL;
+    
+    // Skip Unicode code
+    if (SizeOfTag < 2* sizeof(cmsUInt32Number)) goto Done;
+    if (!_cmsReadUInt32Number(io, &UnicodeCode)) goto Done;                      
+    if (!_cmsReadUInt32Number(io, &UnicodeCount)) goto Done;    
+    SizeOfTag -= 2* sizeof(cmsUInt32Number);
+
+    if (SizeOfTag < UnicodeCount*sizeof(cmsUInt16Number)) goto Done;
+
+    for (i=0; i < UnicodeCount; i++) {
+        if (!io ->Read(io, &Dummy, sizeof(cmsUInt16Number), 1)) goto Done;
+    }
+    SizeOfTag -= UnicodeCount*sizeof(cmsUInt16Number);
+
+    // Skip ScriptCode code if present. Some buggy profiles does have less
+    // data that stricttly required. We need to skip it as this type may come 
+    // embedded in other types.
+    
+    if (SizeOfTag >= sizeof(cmsUInt16Number) + sizeof(cmsUInt8Number) + 67) {
+
+        if (!_cmsReadUInt16Number(io, &ScriptCodeCode)) goto Done;
+        if (!_cmsReadUInt8Number(io,  &ScriptCodeCount)) goto Done;                      
+
+        // Skip rest of tag
+        for (i=0; i < 67; i++) {
+            if (!io ->Read(io, &Dummy, sizeof(cmsUInt8Number), 1)) goto Error;
+        }
+    }
+
+Done:
+
+    *nItems = 1;
+    return mlu;
+
+Error:
+    if (Text) _cmsFree(self ->ContextID, (void*) Text);
+    if (mlu) cmsMLUfree(mlu);
+    return NULL;
+}
+
+
+// This tag can come IN UNALIGNED SIZE. In order to prevent issues, we force zeros on description to align it
+static
+cmsBool  Type_Text_Description_Write(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, void* Ptr, cmsUInt32Number nItems)
+{
+    cmsMLU* mlu = (cmsMLU*) Ptr;
+    char *Text = NULL;
+    wchar_t *Wide = NULL;
+    cmsUInt32Number len, len_aligned, len_filler_alignment;
+    cmsBool  rc = FALSE;
+    char Filler[68];
+
+    // Used below for writting zeroes
+    memset(Filler, 0, sizeof(Filler));
+    
+    // Get the len of string
+    len = cmsMLUgetASCII(mlu, cmsNoLanguage, cmsNoCountry, NULL, 0);
+    
+    // From ICC3.4: It has been found that textDescriptionType can contain misaligned data
+    //(see clause 4.1 for the definition of “aligned”). Because the Unicode language
+    // code and Unicode count immediately follow the ASCII description, their
+    // alignment is not correct if the ASCII count is not a multiple of four. The
+    // ScriptCode code is misaligned when the ASCII count is odd. Profile reading and
+    // writing software must be written carefully in order to handle these alignment
+    // problems. 
+
+    // Compute an aligned size
+    len_aligned = _cmsALIGNLONG(len);
+    len_filler_alignment = len_aligned - len;
+
+    // Null strings
+    if (len <= 0) {
+
+        Text = (char*)    _cmsDupMem(self ->ContextID, "", sizeof(char));
+        Wide = (wchar_t*) _cmsDupMem(self ->ContextID, L"", sizeof(wchar_t));
+    }
+    else {
+        // Create independent buffers
+        Text = (char*) _cmsCalloc(self ->ContextID, len, sizeof(char));
+        if (Text == NULL) goto Error;
+
+        Wide = (wchar_t*) _cmsCalloc(self ->ContextID, len, sizeof(wchar_t));
+        if (Wide == NULL) goto Error;
+
+        // Get both representations. 
+        cmsMLUgetASCII(mlu, cmsNoLanguage, cmsNoCountry,  Text, len * sizeof(char));
+        cmsMLUgetWide(mlu,  cmsNoLanguage, cmsNoCountry,  Wide, len * sizeof(wchar_t));
+    }
+          
+  // * cmsUInt32Number       count;          * Description length
+  // * cmsInt8Number         desc[count]     * NULL terminated ascii string
+  // * cmsUInt32Number       ucLangCode;     * UniCode language code
+  // * cmsUInt32Number       ucCount;        * UniCode description length
+  // * cmsInt16Number        ucDesc[ucCount];* The UniCode description
+  // * cmsUInt16Number       scCode;         * ScriptCode code
+  // * cmsUInt8Number        scCount;        * ScriptCode count
+  // * cmsInt8Number         scDesc[67];     * ScriptCode Description
+    
+    if (!_cmsWriteUInt32Number(io, len_aligned)) goto Error;        
+    if (!io ->Write(io, len, Text)) goto Error;    
+    if (!io ->Write(io, len_filler_alignment, Filler)) goto Error;    
+
+    if (!_cmsWriteUInt32Number(io, 0)) goto Error;  // ucLanguageCode
+    
+    // This part is tricky: we need an aligned tag size, and the ScriptCode part
+    // takes 70 bytes, so we need 2 extra bytes to do the alignment
+
+    if (!_cmsWriteUInt32Number(io, len_aligned+1)) goto Error;  
+
+    // Note that in some compilers sizeof(cmsUInt16Number) != sizeof(wchar_t)
+    if (!_cmsWriteWCharArray(io, len, Wide)) goto Error;            
+    if (!_cmsWriteUInt16Array(io, len_filler_alignment+1, (cmsUInt16Number*) Filler)) goto Error;   
+
+    // ScriptCode Code & count (unused)
+    if (!_cmsWriteUInt16Number(io, 0)) goto Error;  
+    if (!_cmsWriteUInt8Number(io, 0)) goto Error;   
+        
+    if (!io ->Write(io, 67, Filler)) goto Error;
+
+    rc = TRUE;
+
+Error:
+    if (Text) _cmsFree(self ->ContextID, Text);
+    if (Wide) _cmsFree(self ->ContextID, Wide);
+
+    return rc;
+
+    cmsUNUSED_PARAMETER(nItems);
+}
+
+
+static
+void* Type_Text_Description_Dup(struct _cms_typehandler_struct* self, const void *Ptr, cmsUInt32Number n)
+{
+    return (void*) cmsMLUdup((cmsMLU*) Ptr);
+
+    cmsUNUSED_PARAMETER(n);
+    cmsUNUSED_PARAMETER(self);
+}
+
+static
+void Type_Text_Description_Free(struct _cms_typehandler_struct* self, void* Ptr)
+{
+    cmsMLU* mlu = (cmsMLU*) Ptr;
+
+    cmsMLUfree(mlu);
+    return;
+
+    cmsUNUSED_PARAMETER(self);
+}
+
+
+static
+cmsTagTypeSignature DecideTextDescType(cmsFloat64Number ICCVersion, const void *Data)
+{
+    if (ICCVersion >= 4.0) 
+        return cmsSigMultiLocalizedUnicodeType;
+        
+    return cmsSigTextDescriptionType;
+    
+    cmsUNUSED_PARAMETER(Data);
+}
+
+
+// ********************************************************************************
+// Type cmsSigCurveType
+// ********************************************************************************
+
+static
+void *Type_Curve_Read(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsUInt32Number* nItems, cmsUInt32Number SizeOfTag)
+{
+    cmsUInt32Number Count;
+    cmsToneCurve* NewGamma;
+    cmsUInt16Number Linear[2] = { 0, 0xffff };
+
+
+    *nItems = 0;
+    if (!_cmsReadUInt32Number(io, &Count)) return NULL;
+
+    switch (Count) {
+
+           case 0:   // Linear.
+
+               NewGamma = cmsBuildTabulatedToneCurve16(self ->ContextID, 2, Linear);               
+               if (!NewGamma) return NULL;
+               *nItems = 1;
+               return NewGamma;
+
+           case 1:  // Specified as the exponent of gamma function
+               {
+                   cmsUInt16Number SingleGammaFixed;
+                   cmsFloat64Number SingleGamma;
+
+                   if (!_cmsReadUInt16Number(io, &SingleGammaFixed)) return NULL;
+                   SingleGamma = _cms8Fixed8toDouble(SingleGammaFixed);
+
+                   *nItems = 1;
+                   return cmsBuildParametricToneCurve(self ->ContextID, 1, &SingleGamma);
+               }
+
+           default:  // Curve
+
+               NewGamma = cmsBuildTabulatedToneCurve16(self ->ContextID, Count, NULL);
+               if (!NewGamma) return NULL;
+
+               if (!_cmsReadUInt16Array(io, Count, NewGamma -> Table16)) return NULL;         
+
+               *nItems = 1;
+               return NewGamma;
+    }
+
+    cmsUNUSED_PARAMETER(SizeOfTag);
+}
+
+
+static
+cmsBool  Type_Curve_Write(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, void* Ptr, cmsUInt32Number nItems)
+{
+    cmsToneCurve* Curve = (cmsToneCurve*) Ptr;
+
+    if (Curve ->nSegments == 1 && Curve ->Segments[0].Type == 1) {
+
+            // Single gamma, preserve number
+            cmsUInt16Number SingleGammaFixed = _cmsDoubleTo8Fixed8(Curve ->Segments[0].Params[0]);
+
+            if (!_cmsWriteUInt32Number(io, 1)) return FALSE;    
+            if (!_cmsWriteUInt16Number(io, SingleGammaFixed)) return FALSE;
+            return TRUE;
+        
+    }
+
+    if (!_cmsWriteUInt32Number(io, Curve ->nEntries)) return FALSE; 
+    return _cmsWriteUInt16Array(io, Curve ->nEntries, Curve ->Table16);
+
+    cmsUNUSED_PARAMETER(nItems);
+    cmsUNUSED_PARAMETER(self);
+}
+
+
+static
+void* Type_Curve_Dup(struct _cms_typehandler_struct* self, const void *Ptr, cmsUInt32Number n)
+{
+    return (void*) cmsDupToneCurve((cmsToneCurve*) Ptr);
+
+    cmsUNUSED_PARAMETER(n);
+    cmsUNUSED_PARAMETER(self);
+}
+
+static
+void Type_Curve_Free(struct _cms_typehandler_struct* self, void* Ptr)
+{
+    cmsToneCurve* gamma = (cmsToneCurve*) Ptr;
+
+    cmsFreeToneCurve(gamma);
+    return;
+
+    cmsUNUSED_PARAMETER(self);
+}
+
+
+// ********************************************************************************
+// Type cmsSigParametricCurveType
+// ********************************************************************************
+
+
+// Decide which curve type to use on writting
+static
+cmsTagTypeSignature DecideCurveType(cmsFloat64Number ICCVersion, const void *Data)
+{
+    cmsToneCurve* Curve = (cmsToneCurve*) Data;
+
+    if (ICCVersion < 4.0) return cmsSigCurveType;
+    if (Curve ->nSegments != 1) return cmsSigCurveType;          // Only 1-segment curves can be saved as parametric
+    if (Curve ->Segments[0].Type < 0) return cmsSigCurveType;    // Only non-inverted curves
+
+    return cmsSigParametricCurveType;
+}
+
+static
+void *Type_ParametricCurve_Read(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsUInt32Number* nItems, cmsUInt32Number SizeOfTag)
+{
+    static const int ParamsByType[] = { 1, 3, 4, 5, 7 };
+    cmsFloat64Number Params[10];
+    cmsUInt16Number Type;
+    int i, n;
+    cmsToneCurve* NewGamma;
+
+    if (!_cmsReadUInt16Number(io, &Type)) return NULL;
+    if (!_cmsReadUInt16Number(io, NULL)) return NULL;   // Reserved
+
+    if (Type > 4) {
+
+        cmsSignalError(self->ContextID, cmsERROR_UNKNOWN_EXTENSION, "Unknown parametric curve type '%d'", Type);
+        return NULL;
+    }
+
+    memset(Params, 0, sizeof(Params));
+    n = ParamsByType[Type];
+
+    for (i=0; i < n; i++) {
+
+        if (!_cmsRead15Fixed16Number(io, &Params[i])) return NULL;      
+    }
+
+    NewGamma = cmsBuildParametricToneCurve(self ->ContextID, Type+1, Params);
+
+    *nItems = 1;
+    return NewGamma;
+
+    cmsUNUSED_PARAMETER(SizeOfTag);
+}
+    
+
+static
+cmsBool  Type_ParametricCurve_Write(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, void* Ptr, cmsUInt32Number nItems)
+{
+    cmsToneCurve* Curve = (cmsToneCurve*) Ptr;
+    int i, nParams;
+    static const int ParamsByType[] = { 0, 1, 3, 4, 5, 7 };
+      
+
+    if (Curve ->nSegments > 1 || Curve -> Segments[0].Type < 1) {
+
+        cmsSignalError(self->ContextID, 0, "Multisegment or Inverted parametric curves cannot be written");          
+        return FALSE;
+    }
+
+    nParams = ParamsByType[Curve ->Segments[0].Type];
+    
+    if (!_cmsWriteUInt16Number(io, (cmsUInt16Number) (Curve ->Segments[0].Type - 1))) return FALSE;
+    if (!_cmsWriteUInt16Number(io, 0)) return FALSE;        // Reserved
+
+    for (i=0; i < nParams; i++) {
+
+        if (!_cmsWrite15Fixed16Number(io, Curve -> Segments[0].Params[i])) return FALSE;        
+    }
+    
+    return TRUE;
+
+    cmsUNUSED_PARAMETER(nItems);
+}
+
+static
+void* Type_ParametricCurve_Dup(struct _cms_typehandler_struct* self, const void *Ptr, cmsUInt32Number n)
+{
+    return (void*) cmsDupToneCurve((cmsToneCurve*) Ptr);
+
+    cmsUNUSED_PARAMETER(n);
+    cmsUNUSED_PARAMETER(self);
+}
+
+static
+void Type_ParametricCurve_Free(struct _cms_typehandler_struct* self, void* Ptr)
+{
+    cmsToneCurve* gamma = (cmsToneCurve*) Ptr;
+
+    cmsFreeToneCurve(gamma);
+    return;
+
+    cmsUNUSED_PARAMETER(self);
+}
+
+
+// ********************************************************************************
+// Type cmsSigDateTimeType
+// ********************************************************************************
+
+// A 12-byte value representation of the time and date, where the byte usage is assigned 
+// as specified in table 1. The actual values are encoded as 16-bit unsigned integers 
+// (uInt16Number - see 5.1.6).
+//
+// All the dateTimeNumber values in a profile shall be in Coordinated Universal Time 
+// (UTC, also known as GMT or ZULU Time). Profile writers are required to convert local
+// time to UTC when setting these values. Programmes that display these values may show 
+// the dateTimeNumber as UTC, show the equivalent local time (at current locale), or 
+// display both UTC and local versions of the dateTimeNumber.
+
+static
+void *Type_DateTime_Read(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsUInt32Number* nItems, cmsUInt32Number SizeOfTag)
+{
+    cmsDateTimeNumber timestamp;
+    struct tm * NewDateTime;
+
+    *nItems = 0;
+    NewDateTime = (struct tm*) _cmsMalloc(self ->ContextID, sizeof(struct tm));
+    if (NewDateTime == NULL) return NULL;
+
+    if (io->Read(io, &timestamp, sizeof(cmsDateTimeNumber), 1) != 1) return NULL;
+            
+     _cmsDecodeDateTimeNumber(&timestamp, NewDateTime);
+
+     *nItems = 1;
+     return NewDateTime;
+
+     cmsUNUSED_PARAMETER(SizeOfTag);
+}
+
+
+static
+cmsBool  Type_DateTime_Write(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, void* Ptr, cmsUInt32Number nItems)
+{
+    struct tm * DateTime = (struct tm*) Ptr;
+    cmsDateTimeNumber timestamp;
+
+    _cmsEncodeDateTimeNumber(&timestamp, DateTime);
+    if (!io ->Write(io, sizeof(cmsDateTimeNumber), &timestamp)) return FALSE;
+
+    return TRUE;
+
+    cmsUNUSED_PARAMETER(nItems);
+    cmsUNUSED_PARAMETER(self);
+}
+
+static
+void* Type_DateTime_Dup(struct _cms_typehandler_struct* self, const void *Ptr, cmsUInt32Number n)
+{
+    return _cmsDupMem(self ->ContextID, Ptr, sizeof(struct tm));
+
+    cmsUNUSED_PARAMETER(n);
+}
+
+static
+void Type_DateTime_Free(struct _cms_typehandler_struct* self, void* Ptr)
+{
+    _cmsFree(self ->ContextID, Ptr);
+}
+
+
+
+// ********************************************************************************
+// Type icMeasurementType
+// ********************************************************************************
+
+/*
+The measurementType information refers only to the internal profile data and is
+meant to provide profile makers an alternative to the default measurement 
+specifications.
+*/
+
+static
+void *Type_Measurement_Read(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsUInt32Number* nItems, cmsUInt32Number SizeOfTag)
+{
+    cmsICCMeasurementConditions mc;
+
+    if (!_cmsReadUInt32Number(io, &mc.Observer)) return NULL;
+    if (!_cmsReadXYZNumber(io,    &mc.Backing)) return NULL;
+    if (!_cmsReadUInt32Number(io, &mc.Geometry)) return NULL;
+    if (!_cmsRead15Fixed16Number(io, &mc.Flare)) return NULL;
+    if (!_cmsReadUInt32Number(io, &mc.IlluminantType)) return NULL;
+
+    *nItems = 1;
+    return _cmsDupMem(self ->ContextID, &mc, sizeof(cmsICCMeasurementConditions));
+
+    cmsUNUSED_PARAMETER(SizeOfTag);
+}
+
+
+static
+cmsBool  Type_Measurement_Write(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, void* Ptr, cmsUInt32Number nItems)
+{
+    cmsICCMeasurementConditions* mc =(cmsICCMeasurementConditions*) Ptr;
+    
+    if (!_cmsWriteUInt32Number(io, mc->Observer)) return FALSE;
+    if (!_cmsWriteXYZNumber(io,    &mc->Backing)) return FALSE;
+    if (!_cmsWriteUInt32Number(io, mc->Geometry)) return FALSE;
+    if (!_cmsWrite15Fixed16Number(io, mc->Flare)) return FALSE;
+    if (!_cmsWriteUInt32Number(io, mc->IlluminantType)) return FALSE;
+
+    return TRUE;
+
+    cmsUNUSED_PARAMETER(nItems);
+    cmsUNUSED_PARAMETER(self);
+}
+
+static
+void* Type_Measurement_Dup(struct _cms_typehandler_struct* self, const void *Ptr, cmsUInt32Number n)
+{
+     return _cmsDupMem(self ->ContextID, Ptr, sizeof(cmsICCMeasurementConditions));
+
+     cmsUNUSED_PARAMETER(n);
+}
+
+static
+void Type_Measurement_Free(struct _cms_typehandler_struct* self, void* Ptr)
+{
+   _cmsFree(self ->ContextID, Ptr);
+}
+
+
+// ********************************************************************************
+// Type cmsSigMultiLocalizedUnicodeType
+// ********************************************************************************
+//
+//   Do NOT trust SizeOfTag as there is an issue on the definition of profileSequenceDescTag. See the TechNote from 
+//   Max Derhak and Rohit Patil about this: basically the size of the string table should be guessed and cannot be
+//   taken from the size of tag if this tag is embedded as part of bigger structures (profileSequenceDescTag, for instance)
+//
+
+static
+void *Type_MLU_Read(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsUInt32Number* nItems, cmsUInt32Number SizeOfTag)
+{
+    cmsMLU* mlu;
+    cmsUInt32Number Count, RecLen, NumOfWchar;       
+    cmsUInt32Number SizeOfHeader;
+    cmsUInt32Number  Len, Offset;
+    cmsUInt32Number  i;
+    wchar_t*         Block;
+    cmsUInt32Number  BeginOfThisString, EndOfThisString, LargestPosition;
+
+    *nItems = 0;
+    if (!_cmsReadUInt32Number(io, &Count)) return NULL;           
+    if (!_cmsReadUInt32Number(io, &RecLen)) return NULL;           
+
+    if (RecLen != 12) {
+
+        cmsSignalError(self->ContextID, cmsERROR_UNKNOWN_EXTENSION, "multiLocalizedUnicodeType of len != 12 is not supported.");
+        return NULL;
+    }
+
+    mlu = cmsMLUalloc(self ->ContextID, Count);
+    if (mlu == NULL) return NULL;
+
+    mlu ->UsedEntries = Count;
+
+    SizeOfHeader = 12 * Count + sizeof(_cmsTagBase);
+    LargestPosition = 0;
+
+    for (i=0; i < Count; i++) {
+
+        if (!_cmsReadUInt16Number(io, &mlu ->Entries[i].Language)) goto Error;           
+        if (!_cmsReadUInt16Number(io, &mlu ->Entries[i].Country))  goto Error;           
+
+        // Now deal with Len and offset.
+        if (!_cmsReadUInt32Number(io, &Len)) goto Error;      
+        if (!_cmsReadUInt32Number(io, &Offset)) goto Error;
+
+        // Check for overflow
+        if (Offset < (SizeOfHeader + 8)) goto Error;
+
+        // True begin of the string
+        BeginOfThisString = Offset - SizeOfHeader - 8;
+        
+        // Ajust to wchar_t elements
+        mlu ->Entries[i].Len = (Len * sizeof(wchar_t)) / sizeof(cmsUInt16Number);        
+        mlu ->Entries[i].StrW = (BeginOfThisString * sizeof(wchar_t)) / sizeof(cmsUInt16Number);
+
+        // To guess maximum size, add offset + len
+        EndOfThisString = BeginOfThisString + Len;
+        if (EndOfThisString > LargestPosition)
+            LargestPosition = EndOfThisString;
+    }
+
+    // Now read the remaining of tag and fill all strings. Substract the directory
+    SizeOfTag   = (LargestPosition * sizeof(wchar_t)) / sizeof(cmsUInt16Number);
+
+    Block = (wchar_t*) _cmsMalloc(self ->ContextID, SizeOfTag);
+    if (Block == NULL) goto Error;
+
+    NumOfWchar = SizeOfTag / sizeof(wchar_t);
+
+    if (!_cmsReadWCharArray(io, NumOfWchar, Block)) goto Error;
+
+    mlu ->MemPool  = Block;
+    mlu ->PoolSize = SizeOfTag;
+    mlu ->PoolUsed = SizeOfTag;
+
+    *nItems = 1;
+    return (void*) mlu;
+
+Error:          
+    if (mlu) cmsMLUfree(mlu);
+    return NULL;                                                
+}
+
+static
+cmsBool  Type_MLU_Write(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, void* Ptr, cmsUInt32Number nItems)
+{
+    cmsMLU* mlu =(cmsMLU*) Ptr;
+    cmsUInt32Number HeaderSize;
+    cmsUInt32Number  Len, Offset;
+    int i;
+
+    if (Ptr == NULL) {
+
+          // Empty placeholder
+          if (!_cmsWriteUInt32Number(io, 0)) return FALSE;           
+          if (!_cmsWriteUInt32Number(io, 12)) return FALSE;           
+          return TRUE;
+    }
+    
+    if (!_cmsWriteUInt32Number(io, mlu ->UsedEntries)) return FALSE;           
+    if (!_cmsWriteUInt32Number(io, 12)) return FALSE;           
+          
+    HeaderSize = 12 * mlu ->UsedEntries + sizeof(_cmsTagBase);
+
+    for (i=0; i < mlu ->UsedEntries; i++) {
+
+        Len    =  mlu ->Entries[i].Len;
+        Offset =  mlu ->Entries[i].StrW;
+        
+        Len    = (Len * sizeof(cmsUInt16Number)) / sizeof(wchar_t);
+        Offset = (Offset * sizeof(cmsUInt16Number)) / sizeof(wchar_t) + HeaderSize + 8;
+
+        if (!_cmsWriteUInt16Number(io, mlu ->Entries[i].Language)) return FALSE;           
+        if (!_cmsWriteUInt16Number(io, mlu ->Entries[i].Country))  return FALSE;  
+        if (!_cmsWriteUInt32Number(io, Len)) return FALSE;
+        if (!_cmsWriteUInt32Number(io, Offset)) return FALSE;               
+    }
+
+    if (!_cmsWriteWCharArray(io, mlu ->PoolUsed / sizeof(wchar_t), (wchar_t*)  mlu ->MemPool)) return FALSE;
+
+    return TRUE;
+
+    cmsUNUSED_PARAMETER(nItems);
+    cmsUNUSED_PARAMETER(self);
+}
+
+
+static
+void* Type_MLU_Dup(struct _cms_typehandler_struct* self, const void *Ptr, cmsUInt32Number n)
+{
+    return (void*) cmsMLUdup((cmsMLU*) Ptr);
+
+    cmsUNUSED_PARAMETER(n);
+    cmsUNUSED_PARAMETER(self);
+}
+
+static
+void Type_MLU_Free(struct _cms_typehandler_struct* self, void* Ptr)
+{    
+    cmsMLUfree((cmsMLU*) Ptr);
+    return;
+
+    cmsUNUSED_PARAMETER(self);
+}
+
+
+// ********************************************************************************
+// Type cmsSigLut8Type
+// ********************************************************************************
+
+// Decide which LUT type to use on writting
+static
+cmsTagTypeSignature DecideLUTtypeA2B(cmsFloat64Number ICCVersion, const void *Data)
+{
+    cmsPipeline* Lut = (cmsPipeline*) Data;
+
+    if (ICCVersion < 4.0) {
+        if (Lut ->SaveAs8Bits) return cmsSigLut8Type;
+        return cmsSigLut16Type;
+    }
+    else {
+         return cmsSigLutAtoBType;
+    }
+}
+
+static
+cmsTagTypeSignature DecideLUTtypeB2A(cmsFloat64Number ICCVersion, const void *Data)
+{
+    cmsPipeline* Lut = (cmsPipeline*) Data;
+
+    if (ICCVersion < 4.0) {
+        if (Lut ->SaveAs8Bits) return cmsSigLut8Type;
+        return cmsSigLut16Type;
+    }
+    else {
+         return cmsSigLutBtoAType;
+    }
+}
+
+/*
+This structure represents a colour transform using tables of 8-bit precision. 
+This type contains four processing elements: a 3 by 3 matrix (which shall be 
+the identity matrix unless the input colour space is XYZ), a set of one dimensional 
+input tables, a multidimensional lookup table, and a set of one dimensional output 
+tables. Data is processed using these elements via the following sequence:
+(matrix) -> (1d input tables)  -> (multidimensional lookup table - CLUT) -> (1d output tables)
+
+Byte Position   Field Length (bytes)  Content Encoded as...
+8                  1          Number of Input Channels (i)    uInt8Number
+9                  1          Number of Output Channels (o)   uInt8Number
+10                 1          Number of CLUT grid points (identical for each side) (g) uInt8Number
+11                 1          Reserved for padding (fill with 00h)
+
+12..15             4          Encoded e00 parameter   s15Fixed16Number
+*/
+
+
+// Read 8 bit tables as gamma functions
+static 
+cmsBool  Read8bitTables(cmsContext ContextID, cmsIOHANDLER* io, cmsPipeline* lut, int nChannels)
+{
+    cmsStage* mpe;
+    cmsUInt8Number* Temp = NULL;
+    int i, j;
+    cmsToneCurve* Tables[cmsMAXCHANNELS];
+
+    if (nChannels > cmsMAXCHANNELS) return FALSE;
+        
+    memset(Tables, 0, sizeof(Tables));
+
+    Temp = (cmsUInt8Number*) _cmsMalloc(ContextID, 256);
+    if (Temp == NULL) return FALSE;
+
+    for (i=0; i < nChannels; i++) {
+        Tables[i] = cmsBuildTabulatedToneCurve16(ContextID, 256, NULL);
+        if (Tables[i] == NULL) goto Error;
+    }
+
+    for (i=0; i < nChannels; i++) {
+
+        if (io ->Read(io, Temp, 256, 1) != 1) goto Error;
+
+        for (j=0; j < 256; j++)
+            Tables[i]->Table16[j] = (cmsUInt16Number) FROM_8_TO_16(Temp[j]);
+    }
+
+    _cmsFree(ContextID, Temp);
+
+
+    mpe = cmsStageAllocToneCurves(ContextID, nChannels, Tables);
+    if (mpe == NULL) goto Error;
+
+    cmsPipelineInsertStage(lut, cmsAT_END, mpe);
+
+    for (i=0; i < nChannels; i++) 
+        cmsFreeToneCurve(Tables[i]);
+
+    return TRUE;
+
+Error:
+    for (i=0; i < nChannels; i++) {
+        if (Tables[i]) cmsFreeToneCurve(Tables[i]);
+    }
+
+    if (Temp) _cmsFree(ContextID, Temp);
+    return FALSE;       
+}
+
+
+static
+cmsBool Write8bitTables(cmsContext ContextID, cmsIOHANDLER* io, cmsUInt32Number n, _cmsStageToneCurvesData* Tables)
+{
+    int j;
+    cmsUInt32Number i;
+    cmsUInt8Number val;
+
+    for (i=0; i < n; i++) {
+
+        if (Tables) {
+
+            if (Tables ->TheCurves[i]->nEntries != 256) {
+                cmsSignalError(ContextID, cmsERROR_RANGE, "LUT8 needs 256 entries on prelinearization");
+                return FALSE;
+            }
+
+        }
+
+        for (j=0; j < 256; j++) {
+
+            if (Tables != NULL)
+                val = (cmsUInt8Number) FROM_16_TO_8(Tables->TheCurves[i]->Table16[j]);
+            else
+                val = (cmsUInt8Number) j;
+
+            if (!_cmsWriteUInt8Number(io, val)) return FALSE;
+        }
+    }
+    return TRUE;
+}
+
+
+// Check overflow
+static
+unsigned int uipow(cmsUInt32Number n, cmsUInt32Number a, cmsUInt32Number b) 
+{
+    cmsUInt32Number rv = 1, rc;
+
+    if (a == 0) return 0;
+    if (n == 0) return 0;
+
+    for (; b > 0; b--) {
+
+        rv *= a;
+
+        // Check for overflow
+        if (rv > UINT_MAX / a) return 0;
+
+    }
+    
+    rc = rv * n;
+
+    if (rv != rc / n) return 0;
+    return rc;
+}
+
+
+// That will create a MPE LUT with Matrix, pre tables, CLUT and post tables. 
+// 8 bit lut may be scaled easely to v4 PCS, but we need also to properly adjust
+// PCS on BToAxx tags and AtoB if abstract. We need to fix input direction.
+
+static
+void *Type_LUT8_Read(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsUInt32Number* nItems, cmsUInt32Number SizeOfTag)
+{
+    cmsUInt8Number InputChannels, OutputChannels, CLUTpoints;
+    cmsUInt8Number* Temp = NULL;
+    cmsPipeline* NewLUT = NULL;
+    cmsStage *mpemat, *mpeclut;
+    cmsUInt32Number nTabSize, i;    
+    cmsFloat64Number Matrix[3*3];
+
+    *nItems = 0;
+ 
+    if (!_cmsReadUInt8Number(io, &InputChannels)) goto Error;
+    if (!_cmsReadUInt8Number(io, &OutputChannels)) goto Error;
+    if (!_cmsReadUInt8Number(io, &CLUTpoints)) goto Error;
+
+    // Padding
+    if (!_cmsReadUInt8Number(io, NULL)) goto Error;
+
+    // Do some checking
+   
+    if (InputChannels > cmsMAXCHANNELS)  goto Error;
+    if (OutputChannels > cmsMAXCHANNELS) goto Error;
+
+   // Allocates an empty Pipeline
+    NewLUT = cmsPipelineAlloc(self ->ContextID, InputChannels, OutputChannels);
+    if (NewLUT == NULL) goto Error;
+
+    // Read the Matrix
+    if (!_cmsRead15Fixed16Number(io,  &Matrix[0])) goto Error;
+    if (!_cmsRead15Fixed16Number(io,  &Matrix[1])) goto Error;
+    if (!_cmsRead15Fixed16Number(io,  &Matrix[2])) goto Error;
+    if (!_cmsRead15Fixed16Number(io,  &Matrix[3])) goto Error;
+    if (!_cmsRead15Fixed16Number(io,  &Matrix[4])) goto Error;
+    if (!_cmsRead15Fixed16Number(io,  &Matrix[5])) goto Error;
+    if (!_cmsRead15Fixed16Number(io,  &Matrix[6])) goto Error;
+    if (!_cmsRead15Fixed16Number(io,  &Matrix[7])) goto Error;
+    if (!_cmsRead15Fixed16Number(io,  &Matrix[8])) goto Error;
+
+  
+    // Only operates if not identity...
+    if ((InputChannels == 3) && !_cmsMAT3isIdentity((cmsMAT3*) Matrix)) {
+
+        mpemat = cmsStageAllocMatrix(self ->ContextID, 3, 3, Matrix, NULL);
+        if (mpemat == NULL) goto Error;
+        cmsPipelineInsertStage(NewLUT, cmsAT_BEGIN, mpemat);
+    }
+    
+    // Get input tables
+    if (!Read8bitTables(self ->ContextID, io,  NewLUT, InputChannels)) goto Error;
+    
+    // Get 3D CLUT. Check the overflow....
+    nTabSize = uipow(OutputChannels, CLUTpoints, InputChannels);
+    if (nTabSize > 0) {
+
+        cmsUInt16Number *PtrW, *T;
+        cmsUInt32Number Tsize;
+
+        Tsize = (cmsUInt32Number) nTabSize * sizeof(cmsUInt16Number);
+
+        PtrW = T  = (cmsUInt16Number*) _cmsCalloc(self ->ContextID, nTabSize, sizeof(cmsUInt16Number));
+        if (T  == NULL) goto Error;
+
+        Temp = (cmsUInt8Number*) _cmsMalloc(self ->ContextID, nTabSize);             
+        if (Temp == NULL) goto Error;
+
+        if (io ->Read(io, Temp, nTabSize, 1) != 1) goto Error;
+
+        for (i = 0; i < nTabSize; i++) {
+
+            *PtrW++ = FROM_8_TO_16(Temp[i]);
+        }
+        _cmsFree(self ->ContextID, Temp);
+        Temp = NULL;
+
+        
+        mpeclut = cmsStageAllocCLut16bit(self ->ContextID, CLUTpoints, InputChannels, OutputChannels, T);
+        if (mpeclut == NULL) goto Error;
+        cmsPipelineInsertStage(NewLUT, cmsAT_END, mpeclut);
+        _cmsFree(self ->ContextID, T);
+    }
+
+
+    // Get output tables
+    if (!Read8bitTables(self ->ContextID, io,  NewLUT, OutputChannels)) goto Error;
+    
+    *nItems = 1;
+    return NewLUT;
+
+Error:
+    if (NewLUT != NULL) cmsPipelineFree(NewLUT);
+    return NULL;
+
+    cmsUNUSED_PARAMETER(SizeOfTag);
+}
+
+// We only allow a specific MPE structure: Matrix plus prelin, plus clut, plus post-lin.
+static
+cmsBool  Type_LUT8_Write(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, void* Ptr, cmsUInt32Number nItems)
+{
+    cmsUInt32Number j, nTabSize;
+    cmsUInt8Number  val;
+    cmsPipeline* NewLUT = (cmsPipeline*) Ptr;
+    cmsStage* mpe;
+    _cmsStageToneCurvesData* PreMPE = NULL, *PostMPE = NULL;
+    _cmsStageMatrixData* MatMPE = NULL;
+    _cmsStageCLutData* clut = NULL;
+    int clutPoints;
+
+    // Disassemble the LUT into components.
+    mpe = NewLUT -> Elements;
+    if (mpe ->Type == cmsSigMatrixElemType) {
+
+        MatMPE = (_cmsStageMatrixData*) mpe ->Data;
+        mpe = mpe -> Next;
+    }
+
+    if (mpe != NULL && mpe ->Type == cmsSigCurveSetElemType) {
+        PreMPE = (_cmsStageToneCurvesData*) mpe ->Data;
+        mpe = mpe -> Next;
+    }
+
+    if (mpe != NULL && mpe ->Type == cmsSigCLutElemType) {
+        clut  = (_cmsStageCLutData*) mpe -> Data;
+        mpe = mpe ->Next;
+    }
+
+    if (mpe != NULL && mpe ->Type == cmsSigCurveSetElemType) {
+        PostMPE = (_cmsStageToneCurvesData*) mpe ->Data;
+        mpe = mpe -> Next;
+    }
+
+    // That should be all
+    if (mpe != NULL) {
+        cmsSignalError(mpe->ContextID, cmsERROR_UNKNOWN_EXTENSION, "LUT is not suitable to be saved as LUT8");
+        return FALSE;
+    }
+
+
+    if (clut == NULL)
+        clutPoints = 0;
+    else
+        clutPoints    = clut->Params->nSamples[0];
+
+    if (!_cmsWriteUInt8Number(io, (cmsUInt8Number) NewLUT ->InputChannels)) return FALSE;
+    if (!_cmsWriteUInt8Number(io, (cmsUInt8Number) NewLUT ->OutputChannels)) return FALSE;
+    if (!_cmsWriteUInt8Number(io, (cmsUInt8Number) clutPoints)) return FALSE;
+    if (!_cmsWriteUInt8Number(io, 0)) return FALSE; // Padding
+
+
+    if (MatMPE != NULL) {
+
+        if (!_cmsWrite15Fixed16Number(io, MatMPE -> Double[0])) return FALSE;
+        if (!_cmsWrite15Fixed16Number(io, MatMPE -> Double[1])) return FALSE;
+        if (!_cmsWrite15Fixed16Number(io, MatMPE -> Double[2])) return FALSE;
+        if (!_cmsWrite15Fixed16Number(io, MatMPE -> Double[3])) return FALSE;
+        if (!_cmsWrite15Fixed16Number(io, MatMPE -> Double[4])) return FALSE;
+        if (!_cmsWrite15Fixed16Number(io, MatMPE -> Double[5])) return FALSE;
+        if (!_cmsWrite15Fixed16Number(io, MatMPE -> Double[6])) return FALSE;
+        if (!_cmsWrite15Fixed16Number(io, MatMPE -> Double[7])) return FALSE;
+        if (!_cmsWrite15Fixed16Number(io, MatMPE -> Double[8])) return FALSE;
+
+    }
+    else {
+
+        if (!_cmsWrite15Fixed16Number(io, 1)) return FALSE;
+        if (!_cmsWrite15Fixed16Number(io, 0)) return FALSE;
+        if (!_cmsWrite15Fixed16Number(io, 0)) return FALSE;
+        if (!_cmsWrite15Fixed16Number(io, 0)) return FALSE;
+        if (!_cmsWrite15Fixed16Number(io, 1)) return FALSE;
+        if (!_cmsWrite15Fixed16Number(io, 0)) return FALSE;
+        if (!_cmsWrite15Fixed16Number(io, 0)) return FALSE;
+        if (!_cmsWrite15Fixed16Number(io, 0)) return FALSE;
+        if (!_cmsWrite15Fixed16Number(io, 1)) return FALSE;
+    }
+
+    // The prelinearization table
+    if (!Write8bitTables(self ->ContextID, io, NewLUT ->InputChannels, PreMPE)) return FALSE;
+
+    nTabSize = uipow(NewLUT->OutputChannels, clutPoints, NewLUT ->InputChannels);
+    if (nTabSize > 0) {
+
+    // The 3D CLUT.
+    if (clut != NULL) {
+
+        for (j=0; j < nTabSize; j++) {
+
+            val = (cmsUInt8Number) FROM_16_TO_8(clut ->Tab.T[j]);
+            if (!_cmsWriteUInt8Number(io, val)) return FALSE;
+        }
+    }
+    }
+
+    // The postlinearization table
+    if (!Write8bitTables(self ->ContextID, io, NewLUT ->OutputChannels, PostMPE)) return FALSE;
+
+    return TRUE;
+
+    cmsUNUSED_PARAMETER(nItems);
+}
+
+
+static
+void* Type_LUT8_Dup(struct _cms_typehandler_struct* self, const void *Ptr, cmsUInt32Number n)
+{
+    return (void*) cmsPipelineDup((cmsPipeline*) Ptr);
+
+    cmsUNUSED_PARAMETER(n);
+    cmsUNUSED_PARAMETER(self);
+}
+
+static
+void Type_LUT8_Free(struct _cms_typehandler_struct* self, void* Ptr)
+{    
+    cmsPipelineFree((cmsPipeline*) Ptr);
+    return;
+
+    cmsUNUSED_PARAMETER(self);
+}
+
+// ********************************************************************************
+// Type cmsSigLut16Type
+// ********************************************************************************
+
+// Read 16 bit tables as gamma functions
+static 
+cmsBool  Read16bitTables(cmsContext ContextID, cmsIOHANDLER* io, cmsPipeline* lut, int nChannels, int nEntries)
+{
+    cmsStage* mpe;
+    int i;
+    cmsToneCurve* Tables[cmsMAXCHANNELS];
+
+    // Maybe an empty table? (this is a lcms extension)
+    if (nEntries <= 0) return TRUE;
+
+    // Check for malicious profiles
+    if (nChannels > cmsMAXCHANNELS) return FALSE;
+    
+    // Init table to zero
+    memset(Tables, 0, sizeof(Tables));
+
+    for (i=0; i < nChannels; i++) {
+
+        Tables[i] = cmsBuildTabulatedToneCurve16(ContextID, nEntries, NULL);
+        if (Tables[i] == NULL) goto Error;
+
+        if (!_cmsReadUInt16Array(io, nEntries, Tables[i]->Table16)) goto Error;
+    }
+
+
+    // Add the table (which may certainly be an identity, but this is up to the optimizer, not the reading code)
+    mpe = cmsStageAllocToneCurves(ContextID, nChannels, Tables);
+    if (mpe == NULL) goto Error;
+
+    cmsPipelineInsertStage(lut, cmsAT_END, mpe);
+
+    for (i=0; i < nChannels; i++) 
+        cmsFreeToneCurve(Tables[i]);
+
+    return TRUE;
+
+Error:
+    for (i=0; i < nChannels; i++) {
+        if (Tables[i]) cmsFreeToneCurve(Tables[i]);
+    }
+
+    return FALSE;       
+}
+
+static
+cmsBool Write16bitTables(cmsContext ContextID, cmsIOHANDLER* io, _cmsStageToneCurvesData* Tables)
+{
+    int j;
+    cmsUInt32Number i;
+    cmsUInt16Number val;
+    int nEntries = 256;
+
+    nEntries = Tables->TheCurves[0]->nEntries;
+    
+    for (i=0; i < Tables ->nCurves; i++) {
+
+        for (j=0; j < nEntries; j++) {
+
+            if (Tables != NULL)
+                val = Tables->TheCurves[i]->Table16[j];
+            else
+                val = _cmsQuantizeVal(j, nEntries);
+
+            if (!_cmsWriteUInt16Number(io, val)) return FALSE;
+        }
+    }
+    return TRUE;
+
+    cmsUNUSED_PARAMETER(ContextID);
+}
+
+static
+void *Type_LUT16_Read(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsUInt32Number* nItems, cmsUInt32Number SizeOfTag)
+{
+    cmsUInt8Number InputChannels, OutputChannels, CLUTpoints;
+    cmsPipeline* NewLUT = NULL;
+    cmsStage *mpemat,  *mpeclut;
+    cmsUInt32Number nTabSize;
+    cmsFloat64Number Matrix[3*3];
+    cmsUInt16Number InputEntries, OutputEntries;
+    
+    *nItems = 0;
+
+    if (!_cmsReadUInt8Number(io, &InputChannels)) return NULL;
+    if (!_cmsReadUInt8Number(io, &OutputChannels)) return NULL;
+    if (!_cmsReadUInt8Number(io, &CLUTpoints)) return NULL;   // 255 maximum
+    
+    // Padding
+    if (!_cmsReadUInt8Number(io, NULL)) return NULL;
+
+    // Do some checking
+    if (InputChannels > cmsMAXCHANNELS)  goto Error;
+    if (OutputChannels > cmsMAXCHANNELS) goto Error;
+
+    // Allocates an empty LUT
+    NewLUT = cmsPipelineAlloc(self ->ContextID, InputChannels, OutputChannels);
+    if (NewLUT == NULL) goto Error;
+
+    // Read the Matrix
+    if (!_cmsRead15Fixed16Number(io,  &Matrix[0])) goto Error;
+    if (!_cmsRead15Fixed16Number(io,  &Matrix[1])) goto Error;
+    if (!_cmsRead15Fixed16Number(io,  &Matrix[2])) goto Error;
+    if (!_cmsRead15Fixed16Number(io,  &Matrix[3])) goto Error;
+    if (!_cmsRead15Fixed16Number(io,  &Matrix[4])) goto Error;
+    if (!_cmsRead15Fixed16Number(io,  &Matrix[5])) goto Error;
+    if (!_cmsRead15Fixed16Number(io,  &Matrix[6])) goto Error;
+    if (!_cmsRead15Fixed16Number(io,  &Matrix[7])) goto Error;
+    if (!_cmsRead15Fixed16Number(io,  &Matrix[8])) goto Error;
+
+
+    // Only operates on 3 channels
+
+    if ((InputChannels == 3) && !_cmsMAT3isIdentity((cmsMAT3*) Matrix)) {
+
+        mpemat = cmsStageAllocMatrix(self ->ContextID, 3, 3, Matrix, NULL);
+        if (mpemat == NULL) goto Error;
+        cmsPipelineInsertStage(NewLUT, cmsAT_END, mpemat);
+    }
+
+    if (!_cmsReadUInt16Number(io, &InputEntries)) return NULL;
+    if (!_cmsReadUInt16Number(io, &OutputEntries)) return NULL; 
+
+    
+    // Get input tables
+    if (!Read16bitTables(self ->ContextID, io,  NewLUT, InputChannels, InputEntries)) goto Error;
+
+    // Get 3D CLUT
+    nTabSize = uipow(OutputChannels, CLUTpoints, InputChannels);
+    if (nTabSize > 0) {
+
+        cmsUInt16Number *T;
+
+        T  = (cmsUInt16Number*) _cmsCalloc(self ->ContextID, nTabSize, sizeof(cmsUInt16Number));
+        if (T  == NULL) goto Error;
+
+        if (!_cmsReadUInt16Array(io, nTabSize, T)) {
+            _cmsFree(self ->ContextID, T);
+            goto Error;      
+        }
+        
+        mpeclut = cmsStageAllocCLut16bit(self ->ContextID, CLUTpoints, InputChannels, OutputChannels, T);
+        if (mpeclut == NULL) {
+             _cmsFree(self ->ContextID, T);
+            goto Error;
+        }
+
+        cmsPipelineInsertStage(NewLUT, cmsAT_END, mpeclut);
+        _cmsFree(self ->ContextID, T);
+    }
+
+
+    // Get output tables
+    if (!Read16bitTables(self ->ContextID, io,  NewLUT, OutputChannels, OutputEntries)) goto Error;
+
+    *nItems = 1;
+    return NewLUT;
+
+Error:
+    if (NewLUT != NULL) cmsPipelineFree(NewLUT);
+    return NULL;
+
+    cmsUNUSED_PARAMETER(SizeOfTag);
+}
+
+// We only allow some specific MPE structures: Matrix plus prelin, plus clut, plus post-lin. 
+// Some empty defaults are created for missing parts
+
+static
+cmsBool  Type_LUT16_Write(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, void* Ptr, cmsUInt32Number nItems)
+{
+    cmsUInt32Number nTabSize;
+    cmsPipeline* NewLUT = (cmsPipeline*) Ptr;
+    cmsStage* mpe;
+    _cmsStageToneCurvesData* PreMPE = NULL, *PostMPE = NULL;
+    _cmsStageMatrixData* MatMPE = NULL;
+    _cmsStageCLutData* clut = NULL;
+    int InputChannels, OutputChannels, clutPoints;
+
+    // Disassemble the LUT into components.
+    mpe = NewLUT -> Elements;
+    if (mpe != NULL && mpe ->Type == cmsSigMatrixElemType) {
+
+        MatMPE = (_cmsStageMatrixData*) mpe ->Data;
+        mpe = mpe -> Next;
+    }
+    
+
+    if (mpe != NULL && mpe ->Type == cmsSigCurveSetElemType) {
+        PreMPE = (_cmsStageToneCurvesData*) mpe ->Data;
+        mpe = mpe -> Next;
+    }
+    
+    if (mpe != NULL && mpe ->Type == cmsSigCLutElemType) {
+        clut  = (_cmsStageCLutData*) mpe -> Data;
+        mpe = mpe ->Next;
+    }
+    
+    if (mpe != NULL && mpe ->Type == cmsSigCurveSetElemType) {
+        PostMPE = (_cmsStageToneCurvesData*) mpe ->Data;
+        mpe = mpe -> Next;
+    }
+
+    // That should be all
+    if (mpe != NULL) {
+        cmsSignalError(mpe->ContextID, cmsERROR_UNKNOWN_EXTENSION, "LUT is not suitable to be saved as LUT16");
+        return FALSE;
+    }
+
+    InputChannels  = cmsPipelineInputChannels(NewLUT);
+    OutputChannels = cmsPipelineOutputChannels(NewLUT);
+    
+    if (clut == NULL)
+        clutPoints = 0;
+    else
+        clutPoints    = clut->Params->nSamples[0];
+
+    if (!_cmsWriteUInt8Number(io, (cmsUInt8Number) InputChannels)) return FALSE;
+    if (!_cmsWriteUInt8Number(io, (cmsUInt8Number) OutputChannels)) return FALSE;
+    if (!_cmsWriteUInt8Number(io, (cmsUInt8Number) clutPoints)) return FALSE;
+    if (!_cmsWriteUInt8Number(io, 0)) return FALSE; // Padding
+
+
+    if (MatMPE != NULL) {
+    
+        if (!_cmsWrite15Fixed16Number(io, MatMPE -> Double[0])) return FALSE;
+        if (!_cmsWrite15Fixed16Number(io, MatMPE -> Double[1])) return FALSE;
+        if (!_cmsWrite15Fixed16Number(io, MatMPE -> Double[2])) return FALSE;
+        if (!_cmsWrite15Fixed16Number(io, MatMPE -> Double[3])) return FALSE;
+        if (!_cmsWrite15Fixed16Number(io, MatMPE -> Double[4])) return FALSE;
+        if (!_cmsWrite15Fixed16Number(io, MatMPE -> Double[5])) return FALSE;
+        if (!_cmsWrite15Fixed16Number(io, MatMPE -> Double[6])) return FALSE;
+        if (!_cmsWrite15Fixed16Number(io, MatMPE -> Double[7])) return FALSE;
+        if (!_cmsWrite15Fixed16Number(io, MatMPE -> Double[8])) return FALSE;
+    }
+    else {
+
+        if (!_cmsWrite15Fixed16Number(io, 1)) return FALSE;
+        if (!_cmsWrite15Fixed16Number(io, 0)) return FALSE;
+        if (!_cmsWrite15Fixed16Number(io, 0)) return FALSE;
+        if (!_cmsWrite15Fixed16Number(io, 0)) return FALSE;
+        if (!_cmsWrite15Fixed16Number(io, 1)) return FALSE;
+        if (!_cmsWrite15Fixed16Number(io, 0)) return FALSE;
+        if (!_cmsWrite15Fixed16Number(io, 0)) return FALSE;
+        if (!_cmsWrite15Fixed16Number(io, 0)) return FALSE;
+        if (!_cmsWrite15Fixed16Number(io, 1)) return FALSE;
+    }
+
+
+    if (PreMPE != NULL) {
+        if (!_cmsWriteUInt16Number(io, (cmsUInt16Number) PreMPE ->TheCurves[0]->nEntries)) return FALSE;
+    } else {
+            if (!_cmsWriteUInt16Number(io, 0)) return FALSE;
+    }
+
+    if (PostMPE != NULL) {
+        if (!_cmsWriteUInt16Number(io, (cmsUInt16Number) PostMPE ->TheCurves[0]->nEntries)) return FALSE;
+    } else {
+        if (!_cmsWriteUInt16Number(io, 0)) return FALSE;
+    
+    }
+
+    // The prelinearization table
+
+    if (PreMPE != NULL) {
+        if (!Write16bitTables(self ->ContextID, io, PreMPE)) return FALSE;
+    }
+
+    nTabSize = uipow(OutputChannels, clutPoints, InputChannels);
+
+    if (nTabSize > 0) {
+    // The 3D CLUT.
+    if (clut != NULL) {
+        if (!_cmsWriteUInt16Array(io, nTabSize, clut->Tab.T)) return FALSE;
+    }
+    }
+
+    // The postlinearization table
+    if (PostMPE != NULL) {
+        if (!Write16bitTables(self ->ContextID, io, PostMPE)) return FALSE;
+    }
+
+
+    return TRUE;
+
+    cmsUNUSED_PARAMETER(nItems);
+}
+
+static
+void* Type_LUT16_Dup(struct _cms_typehandler_struct* self, const void *Ptr, cmsUInt32Number n)
+{
+    return (void*) cmsPipelineDup((cmsPipeline*) Ptr);
+
+    cmsUNUSED_PARAMETER(n);
+    cmsUNUSED_PARAMETER(self);
+}
+
+static
+void Type_LUT16_Free(struct _cms_typehandler_struct* self, void* Ptr)
+{    
+    cmsPipelineFree((cmsPipeline*) Ptr);
+    return;
+
+    cmsUNUSED_PARAMETER(self);
+}
+
+
+// ********************************************************************************
+// Type cmsSigLutAToBType
+// ********************************************************************************
+
+
+// V4 stuff. Read matrix for LutAtoB and LutBtoA
+
+static
+cmsStage* ReadMatrix(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsUInt32Number Offset)
+{
+    cmsFloat64Number dMat[3*3];
+    cmsFloat64Number dOff[3];
+    cmsStage* Mat;
+
+    // Go to address 
+    if (!io -> Seek(io, Offset)) return NULL;
+
+    // Read the Matrix
+    if (!_cmsRead15Fixed16Number(io, &dMat[0])) return NULL;
+    if (!_cmsRead15Fixed16Number(io, &dMat[1])) return NULL;
+    if (!_cmsRead15Fixed16Number(io, &dMat[2])) return NULL;
+    if (!_cmsRead15Fixed16Number(io, &dMat[3])) return NULL;
+    if (!_cmsRead15Fixed16Number(io, &dMat[4])) return NULL;
+    if (!_cmsRead15Fixed16Number(io, &dMat[5])) return NULL;
+    if (!_cmsRead15Fixed16Number(io, &dMat[6])) return NULL;
+    if (!_cmsRead15Fixed16Number(io, &dMat[7])) return NULL;
+    if (!_cmsRead15Fixed16Number(io, &dMat[8])) return NULL;
+
+    if (!_cmsRead15Fixed16Number(io, &dOff[0])) return NULL;
+    if (!_cmsRead15Fixed16Number(io, &dOff[1])) return NULL;
+    if (!_cmsRead15Fixed16Number(io, &dOff[2])) return NULL;
+    
+    Mat = cmsStageAllocMatrix(self ->ContextID, 3, 3, dMat, dOff);
+
+     return Mat;
+}
+
+
+
+
+//  V4 stuff. Read CLUT part for LutAtoB and LutBtoA
+
+static
+cmsStage* ReadCLUT(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsUInt32Number Offset, int InputChannels, int OutputChannels)
+{
+    cmsUInt8Number  gridPoints8[cmsMAXCHANNELS]; // Number of grid points in each dimension.  
+    cmsUInt32Number GridPoints[cmsMAXCHANNELS], i;
+    cmsUInt8Number  Precision;
+    cmsStage* CLUT;
+    _cmsStageCLutData* Data;
+
+    if (!io -> Seek(io, Offset)) return NULL;
+    if (io -> Read(io, gridPoints8, cmsMAXCHANNELS, 1) != 1) return NULL;
+
+    for (i=0; i < cmsMAXCHANNELS; i++)
+        GridPoints[i] = gridPoints8[i];
+
+    if (!_cmsReadUInt8Number(io, &Precision)) return NULL;
+
+    if (!_cmsReadUInt8Number(io, NULL)) return NULL;
+    if (!_cmsReadUInt8Number(io, NULL)) return NULL;
+    if (!_cmsReadUInt8Number(io, NULL)) return NULL;
+
+    CLUT = cmsStageAllocCLut16bitGranular(self ->ContextID, GridPoints, InputChannels, OutputChannels, NULL);
+    if (CLUT == NULL) return NULL;
+
+    Data = (_cmsStageCLutData*) CLUT ->Data;
+
+    // Precision can be 1 or 2 bytes
+    if (Precision == 1) {
+
+       cmsUInt8Number  v;
+  
+        for (i=0; i < Data ->nEntries; i++) {
+
+                if (io ->Read(io, &v, sizeof(cmsUInt8Number), 1) != 1) return NULL;
+                Data ->Tab.T[i] = FROM_8_TO_16(v);
+        }
+                
+    }
+    else  
+        if (Precision == 2) {
+        
+            if (!_cmsReadUInt16Array(io, Data->nEntries, Data ->Tab.T)) return NULL;
+    }
+    else {
+        cmsSignalError(self ->ContextID, cmsERROR_UNKNOWN_EXTENSION, "Unknown precision of '%d'", Precision); 
+        return NULL;
+    }
+
+    
+    return CLUT;
+}
+
+static
+cmsToneCurve* ReadEmbeddedCurve(struct _cms_typehandler_struct* self, cmsIOHANDLER* io)
+{
+    cmsTagTypeSignature  BaseType;
+    cmsUInt32Number nItems;
+
+    BaseType = _cmsReadTypeBase(io);       
+    switch (BaseType) {
+
+            case cmsSigCurveType:
+                return (cmsToneCurve*) Type_Curve_Read(self, io, &nItems, 0);
+
+            case cmsSigParametricCurveType:
+                return (cmsToneCurve*) Type_ParametricCurve_Read(self, io, &nItems, 0);
+
+            default: 
+                {
+                    char String[5];
+
+                    _cmsTagSignature2String(String, (cmsTagSignature) BaseType);
+                    cmsSignalError(self ->ContextID, cmsERROR_UNKNOWN_EXTENSION, "Unknown curve type '%s'", String);
+                }
+                return NULL;
+    }
+}
+
+
+// Read a set of curves from specific offset
+static
+cmsStage* ReadSetOfCurves(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsUInt32Number Offset, cmsUInt32Number nCurves)
+{     
+    cmsToneCurve* Curves[cmsMAXCHANNELS];
+    cmsUInt32Number i;
+    cmsStage* Lin = NULL;
+
+    if (nCurves > cmsMAXCHANNELS) return FALSE;
+
+    if (!io -> Seek(io, Offset)) return FALSE;
+    
+    for (i=0; i < nCurves; i++) 
+        Curves[i] = NULL;
+
+    for (i=0; i < nCurves; i++) {
+
+        Curves[i] = ReadEmbeddedCurve(self, io);                     
+        if (Curves[i] == NULL) goto Error;
+        if (!_cmsReadAlignment(io)) goto Error;   
+    }
+    
+    Lin = cmsStageAllocToneCurves(self ->ContextID, nCurves, Curves);
+    
+Error:
+    for (i=0; i < nCurves; i++) 
+        cmsFreeToneCurve(Curves[i]);
+
+    return Lin;
+}
+
+
+// LutAtoB type 
+
+// This structure represents a colour transform. The type contains up to five processing 
+// elements which are stored in the AtoBTag tag in the following order: a set of one 
+// dimensional curves, a 3 by 3 matrix with offset terms, a set of one dimensional curves, 
+// a multidimensional lookup table, and a set of one dimensional output curves.
+// Data are processed using these elements via the following sequence:
+//
+//("A" curves) -> (multidimensional lookup table - CLUT) -> ("M" curves) -> (matrix) -> ("B" curves).
+//
+/*
+It is possible to use any or all of these processing elements. At least one processing element 
+must be included.Only the following combinations are allowed:
+
+B
+M - Matrix - B
+A - CLUT - B
+A - CLUT - M - Matrix - B
+
+*/
+
+static
+void* Type_LUTA2B_Read(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsUInt32Number* nItems, cmsUInt32Number SizeOfTag)
+{
+    cmsUInt32Number      BaseOffset;
+    cmsUInt8Number       inputChan;      // Number of input channels     
+    cmsUInt8Number       outputChan;     // Number of output channels     
+    cmsUInt32Number      offsetB;        // Offset to first "B" curve    
+    cmsUInt32Number      offsetMat;      // Offset to matrix             
+    cmsUInt32Number      offsetM;        // Offset to first "M" curve    
+    cmsUInt32Number      offsetC;        // Offset to CLUT               
+    cmsUInt32Number      offsetA;        // Offset to first "A" curve
+    cmsStage* mpe;
+    cmsPipeline* NewLUT = NULL;
+
+
+    BaseOffset = io ->Tell(io) - sizeof(_cmsTagBase);
+
+    if (!_cmsReadUInt8Number(io, &inputChan)) return NULL;
+    if (!_cmsReadUInt8Number(io, &outputChan)) return NULL;
+    
+    if (!_cmsReadUInt16Number(io, NULL)) return NULL;
+    
+    if (!_cmsReadUInt32Number(io, &offsetB)) return NULL;
+    if (!_cmsReadUInt32Number(io, &offsetMat)) return NULL;
+    if (!_cmsReadUInt32Number(io, &offsetM)) return NULL;
+    if (!_cmsReadUInt32Number(io, &offsetC)) return NULL;
+    if (!_cmsReadUInt32Number(io, &offsetA)) return NULL;
+
+   // Allocates an empty LUT
+    NewLUT = cmsPipelineAlloc(self ->ContextID, inputChan, outputChan);
+    if (NewLUT == NULL) return NULL;
+
+    if (offsetA!= 0) {
+        mpe = ReadSetOfCurves(self, io, BaseOffset + offsetA, inputChan);
+        cmsPipelineInsertStage(NewLUT, cmsAT_END, mpe);
+    }
+
+    if (offsetC != 0) {
+        mpe = ReadCLUT(self, io, BaseOffset + offsetC, inputChan, outputChan);
+        if (mpe != NULL) cmsPipelineInsertStage(NewLUT, cmsAT_END, mpe);
+    }
+
+    if (offsetM != 0) {
+        mpe = ReadSetOfCurves(self, io, BaseOffset + offsetM, outputChan);
+        if (mpe != NULL) cmsPipelineInsertStage(NewLUT, cmsAT_END, mpe);
+    }
+
+    if (offsetMat != 0) {           
+        mpe = ReadMatrix(self, io, BaseOffset + offsetMat);
+        if (mpe != NULL) cmsPipelineInsertStage(NewLUT, cmsAT_END, mpe);
+    }
+
+    if (offsetB != 0) {                                        
+        mpe = ReadSetOfCurves(self, io, BaseOffset + offsetB, outputChan);
+        if (mpe != NULL) cmsPipelineInsertStage(NewLUT, cmsAT_END, mpe);
+    }
+      
+    *nItems = 1;
+    return NewLUT;
+
+    cmsUNUSED_PARAMETER(SizeOfTag);
+}
+
+// Write a set of curves
+static
+cmsBool  WriteMatrix(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsStage* mpe)
+{   
+    _cmsStageMatrixData* m = (_cmsStageMatrixData*) mpe -> Data;
+
+    // Write the Matrix
+    if (!_cmsWrite15Fixed16Number(io, m -> Double[0])) return FALSE;
+    if (!_cmsWrite15Fixed16Number(io, m -> Double[1])) return FALSE;
+    if (!_cmsWrite15Fixed16Number(io, m -> Double[2])) return FALSE;
+    if (!_cmsWrite15Fixed16Number(io, m -> Double[3])) return FALSE;
+    if (!_cmsWrite15Fixed16Number(io, m -> Double[4])) return FALSE;
+    if (!_cmsWrite15Fixed16Number(io, m -> Double[5])) return FALSE;
+    if (!_cmsWrite15Fixed16Number(io, m -> Double[6])) return FALSE;
+    if (!_cmsWrite15Fixed16Number(io, m -> Double[7])) return FALSE;
+    if (!_cmsWrite15Fixed16Number(io, m -> Double[8])) return FALSE;
+
+    if (m ->Offset != NULL) {
+
+    if (!_cmsWrite15Fixed16Number(io, m -> Offset[0])) return FALSE;
+    if (!_cmsWrite15Fixed16Number(io, m -> Offset[1])) return FALSE;
+    if (!_cmsWrite15Fixed16Number(io, m -> Offset[2])) return FALSE;
+    }
+    else {
+        if (!_cmsWrite15Fixed16Number(io, 0)) return FALSE;
+        if (!_cmsWrite15Fixed16Number(io, 0)) return FALSE;
+        if (!_cmsWrite15Fixed16Number(io, 0)) return FALSE;
+
+    }
+
+
+    return TRUE;
+
+    cmsUNUSED_PARAMETER(self);
+}
+
+
+// Write a set of curves
+static
+cmsBool WriteSetOfCurves(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsTagTypeSignature Type, cmsStage* mpe)
+{   
+    cmsUInt32Number i, n;
+    cmsTagTypeSignature CurrentType;
+    cmsToneCurve** Curves;
+
+
+    n      = cmsStageOutputChannels(mpe);
+    Curves = _cmsStageGetPtrToCurveSet(mpe);
+
+    for (i=0; i < n; i++) {
+
+        // If this is a table-based curve, use curve type even on V4
+        CurrentType = Type;
+
+        if (Curves[i] ->nSegments == 0)
+            CurrentType = cmsSigCurveType;
+        else
+        if (Curves[i] ->Segments[0].Type < 0)
+            CurrentType = cmsSigCurveType;
+
+        if (!_cmsWriteTypeBase(io, CurrentType)) return FALSE;
+
+        switch (CurrentType) {
+
+            case cmsSigCurveType:
+                if (!Type_Curve_Write(self, io, Curves[i], 1)) return FALSE;
+                break;
+
+            case cmsSigParametricCurveType:
+                if (!Type_ParametricCurve_Write(self, io, Curves[i], 1)) return FALSE;
+                break;
+
+            default: 
+                {
+                    char String[5];
+
+                    _cmsTagSignature2String(String, (cmsTagSignature) Type);
+                    cmsSignalError(self ->ContextID, cmsERROR_UNKNOWN_EXTENSION, "Unknown curve type '%s'", String);
+                }               
+                return FALSE;
+        }
+
+        if (!_cmsWriteAlignment(io)) return FALSE;
+    }
+
+
+    return TRUE;
+}
+
+
+static
+cmsBool WriteCLUT(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsUInt8Number  Precision, cmsStage* mpe)
+{
+    cmsUInt8Number  gridPoints[cmsMAXCHANNELS]; // Number of grid points in each dimension.  
+    cmsUInt32Number i;    
+    _cmsStageCLutData* CLUT = ( _cmsStageCLutData*) mpe -> Data;
+
+    if (CLUT ->HasFloatValues) {
+         cmsSignalError(self ->ContextID, cmsERROR_NOT_SUITABLE, "Cannot save floating point data, CLUT are 8 or 16 bit only"); 
+         return FALSE;
+    }
+
+    memset(gridPoints, 0, sizeof(gridPoints));
+    for (i=0; i < (cmsUInt32Number) CLUT ->Params ->nInputs; i++) 
+        gridPoints[i] = (cmsUInt8Number) CLUT ->Params ->nSamples[i];
+
+    if (!io -> Write(io, cmsMAXCHANNELS*sizeof(cmsUInt8Number), gridPoints)) return FALSE;
+
+    if (!_cmsWriteUInt8Number(io, (cmsUInt8Number) Precision)) return FALSE;
+    if (!_cmsWriteUInt8Number(io, 0)) return FALSE;
+    if (!_cmsWriteUInt8Number(io, 0)) return FALSE;
+    if (!_cmsWriteUInt8Number(io, 0)) return FALSE;
+
+    // Precision can be 1 or 2 bytes
+    if (Precision == 1) {
+
+        for (i=0; i < CLUT->nEntries; i++) {
+
+            if (!_cmsWriteUInt8Number(io, FROM_16_TO_8(CLUT->Tab.T[i]))) return FALSE;                
+        }
+    }
+    else  
+        if (Precision == 2) {
+
+            if (!_cmsWriteUInt16Array(io, CLUT->nEntries, CLUT ->Tab.T)) return FALSE;
+        }
+        else {
+             cmsSignalError(self ->ContextID, cmsERROR_UNKNOWN_EXTENSION, "Unknown precision of '%d'", Precision); 
+            return FALSE;
+        }
+
+        if (!_cmsWriteAlignment(io)) return FALSE;
+
+        return TRUE;
+}
+
+
+
+
+static
+cmsBool Type_LUTA2B_Write(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, void* Ptr, cmsUInt32Number nItems)
+{
+    cmsPipeline* Lut = (cmsPipeline*) Ptr;
+    int inputChan, outputChan;
+    cmsStage *A = NULL, *B = NULL, *M = NULL;
+    cmsStage * Matrix = NULL;
+    cmsStage * CLUT = NULL;
+    cmsUInt32Number offsetB = 0, offsetMat = 0, offsetM = 0, offsetC = 0, offsetA = 0;
+    cmsUInt32Number BaseOffset, DirectoryPos, CurrentPos;
+
+    // Get the base for all offsets
+    BaseOffset = io ->Tell(io) - sizeof(_cmsTagBase);
+
+    if (Lut ->Elements != NULL)
+        if (!cmsPipelineCheckAndRetreiveStages(Lut, 1, cmsSigCurveSetElemType, &B))
+            if (!cmsPipelineCheckAndRetreiveStages(Lut, 3, cmsSigCurveSetElemType, cmsSigMatrixElemType, cmsSigCurveSetElemType, &M, &Matrix, &B))          
+                if (!cmsPipelineCheckAndRetreiveStages(Lut, 3, cmsSigCurveSetElemType, cmsSigCLutElemType, cmsSigCurveSetElemType, &A, &CLUT, &B))              
+                    if (!cmsPipelineCheckAndRetreiveStages(Lut, 5, cmsSigCurveSetElemType, cmsSigCLutElemType, cmsSigCurveSetElemType, 
+                        cmsSigMatrixElemType, cmsSigCurveSetElemType, &A, &CLUT, &M, &Matrix, &B)) {    
+
+                            cmsSignalError(self->ContextID, cmsERROR_NOT_SUITABLE, "LUT is not suitable to be saved as LutAToB");
+                            return FALSE;
+                    }
+
+    // Get input, output channels
+    inputChan  = cmsPipelineInputChannels(Lut);
+    outputChan = cmsPipelineOutputChannels(Lut);
+
+    // Write channel count
+    if (!_cmsWriteUInt8Number(io, (cmsUInt8Number) inputChan)) return FALSE;
+    if (!_cmsWriteUInt8Number(io, (cmsUInt8Number) outputChan)) return FALSE;
+    if (!_cmsWriteUInt16Number(io, 0)) return FALSE;
+
+    // Keep directory to be filled latter
+    DirectoryPos = io ->Tell(io);
+
+    // Write the directory
+    if (!_cmsWriteUInt32Number(io, 0)) return FALSE;
+    if (!_cmsWriteUInt32Number(io, 0)) return FALSE;
+    if (!_cmsWriteUInt32Number(io, 0)) return FALSE;
+    if (!_cmsWriteUInt32Number(io, 0)) return FALSE;
+    if (!_cmsWriteUInt32Number(io, 0)) return FALSE;
+
+    if (A != NULL) {
+
+        offsetA = io ->Tell(io) - BaseOffset;
+        if (!WriteSetOfCurves(self, io, cmsSigParametricCurveType, A)) return FALSE;
+    }
+    
+    if (CLUT != NULL) {
+        offsetC = io ->Tell(io) - BaseOffset;
+        if (!WriteCLUT(self, io, Lut ->SaveAs8Bits ? 1 : 2, CLUT)) return FALSE;
+
+    }
+    if (M != NULL) {
+
+        offsetM = io ->Tell(io) - BaseOffset;
+        if (!WriteSetOfCurves(self, io, cmsSigParametricCurveType, M)) return FALSE;
+    }
+    
+    if (Matrix != NULL) {
+        offsetMat = io ->Tell(io) - BaseOffset;
+        if (!WriteMatrix(self, io, Matrix)) return FALSE;
+    }
+
+    if (B != NULL) {
+
+        offsetB = io ->Tell(io) - BaseOffset;
+        if (!WriteSetOfCurves(self, io, cmsSigParametricCurveType, B)) return FALSE;
+    }
+    
+    CurrentPos = io ->Tell(io);
+
+    if (!io ->Seek(io, DirectoryPos)) return FALSE;
+
+    if (!_cmsWriteUInt32Number(io, offsetB)) return FALSE;
+    if (!_cmsWriteUInt32Number(io, offsetMat)) return FALSE;
+    if (!_cmsWriteUInt32Number(io, offsetM)) return FALSE;
+    if (!_cmsWriteUInt32Number(io, offsetC)) return FALSE;
+    if (!_cmsWriteUInt32Number(io, offsetA)) return FALSE;
+
+    if (!io ->Seek(io, CurrentPos)) return FALSE;
+    
+    return TRUE;
+
+    cmsUNUSED_PARAMETER(nItems);
+}
+
+
+static
+void* Type_LUTA2B_Dup(struct _cms_typehandler_struct* self, const void *Ptr, cmsUInt32Number n)
+{
+    return (void*) cmsPipelineDup((cmsPipeline*) Ptr);
+
+    cmsUNUSED_PARAMETER(n);
+    cmsUNUSED_PARAMETER(self);
+}
+
+static
+void Type_LUTA2B_Free(struct _cms_typehandler_struct* self, void* Ptr)
+{    
+    cmsPipelineFree((cmsPipeline*) Ptr);
+    return;
+
+    cmsUNUSED_PARAMETER(self);
+}
+
+
+// LutBToA type 
+
+static
+void* Type_LUTB2A_Read(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsUInt32Number* nItems, cmsUInt32Number SizeOfTag)
+{
+    cmsUInt8Number       inputChan;      // Number of input channels     
+    cmsUInt8Number       outputChan;     // Number of output channels     
+    cmsUInt32Number      BaseOffset;     // Actual position in file
+    cmsUInt32Number      offsetB;        // Offset to first "B" curve    
+    cmsUInt32Number      offsetMat;      // Offset to matrix             
+    cmsUInt32Number      offsetM;        // Offset to first "M" curve    
+    cmsUInt32Number      offsetC;        // Offset to CLUT               
+    cmsUInt32Number      offsetA;        // Offset to first "A" curve
+    cmsStage* mpe;
+    cmsPipeline* NewLUT = NULL;
+
+
+    BaseOffset = io ->Tell(io) - sizeof(_cmsTagBase);
+
+    if (!_cmsReadUInt8Number(io, &inputChan)) return NULL;
+    if (!_cmsReadUInt8Number(io, &outputChan)) return NULL;
+
+    // Padding
+    if (!_cmsReadUInt16Number(io, NULL)) return NULL;
+    
+    if (!_cmsReadUInt32Number(io, &offsetB)) return NULL;
+    if (!_cmsReadUInt32Number(io, &offsetMat)) return NULL;
+    if (!_cmsReadUInt32Number(io, &offsetM)) return NULL;
+    if (!_cmsReadUInt32Number(io, &offsetC)) return NULL;
+    if (!_cmsReadUInt32Number(io, &offsetA)) return NULL;
+
+    // Allocates an empty LUT
+    NewLUT = cmsPipelineAlloc(self ->ContextID, inputChan, outputChan);
+    if (NewLUT == NULL) return NULL;
+    
+    if (offsetB != 0) {                                        
+        mpe = ReadSetOfCurves(self, io, BaseOffset + offsetB, inputChan);
+        if (mpe != NULL) cmsPipelineInsertStage(NewLUT, cmsAT_END, mpe);
+    }
+
+    if (offsetMat != 0) {           
+        mpe = ReadMatrix(self, io, BaseOffset + offsetMat);
+        if (mpe != NULL) cmsPipelineInsertStage(NewLUT, cmsAT_END, mpe);
+    }
+
+    if (offsetM != 0) {
+        mpe = ReadSetOfCurves(self, io, BaseOffset + offsetM, inputChan);
+        if (mpe != NULL) cmsPipelineInsertStage(NewLUT, cmsAT_END, mpe);
+    }
+
+    if (offsetC != 0) {
+        mpe = ReadCLUT(self, io, BaseOffset + offsetC, inputChan, outputChan);
+        if (mpe != NULL) cmsPipelineInsertStage(NewLUT, cmsAT_END, mpe);
+    }
+
+    if (offsetA!= 0) {
+        mpe = ReadSetOfCurves(self, io, BaseOffset + offsetA, outputChan);
+        if (mpe != NULL) cmsPipelineInsertStage(NewLUT, cmsAT_END, mpe);
+    }
+
+    *nItems = 1;
+    return NewLUT;
+
+    cmsUNUSED_PARAMETER(SizeOfTag);
+}
+
+
+/*
+B
+B - Matrix - M
+B - CLUT - A
+B - Matrix - M - CLUT - A
+*/
+
+static
+cmsBool  Type_LUTB2A_Write(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, void* Ptr, cmsUInt32Number nItems)
+{
+    cmsPipeline* Lut = (cmsPipeline*) Ptr;
+    int inputChan, outputChan;
+    cmsStage *A = NULL, *B = NULL, *M = NULL;
+    cmsStage *Matrix = NULL;
+    cmsStage *CLUT = NULL;
+    cmsUInt32Number offsetB = 0, offsetMat = 0, offsetM = 0, offsetC = 0, offsetA = 0;
+    cmsUInt32Number BaseOffset, DirectoryPos, CurrentPos;
+
+
+    BaseOffset = io ->Tell(io) - sizeof(_cmsTagBase);
+
+    if (!cmsPipelineCheckAndRetreiveStages(Lut, 1, cmsSigCurveSetElemType, &B))
+        if (!cmsPipelineCheckAndRetreiveStages(Lut, 3, cmsSigCurveSetElemType, cmsSigMatrixElemType, cmsSigCurveSetElemType, &B, &Matrix, &M))          
+            if (!cmsPipelineCheckAndRetreiveStages(Lut, 3, cmsSigCurveSetElemType, cmsSigCLutElemType, cmsSigCurveSetElemType, &B, &CLUT, &A))              
+                if (!cmsPipelineCheckAndRetreiveStages(Lut, 5, cmsSigCurveSetElemType, cmsSigMatrixElemType, cmsSigCurveSetElemType, 
+                    cmsSigCLutElemType, cmsSigCurveSetElemType, &B, &Matrix, &M, &CLUT, &A)) {                 
+                        cmsSignalError(self->ContextID, cmsERROR_NOT_SUITABLE, "LUT is not suitable to be saved as LutBToA");
+                        return FALSE;
+                }
+
+    inputChan  = cmsPipelineInputChannels(Lut);
+    outputChan = cmsPipelineOutputChannels(Lut);
+    
+    if (!_cmsWriteUInt8Number(io, (cmsUInt8Number) inputChan)) return FALSE;
+    if (!_cmsWriteUInt8Number(io, (cmsUInt8Number) outputChan)) return FALSE;
+    if (!_cmsWriteUInt16Number(io, 0)) return FALSE;
+
+    DirectoryPos = io ->Tell(io);
+
+    if (!_cmsWriteUInt32Number(io, 0)) return FALSE;  
+    if (!_cmsWriteUInt32Number(io, 0)) return FALSE;  
+    if (!_cmsWriteUInt32Number(io, 0)) return FALSE;
+    if (!_cmsWriteUInt32Number(io, 0)) return FALSE;
+    if (!_cmsWriteUInt32Number(io, 0)) return FALSE;
+
+    if (A != NULL) {
+
+        offsetA = io ->Tell(io) - BaseOffset;
+        if (!WriteSetOfCurves(self, io, cmsSigParametricCurveType, A)) return FALSE;
+    }
+    
+    if (CLUT != NULL) {
+        offsetC = io ->Tell(io) - BaseOffset;
+        if (!WriteCLUT(self, io, Lut ->SaveAs8Bits ? 1 : 2, CLUT)) return FALSE;
+
+    }
+    if (M != NULL) {
+
+        offsetM = io ->Tell(io) - BaseOffset;
+        if (!WriteSetOfCurves(self, io, cmsSigParametricCurveType, M)) return FALSE;
+    }
+    
+    if (Matrix != NULL) {
+        offsetMat = io ->Tell(io) - BaseOffset;
+        if (!WriteMatrix(self, io, Matrix)) return FALSE;    
+    }
+
+    if (B != NULL) {
+
+        offsetB = io ->Tell(io) - BaseOffset;
+        if (!WriteSetOfCurves(self, io, cmsSigParametricCurveType, B)) return FALSE;
+    }
+    
+    CurrentPos = io ->Tell(io);
+
+    if (!io ->Seek(io, DirectoryPos)) return FALSE;
+
+    if (!_cmsWriteUInt32Number(io, offsetB)) return FALSE;
+    if (!_cmsWriteUInt32Number(io, offsetMat)) return FALSE;
+    if (!_cmsWriteUInt32Number(io, offsetM)) return FALSE;
+    if (!_cmsWriteUInt32Number(io, offsetC)) return FALSE;
+    if (!_cmsWriteUInt32Number(io, offsetA)) return FALSE;
+
+    if (!io ->Seek(io, CurrentPos)) return FALSE;
+    
+    return TRUE;
+
+    cmsUNUSED_PARAMETER(nItems);
+}
+
+
+
+static
+void* Type_LUTB2A_Dup(struct _cms_typehandler_struct* self, const void *Ptr, cmsUInt32Number n)
+{
+    return (void*) cmsPipelineDup((cmsPipeline*) Ptr);
+
+    cmsUNUSED_PARAMETER(n);
+    cmsUNUSED_PARAMETER(self);
+}
+
+static
+void Type_LUTB2A_Free(struct _cms_typehandler_struct* self, void* Ptr)
+{    
+    cmsPipelineFree((cmsPipeline*) Ptr);
+    return;
+
+    cmsUNUSED_PARAMETER(self);
+}
+
+
+
+// ********************************************************************************
+// Type cmsSigColorantTableType
+// ********************************************************************************
+/*
+The purpose of this tag is to identify the colorants used in the profile by a 
+unique name and set of XYZ or L*a*b* values to give the colorant an unambiguous 
+value. The first colorant listed is the colorant of the first device channel of
+a lut tag. The second colorant listed is the colorant of the second device channel
+of a lut tag, and so on.
+*/
+
+static
+void *Type_ColorantTable_Read(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsUInt32Number* nItems, cmsUInt32Number SizeOfTag)
+{
+    cmsUInt32Number i, Count;
+    cmsNAMEDCOLORLIST* List;
+    char Name[34];
+    cmsUInt16Number PCS[3]; 
+
+
+    if (!_cmsReadUInt32Number(io, &Count)) return NULL;
+
+    if (Count > cmsMAXCHANNELS) {
+        cmsSignalError(self->ContextID, cmsERROR_RANGE, "Too many colorants '%d'", Count);
+        return NULL;
+    }
+
+    List = cmsAllocNamedColorList(self ->ContextID, Count, 0, "", "");
+    for (i=0; i < Count; i++) {
+
+        if (io ->Read(io, Name, 32, 1) != 1) goto Error;
+        Name[33] = 0;
+
+        if (!_cmsReadUInt16Array(io, 3, PCS)) goto Error;
+
+        if (!cmsAppendNamedColor(List, Name, PCS, NULL)) goto Error; 
+                                                            
+    }
+
+    *nItems = 1;
+    return List;
+
+Error:
+    *nItems = 0;
+    cmsFreeNamedColorList(List);
+    return NULL;
+
+    cmsUNUSED_PARAMETER(SizeOfTag);
+}
+
+
+
+// Saves a colorant table. It is using the named color structure for simplicity sake
+static
+cmsBool  Type_ColorantTable_Write(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, void* Ptr, cmsUInt32Number nItems)
+{
+    cmsNAMEDCOLORLIST* NamedColorList = (cmsNAMEDCOLORLIST*) Ptr; 
+    int i, nColors;
+
+    nColors = cmsNamedColorCount(NamedColorList);
+
+    if (!_cmsWriteUInt32Number(io, nColors)) return FALSE;
+
+    for (i=0; i < nColors; i++) {
+
+        char root[33];
+        cmsUInt16Number PCS[3]; 
+
+        if (!cmsNamedColorInfo(NamedColorList, i, root, NULL, NULL, PCS, NULL)) return 0;
+        root[32] = 0;
+
+        if (!io ->Write(io, 32, root)) return FALSE;
+        if (!_cmsWriteUInt16Array(io, 3, PCS)) return FALSE;
+    }
+
+    return TRUE;
+
+    cmsUNUSED_PARAMETER(nItems);
+    cmsUNUSED_PARAMETER(self);
+}
+
+
+static
+void* Type_ColorantTable_Dup(struct _cms_typehandler_struct* self, const void* Ptr, cmsUInt32Number n)
+{
+    cmsNAMEDCOLORLIST* nc = (cmsNAMEDCOLORLIST*) Ptr;
+    return (void*) cmsDupNamedColorList(nc); 
+
+    cmsUNUSED_PARAMETER(n);
+    cmsUNUSED_PARAMETER(self);
+}
+
+
+static
+void Type_ColorantTable_Free(struct _cms_typehandler_struct* self, void* Ptr)
+{    
+    cmsFreeNamedColorList((cmsNAMEDCOLORLIST*) Ptr);
+    return;
+
+    cmsUNUSED_PARAMETER(self);
+}
+
+
+// ********************************************************************************
+// Type cmsSigNamedColor2Type
+// ********************************************************************************
+//
+//The namedColor2Type is a count value and array of structures that provide color 
+//coordinates for 7-bit ASCII color names. For each named color, a PCS and optional 
+//device representation of the color are given. Both representations are 16-bit values. 
+//The device representation corresponds to the header’s “color space of data” field. 
+//This representation should be consistent with the “number of device components”
+//field in the namedColor2Type. If this field is 0, device coordinates are not provided.
+//The PCS representation corresponds to the header’s PCS field. The PCS representation 
+//is always provided. Color names are fixed-length, 32-byte fields including null 
+//termination. In order to maintain maximum portability, it is strongly recommended 
+//that special characters of the 7-bit ASCII set not be used.
+
+static
+void *Type_NamedColor_Read(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsUInt32Number* nItems, cmsUInt32Number SizeOfTag)
+{
+
+    cmsUInt32Number      vendorFlag;     // Bottom 16 bits for ICC use 
+    cmsUInt32Number      count;          // Count of named colors 
+    cmsUInt32Number      nDeviceCoords;  // Num of device coordinates 
+    char                 prefix[32];     // Prefix for each color name 
+    char                 suffix[32];     // Suffix for each color name 
+    cmsNAMEDCOLORLIST*  v;
+    cmsUInt32Number i;
+
+
+    *nItems = 0;
+    if (!_cmsReadUInt32Number(io, &vendorFlag)) return NULL;
+    if (!_cmsReadUInt32Number(io, &count)) return NULL;
+    if (!_cmsReadUInt32Number(io, &nDeviceCoords)) return NULL;
+
+    if (io -> Read(io, prefix, 32, 1) != 1) return NULL;
+    if (io -> Read(io, suffix, 32, 1) != 1) return NULL;
+
+    prefix[31] = suffix[31] = 0;
+
+    v = cmsAllocNamedColorList(self ->ContextID, count, nDeviceCoords, prefix, suffix);
+    if (v == NULL) {
+        cmsSignalError(self->ContextID, cmsERROR_RANGE, "Too many named colors '%d'", count);
+        return NULL;
+    }
+
+    if (nDeviceCoords > cmsMAXCHANNELS) {
+        cmsSignalError(self->ContextID, cmsERROR_RANGE, "Too many device coordinates '%d'", nDeviceCoords);
+        return 0;
+    }
+    for (i=0; i < count; i++) {
+
+        cmsUInt16Number PCS[3];
+        cmsUInt16Number Colorant[cmsMAXCHANNELS];
+        char Root[33];
+
+        memset(Colorant, 0, sizeof(Colorant));
+        if (io -> Read(io, Root, 32, 1) != 1) return NULL;
+        if (!_cmsReadUInt16Array(io, 3, PCS)) goto Error;
+        if (!_cmsReadUInt16Array(io, nDeviceCoords, Colorant)) goto Error;
+
+        if (!cmsAppendNamedColor(v, Root, PCS, Colorant)) goto Error;
+    }
+
+    *nItems = 1;
+    return (void*) v ;
+
+Error:
+    cmsFreeNamedColorList(v);
+    return NULL;
+
+    cmsUNUSED_PARAMETER(SizeOfTag);
+}
+
+
+// Saves a named color list into a named color profile
+static
+cmsBool Type_NamedColor_Write(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, void* Ptr, cmsUInt32Number nItems)
+{
+    cmsNAMEDCOLORLIST* NamedColorList = (cmsNAMEDCOLORLIST*) Ptr;
+    char                prefix[32];     // Prefix for each color name 
+    char                suffix[32];     // Suffix for each color name 
+    int i, nColors;
+
+    nColors = cmsNamedColorCount(NamedColorList);
+
+    if (!_cmsWriteUInt32Number(io, 0)) return FALSE;
+    if (!_cmsWriteUInt32Number(io, nColors)) return FALSE;
+    if (!_cmsWriteUInt32Number(io, NamedColorList ->ColorantCount)) return FALSE;
+
+    strncpy(prefix, (const char*) NamedColorList->Prefix, 32);
+    strncpy(suffix, (const char*) NamedColorList->Suffix, 32);
+
+    suffix[31] = prefix[31] = 0;
+
+    if (!io ->Write(io, 32, prefix)) return FALSE;
+    if (!io ->Write(io, 32, suffix)) return FALSE;
+
+    for (i=0; i < nColors; i++) {
+
+       cmsUInt16Number PCS[3];
+       cmsUInt16Number Colorant[cmsMAXCHANNELS];
+       char Root[33];
+
+        if (!cmsNamedColorInfo(NamedColorList, i, Root, NULL, NULL, PCS, Colorant)) return 0;
+        if (!io ->Write(io, 32 , Root)) return FALSE;
+        if (!_cmsWriteUInt16Array(io, 3, PCS)) return FALSE;   
+        if (!_cmsWriteUInt16Array(io, NamedColorList ->ColorantCount, Colorant)) return FALSE;
+    }
+
+    return TRUE;
+
+    cmsUNUSED_PARAMETER(nItems);
+    cmsUNUSED_PARAMETER(self);
+}
+
+static
+void* Type_NamedColor_Dup(struct _cms_typehandler_struct* self, const void* Ptr, cmsUInt32Number n)
+{
+    cmsNAMEDCOLORLIST* nc = (cmsNAMEDCOLORLIST*) Ptr;
+
+    return (void*) cmsDupNamedColorList(nc);
+
+    cmsUNUSED_PARAMETER(n);
+    cmsUNUSED_PARAMETER(self);
+}
+
+
+static
+void Type_NamedColor_Free(struct _cms_typehandler_struct* self, void* Ptr)
+{    
+    cmsFreeNamedColorList((cmsNAMEDCOLORLIST*) Ptr);
+    return;
+
+    cmsUNUSED_PARAMETER(self);
+}
+
+
+// ********************************************************************************
+// Type cmsSigProfileSequenceDescType
+// ********************************************************************************
+
+// This type is an array of structures, each of which contains information from the 
+// header fields and tags from the original profiles which were combined to create 
+// the final profile. The order of the structures is the order in which the profiles 
+// were combined and includes a structure for the final profile. This provides a 
+// description of the profile sequence from source to destination, 
+// typically used with the DeviceLink profile.
+
+static
+cmsBool ReadEmbeddedText(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsMLU** mlu, cmsUInt32Number SizeOfTag)
+{
+    cmsTagTypeSignature  BaseType;
+    cmsUInt32Number nItems;
+
+    BaseType = _cmsReadTypeBase(io); 
+
+    switch (BaseType) {
+
+       case cmsSigTextType:
+           if (*mlu) cmsMLUfree(*mlu);
+           *mlu = (cmsMLU*)Type_Text_Read(self, io, &nItems, SizeOfTag);
+           return (*mlu != NULL); 
+
+       case cmsSigTextDescriptionType:
+           if (*mlu) cmsMLUfree(*mlu);
+           *mlu =  (cmsMLU*) Type_Text_Description_Read(self, io, &nItems, SizeOfTag);
+           return (*mlu != NULL); 
+
+           /*
+           TBD: Size is needed for MLU, and we have no idea on which is the available size
+           */
+
+       case cmsSigMultiLocalizedUnicodeType:
+           if (*mlu) cmsMLUfree(*mlu);
+           *mlu =  (cmsMLU*) Type_MLU_Read(self, io, &nItems, SizeOfTag);
+           return (*mlu != NULL);
+
+       default: return FALSE;
+    }
+}
+
+
+static
+void *Type_ProfileSequenceDesc_Read(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsUInt32Number* nItems, cmsUInt32Number SizeOfTag)
+{
+    cmsSEQ* OutSeq;
+    cmsUInt32Number i, Count;
+    
+    *nItems = 0;
+    
+    if (!_cmsReadUInt32Number(io, &Count)) return NULL;  
+
+    if (SizeOfTag < sizeof(cmsUInt32Number)) return NULL;
+    SizeOfTag -= sizeof(cmsUInt32Number);
+
+   
+    OutSeq = cmsAllocProfileSequenceDescription(self ->ContextID, Count);
+    if (OutSeq == NULL) return NULL;
+
+    OutSeq ->n = Count;
+    
+    // Get structures as well
+
+    for (i=0; i < Count; i++) {
+        
+        cmsPSEQDESC* sec = &OutSeq -> seq[i];
+
+        if (!_cmsReadUInt32Number(io, &sec ->deviceMfg)) return NULL;
+        if (SizeOfTag < sizeof(cmsUInt32Number)) return NULL;
+        SizeOfTag -= sizeof(cmsUInt32Number);
+
+        if (!_cmsReadUInt32Number(io, &sec ->deviceModel)) return NULL;
+        if (SizeOfTag < sizeof(cmsUInt32Number)) return NULL;
+        SizeOfTag -= sizeof(cmsUInt32Number);
+
+        if (!_cmsReadUInt64Number(io, &sec ->attributes)) return NULL;        
+        if (SizeOfTag < sizeof(cmsUInt32Number)) return NULL;
+        SizeOfTag -= sizeof(cmsUInt64Number);
+
+        if (!_cmsReadUInt32Number(io, (cmsUInt32Number *)&sec ->technology)) return NULL;
+        if (SizeOfTag < sizeof(cmsUInt32Number)) return NULL;
+        SizeOfTag -= sizeof(cmsUInt32Number);
+
+        if (!ReadEmbeddedText(self, io, &sec ->Manufacturer, SizeOfTag)) return NULL;
+        if (!ReadEmbeddedText(self, io, &sec ->Model, SizeOfTag)) return NULL;
+    }
+
+    *nItems = 1;
+    return OutSeq;
+}
+
+
+// Aux--Embed a text description type. It can be of type text description or multilocalized unicode
+// and it depends of the version number passed on cmsTagDescriptor structure instead of stack
+static
+cmsBool  SaveDescription(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsMLU* Text)
+{
+    if (self ->ICCVersion < 0x4000000) { 
+        
+        if (!_cmsWriteTypeBase(io, cmsSigTextDescriptionType)) return FALSE;
+        return Type_Text_Description_Write(self, io, Text, 1);
+    }
+    else {
+        if (!_cmsWriteTypeBase(io, cmsSigMultiLocalizedUnicodeType)) return FALSE;
+        return Type_MLU_Write(self, io, Text, 1);
+    }
+}
+
+
+static
+cmsBool  Type_ProfileSequenceDesc_Write(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, void* Ptr, cmsUInt32Number nItems)
+{
+    cmsSEQ* Seq = (cmsSEQ*) Ptr;
+    cmsUInt32Number i;
+
+    if (!_cmsWriteUInt32Number(io, Seq->n)) return FALSE;
+    
+    for (i=0; i < Seq ->n; i++) {
+
+        cmsPSEQDESC* sec = &Seq -> seq[i];
+
+        if (!_cmsWriteUInt32Number(io, sec ->deviceMfg)) return FALSE;
+        if (!_cmsWriteUInt32Number(io, sec ->deviceModel)) return FALSE;
+        if (!_cmsWriteUInt64Number(io, sec ->attributes)) return FALSE;        
+        if (!_cmsWriteUInt32Number(io, sec ->technology)) return FALSE;
+        
+        if (!SaveDescription(self, io, sec ->Manufacturer)) return FALSE;       
+        if (!SaveDescription(self, io, sec ->Model)) return FALSE;          
+    }
+    
+     return TRUE;  
+
+     cmsUNUSED_PARAMETER(nItems);
+}
+
+
+static
+void* Type_ProfileSequenceDesc_Dup(struct _cms_typehandler_struct* self, const void* Ptr, cmsUInt32Number n)
+{   
+    return (void*) cmsDupProfileSequenceDescription((cmsSEQ*) Ptr);
+
+    cmsUNUSED_PARAMETER(n);
+    cmsUNUSED_PARAMETER(self);
+}
+
+static
+void Type_ProfileSequenceDesc_Free(struct _cms_typehandler_struct* self, void* Ptr)
+{    
+    cmsFreeProfileSequenceDescription((cmsSEQ*) Ptr);
+    return;
+
+    cmsUNUSED_PARAMETER(self);
+}
+
+
+// ********************************************************************************
+// Type cmsSigProfileSequenceIdType
+// ********************************************************************************
+/*
+In certain workflows using ICC Device Link Profiles, it is necessary to identify the 
+original profiles that were combined to create the Device Link Profile.
+This type is an array of structures, each of which contains information for 
+identification of a profile used in a sequence
+*/
+
+
+static
+cmsBool ReadSeqID(struct _cms_typehandler_struct* self, 
+                                             cmsIOHANDLER* io,
+                                             void* Cargo,
+                                             cmsUInt32Number n, 
+                                             cmsUInt32Number SizeOfTag)
+{
+    cmsSEQ* OutSeq = (cmsSEQ*) Cargo;
+    cmsPSEQDESC* seq = &OutSeq ->seq[n];
+    
+    if (io -> Read(io, seq ->ProfileID.ID8, 16, 1) != 1) return FALSE;
+    if (!ReadEmbeddedText(self, io, &seq ->Description, SizeOfTag)) return FALSE;
+
+    return TRUE;
+}
+
+
+
+static
+void *Type_ProfileSequenceId_Read(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsUInt32Number* nItems, cmsUInt32Number SizeOfTag)
+{
+    cmsSEQ* OutSeq;
+    cmsUInt32Number Count;
+    cmsUInt32Number BaseOffset;
+
+    *nItems = 0;
+
+    // Get actual position as a basis for element offsets
+    BaseOffset = io ->Tell(io) - sizeof(_cmsTagBase);
+
+    // Get table count
+    if (!_cmsReadUInt32Number(io, &Count)) return NULL;  
+    SizeOfTag -= sizeof(cmsUInt32Number);
+
+    // Allocate an empty structure
+    OutSeq = cmsAllocProfileSequenceDescription(self ->ContextID, Count);
+    if (OutSeq == NULL) return NULL;
+
+
+    // Read the position table
+    if (!ReadPositionTable(self, io, Count, BaseOffset, OutSeq, ReadSeqID)) {
+
+        cmsFreeProfileSequenceDescription(OutSeq);
+        return NULL;
+    }
+
+    // Success
+    *nItems = 1;
+    return OutSeq;
+
+}
+
+
+static
+cmsBool WriteSeqID(struct _cms_typehandler_struct* self, 
+                                             cmsIOHANDLER* io,
+                                             void* Cargo,
+                                             cmsUInt32Number n, 
+                                             cmsUInt32Number SizeOfTag)
+{
+    cmsSEQ* Seq = (cmsSEQ*) Cargo;
+    
+    if (!io ->Write(io, 16, Seq ->seq[n].ProfileID.ID8)) return FALSE;
+
+    // Store here the MLU
+    if (!SaveDescription(self, io, Seq ->seq[n].Description)) return FALSE;  
+
+    return TRUE;
+
+    cmsUNUSED_PARAMETER(SizeOfTag);
+}
+
+static
+cmsBool  Type_ProfileSequenceId_Write(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, void* Ptr, cmsUInt32Number nItems)
+{
+    cmsSEQ* Seq = (cmsSEQ*) Ptr;
+    cmsUInt32Number BaseOffset;
+
+    // Keep the base offset
+    BaseOffset = io ->Tell(io) - sizeof(_cmsTagBase);
+    
+    // This is the table count
+    if (!_cmsWriteUInt32Number(io, Seq ->n)) return FALSE;
+
+    // This is the position table and content
+    if (!WritePositionTable(self, io, 0, Seq ->n, BaseOffset, Seq, WriteSeqID)) return FALSE;
+          
+    return TRUE;
+
+    cmsUNUSED_PARAMETER(nItems);
+}
+
+static
+void* Type_ProfileSequenceId_Dup(struct _cms_typehandler_struct* self, const void* Ptr, cmsUInt32Number n)
+{   
+    return (void*) cmsDupProfileSequenceDescription((cmsSEQ*) Ptr);
+
+    cmsUNUSED_PARAMETER(n);
+    cmsUNUSED_PARAMETER(self);
+}
+
+static
+void Type_ProfileSequenceId_Free(struct _cms_typehandler_struct* self, void* Ptr)
+{    
+    cmsFreeProfileSequenceDescription((cmsSEQ*) Ptr);
+    return;
+
+    cmsUNUSED_PARAMETER(self);
+}
+
+
+// ********************************************************************************
+// Type cmsSigUcrBgType
+// ********************************************************************************
+/*
+This type contains curves representing the under color removal and black
+generation and a text string which is a general description of the method used
+for the ucr/bg.
+*/
+
+static
+void *Type_UcrBg_Read(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsUInt32Number* nItems, cmsUInt32Number SizeOfTag)
+{
+    cmsUcrBg* n = (cmsUcrBg*) _cmsMallocZero(self ->ContextID, sizeof(cmsUcrBg));
+    cmsUInt32Number CountUcr, CountBg;
+    char* ASCIIString;
+
+    *nItems = 0;
+    if (n == NULL) return NULL;
+
+    // First curve is Under color removal
+    if (!_cmsReadUInt32Number(io, &CountUcr)) return NULL;
+    if (SizeOfTag < sizeof(cmsUInt32Number)) return NULL;
+    SizeOfTag -= sizeof(cmsUInt32Number);
+
+    n ->Ucr = cmsBuildTabulatedToneCurve16(self ->ContextID, CountUcr, NULL);
+    if (n ->Ucr == NULL) return NULL;
+
+    if (!_cmsReadUInt16Array(io, CountUcr, n ->Ucr->Table16)) return NULL;
+    if (SizeOfTag < sizeof(cmsUInt32Number)) return NULL;
+    SizeOfTag -= CountUcr * sizeof(cmsUInt16Number);
+
+    // Second curve is Black generation
+    if (!_cmsReadUInt32Number(io, &CountBg)) return NULL;
+    if (SizeOfTag < sizeof(cmsUInt32Number)) return NULL;
+    SizeOfTag -= sizeof(cmsUInt32Number);
+
+    n ->Bg = cmsBuildTabulatedToneCurve16(self ->ContextID, CountBg, NULL);
+    if (n ->Bg == NULL) return NULL;
+    if (!_cmsReadUInt16Array(io, CountBg, n ->Bg->Table16)) return NULL;
+    if (SizeOfTag < CountBg * sizeof(cmsUInt16Number)) return NULL;
+    SizeOfTag -= CountBg * sizeof(cmsUInt16Number);
+    if (SizeOfTag == UINT_MAX) return NULL;
+
+    // Now comes the text. The length is specified by the tag size
+    n ->Desc = cmsMLUalloc(self ->ContextID, 1);
+    if (n ->Desc == NULL) return NULL;
+
+    ASCIIString = (char*) _cmsMalloc(self ->ContextID, SizeOfTag + 1);
+    if (io ->Read(io, ASCIIString, sizeof(char), SizeOfTag) != SizeOfTag) return NULL;
+    ASCIIString[SizeOfTag] = 0;
+    cmsMLUsetASCII(n ->Desc, cmsNoLanguage, cmsNoCountry, ASCIIString);
+    _cmsFree(self ->ContextID, ASCIIString);
+
+    *nItems = 1;
+    return (void*) n;
+}
+
+static
+cmsBool  Type_UcrBg_Write(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, void* Ptr, cmsUInt32Number nItems)
+{
+    cmsUcrBg* Value = (cmsUcrBg*) Ptr;
+    cmsUInt32Number TextSize;
+    char* Text;
+
+    // First curve is Under color removal
+    if (!_cmsWriteUInt32Number(io, Value ->Ucr ->nEntries)) return FALSE;
+    if (!_cmsWriteUInt16Array(io, Value ->Ucr ->nEntries, Value ->Ucr ->Table16)) return FALSE;
+
+    // Then black generation    
+    if (!_cmsWriteUInt32Number(io, Value ->Bg ->nEntries)) return FALSE;
+    if (!_cmsWriteUInt16Array(io, Value ->Bg ->nEntries, Value ->Bg ->Table16)) return FALSE;
+
+    // Now comes the text. The length is specified by the tag size
+    TextSize = cmsMLUgetASCII(Value ->Desc, cmsNoLanguage, cmsNoCountry, NULL, 0);
+    Text     = (char*) _cmsMalloc(self ->ContextID, TextSize);
+    if (cmsMLUgetASCII(Value ->Desc, cmsNoLanguage, cmsNoCountry, Text, TextSize) != TextSize) return FALSE;
+
+    if (!io ->Write(io, TextSize, Text)) return FALSE;      
+    _cmsFree(self ->ContextID, Text);
+
+    return TRUE;
+
+    cmsUNUSED_PARAMETER(nItems);
+}
+
+static
+void* Type_UcrBg_Dup(struct _cms_typehandler_struct* self, const void *Ptr, cmsUInt32Number n)
+{
+    cmsUcrBg* Src = (cmsUcrBg*) Ptr;
+    cmsUcrBg* NewUcrBg = (cmsUcrBg*) _cmsMallocZero(self ->ContextID, sizeof(cmsUcrBg));
+
+    if (NewUcrBg == NULL) return NULL;
+
+    NewUcrBg ->Bg   = cmsDupToneCurve(Src ->Bg);
+    NewUcrBg ->Ucr  = cmsDupToneCurve(Src ->Ucr);
+    NewUcrBg ->Desc = cmsMLUdup(Src ->Desc);
+
+    return (void*) NewUcrBg;
+
+    cmsUNUSED_PARAMETER(n);
+}
+
+static
+void Type_UcrBg_Free(struct _cms_typehandler_struct* self, void *Ptr)
+{
+   cmsUcrBg* Src = (cmsUcrBg*) Ptr;
+   
+   if (Src ->Ucr) cmsFreeToneCurve(Src ->Ucr);
+   if (Src ->Bg)  cmsFreeToneCurve(Src ->Bg);
+   if (Src ->Desc) cmsMLUfree(Src ->Desc);
+   
+   _cmsFree(self ->ContextID, Ptr);
+}
+
+// ********************************************************************************
+// Type cmsSigCrdInfoType
+// ********************************************************************************
+
+/*
+This type contains the PostScript product name to which this profile corresponds
+and the names of the companion CRDs. Recall that a single profile can generate
+multiple CRDs. It is implemented as a MLU being the language code "PS" and then
+country varies for each element:
+
+                nm: PostScript product name
+                #0: Rendering intent 0 CRD name
+                #1: Rendering intent 1 CRD name
+                #2: Rendering intent 2 CRD name
+                #3: Rendering intent 3 CRD name
+*/
+
+
+
+// Auxiliar, read an string specified as count + string
+static
+cmsBool  ReadCountAndSting(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsMLU* mlu, cmsUInt32Number* SizeOfTag, const char* Section)
+{
+    cmsUInt32Number Count;
+    char* Text;
+
+    if (*SizeOfTag < sizeof(cmsUInt32Number)) return FALSE; 
+    
+    if (!_cmsReadUInt32Number(io, &Count)) return FALSE;
+
+    if (Count > UINT_MAX - sizeof(cmsUInt32Number)) return FALSE;
+    if (*SizeOfTag < Count + sizeof(cmsUInt32Number)) return FALSE; 
+
+    Text     = (char*) _cmsMalloc(self ->ContextID, Count+1);
+    if (Text == NULL) return FALSE;
+
+    if (io ->Read(io, Text, sizeof(cmsUInt8Number), Count) != Count) {
+        _cmsFree(self ->ContextID, Text);
+        return FALSE;
+    }
+
+    Text[Count] = 0;
+   
+    cmsMLUsetASCII(mlu, "PS", Section, Text);
+    _cmsFree(self ->ContextID, Text);
+
+    *SizeOfTag -= (Count + sizeof(cmsUInt32Number));
+    return TRUE;    
+}
+
+static
+cmsBool  WriteCountAndSting(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsMLU* mlu, const char* Section)
+{
+ cmsUInt32Number TextSize;
+ char* Text;    
+    
+    TextSize = cmsMLUgetASCII(mlu, "PS", Section, NULL, 0);
+    Text     = (char*) _cmsMalloc(self ->ContextID, TextSize);
+
+    if (!_cmsWriteUInt32Number(io, TextSize)) return FALSE; 
+
+    if (cmsMLUgetASCII(mlu, "PS", Section, Text, TextSize) == 0) return FALSE;
+
+    if (!io ->Write(io, TextSize, Text)) return FALSE;      
+    _cmsFree(self ->ContextID, Text);
+
+    return TRUE;
+}
+
+static
+void *Type_CrdInfo_Read(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsUInt32Number* nItems, cmsUInt32Number SizeOfTag)
+{
+    cmsMLU* mlu = cmsMLUalloc(self ->ContextID, 5);
+
+    *nItems = 0;
+    if (!ReadCountAndSting(self, io, mlu, &SizeOfTag, "nm")) goto Error;
+    if (!ReadCountAndSting(self, io, mlu, &SizeOfTag, "#0")) goto Error;
+    if (!ReadCountAndSting(self, io, mlu, &SizeOfTag, "#1")) goto Error;
+    if (!ReadCountAndSting(self, io, mlu, &SizeOfTag, "#2")) goto Error;
+    if (!ReadCountAndSting(self, io, mlu, &SizeOfTag, "#3")) goto Error;
+  
+    *nItems = 1;
+    return (void*) mlu;
+
+Error:
+    cmsMLUfree(mlu);
+    return NULL;
+
+}
+
+static
+cmsBool  Type_CrdInfo_Write(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, void* Ptr, cmsUInt32Number nItems)
+{
+
+    cmsMLU* mlu = (cmsMLU*) Ptr;
+
+    if (!WriteCountAndSting(self, io, mlu, "nm")) goto Error;
+    if (!WriteCountAndSting(self, io, mlu, "#0")) goto Error;
+    if (!WriteCountAndSting(self, io, mlu, "#1")) goto Error;
+    if (!WriteCountAndSting(self, io, mlu, "#2")) goto Error;
+    if (!WriteCountAndSting(self, io, mlu, "#3")) goto Error;
+
+    return TRUE;
+
+Error:
+    return FALSE;
+
+    cmsUNUSED_PARAMETER(nItems);
+}
+
+
+static
+void* Type_CrdInfo_Dup(struct _cms_typehandler_struct* self, const void *Ptr, cmsUInt32Number n)
+{
+    return (void*) cmsMLUdup((cmsMLU*) Ptr);
+
+    cmsUNUSED_PARAMETER(n);
+    cmsUNUSED_PARAMETER(self);
+}
+
+static
+void Type_CrdInfo_Free(struct _cms_typehandler_struct* self, void *Ptr)
+{    
+    cmsMLUfree((cmsMLU*) Ptr);
+    return;
+
+    cmsUNUSED_PARAMETER(self);
+}
+
+// ********************************************************************************
+// Type cmsSigScreeningType
+// ********************************************************************************
+//
+//The screeningType describes various screening parameters including screen
+//frequency, screening angle, and spot shape.
+
+static
+void *Type_Screening_Read(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsUInt32Number* nItems, cmsUInt32Number SizeOfTag)
+{
+    cmsScreening* sc = NULL;
+    cmsUInt32Number i;
+    
+    sc = (cmsScreening*) _cmsMallocZero(self ->ContextID, sizeof(cmsScreening));
+    if (sc == NULL) return NULL;
+
+    *nItems = 0;
+    
+    if (!_cmsReadUInt32Number(io, &sc ->Flag)) goto Error;
+    if (!_cmsReadUInt32Number(io, &sc ->nChannels)) goto Error;
+
+    if (sc ->nChannels > cmsMAXCHANNELS - 1)
+        sc ->nChannels = cmsMAXCHANNELS - 1;
+
+    for (i=0; i < sc ->nChannels; i++) {
+
+        if (!_cmsRead15Fixed16Number(io, &sc ->Channels[i].Frequency)) goto Error;      
+        if (!_cmsRead15Fixed16Number(io, &sc ->Channels[i].ScreenAngle)) goto Error;        
+        if (!_cmsReadUInt32Number(io, &sc ->Channels[i].SpotShape)) goto Error;
+    }
+
+
+    *nItems = 1;
+
+    return (void*) sc;
+
+Error:
+    if (sc != NULL) 
+        _cmsFree(self ->ContextID, sc);
+
+    return NULL;
+
+    cmsUNUSED_PARAMETER(SizeOfTag);
+}
+
+
+static
+cmsBool Type_Screening_Write(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, void* Ptr, cmsUInt32Number nItems)
+{
+    cmsScreening* sc = (cmsScreening* ) Ptr; 
+    cmsUInt32Number i;
+        
+    if (!_cmsWriteUInt32Number(io, sc ->Flag)) return FALSE;
+    if (!_cmsWriteUInt32Number(io, sc ->nChannels)) return FALSE;
+
+    for (i=0; i < sc ->nChannels; i++) {
+
+        if (!_cmsWrite15Fixed16Number(io, sc ->Channels[i].Frequency)) return FALSE;
+        if (!_cmsWrite15Fixed16Number(io, sc ->Channels[i].ScreenAngle)) return FALSE;
+        if (!_cmsWriteUInt32Number(io, sc ->Channels[i].SpotShape)) return FALSE;
+    }
+
+    return TRUE;
+
+    cmsUNUSED_PARAMETER(nItems);
+    cmsUNUSED_PARAMETER(self);
+}
+
+
+static
+void* Type_Screening_Dup(struct _cms_typehandler_struct* self, const void *Ptr, cmsUInt32Number n)
+{
+   return _cmsDupMem(self ->ContextID, Ptr, sizeof(cmsScreening));
+
+   cmsUNUSED_PARAMETER(n);
+}
+
+
+static
+void Type_Screening_Free(struct _cms_typehandler_struct* self, void* Ptr)
+{
+   _cmsFree(self ->ContextID, Ptr);
+}
+
+// ********************************************************************************
+// Type cmsSigViewingConditionsType
+// ********************************************************************************
+//
+//This type represents a set of viewing condition parameters including: 
+//CIE ’absolute’ illuminant white point tristimulus values and CIE ’absolute’ 
+//surround tristimulus values.
+
+static
+void *Type_ViewingConditions_Read(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsUInt32Number* nItems, cmsUInt32Number SizeOfTag)
+{
+    cmsICCViewingConditions* vc = NULL;
+    
+    vc = (cmsICCViewingConditions*) _cmsMallocZero(self ->ContextID, sizeof(cmsICCViewingConditions));
+    if (vc == NULL) return NULL;
+
+    *nItems = 0;
+        
+    if (!_cmsReadXYZNumber(io, &vc ->IlluminantXYZ)) goto Error;
+    if (!_cmsReadXYZNumber(io, &vc ->SurroundXYZ)) goto Error;
+    if (!_cmsReadUInt32Number(io, &vc ->IlluminantType)) goto Error;
+
+    *nItems = 1;
+
+    return (void*) vc;
+
+Error:
+    if (vc != NULL) 
+        _cmsFree(self ->ContextID, vc);
+
+    return NULL;
+
+    cmsUNUSED_PARAMETER(SizeOfTag);
+}
+
+
+static
+cmsBool Type_ViewingConditions_Write(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, void* Ptr, cmsUInt32Number nItems)
+{
+    cmsICCViewingConditions* sc = (cmsICCViewingConditions* ) Ptr; 
+            
+    if (!_cmsWriteXYZNumber(io, &sc ->IlluminantXYZ)) return FALSE;
+    if (!_cmsWriteXYZNumber(io, &sc ->SurroundXYZ)) return FALSE;
+    if (!_cmsWriteUInt32Number(io, sc ->IlluminantType)) return FALSE;
+
+    return TRUE;
+
+    cmsUNUSED_PARAMETER(nItems);
+    cmsUNUSED_PARAMETER(self);
+}
+
+
+static
+void* Type_ViewingConditions_Dup(struct _cms_typehandler_struct* self, const void *Ptr, cmsUInt32Number n)
+{
+   return _cmsDupMem(self ->ContextID, Ptr, sizeof(cmsScreening));
+
+   cmsUNUSED_PARAMETER(n);
+}
+
+
+static
+void Type_ViewingConditions_Free(struct _cms_typehandler_struct* self, void* Ptr)
+{
+   _cmsFree(self ->ContextID, Ptr);
+}
+
+
+// ********************************************************************************
+// Type cmsSigMultiProcessElementType
+// ********************************************************************************
+
+
+static
+void* GenericMPEdup(struct _cms_typehandler_struct* self, const void *Ptr, cmsUInt32Number n)
+{
+    return (void*) cmsStageDup((cmsStage*) Ptr);
+
+    cmsUNUSED_PARAMETER(n);
+    cmsUNUSED_PARAMETER(self);
+}
+
+static
+void GenericMPEfree(struct _cms_typehandler_struct* self, void *Ptr)
+{    
+    cmsStageFree((cmsStage*) Ptr);
+    return;
+
+    cmsUNUSED_PARAMETER(self);
+}
+
+// Each curve is stored in one or more curve segments, with break-points specified between curve segments.
+// The first curve segment always starts at –Infinity, and the last curve segment always ends at +Infinity. The
+// first and last curve segments shall be specified in terms of a formula, whereas the other segments shall be
+// specified either in terms of a formula, or by a sampled curve.
+
+
+// Read an embedded segmented curve
+static
+cmsToneCurve* ReadSegmentedCurve(struct _cms_typehandler_struct* self, cmsIOHANDLER* io)
+{   
+    cmsCurveSegSignature ElementSig;
+    cmsUInt32Number i, j;
+    cmsUInt16Number nSegments;
+    cmsCurveSegment*  Segments;
+    cmsToneCurve* Curve;
+    cmsFloat32Number PrevBreak = -1E22F;    // - infinite
+ 
+    // Take signature and channels for each element.
+     if (!_cmsReadUInt32Number(io, (cmsUInt32Number*) &ElementSig)) return NULL;    
+
+     // That should be a segmented curve
+     if (ElementSig != cmsSigSegmentedCurve) return NULL;
+
+     if (!_cmsReadUInt32Number(io, NULL)) return NULL;
+     if (!_cmsReadUInt16Number(io, &nSegments)) return NULL;
+     if (!_cmsReadUInt16Number(io, NULL)) return NULL;
+
+     if (nSegments < 1) return NULL;
+     Segments = (cmsCurveSegment*) _cmsCalloc(self ->ContextID, nSegments, sizeof(cmsCurveSegment));
+     if (Segments == NULL) return NULL;
+
+     // Read breakpoints
+     for (i=0; i < (cmsUInt32Number) nSegments - 1; i++) {
+
+         Segments[i].x0 = PrevBreak;
+         if (!_cmsReadFloat32Number(io, &Segments[i].x1)) goto Error;
+         PrevBreak = Segments[i].x1;
+     }
+
+     Segments[nSegments-1].x0 = PrevBreak;
+     Segments[nSegments-1].x1 = 1E22F;     // A big cmsFloat32Number number
+
+     // Read segments
+     for (i=0; i < nSegments; i++) {
+
+          if (!_cmsReadUInt32Number(io, (cmsUInt32Number*) &ElementSig)) goto Error;
+          if (!_cmsReadUInt32Number(io, NULL)) goto Error;
+         
+           switch (ElementSig) {
+
+            case cmsSigFormulaCurveSeg: {
+                                       
+                cmsUInt16Number Type;
+                cmsUInt32Number ParamsByType[] = {4, 5, 5 };
+
+                if (!_cmsReadUInt16Number(io, &Type)) goto Error;
+                if (!_cmsReadUInt16Number(io, NULL)) goto Error;
+
+                Segments[i].Type = Type + 6;
+                if (Type > 2) goto Error;
+
+                for (j=0; j < ParamsByType[Type]; j++) {
+
+                    cmsFloat32Number f;
+                    if (!_cmsReadFloat32Number(io, &f)) goto Error;
+                    Segments[i].Params[j] = f;
+                }
+                }
+                break;
+
+
+            case cmsSigSampledCurveSeg: {
+                cmsUInt32Number Count;
+
+                if (!_cmsReadUInt32Number(io, &Count)) return NULL;
+
+                Segments[i].nGridPoints = Count;
+                Segments[i].SampledPoints = (cmsFloat32Number*) _cmsCalloc(self ->ContextID, Count, sizeof(cmsFloat32Number));
+                if (Segments[i].SampledPoints == NULL) goto Error;
+
+                for (j=0; j < Count; j++) {
+                    if (!_cmsReadFloat32Number(io, &Segments[i].SampledPoints[j])) goto Error;
+                }
+                }
+                break;
+
+            default:
+                {
+                char String[5];
+
+                _cmsTagSignature2String(String, (cmsTagSignature) ElementSig);
+                cmsSignalError(self->ContextID, cmsERROR_UNKNOWN_EXTENSION, "Unknown curve element type '%s' found.", String);
+                }
+                return NULL;                
+
+         }
+     }
+
+     Curve = cmsBuildSegmentedToneCurve(self ->ContextID, nSegments, Segments);
+
+     for (i=0; i < nSegments; i++) {
+         if (Segments[i].SampledPoints) _cmsFree(self ->ContextID, Segments[i].SampledPoints);
+     }
+     _cmsFree(self ->ContextID, Segments);
+     return Curve;
+
+Error:
+     if (Segments) _cmsFree(self ->ContextID, Segments);
+     return NULL;
+}
+
+
+static
+cmsBool ReadMPECurve(struct _cms_typehandler_struct* self, 
+                     cmsIOHANDLER* io,
+                     void* Cargo,
+                     cmsUInt32Number n, 
+                     cmsUInt32Number SizeOfTag)
+{
+      cmsToneCurve** GammaTables = ( cmsToneCurve**) Cargo;
+
+      GammaTables[n] = ReadSegmentedCurve(self, io);
+      return (GammaTables[n] != NULL);
+
+      cmsUNUSED_PARAMETER(SizeOfTag);
+}
+
+static
+void *Type_MPEcurve_Read(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsUInt32Number* nItems, cmsUInt32Number SizeOfTag)
+{
+    cmsStage* mpe = NULL;
+    cmsUInt16Number InputChans, OutputChans;
+    cmsUInt32Number i, BaseOffset;
+    cmsToneCurve** GammaTables;
+
+    *nItems = 0;
+
+    // Get actual position as a basis for element offsets
+    BaseOffset = io ->Tell(io) - sizeof(_cmsTagBase);
+
+    if (!_cmsReadUInt16Number(io, &InputChans)) return NULL;
+    if (!_cmsReadUInt16Number(io, &OutputChans)) return NULL;
+
+    if (InputChans != OutputChans) return NULL;
+
+    GammaTables = (cmsToneCurve**) _cmsCalloc(self ->ContextID, InputChans, sizeof(cmsToneCurve*));
+    if (GammaTables == NULL) return NULL;
+
+    if (ReadPositionTable(self, io, InputChans, BaseOffset, GammaTables, ReadMPECurve)) {
+
+        mpe = cmsStageAllocToneCurves(self ->ContextID, InputChans, GammaTables);
+    }
+    else {
+        mpe = NULL;
+    }
+
+    for (i=0; i < InputChans; i++) {
+        if (GammaTables[i]) cmsFreeToneCurve(GammaTables[i]);
+    }
+
+    _cmsFree(self ->ContextID, GammaTables);
+    *nItems = (mpe != NULL) ? 1 : 0;
+    return mpe;
+
+    cmsUNUSED_PARAMETER(SizeOfTag);
+}
+
+
+// Write a single segmented curve. NO CHECK IS PERFORMED ON VALIDITY
+static
+cmsBool WriteSegmentedCurve(cmsIOHANDLER* io, cmsToneCurve* g)
+{
+    cmsUInt32Number i, j;
+    cmsCurveSegment* Segments = g ->Segments;
+    cmsUInt32Number nSegments = g ->nSegments;
+
+    if (!_cmsWriteUInt32Number(io, cmsSigSegmentedCurve)) goto Error;    
+    if (!_cmsWriteUInt32Number(io, 0)) goto Error;  
+    if (!_cmsWriteUInt16Number(io, (cmsUInt16Number) nSegments)) goto Error;    
+    if (!_cmsWriteUInt16Number(io, 0)) goto Error;  
+
+    // Write the break-points
+    for (i=0; i < nSegments - 1; i++) {
+        if (!_cmsWriteFloat32Number(io, Segments[i].x1)) goto Error;
+    }
+
+    // Write the segments
+    for (i=0; i < g ->nSegments; i++) {
+
+        cmsCurveSegment* ActualSeg = Segments + i;
+
+        if (ActualSeg -> Type == 0) {
+
+            // This is a sampled curve
+            if (!_cmsWriteUInt32Number(io, (cmsUInt32Number) cmsSigSampledCurveSeg)) goto Error;
+            if (!_cmsWriteUInt32Number(io, 0)) goto Error;
+            if (!_cmsWriteUInt32Number(io, ActualSeg -> nGridPoints)) goto Error;
+
+            for (j=0; j < g ->Segments[i].nGridPoints; j++) {
+                if (!_cmsWriteFloat32Number(io, ActualSeg -> SampledPoints[j])) goto Error;
+            }
+
+        }
+        else {
+            int Type;
+            cmsUInt32Number ParamsByType[] = { 4, 5, 5 };
+
+            // This is a formula-based
+            if (!_cmsWriteUInt32Number(io, (cmsUInt32Number) cmsSigFormulaCurveSeg)) goto Error;
+            if (!_cmsWriteUInt32Number(io, 0)) goto Error;
+
+            // We only allow 1, 2 and 3 as types
+            Type = ActualSeg ->Type - 6;
+            if (Type > 2 || Type < 0) goto Error;
+
+            if (!_cmsWriteUInt16Number(io, (cmsUInt16Number) Type)) goto Error;
+            if (!_cmsWriteUInt16Number(io, 0)) goto Error;
+
+            for (j=0; j < ParamsByType[Type]; j++) {
+                if (!_cmsWriteFloat32Number(io, (cmsFloat32Number) ActualSeg ->Params[j])) goto Error;
+            }
+        }         
+
+        // It seems there is no need to align. Code is here, and for safety commented out
+        // if (!_cmsWriteAlignment(io)) goto Error;
+    }
+
+    return TRUE;
+
+Error:
+    return FALSE;
+}
+
+
+static
+cmsBool WriteMPECurve(struct _cms_typehandler_struct* self, 
+                      cmsIOHANDLER* io,
+                      void* Cargo,
+                      cmsUInt32Number n, 
+                      cmsUInt32Number SizeOfTag)
+{
+    _cmsStageToneCurvesData* Curves  = (_cmsStageToneCurvesData*) Cargo;
+
+    return WriteSegmentedCurve(io, Curves ->TheCurves[n]);
+
+    cmsUNUSED_PARAMETER(SizeOfTag);
+    cmsUNUSED_PARAMETER(self);
+}
+
+// Write a curve, checking first for validity
+static
+cmsBool  Type_MPEcurve_Write(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, void* Ptr, cmsUInt32Number nItems)
+{   
+    cmsUInt32Number BaseOffset;
+    cmsStage* mpe = (cmsStage*) Ptr;
+    _cmsStageToneCurvesData* Curves = (_cmsStageToneCurvesData*) mpe ->Data;
+
+    BaseOffset = io ->Tell(io) - sizeof(_cmsTagBase);
+
+    // Write the header. Since those are curves, input and output channels are same
+    if (!_cmsWriteUInt16Number(io, (cmsUInt16Number) mpe ->InputChannels)) return FALSE;
+    if (!_cmsWriteUInt16Number(io, (cmsUInt16Number) mpe ->InputChannels)) return FALSE;
+    
+    if (!WritePositionTable(self, io, 0, 
+                                mpe ->InputChannels, BaseOffset, Curves, WriteMPECurve)) return FALSE;
+
+
+    return TRUE;
+
+    cmsUNUSED_PARAMETER(nItems);
+}
+
+
+
+// The matrix is organized as an array of PxQ+Q elements, where P is the number of input channels to the
+// matrix, and Q is the number of output channels. The matrix elements are each float32Numbers. The array
+// is organized as follows:
+// array = [e11, e12, …, e1P, e21, e22, …, e2P, …, eQ1, eQ2, …, eQP, e1, e2, …, eQ]
+
+static
+void *Type_MPEmatrix_Read(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsUInt32Number* nItems, cmsUInt32Number SizeOfTag)
+{
+    cmsStage* mpe;
+    cmsUInt16Number   InputChans, OutputChans;
+    cmsUInt32Number   nElems, i;  
+    cmsFloat64Number* Matrix;
+    cmsFloat64Number* Offsets;
+
+    if (!_cmsReadUInt16Number(io, &InputChans)) return NULL;
+    if (!_cmsReadUInt16Number(io, &OutputChans)) return NULL;
+
+    
+    nElems = InputChans * OutputChans;
+
+    // Input and output chans may be ANY (up to 0xffff)
+    Matrix = (cmsFloat64Number*) _cmsCalloc(self ->ContextID, nElems, sizeof(cmsFloat64Number));
+    if (Matrix == NULL) return NULL;
+
+    Offsets = (cmsFloat64Number*) _cmsCalloc(self ->ContextID, OutputChans, sizeof(cmsFloat64Number));
+    if (Offsets == NULL) {
+    
+        _cmsFree(self ->ContextID, Matrix);
+        return NULL;
+    }
+
+    for (i=0; i < nElems; i++) {
+
+        cmsFloat32Number v;
+
+        if (!_cmsReadFloat32Number(io, &v)) return NULL;        
+        Matrix[i] = v;
+    }
+
+
+    for (i=0; i < OutputChans; i++) {
+
+        cmsFloat32Number v;
+
+        if (!_cmsReadFloat32Number(io, &v)) return NULL;        
+        Offsets[i] = v;
+    }
+
+
+    mpe = cmsStageAllocMatrix(self ->ContextID, OutputChans, InputChans, Matrix, Offsets);
+    _cmsFree(self ->ContextID, Matrix);
+    _cmsFree(self ->ContextID, Offsets);
+
+    *nItems = 1;
+
+    return mpe;
+
+    cmsUNUSED_PARAMETER(SizeOfTag);
+}
+
+static
+cmsBool  Type_MPEmatrix_Write(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, void* Ptr, cmsUInt32Number nItems)
+{
+    cmsUInt32Number i, nElems;
+    cmsStage* mpe = (cmsStage*) Ptr;
+    _cmsStageMatrixData* Matrix = (_cmsStageMatrixData*) mpe ->Data;
+
+    if (!_cmsWriteUInt16Number(io, (cmsUInt16Number) mpe ->InputChannels)) return FALSE;    
+    if (!_cmsWriteUInt16Number(io, (cmsUInt16Number) mpe ->OutputChannels)) return FALSE;
+
+    nElems = mpe ->InputChannels * mpe ->OutputChannels;
+
+    for (i=0; i < nElems; i++) {
+        if (!_cmsWriteFloat32Number(io, (cmsFloat32Number) Matrix->Double[i])) return FALSE;        
+    }
+
+
+    for (i=0; i < mpe ->OutputChannels; i++) {
+
+        if (Matrix ->Offset == NULL) {
+
+               if (!_cmsWriteFloat32Number(io, 0)) return FALSE;        
+        }
+        else {
+               if (!_cmsWriteFloat32Number(io, (cmsFloat32Number) Matrix->Offset[i])) return FALSE;        
+        }
+    }
+
+    return TRUE;
+
+    cmsUNUSED_PARAMETER(nItems);
+    cmsUNUSED_PARAMETER(self);
+}
+
+
+
+static
+void *Type_MPEclut_Read(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsUInt32Number* nItems, cmsUInt32Number SizeOfTag)
+{
+    cmsStage* mpe = NULL;
+    cmsUInt16Number InputChans, OutputChans;
+    cmsUInt8Number Dimensions8[16];
+    cmsUInt32Number i, nMaxGrids, GridPoints[MAX_INPUT_DIMENSIONS];
+    _cmsStageCLutData* clut;
+    
+    if (!_cmsReadUInt16Number(io, &InputChans)) return NULL;
+    if (!_cmsReadUInt16Number(io, &OutputChans)) return NULL;
+    
+    if (io ->Read(io, Dimensions8, sizeof(cmsUInt8Number), 16) != 16)
+        goto Error;
+    
+    // Copy MAX_INPUT_DIMENSIONS at most. Expand to cmsUInt32Number
+    nMaxGrids = InputChans > MAX_INPUT_DIMENSIONS ? MAX_INPUT_DIMENSIONS : InputChans;
+    for (i=0; i < nMaxGrids; i++) GridPoints[i] = (cmsUInt32Number) Dimensions8[i];
+
+    // Allocate the true CLUT
+    mpe = cmsStageAllocCLutFloatGranular(self ->ContextID, GridPoints, InputChans, OutputChans, NULL);
+    if (mpe == NULL) goto Error;
+
+    // Read the data
+    clut = (_cmsStageCLutData*) mpe ->Data;
+    for (i=0; i < clut ->nEntries; i++) {
+        
+        if (!_cmsReadFloat32Number(io, &clut ->Tab.TFloat[i])) goto Error;
+    }
+
+    *nItems = 1;    
+    return mpe;
+
+Error:
+    *nItems = 0;
+    if (mpe != NULL) cmsStageFree(mpe);
+    return NULL;
+
+    cmsUNUSED_PARAMETER(SizeOfTag);
+}
+
+// Write a CLUT in floating point
+static
+cmsBool  Type_MPEclut_Write(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, void* Ptr, cmsUInt32Number nItems)
+{   
+    cmsUInt8Number Dimensions8[16];
+    cmsUInt32Number i;
+    cmsStage* mpe = (cmsStage*) Ptr;
+    _cmsStageCLutData* clut = (_cmsStageCLutData*) mpe ->Data;
+
+    // Check for maximum number of channels
+    if (mpe -> InputChannels > 15) return FALSE;
+    
+    // Only floats are supported in MPE
+    if (clut ->HasFloatValues == FALSE) return FALSE;
+
+    if (!_cmsWriteUInt16Number(io, (cmsUInt16Number) mpe ->InputChannels)) return FALSE;
+    if (!_cmsWriteUInt16Number(io, (cmsUInt16Number) mpe ->OutputChannels)) return FALSE;
+
+    memset(Dimensions8, 0, sizeof(Dimensions8));
+
+    for (i=0; i < mpe ->InputChannels; i++) 
+        Dimensions8[i] = (cmsUInt8Number) clut ->Params ->nSamples[i];
+
+    if (!io ->Write(io, 16, Dimensions8)) return FALSE;
+
+    for (i=0; i < clut ->nEntries; i++) {
+        
+        if (!_cmsWriteFloat32Number(io, clut ->Tab.TFloat[i])) return FALSE;
+    }
+
+    return TRUE;
+
+    cmsUNUSED_PARAMETER(nItems);
+    cmsUNUSED_PARAMETER(self);
+}
+
+
+
+// This is the list of built-in MPE types
+static _cmsTagTypeLinkedList SupportedMPEtypes[] = {
+
+{{ (cmsTagTypeSignature) cmsSigBAcsElemType, NULL, NULL, NULL, NULL }, &SupportedMPEtypes[1] },   // Ignore those elements for now
+{{ (cmsTagTypeSignature) cmsSigEAcsElemType, NULL, NULL, NULL, NULL }, &SupportedMPEtypes[2] },   // (That's what the spec says)
+
+{TYPE_MPE_HANDLER((cmsTagTypeSignature) cmsSigCurveSetElemType,     MPEcurve),      &SupportedMPEtypes[3] },
+{TYPE_MPE_HANDLER((cmsTagTypeSignature) cmsSigMatrixElemType,       MPEmatrix),     &SupportedMPEtypes[4] },
+{TYPE_MPE_HANDLER((cmsTagTypeSignature) cmsSigCLutElemType,         MPEclut),        NULL },
+};
+
+#define DEFAULT_MPE_TYPE_COUNT  (sizeof(SupportedMPEtypes) / sizeof(_cmsTagTypeLinkedList))
+
+static
+cmsBool ReadMPEElem(struct _cms_typehandler_struct* self, 
+                    cmsIOHANDLER* io,
+                    void* Cargo,
+                    cmsUInt32Number n, 
+                    cmsUInt32Number SizeOfTag)
+{
+    cmsStageSignature ElementSig;
+    cmsTagTypeHandler* TypeHandler;
+    cmsStage *mpe = NULL;
+    cmsUInt32Number nItems;
+    cmsPipeline *NewLUT = (cmsPipeline *) Cargo;
+
+    // Take signature and channels for each element.
+    if (!_cmsReadUInt32Number(io, (cmsUInt32Number*) &ElementSig)) return FALSE;
+
+    // The reserved placeholder
+    if (!_cmsReadUInt32Number(io, NULL)) return FALSE;
+
+    // Read diverse MPE types
+    TypeHandler = GetHandler((cmsTagTypeSignature) ElementSig, SupportedMPEtypes);
+    if (TypeHandler == NULL)  {
+
+        char String[5];
+
+        _cmsTagSignature2String(String, (cmsTagSignature) ElementSig);
+
+        // An unknown element was found. 
+        cmsSignalError(self ->ContextID, cmsERROR_UNKNOWN_EXTENSION, "Unknown MPE type '%s' found.", String);
+        return FALSE;
+    }
+
+    // If no read method, just ignore the element (valid for cmsSigBAcsElemType and cmsSigEAcsElemType)
+    // Read the MPE. No size is given
+    if (TypeHandler ->ReadPtr != NULL) {
+
+        // This is a real element which should be read and processed
+        mpe = (cmsStage*) TypeHandler ->ReadPtr(self, io, &nItems, SizeOfTag);
+        if (mpe == NULL) return FALSE;
+
+        // All seems ok, insert element
+        cmsPipelineInsertStage(NewLUT, cmsAT_END, mpe);
+    }
+
+    return TRUE;
+
+    cmsUNUSED_PARAMETER(SizeOfTag);
+    cmsUNUSED_PARAMETER(n);
+}
+
+
+// This is the main dispatcher for MPE
+static
+void *Type_MPE_Read(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, cmsUInt32Number* nItems, cmsUInt32Number SizeOfTag)
+{   
+    cmsUInt16Number InputChans, OutputChans;
+    cmsUInt32Number ElementCount;
+    cmsPipeline *NewLUT = NULL;
+    cmsUInt32Number BaseOffset;
+
+    // Get actual position as a basis for element offsets
+    BaseOffset = io ->Tell(io) - sizeof(_cmsTagBase);
+
+    // Read channels and element count
+    if (!_cmsReadUInt16Number(io, &InputChans)) return NULL;
+    if (!_cmsReadUInt16Number(io, &OutputChans)) return NULL; 
+    
+    // Allocates an empty LUT
+    NewLUT = cmsPipelineAlloc(self ->ContextID, InputChans, OutputChans);
+    if (NewLUT == NULL) return NULL;
+
+    if (!_cmsReadUInt32Number(io, &ElementCount)) return NULL;
+
+    if (!ReadPositionTable(self, io, ElementCount, BaseOffset, NewLUT, ReadMPEElem)) {
+        if (NewLUT != NULL) cmsPipelineFree(NewLUT);
+        *nItems = 0;
+        return NULL;
+    }
+
+    // Success
+    *nItems = 1;
+    return NewLUT;
+
+    cmsUNUSED_PARAMETER(SizeOfTag);
+}
+
+
+
+// This one is a liitle bit more complex, so we don't use position tables this time.
+static
+cmsBool Type_MPE_Write(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, void* Ptr, cmsUInt32Number nItems)
+{   
+    cmsUInt32Number i, BaseOffset, DirectoryPos, CurrentPos;
+    int inputChan, outputChan;
+    cmsUInt32Number ElemCount;
+    cmsUInt32Number *ElementOffsets = NULL, *ElementSizes = NULL, Before;
+    cmsStageSignature ElementSig;
+    cmsPipeline* Lut = (cmsPipeline*) Ptr;
+    cmsStage* Elem = Lut ->Elements;
+    cmsTagTypeHandler* TypeHandler;
+
+    BaseOffset = io ->Tell(io) - sizeof(_cmsTagBase);
+
+    inputChan  = cmsPipelineInputChannels(Lut);
+    outputChan = cmsPipelineOutputChannels(Lut);
+    ElemCount  = cmsPipelineStageCount(Lut);
+
+    ElementOffsets = (cmsUInt32Number *) _cmsCalloc(self ->ContextID, ElemCount, sizeof(cmsUInt32Number *));
+    if (ElementOffsets == NULL) goto Error;
+
+    ElementSizes = (cmsUInt32Number *) _cmsCalloc(self ->ContextID, ElemCount, sizeof(cmsUInt32Number *));
+    if (ElementSizes == NULL) goto Error;
+
+    // Write the head
+    if (!_cmsWriteUInt16Number(io, (cmsUInt16Number) inputChan)) goto Error;
+    if (!_cmsWriteUInt16Number(io, (cmsUInt16Number) outputChan)) goto Error;
+    if (!_cmsWriteUInt32Number(io, (cmsUInt16Number) ElemCount)) goto Error;
+
+    DirectoryPos = io ->Tell(io);
+
+    // Write a fake directory to be filled latter on
+    for (i=0; i < ElemCount; i++) {
+        if (!_cmsWriteUInt32Number(io, 0)) goto Error;  // Offset 
+        if (!_cmsWriteUInt32Number(io, 0)) goto Error;  // size
+    }
+
+    // Write each single tag. Keep track of the size as well.
+    for (i=0; i < ElemCount; i++) {
+
+        ElementOffsets[i] = io ->Tell(io) - BaseOffset;
+
+        ElementSig = Elem ->Type;
+
+        TypeHandler = GetHandler((cmsTagTypeSignature) ElementSig, SupportedMPEtypes);
+        if (TypeHandler == NULL)  {
+
+                char String[5];
+
+                _cmsTagSignature2String(String, (cmsTagSignature) ElementSig);
+
+                 // An unknow element was found. 
+                 cmsSignalError(self->ContextID, cmsERROR_UNKNOWN_EXTENSION, "Found unknown MPE type '%s'", String);
+                 goto Error;
+        }
+
+        if (!_cmsWriteUInt32Number(io, ElementSig)) goto Error;  
+        if (!_cmsWriteUInt32Number(io, 0)) goto Error;  
+        Before = io ->Tell(io);
+        if (!TypeHandler ->WritePtr(self, io, Elem, 1)) goto Error;
+        if (!_cmsWriteAlignment(io)) goto Error;
+
+        ElementSizes[i] = io ->Tell(io) - Before;
+
+        Elem = Elem ->Next;
+    }
+
+    // Write the directory
+    CurrentPos = io ->Tell(io);
+
+    if (!io ->Seek(io, DirectoryPos)) goto Error;
+
+    for (i=0; i < ElemCount; i++) {
+        if (!_cmsWriteUInt32Number(io, ElementOffsets[i])) goto Error;  
+        if (!_cmsWriteUInt32Number(io, ElementSizes[i])) goto Error;  
+    }
+
+    if (!io ->Seek(io, CurrentPos)) goto Error;
+
+    if (ElementOffsets != NULL) _cmsFree(self ->ContextID, ElementOffsets);
+    if (ElementSizes != NULL) _cmsFree(self ->ContextID, ElementSizes);
+    return TRUE;
+
+Error:
+    if (ElementOffsets != NULL) _cmsFree(self ->ContextID, ElementOffsets);
+    if (ElementSizes != NULL) _cmsFree(self ->ContextID, ElementSizes);
+    return FALSE;
+
+    cmsUNUSED_PARAMETER(nItems);
+}
+
+
+static
+void* Type_MPE_Dup(struct _cms_typehandler_struct* self, const void *Ptr, cmsUInt32Number n)
+{
+    return (void*) cmsPipelineDup((cmsPipeline*) Ptr);
+
+    cmsUNUSED_PARAMETER(n);
+    cmsUNUSED_PARAMETER(self);
+}
+
+static
+void Type_MPE_Free(struct _cms_typehandler_struct* self, void *Ptr)
+{    
+    cmsPipelineFree((cmsPipeline*) Ptr);
+    return;
+
+    cmsUNUSED_PARAMETER(self);
+}
+
+
+// ********************************************************************************
+// Type cmsSigVcgtType
+// ********************************************************************************
+
+
+#define cmsVideoCardGammaTableType    0
+#define cmsVideoCardGammaFormulaType  1
+
+// Used internally
+typedef struct {
+    double Gamma;
+    double Min;
+    double Max;
+} _cmsVCGTGAMMA;
+
+
+static
+void *Type_vcgt_Read(struct _cms_typehandler_struct* self, 
+                     cmsIOHANDLER* io, 
+                     cmsUInt32Number* nItems, 
+                     cmsUInt32Number SizeOfTag)
+{
+    cmsUInt32Number TagType, n, i;
+    cmsToneCurve** Curves;
+
+    *nItems = 0;
+
+    // Read tag type
+    if (!_cmsReadUInt32Number(io, &TagType)) return NULL;
+ 
+    // Allocate space for the array
+    Curves = ( cmsToneCurve**) _cmsCalloc(self ->ContextID, 3, sizeof(cmsToneCurve*));
+    if (Curves == NULL) return NULL;
+
+    // There are two possible flavors
+    switch (TagType) {
+
+    // Gamma is stored as a table
+    case cmsVideoCardGammaTableType: 
+    {
+       cmsUInt16Number nChannels, nElems, nBytes;
+
+       // Check channel count, which should be 3 (we don't support monochrome this time)
+       if (!_cmsReadUInt16Number(io, &nChannels)) goto Error;
+
+       if (nChannels != 3) {
+           cmsSignalError(self->ContextID, cmsERROR_UNKNOWN_EXTENSION, "Unsupported number of channels for VCGT '%d'", nChannels);
+           goto Error;                               
+       }
+
+       // Get Table element count and bytes per element
+       if (!_cmsReadUInt16Number(io, &nElems)) goto Error;
+       if (!_cmsReadUInt16Number(io, &nBytes)) goto Error;
+       
+	   // Adobe's quirk fixup. Fixing broken profiles...
+	   if (nElems == 256 && nBytes == 1 && SizeOfTag == 1576)
+		   nBytes = 2;
+
+
+       // Populate tone curves
+       for (n=0; n < 3; n++) {
+
+           Curves[n] = cmsBuildTabulatedToneCurve16(self ->ContextID, nElems, NULL);
+           if (Curves[n] == NULL) goto Error;
+
+           // On depending on byte depth
+           switch (nBytes) {
+
+           // One byte, 0..255
+           case 1:
+               for (i=0; i < nElems; i++) {
+
+                   cmsUInt8Number v;
+
+                      if (!_cmsReadUInt8Number(io, &v)) goto Error;
+                      Curves[n] ->Table16[i] = FROM_8_TO_16(v);
+               }
+               break;
+
+           // One word 0..65535
+           case 2:
+              if (!_cmsReadUInt16Array(io, nElems, Curves[n]->Table16)) goto Error;
+              break;
+
+          // Unsupported
+           default:
+              cmsSignalError(self->ContextID, cmsERROR_UNKNOWN_EXTENSION, "Unsupported bit depth for VCGT '%d'", nBytes * 8);
+              goto Error;       
+           }
+       } // For all 3 channels
+    }
+    break;
+
+   // In this case, gamma is stored as a formula
+   case cmsVideoCardGammaFormulaType: 
+   {
+       _cmsVCGTGAMMA Colorant[3];
+       
+        // Populate tone curves
+       for (n=0; n < 3; n++) {
+
+           double Params[10];
+
+           if (!_cmsRead15Fixed16Number(io, &Colorant[n].Gamma)) goto Error;
+           if (!_cmsRead15Fixed16Number(io, &Colorant[n].Min)) goto Error;
+           if (!_cmsRead15Fixed16Number(io, &Colorant[n].Max)) goto Error;
+           
+            // Parametric curve type 5 is:
+            // Y = (aX + b)^Gamma + e | X >= d
+            // Y = cX + f             | X < d
+
+            // vcgt formula is:
+            // Y = (Max – Min) * (X ^ Gamma) + Min
+            
+            // So, the translation is
+            // a = (Max – Min) ^ ( 1 / Gamma) 
+            // e = Min
+            // b=c=d=f=0
+
+           Params[0] = Colorant[n].Gamma;
+           Params[1] = pow((Colorant[n].Max - Colorant[n].Min), (1.0 / Colorant[n].Gamma));
+           Params[2] = 0;
+           Params[3] = 0;
+           Params[4] = 0;
+           Params[5] = Colorant[n].Min;
+           Params[6] = 0;
+    
+           Curves[n] = cmsBuildParametricToneCurve(self ->ContextID, 5, Params);
+           if (Curves[n] == NULL) goto Error;
+       }
+   }
+   break;
+
+   // Unsupported
+   default: 
+      cmsSignalError(self->ContextID, cmsERROR_UNKNOWN_EXTENSION, "Unsupported tag type for VCGT '%d'", TagType);
+      goto Error;
+   }
+
+   *nItems = 1;
+   return (void*) Curves;
+
+// Regret,  free all resources
+Error:
+
+    cmsFreeToneCurveTriple(Curves);
+    _cmsFree(self ->ContextID, Curves);
+    return NULL;
+
+     cmsUNUSED_PARAMETER(SizeOfTag);
+}
+
+
+// We don't support all flavors, only 16bits tables and formula
+static
+cmsBool Type_vcgt_Write(struct _cms_typehandler_struct* self, cmsIOHANDLER* io, void* Ptr, cmsUInt32Number nItems)
+{
+    cmsToneCurve** Curves =  (cmsToneCurve**) Ptr;
+    cmsUInt32Number i, j;
+
+    if (cmsGetToneCurveParametricType(Curves[0]) == 5 &&
+        cmsGetToneCurveParametricType(Curves[1]) == 5 &&
+        cmsGetToneCurveParametricType(Curves[2]) == 5) {
+
+            if (!_cmsWriteUInt32Number(io, cmsVideoCardGammaFormulaType)) return FALSE;  
+
+            // Save parameters
+            for (i=0; i < 3; i++) {
+
+                _cmsVCGTGAMMA v;
+
+                v.Gamma = Curves[i] ->Segments[0].Params[0];
+                v.Min   = Curves[i] ->Segments[0].Params[5];
+                v.Max   = pow(Curves[i] ->Segments[0].Params[1], v.Gamma) + v.Min;
+
+                if (!_cmsWrite15Fixed16Number(io, v.Gamma)) return FALSE;
+                if (!_cmsWrite15Fixed16Number(io, v.Min)) return FALSE;
+                if (!_cmsWrite15Fixed16Number(io, v.Max)) return FALSE;
+            }
+    }
+
+    else {
+
+        // Always store as a table of 256 words
+        if (!_cmsWriteUInt32Number(io, cmsVideoCardGammaTableType)) return FALSE;    
+        if (!_cmsWriteUInt16Number(io, 3)) return FALSE;    
+        if (!_cmsWriteUInt16Number(io, 256)) return FALSE;    
+        if (!_cmsWriteUInt16Number(io, 2)) return FALSE;    
+
+        for (i=0; i < 3; i++) {
+            for (j=0; j < 256; j++) {
+
+                cmsFloat32Number v = cmsEvalToneCurveFloat(Curves[i], (cmsFloat32Number) (j / 255.0));
+                cmsUInt16Number  n = _cmsQuickSaturateWord(v * 65535.0);
+
+                if (!_cmsWriteUInt16Number(io, n)) return FALSE;    
+            }
+        }
+    }
+
+    return TRUE;
+
+    cmsUNUSED_PARAMETER(self);
+    cmsUNUSED_PARAMETER(nItems);
+}
+
+static
+void* Type_vcgt_Dup(struct _cms_typehandler_struct* self, const void *Ptr, cmsUInt32Number n)
+{
+    cmsToneCurve** OldCurves =  (cmsToneCurve**) Ptr;
+    cmsToneCurve** NewCurves;
+
+    NewCurves = ( cmsToneCurve**) _cmsCalloc(self ->ContextID, 3, sizeof(cmsToneCurve*));
+    if (NewCurves == NULL) return NULL;
+
+    NewCurves[0] = cmsDupToneCurve(OldCurves[0]);
+    NewCurves[1] = cmsDupToneCurve(OldCurves[1]);
+    NewCurves[2] = cmsDupToneCurve(OldCurves[2]);
+
+    return (void*) NewCurves;
+
+    cmsUNUSED_PARAMETER(n);
+}
+
+
+static
+void Type_vcgt_Free(struct _cms_typehandler_struct* self, void* Ptr)
+{
+    cmsFreeToneCurveTriple((cmsToneCurve**) Ptr);
+    _cmsFree(self ->ContextID, Ptr);
+}
+
+// ********************************************************************************
+// Type support main routines
+// ********************************************************************************
+
+
+// This is the list of built-in types
+static _cmsTagTypeLinkedList SupportedTagTypes[] = {
+
+{TYPE_HANDLER(cmsSigChromaticityType,          Chromaticity),        &SupportedTagTypes[1] },
+{TYPE_HANDLER(cmsSigColorantOrderType,         ColorantOrderType),   &SupportedTagTypes[2] },
+{TYPE_HANDLER(cmsSigS15Fixed16ArrayType,       S15Fixed16),          &SupportedTagTypes[3] },
+{TYPE_HANDLER(cmsSigU16Fixed16ArrayType,       U16Fixed16),          &SupportedTagTypes[4] },
+{TYPE_HANDLER(cmsSigTextType,                  Text),                &SupportedTagTypes[5] },
+{TYPE_HANDLER(cmsSigTextDescriptionType,       Text_Description),    &SupportedTagTypes[6] },
+{TYPE_HANDLER(cmsSigCurveType,                 Curve),               &SupportedTagTypes[7] },
+{TYPE_HANDLER(cmsSigParametricCurveType,       ParametricCurve),     &SupportedTagTypes[8] },
+{TYPE_HANDLER(cmsSigDateTimeType,              DateTime),            &SupportedTagTypes[9] },
+{TYPE_HANDLER(cmsSigLut8Type,                  LUT8),                &SupportedTagTypes[10] },
+{TYPE_HANDLER(cmsSigLut16Type,                 LUT16),               &SupportedTagTypes[11] },
+{TYPE_HANDLER(cmsSigColorantTableType,         ColorantTable),       &SupportedTagTypes[12] },
+{TYPE_HANDLER(cmsSigNamedColor2Type,           NamedColor),          &SupportedTagTypes[13] },
+{TYPE_HANDLER(cmsSigMultiLocalizedUnicodeType, MLU),                 &SupportedTagTypes[14] },
+{TYPE_HANDLER(cmsSigProfileSequenceDescType,   ProfileSequenceDesc), &SupportedTagTypes[15] },
+{TYPE_HANDLER(cmsSigSignatureType,             Signature),           &SupportedTagTypes[16] },
+{TYPE_HANDLER(cmsSigMeasurementType,           Measurement),         &SupportedTagTypes[17] },
+{TYPE_HANDLER(cmsSigDataType,                  Data),                &SupportedTagTypes[18] },
+{TYPE_HANDLER(cmsSigLutAtoBType,               LUTA2B),              &SupportedTagTypes[19] },
+{TYPE_HANDLER(cmsSigLutBtoAType,               LUTB2A),              &SupportedTagTypes[20] },
+{TYPE_HANDLER(cmsSigUcrBgType,                 UcrBg),               &SupportedTagTypes[21] },
+{TYPE_HANDLER(cmsSigCrdInfoType,               CrdInfo),             &SupportedTagTypes[22] },
+{TYPE_HANDLER(cmsSigMultiProcessElementType,   MPE),                 &SupportedTagTypes[23] },
+{TYPE_HANDLER(cmsSigScreeningType,             Screening),           &SupportedTagTypes[24] },
+{TYPE_HANDLER(cmsSigViewingConditionsType,     ViewingConditions),   &SupportedTagTypes[25] },
+{TYPE_HANDLER(cmsSigXYZType,                   XYZ),                 &SupportedTagTypes[26] },
+{TYPE_HANDLER(cmsCorbisBrokenXYZtype,          XYZ),                 &SupportedTagTypes[27] },
+{TYPE_HANDLER(cmsMonacoBrokenCurveType,        Curve),               &SupportedTagTypes[28] },
+{TYPE_HANDLER(cmsSigProfileSequenceIdType,     ProfileSequenceId),   &SupportedTagTypes[29] },
+{TYPE_HANDLER(cmsSigVcgtType,                  vcgt),                NULL }
+};
+
+#define DEFAULT_TAG_TYPE_COUNT  (sizeof(SupportedTagTypes) / sizeof(_cmsTagTypeLinkedList))
+                                
+// Both kind of plug-ins share same structure
+cmsBool  _cmsRegisterTagTypePlugin(cmsPluginBase* Data)
+{
+    return RegisterTypesPlugin(Data, SupportedTagTypes, DEFAULT_TAG_TYPE_COUNT);
+}
+
+cmsBool  _cmsRegisterMultiProcessElementPlugin(cmsPluginBase* Data)
+{
+    return RegisterTypesPlugin(Data, SupportedMPEtypes, DEFAULT_MPE_TYPE_COUNT);
+}
+
+
+// Wrapper for tag types
+cmsTagTypeHandler* _cmsGetTagTypeHandler(cmsTagTypeSignature sig)
+{
+    return GetHandler(sig, SupportedTagTypes);
+}
+    
+// ********************************************************************************
+// Tag support main routines
+// ********************************************************************************
+
+typedef struct _cmsTagLinkedList_st {
+
+            cmsTagSignature Signature;
+            cmsTagDescriptor Descriptor;
+            struct _cmsTagLinkedList_st* Next;
+
+} _cmsTagLinkedList;
+
+// This is the list of built-in tags
+static _cmsTagLinkedList SupportedTags[] = {
+
+    { cmsSigAToB0Tag,               { 1, 3,  { cmsSigLut16Type,  cmsSigLutAtoBType, cmsSigLut8Type}, DecideLUTtypeA2B}, &SupportedTags[1]},   
+    { cmsSigAToB1Tag,               { 1, 3,  { cmsSigLut16Type,  cmsSigLutAtoBType, cmsSigLut8Type}, DecideLUTtypeA2B}, &SupportedTags[2]},
+    { cmsSigAToB2Tag,               { 1, 3,  { cmsSigLut16Type,  cmsSigLutAtoBType, cmsSigLut8Type}, DecideLUTtypeA2B}, &SupportedTags[3]},
+    { cmsSigBToA0Tag,               { 1, 3,  { cmsSigLut16Type,  cmsSigLutBtoAType, cmsSigLut8Type}, DecideLUTtypeB2A}, &SupportedTags[4]},
+    { cmsSigBToA1Tag,               { 1, 3,  { cmsSigLut16Type,  cmsSigLutBtoAType, cmsSigLut8Type}, DecideLUTtypeB2A}, &SupportedTags[5]},
+    { cmsSigBToA2Tag,               { 1, 3,  { cmsSigLut16Type,  cmsSigLutBtoAType, cmsSigLut8Type}, DecideLUTtypeB2A}, &SupportedTags[6]},
+     
+    // Allow corbis  and its broken XYZ type
+    { cmsSigRedColorantTag,         { 1, 2, { cmsSigXYZType, cmsCorbisBrokenXYZtype }, DecideXYZtype}, &SupportedTags[7]},
+    { cmsSigGreenColorantTag,       { 1, 2, { cmsSigXYZType, cmsCorbisBrokenXYZtype }, DecideXYZtype}, &SupportedTags[8]},
+    { cmsSigBlueColorantTag,        { 1, 2, { cmsSigXYZType, cmsCorbisBrokenXYZtype }, DecideXYZtype}, &SupportedTags[9]},
+                                   
+    { cmsSigRedTRCTag,              { 1, 3, { cmsSigCurveType, cmsSigParametricCurveType, cmsMonacoBrokenCurveType }, DecideCurveType}, &SupportedTags[10]},
+    { cmsSigGreenTRCTag,            { 1, 3, { cmsSigCurveType, cmsSigParametricCurveType, cmsMonacoBrokenCurveType }, DecideCurveType}, &SupportedTags[11]},
+    { cmsSigBlueTRCTag,             { 1, 3, { cmsSigCurveType, cmsSigParametricCurveType, cmsMonacoBrokenCurveType }, DecideCurveType}, &SupportedTags[12]}, 
+
+    { cmsSigCalibrationDateTimeTag, { 1, 1, { cmsSigDateTimeType }, NULL}, &SupportedTags[13]},
+    { cmsSigCharTargetTag,          { 1, 1, { cmsSigTextType },     NULL}, &SupportedTags[14]},
+
+    { cmsSigChromaticAdaptationTag, { 9, 1, { cmsSigS15Fixed16ArrayType }, NULL}, &SupportedTags[15]},
+    { cmsSigChromaticityTag,        { 1, 1, { cmsSigChromaticityType    }, NULL}, &SupportedTags[16]},
+    { cmsSigColorantOrderTag,       { 1, 1, { cmsSigColorantOrderType   }, NULL}, &SupportedTags[17]},
+    { cmsSigColorantTableTag,       { 1, 1, { cmsSigColorantTableType   }, NULL}, &SupportedTags[18]},
+    { cmsSigColorantTableOutTag,    { 1, 1, { cmsSigColorantTableType   }, NULL}, &SupportedTags[19]},
+
+    { cmsSigCopyrightTag,           { 1, 3, { cmsSigTextType,  cmsSigMultiLocalizedUnicodeType, cmsSigTextDescriptionType}, DecideTextType}, &SupportedTags[20]},
+    { cmsSigDateTimeTag,            { 1, 1, { cmsSigDateTimeType }, NULL}, &SupportedTags[21]},
+
+    { cmsSigDeviceMfgDescTag,       { 1, 3, { cmsSigTextDescriptionType, cmsSigMultiLocalizedUnicodeType, cmsSigTextType}, DecideTextDescType}, &SupportedTags[22]},
+    { cmsSigDeviceModelDescTag,     { 1, 3, { cmsSigTextDescriptionType, cmsSigMultiLocalizedUnicodeType, cmsSigTextType}, DecideTextDescType}, &SupportedTags[23]},
+
+    { cmsSigGamutTag,               { 1, 3, { cmsSigLut16Type, cmsSigLutBtoAType, cmsSigLut8Type }, DecideLUTtypeB2A}, &SupportedTags[24]},
+
+    { cmsSigGrayTRCTag,             { 1, 2, { cmsSigCurveType, cmsSigParametricCurveType }, DecideCurveType}, &SupportedTags[25]},
+    { cmsSigLuminanceTag,           { 1, 1, { cmsSigXYZType }, NULL}, &SupportedTags[26]},
+
+    { cmsSigMediaBlackPointTag,     { 1, 2, { cmsSigXYZType, cmsCorbisBrokenXYZtype }, NULL}, &SupportedTags[27]},
+    { cmsSigMediaWhitePointTag,     { 1, 2, { cmsSigXYZType, cmsCorbisBrokenXYZtype }, NULL}, &SupportedTags[28]},
+
+    { cmsSigNamedColor2Tag,         { 1, 1, { cmsSigNamedColor2Type }, NULL}, &SupportedTags[29]},
+
+    { cmsSigPreview0Tag,            { 1, 3,  { cmsSigLut16Type, cmsSigLutBtoAType, cmsSigLut8Type }, DecideLUTtypeB2A}, &SupportedTags[30]},
+    { cmsSigPreview1Tag,            { 1, 3,  { cmsSigLut16Type, cmsSigLutBtoAType, cmsSigLut8Type }, DecideLUTtypeB2A}, &SupportedTags[31]},
+    { cmsSigPreview2Tag,            { 1, 3,  { cmsSigLut16Type, cmsSigLutBtoAType, cmsSigLut8Type }, DecideLUTtypeB2A}, &SupportedTags[32]},
+
+    { cmsSigProfileDescriptionTag,  { 1, 3, { cmsSigTextDescriptionType, cmsSigMultiLocalizedUnicodeType, cmsSigTextType}, DecideTextDescType}, &SupportedTags[33]},
+    { cmsSigProfileSequenceDescTag, { 1, 1, { cmsSigProfileSequenceDescType }, NULL}, &SupportedTags[34]},
+    { cmsSigTechnologyTag,          { 1, 1, { cmsSigSignatureType }, NULL},  &SupportedTags[35]},
+
+    { cmsSigColorimetricIntentImageStateTag,   { 1, 1, { cmsSigSignatureType }, NULL}, &SupportedTags[36]},
+    { cmsSigPerceptualRenderingIntentGamutTag, { 1, 1, { cmsSigSignatureType }, NULL}, &SupportedTags[37]},
+    { cmsSigSaturationRenderingIntentGamutTag, { 1, 1, { cmsSigSignatureType }, NULL}, &SupportedTags[38]},
+
+    { cmsSigMeasurementTag,         { 1, 1, { cmsSigMeasurementType }, NULL}, &SupportedTags[39]},
+
+    { cmsSigPs2CRD0Tag,             { 1, 1, { cmsSigDataType }, NULL}, &SupportedTags[40]},          
+    { cmsSigPs2CRD1Tag,             { 1, 1, { cmsSigDataType }, NULL}, &SupportedTags[41]},                     
+    { cmsSigPs2CRD2Tag,             { 1, 1, { cmsSigDataType }, NULL}, &SupportedTags[42]},                    
+    { cmsSigPs2CRD3Tag,             { 1, 1, { cmsSigDataType }, NULL}, &SupportedTags[43]},                    
+    { cmsSigPs2CSATag,              { 1, 1, { cmsSigDataType }, NULL}, &SupportedTags[44]},                    
+    { cmsSigPs2RenderingIntentTag,  { 1, 1, { cmsSigDataType }, NULL}, &SupportedTags[45]},     
+
+    { cmsSigViewingCondDescTag,     { 1, 3, { cmsSigTextDescriptionType, cmsSigMultiLocalizedUnicodeType, cmsSigTextType}, DecideTextDescType}, &SupportedTags[46]},
+
+    { cmsSigUcrBgTag,               { 1, 1, { cmsSigUcrBgType}, NULL},    &SupportedTags[47]}, 
+    { cmsSigCrdInfoTag,             { 1, 1, { cmsSigCrdInfoType}, NULL},  &SupportedTags[48]}, 
+
+    { cmsSigDToB0Tag,               { 1, 1, { cmsSigMultiProcessElementType}, NULL}, &SupportedTags[49]}, 
+    { cmsSigDToB1Tag,               { 1, 1, { cmsSigMultiProcessElementType}, NULL}, &SupportedTags[50]}, 
+    { cmsSigDToB2Tag,               { 1, 1, { cmsSigMultiProcessElementType}, NULL}, &SupportedTags[51]}, 
+    { cmsSigDToB3Tag,               { 1, 1, { cmsSigMultiProcessElementType}, NULL}, &SupportedTags[52]}, 
+    { cmsSigBToD0Tag,               { 1, 1, { cmsSigMultiProcessElementType}, NULL}, &SupportedTags[53]}, 
+    { cmsSigBToD1Tag,               { 1, 1, { cmsSigMultiProcessElementType}, NULL}, &SupportedTags[54]}, 
+    { cmsSigBToD2Tag,               { 1, 1, { cmsSigMultiProcessElementType}, NULL}, &SupportedTags[55]}, 
+    { cmsSigBToD3Tag,               { 1, 1, { cmsSigMultiProcessElementType}, NULL}, &SupportedTags[56]}, 
+        
+    { cmsSigScreeningDescTag,       { 1, 1, { cmsSigTextDescriptionType },    NULL}, &SupportedTags[57]}, 
+    { cmsSigViewingConditionsTag,   { 1, 1, { cmsSigViewingConditionsType },  NULL}, &SupportedTags[58]}, 
+
+    { cmsSigScreeningTag,           { 1, 1, { cmsSigScreeningType},          NULL }, &SupportedTags[59]},  
+    { cmsSigVcgtTag,                { 1, 1, { cmsSigVcgtType},               NULL }, &SupportedTags[60]}, 
+    { cmsSigProfileSequenceIdTag,   { 1, 1, { cmsSigProfileSequenceIdType},  NULL},  NULL}  
+
+};
+
+/*
+    Not supported                 Why
+    =======================       =========================================                
+    cmsSigOutputResponseTag   ==> WARNING, POSSIBLE PATENT ON THIS SUBJECT!                
+    cmsSigNamedColorTag       ==> Deprecated                                 
+    cmsSigDataTag             ==> Ancient, unused             
+    cmsSigDeviceSettingsTag   ==> Deprecated, useless     
+*/               
+
+#define DEFAULT_TAG_COUNT  (sizeof(SupportedTags) / sizeof(_cmsTagLinkedList))
+
+cmsBool  _cmsRegisterTagPlugin(cmsPluginBase* Data)
+{
+    cmsPluginTag* Plugin = (cmsPluginTag*) Data;
+    _cmsTagLinkedList *pt, *Anterior;
+
+
+    if (Data == NULL) {
+    
+        SupportedTags[DEFAULT_TAG_COUNT-1].Next = NULL;
+        return TRUE;
+    }
+
+    pt = Anterior = SupportedTags; 
+    while (pt != NULL) {
+
+        if (Plugin->Signature == pt -> Signature) {
+            pt ->Descriptor = Plugin ->Descriptor;  // Replace old behaviour
+            return TRUE;
+        }   
+
+        Anterior = pt;          
+        pt = pt ->Next;
+    }
+
+    pt = (_cmsTagLinkedList*) _cmsPluginMalloc(sizeof(_cmsTagLinkedList));
+    if (pt == NULL) return FALSE;
+
+    pt ->Signature  = Plugin ->Signature;
+    pt ->Descriptor = Plugin ->Descriptor;  
+    pt ->Next       = NULL;
+
+    if (Anterior != NULL) Anterior -> Next = pt;
+
+    return TRUE;
+}
+
+// Return a descriptor for a given tag or NULL
+cmsTagDescriptor* _cmsGetTagDescriptor(cmsTagSignature sig)
+{
+    _cmsTagLinkedList* pt;
+
+    for (pt = SupportedTags; 
+            pt != NULL;
+            pt = pt ->Next) {
+
+                if (sig == pt -> Signature) return &pt ->Descriptor;
+    }
+
+    return NULL;
+}
+
diff --git a/thirdparty/liblcms2/src/cmsvirt.c b/thirdparty/liblcms2/src/cmsvirt.c
new file mode 100644
index 00000000..807bd223
--- /dev/null
+++ b/thirdparty/liblcms2/src/cmsvirt.c
@@ -0,0 +1,1148 @@
+//---------------------------------------------------------------------------------
+//
+//  Little Color Management System
+//  Copyright (c) 1998-2010 Marti Maria Saguer
+//
+// Permission is hereby granted, free of charge, to any person obtaining 
+// a copy of this software and associated documentation files (the "Software"), 
+// to deal in the Software without restriction, including without limitation 
+// the rights to use, copy, modify, merge, publish, distribute, sublicense, 
+// and/or sell copies of the Software, and to permit persons to whom the Software 
+// is furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in 
+// all copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 
+// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO 
+// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE 
+// LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 
+// OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 
+// WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+//
+//---------------------------------------------------------------------------------
+//
+
+#include "lcms2_internal.h"
+
+// Virtual (built-in) profiles
+// -----------------------------------------------------------------------------------
+
+static
+cmsBool SetTextTags(cmsHPROFILE hProfile, const wchar_t* Description)
+{
+    cmsMLU *DescriptionMLU, *CopyrightMLU;
+    cmsBool  rc = FALSE;
+    cmsContext ContextID = cmsGetProfileContextID(hProfile);
+    
+    DescriptionMLU  = cmsMLUalloc(ContextID, 1);
+    CopyrightMLU    = cmsMLUalloc(ContextID, 1);
+
+    if (DescriptionMLU == NULL || CopyrightMLU == NULL) goto Error;
+
+    if (!cmsMLUsetWide(DescriptionMLU,  "en", "US", Description)) goto Error;
+    if (!cmsMLUsetWide(CopyrightMLU,    "en", "US", L"No copyright, use freely")) goto Error;
+
+    if (!cmsWriteTag(hProfile, cmsSigProfileDescriptionTag,  DescriptionMLU)) goto Error;
+    if (!cmsWriteTag(hProfile, cmsSigCopyrightTag,           CopyrightMLU)) goto Error;     
+        
+    rc = TRUE;
+
+Error:
+
+    if (DescriptionMLU)
+        cmsMLUfree(DescriptionMLU);
+    if (CopyrightMLU)
+        cmsMLUfree(CopyrightMLU);
+    return rc;
+}
+    
+
+static
+cmsBool  SetSeqDescTag(cmsHPROFILE hProfile, const char* Model)
+{
+    cmsBool  rc = FALSE;
+    cmsContext ContextID = cmsGetProfileContextID(hProfile);
+    cmsSEQ* Seq = cmsAllocProfileSequenceDescription(ContextID, 1);
+
+    if (Seq == NULL) return FALSE;
+
+    Seq->seq[0].deviceMfg = (cmsSignature) 0;
+    Seq->seq[0].deviceModel = (cmsSignature) 0;
+
+#ifdef CMS_DONT_USE_INT64
+    Seq->seq[0].attributes[0] = 0;  
+    Seq->seq[0].attributes[1] = 0;  
+#else
+    Seq->seq[0].attributes = 0; 
+#endif
+
+    Seq->seq[0].technology = (cmsTechnologySignature) 0;
+
+    cmsMLUsetASCII( Seq->seq[0].Manufacturer, cmsNoLanguage, cmsNoCountry, "Little CMS");
+    cmsMLUsetASCII( Seq->seq[0].Model,        cmsNoLanguage, cmsNoCountry, Model);
+
+    if (!_cmsWriteProfileSequence(hProfile, Seq)) goto Error;
+    
+    rc = TRUE;
+
+Error:
+    if (Seq) 
+        cmsFreeProfileSequenceDescription(Seq);
+
+    return rc;
+}
+
+
+
+// This function creates a profile based on White point, primaries and
+// transfer functions.
+cmsHPROFILE CMSEXPORT cmsCreateRGBProfileTHR(cmsContext ContextID,
+                                          const cmsCIExyY* WhitePoint,
+                                          const cmsCIExyYTRIPLE* Primaries,
+                                          cmsToneCurve* const TransferFunction[3])
+{
+    cmsHPROFILE hICC;   
+    cmsMAT3 MColorants;
+    cmsCIEXYZTRIPLE Colorants;
+    cmsCIExyY MaxWhite;
+    cmsMAT3 CHAD;
+    cmsCIEXYZ WhitePointXYZ;
+
+    hICC = cmsCreateProfilePlaceholder(ContextID);
+    if (!hICC)                          // can't allocate
+        return NULL;
+
+    cmsSetProfileVersion(hICC, 4.2);
+
+    cmsSetDeviceClass(hICC,      cmsSigDisplayClass);
+    cmsSetColorSpace(hICC,       cmsSigRgbData);
+    cmsSetPCS(hICC,              cmsSigXYZData);
+
+    cmsSetHeaderRenderingIntent(hICC,  INTENT_PERCEPTUAL); 
+
+
+    // Implement profile using following tags:
+    //
+    //  1 cmsSigProfileDescriptionTag
+    //  2 cmsSigMediaWhitePointTag
+    //  3 cmsSigRedColorantTag
+    //  4 cmsSigGreenColorantTag
+    //  5 cmsSigBlueColorantTag
+    //  6 cmsSigRedTRCTag
+    //  7 cmsSigGreenTRCTag
+    //  8 cmsSigBlueTRCTag
+    //  9 Chromatic adaptation Tag
+    // This conforms a standard RGB DisplayProfile as says ICC, and then I add (As per addendum II)
+    // 10 cmsSigChromaticityTag
+
+    
+    if (!SetTextTags(hICC, L"RGB built-in")) goto Error;
+
+    if (WhitePoint) {
+
+        if (!cmsWriteTag(hICC, cmsSigMediaWhitePointTag, cmsD50_XYZ())) goto Error;
+
+        cmsxyY2XYZ(&WhitePointXYZ, WhitePoint);
+        _cmsAdaptationMatrix(&CHAD, NULL, &WhitePointXYZ, cmsD50_XYZ());     
+
+        // This is a V4 tag, but many CMM does read and understand it no matter which version       
+        if (!cmsWriteTag(hICC, cmsSigChromaticAdaptationTag, (void*) &CHAD)) goto Error;
+    }
+
+    if (WhitePoint && Primaries) {
+
+        MaxWhite.x =  WhitePoint -> x;
+        MaxWhite.y =  WhitePoint -> y;
+        MaxWhite.Y =  1.0;
+
+        if (!_cmsBuildRGB2XYZtransferMatrix(&MColorants, &MaxWhite, Primaries)) goto Error;         
+        
+        Colorants.Red.X   = MColorants.v[0].n[0];
+        Colorants.Red.Y   = MColorants.v[1].n[0];
+        Colorants.Red.Z   = MColorants.v[2].n[0];
+
+        Colorants.Green.X = MColorants.v[0].n[1];
+        Colorants.Green.Y = MColorants.v[1].n[1];
+        Colorants.Green.Z = MColorants.v[2].n[1];
+
+        Colorants.Blue.X  = MColorants.v[0].n[2];
+        Colorants.Blue.Y  = MColorants.v[1].n[2];
+        Colorants.Blue.Z  = MColorants.v[2].n[2];
+
+        if (!cmsWriteTag(hICC, cmsSigRedColorantTag,   (void*) &Colorants.Red)) goto Error;
+        if (!cmsWriteTag(hICC, cmsSigBlueColorantTag,  (void*) &Colorants.Blue)) goto Error;
+        if (!cmsWriteTag(hICC, cmsSigGreenColorantTag, (void*) &Colorants.Green)) goto Error;
+    }
+
+
+    if (TransferFunction) {
+    
+        if (!cmsWriteTag(hICC, cmsSigRedTRCTag,   (void*) TransferFunction[0])) goto Error;
+        if (!cmsWriteTag(hICC, cmsSigGreenTRCTag, (void*) TransferFunction[1])) goto Error;
+        if (!cmsWriteTag(hICC, cmsSigBlueTRCTag,  (void*) TransferFunction[2])) goto Error;
+    }
+
+    if (Primaries) {
+        if (!cmsWriteTag(hICC, cmsSigChromaticityTag, (void*) Primaries)) goto Error;
+    }
+
+
+    return hICC;
+
+Error:
+    if (hICC)
+        cmsCloseProfile(hICC);
+    return NULL;
+}
+
+cmsHPROFILE CMSEXPORT cmsCreateRGBProfile(const cmsCIExyY* WhitePoint,
+                                          const cmsCIExyYTRIPLE* Primaries,
+                                          cmsToneCurve* const TransferFunction[3])
+{
+    return cmsCreateRGBProfileTHR(NULL, WhitePoint, Primaries, TransferFunction);
+}
+
+
+
+// This function creates a profile based on White point and transfer function.
+cmsHPROFILE CMSEXPORT cmsCreateGrayProfileTHR(cmsContext ContextID,
+                                           const cmsCIExyY* WhitePoint,
+                                           const cmsToneCurve* TransferFunction)
+{
+    cmsHPROFILE hICC;
+    cmsCIEXYZ tmp;              
+
+    hICC = cmsCreateProfilePlaceholder(ContextID);
+    if (!hICC)                          // can't allocate
+        return NULL;
+
+    cmsSetProfileVersion(hICC, 4.2);
+
+    cmsSetDeviceClass(hICC,      cmsSigDisplayClass);
+    cmsSetColorSpace(hICC,       cmsSigGrayData);
+    cmsSetPCS(hICC,              cmsSigXYZData);
+    cmsSetHeaderRenderingIntent(hICC,  INTENT_PERCEPTUAL); 
+
+
+    // Implement profile using following tags:
+    //
+    //  1 cmsSigProfileDescriptionTag
+    //  2 cmsSigMediaWhitePointTag
+    //  3 cmsSigGrayTRCTag
+
+    // This conforms a standard Gray DisplayProfile 
+
+    // Fill-in the tags
+
+    if (!SetTextTags(hICC, L"gray built-in")) goto Error;
+
+
+    if (WhitePoint) {
+
+        cmsxyY2XYZ(&tmp, WhitePoint);
+        if (!cmsWriteTag(hICC, cmsSigMediaWhitePointTag, (void*) &tmp)) goto Error;
+    }
+
+    if (TransferFunction) {
+
+        if (!cmsWriteTag(hICC, cmsSigGrayTRCTag, (void*) TransferFunction)) goto Error;
+    }
+
+    return hICC;
+
+Error:
+    if (hICC)
+        cmsCloseProfile(hICC);
+    return NULL;
+}
+
+
+
+cmsHPROFILE CMSEXPORT cmsCreateGrayProfile(const cmsCIExyY* WhitePoint,
+                                                    const cmsToneCurve* TransferFunction)
+{
+    return cmsCreateGrayProfileTHR(NULL, WhitePoint, TransferFunction);
+}
+
+// This is a devicelink operating in the target colorspace with as many transfer functions as components
+
+cmsHPROFILE CMSEXPORT cmsCreateLinearizationDeviceLinkTHR(cmsContext ContextID,
+                                                          cmsColorSpaceSignature ColorSpace,
+                                                          cmsToneCurve* const TransferFunctions[])
+{
+    cmsHPROFILE hICC;
+    cmsPipeline* Pipeline;
+    cmsStage* Lin;
+    int nChannels;
+
+    hICC = cmsCreateProfilePlaceholder(ContextID);
+    if (!hICC)                          
+        return NULL;
+
+    cmsSetProfileVersion(hICC, 4.2);
+
+    cmsSetDeviceClass(hICC,      cmsSigLinkClass);
+    cmsSetColorSpace(hICC,       ColorSpace);
+    cmsSetPCS(hICC,              ColorSpace);
+
+    cmsSetHeaderRenderingIntent(hICC,  INTENT_PERCEPTUAL); 
+
+    // Set up channels
+    nChannels = cmsChannelsOf(ColorSpace);
+
+    // Creates a Pipeline with prelinearization step only
+    Pipeline = cmsPipelineAlloc(ContextID, nChannels, nChannels);
+    if (Pipeline == NULL) goto Error;
+
+
+    // Copy tables to Pipeline
+    Lin = cmsStageAllocToneCurves(ContextID, nChannels, TransferFunctions);
+    if (Lin == NULL) goto Error;
+
+    cmsPipelineInsertStage(Pipeline, cmsAT_BEGIN, Lin);
+
+    // Create tags       
+    if (!SetTextTags(hICC, L"Linearization built-in")) goto Error;    
+    if (!cmsWriteTag(hICC, cmsSigAToB0Tag, (void*) Pipeline)) goto Error;       
+    if (!SetSeqDescTag(hICC, "Linearization built-in")) goto Error;
+
+    // Pipeline is already on virtual profile
+    cmsPipelineFree(Pipeline);
+
+    // Ok, done
+    return hICC;
+
+Error:
+    if (hICC)
+        cmsCloseProfile(hICC);
+
+
+    return NULL;
+}
+
+cmsHPROFILE CMSEXPORT cmsCreateLinearizationDeviceLink(cmsColorSpaceSignature ColorSpace,
+                                                                 cmsToneCurve* const TransferFunctions[])
+{
+    return cmsCreateLinearizationDeviceLinkTHR(NULL, ColorSpace, TransferFunctions);
+}
+
+// Ink-limiting algorithm
+//
+//  Sum = C + M + Y + K 
+//  If Sum > InkLimit 
+//        Ratio= 1 - (Sum - InkLimit) / (C + M + Y)
+//        if Ratio <0 
+//              Ratio=0
+//        endif     
+//     Else 
+//         Ratio=1
+//     endif
+//
+//     C = Ratio * C
+//     M = Ratio * M
+//     Y = Ratio * Y
+//     K: Does not change
+
+static
+int InkLimitingSampler(register const cmsUInt16Number In[], register cmsUInt16Number Out[], register void* Cargo)
+{
+    cmsFloat64Number InkLimit = *(cmsFloat64Number *) Cargo;
+    cmsFloat64Number SumCMY, SumCMYK, Ratio;
+
+    InkLimit = (InkLimit * 655.35);
+
+    SumCMY   = In[0]  + In[1] + In[2];
+    SumCMYK  = SumCMY + In[3];      
+
+    if (SumCMYK > InkLimit) {
+
+        Ratio = 1 - ((SumCMYK - InkLimit) / SumCMY);
+        if (Ratio < 0)
+            Ratio = 0;
+    }
+    else Ratio = 1;
+
+    Out[0] = _cmsQuickSaturateWord(In[0] * Ratio);     // C
+    Out[1] = _cmsQuickSaturateWord(In[1] * Ratio);     // M
+    Out[2] = _cmsQuickSaturateWord(In[2] * Ratio);     // Y
+
+    Out[3] = In[3];                                 // K (untouched)
+
+    return TRUE;
+}
+
+// This is a devicelink operating in CMYK for ink-limiting
+
+cmsHPROFILE CMSEXPORT cmsCreateInkLimitingDeviceLinkTHR(cmsContext ContextID, 
+                                                     cmsColorSpaceSignature ColorSpace,
+                                                     cmsFloat64Number Limit)
+{
+    cmsHPROFILE hICC;
+    cmsPipeline* LUT;
+    cmsStage* CLUT;
+    int nChannels;
+    
+    if (ColorSpace != cmsSigCmykData) {
+        cmsSignalError(ContextID, cmsERROR_COLORSPACE_CHECK, "InkLimiting: Only CMYK currently supported");
+        return NULL;
+    }
+
+    if (Limit < 0.0 || Limit > 400) {
+
+        cmsSignalError(ContextID, cmsERROR_RANGE, "InkLimiting: Limit should be between 0..400");        
+        if (Limit < 0) Limit = 0;
+        if (Limit > 400) Limit = 400;
+
+    }
+
+    hICC = cmsCreateProfilePlaceholder(ContextID);
+    if (!hICC)                          // can't allocate
+        return NULL;
+
+    cmsSetProfileVersion(hICC, 4.2);
+
+    cmsSetDeviceClass(hICC,      cmsSigLinkClass);
+    cmsSetColorSpace(hICC,       ColorSpace);
+    cmsSetPCS(hICC,              ColorSpace);
+
+    cmsSetHeaderRenderingIntent(hICC,  INTENT_PERCEPTUAL); 
+
+
+    // Creates a Pipeline with 3D grid only
+    LUT = cmsPipelineAlloc(ContextID, 4, 4);
+    if (LUT == NULL) goto Error;
+    
+
+    nChannels = cmsChannelsOf(ColorSpace);
+    
+    CLUT = cmsStageAllocCLut16bit(ContextID, 17, nChannels, nChannels, NULL);
+    if (CLUT == NULL) goto Error;
+
+    if (!cmsStageSampleCLut16bit(CLUT, InkLimitingSampler, (void*) &Limit, 0)) goto Error;
+
+    cmsPipelineInsertStage(LUT, cmsAT_BEGIN, _cmsStageAllocIdentityCurves(ContextID, nChannels));  
+    cmsPipelineInsertStage(LUT, cmsAT_END, CLUT);
+    cmsPipelineInsertStage(LUT, cmsAT_END, _cmsStageAllocIdentityCurves(ContextID, nChannels));
+
+    // Create tags
+    if (!SetTextTags(hICC, L"ink-limiting built-in")) goto Error;
+
+    if (!cmsWriteTag(hICC, cmsSigAToB0Tag, (void*) LUT))  goto Error;
+    if (!SetSeqDescTag(hICC, "ink-limiting built-in")) goto Error;
+     
+    // cmsPipeline is already on virtual profile
+    cmsPipelineFree(LUT);
+
+    // Ok, done
+    return hICC;
+
+Error:
+    if (LUT != NULL)
+        cmsPipelineFree(LUT);
+
+    if (hICC != NULL)
+        cmsCloseProfile(hICC);
+
+    return NULL;
+}
+
+cmsHPROFILE CMSEXPORT cmsCreateInkLimitingDeviceLink(cmsColorSpaceSignature ColorSpace, cmsFloat64Number Limit)
+{
+    return cmsCreateInkLimitingDeviceLinkTHR(NULL, ColorSpace, Limit);
+}
+
+
+// Creates a fake Lab identity.
+cmsHPROFILE CMSEXPORT cmsCreateLab2ProfileTHR(cmsContext ContextID, const cmsCIExyY* WhitePoint)
+{
+    cmsHPROFILE hProfile;        
+    cmsPipeline* LUT = NULL;
+    
+    hProfile = cmsCreateRGBProfileTHR(ContextID, WhitePoint == NULL ? cmsD50_xyY() : WhitePoint, NULL, NULL);
+    if (hProfile == NULL) return NULL;
+
+    cmsSetProfileVersion(hProfile, 2.1);
+
+    cmsSetDeviceClass(hProfile, cmsSigAbstractClass);
+    cmsSetColorSpace(hProfile,  cmsSigLabData);
+    cmsSetPCS(hProfile,         cmsSigLabData);
+
+    if (!SetTextTags(hProfile, L"Lab identity built-in")) return NULL;
+
+    // An identity LUT is all we need
+    LUT = cmsPipelineAlloc(ContextID, 3, 3);
+    if (LUT == NULL) goto Error;
+
+    cmsPipelineInsertStage(LUT, cmsAT_BEGIN, _cmsStageAllocIdentityCLut(ContextID, 3));
+
+    if (!cmsWriteTag(hProfile, cmsSigAToB0Tag, LUT)) goto Error;
+    cmsPipelineFree(LUT);
+
+    return hProfile;
+
+Error:
+
+    if (LUT != NULL)
+        cmsPipelineFree(LUT);
+
+    if (hProfile != NULL)
+        cmsCloseProfile(hProfile);
+
+    return NULL;
+}
+
+
+cmsHPROFILE CMSEXPORT cmsCreateLab2Profile(const cmsCIExyY* WhitePoint)
+{
+    return cmsCreateLab2ProfileTHR(NULL, WhitePoint);
+}
+
+
+// Creates a fake Lab V4 identity.
+cmsHPROFILE CMSEXPORT cmsCreateLab4ProfileTHR(cmsContext ContextID, const cmsCIExyY* WhitePoint)
+{
+    cmsHPROFILE hProfile;        
+    cmsPipeline* LUT = NULL;
+    
+    hProfile = cmsCreateRGBProfileTHR(ContextID, WhitePoint == NULL ? cmsD50_xyY() : WhitePoint, NULL, NULL);
+    if (hProfile == NULL) return NULL;
+
+    cmsSetProfileVersion(hProfile, 4.2);
+
+    cmsSetDeviceClass(hProfile, cmsSigAbstractClass);
+    cmsSetColorSpace(hProfile,  cmsSigLabData);
+    cmsSetPCS(hProfile,         cmsSigLabData);
+
+    if (!SetTextTags(hProfile, L"Lab identity built-in")) goto Error;
+
+    // An empty LUTs is all we need
+    LUT = cmsPipelineAlloc(ContextID, 3, 3);
+    if (LUT == NULL) goto Error;
+
+    cmsPipelineInsertStage(LUT, cmsAT_BEGIN, _cmsStageAllocIdentityCurves(ContextID, 3));
+
+    if (!cmsWriteTag(hProfile, cmsSigAToB0Tag, LUT)) goto Error;
+    cmsPipelineFree(LUT);
+
+    return hProfile;
+
+Error:
+
+    if (LUT != NULL)
+        cmsPipelineFree(LUT);
+
+    if (hProfile != NULL)
+        cmsCloseProfile(hProfile);
+
+    return NULL;
+}
+
+cmsHPROFILE CMSEXPORT cmsCreateLab4Profile(const cmsCIExyY* WhitePoint)
+{
+    return cmsCreateLab4ProfileTHR(NULL, WhitePoint);
+}
+
+
+// Creates a fake XYZ identity
+cmsHPROFILE CMSEXPORT cmsCreateXYZProfileTHR(cmsContext ContextID)
+{
+    cmsHPROFILE hProfile;        
+    cmsPipeline* LUT = NULL;
+    
+    hProfile = cmsCreateRGBProfileTHR(ContextID, cmsD50_xyY(), NULL, NULL);
+    if (hProfile == NULL) return NULL;
+
+    cmsSetProfileVersion(hProfile, 4.2);
+
+    cmsSetDeviceClass(hProfile, cmsSigAbstractClass);
+    cmsSetColorSpace(hProfile,  cmsSigXYZData);
+    cmsSetPCS(hProfile,         cmsSigXYZData);
+
+    if (!SetTextTags(hProfile, L"XYZ identity built-in")) goto Error;
+
+    // An identity LUT is all we need
+    LUT = cmsPipelineAlloc(ContextID, 3, 3);
+    if (LUT == NULL) goto Error;
+
+    cmsPipelineInsertStage(LUT, cmsAT_BEGIN, _cmsStageAllocIdentityCurves(ContextID, 3));
+
+    if (!cmsWriteTag(hProfile, cmsSigAToB0Tag, LUT)) goto Error;
+    cmsPipelineFree(LUT);
+
+    return hProfile;
+
+Error:
+
+    if (LUT != NULL)
+        cmsPipelineFree(LUT);
+
+    if (hProfile != NULL)
+        cmsCloseProfile(hProfile);
+
+    return NULL;
+}
+
+
+cmsHPROFILE CMSEXPORT cmsCreateXYZProfile(void)
+{
+    return cmsCreateXYZProfileTHR(NULL);
+}
+
+
+//sRGB Curves are defined by:
+//
+//If  R’sRGB,G’sRGB, B’sRGB < 0.04045
+//
+//    R =  R’sRGB / 12.92
+//    G =  G’sRGB / 12.92
+//    B =  B’sRGB / 12.92
+//
+//
+//else if  R’sRGB,G’sRGB, B’sRGB >= 0.04045
+//
+//    R = ((R’sRGB + 0.055) / 1.055)^2.4
+//    G = ((G’sRGB + 0.055) / 1.055)^2.4
+//    B = ((B’sRGB + 0.055) / 1.055)^2.4
+
+static
+cmsToneCurve* Build_sRGBGamma(cmsContext ContextID)
+{
+    cmsFloat64Number Parameters[5];
+
+    Parameters[0] = 2.4;
+    Parameters[1] = 1. / 1.055;
+    Parameters[2] = 0.055 / 1.055;
+    Parameters[3] = 1. / 12.92;
+    Parameters[4] = 0.04045;    
+
+    return cmsBuildParametricToneCurve(ContextID, 4, Parameters);
+}
+
+// Create the ICC virtual profile for sRGB space 
+cmsHPROFILE CMSEXPORT cmsCreate_sRGBProfileTHR(cmsContext ContextID)
+{
+       cmsCIExyY       D65;
+       cmsCIExyYTRIPLE Rec709Primaries = {
+                                   {0.6400, 0.3300, 1.0},
+                                   {0.3000, 0.6000, 1.0},
+                                   {0.1500, 0.0600, 1.0}
+                                   };
+       cmsToneCurve* Gamma22[3];
+       cmsHPROFILE  hsRGB;
+ 
+       cmsWhitePointFromTemp(&D65, 6504);
+       Gamma22[0] = Gamma22[1] = Gamma22[2] = Build_sRGBGamma(ContextID);
+       if (Gamma22[0] == NULL) return NULL;
+           
+       hsRGB = cmsCreateRGBProfileTHR(ContextID, &D65, &Rec709Primaries, Gamma22);
+       cmsFreeToneCurve(Gamma22[0]);
+       if (hsRGB == NULL) return NULL;
+
+       if (!SetTextTags(hsRGB, L"sRGB built-in")) {
+           cmsCloseProfile(hsRGB);
+           return NULL;
+       }
+
+       return hsRGB;
+}
+
+cmsHPROFILE CMSEXPORT cmsCreate_sRGBProfile(void)
+{
+    return cmsCreate_sRGBProfileTHR(NULL);
+}
+
+
+
+typedef struct {
+                cmsFloat64Number Brightness;
+                cmsFloat64Number Contrast;
+                cmsFloat64Number Hue;
+                cmsFloat64Number Saturation;
+                cmsCIEXYZ WPsrc, WPdest;
+
+} BCHSWADJUSTS, *LPBCHSWADJUSTS;
+
+
+static
+int bchswSampler(register const cmsUInt16Number In[], register cmsUInt16Number Out[], register void* Cargo)
+{
+    cmsCIELab LabIn, LabOut;
+    cmsCIELCh LChIn, LChOut;
+    cmsCIEXYZ XYZ;
+    LPBCHSWADJUSTS bchsw = (LPBCHSWADJUSTS) Cargo;
+    
+
+    cmsLabEncoded2Float(&LabIn, In);
+    
+
+    cmsLab2LCh(&LChIn, &LabIn);
+
+    // Do some adjusts on LCh
+    
+    LChOut.L = LChIn.L * bchsw ->Contrast + bchsw ->Brightness;
+    LChOut.C = LChIn.C + bchsw -> Saturation;
+    LChOut.h = LChIn.h + bchsw -> Hue;
+    
+        
+    cmsLCh2Lab(&LabOut, &LChOut);
+
+    // Move white point in Lab
+
+    cmsLab2XYZ(&bchsw ->WPsrc,  &XYZ, &LabOut);
+    cmsXYZ2Lab(&bchsw ->WPdest, &LabOut, &XYZ);
+    
+    // Back to encoded
+
+    cmsFloat2LabEncoded(Out, &LabOut);
+    
+    return TRUE;
+}
+
+
+// Creates an abstract profile operating in Lab space for Brightness,
+// contrast, Saturation and white point displacement
+
+cmsHPROFILE CMSEXPORT cmsCreateBCHSWabstractProfileTHR(cmsContext ContextID,
+                                                     int nLUTPoints,
+                                                     cmsFloat64Number Bright, 
+                                                     cmsFloat64Number Contrast,
+                                                     cmsFloat64Number Hue,
+                                                     cmsFloat64Number Saturation,
+                                                     int TempSrc, 
+                                                     int TempDest)
+{
+     cmsHPROFILE hICC;
+     cmsPipeline* Pipeline;
+     BCHSWADJUSTS bchsw;
+     cmsCIExyY WhitePnt;
+     cmsStage* CLUT;
+     cmsUInt32Number Dimensions[MAX_INPUT_DIMENSIONS];
+     int i;
+
+
+     bchsw.Brightness = Bright;
+     bchsw.Contrast   = Contrast;
+     bchsw.Hue        = Hue;
+     bchsw.Saturation = Saturation;
+     
+     cmsWhitePointFromTemp(&WhitePnt, TempSrc );         
+     cmsxyY2XYZ(&bchsw.WPsrc, &WhitePnt);
+
+     cmsWhitePointFromTemp(&WhitePnt, TempDest);
+     cmsxyY2XYZ(&bchsw.WPdest, &WhitePnt);
+    
+      hICC = cmsCreateProfilePlaceholder(ContextID);
+       if (!hICC)                          // can't allocate
+            return NULL;
+              
+
+       cmsSetDeviceClass(hICC,      cmsSigAbstractClass);
+       cmsSetColorSpace(hICC,       cmsSigLabData);
+       cmsSetPCS(hICC,              cmsSigLabData);
+
+       cmsSetHeaderRenderingIntent(hICC,  INTENT_PERCEPTUAL); 
+
+      
+       // Creates a Pipeline with 3D grid only
+       Pipeline = cmsPipelineAlloc(ContextID, 3, 3);
+       if (Pipeline == NULL) {
+           cmsCloseProfile(hICC);
+           return NULL;
+           }
+
+       for (i=0; i < MAX_INPUT_DIMENSIONS; i++) Dimensions[i] = nLUTPoints;
+       CLUT = cmsStageAllocCLut16bitGranular(ContextID, Dimensions, 3, 3, NULL);
+       if (CLUT == NULL) return NULL;
+    
+       
+       if (!cmsStageSampleCLut16bit(CLUT, bchswSampler, (void*) &bchsw, 0)) {
+
+                // Shouldn't reach here
+                cmsPipelineFree(Pipeline);
+                cmsCloseProfile(hICC);
+                return NULL;
+       }    
+
+       cmsPipelineInsertStage(Pipeline, cmsAT_END, CLUT);
+
+       // Create tags
+        
+       if (!SetTextTags(hICC, L"BCHS built-in")) return NULL;    
+       
+       cmsWriteTag(hICC, cmsSigMediaWhitePointTag, (void*) cmsD50_XYZ());
+
+       cmsWriteTag(hICC, cmsSigAToB0Tag, (void*) Pipeline);
+       
+       // Pipeline is already on virtual profile
+       cmsPipelineFree(Pipeline);
+
+       // Ok, done
+       return hICC;
+}
+
+
+CMSAPI cmsHPROFILE   CMSEXPORT cmsCreateBCHSWabstractProfile(int nLUTPoints,
+                                                             cmsFloat64Number Bright, 
+                                                             cmsFloat64Number Contrast,
+                                                             cmsFloat64Number Hue,
+                                                             cmsFloat64Number Saturation,
+                                                             int TempSrc, 
+                                                             int TempDest)
+{
+    return cmsCreateBCHSWabstractProfileTHR(NULL, nLUTPoints, Bright, Contrast, Hue, Saturation, TempSrc, TempDest);
+}
+
+
+// Creates a fake NULL profile. This profile return 1 channel as always 0. 
+// Is useful only for gamut checking tricks
+cmsHPROFILE CMSEXPORT cmsCreateNULLProfileTHR(cmsContext ContextID)
+{
+    cmsHPROFILE hProfile;        
+    cmsPipeline* LUT = NULL;
+    cmsStage* PostLin;
+    cmsToneCurve* EmptyTab;
+    cmsUInt16Number Zero[2] = { 0, 0 };
+
+    hProfile = cmsCreateProfilePlaceholder(ContextID);
+    if (!hProfile)                          // can't allocate
+        return NULL;
+
+    cmsSetProfileVersion(hProfile, 4.2);
+
+    if (!SetTextTags(hProfile, L"NULL profile built-in")) goto Error;    
+
+    
+
+    cmsSetDeviceClass(hProfile, cmsSigOutputClass);
+    cmsSetColorSpace(hProfile,  cmsSigGrayData);
+    cmsSetPCS(hProfile,         cmsSigLabData);
+
+    // An empty LUTs is all we need
+    LUT = cmsPipelineAlloc(ContextID, 1, 1);
+    if (LUT == NULL) goto Error;
+
+    EmptyTab = cmsBuildTabulatedToneCurve16(ContextID, 2, Zero);       
+    PostLin = cmsStageAllocToneCurves(ContextID, 1, &EmptyTab);
+    cmsFreeToneCurve(EmptyTab);
+
+    cmsPipelineInsertStage(LUT, cmsAT_END, PostLin);
+
+    if (!cmsWriteTag(hProfile, cmsSigBToA0Tag, (void*) LUT)) goto Error;
+    if (!cmsWriteTag(hProfile, cmsSigMediaWhitePointTag, cmsD50_XYZ())) goto Error;
+
+    cmsPipelineFree(LUT);       
+    return hProfile;
+
+Error:
+
+    if (LUT != NULL)
+        cmsPipelineFree(LUT);
+
+    if (hProfile != NULL)
+        cmsCloseProfile(hProfile);
+
+    return NULL;
+}
+
+cmsHPROFILE CMSEXPORT cmsCreateNULLProfile(void)
+{
+    return cmsCreateNULLProfileTHR(NULL);
+}
+
+
+static
+int IsPCS(cmsColorSpaceSignature ColorSpace)
+{
+    return (ColorSpace == cmsSigXYZData ||
+            ColorSpace == cmsSigLabData);
+}
+
+
+static
+void FixColorSpaces(cmsHPROFILE hProfile, 
+                              cmsColorSpaceSignature ColorSpace, 
+                              cmsColorSpaceSignature PCS,
+                              cmsUInt32Number dwFlags)
+{
+    if (dwFlags & cmsFLAGS_GUESSDEVICECLASS) {
+
+            if (IsPCS(ColorSpace) && IsPCS(PCS)) { 
+
+                    cmsSetDeviceClass(hProfile,      cmsSigAbstractClass);
+                    cmsSetColorSpace(hProfile,       ColorSpace);
+                    cmsSetPCS(hProfile,              PCS);
+                    return;
+            }
+
+            if (IsPCS(ColorSpace) && !IsPCS(PCS)) {
+
+                    cmsSetDeviceClass(hProfile, cmsSigOutputClass);
+                    cmsSetPCS(hProfile,         ColorSpace);
+                    cmsSetColorSpace(hProfile,  PCS);        
+                    return;
+            }
+
+            if (IsPCS(PCS) && !IsPCS(ColorSpace)) {
+
+                   cmsSetDeviceClass(hProfile,  cmsSigInputClass);
+                   cmsSetColorSpace(hProfile,   ColorSpace);
+                   cmsSetPCS(hProfile,          PCS);
+                   return;
+            }
+    }
+
+    cmsSetDeviceClass(hProfile,      cmsSigLinkClass);
+    cmsSetColorSpace(hProfile,       ColorSpace);
+    cmsSetPCS(hProfile,              PCS);
+}
+
+
+
+// This function creates a named color profile dumping all the contents of transform to a single profile
+// In this way, LittleCMS may be used to "group" several named color databases into a single profile.
+// It has, however, several minor limitations. PCS is always Lab, which is not very critic since this
+// is the normal PCS for named color profiles.
+static
+cmsHPROFILE CreateNamedColorDevicelink(cmsHTRANSFORM xform)
+{
+    _cmsTRANSFORM* v = (_cmsTRANSFORM*) xform;
+    cmsHPROFILE hICC = NULL;  
+    int i, nColors;
+    cmsNAMEDCOLORLIST *nc2 = NULL, *Original = NULL;
+
+    // Create an empty placeholder
+    hICC = cmsCreateProfilePlaceholder(v->ContextID);
+    if (hICC == NULL) return NULL;
+
+    // Critical information
+    cmsSetDeviceClass(hICC, cmsSigNamedColorClass);
+    cmsSetColorSpace(hICC, v ->ExitColorSpace);
+    cmsSetPCS(hICC, cmsSigLabData);
+
+    // Tag profile with information
+    if (!SetTextTags(hICC, L"Named color devicelink")) goto Error;    
+
+    Original = cmsGetNamedColorList(xform);
+    if (Original == NULL) goto Error;
+
+    nColors = cmsNamedColorCount(Original);
+    nc2     = cmsDupNamedColorList(Original);
+    if (nc2 == NULL) goto Error;
+
+    // Colorant count now depends on the output space 
+    nc2 ->ColorantCount = cmsPipelineOutputChannels(v ->Lut);
+
+    // Apply the transfor to colorants.
+    for (i=0; i < nColors; i++) {
+        cmsDoTransform(xform, &i, nc2 ->List[i].DeviceColorant, 1);
+    }
+
+    if (!cmsWriteTag(hICC, cmsSigNamedColor2Tag, (void*) nc2)) goto Error;
+    cmsFreeNamedColorList(nc2);
+
+    return hICC;
+
+Error:
+    if (hICC != NULL) cmsCloseProfile(hICC);
+    return NULL;
+}
+
+
+// This structure holds information about which MPU can be stored on a profile based on the version
+
+typedef struct {
+    cmsBool              IsV4;             // Is a V4 tag?
+    cmsTagSignature      RequiredTag;      // Set to 0 for both types
+    cmsTagTypeSignature  LutType;          // The LUT type
+    int                  nTypes;           // Number of types (up to 5)
+    cmsStageSignature    MpeTypes[5];      // 5 is the maximum number
+
+} cmsAllowedLUT;
+
+static const cmsAllowedLUT AllowedLUTTypes[] = {
+
+    { FALSE, 0,              cmsSigLut16Type,    4,  { cmsSigMatrixElemType,  cmsSigCurveSetElemType, cmsSigCLutElemType, cmsSigCurveSetElemType}},
+    { FALSE, 0,              cmsSigLut16Type,    3,  { cmsSigCurveSetElemType, cmsSigCLutElemType, cmsSigCurveSetElemType}},
+    { TRUE , 0,              cmsSigLutAtoBType,  1,  { cmsSigCurveSetElemType }},
+    { TRUE , cmsSigAToB0Tag, cmsSigLutAtoBType,  3,  { cmsSigCurveSetElemType, cmsSigMatrixElemType, cmsSigCurveSetElemType } },
+    { TRUE , cmsSigAToB0Tag, cmsSigLutAtoBType,  3,  { cmsSigCurveSetElemType, cmsSigCLutElemType, cmsSigCurveSetElemType   } },
+    { TRUE , cmsSigAToB0Tag, cmsSigLutAtoBType,  5,  { cmsSigCurveSetElemType, cmsSigCLutElemType, cmsSigCurveSetElemType, cmsSigMatrixElemType, cmsSigCurveSetElemType }},
+    { TRUE , cmsSigBToA0Tag, cmsSigLutBtoAType,  1,  { cmsSigCurveSetElemType }},
+    { TRUE , cmsSigBToA0Tag, cmsSigLutBtoAType,  3,  { cmsSigCurveSetElemType, cmsSigMatrixElemType, cmsSigCurveSetElemType }},
+    { TRUE , cmsSigBToA0Tag, cmsSigLutBtoAType,  3,  { cmsSigCurveSetElemType, cmsSigCLutElemType, cmsSigCurveSetElemType }},
+    { TRUE , cmsSigBToA0Tag, cmsSigLutBtoAType,  5,  { cmsSigCurveSetElemType, cmsSigMatrixElemType, cmsSigCurveSetElemType, cmsSigCLutElemType, cmsSigCurveSetElemType }}
+};
+
+#define SIZE_OF_ALLOWED_LUT (sizeof(AllowedLUTTypes)/sizeof(cmsAllowedLUT))
+
+// Check a single entry
+static
+cmsBool CheckOne(const cmsAllowedLUT* Tab, const cmsPipeline* Lut)
+{
+    cmsStage* mpe;
+    int n;
+
+    for (n=0, mpe = Lut ->Elements; mpe != NULL; mpe = mpe ->Next, n++) {
+
+        if (n > Tab ->nTypes) return FALSE;
+        if (cmsStageType(mpe) != Tab ->MpeTypes[n]) return FALSE;             
+    }
+
+    return (n == Tab ->nTypes);
+}
+
+
+static 
+const cmsAllowedLUT* FindCombination(const cmsPipeline* Lut, cmsBool IsV4, cmsTagSignature DestinationTag)
+{
+    int n;
+
+    for (n=0; n < SIZE_OF_ALLOWED_LUT; n++) {
+
+        const cmsAllowedLUT* Tab = AllowedLUTTypes + n;
+
+        if (IsV4 ^ Tab -> IsV4) continue;
+        if ((Tab ->RequiredTag != 0) && (Tab ->RequiredTag != DestinationTag)) continue;
+
+        if (CheckOne(Tab, Lut)) return Tab;
+    }
+
+    return NULL;
+}
+    
+
+// Does convert a transform into a device link profile
+cmsHPROFILE CMSEXPORT cmsTransform2DeviceLink(cmsHTRANSFORM hTransform, cmsFloat64Number Version, cmsUInt32Number dwFlags)
+{
+    cmsHPROFILE hProfile = NULL;
+    cmsUInt32Number FrmIn, FrmOut, ChansIn, ChansOut;
+    cmsUInt32Number ColorSpaceBitsIn, ColorSpaceBitsOut;
+    _cmsTRANSFORM* xform = (_cmsTRANSFORM*) hTransform;
+    cmsPipeline* LUT = NULL;
+    cmsStage* mpe;
+    cmsContext ContextID = cmsGetTransformContextID(hTransform);
+    const cmsAllowedLUT* AllowedLUT;
+    cmsTagSignature DestinationTag;
+
+    _cmsAssert(hTransform != NULL);
+    
+    // Get the first mpe to check for named color
+    mpe = cmsPipelineGetPtrToFirstStage(xform ->Lut);
+    
+    // Check if is a named color transform
+    if (mpe != NULL) {
+
+        if (cmsStageType(mpe) == cmsSigNamedColorElemType) {
+            return CreateNamedColorDevicelink(hTransform);
+        }
+    }
+
+    // First thing to do is to get a copy of the transformation
+    LUT = cmsPipelineDup(xform ->Lut);
+    if (LUT == NULL) return NULL;
+
+    // Time to fix the Lab2/Lab4 issue.
+    if ((xform ->EntryColorSpace == cmsSigLabData) && (Version < 4.0)) {
+
+        cmsPipelineInsertStage(LUT, cmsAT_BEGIN, _cmsStageAllocLabV2ToV4curves(ContextID));        
+    }
+
+    // On the output side too
+    if ((xform ->ExitColorSpace) == cmsSigLabData && (Version < 4.0)) {
+
+        cmsPipelineInsertStage(LUT, cmsAT_END, _cmsStageAllocLabV4ToV2(ContextID));        
+    }
+    
+
+    hProfile = cmsCreateProfilePlaceholder(ContextID);
+    if (!hProfile) goto Error;                    // can't allocate        
+    
+    cmsSetProfileVersion(hProfile, Version);
+
+    FixColorSpaces(hProfile, xform -> EntryColorSpace, xform -> ExitColorSpace, dwFlags);   
+
+    // Optimize the LUT and precalculate a devicelink
+
+    ChansIn  = cmsChannelsOf(xform -> EntryColorSpace);
+    ChansOut = cmsChannelsOf(xform -> ExitColorSpace);
+
+    ColorSpaceBitsIn  = _cmsLCMScolorSpace(xform -> EntryColorSpace);
+    ColorSpaceBitsOut = _cmsLCMScolorSpace(xform -> ExitColorSpace);
+
+    FrmIn  = COLORSPACE_SH(ColorSpaceBitsIn) | CHANNELS_SH(ChansIn)|BYTES_SH(2);
+    FrmOut = COLORSPACE_SH(ColorSpaceBitsOut) | CHANNELS_SH(ChansOut)|BYTES_SH(2);
+
+
+     if (cmsGetDeviceClass(hProfile) == cmsSigOutputClass)
+         DestinationTag = cmsSigBToA0Tag;
+     else
+         DestinationTag = cmsSigAToB0Tag;
+
+    // Check if the profile/version can store the result
+    if (dwFlags & cmsFLAGS_FORCE_CLUT)
+        AllowedLUT = NULL;
+    else
+    AllowedLUT = FindCombination(LUT, Version >= 4.0, DestinationTag);
+
+    if (AllowedLUT == NULL) {
+
+        // Try to optimize
+        _cmsOptimizePipeline(&LUT, xform ->RenderingIntent, &FrmIn, &FrmOut, &dwFlags);
+        AllowedLUT = FindCombination(LUT, Version >= 4.0, DestinationTag);
+
+    }
+
+    // If no way, then force CLUT that for sure can be written
+    if (AllowedLUT == NULL) {
+
+        dwFlags |= cmsFLAGS_FORCE_CLUT;
+        _cmsOptimizePipeline(&LUT, xform ->RenderingIntent, &FrmIn, &FrmOut, &dwFlags);
+
+        // Put identity curves if needed
+        if (cmsPipelineStageCount(LUT) == 1) {
+
+            cmsPipelineInsertStage(LUT, cmsAT_BEGIN, _cmsStageAllocIdentityCurves(ContextID, ChansIn));    
+            cmsPipelineInsertStage(LUT, cmsAT_END,   _cmsStageAllocIdentityCurves(ContextID, ChansOut));   
+        }
+
+        AllowedLUT = FindCombination(LUT, Version >= 4.0, DestinationTag);
+    }
+
+    // Somethings is wrong...
+    if (AllowedLUT == NULL) {
+        goto Error;
+    }
+
+
+    if (dwFlags & cmsFLAGS_8BITS_DEVICELINK) 
+                     cmsPipelineSetSaveAs8bitsFlag(LUT, TRUE);
+        
+    // Tag profile with information
+    if (!SetTextTags(hProfile, L"devicelink")) goto Error;    
+                
+    // Store result            
+    if (!cmsWriteTag(hProfile, DestinationTag, LUT)) goto Error;
+    
+    
+    if (xform -> InputColorant != NULL) {
+           if (!cmsWriteTag(hProfile, cmsSigColorantTableTag, xform->InputColorant)) goto Error;
+    }
+
+    if (xform -> OutputColorant != NULL) {
+           if (!cmsWriteTag(hProfile, cmsSigColorantTableOutTag, xform->OutputColorant)) goto Error;
+    }
+    
+    if (xform ->Sequence != NULL) {
+        if (!_cmsWriteProfileSequence(hProfile, xform ->Sequence)) goto Error;
+    }
+
+    cmsPipelineFree(LUT);       
+    return hProfile;
+
+Error:
+    if (LUT != NULL) cmsPipelineFree(LUT);       
+    cmsCloseProfile(hProfile);
+    return NULL;
+}
diff --git a/thirdparty/liblcms2/src/cmswtpnt.c b/thirdparty/liblcms2/src/cmswtpnt.c
new file mode 100644
index 00000000..6319984c
--- /dev/null
+++ b/thirdparty/liblcms2/src/cmswtpnt.c
@@ -0,0 +1,351 @@
+//---------------------------------------------------------------------------------
+//
+//  Little Color Management System
+//  Copyright (c) 1998-2010 Marti Maria Saguer
+//
+// Permission is hereby granted, free of charge, to any person obtaining 
+// a copy of this software and associated documentation files (the "Software"), 
+// to deal in the Software without restriction, including without limitation 
+// the rights to use, copy, modify, merge, publish, distribute, sublicense, 
+// and/or sell copies of the Software, and to permit persons to whom the Software 
+// is furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in 
+// all copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 
+// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO 
+// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE 
+// LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 
+// OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 
+// WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+//
+//---------------------------------------------------------------------------------
+//
+
+#include "lcms2_internal.h"
+
+
+// D50 - Widely used
+const cmsCIEXYZ* CMSEXPORT cmsD50_XYZ(void)
+{
+    static cmsCIEXYZ D50XYZ = {cmsD50X, cmsD50Y, cmsD50Z};
+
+    return &D50XYZ;
+}
+
+const cmsCIExyY* CMSEXPORT cmsD50_xyY(void)
+{
+    static cmsCIExyY D50xyY;
+
+    cmsXYZ2xyY(&D50xyY, cmsD50_XYZ());
+
+    return &D50xyY;
+}
+
+// Obtains WhitePoint from Temperature
+cmsBool  CMSEXPORT cmsWhitePointFromTemp(cmsCIExyY* WhitePoint, cmsFloat64Number TempK)
+{
+       cmsFloat64Number x, y;
+       cmsFloat64Number T, T2, T3;
+       // cmsFloat64Number M1, M2;
+
+	   _cmsAssert(WhitePoint != NULL);
+
+       T = TempK;
+       T2 = T*T;            // Square
+       T3 = T2*T;           // Cube
+
+       // For correlated color temperature (T) between 4000K and 7000K:
+
+       if (T >= 4000. && T <= 7000.)
+       {
+              x = -4.6070*(1E9/T3) + 2.9678*(1E6/T2) + 0.09911*(1E3/T) + 0.244063;
+       }
+       else
+              // or for correlated color temperature (T) between 7000K and 25000K:
+
+       if (T > 7000.0 && T <= 25000.0)
+       {
+              x = -2.0064*(1E9/T3) + 1.9018*(1E6/T2) + 0.24748*(1E3/T) + 0.237040;
+       }
+       else {
+              cmsSignalError(0, cmsERROR_RANGE, "cmsWhitePointFromTemp: invalid temp");
+              return FALSE;
+              }
+
+       // Obtain y(x)
+
+       y = -3.000*(x*x) + 2.870*x - 0.275;
+
+       // wave factors (not used, but here for futures extensions)
+
+       // M1 = (-1.3515 - 1.7703*x + 5.9114 *y)/(0.0241 + 0.2562*x - 0.7341*y);
+       // M2 = (0.0300 - 31.4424*x + 30.0717*y)/(0.0241 + 0.2562*x - 0.7341*y);
+
+       WhitePoint -> x = x;
+       WhitePoint -> y = y;
+       WhitePoint -> Y = 1.0;
+
+       return TRUE;
+}
+
+
+
+typedef struct {
+
+    cmsFloat64Number mirek;  // temp (in microreciprocal kelvin) 
+    cmsFloat64Number ut;     // u coord of intersection w/ blackbody locus  
+    cmsFloat64Number vt;     // v coord of intersection w/ blackbody locus 
+    cmsFloat64Number tt;     // slope of ISOTEMPERATURE. line 
+
+    } ISOTEMPERATURE;
+
+static ISOTEMPERATURE isotempdata[] = {
+//  {Mirek, Ut,       Vt,      Tt      } 
+    {0,     0.18006,  0.26352,  -0.24341},
+    {10,    0.18066,  0.26589,  -0.25479},
+    {20,    0.18133,  0.26846,  -0.26876},
+    {30,    0.18208,  0.27119,  -0.28539},
+    {40,    0.18293,  0.27407,  -0.30470},
+    {50,    0.18388,  0.27709,  -0.32675},
+    {60,    0.18494,  0.28021,  -0.35156},
+    {70,    0.18611,  0.28342,  -0.37915},
+    {80,    0.18740,  0.28668,  -0.40955},
+    {90,    0.18880,  0.28997,  -0.44278},
+    {100,   0.19032,  0.29326,  -0.47888},
+    {125,   0.19462,  0.30141,  -0.58204},
+    {150,   0.19962,  0.30921,  -0.70471},
+    {175,   0.20525,  0.31647,  -0.84901},
+    {200,   0.21142,  0.32312,  -1.0182 },
+    {225,   0.21807,  0.32909,  -1.2168 },
+    {250,   0.22511,  0.33439,  -1.4512 },
+    {275,   0.23247,  0.33904,  -1.7298 },
+    {300,   0.24010,  0.34308,  -2.0637 },
+    {325,   0.24702,  0.34655,  -2.4681 },
+    {350,   0.25591,  0.34951,  -2.9641 },
+    {375,   0.26400,  0.35200,  -3.5814 },
+    {400,   0.27218,  0.35407,  -4.3633 },
+    {425,   0.28039,  0.35577,  -5.3762 },
+    {450,   0.28863,  0.35714,  -6.7262 },
+    {475,   0.29685,  0.35823,  -8.5955 },
+    {500,   0.30505,  0.35907,  -11.324 },
+    {525,   0.31320,  0.35968,  -15.628 },
+    {550,   0.32129,  0.36011,  -23.325 },
+    {575,   0.32931,  0.36038,  -40.770 },
+    {600,   0.33724,  0.36051,  -116.45  }
+};
+
+#define NISO sizeof(isotempdata)/sizeof(ISOTEMPERATURE)
+
+
+// Robertson's method
+cmsBool  CMSEXPORT cmsTempFromWhitePoint(cmsFloat64Number* TempK, const cmsCIExyY* WhitePoint)
+{
+	int j;
+	cmsFloat64Number us,vs;
+	cmsFloat64Number uj,vj,tj,di,dj,mi,mj;
+	cmsFloat64Number xs, ys;
+
+	_cmsAssert(WhitePoint != NULL);
+    _cmsAssert(TempK != NULL);
+
+	di = mi = 0;
+	xs = WhitePoint -> x;
+	ys = WhitePoint -> y;
+
+	// convert (x,y) to CIE 1960 (u,WhitePoint) 
+
+	us = (2*xs) / (-xs + 6*ys + 1.5);
+	vs = (3*ys) / (-xs + 6*ys + 1.5);
+
+
+	for (j=0; j < NISO; j++) {
+
+		uj = isotempdata[j].ut;
+		vj = isotempdata[j].vt;
+		tj = isotempdata[j].tt;
+		mj = isotempdata[j].mirek;
+
+		dj = ((vs - vj) - tj * (us - uj)) / sqrt(1.0 + tj * tj);
+
+		if ((j != 0) && (di/dj < 0.0)) {
+
+			// Found a match
+			*TempK = 1000000.0 / (mi + (di / (di - dj)) * (mj - mi));
+			return TRUE;
+		}
+
+		di = dj;
+		mi = mj;
+	}
+
+	// Not found
+	return FALSE;
+}
+
+
+// Compute chromatic adaptation matrix using Chad as cone matrix 
+
+static
+cmsBool ComputeChromaticAdaptation(cmsMAT3* Conversion,
+                                const cmsCIEXYZ* SourceWhitePoint,
+                                const cmsCIEXYZ* DestWhitePoint,
+                                const cmsMAT3* Chad)
+
+{
+      
+    cmsMAT3 Chad_Inv;
+    cmsVEC3 ConeSourceXYZ, ConeSourceRGB;
+    cmsVEC3 ConeDestXYZ, ConeDestRGB;
+    cmsMAT3 Cone, Tmp;
+
+
+    Tmp = *Chad;
+    if (!_cmsMAT3inverse(&Tmp, &Chad_Inv)) return FALSE;
+
+    _cmsVEC3init(&ConeSourceXYZ, SourceWhitePoint -> X,
+                             SourceWhitePoint -> Y,
+                             SourceWhitePoint -> Z);
+
+    _cmsVEC3init(&ConeDestXYZ,   DestWhitePoint -> X,
+                             DestWhitePoint -> Y,
+                             DestWhitePoint -> Z);
+
+    _cmsMAT3eval(&ConeSourceRGB, Chad, &ConeSourceXYZ);
+    _cmsMAT3eval(&ConeDestRGB,   Chad, &ConeDestXYZ);
+
+    // Build matrix
+    _cmsVEC3init(&Cone.v[0], ConeDestRGB.n[0]/ConeSourceRGB.n[0],    0.0,  0.0);
+    _cmsVEC3init(&Cone.v[1], 0.0,   ConeDestRGB.n[1]/ConeSourceRGB.n[1],   0.0);
+    _cmsVEC3init(&Cone.v[2], 0.0,   0.0,   ConeDestRGB.n[2]/ConeSourceRGB.n[2]);
+
+
+    // Normalize
+    _cmsMAT3per(&Tmp, &Cone, Chad);
+    _cmsMAT3per(Conversion, &Chad_Inv, &Tmp);
+
+	return TRUE;
+}
+
+// Returns the final chrmatic adaptation from illuminant FromIll to Illuminant ToIll
+// The cone matrix can be specified in ConeMatrix. If NULL, Bradford is assumed
+cmsBool  _cmsAdaptationMatrix(cmsMAT3* r, const cmsMAT3* ConeMatrix, const cmsCIEXYZ* FromIll, const cmsCIEXYZ* ToIll)
+{
+	cmsMAT3 LamRigg   = {{ // Bradford matrix
+		{{  0.8951,  0.2664, -0.1614 }},
+		{{ -0.7502,  1.7135,  0.0367 }},
+		{{  0.0389, -0.0685,  1.0296 }}
+	}};
+
+	if (ConeMatrix == NULL)
+		ConeMatrix = &LamRigg;
+
+	return ComputeChromaticAdaptation(r, FromIll, ToIll, ConeMatrix);	
+}
+
+// Same as anterior, but assuming D50 destination. White point is given in xyY
+static
+cmsBool _cmsAdaptMatrixToD50(cmsMAT3* r, const cmsCIExyY* SourceWhitePt)
+{
+	cmsCIEXYZ Dn;      
+	cmsMAT3 Bradford;
+	cmsMAT3 Tmp;
+
+	cmsxyY2XYZ(&Dn, SourceWhitePt);
+
+	if (!_cmsAdaptationMatrix(&Bradford, NULL, &Dn, cmsD50_XYZ())) return FALSE;
+
+	Tmp = *r;
+	_cmsMAT3per(r, &Bradford, &Tmp);
+
+	return TRUE;
+}
+
+// Build a White point, primary chromas transfer matrix from RGB to CIE XYZ
+// This is just an approximation, I am not handling all the non-linear
+// aspects of the RGB to XYZ process, and assumming that the gamma correction
+// has transitive property in the tranformation chain.
+//
+// the alghoritm:
+//
+//            - First I build the absolute conversion matrix using
+//              primaries in XYZ. This matrix is next inverted
+//            - Then I eval the source white point across this matrix
+//              obtaining the coeficients of the transformation
+//            - Then, I apply these coeficients to the original matrix
+//
+cmsBool _cmsBuildRGB2XYZtransferMatrix(cmsMAT3* r, const cmsCIExyY* WhitePt, const cmsCIExyYTRIPLE* Primrs)
+{
+	cmsVEC3 WhitePoint, Coef;
+	cmsMAT3 Result, Primaries;
+	cmsFloat64Number xn, yn;
+	cmsFloat64Number xr, yr;
+	cmsFloat64Number xg, yg;
+	cmsFloat64Number xb, yb;
+
+	xn = WhitePt -> x;
+	yn = WhitePt -> y;
+	xr = Primrs -> Red.x;
+	yr = Primrs -> Red.y;
+	xg = Primrs -> Green.x;
+	yg = Primrs -> Green.y;
+	xb = Primrs -> Blue.x;
+	yb = Primrs -> Blue.y;
+
+	// Build Primaries matrix
+	_cmsVEC3init(&Primaries.v[0], xr,        xg,         xb);
+	_cmsVEC3init(&Primaries.v[1], yr,        yg,         yb);
+	_cmsVEC3init(&Primaries.v[2], (1-xr-yr), (1-xg-yg),  (1-xb-yb));
+
+
+	// Result = Primaries ^ (-1) inverse matrix
+	if (!_cmsMAT3inverse(&Primaries, &Result))
+		return FALSE;
+
+
+	_cmsVEC3init(&WhitePoint, xn/yn, 1.0, (1.0-xn-yn)/yn);
+
+	// Across inverse primaries ...
+	_cmsMAT3eval(&Coef, &Result, &WhitePoint);
+
+	// Give us the Coefs, then I build transformation matrix
+	_cmsVEC3init(&r -> v[0], Coef.n[VX]*xr,          Coef.n[VY]*xg,          Coef.n[VZ]*xb);
+	_cmsVEC3init(&r -> v[1], Coef.n[VX]*yr,          Coef.n[VY]*yg,          Coef.n[VZ]*yb);
+	_cmsVEC3init(&r -> v[2], Coef.n[VX]*(1.0-xr-yr), Coef.n[VY]*(1.0-xg-yg), Coef.n[VZ]*(1.0-xb-yb));
+
+
+	return _cmsAdaptMatrixToD50(r, WhitePt);
+	
+}
+
+
+// Adapts a color to a given illuminant. Original color is expected to have
+// a SourceWhitePt white point. 
+cmsBool CMSEXPORT cmsAdaptToIlluminant(cmsCIEXYZ* Result, 
+							 const cmsCIEXYZ* SourceWhitePt, 
+							 const cmsCIEXYZ* Illuminant, 
+							 const cmsCIEXYZ* Value)
+{
+	cmsMAT3 Bradford;
+	cmsVEC3 In, Out;
+
+	_cmsAssert(Result != NULL);
+	_cmsAssert(SourceWhitePt != NULL);
+	_cmsAssert(Illuminant != NULL);
+	_cmsAssert(Value != NULL);
+
+	if (!_cmsAdaptationMatrix(&Bradford, NULL, SourceWhitePt, Illuminant)) return FALSE;
+
+	_cmsVEC3init(&In, Value -> X, Value -> Y, Value -> Z);
+	_cmsMAT3eval(&Out, &Bradford, &In);
+
+	Result -> X = Out.n[0];
+	Result -> Y = Out.n[1];
+	Result -> Z = Out.n[2];
+
+	return TRUE;
+}
+
+
diff --git a/thirdparty/liblcms2/src/cmsxform.c b/thirdparty/liblcms2/src/cmsxform.c
new file mode 100644
index 00000000..7f5cc947
--- /dev/null
+++ b/thirdparty/liblcms2/src/cmsxform.c
@@ -0,0 +1,809 @@
+//---------------------------------------------------------------------------------
+//
+//  Little Color Management System
+//  Copyright (c) 1998-2010 Marti Maria Saguer
+//
+// Permission is hereby granted, free of charge, to any person obtaining 
+// a copy of this software and associated documentation files (the "Software"), 
+// to deal in the Software without restriction, including without limitation 
+// the rights to use, copy, modify, merge, publish, distribute, sublicense, 
+// and/or sell copies of the Software, and to permit persons to whom the Software 
+// is furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in 
+// all copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 
+// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO 
+// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE 
+// LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 
+// OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 
+// WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+//
+//---------------------------------------------------------------------------------
+//
+
+#include "lcms2_internal.h"
+
+// Transformations stuff
+// -----------------------------------------------------------------------
+
+// Alarm codes for 16-bit transformations, because the fixed range of containers there are
+// no values left to mark out of gamut. volatile is C99 per 6.2.5
+static volatile cmsUInt16Number Alarm[cmsMAXCHANNELS];
+static volatile cmsFloat64Number GlobalAdaptationState = 0;
+
+// The adaptation state may be defaulted by this function. If you don't like it, use the extended transform routine
+cmsFloat64Number CMSEXPORT cmsSetAdaptationState(cmsFloat64Number d)
+{
+    cmsFloat64Number OldVal = GlobalAdaptationState;
+
+    if (d >= 0) 
+        GlobalAdaptationState = d;
+
+    return OldVal;  
+}
+
+// Alarm codes are always global
+void CMSEXPORT cmsSetAlarmCodes(cmsUInt16Number NewAlarm[cmsMAXCHANNELS])
+{
+    int i;
+
+    _cmsAssert(NewAlarm != NULL);
+
+    for (i=0; i < cmsMAXCHANNELS; i++) 
+        Alarm[i] = NewAlarm[i];
+}
+
+// You can get the codes cas well
+void CMSEXPORT cmsGetAlarmCodes(cmsUInt16Number OldAlarm[cmsMAXCHANNELS])
+{
+    int i;
+
+    _cmsAssert(OldAlarm != NULL);
+
+    for (i=0; i < cmsMAXCHANNELS; i++) 
+        OldAlarm[i] = Alarm[i];
+}
+
+// Get rid of transform resources
+void CMSEXPORT cmsDeleteTransform(cmsHTRANSFORM hTransform)
+{
+    _cmsTRANSFORM* p = (_cmsTRANSFORM*) hTransform;
+
+    _cmsAssert(p != NULL);
+
+    if (p -> GamutCheck)
+        cmsPipelineFree(p -> GamutCheck);
+
+    if (p -> Lut)
+        cmsPipelineFree(p -> Lut);
+
+    if (p ->InputColorant)
+        cmsFreeNamedColorList(p ->InputColorant);
+
+    if (p -> OutputColorant)
+        cmsFreeNamedColorList(p ->OutputColorant);
+
+    if (p ->Sequence)
+        cmsFreeProfileSequenceDescription(p ->Sequence);
+
+    LCMS_FREE_LOCK(&p->rwlock);
+    _cmsFree(p ->ContextID, (void *) p);
+}
+
+// Apply transform 
+void CMSEXPORT cmsDoTransform(cmsHTRANSFORM  Transform,
+                              const void* InputBuffer,
+                              void* OutputBuffer, 
+                              cmsUInt32Number Size)
+
+{
+    _cmsTRANSFORM* p = (_cmsTRANSFORM*) Transform;
+
+    p -> xform(p, InputBuffer, OutputBuffer, Size);
+}
+
+
+// Transform routines ----------------------------------------------------------------------------------------------------------
+
+// Float xform converts floats. Since there are no performance issues, one routine does all job, including gamut check.
+// Note that because extended range, we can use a -1.0 value for out of gamut in this case.
+static
+void FloatXFORM(_cmsTRANSFORM* p,
+                const void* in,
+                void* out, cmsUInt32Number Size)
+{
+    cmsUInt8Number* accum;
+    cmsUInt8Number* output;
+    cmsFloat32Number fIn[cmsMAXCHANNELS], fOut[cmsMAXCHANNELS];
+    cmsFloat32Number OutOfGamut;
+    cmsUInt32Number i, j;
+
+    accum  = (cmsUInt8Number*)  in;
+    output = (cmsUInt8Number*)  out;
+
+    for (i=0; i < Size; i++) {
+
+        accum = p -> FromInputFloat(p, fIn, accum, Size);         
+
+        // Any gamut chack to do?
+        if (p ->GamutCheck != NULL) {
+
+            // Evaluate gamut marker.
+            cmsPipelineEvalFloat( fIn, &OutOfGamut, p ->GamutCheck);
+
+            // Is current color out of gamut?
+            if (OutOfGamut > 0.0) {
+
+                // Certainly, out of gamut
+                for (j=0; j < cmsMAXCHANNELS; j++)
+                    fOut[j] = -1.0;
+
+            }
+            else {
+                // No, proceed normally
+                cmsPipelineEvalFloat(fIn, fOut, p -> Lut); 
+            }
+        }
+        else {
+
+            // No gamut check at all
+            cmsPipelineEvalFloat(fIn, fOut, p -> Lut);     
+        }
+
+        // Back to asked representation
+        output = p -> ToOutputFloat(p, fOut, output, Size);
+    }
+}
+
+// 16 bit precision -----------------------------------------------------------------------------------------------------------
+
+// Null transformation, only applies formatters. No caché
+static
+void NullXFORM(_cmsTRANSFORM* p,
+               const void* in,
+               void* out, cmsUInt32Number Size)
+{
+    cmsUInt8Number* accum;
+    cmsUInt8Number* output;
+    cmsUInt16Number wIn[cmsMAXCHANNELS];
+    cmsUInt32Number i, n;
+
+    accum  = (cmsUInt8Number*)  in;
+    output = (cmsUInt8Number*)  out;
+    n = Size;                    // Buffer len
+
+    for (i=0; i < n; i++) {
+
+        accum  = p -> FromInput(p, wIn, accum, Size);
+        output = p -> ToOutput(p, wIn, output, Size);
+    }
+}
+
+
+// No gamut check, no cache, 16 bits
+static
+void PrecalculatedXFORM(_cmsTRANSFORM* p,
+                        const void* in,
+                        void* out, cmsUInt32Number Size)
+{
+    register cmsUInt8Number* accum;
+    register cmsUInt8Number* output;
+    cmsUInt16Number wIn[cmsMAXCHANNELS], wOut[cmsMAXCHANNELS];
+    cmsUInt32Number i, n;
+
+    accum  = (cmsUInt8Number*)  in;
+    output = (cmsUInt8Number*)  out;
+    n = Size;                    
+
+    for (i=0; i < n; i++) {
+
+        accum = p -> FromInput(p, wIn, accum, Size);         
+        p ->Lut ->Eval16Fn(wIn, wOut, p -> Lut->Data);     
+        output = p -> ToOutput(p, wOut, output, Size);
+    }
+}
+
+
+// Auxiliar: Handle precalculated gamut check
+static
+void TransformOnePixelWithGamutCheck(_cmsTRANSFORM* p, 
+                                     const cmsUInt16Number wIn[], 
+                                     cmsUInt16Number wOut[])
+{
+    cmsUInt16Number wOutOfGamut;
+
+    p ->GamutCheck ->Eval16Fn(wIn, &wOutOfGamut, p ->GamutCheck ->Data);   
+    if (wOutOfGamut >= 1) {
+
+        cmsUInt16Number i;
+
+        for (i=0; i < p ->Lut->OutputChannels; i++)
+            wOut[i] = Alarm[i];                      
+    }
+    else
+        p ->Lut ->Eval16Fn(wIn, wOut, p -> Lut->Data);   
+}
+
+// Gamut check, No caché, 16 bits.
+static
+void PrecalculatedXFORMGamutCheck(_cmsTRANSFORM* p,
+                                  const void* in,
+                                  void* out, cmsUInt32Number Size)
+{
+    cmsUInt8Number* accum;
+    cmsUInt8Number* output;
+    cmsUInt16Number wIn[cmsMAXCHANNELS], wOut[cmsMAXCHANNELS];
+    cmsUInt32Number i, n;
+
+    accum  = (cmsUInt8Number*)  in;
+    output = (cmsUInt8Number*)  out;
+    n = Size;                    // Buffer len
+
+    for (i=0; i < n; i++) {
+
+        accum = p -> FromInput(p, wIn, accum, Size);
+        TransformOnePixelWithGamutCheck(p, wIn, wOut);
+        output = p -> ToOutput(p, wOut, output, Size);
+    }
+}
+
+
+// No gamut check, Caché, 16 bits, 
+static
+void CachedXFORM(_cmsTRANSFORM* p,
+                 const void* in,
+                 void* out, cmsUInt32Number Size)
+{
+    cmsUInt8Number* accum;
+    cmsUInt8Number* output;
+    cmsUInt16Number wIn[cmsMAXCHANNELS], wOut[cmsMAXCHANNELS];
+    cmsUInt32Number i, n;
+    cmsUInt16Number CacheIn[cmsMAXCHANNELS], CacheOut[cmsMAXCHANNELS];
+
+    accum  = (cmsUInt8Number*)  in;
+    output = (cmsUInt8Number*)  out;
+    n = Size;                    // Buffer len
+
+    // Empty buffers for quick memcmp
+    memset(wIn,  0, sizeof(wIn));
+    memset(wOut, 0, sizeof(wOut));
+
+
+    LCMS_READ_LOCK(&p ->rwlock);
+    memmove(CacheIn,  p ->CacheIn, sizeof(CacheIn));
+    memmove(CacheOut, p ->CacheOut, sizeof(CacheOut));
+    LCMS_UNLOCK(&p ->rwlock);
+
+    for (i=0; i < n; i++) {
+
+        accum = p -> FromInput(p, wIn, accum, Size);
+
+        if (memcmp(wIn, CacheIn, sizeof(CacheIn)) == 0) {
+
+            memmove(wOut, CacheOut, sizeof(CacheOut));
+        }
+        else {   
+
+            p ->Lut ->Eval16Fn(wIn, wOut, p -> Lut->Data);  
+
+            memmove(CacheIn,  wIn,  sizeof(CacheIn));
+            memmove(CacheOut, wOut, sizeof(CacheOut));
+        }
+
+        output = p -> ToOutput(p, wOut, output, Size);            
+    }
+
+
+    LCMS_WRITE_LOCK(&p ->rwlock);
+    memmove(p->CacheIn,  CacheIn, sizeof(CacheIn));
+    memmove(p->CacheOut, CacheOut, sizeof(CacheOut));
+    LCMS_UNLOCK(&p ->rwlock);
+}
+
+
+// All those nice features together
+static
+void CachedXFORMGamutCheck(_cmsTRANSFORM* p,
+                           const void* in,
+                           void* out, cmsUInt32Number Size)
+{
+       cmsUInt8Number* accum;
+       cmsUInt8Number* output;
+       cmsUInt16Number wIn[cmsMAXCHANNELS], wOut[cmsMAXCHANNELS];
+       cmsUInt32Number i, n;
+       cmsUInt16Number CacheIn[cmsMAXCHANNELS], CacheOut[cmsMAXCHANNELS];
+
+       accum  = (cmsUInt8Number*)  in;
+       output = (cmsUInt8Number*)  out;
+       n = Size;                    // Buffer len
+
+       // Empty buffers for quick memcmp
+       memset(wIn,  0, sizeof(cmsUInt16Number) * cmsMAXCHANNELS);
+       memset(wOut, 0, sizeof(cmsUInt16Number) * cmsMAXCHANNELS);
+
+       LCMS_READ_LOCK(&p ->rwlock);
+           memmove(CacheIn,  p ->CacheIn, sizeof(cmsUInt16Number) * cmsMAXCHANNELS);
+           memmove(CacheOut, p ->CacheOut, sizeof(cmsUInt16Number) * cmsMAXCHANNELS);
+       LCMS_UNLOCK(&p ->rwlock);
+
+
+       for (i=0; i < n; i++) {
+
+            accum = p -> FromInput(p, wIn, accum, Size);
+     
+            if (memcmp(wIn, CacheIn, sizeof(cmsUInt16Number) * cmsMAXCHANNELS) == 0) {
+                    memmove(wOut, CacheOut, sizeof(cmsUInt16Number) * cmsMAXCHANNELS);
+            }
+            else {            
+                    TransformOnePixelWithGamutCheck(p, wIn, wOut);
+                    memmove(CacheIn, wIn, sizeof(cmsUInt16Number) * cmsMAXCHANNELS);
+                    memmove(CacheOut, wOut, sizeof(cmsUInt16Number) * cmsMAXCHANNELS);
+            }
+
+            output = p -> ToOutput(p, wOut, output, Size);
+       }
+
+        LCMS_WRITE_LOCK(&p ->rwlock);
+           memmove(p->CacheIn,  CacheIn, sizeof(cmsUInt16Number) * cmsMAXCHANNELS);
+           memmove(p->CacheOut, CacheOut, sizeof(cmsUInt16Number) * cmsMAXCHANNELS);
+        LCMS_UNLOCK(&p ->rwlock);
+}
+
+
+
+
+// Allocate transform struct and set it to defaults
+static
+_cmsTRANSFORM* AllocEmptyTransform(cmsContext ContextID, cmsUInt32Number InputFormat, cmsUInt32Number OutputFormat, cmsUInt32Number dwFlags)
+{
+    // Allocate needed memory
+    _cmsTRANSFORM* p = (_cmsTRANSFORM*) _cmsMallocZero(ContextID, sizeof(_cmsTRANSFORM));
+    if (!p) return NULL;
+
+    // Check whatever this is a true floating point transform
+    if (_cmsFormatterIsFloat(InputFormat) && _cmsFormatterIsFloat(OutputFormat)) {
+
+        // Get formatter function always return a valid union, but the contents of this union may be NULL.
+        p ->FromInputFloat = _cmsGetFormatter(InputFormat,  cmsFormatterInput, CMS_PACK_FLAGS_FLOAT).FmtFloat;
+        p ->ToOutputFloat  = _cmsGetFormatter(OutputFormat, cmsFormatterOutput, CMS_PACK_FLAGS_FLOAT).FmtFloat;
+        dwFlags |= cmsFLAGS_CAN_CHANGE_FORMATTER;
+
+        if (p ->FromInputFloat == NULL || p ->ToOutputFloat == NULL) {
+        
+            cmsSignalError(ContextID, cmsERROR_UNKNOWN_EXTENSION, "Unsupported raster format");
+            _cmsFree(ContextID, p);
+            return NULL;
+        }
+
+        // Float transforms don't use caché, always are non-NULL
+        p ->xform = FloatXFORM;
+    }
+    else {
+
+        if (InputFormat == 0 && OutputFormat == 0) {
+            p ->FromInput = p ->ToOutput = NULL;
+        }
+        else {
+
+            int BytesPerPixelInput;
+
+            p ->FromInput = _cmsGetFormatter(InputFormat,  cmsFormatterInput, CMS_PACK_FLAGS_16BITS).Fmt16;
+            p ->ToOutput  = _cmsGetFormatter(OutputFormat, cmsFormatterOutput, CMS_PACK_FLAGS_16BITS).Fmt16;
+
+
+            if (p ->FromInput == NULL || p ->ToOutput == NULL) {
+
+                cmsSignalError(ContextID, cmsERROR_UNKNOWN_EXTENSION, "Unsupported raster format");
+                _cmsFree(ContextID, p);
+                return NULL;
+            }
+
+            BytesPerPixelInput = T_BYTES(p ->InputFormat);
+            if (BytesPerPixelInput == 0 || BytesPerPixelInput >= 2) 
+                   dwFlags |= cmsFLAGS_CAN_CHANGE_FORMATTER;
+
+        }
+
+        if (dwFlags & cmsFLAGS_NULLTRANSFORM) {
+
+            p ->xform = NullXFORM;
+        }
+        else {
+            if (dwFlags & cmsFLAGS_NOCACHE) {
+
+                if (dwFlags & cmsFLAGS_GAMUTCHECK) 
+                    p ->xform = PrecalculatedXFORMGamutCheck;  // Gamut check, no caché
+                else
+                    p ->xform = PrecalculatedXFORM;  // No caché, no gamut check
+            }
+            else {
+
+                if (dwFlags & cmsFLAGS_GAMUTCHECK) 
+                    p ->xform = CachedXFORMGamutCheck;    // Gamut check, caché
+                else
+                    p ->xform = CachedXFORM;  // No gamut check, caché
+
+            }
+        }
+    }
+
+    
+    // Create a mutex for shared memory
+    LCMS_CREATE_LOCK(&p->rwlock);
+
+    p ->InputFormat     = InputFormat;
+    p ->OutputFormat    = OutputFormat;
+    p ->dwOriginalFlags = dwFlags;
+    p ->ContextID       = ContextID;
+    return p;
+}
+
+static
+cmsBool GetXFormColorSpaces(int nProfiles, cmsHPROFILE hProfiles[], cmsColorSpaceSignature* Input, cmsColorSpaceSignature* Output) 
+{    
+    cmsColorSpaceSignature ColorSpaceIn, ColorSpaceOut;   
+    cmsColorSpaceSignature PostColorSpace;   
+    int i;
+
+	if (hProfiles[0] == NULL) return FALSE;
+
+    *Input = PostColorSpace = cmsGetColorSpace(hProfiles[0]);
+
+    for (i=0; i < nProfiles; i++) {
+
+        cmsHPROFILE hProfile = hProfiles[i];
+
+        int lIsInput = (PostColorSpace != cmsSigXYZData) &&
+                       (PostColorSpace != cmsSigLabData);
+
+		if (hProfile == NULL) return FALSE;
+
+        if (lIsInput) {
+
+            ColorSpaceIn    = cmsGetColorSpace(hProfile);
+            ColorSpaceOut   = cmsGetPCS(hProfile);
+        }
+        else {
+
+            ColorSpaceIn    = cmsGetPCS(hProfile);
+            ColorSpaceOut   = cmsGetColorSpace(hProfile);
+        }
+
+        PostColorSpace = ColorSpaceOut;     
+    }  
+
+    *Output = PostColorSpace;
+
+	return TRUE;
+}
+
+// Check colorspace
+static
+cmsBool  IsProperColorSpace(cmsColorSpaceSignature Check, cmsUInt32Number dwFormat)
+{
+    int Space1 = T_COLORSPACE(dwFormat);
+    int Space2 = _cmsLCMScolorSpace(Check);
+
+    if (Space1 == PT_ANY) return TRUE;
+    if (Space1 == Space2) return TRUE;
+
+    if (Space1 == PT_LabV2 && Space2 == PT_Lab) return TRUE;
+    if (Space1 == PT_Lab   && Space2 == PT_LabV2) return TRUE;
+
+    return FALSE;
+}
+
+// ----------------------------------------------------------------------------------------------------------------
+
+// New to lcms 2.0 -- have all parameters available.
+cmsHTRANSFORM CMSEXPORT cmsCreateExtendedTransform(cmsContext ContextID,
+                                                   cmsUInt32Number nProfiles, cmsHPROFILE hProfiles[], 
+                                                   cmsBool  BPC[], 
+                                                   cmsUInt32Number Intents[], 
+                                                   cmsFloat64Number AdaptationStates[],
+                                                   cmsHPROFILE hGamutProfile,
+                                                   cmsUInt32Number nGamutPCSposition,           
+                                                   cmsUInt32Number InputFormat,
+                                                   cmsUInt32Number OutputFormat,
+                                                   cmsUInt32Number dwFlags)
+{
+    _cmsTRANSFORM* xform;
+    cmsBool  FloatTransform;
+    cmsColorSpaceSignature EntryColorSpace;
+    cmsColorSpaceSignature ExitColorSpace;
+    cmsPipeline* Lut;
+    cmsUInt32Number LastIntent = Intents[nProfiles-1];
+
+    // If gamut check is requested, make sure we have a gamut profile
+    if (dwFlags & cmsFLAGS_GAMUTCHECK) {
+        if (hGamutProfile == NULL) dwFlags &= ~cmsFLAGS_GAMUTCHECK;
+    }
+
+    // On floating point transforms, inhibit optimizations 
+    FloatTransform = (_cmsFormatterIsFloat(InputFormat) && _cmsFormatterIsFloat(OutputFormat));
+
+    if (_cmsFormatterIsFloat(InputFormat) || _cmsFormatterIsFloat(OutputFormat))
+        dwFlags |= cmsFLAGS_NOCACHE;
+
+    // Mark entry/exit spaces
+	if (!GetXFormColorSpaces(nProfiles, hProfiles, &EntryColorSpace, &ExitColorSpace)) {
+		cmsSignalError(ContextID, cmsERROR_NULL, "NULL input profiles on transform");        
+		return NULL;
+	}
+
+    // Check if proper colorspaces
+    if (!IsProperColorSpace(EntryColorSpace, InputFormat)) {        
+        cmsSignalError(ContextID, cmsERROR_COLORSPACE_CHECK, "Wrong input color space on transform");        
+        return NULL;
+    }
+
+    if (!IsProperColorSpace(ExitColorSpace, OutputFormat)) {
+        cmsSignalError(ContextID, cmsERROR_COLORSPACE_CHECK, "Wrong output color space on transform");       
+        return NULL;
+    }
+
+    // Create a pipeline with all transformations
+    Lut = _cmsLinkProfiles(ContextID, nProfiles, Intents, hProfiles, BPC, AdaptationStates, dwFlags);
+    if (Lut == NULL) {
+        cmsSignalError(ContextID, cmsERROR_NOT_SUITABLE, "Couldn't link the profiles");   
+        return NULL;
+    }
+
+    // Optimize the LUT if possible
+    _cmsOptimizePipeline(&Lut, LastIntent, &InputFormat, &OutputFormat, &dwFlags);       
+
+
+    // All seems ok
+    xform = AllocEmptyTransform(ContextID, InputFormat, OutputFormat, dwFlags);
+    if (xform == NULL) {
+        cmsPipelineFree(Lut);
+        return NULL;
+    }
+
+    // Keep values
+    xform ->EntryColorSpace = EntryColorSpace;
+    xform ->ExitColorSpace  = ExitColorSpace;
+    xform ->Lut             = Lut;
+
+
+    // Create a gamut check LUT if requested
+    if (hGamutProfile != NULL && (dwFlags & cmsFLAGS_GAMUTCHECK))       
+        xform ->GamutCheck  = _cmsCreateGamutCheckPipeline(ContextID, hProfiles, 
+                                                        BPC, Intents, 
+                                                        AdaptationStates, 
+                                                        nGamutPCSposition, 
+                                                        hGamutProfile);
+
+
+    // Try to read input and output colorant table
+    if (cmsIsTag(hProfiles[0], cmsSigColorantTableTag)) {
+
+        // Input table can only come in this way.       
+        xform ->InputColorant = cmsDupNamedColorList((cmsNAMEDCOLORLIST*) cmsReadTag(hProfiles[0], cmsSigColorantTableTag));
+    }
+
+    // Output is a little bit more complex.    
+    if (cmsGetDeviceClass(hProfiles[nProfiles-1]) == cmsSigLinkClass) {
+
+        // This tag may exist only on devicelink profiles.        
+        if (cmsIsTag(hProfiles[nProfiles-1], cmsSigColorantTableOutTag)) {
+
+            // It may be NULL if error
+            xform ->OutputColorant = cmsDupNamedColorList((cmsNAMEDCOLORLIST*) cmsReadTag(hProfiles[nProfiles-1], cmsSigColorantTableOutTag));
+        }
+
+    } else {
+
+        if (cmsIsTag(hProfiles[nProfiles-1], cmsSigColorantTableTag)) {
+
+            xform -> OutputColorant = cmsDupNamedColorList((cmsNAMEDCOLORLIST*) cmsReadTag(hProfiles[nProfiles-1], cmsSigColorantTableTag));
+        }     
+    }
+
+    // Store the sequence of profiles
+    if (dwFlags & cmsFLAGS_KEEP_SEQUENCE) {
+        xform ->Sequence = _cmsCompileProfileSequence(ContextID, nProfiles, hProfiles);
+    }
+    else 
+        xform ->Sequence = NULL;
+
+    // If this is a cached transform, init first value, which is zero (16 bits only)
+    if (!(dwFlags & cmsFLAGS_NOCACHE)) {
+
+        memset(&xform ->CacheIn, 0, sizeof(xform ->CacheIn));
+
+        if (xform ->GamutCheck != NULL) {
+            TransformOnePixelWithGamutCheck(xform, xform ->CacheIn, xform->CacheOut);
+        }
+        else {
+
+            xform ->Lut ->Eval16Fn(xform ->CacheIn, xform->CacheOut, xform -> Lut->Data);  
+        }
+
+    }
+
+    return (cmsHTRANSFORM) xform; 
+}
+
+// Multiprofile transforms: Gamut check is not available here, as it is unclear from which profile the gamut comes.
+
+cmsHTRANSFORM CMSEXPORT cmsCreateMultiprofileTransformTHR(cmsContext ContextID,
+                                                       cmsHPROFILE hProfiles[],
+                                                       cmsUInt32Number nProfiles,
+                                                       cmsUInt32Number InputFormat,
+                                                       cmsUInt32Number OutputFormat,
+                                                       cmsUInt32Number Intent,
+                                                       cmsUInt32Number dwFlags)
+{
+    cmsUInt32Number i;
+    cmsBool BPC[256];
+    cmsUInt32Number Intents[256];
+    cmsFloat64Number AdaptationStates[256];
+
+    if (nProfiles <= 0 || nProfiles > 255) {
+         cmsSignalError(ContextID, cmsERROR_RANGE, "Wrong number of profiles. 1..255 expected, %d found.", nProfiles);
+        return NULL;
+    }
+
+    for (i=0; i < nProfiles; i++) {
+        BPC[i] = dwFlags & cmsFLAGS_BLACKPOINTCOMPENSATION ? TRUE : FALSE;
+        Intents[i] = Intent;
+        AdaptationStates[i] = GlobalAdaptationState;
+    }
+
+
+    return cmsCreateExtendedTransform(ContextID, nProfiles, hProfiles, BPC, Intents, AdaptationStates, NULL, 0, InputFormat, OutputFormat, dwFlags);
+}
+
+
+
+cmsHTRANSFORM CMSEXPORT cmsCreateMultiprofileTransform(cmsHPROFILE hProfiles[],
+                                                  cmsUInt32Number nProfiles,
+                                                  cmsUInt32Number InputFormat,
+                                                  cmsUInt32Number OutputFormat,
+                                                  cmsUInt32Number Intent,
+                                                  cmsUInt32Number dwFlags)
+{
+
+    if (nProfiles <= 0 || nProfiles > 255) {
+         cmsSignalError(NULL, cmsERROR_RANGE, "Wrong number of profiles. 1..255 expected, %d found.", nProfiles);
+         return NULL;
+    }
+
+    return cmsCreateMultiprofileTransformTHR(cmsGetProfileContextID(hProfiles[0]),
+                                                  hProfiles,
+                                                  nProfiles,
+                                                  InputFormat,
+                                                  OutputFormat,
+                                                  Intent,
+                                                  dwFlags);
+}
+
+cmsHTRANSFORM CMSEXPORT cmsCreateTransformTHR(cmsContext ContextID,
+                                              cmsHPROFILE Input,
+                                              cmsUInt32Number InputFormat,
+                                              cmsHPROFILE Output,
+                                              cmsUInt32Number OutputFormat,
+                                              cmsUInt32Number Intent,
+                                              cmsUInt32Number dwFlags)
+{
+
+    cmsHPROFILE hArray[2];
+    
+    hArray[0] = Input;
+    hArray[1] = Output;
+
+    return cmsCreateMultiprofileTransformTHR(ContextID, hArray, Output == NULL ? 1 : 2, InputFormat, OutputFormat, Intent, dwFlags);
+}
+
+CMSAPI cmsHTRANSFORM CMSEXPORT cmsCreateTransform(cmsHPROFILE Input,
+                                                  cmsUInt32Number InputFormat,
+                                                  cmsHPROFILE Output,
+                                                  cmsUInt32Number OutputFormat,
+                                                  cmsUInt32Number Intent,
+                                                  cmsUInt32Number dwFlags)
+{
+    return cmsCreateTransformTHR(cmsGetProfileContextID(Input), Input, InputFormat, Output, OutputFormat, Intent, dwFlags);
+}
+
+
+cmsHTRANSFORM CMSEXPORT cmsCreateProofingTransformTHR(cmsContext ContextID,
+                                                   cmsHPROFILE InputProfile,
+                                                   cmsUInt32Number InputFormat,
+                                                   cmsHPROFILE OutputProfile,
+                                                   cmsUInt32Number OutputFormat,
+                                                   cmsHPROFILE ProofingProfile,
+                                                   cmsUInt32Number nIntent,
+                                                   cmsUInt32Number ProofingIntent,
+                                                   cmsUInt32Number dwFlags)
+{   
+    cmsHPROFILE hArray[4];
+    cmsUInt32Number Intents[4];
+    cmsBool  BPC[4];
+    cmsFloat64Number Adaptation[4];
+    cmsBool  DoBPC = (dwFlags & cmsFLAGS_BLACKPOINTCOMPENSATION) ? TRUE : FALSE;
+
+
+    hArray[0]  = InputProfile; hArray[1] = ProofingProfile; hArray[2]  = ProofingProfile;               hArray[3] = OutputProfile;
+    Intents[0] = nIntent;      Intents[1] = nIntent;        Intents[2] = INTENT_RELATIVE_COLORIMETRIC;  Intents[3] = ProofingIntent;
+    BPC[0]     = DoBPC;        BPC[1] = DoBPC;              BPC[2] = 0;                                 BPC[3] = 0;
+    
+    Adaptation[0] = Adaptation[1] = Adaptation[2] = Adaptation[3] = GlobalAdaptationState;
+
+    if (!(dwFlags & (cmsFLAGS_SOFTPROOFING|cmsFLAGS_GAMUTCHECK))) 
+        return cmsCreateTransformTHR(ContextID, InputProfile, InputFormat, OutputProfile, OutputFormat, nIntent, dwFlags);
+  
+    return cmsCreateExtendedTransform(ContextID, 4, hArray, BPC, Intents, Adaptation, 
+                                        ProofingProfile, 1, InputFormat, OutputFormat, dwFlags);
+
+}
+
+
+cmsHTRANSFORM CMSEXPORT cmsCreateProofingTransform(cmsHPROFILE InputProfile,
+                                                   cmsUInt32Number InputFormat,
+                                                   cmsHPROFILE OutputProfile,
+                                                   cmsUInt32Number OutputFormat,
+                                                   cmsHPROFILE ProofingProfile,
+                                                   cmsUInt32Number nIntent,
+                                                   cmsUInt32Number ProofingIntent,
+                                                   cmsUInt32Number dwFlags)
+{
+    return cmsCreateProofingTransformTHR(cmsGetProfileContextID(InputProfile), 
+                                                   InputProfile,
+                                                   InputFormat,
+                                                   OutputProfile,
+                                                   OutputFormat,
+                                                   ProofingProfile,
+                                                   nIntent,
+                                                   ProofingIntent,
+                                                   dwFlags);
+}
+
+
+// Grab the ContextID from an open transform. Returns NULL if a NULL transform is passed
+cmsContext CMSEXPORT cmsGetTransformContextID(cmsHTRANSFORM hTransform)
+{
+    _cmsTRANSFORM* xform = (_cmsTRANSFORM*) hTransform;
+
+    if (xform == NULL) return NULL;
+    return xform -> ContextID;
+}
+
+
+
+// For backwards compatibility
+cmsBool CMSEXPORT cmsChangeBuffersFormat(cmsHTRANSFORM hTransform, 
+                                         cmsUInt32Number InputFormat, 
+                                         cmsUInt32Number OutputFormat)
+{
+
+    _cmsTRANSFORM* xform = (_cmsTRANSFORM*) hTransform;
+    cmsFormatter16 FromInput, ToOutput;
+    cmsUInt32Number BytesPerPixelInput;
+
+    // We only can afford to change formatters if previous transform is at least 16 bits
+    BytesPerPixelInput = T_BYTES(xform ->InputFormat);
+    if (!(xform ->dwOriginalFlags & cmsFLAGS_CAN_CHANGE_FORMATTER)) {
+
+        cmsSignalError(xform ->ContextID, cmsERROR_NOT_SUITABLE, "cmsChangeBuffersFormat works only on transforms created originally with at least 16 bits of precision");
+        return FALSE;
+    }
+
+    FromInput = _cmsGetFormatter(InputFormat,  cmsFormatterInput, CMS_PACK_FLAGS_16BITS).Fmt16;
+    ToOutput  = _cmsGetFormatter(OutputFormat, cmsFormatterOutput, CMS_PACK_FLAGS_16BITS).Fmt16;
+
+    if (FromInput == NULL || ToOutput == NULL) {
+
+        cmsSignalError(xform -> ContextID, cmsERROR_UNKNOWN_EXTENSION, "Unsupported raster format");
+        return FALSE;
+    }
+
+    xform ->InputFormat  = InputFormat;
+    xform ->OutputFormat = OutputFormat;
+    xform ->FromInput = FromInput;
+    xform ->ToOutput  = ToOutput;
+    return TRUE;
+}
diff --git a/thirdparty/liblcms2/src/lcms2.def b/thirdparty/liblcms2/src/lcms2.def
new file mode 100644
index 00000000..d4075021
--- /dev/null
+++ b/thirdparty/liblcms2/src/lcms2.def
@@ -0,0 +1,300 @@
+LIBRARY     LCMS2.DLL
+
+EXPORTS
+
+_cms15Fixed16toDouble                    =   _cms15Fixed16toDouble
+_cms8Fixed8toDouble                      =   _cms8Fixed8toDouble
+cmsAdaptToIlluminant                     =    cmsAdaptToIlluminant
+_cmsAdjustEndianess16                    =   _cmsAdjustEndianess16
+_cmsAdjustEndianess32                    =   _cmsAdjustEndianess32
+_cmsAdjustEndianess64                    =   _cmsAdjustEndianess64
+cmsAllocNamedColorList                   =   cmsAllocNamedColorList
+cmsAllocProfileSequenceDescription       =   cmsAllocProfileSequenceDescription
+cmsAppendNamedColor                      =   cmsAppendNamedColor
+cmsBFDdeltaE                             =   cmsBFDdeltaE
+cmsBuildGamma                            =   cmsBuildGamma
+cmsBuildParametricToneCurve              =   cmsBuildParametricToneCurve
+cmsBuildSegmentedToneCurve               =   cmsBuildSegmentedToneCurve
+cmsBuildTabulatedToneCurve16             =   cmsBuildTabulatedToneCurve16
+cmsBuildTabulatedToneCurveFloat          =   cmsBuildTabulatedToneCurveFloat
+_cmsCalloc                               =   _cmsCalloc
+cmsChannelsOf                            =    cmsChannelsOf
+cmsCIE2000DeltaE                         =    cmsCIE2000DeltaE
+cmsCIE94DeltaE                           =    cmsCIE94DeltaE
+cmsCIECAM02Done                          =    cmsCIECAM02Done
+cmsCIECAM02Forward                       =    cmsCIECAM02Forward
+cmsCIECAM02Init                          =    cmsCIECAM02Init
+cmsCIECAM02Reverse                       =    cmsCIECAM02Reverse
+cmsCloseIOhandler                        =    cmsCloseIOhandler
+cmsCloseProfile                          =    cmsCloseProfile
+cmsCMCdeltaE                             =    cmsCMCdeltaE
+cmsCreate_sRGBProfile                    =    cmsCreate_sRGBProfile
+cmsCreate_sRGBProfileTHR                 =    cmsCreate_sRGBProfileTHR
+cmsCreateBCHSWabstractProfile            =    cmsCreateBCHSWabstractProfile
+cmsCreateBCHSWabstractProfileTHR         =    cmsCreateBCHSWabstractProfileTHR
+cmsCreateExtendedTransform               =    cmsCreateExtendedTransform
+cmsCreateGrayProfile                     =    cmsCreateGrayProfile
+cmsCreateGrayProfileTHR                  =    cmsCreateGrayProfileTHR
+cmsCreateInkLimitingDeviceLink           =    cmsCreateInkLimitingDeviceLink
+cmsCreateInkLimitingDeviceLinkTHR        =    cmsCreateInkLimitingDeviceLinkTHR
+cmsCreateLab2Profile                     =    cmsCreateLab2Profile
+cmsCreateLab2ProfileTHR                  =    cmsCreateLab2ProfileTHR
+cmsCreateLab4Profile                     =    cmsCreateLab4Profile
+cmsCreateLab4ProfileTHR                  =    cmsCreateLab4ProfileTHR
+cmsCreateLinearizationDeviceLink         =    cmsCreateLinearizationDeviceLink
+cmsCreateLinearizationDeviceLinkTHR      =    cmsCreateLinearizationDeviceLinkTHR
+cmsCreateMultiprofileTransform           =    cmsCreateMultiprofileTransform
+cmsCreateMultiprofileTransformTHR        =    cmsCreateMultiprofileTransformTHR
+cmsCreateNULLProfile                     =    cmsCreateNULLProfile
+cmsCreateNULLProfileTHR                  =    cmsCreateNULLProfileTHR
+cmsCreateProfilePlaceholder              =    cmsCreateProfilePlaceholder
+cmsCreateProofingTransform               =    cmsCreateProofingTransform
+cmsCreateProofingTransformTHR            =    cmsCreateProofingTransformTHR
+cmsCreateRGBProfile                      =    cmsCreateRGBProfile
+cmsCreateRGBProfileTHR                   =    cmsCreateRGBProfileTHR
+cmsCreateTransform                       =    cmsCreateTransform
+cmsCreateTransformTHR                    =    cmsCreateTransformTHR
+cmsCreateXYZProfile                      =    cmsCreateXYZProfile
+cmsCreateXYZProfileTHR                   =    cmsCreateXYZProfileTHR
+cmsD50_xyY                               =    cmsD50_xyY
+cmsD50_XYZ                               =    cmsD50_XYZ
+_cmsDecodeDateTimeNumber                 =    _cmsDecodeDateTimeNumber
+_cmsDefaultICCintents                    =    _cmsDefaultICCintents
+cmsDeleteTransform                       =    cmsDeleteTransform
+cmsDeltaE                                =    cmsDeltaE
+cmsDetectBlackPoint                      =    cmsDetectBlackPoint
+cmsDetectTAC                             =    cmsDetectTAC
+cmsDesaturateLab                         =    cmsDesaturateLab
+cmsDoTransform                           =    cmsDoTransform
+_cmsDoubleTo15Fixed16                    =    _cmsDoubleTo15Fixed16
+_cmsDoubleTo8Fixed8                      =    _cmsDoubleTo8Fixed8
+_cmsDupMem                               =    _cmsDupMem
+cmsDupNamedColorList                     =    cmsDupNamedColorList
+cmsDupProfileSequenceDescription         =    cmsDupProfileSequenceDescription
+cmsDupToneCurve                          =    cmsDupToneCurve
+_cmsEncodeDateTimeNumber                 =    _cmsEncodeDateTimeNumber
+cmsEstimateGamma                         =    cmsEstimateGamma
+cmsEvalToneCurve16                       =    cmsEvalToneCurve16
+cmsEvalToneCurveFloat                    =    cmsEvalToneCurveFloat
+cmsfilelength                            =    cmsfilelength
+cmsFloat2LabEncoded                      =    cmsFloat2LabEncoded
+cmsFloat2LabEncodedV2                    =    cmsFloat2LabEncodedV2
+cmsFloat2XYZEncoded                      =    cmsFloat2XYZEncoded
+cmsFormatterForColorspaceOfProfile       =    cmsFormatterForColorspaceOfProfile
+cmsFormatterForPCSOfProfile              =    cmsFormatterForPCSOfProfile
+_cmsFree                                 =    _cmsFree
+cmsFreeNamedColorList                    =    cmsFreeNamedColorList
+cmsFreeProfileSequenceDescription        =    cmsFreeProfileSequenceDescription
+cmsFreeToneCurve                         =    cmsFreeToneCurve
+cmsFreeToneCurveTriple                   =    cmsFreeToneCurveTriple
+cmsGBDAlloc                              =    cmsGBDAlloc
+cmsGBDFree                               =    cmsGBDFree
+cmsGDBAddPoint                           =    cmsGDBAddPoint
+cmsGDBCheckPoint                         =    cmsGDBCheckPoint
+cmsGDBCompute                            =    cmsGDBCompute
+cmsGetAlarmCodes                         =    cmsGetAlarmCodes
+cmsGetColorSpace                         =    cmsGetColorSpace
+cmsGetDeviceClass                        =    cmsGetDeviceClass
+cmsGetEncodedICCversion                  =    cmsGetEncodedICCversion
+cmsGetHeaderAttributes                   =    cmsGetHeaderAttributes
+cmsGetHeaderCreationDateTime             =    cmsGetHeaderCreationDateTime
+cmsGetHeaderFlags                        =    cmsGetHeaderFlags
+cmsGetHeaderManufacturer                 =    cmsGetHeaderManufacturer
+cmsGetHeaderModel                        =    cmsGetHeaderModel
+cmsGetHeaderProfileID                    =    cmsGetHeaderProfileID
+cmsGetHeaderRenderingIntent              =    cmsGetHeaderRenderingIntent
+cmsGetNamedColorList                     =    cmsGetNamedColorList
+cmsGetPCS                                =    cmsGetPCS
+cmsGetPostScriptColorResource            =    cmsGetPostScriptColorResource
+cmsGetPostScriptCRD                      =    cmsGetPostScriptCRD
+cmsGetPostScriptCSA                      =    cmsGetPostScriptCSA
+cmsGetProfileInfo                        =    cmsGetProfileInfo
+cmsGetProfileInfoASCII                   =    cmsGetProfileInfoASCII
+cmsGetProfileContextID                    =    cmsGetProfileContextID
+cmsGetProfileVersion                     =    cmsGetProfileVersion
+cmsGetSupportedIntents                   =    cmsGetSupportedIntents
+cmsGetTagCount                           =    cmsGetTagCount
+cmsGetTagSignature                       =    cmsGetTagSignature
+cmsGetTransformContextID                  =    cmsGetTransformContextID
+_cmsICCcolorSpace                        =    _cmsICCcolorSpace
+_cmsIOPrintf                             =    _cmsIOPrintf
+cmsIsCLUT                                =    cmsIsCLUT
+cmsIsIntentSupported                     =    cmsIsIntentSupported
+cmsIsMatrixShaper                        =    cmsIsMatrixShaper
+cmsIsTag                                 =    cmsIsTag
+cmsIsToneCurveDescending                 =    cmsIsToneCurveDescending
+cmsIsToneCurveLinear                     =    cmsIsToneCurveLinear
+cmsIsToneCurveMonotonic                  =    cmsIsToneCurveMonotonic
+cmsIsToneCurveMultisegment               =    cmsIsToneCurveMultisegment
+cmsGetToneCurveParametricType            =    cmsGetToneCurveParametricType
+cmsIT8Alloc                              =    cmsIT8Alloc
+cmsIT8DefineDblFormat                    =    cmsIT8DefineDblFormat
+cmsIT8EnumDataFormat                     =    cmsIT8EnumDataFormat
+cmsIT8EnumProperties                     =    cmsIT8EnumProperties
+cmsIT8Free                               =    cmsIT8Free
+cmsIT8GetData                            =    cmsIT8GetData
+cmsIT8GetDataDbl                         =    cmsIT8GetDataDbl
+cmsIT8FindDataFormat                      =    cmsIT8FindDataFormat
+cmsIT8GetDataRowCol                      =    cmsIT8GetDataRowCol
+cmsIT8GetDataRowColDbl                   =    cmsIT8GetDataRowColDbl
+cmsIT8GetPatchName                       =    cmsIT8GetPatchName
+cmsIT8GetProperty                        =    cmsIT8GetProperty
+cmsIT8GetPropertyDbl                     =    cmsIT8GetPropertyDbl
+cmsIT8GetSheetType                       =    cmsIT8GetSheetType
+cmsIT8LoadFromFile                       =    cmsIT8LoadFromFile
+cmsIT8LoadFromMem                        =    cmsIT8LoadFromMem
+cmsIT8SaveToFile                         =    cmsIT8SaveToFile
+cmsIT8SaveToMem                          =    cmsIT8SaveToMem
+cmsIT8SetComment                         =    cmsIT8SetComment
+cmsIT8SetData                            =    cmsIT8SetData
+cmsIT8SetDataDbl                         =    cmsIT8SetDataDbl
+cmsIT8SetDataFormat                      =    cmsIT8SetDataFormat
+cmsIT8SetDataRowCol                      =    cmsIT8SetDataRowCol
+cmsIT8SetDataRowColDbl                   =    cmsIT8SetDataRowColDbl
+cmsIT8SetPropertyDbl                     =    cmsIT8SetPropertyDbl
+cmsIT8SetPropertyHex                     =    cmsIT8SetPropertyHex
+cmsIT8SetPropertyStr                     =    cmsIT8SetPropertyStr
+cmsIT8SetPropertyUncooked                =    cmsIT8SetPropertyUncooked
+cmsIT8SetSheetType                       =    cmsIT8SetSheetType
+cmsIT8SetTable                           =    cmsIT8SetTable
+cmsIT8SetTableByLabel                    =    cmsIT8SetTableByLabel
+cmsIT8TableCount                         =    cmsIT8TableCount
+cmsJoinToneCurve                         =    cmsJoinToneCurve
+cmsLab2LCh                               =    cmsLab2LCh
+cmsLab2XYZ                               =    cmsLab2XYZ
+cmsLabEncoded2Float                      =    cmsLabEncoded2Float
+cmsLabEncoded2FloatV2                    =    cmsLabEncoded2FloatV2
+cmsLCh2Lab                               =    cmsLCh2Lab
+_cmsLCMScolorSpace                       =    _cmsLCMScolorSpace
+cmsLinkTag                               =    cmsLinkTag
+cmsPipelineAlloc                         =    cmsPipelineAlloc
+cmsPipelineCat                           =    cmsPipelineCat
+cmsPipelineCheckAndRetreiveStages        =    cmsPipelineCheckAndRetreiveStages
+cmsPipelineDup                           =    cmsPipelineDup
+cmsPipelineStageCount                    =    cmsPipelineStageCount
+cmsPipelineEval16                        =    cmsPipelineEval16
+cmsPipelineEvalFloat                     =    cmsPipelineEvalFloat
+cmsPipelineEvalReverseFloat              =    cmsPipelineEvalReverseFloat
+cmsPipelineFree                          =    cmsPipelineFree
+cmsPipelineGetPtrToFirstStage            =    cmsPipelineGetPtrToFirstStage
+cmsPipelineGetPtrToLastStage             =    cmsPipelineGetPtrToLastStage
+cmsPipelineInputChannels                 =    cmsPipelineInputChannels
+cmsPipelineInsertStage                   =    cmsPipelineInsertStage
+cmsPipelineOutputChannels                =    cmsPipelineOutputChannels
+cmsPipelineSetSaveAs8bitsFlag            =    cmsPipelineSetSaveAs8bitsFlag
+_cmsPipelineSetOptimizationParameters    =    _cmsPipelineSetOptimizationParameters
+cmsPipelineUnlinkStage                   =    cmsPipelineUnlinkStage
+_cmsMalloc                               =    _cmsMalloc
+_cmsMallocZero                           =    _cmsMallocZero
+_cmsMAT3eval                             =    _cmsMAT3eval
+_cmsMAT3identity                         =    _cmsMAT3identity
+_cmsMAT3inverse                          =    _cmsMAT3inverse
+_cmsMAT3isIdentity                       =    _cmsMAT3isIdentity
+_cmsMAT3per                              =    _cmsMAT3per
+_cmsMAT3solve                            =    _cmsMAT3solve
+cmsMD5computeID                          =    cmsMD5computeID
+cmsMLUalloc                              =    cmsMLUalloc
+cmsMLUdup                                =    cmsMLUdup
+cmsMLUfree                               =    cmsMLUfree
+cmsMLUgetASCII                           =    cmsMLUgetASCII
+cmsMLUgetTranslation                     =    cmsMLUgetTranslation
+cmsMLUgetWide                            =    cmsMLUgetWide
+cmsMLUsetASCII                           =    cmsMLUsetASCII
+cmsMLUsetWide                            =    cmsMLUsetWide
+cmsStageAllocCLut16bit                   =    cmsStageAllocCLut16bit
+cmsStageAllocCLut16bitGranular           =    cmsStageAllocCLut16bitGranular
+cmsStageAllocCLutFloat                   =    cmsStageAllocCLutFloat
+cmsStageAllocCLutFloatGranular           =    cmsStageAllocCLutFloatGranular
+cmsStageAllocToneCurves                  =    cmsStageAllocToneCurves
+cmsStageAllocIdentity                    =    cmsStageAllocIdentity
+cmsStageAllocMatrix                      =    cmsStageAllocMatrix
+_cmsStageAllocPlaceholder                =    _cmsStageAllocPlaceholder
+cmsStageDup                              =    cmsStageDup
+cmsStageFree                             =    cmsStageFree
+cmsStageNext                             =    cmsStageNext
+cmsStageInputChannels                    =    cmsStageInputChannels
+cmsStageOutputChannels                   =    cmsStageOutputChannels
+cmsStageSampleCLut16bit                  =    cmsStageSampleCLut16bit
+cmsStageSampleCLutFloat                  =    cmsStageSampleCLutFloat
+cmsStageType                             =    cmsStageType
+cmsStageData                             =    cmsStageData
+cmsNamedColorCount                       =    cmsNamedColorCount
+cmsNamedColorIndex                       =    cmsNamedColorIndex
+cmsNamedColorInfo                        =    cmsNamedColorInfo
+cmsOpenIOhandlerFromFile                 =    cmsOpenIOhandlerFromFile
+cmsOpenIOhandlerFromMem                  =    cmsOpenIOhandlerFromMem
+cmsOpenIOhandlerFromNULL                 =    cmsOpenIOhandlerFromNULL
+cmsOpenIOhandlerFromStream               =    cmsOpenIOhandlerFromStream
+cmsOpenProfileFromFile                   =    cmsOpenProfileFromFile
+cmsOpenProfileFromFileTHR                =    cmsOpenProfileFromFileTHR
+cmsOpenProfileFromIOhandlerTHR           =    cmsOpenProfileFromIOhandlerTHR
+cmsOpenProfileFromMem                    =    cmsOpenProfileFromMem
+cmsOpenProfileFromMemTHR                 =    cmsOpenProfileFromMemTHR
+cmsOpenProfileFromStream                 =    cmsOpenProfileFromStream
+cmsOpenProfileFromStreamTHR              =    cmsOpenProfileFromStreamTHR
+cmsPlugin                                =    cmsPlugin
+_cmsRead15Fixed16Number                  =    _cmsRead15Fixed16Number
+_cmsReadAlignment                        =    _cmsReadAlignment
+_cmsReadFloat32Number                    =    _cmsReadFloat32Number
+cmsReadRawTag                            =    cmsReadRawTag
+cmsReadTag                               =    cmsReadTag
+_cmsReadTypeBase                         =    _cmsReadTypeBase
+_cmsReadUInt16Array                      =    _cmsReadUInt16Array
+_cmsReadUInt16Number                     =    _cmsReadUInt16Number
+_cmsReadUInt32Number                     =    _cmsReadUInt32Number
+_cmsReadUInt64Number                     =    _cmsReadUInt64Number
+_cmsReadUInt8Number                      =    _cmsReadUInt8Number
+_cmsReadXYZNumber                        =    _cmsReadXYZNumber
+_cmsRealloc                              =    _cmsRealloc
+cmsReverseToneCurve                      =    cmsReverseToneCurve
+cmsReverseToneCurveEx                    =    cmsReverseToneCurveEx
+cmsSaveProfileToFile                     =    cmsSaveProfileToFile
+cmsSaveProfileToIOhandler                =    cmsSaveProfileToIOhandler
+cmsSaveProfileToMem                      =    cmsSaveProfileToMem
+cmsSaveProfileToStream                   =    cmsSaveProfileToStream
+cmsSetAdaptationState                    =    cmsSetAdaptationState
+cmsSetAlarmCodes                         =    cmsSetAlarmCodes
+cmsSetColorSpace                         =    cmsSetColorSpace
+cmsSetDeviceClass                        =    cmsSetDeviceClass
+cmsSetEncodedICCversion                  =    cmsSetEncodedICCversion
+cmsSetHeaderAttributes                   =    cmsSetHeaderAttributes
+cmsSetHeaderFlags                        =    cmsSetHeaderFlags
+cmsSetHeaderManufacturer                 =    cmsSetHeaderManufacturer
+cmsSetHeaderModel                        =    cmsSetHeaderModel
+cmsSetHeaderProfileID                    =    cmsSetHeaderProfileID
+cmsSetHeaderRenderingIntent              =    cmsSetHeaderRenderingIntent
+cmsSetLogErrorHandler                    =    cmsSetLogErrorHandler
+cmsSetPCS                                =    cmsSetPCS
+cmsSetProfileVersion                     =    cmsSetProfileVersion
+cmsSignalError                           =    cmsSignalError
+cmsSmoothToneCurve                       =    cmsSmoothToneCurve
+cmsstrcasecmp                            =    cmsstrcasecmp
+cmsTempFromWhitePoint                    =    cmsTempFromWhitePoint
+cmsTransform2DeviceLink                  =    cmsTransform2DeviceLink
+cmsUnregisterPlugins                     =    cmsUnregisterPlugins
+_cmsVEC3cross                            =    _cmsVEC3cross
+_cmsVEC3distance                         =    _cmsVEC3distance
+_cmsVEC3dot                              =    _cmsVEC3dot
+_cmsVEC3init                             =    _cmsVEC3init
+_cmsVEC3length                           =    _cmsVEC3length
+_cmsVEC3minus                            =    _cmsVEC3minus
+cmsWhitePointFromTemp                    =    cmsWhitePointFromTemp
+_cmsWrite15Fixed16Number                 =    _cmsWrite15Fixed16Number
+_cmsWriteAlignment                       =    _cmsWriteAlignment
+_cmsWriteFloat32Number                   =    _cmsWriteFloat32Number
+cmsWriteRawTag                           =    cmsWriteRawTag
+cmsWriteTag                              =    cmsWriteTag
+_cmsWriteTypeBase                        =    _cmsWriteTypeBase
+_cmsWriteUInt16Array                     =    _cmsWriteUInt16Array
+_cmsWriteUInt16Number                    =    _cmsWriteUInt16Number
+_cmsWriteUInt32Number                    =    _cmsWriteUInt32Number
+_cmsWriteUInt64Number                    =    _cmsWriteUInt64Number
+_cmsWriteUInt8Number                     =    _cmsWriteUInt8Number
+_cmsWriteXYZNumber                       =    _cmsWriteXYZNumber
+cmsxyY2XYZ                               =    cmsxyY2XYZ
+cmsXYZ2Lab                               =    cmsXYZ2Lab
+cmsXYZ2xyY                               =   cmsXYZ2xyY
+cmsXYZEncoded2Float                      =   cmsXYZEncoded2Float
+cmsSliceSpace16                          =   cmsSliceSpace16
+cmsSliceSpaceFloat                       =   cmsSliceSpaceFloat
+cmsChangeBuffersFormat                   =   cmsChangeBuffersFormat
diff --git a/thirdparty/liblcms2/src/lcms2_internal.h b/thirdparty/liblcms2/src/lcms2_internal.h
new file mode 100644
index 00000000..84e5a80f
--- /dev/null
+++ b/thirdparty/liblcms2/src/lcms2_internal.h
@@ -0,0 +1,652 @@
+
+//
+//  Little Color Management System
+//  Copyright (c) 1998-2010 Marti Maria Saguer
+//
+// Permission is hereby granted, free of charge, to any person obtaining 
+// a copy of this software and associated documentation files (the "Software"), 
+// to deal in the Software without restriction, including without limitation 
+// the rights to use, copy, modify, merge, publish, distribute, sublicense, 
+// and/or sell copies of the Software, and to permit persons to whom the Software 
+// is furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in 
+// all copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 
+// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO 
+// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE 
+// LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 
+// OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 
+// WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+//
+//---------------------------------------------------------------------------------
+//
+
+#ifndef _lcms_internal_H
+
+// Include plug-in foundation
+#ifndef _lcms_plugin_H
+#   include "lcms2_plugin.h"
+#endif
+
+// ctype is part of C99 as per 7.1.2
+#include <ctype.h>
+
+// assert macro is part of C99 as per 7.2
+#include <assert.h>
+
+// Some needed constants
+#ifndef M_PI
+#       define M_PI        3.14159265358979323846
+#endif
+
+#ifndef M_LOG10E
+#       define M_LOG10E    0.434294481903251827651
+#endif
+
+// BorlandC 5.5 is broken on that
+#ifdef __BORLANDC__
+#define sinf(x) (float)sin((float)x)
+#define sqrtf(x) (float)sqrt((float)x)
+#endif
+
+
+// Alignment of ICC file format uses 4 bytes (cmsUInt32Number)
+#define _cmsSIZEOFLONGMINUS1    (sizeof(cmsUInt32Number)-1)
+#define _cmsALIGNLONG(x) (((x)+_cmsSIZEOFLONGMINUS1) & ~(_cmsSIZEOFLONGMINUS1))     
+
+// Maximum encodeable values in floating point
+#define MAX_ENCODEABLE_XYZ  (1.0 + 32767.0/32768.0)
+#define MIN_ENCODEABLE_ab2  (-128.0)
+#define MAX_ENCODEABLE_ab2  ((65535.0/256.0) - 128.0)
+#define MIN_ENCODEABLE_ab4  (-128.0)
+#define MAX_ENCODEABLE_ab4  (127.0)
+
+// Maximum of channels for internal pipeline evaluation
+#define MAX_STAGE_CHANNELS	128
+
+// Unused parameter warning supression
+#define cmsUNUSED_PARAMETER(x) ((void)x) 
+
+// The specification for "inline" is section 6.7.4 of the C99 standard (ISO/IEC 9899:1999).
+// unfortunately VisualC++ does not conform that
+#if defined(_MSC_VER) || defined(__BORLANDC__)
+#   define cmsINLINE __inline
+#else
+#   define cmsINLINE static inline
+#endif
+
+// Other replacement functions
+#ifdef _MSC_VER
+# ifndef snprintf
+#       define snprintf  _snprintf
+# endif
+# ifndef vsnprintf
+#       define vsnprintf  _vsnprintf
+# endif
+#endif
+
+// Pthreads. In windows we use the native WIN32 API instead
+#ifdef CMS_DONT_USE_PTHREADS
+typedef int LCMS_RWLOCK_T;             
+#   define LCMS_CREATE_LOCK(x)       
+#   define LCMS_FREE_LOCK(x)         
+#   define LCMS_READ_LOCK(x)         
+#   define LCMS_WRITE_LOCK(x)        
+#   define LCMS_UNLOCK(x)            
+#else
+#ifdef CMS_IS_WINDOWS_
+#   ifndef WIN32_LEAN_AND_MEAN
+#       define WIN32_LEAN_AND_MEAN
+#   endif
+#   include <windows.h>
+    typedef CRITICAL_SECTION LCMS_RWLOCK_T;
+#   define LCMS_CREATE_LOCK(x)       InitializeCriticalSection((x))
+#   define LCMS_FREE_LOCK(x)         DeleteCriticalSection((x))
+#   define LCMS_READ_LOCK(x)         EnterCriticalSection((x))
+#   define LCMS_WRITE_LOCK(x)        EnterCriticalSection((x))
+#   define LCMS_UNLOCK(x)            LeaveCriticalSection((x))
+#else
+#   include <pthread.h>
+    typedef    pthread_rwlock_t      LCMS_RWLOCK_T;
+#   define LCMS_CREATE_LOCK(x)       pthread_rwlock_init((x), NULL) 
+#   define LCMS_FREE_LOCK(x)         pthread_rwlock_destroy((x))
+#   define LCMS_READ_LOCK(x)         pthread_rwlock_rdlock((x))
+#   define LCMS_WRITE_LOCK(x)        pthread_rwlock_wrlock((x))
+#   define LCMS_UNLOCK(x)            pthread_rwlock_unlock((x))
+#endif
+#endif
+
+// A fast way to convert from/to 16 <-> 8 bits
+#define FROM_8_TO_16(rgb) (cmsUInt16Number) ((((cmsUInt16Number) (rgb)) << 8)|(rgb)) 
+#define FROM_16_TO_8(rgb) (cmsUInt8Number) ((((rgb) * 65281 + 8388608) >> 24) & 0xFF)
+
+// Code analysis is broken on asserts
+#ifdef _MSC_VER
+#    if (_MSC_VER >= 1500)
+#            define _cmsAssert(a)  { assert((a)); __analysis_assume((a)); }
+#     else
+#            define _cmsAssert(a)   assert((a))
+#     endif
+#else
+#      define _cmsAssert(a)   assert((a))
+#endif
+
+//---------------------------------------------------------------------------------
+
+// Determinant lower than that are assumed zero (used on matrix invert)
+#define MATRIX_DET_TOLERANCE    0.0001
+
+//---------------------------------------------------------------------------------
+
+// Fixed point
+#define FIXED_TO_INT(x)         ((x)>>16)
+#define FIXED_REST_TO_INT(x)    ((x)&0xFFFFU)
+#define ROUND_FIXED_TO_INT(x)   (((x)+0x8000)>>16)
+
+cmsINLINE cmsS15Fixed16Number _cmsToFixedDomain(int a)                   { return a + ((a + 0x7fff) / 0xffff); }
+cmsINLINE int                 _cmsFromFixedDomain(cmsS15Fixed16Number a) { return a - ((a + 0x7fff) >> 16); }   
+
+// -----------------------------------------------------------------------------------------------------------
+
+// Fast floor conversion logic. Thanks to Sree Kotay and Stuart Nixon 
+// note than this only works in the range ..-32767...+32767 because 
+// mantissa is interpreted as 15.16 fixed point.
+// The union is to avoid pointer aliasing overoptimization.
+cmsINLINE int _cmsQuickFloor(cmsFloat64Number val)
+{
+#ifdef CMS_DONT_USE_FAST_FLOOR
+    return (int) floor(val);
+#else
+    const cmsFloat64Number _lcms_double2fixmagic = 68719476736.0 * 1.5;  // 2^36 * 1.5, (52-16=36) uses limited precision to floor
+    union {
+        cmsFloat64Number val;
+        int halves[2];
+    } temp;
+    
+    temp.val = val + _lcms_double2fixmagic;
+    
+#ifdef CMS_USE_BIG_ENDIAN
+    return temp.halves[1] >> 16;
+#else
+    return temp.halves[0] >> 16;
+#endif
+#endif
+}
+
+// Fast floor restricted to 0..65535.0
+cmsINLINE cmsUInt16Number _cmsQuickFloorWord(cmsFloat64Number d) 
+{ 
+    return (cmsUInt16Number) _cmsQuickFloor(d - 32767.0) + 32767U; 
+}
+
+// Floor to word, taking care of saturation
+cmsINLINE cmsUInt16Number _cmsQuickSaturateWord(cmsFloat64Number d) 
+{
+    d += 0.5;
+    if (d <= 0) return 0;
+    if (d >= 65535.0) return 0xffff;
+
+    return _cmsQuickFloorWord(d);
+}
+
+// Plug-In registering ---------------------------------------------------------------
+
+// Specialized function for plug-in memory management. No pairing free() since whole pool is freed at once.
+void* _cmsPluginMalloc(cmsUInt32Number size);
+
+// Memory management
+cmsBool   _cmsRegisterMemHandlerPlugin(cmsPluginBase* Plugin);
+
+// Interpolation
+cmsBool  _cmsRegisterInterpPlugin(cmsPluginBase* Plugin);
+
+// Parametric curves
+cmsBool  _cmsRegisterParametricCurvesPlugin(cmsPluginBase* Plugin);
+
+// Formatters management
+cmsBool  _cmsRegisterFormattersPlugin(cmsPluginBase* Plugin);
+
+// Tag type management
+cmsBool  _cmsRegisterTagTypePlugin(cmsPluginBase* Plugin);
+
+// Tag management
+cmsBool  _cmsRegisterTagPlugin(cmsPluginBase* Plugin);
+
+// Intent management
+cmsBool  _cmsRegisterRenderingIntentPlugin(cmsPluginBase* Plugin);
+
+// Multi Process elements
+cmsBool  _cmsRegisterMultiProcessElementPlugin(cmsPluginBase* Plugin);
+
+// Optimization
+cmsBool  _cmsRegisterOptimizationPlugin(cmsPluginBase* Plugin);
+
+
+// ---------------------------------------------------------------------------------------------------------
+
+// Suballocators. Those are blocks of memory that is freed at the end on whole block.
+typedef struct _cmsSubAllocator_chunk_st {
+
+    cmsUInt8Number* Block;
+    cmsUInt32Number BlockSize;
+    cmsUInt32Number Used;
+
+    struct _cmsSubAllocator_chunk_st* next;
+
+} _cmsSubAllocator_chunk;
+
+
+typedef struct {
+
+    cmsContext ContextID;
+    _cmsSubAllocator_chunk* h;
+
+} _cmsSubAllocator;
+
+
+_cmsSubAllocator* _cmsCreateSubAlloc(cmsContext ContextID, cmsUInt32Number Initial);
+void              _cmsSubAllocDestroy(_cmsSubAllocator* s);
+void*             _cmsSubAlloc(_cmsSubAllocator* s, cmsUInt32Number size);
+
+// ----------------------------------------------------------------------------------
+
+// MLU internal representation
+typedef struct {
+
+    cmsUInt16Number Language;
+    cmsUInt16Number Country;   
+
+    cmsUInt32Number StrW;       // Offset to current unicode string
+    cmsUInt32Number Len;        // Lenght in bytes
+
+} _cmsMLUentry;
+
+struct _cms_MLU_struct {
+ 
+    cmsContext ContextID;
+
+    // The directory
+    int AllocatedEntries; 
+    int UsedEntries;
+    _cmsMLUentry* Entries;     // Array of pointers to strings allocated in MemPool
+
+    // The Pool
+    cmsUInt32Number PoolSize;  // The maximum allocated size
+    cmsUInt32Number PoolUsed;  // The used size
+    void*  MemPool;            // Pointer to begin of memory pool  
+};
+
+// Named color list internal representation
+typedef struct {    
+
+    char Name[cmsMAX_PATH];
+    cmsUInt16Number PCS[3];
+    cmsUInt16Number DeviceColorant[cmsMAXCHANNELS];
+
+} _cmsNAMEDCOLOR;
+
+struct _cms_NAMEDCOLORLIST_struct {
+
+    cmsUInt32Number nColors;                
+    cmsUInt32Number Allocated;
+    cmsUInt32Number ColorantCount;  
+
+    char Prefix[33];      // Prefix and suffix are defined to be 32 characters at most
+    char Suffix[33];     
+
+    _cmsNAMEDCOLOR* List;
+
+    cmsContext ContextID;
+};
+
+
+// ----------------------------------------------------------------------------------
+
+// This is the internal struct holding profile details.
+
+// Maximum supported tags in a profile
+#define MAX_TABLE_TAG       100
+
+typedef struct _cms_iccprofile_struct {
+
+    // I/O handler
+    cmsIOHANDLER*            IOhandler;
+
+    // The thread ID
+    cmsContext               ContextID;
+
+    // Creation time
+    struct tm                Created;
+
+    // Only most important items found in ICC profiles   
+    cmsUInt32Number          Version;
+    cmsProfileClassSignature DeviceClass;
+    cmsColorSpaceSignature   ColorSpace;
+    cmsColorSpaceSignature   PCS;
+    cmsUInt32Number          RenderingIntent;
+    cmsUInt32Number          flags;
+    cmsUInt32Number          manufacturer, model;
+    cmsUInt64Number          attributes;
+
+    cmsProfileID             ProfileID;
+
+    // Dictionary
+    cmsUInt32Number          TagCount;
+    cmsTagSignature          TagNames[MAX_TABLE_TAG];
+    cmsTagSignature          TagLinked[MAX_TABLE_TAG];           // The tag to wich is linked (0=none)
+    cmsUInt32Number          TagSizes[MAX_TABLE_TAG];            // Size on disk
+    cmsUInt32Number          TagOffsets[MAX_TABLE_TAG];
+    cmsBool                  TagSaveAsRaw[MAX_TABLE_TAG];        // True to write uncooked
+    void *                   TagPtrs[MAX_TABLE_TAG];
+    cmsTagTypeHandler*       TagTypeHandlers[MAX_TABLE_TAG];     // Same structure may be serialized on different types
+                                                                 // depending on profile version, so we keep track of the                                                             // type handler for each tag in the list.
+    // Special
+    cmsBool                  IsWrite;
+    
+} _cmsICCPROFILE;
+
+// IO helpers for profiles
+cmsBool              _cmsReadHeader(_cmsICCPROFILE* Icc);
+cmsBool              _cmsWriteHeader(_cmsICCPROFILE* Icc, cmsUInt32Number UsedSpace);
+int                  _cmsSearchTag(_cmsICCPROFILE* Icc, cmsTagSignature sig, cmsBool lFollowLinks);
+
+// Tag types
+cmsTagTypeHandler*   _cmsGetTagTypeHandler(cmsTagTypeSignature sig);
+cmsTagTypeSignature  _cmsGetTagTrueType(cmsHPROFILE hProfile, cmsTagSignature sig);
+cmsTagDescriptor*    _cmsGetTagDescriptor(cmsTagSignature sig);
+
+// Error logging ---------------------------------------------------------------------------------------------------------
+
+void                 _cmsTagSignature2String(char String[5], cmsTagSignature sig);
+
+// Interpolation ---------------------------------------------------------------------------------------------------------
+
+cmsInterpParams*     _cmsComputeInterpParams(cmsContext ContextID, int nSamples, int InputChan, int OutputChan, const void* Table, cmsUInt32Number dwFlags);
+cmsInterpParams*     _cmsComputeInterpParamsEx(cmsContext ContextID, const cmsUInt32Number nSamples[], int InputChan, int OutputChan, const void* Table, cmsUInt32Number dwFlags);
+void                 _cmsFreeInterpParams(cmsInterpParams* p);
+cmsBool              _cmsSetInterpolationRoutine(cmsInterpParams* p);
+
+// Curves ----------------------------------------------------------------------------------------------------------------
+
+// This struct holds information about a segment, plus a pointer to the function that implements the evaluation.
+// In the case of table-based, Eval pointer is set to NULL
+
+// The gamma function main structure
+struct _cms_curve_struct {
+
+    cmsInterpParams*  InterpParams;  // Private optimizations for interpolation
+
+    cmsUInt32Number   nSegments;     // Number of segments in the curve. Zero for a 16-bit based tables
+    cmsCurveSegment*  Segments;      // The segments
+    cmsInterpParams** SegInterp;     // Array of private optimizations for interpolation in table-based segments
+
+    cmsParametricCurveEvaluator* Evals;  // Evaluators (one per segment)
+
+    // 16 bit Table-based representation follows    
+    cmsUInt32Number    nEntries;      // Number of table elements
+    cmsUInt16Number*   Table16;       // The table itself. 
+}; 
+
+
+//  Pipelines & Stages ---------------------------------------------------------------------------------------------
+
+// A single stage
+struct _cmsStage_struct {
+    
+    cmsContext          ContextID;
+    
+    cmsStageSignature   Type;           // Identifies the stage
+    cmsStageSignature   Implements;     // Identifies the *function* of the stage (for optimizations)
+
+    cmsUInt32Number     InputChannels;  // Input channels -- for optimization purposes
+    cmsUInt32Number     OutputChannels; // Output channels -- for optimization purposes
+
+    _cmsStageEvalFn     EvalPtr;        // Points to fn that evaluates the stage (always in floating point)
+    _cmsStageDupElemFn  DupElemPtr;     // Points to a fn that duplicates the *data* of the stage
+    _cmsStageFreeElemFn FreePtr;        // Points to a fn that sets the *data* of the stage free
+
+    // A generic pointer to whatever memory needed by the stage
+    void*               Data;
+
+    // Maintains linked list (used internally)
+    struct _cmsStage_struct* Next;
+};
+
+// Data kept in "Element" member of cmsStage
+
+// Curves
+typedef struct {                
+    cmsUInt32Number nCurves;
+    cmsToneCurve**  TheCurves;
+
+} _cmsStageToneCurvesData;
+
+// Matrix
+typedef struct {                
+    cmsFloat64Number*  Double;          // floating point for the matrix
+    cmsFloat64Number*  Offset;          // The offset
+
+} _cmsStageMatrixData;
+
+// CLUT
+typedef struct {                    
+    
+    union {                       // Can have only one of both representations at same time
+        cmsUInt16Number*  T;      // Points to the table 16 bits table
+        cmsFloat32Number* TFloat; // Points to the cmsFloat32Number table
+
+    } Tab;
+
+    cmsInterpParams* Params;
+    cmsUInt32Number  nEntries;
+    cmsBool          HasFloatValues;
+
+} _cmsStageCLutData;
+
+
+// Special Stages (cannot be saved)
+cmsStage*        _cmsStageAllocLab2XYZ(cmsContext ContextID);
+cmsStage*        _cmsStageAllocXYZ2Lab(cmsContext ContextID);
+cmsStage*        _cmsStageAllocLabPrelin(cmsContext ContextID);
+cmsStage*        _cmsStageAllocLabV2ToV4(cmsContext ContextID);
+cmsStage*        _cmsStageAllocLabV2ToV4curves(cmsContext ContextID);
+cmsStage*        _cmsStageAllocLabV4ToV2(cmsContext ContextID);
+cmsStage*        _cmsStageAllocNamedColor(cmsNAMEDCOLORLIST* NamedColorList);
+cmsStage*        _cmsStageAllocIdentityCurves(cmsContext ContextID, int nChannels);
+cmsStage*        _cmsStageAllocIdentityCLut(cmsContext ContextID, int nChan);
+
+// For curve set only  
+cmsToneCurve**     _cmsStageGetPtrToCurveSet(const cmsStage* mpe);
+
+
+// Pipeline Evaluator (in floating point)
+typedef void (* _cmsPipelineEvalFloatFn)(const cmsFloat32Number In[], 
+                                         cmsFloat32Number Out[], 
+                                         const void* Data); 
+
+struct _cmsPipeline_struct {
+
+    cmsStage* Elements;                                // Points to elements chain 
+    cmsUInt32Number InputChannels, OutputChannels;  
+   
+    // Data & evaluators
+    void *Data;
+
+   _cmsOPTeval16Fn         Eval16Fn;
+   _cmsPipelineEvalFloatFn EvalFloatFn;
+   _cmsOPTfreeDataFn       FreeDataFn;
+   _cmsOPTdupDataFn        DupDataFn;
+    
+    cmsContext ContextID;            // Environment
+
+    cmsBool  SaveAs8Bits;            // Implemntation-specific: save as 8 bits if possible
+};
+
+// LUT reading & creation -------------------------------------------------------------------------------------------
+
+// Read tags using low-level function, provide necessary glue code to adapt versions, etc. All those return a brand new copy
+// of the LUTS, since ownership of original is up to the profile. The user should free allocated resources.
+
+cmsPipeline*      _cmsReadInputLUT(cmsHPROFILE hProfile, int Intent);
+cmsPipeline*      _cmsReadOutputLUT(cmsHPROFILE hProfile, int Intent);
+cmsPipeline*      _cmsReadDevicelinkLUT(cmsHPROFILE hProfile, int Intent);
+
+// Special values
+cmsBool           _cmsReadMediaWhitePoint(cmsCIEXYZ* Dest, cmsHPROFILE hProfile);
+cmsBool           _cmsReadCHAD(cmsMAT3* Dest, cmsHPROFILE hProfile);
+
+// Profile linker --------------------------------------------------------------------------------------------------
+
+cmsPipeline* _cmsLinkProfiles(cmsContext         ContextID, 
+                              cmsUInt32Number    nProfiles,
+                              cmsUInt32Number    TheIntents[], 
+                              cmsHPROFILE        hProfiles[], 
+                              cmsBool            BPC[],
+                              cmsFloat64Number   AdaptationStates[],
+                              cmsUInt32Number    dwFlags);
+
+// Sequence --------------------------------------------------------------------------------------------------------
+
+cmsSEQ* _cmsReadProfileSequence(cmsHPROFILE hProfile);
+cmsBool _cmsWriteProfileSequence(cmsHPROFILE hProfile, const cmsSEQ* seq);
+cmsSEQ* _cmsCompileProfileSequence(cmsContext ContextID, cmsUInt32Number nProfiles, cmsHPROFILE hProfiles[]);
+
+
+// LUT optimization ------------------------------------------------------------------------------------------------
+
+cmsUInt16Number  _cmsQuantizeVal(cmsFloat64Number i, int MaxSamples);
+int              _cmsReasonableGridpointsByColorspace(cmsColorSpaceSignature Colorspace, cmsUInt32Number dwFlags);
+
+cmsBool          _cmsEndPointsBySpace(cmsColorSpaceSignature Space, 
+                                      cmsUInt16Number **White, 
+                                      cmsUInt16Number **Black,
+                                      cmsUInt32Number *nOutputs);
+
+cmsBool          _cmsOptimizePipeline(cmsPipeline**    Lut,                                              
+                                      int              Intent,
+                                      cmsUInt32Number* InputFormat, 
+                                      cmsUInt32Number* OutputFormat,
+                                      cmsUInt32Number* dwFlags );
+
+
+// Hi level LUT building ----------------------------------------------------------------------------------------------
+
+cmsPipeline*     _cmsCreateGamutCheckPipeline(cmsContext ContextID,
+                                              cmsHPROFILE hProfiles[], 
+                                              cmsBool  BPC[], 
+                                              cmsUInt32Number Intents[], 
+                                              cmsFloat64Number AdaptationStates[],
+                                              cmsUInt32Number nGamutPCSposition, 
+                                              cmsHPROFILE hGamut);
+
+
+// Formatters ------------------------------------------------------------------------------------------------------------
+
+#define cmsFLAGS_CAN_CHANGE_FORMATTER     0x02000000   // Allow change buffer format
+
+cmsBool         _cmsFormatterIsFloat(cmsUInt32Number Type);
+cmsBool         _cmsFormatterIs8bit(cmsUInt32Number Type);
+
+cmsFormatter    _cmsGetFormatter(cmsUInt32Number Type,          // Specific type, i.e. TYPE_RGB_8
+                                 cmsFormatterDirection Dir, 
+                                 cmsUInt32Number dwFlags);
+
+
+// Transform logic ------------------------------------------------------------------------------------------------------
+
+struct _cmstransform_struct;
+
+// Full xform
+typedef void (* _cmsTransformFn)(struct _cmstransform_struct *Transform,
+                                 const void* InputBuffer,
+                                 void* OutputBuffer, cmsUInt32Number Size);
+
+typedef struct {
+
+    cmsUInt32Number InputFormat, OutputFormat; // Keep formats for further reference
+    cmsUInt32Number StrideIn, StrideOut;       // Planar support
+
+} cmsFormatterInfo;
+
+// Transformation
+typedef struct _cmstransform_struct {
+
+    cmsUInt32Number InputFormat, OutputFormat; // Keep formats for further reference
+
+    // Points to transform code
+    _cmsTransformFn xform;
+
+    // Formatters, cannot be embedded into LUT because cache
+    cmsFormatter16 FromInput;
+    cmsFormatter16 ToOutput;
+
+    cmsFormatterFloat FromInputFloat;
+    cmsFormatterFloat ToOutputFloat;
+    
+    // 1-pixel cache (16 bits only)
+    cmsUInt16Number CacheIn[cmsMAXCHANNELS];
+    cmsUInt16Number CacheOut[cmsMAXCHANNELS];
+
+    // Semaphor for cache
+    LCMS_RWLOCK_T rwlock;
+    
+    // A MPE LUT holding the full (optimized) transform
+    cmsPipeline* Lut;
+    
+    // A MPE LUT holding the gamut check. It goes from the input space to bilevel
+    cmsPipeline* GamutCheck;
+
+    // Colorant tables
+    cmsNAMEDCOLORLIST* InputColorant;       // Input Colorant table
+    cmsNAMEDCOLORLIST* OutputColorant;      // Colorant table (for n chans > CMYK)
+
+    // Informational only
+    cmsColorSpaceSignature EntryColorSpace;
+    cmsColorSpaceSignature ExitColorSpace;
+    
+    // Profiles used to create the transform
+    cmsSEQ* Sequence;
+
+    cmsUInt32Number  dwOriginalFlags;      
+    cmsFloat64Number AdaptationState;              
+
+    // The intent of this transform. That is usually the last intent in the profilechain, but may differ
+    cmsUInt32Number RenderingIntent;
+
+    // An id that uniquely identifies the running context. May be null.
+    cmsContext ContextID;
+
+} _cmsTRANSFORM;
+
+// --------------------------------------------------------------------------------------------------
+
+cmsHTRANSFORM _cmsChain2Lab(cmsContext             ContextID,
+                            cmsUInt32Number        nProfiles,
+                            cmsUInt32Number        InputFormat,
+                            cmsUInt32Number        OutputFormat,
+                            const cmsUInt32Number  Intents[], 
+                            const cmsHPROFILE      hProfiles[], 
+                            const cmsBool          BPC[],
+                            const cmsFloat64Number AdaptationStates[],
+                            cmsUInt32Number        dwFlags);
+
+
+cmsToneCurve* _cmsBuildKToneCurve(cmsContext       ContextID, 
+                            cmsUInt32Number        nPoints,
+                            cmsUInt32Number        nProfiles,
+                            const cmsUInt32Number  Intents[], 
+                            const cmsHPROFILE      hProfiles[], 
+                            const cmsBool          BPC[],
+                            const cmsFloat64Number AdaptationStates[],
+                            cmsUInt32Number        dwFlags);
+
+cmsBool   _cmsAdaptationMatrix(cmsMAT3* r, const cmsMAT3* ConeMatrix, const cmsCIEXYZ* FromIll, const cmsCIEXYZ* ToIll);
+
+cmsBool   _cmsBuildRGB2XYZtransferMatrix(cmsMAT3* r, const cmsCIExyY* WhitePoint, const cmsCIExyYTRIPLE* Primaries);
+
+
+#define _lcms_internal_H
+#endif