#include "compose.hpp"
+#include "dcpomatic_assert.h"
#include "dcpomatic_socket.h"
+#include "enum_indexed_vector.h"
#include "exceptions.h"
#include "image.h"
+#include "maths_util.h"
+#include "memory_util.h"
#include "rect.h"
#include "timer.h"
-#include "util.h"
-#include "warnings.h"
#include <dcp/rgb_xyz.h>
#include <dcp/transfer_function.h>
-DCPOMATIC_DISABLE_WARNINGS
+#include <dcp/warnings.h>
+LIBDCP_DISABLE_WARNINGS
extern "C" {
#include <libavutil/frame.h>
#include <libavutil/pixdesc.h>
#include <libavutil/pixfmt.h>
#include <libswscale/swscale.h>
}
-DCPOMATIC_ENABLE_WARNINGS
-#include <png.h>
+LIBDCP_ENABLE_WARNINGS
#if HAVE_VALGRIND_MEMCHECK_H
#include <valgrind/memcheck.h>
#endif
using dcp::Size;
-/** The memory alignment, in bytes, used for each row of an image if aligment is requested */
-#define ALIGNMENT 64
+/** The memory alignment, in bytes, used for each row of an image if Alignment::PADDED is requested */
+int constexpr ALIGNMENT = 64;
/* U/V black value for 8-bit colour */
static uint8_t const eight_bit_uv = (1 << 7) - 1;
return lrintf(powf(2.0f, d->log2_chroma_w));
}
+
/** @param n Component index.
* @return Number of samples (i.e. pixels, unless sub-sampled) in each direction for this component.
*/
);
}
+
/** @return Number of planes */
int
Image::planes () const
{
+ if (_pixel_format == AV_PIX_FMT_PAL8) {
+ return 2;
+ }
+
auto d = av_pix_fmt_desc_get(_pixel_format);
if (!d) {
throw PixelFormatError ("planes()", _pixel_format);
}
- if (_pixel_format == AV_PIX_FMT_PAL8) {
- return 2;
- }
-
if ((d->flags & AV_PIX_FMT_FLAG_PLANAR) == 0) {
return 1;
}
VideoRange video_range,
AVPixelFormat out_format,
VideoRange out_video_range,
- bool out_aligned,
+ Alignment out_alignment,
bool fast
) const
{
/* Empirical testing suggests that sws_scale() will crash if
- the input image is not aligned.
+ the input image is not padded.
*/
- DCPOMATIC_ASSERT (aligned ());
+ DCPOMATIC_ASSERT (alignment() == Alignment::PADDED);
DCPOMATIC_ASSERT (out_size.width >= inter_size.width);
DCPOMATIC_ASSERT (out_size.height >= inter_size.height);
- auto out = make_shared<Image>(out_format, out_size, out_aligned);
+ auto out = make_shared<Image>(out_format, out_size, out_alignment);
out->make_black ();
auto in_desc = av_pix_fmt_desc_get (_pixel_format);
}
DCPOMATIC_ASSERT (yuv_to_rgb < dcp::YUVToRGB::COUNT);
- int const lut[static_cast<int>(dcp::YUVToRGB::COUNT)] = {
- SWS_CS_ITU601,
- SWS_CS_ITU709
- };
+ EnumIndexedVector<int, dcp::YUVToRGB> lut;
+ lut[dcp::YUVToRGB::REC601] = SWS_CS_ITU601;
+ lut[dcp::YUVToRGB::REC709] = SWS_CS_ITU709;
+ lut[dcp::YUVToRGB::REC2020] = SWS_CS_BT2020;
/* The 3rd parameter here is:
0 -> source range MPEG (i.e. "video", 16-235)
*/
sws_setColorspaceDetails (
scale_context,
- sws_getCoefficients (lut[static_cast<int>(yuv_to_rgb)]), video_range == VideoRange::VIDEO ? 0 : 1,
- sws_getCoefficients (lut[static_cast<int>(yuv_to_rgb)]), out_video_range == VideoRange::VIDEO ? 0 : 1,
+ sws_getCoefficients(lut[yuv_to_rgb]), video_range == VideoRange::VIDEO ? 0 : 1,
+ sws_getCoefficients(lut[yuv_to_rgb]), out_video_range == VideoRange::VIDEO ? 0 : 1,
0, 1 << 16, 1 << 16
);
sws_freeContext (scale_context);
- if (corrected_crop != Crop() && cropped_size == inter_size) {
- /* We are cropping without any scaling or pixel format conversion, so FFmpeg may have left some
- data behind in our image. Clear it out. It may get to the point where we should just stop
- trying to be clever with cropping.
- */
- out->make_part_black (corner.x + cropped_size.width, out_size.width - cropped_size.width);
+ /* There are some cases where there will be unwanted image data left in the image at this point:
+ *
+ * 1. When we are cropping without any scaling or pixel format conversion.
+ * 2. When we are scaling to certain sizes and placing the result into a larger
+ * black frame.
+ *
+ * Clear out the sides of the image to take care of those cases.
+ */
+ auto const pad = (out_size.width - inter_size.width) / 2;
+ out->make_part_black(0, pad);
+ out->make_part_black(corner.x + inter_size.width, pad);
+
+ if (
+ video_range == VideoRange::VIDEO &&
+ out_video_range == VideoRange::FULL &&
+ av_pix_fmt_desc_get(_pixel_format)->flags & AV_PIX_FMT_FLAG_RGB
+ ) {
+ /* libswscale will not convert video range for RGB sources, so we have to do it ourselves */
+ out->video_range_to_full_range ();
}
return out;
}
+
shared_ptr<Image>
-Image::convert_pixel_format (dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_format, bool out_aligned, bool fast) const
+Image::convert_pixel_format (dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_format, Alignment out_alignment, bool fast) const
{
- return scale(size(), yuv_to_rgb, out_format, out_aligned, fast);
+ return scale(size(), yuv_to_rgb, out_format, out_alignment, fast);
}
+
/** @param out_size Size to scale to.
* @param yuv_to_rgb YUVToRGB transform transform to use, if required.
* @param out_format Output pixel format.
- * @param out_aligned true to make an aligned output image.
+ * @param out_alignment Output alignment.
* @param fast Try to be fast at the possible expense of quality; at present this means using
* fast bilinear rather than bicubic scaling.
*/
shared_ptr<Image>
-Image::scale (dcp::Size out_size, dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_format, bool out_aligned, bool fast) const
+Image::scale (dcp::Size out_size, dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_format, Alignment out_alignment, bool fast) const
{
/* Empirical testing suggests that sws_scale() will crash if
- the input image is not aligned.
+ the input image alignment is not PADDED.
*/
- DCPOMATIC_ASSERT (aligned ());
+ DCPOMATIC_ASSERT (alignment() == Alignment::PADDED);
- auto scaled = make_shared<Image>(out_format, out_size, out_aligned);
+ auto scaled = make_shared<Image>(out_format, out_size, out_alignment);
auto scale_context = sws_getContext (
size().width, size().height, pixel_format(),
out_size.width, out_size.height, out_format,
);
DCPOMATIC_ASSERT (yuv_to_rgb < dcp::YUVToRGB::COUNT);
- int const lut[static_cast<int>(dcp::YUVToRGB::COUNT)] = {
- SWS_CS_ITU601,
- SWS_CS_ITU709
- };
+ EnumIndexedVector<int, dcp::YUVToRGB> lut;
+ lut[dcp::YUVToRGB::REC601] = SWS_CS_ITU601;
+ lut[dcp::YUVToRGB::REC709] = SWS_CS_ITU709;
+ lut[dcp::YUVToRGB::REC2020] = SWS_CS_BT2020;
/* The 3rd parameter here is:
0 -> source range MPEG (i.e. "video", 16-235)
*/
sws_setColorspaceDetails (
scale_context,
- sws_getCoefficients (lut[static_cast<int>(yuv_to_rgb)]), 0,
- sws_getCoefficients (lut[static_cast<int>(yuv_to_rgb)]), 0,
+ sws_getCoefficients(lut[yuv_to_rgb]), 0,
+ sws_getCoefficients(lut[yuv_to_rgb]), 0,
0, 1 << 16, 1 << 16
);
return scaled;
}
+
/** Blacken a YUV image whose bits per pixel is rounded up to 16 */
void
Image::yuv_16_black (uint16_t v, bool alpha)
}
}
+
uint16_t
Image::swap_16 (uint16_t v)
{
return ((v >> 8) & 0xff) | ((v & 0xff) << 8);
}
+
void
Image::make_part_black (int const start, int const width)
{
}
break;
}
+ case AV_PIX_FMT_YUV444P10LE:
+ {
+ y_part();
+ for (int i = 1; i < 3; ++i) {
+ auto p = reinterpret_cast<int16_t*>(data()[i]);
+ int const h = sample_size(i).height;
+ for (int y = 0; y < h; ++y) {
+ for (int x = start; x < (start + width); ++x) {
+ p[x] = ten_bit_uv;
+ }
+ p += stride()[i] / 2;
+ }
+ }
+ break;
+ }
default:
throw PixelFormatError ("make_part_black()", _pixel_format);
}
}
+
void
Image::make_black ()
{
}
}
+
void
Image::make_transparent ()
{
- if (_pixel_format != AV_PIX_FMT_BGRA && _pixel_format != AV_PIX_FMT_RGBA) {
+ if (_pixel_format != AV_PIX_FMT_BGRA && _pixel_format != AV_PIX_FMT_RGBA && _pixel_format != AV_PIX_FMT_RGBA64BE) {
throw PixelFormatError ("make_transparent()", _pixel_format);
}
memset (data()[0], 0, sample_size(0).height * stride()[0]);
}
+
+struct TargetParams
+{
+ int start_x;
+ int start_y;
+ dcp::Size size;
+ uint8_t* const* data;
+ int const* stride;
+ int bpp;
+
+ uint8_t* line_pointer(int y) const {
+ return data[0] + y * stride[0] + start_x * bpp;
+ }
+};
+
+
+struct OtherParams
+{
+ int start_x;
+ int start_y;
+ dcp::Size size;
+ uint8_t* const* data;
+ int const* stride;
+ int bpp;
+
+ uint8_t* line_pointer(int y) const {
+ return data[0] + y * stride[0];
+ }
+
+ float alpha_divisor() const {
+ return pow(2, bpp * 2) - 1;
+ }
+};
+
+
+template <class OtherType>
+void
+alpha_blend_onto_rgb24(TargetParams const& target, OtherParams const& other, int red, int blue, std::function<float (OtherType*)> get, int value_divisor)
+{
+ /* Going onto RGB24. First byte is red, second green, third blue */
+ auto const alpha_divisor = other.alpha_divisor();
+ for (int ty = target.start_y, oy = other.start_y; ty < target.size.height && oy < other.size.height; ++ty, ++oy) {
+ auto tp = target.line_pointer(ty);
+ auto op = reinterpret_cast<OtherType*>(other.line_pointer(oy));
+ for (int tx = target.start_x, ox = other.start_x; tx < target.size.width && ox < other.size.width; ++tx, ++ox) {
+ float const alpha = get(op + 3) / alpha_divisor;
+ tp[0] = (get(op + red) / value_divisor) * alpha + tp[0] * (1 - alpha);
+ tp[1] = (get(op + 1) / value_divisor) * alpha + tp[1] * (1 - alpha);
+ tp[2] = (get(op + blue) / value_divisor) * alpha + tp[2] * (1 - alpha);
+
+ tp += target.bpp;
+ op += other.bpp / sizeof(OtherType);
+ }
+ }
+}
+
+
+template <class OtherType>
+void
+alpha_blend_onto_bgra(TargetParams const& target, OtherParams const& other, int red, int blue, std::function<float (OtherType*)> get, int value_divisor)
+{
+ auto const alpha_divisor = other.alpha_divisor();
+ for (int ty = target.start_y, oy = other.start_y; ty < target.size.height && oy < other.size.height; ++ty, ++oy) {
+ auto tp = target.line_pointer(ty);
+ auto op = reinterpret_cast<OtherType*>(other.line_pointer(oy));
+ for (int tx = target.start_x, ox = other.start_x; tx < target.size.width && ox < other.size.width; ++tx, ++ox) {
+ float const alpha = get(op + 3) / alpha_divisor;
+ tp[0] = (get(op + blue) / value_divisor) * alpha + tp[0] * (1 - alpha);
+ tp[1] = (get(op + 1) / value_divisor) * alpha + tp[1] * (1 - alpha);
+ tp[2] = (get(op + red) / value_divisor) * alpha + tp[2] * (1 - alpha);
+ tp[3] = (get(op + 3) / value_divisor) * alpha + tp[3] * (1 - alpha);
+
+ tp += target.bpp;
+ op += other.bpp / sizeof(OtherType);
+ }
+ }
+}
+
+
+template <class OtherType>
+void
+alpha_blend_onto_rgba(TargetParams const& target, OtherParams const& other, int red, int blue, std::function<float (OtherType*)> get, int value_divisor)
+{
+ auto const alpha_divisor = other.alpha_divisor();
+ for (int ty = target.start_y, oy = other.start_y; ty < target.size.height && oy < other.size.height; ++ty, ++oy) {
+ auto tp = target.line_pointer(ty);
+ auto op = reinterpret_cast<OtherType*>(other.line_pointer(oy));
+ for (int tx = target.start_x, ox = other.start_x; tx < target.size.width && ox < other.size.width; ++tx, ++ox) {
+ float const alpha = get(op + 3) / alpha_divisor;
+ tp[0] = (get(op + red) / value_divisor) * alpha + tp[0] * (1 - alpha);
+ tp[1] = (get(op + 1) / value_divisor) * alpha + tp[1] * (1 - alpha);
+ tp[2] = (get(op + blue) / value_divisor) * alpha + tp[2] * (1 - alpha);
+ tp[3] = (get(op + 3) / value_divisor) * alpha + tp[3] * (1 - alpha);
+
+ tp += target.bpp;
+ op += other.bpp / sizeof(OtherType);
+ }
+ }
+}
+
+
+template <class OtherType>
+void
+alpha_blend_onto_rgb48le(TargetParams const& target, OtherParams const& other, int red, int blue, std::function<float (OtherType*)> get, int value_scale)
+{
+ auto const alpha_divisor = other.alpha_divisor();
+ for (int ty = target.start_y, oy = other.start_y; ty < target.size.height && oy < other.size.height; ++ty, ++oy) {
+ auto tp = reinterpret_cast<uint16_t*>(target.line_pointer(ty));
+ auto op = reinterpret_cast<OtherType*>(other.line_pointer(oy));
+ for (int tx = target.start_x, ox = other.start_x; tx < target.size.width && ox < other.size.width; ++tx, ++ox) {
+ float const alpha = get(op + 3) / alpha_divisor;
+ tp[0] = get(op + red) * value_scale * alpha + tp[0] * (1 - alpha);
+ tp[1] = get(op + 1) * value_scale * alpha + tp[1] * (1 - alpha);
+ tp[2] = get(op + blue) * value_scale * alpha + tp[2] * (1 - alpha);
+
+ tp += target.bpp / 2;
+ op += other.bpp / sizeof(OtherType);
+ }
+ }
+}
+
+
+template <class OtherType>
+void
+alpha_blend_onto_xyz12le(TargetParams const& target, OtherParams const& other, int red, int blue, std::function<float (OtherType*)> get, int value_divisor)
+{
+ auto const alpha_divisor = other.alpha_divisor();
+ auto conv = dcp::ColourConversion::srgb_to_xyz();
+ double fast_matrix[9];
+ dcp::combined_rgb_to_xyz(conv, fast_matrix);
+ auto lut_in = conv.in()->double_lut(0, 1, 8, false);
+ auto lut_out = conv.out()->int_lut(0, 1, 16, true, 65535);
+ for (int ty = target.start_y, oy = other.start_y; ty < target.size.height && oy < other.size.height; ++ty, ++oy) {
+ auto tp = reinterpret_cast<uint16_t*>(target.data[0] + ty * target.stride[0] + target.start_x * target.bpp);
+ auto op = reinterpret_cast<OtherType*>(other.data[0] + oy * other.stride[0]);
+ for (int tx = target.start_x, ox = other.start_x; tx < target.size.width && ox < other.size.width; ++tx, ++ox) {
+ float const alpha = get(op + 3) / alpha_divisor;
+
+ /* Convert sRGB to XYZ; op is BGRA. First, input gamma LUT */
+ double const r = lut_in[get(op + red) / value_divisor];
+ double const g = lut_in[get(op + 1) / value_divisor];
+ double const b = lut_in[get(op + blue) / value_divisor];
+
+ /* RGB to XYZ, including Bradford transform and DCI companding */
+ double const x = max(0.0, min(1.0, r * fast_matrix[0] + g * fast_matrix[1] + b * fast_matrix[2]));
+ double const y = max(0.0, min(1.0, r * fast_matrix[3] + g * fast_matrix[4] + b * fast_matrix[5]));
+ double const z = max(0.0, min(1.0, r * fast_matrix[6] + g * fast_matrix[7] + b * fast_matrix[8]));
+
+ /* Out gamma LUT and blend */
+ tp[0] = lut_out[lrint(x * 65535)] * alpha + tp[0] * (1 - alpha);
+ tp[1] = lut_out[lrint(y * 65535)] * alpha + tp[1] * (1 - alpha);
+ tp[2] = lut_out[lrint(z * 65535)] * alpha + tp[2] * (1 - alpha);
+
+ tp += target.bpp / 2;
+ op += other.bpp / sizeof(OtherType);
+ }
+ }
+}
+
+
+static
+void
+alpha_blend_onto_yuv420p(TargetParams const& target, OtherParams const& other, uint8_t* const* alpha_data, int const* alpha_stride)
+{
+ auto const ts = target.size;
+ auto const os = other.size;
+ for (int ty = target.start_y, oy = other.start_y; ty < ts.height && oy < os.height; ++ty, ++oy) {
+ int const hty = ty / 2;
+ int const hoy = oy / 2;
+ uint8_t* tY = target.data[0] + (ty * target.stride[0]) + target.start_x;
+ uint8_t* tU = target.data[1] + (hty * target.stride[1]) + target.start_x / 2;
+ uint8_t* tV = target.data[2] + (hty * target.stride[2]) + target.start_x / 2;
+ uint8_t* oY = other.data[0] + (oy * other.stride[0]) + other.start_x;
+ uint8_t* oU = other.data[1] + (hoy * other.stride[1]) + other.start_x / 2;
+ uint8_t* oV = other.data[2] + (hoy * other.stride[2]) + other.start_x / 2;
+ uint8_t* alpha = alpha_data[0] + (oy * alpha_stride[0]) + other.start_x * 4;
+ for (int tx = target.start_x, ox = other.start_x; tx < ts.width && ox < os.width; ++tx, ++ox) {
+ float const a = float(alpha[3]) / 255;
+ *tY = *oY * a + *tY * (1 - a);
+ *tU = *oU * a + *tU * (1 - a);
+ *tV = *oV * a + *tV * (1 - a);
+ ++tY;
+ ++oY;
+ if (tx % 2) {
+ ++tU;
+ ++tV;
+ }
+ if (ox % 2) {
+ ++oU;
+ ++oV;
+ }
+ alpha += 4;
+ }
+ }
+}
+
+
+static
+void
+alpha_blend_onto_yuv420p10(TargetParams const& target, OtherParams const& other, uint8_t* const* alpha_data, int const* alpha_stride)
+{
+ auto const ts = target.size;
+ auto const os = other.size;
+ for (int ty = target.start_y, oy = other.start_y; ty < ts.height && oy < os.height; ++ty, ++oy) {
+ int const hty = ty / 2;
+ int const hoy = oy / 2;
+ uint16_t* tY = reinterpret_cast<uint16_t*>(target.data[0] + (ty * target.stride[0])) + target.start_x;
+ uint16_t* tU = reinterpret_cast<uint16_t*>(target.data[1] + (hty * target.stride[1])) + target.start_x / 2;
+ uint16_t* tV = reinterpret_cast<uint16_t*>(target.data[2] + (hty * target.stride[2])) + target.start_x / 2;
+ uint16_t* oY = reinterpret_cast<uint16_t*>(other.data[0] + (oy * other.stride[0])) + other.start_x;
+ uint16_t* oU = reinterpret_cast<uint16_t*>(other.data[1] + (hoy * other.stride[1])) + other.start_x / 2;
+ uint16_t* oV = reinterpret_cast<uint16_t*>(other.data[2] + (hoy * other.stride[2])) + other.start_x / 2;
+ uint8_t* alpha = alpha_data[0] + (oy * alpha_stride[0]) + other.start_x * 4;
+ for (int tx = target.start_x, ox = other.start_x; tx < ts.width && ox < os.width; ++tx, ++ox) {
+ float const a = float(alpha[3]) / 255;
+ *tY = *oY * a + *tY * (1 - a);
+ *tU = *oU * a + *tU * (1 - a);
+ *tV = *oV * a + *tV * (1 - a);
+ ++tY;
+ ++oY;
+ if (tx % 2) {
+ ++tU;
+ ++tV;
+ }
+ if (ox % 2) {
+ ++oU;
+ ++oV;
+ }
+ alpha += 4;
+ }
+ }
+}
+
+
+static
+void
+alpha_blend_onto_yuv422p10le(TargetParams const& target, OtherParams const& other, uint8_t* const* alpha_data, int const* alpha_stride)
+{
+ auto const ts = target.size;
+ auto const os = other.size;
+ for (int ty = target.start_y, oy = other.start_y; ty < ts.height && oy < os.height; ++ty, ++oy) {
+ uint16_t* tY = reinterpret_cast<uint16_t*>(target.data[0] + (ty * target.stride[0])) + target.start_x;
+ uint16_t* tU = reinterpret_cast<uint16_t*>(target.data[1] + (ty * target.stride[1])) + target.start_x / 2;
+ uint16_t* tV = reinterpret_cast<uint16_t*>(target.data[2] + (ty * target.stride[2])) + target.start_x / 2;
+ uint16_t* oY = reinterpret_cast<uint16_t*>(other.data[0] + (oy * other.stride[0])) + other.start_x;
+ uint16_t* oU = reinterpret_cast<uint16_t*>(other.data[1] + (oy * other.stride[1])) + other.start_x / 2;
+ uint16_t* oV = reinterpret_cast<uint16_t*>(other.data[2] + (oy * other.stride[2])) + other.start_x / 2;
+ uint8_t* alpha = alpha_data[0] + (oy * alpha_stride[0]) + other.start_x * 4;
+ for (int tx = target.start_x, ox = other.start_x; tx < ts.width && ox < os.width; ++tx, ++ox) {
+ float const a = float(alpha[3]) / 255;
+ *tY = *oY * a + *tY * (1 - a);
+ *tU = *oU * a + *tU * (1 - a);
+ *tV = *oV * a + *tV * (1 - a);
+ ++tY;
+ ++oY;
+ if (tx % 2) {
+ ++tU;
+ ++tV;
+ }
+ if (ox % 2) {
+ ++oU;
+ ++oV;
+ }
+ alpha += 4;
+ }
+ }
+}
+
+
void
Image::alpha_blend (shared_ptr<const Image> other, Position<int> position)
{
- /* We're blending RGBA or BGRA images */
- DCPOMATIC_ASSERT (other->pixel_format() == AV_PIX_FMT_BGRA || other->pixel_format() == AV_PIX_FMT_RGBA);
+ DCPOMATIC_ASSERT(
+ other->pixel_format() == AV_PIX_FMT_BGRA ||
+ other->pixel_format() == AV_PIX_FMT_RGBA ||
+ other->pixel_format() == AV_PIX_FMT_RGBA64BE
+ );
+
int const blue = other->pixel_format() == AV_PIX_FMT_BGRA ? 0 : 2;
int const red = other->pixel_format() == AV_PIX_FMT_BGRA ? 2 : 0;
- int const other_bpp = 4;
-
int start_tx = position.x;
int start_ox = 0;
start_ty = 0;
}
+ TargetParams target_params = {
+ start_tx,
+ start_ty,
+ size(),
+ data(),
+ stride(),
+ 0
+ };
+
+ OtherParams other_params = {
+ start_ox,
+ start_oy,
+ other->size(),
+ other->data(),
+ other->stride(),
+ other->pixel_format() == AV_PIX_FMT_RGBA64BE ? 8 : 4
+ };
+
+ auto byteswap = [](uint16_t* p) {
+ return (*p >> 8) | ((*p & 0xff) << 8);
+ };
+
+ auto pass = [](uint8_t* p) {
+ return *p;
+ };
+
switch (_pixel_format) {
case AV_PIX_FMT_RGB24:
- {
- /* Going onto RGB24. First byte is red, second green, third blue */
- int const this_bpp = 3;
- for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
- uint8_t* tp = data()[0] + ty * stride()[0] + start_tx * this_bpp;
- uint8_t* op = other->data()[0] + oy * other->stride()[0];
- for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
- float const alpha = float (op[3]) / 255;
- tp[0] = op[red] * alpha + tp[0] * (1 - alpha);
- tp[1] = op[1] * alpha + tp[1] * (1 - alpha);
- tp[2] = op[blue] * alpha + tp[2] * (1 - alpha);
-
- tp += this_bpp;
- op += other_bpp;
- }
+ target_params.bpp = 3;
+ if (other->pixel_format() == AV_PIX_FMT_RGBA64BE) {
+ alpha_blend_onto_rgb24<uint16_t>(target_params, other_params, red, blue, byteswap, 256);
+ } else {
+ alpha_blend_onto_rgb24<uint8_t>(target_params, other_params, red, blue, pass, 1);
}
break;
- }
case AV_PIX_FMT_BGRA:
- {
- int const this_bpp = 4;
- for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
- uint8_t* tp = data()[0] + ty * stride()[0] + start_tx * this_bpp;
- uint8_t* op = other->data()[0] + oy * other->stride()[0];
- for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
- float const alpha = float (op[3]) / 255;
- tp[0] = op[blue] * alpha + tp[0] * (1 - alpha);
- tp[1] = op[1] * alpha + tp[1] * (1 - alpha);
- tp[2] = op[red] * alpha + tp[2] * (1 - alpha);
- tp[3] = op[3] * alpha + tp[3] * (1 - alpha);
-
- tp += this_bpp;
- op += other_bpp;
- }
+ target_params.bpp = 4;
+ if (other->pixel_format() == AV_PIX_FMT_RGBA64BE) {
+ alpha_blend_onto_bgra<uint16_t>(target_params, other_params, red, blue, byteswap, 256);
+ } else {
+ alpha_blend_onto_bgra<uint8_t>(target_params, other_params, red, blue, pass, 1);
}
break;
- }
case AV_PIX_FMT_RGBA:
- {
- int const this_bpp = 4;
- for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
- uint8_t* tp = data()[0] + ty * stride()[0] + start_tx * this_bpp;
- uint8_t* op = other->data()[0] + oy * other->stride()[0];
- for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
- float const alpha = float (op[3]) / 255;
- tp[0] = op[red] * alpha + tp[0] * (1 - alpha);
- tp[1] = op[1] * alpha + tp[1] * (1 - alpha);
- tp[2] = op[blue] * alpha + tp[2] * (1 - alpha);
- tp[3] = op[3] * alpha + tp[3] * (1 - alpha);
-
- tp += this_bpp;
- op += other_bpp;
- }
+ target_params.bpp = 4;
+ if (other->pixel_format() == AV_PIX_FMT_RGBA64BE) {
+ alpha_blend_onto_rgba<uint16_t>(target_params, other_params, red, blue, byteswap, 256);
+ } else {
+ alpha_blend_onto_rgba<uint8_t>(target_params, other_params, red, blue, pass, 1);
}
break;
- }
case AV_PIX_FMT_RGB48LE:
- {
- int const this_bpp = 6;
- for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
- uint8_t* tp = data()[0] + ty * stride()[0] + start_tx * this_bpp;
- uint8_t* op = other->data()[0] + oy * other->stride()[0];
- for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
- float const alpha = float (op[3]) / 255;
- /* Blend high bytes */
- tp[1] = op[red] * alpha + tp[1] * (1 - alpha);
- tp[3] = op[1] * alpha + tp[3] * (1 - alpha);
- tp[5] = op[blue] * alpha + tp[5] * (1 - alpha);
-
- tp += this_bpp;
- op += other_bpp;
- }
+ target_params.bpp = 6;
+ if (other->pixel_format() == AV_PIX_FMT_RGBA64BE) {
+ alpha_blend_onto_rgb48le<uint16_t>(target_params, other_params, red, blue, byteswap, 1);
+ } else {
+ alpha_blend_onto_rgb48le<uint8_t>(target_params, other_params, red, blue, pass, 256);
}
break;
- }
case AV_PIX_FMT_XYZ12LE:
- {
- auto conv = dcp::ColourConversion::srgb_to_xyz();
- double fast_matrix[9];
- dcp::combined_rgb_to_xyz (conv, fast_matrix);
- double const * lut_in = conv.in()->lut (8, false);
- double const * lut_out = conv.out()->lut (16, true);
- int const this_bpp = 6;
- for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
- uint16_t* tp = reinterpret_cast<uint16_t*> (data()[0] + ty * stride()[0] + start_tx * this_bpp);
- uint8_t* op = other->data()[0] + oy * other->stride()[0];
- for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
- float const alpha = float (op[3]) / 255;
-
- /* Convert sRGB to XYZ; op is BGRA. First, input gamma LUT */
- double const r = lut_in[op[red]];
- double const g = lut_in[op[1]];
- double const b = lut_in[op[blue]];
-
- /* RGB to XYZ, including Bradford transform and DCI companding */
- double const x = max (0.0, min (65535.0, r * fast_matrix[0] + g * fast_matrix[1] + b * fast_matrix[2]));
- double const y = max (0.0, min (65535.0, r * fast_matrix[3] + g * fast_matrix[4] + b * fast_matrix[5]));
- double const z = max (0.0, min (65535.0, r * fast_matrix[6] + g * fast_matrix[7] + b * fast_matrix[8]));
-
- /* Out gamma LUT and blend */
- tp[0] = lrint(lut_out[lrint(x)] * 65535) * alpha + tp[0] * (1 - alpha);
- tp[1] = lrint(lut_out[lrint(y)] * 65535) * alpha + tp[1] * (1 - alpha);
- tp[2] = lrint(lut_out[lrint(z)] * 65535) * alpha + tp[2] * (1 - alpha);
-
- tp += this_bpp / 2;
- op += other_bpp;
- }
+ target_params.bpp = 6;
+ if (other->pixel_format() == AV_PIX_FMT_RGBA64BE) {
+ alpha_blend_onto_xyz12le<uint16_t>(target_params, other_params, red, blue, byteswap, 256);
+ } else {
+ alpha_blend_onto_xyz12le<uint8_t>(target_params, other_params, red, blue, pass, 1);
}
break;
- }
case AV_PIX_FMT_YUV420P:
{
- auto yuv = other->convert_pixel_format (dcp::YUVToRGB::REC709, _pixel_format, false, false);
- dcp::Size const ts = size();
- dcp::Size const os = yuv->size();
- for (int ty = start_ty, oy = start_oy; ty < ts.height && oy < os.height; ++ty, ++oy) {
- int const hty = ty / 2;
- int const hoy = oy / 2;
- uint8_t* tY = data()[0] + (ty * stride()[0]) + start_tx;
- uint8_t* tU = data()[1] + (hty * stride()[1]) + start_tx / 2;
- uint8_t* tV = data()[2] + (hty * stride()[2]) + start_tx / 2;
- uint8_t* oY = yuv->data()[0] + (oy * yuv->stride()[0]) + start_ox;
- uint8_t* oU = yuv->data()[1] + (hoy * yuv->stride()[1]) + start_ox / 2;
- uint8_t* oV = yuv->data()[2] + (hoy * yuv->stride()[2]) + start_ox / 2;
- uint8_t* alpha = other->data()[0] + (oy * other->stride()[0]) + start_ox * 4;
- for (int tx = start_tx, ox = start_ox; tx < ts.width && ox < os.width; ++tx, ++ox) {
- float const a = float(alpha[3]) / 255;
- *tY = *oY * a + *tY * (1 - a);
- *tU = *oU * a + *tU * (1 - a);
- *tV = *oV * a + *tV * (1 - a);
- ++tY;
- ++oY;
- if (tx % 2) {
- ++tU;
- ++tV;
- }
- if (ox % 2) {
- ++oU;
- ++oV;
- }
- alpha += 4;
- }
- }
+ auto yuv = other->convert_pixel_format (dcp::YUVToRGB::REC709, _pixel_format, Alignment::COMPACT, false);
+ other_params.data = yuv->data();
+ other_params.stride = yuv->stride();
+ alpha_blend_onto_yuv420p(target_params, other_params, other->data(), other->stride());
break;
}
case AV_PIX_FMT_YUV420P10:
{
- auto yuv = other->convert_pixel_format (dcp::YUVToRGB::REC709, _pixel_format, false, false);
- dcp::Size const ts = size();
- dcp::Size const os = yuv->size();
- for (int ty = start_ty, oy = start_oy; ty < ts.height && oy < os.height; ++ty, ++oy) {
- int const hty = ty / 2;
- int const hoy = oy / 2;
- uint16_t* tY = ((uint16_t *) (data()[0] + (ty * stride()[0]))) + start_tx;
- uint16_t* tU = ((uint16_t *) (data()[1] + (hty * stride()[1]))) + start_tx / 2;
- uint16_t* tV = ((uint16_t *) (data()[2] + (hty * stride()[2]))) + start_tx / 2;
- uint16_t* oY = ((uint16_t *) (yuv->data()[0] + (oy * yuv->stride()[0]))) + start_ox;
- uint16_t* oU = ((uint16_t *) (yuv->data()[1] + (hoy * yuv->stride()[1]))) + start_ox / 2;
- uint16_t* oV = ((uint16_t *) (yuv->data()[2] + (hoy * yuv->stride()[2]))) + start_ox / 2;
- uint8_t* alpha = other->data()[0] + (oy * other->stride()[0]) + start_ox * 4;
- for (int tx = start_tx, ox = start_ox; tx < ts.width && ox < os.width; ++tx, ++ox) {
- float const a = float(alpha[3]) / 255;
- *tY = *oY * a + *tY * (1 - a);
- *tU = *oU * a + *tU * (1 - a);
- *tV = *oV * a + *tV * (1 - a);
- ++tY;
- ++oY;
- if (tx % 2) {
- ++tU;
- ++tV;
- }
- if (ox % 2) {
- ++oU;
- ++oV;
- }
- alpha += 4;
- }
- }
+ auto yuv = other->convert_pixel_format (dcp::YUVToRGB::REC709, _pixel_format, Alignment::COMPACT, false);
+ other_params.data = yuv->data();
+ other_params.stride = yuv->stride();
+ alpha_blend_onto_yuv420p10(target_params, other_params, other->data(), other->stride());
break;
}
case AV_PIX_FMT_YUV422P10LE:
{
- auto yuv = other->convert_pixel_format (dcp::YUVToRGB::REC709, _pixel_format, false, false);
- dcp::Size const ts = size();
- dcp::Size const os = yuv->size();
- for (int ty = start_ty, oy = start_oy; ty < ts.height && oy < os.height; ++ty, ++oy) {
- uint16_t* tY = ((uint16_t *) (data()[0] + (ty * stride()[0]))) + start_tx;
- uint16_t* tU = ((uint16_t *) (data()[1] + (ty * stride()[1]))) + start_tx / 2;
- uint16_t* tV = ((uint16_t *) (data()[2] + (ty * stride()[2]))) + start_tx / 2;
- uint16_t* oY = ((uint16_t *) (yuv->data()[0] + (oy * yuv->stride()[0]))) + start_ox;
- uint16_t* oU = ((uint16_t *) (yuv->data()[1] + (oy * yuv->stride()[1]))) + start_ox / 2;
- uint16_t* oV = ((uint16_t *) (yuv->data()[2] + (oy * yuv->stride()[2]))) + start_ox / 2;
- uint8_t* alpha = other->data()[0] + (oy * other->stride()[0]) + start_ox * 4;
- for (int tx = start_tx, ox = start_ox; tx < ts.width && ox < os.width; ++tx, ++ox) {
- float const a = float(alpha[3]) / 255;
- *tY = *oY * a + *tY * (1 - a);
- *tU = *oU * a + *tU * (1 - a);
- *tV = *oV * a + *tV * (1 - a);
- ++tY;
- ++oY;
- if (tx % 2) {
- ++tU;
- ++tV;
- }
- if (ox % 2) {
- ++oU;
- ++oV;
- }
- alpha += 4;
- }
- }
+ auto yuv = other->convert_pixel_format (dcp::YUVToRGB::REC709, _pixel_format, Alignment::COMPACT, false);
+ other_params.data = yuv->data();
+ other_params.stride = yuv->stride();
+ alpha_blend_onto_yuv422p10le(target_params, other_params, other->data(), other->stride());
break;
}
default:
}
}
+
void
Image::copy (shared_ptr<const Image> other, Position<int> position)
{
}
}
+
void
Image::read_from_socket (shared_ptr<Socket> socket)
{
}
}
+
void
Image::write_to_socket (shared_ptr<Socket> socket) const
{
}
}
+
float
Image::bytes_per_pixel (int c) const
{
return bpp[c];
}
+
/** Construct a Image of a given size and format, allocating memory
* as required.
*
* @param p Pixel format.
* @param s Size in pixels.
- * @param aligned true to make each row of this image aligned to a ALIGNMENT-byte boundary.
+ * @param alignment PADDED to make each row of this image aligned to a ALIGNMENT-byte boundary, otherwise COMPACT.
*/
-Image::Image (AVPixelFormat p, dcp::Size s, bool aligned)
+Image::Image (AVPixelFormat p, dcp::Size s, Alignment alignment)
: _size (s)
, _pixel_format (p)
- , _aligned (aligned)
+ , _alignment (alignment)
{
allocate ();
}
+
void
Image::allocate ()
{
_stride = (int *) wrapped_av_malloc (4 * sizeof (int));
_stride[0] = _stride[1] = _stride[2] = _stride[3] = 0;
+ auto stride_round_up = [](int stride, int t) {
+ int const a = stride + (t - 1);
+ return a - (a % t);
+ };
+
for (int i = 0; i < planes(); ++i) {
_line_size[i] = ceil (_size.width * bytes_per_pixel(i));
- _stride[i] = stride_round_up (i, _line_size, _aligned ? ALIGNMENT : 1);
+ _stride[i] = stride_round_up (_line_size[i], _alignment == Alignment::PADDED ? ALIGNMENT : 1);
/* The assembler function ff_rgb24ToY_avx (in libswscale/x86/input.asm)
uses a 16-byte fetch to read three bytes (R/G/B) of image data.
}
}
+
Image::Image (Image const & other)
: std::enable_shared_from_this<Image>(other)
, _size (other._size)
, _pixel_format (other._pixel_format)
- , _aligned (other._aligned)
+ , _alignment (other._alignment)
{
allocate ();
}
}
-Image::Image (AVFrame* frame)
+
+Image::Image (AVFrame const * frame, Alignment alignment)
: _size (frame->width, frame->height)
, _pixel_format (static_cast<AVPixelFormat>(frame->format))
- , _aligned (true)
+ , _alignment (alignment)
{
DCPOMATIC_ASSERT (_pixel_format != AV_PIX_FMT_NONE);
}
}
-Image::Image (shared_ptr<const Image> other, bool aligned)
+
+Image::Image (shared_ptr<const Image> other, Alignment alignment)
: _size (other->_size)
, _pixel_format (other->_pixel_format)
- , _aligned (aligned)
+ , _alignment (alignment)
{
allocate ();
}
}
+
Image&
Image::operator= (Image const & other)
{
return *this;
}
+
void
Image::swap (Image & other)
{
std::swap (_stride[i], other._stride[i]);
}
- std::swap (_aligned, other._aligned);
+ std::swap (_alignment, other._alignment);
}
+
Image::~Image ()
{
for (int i = 0; i < planes(); ++i) {
av_free (_stride);
}
+
uint8_t * const *
Image::data () const
{
return _data;
}
+
int const *
Image::line_size () const
{
return _line_size;
}
+
int const *
Image::stride () const
{
return _stride;
}
+
dcp::Size
Image::size () const
{
return _size;
}
-bool
-Image::aligned () const
+
+Image::Alignment
+Image::alignment () const
{
- return _aligned;
+ return _alignment;
}
PositionImage
-merge (list<PositionImage> images)
+merge (list<PositionImage> images, Image::Alignment alignment)
{
if (images.empty ()) {
return {};
}
if (images.size() == 1) {
- return images.front ();
+ images.front().image = Image::ensure_alignment(images.front().image, alignment);
+ return images.front();
}
dcpomatic::Rect<int> all (images.front().position, images.front().image->size().width, images.front().image->size().height);
all.extend (dcpomatic::Rect<int>(i.position, i.image->size().width, i.image->size().height));
}
- auto merged = make_shared<Image>(images.front().image->pixel_format(), dcp::Size(all.width, all.height), true);
+ auto merged = make_shared<Image>(images.front().image->pixel_format(), dcp::Size(all.width, all.height), alignment);
merged->make_transparent ();
for (auto const& i: images) {
merged->alpha_blend (i.image, i.position - all.position());
bool
operator== (Image const & a, Image const & b)
{
- if (a.planes() != b.planes() || a.pixel_format() != b.pixel_format() || a.aligned() != b.aligned()) {
+ if (a.planes() != b.planes() || a.pixel_format() != b.pixel_format() || a.alignment() != b.alignment()) {
return false;
}
return true;
}
+
/** Fade the image.
* @param f Amount to fade by; 0 is black, 1 is no fade.
*/
}
}
+
shared_ptr<const Image>
-Image::ensure_aligned (shared_ptr<const Image> image)
+Image::ensure_alignment (shared_ptr<const Image> image, Image::Alignment alignment)
{
- if (image->aligned()) {
+ if (image->alignment() == alignment) {
return image;
}
- return make_shared<Image>(image, true);
+ return make_shared<Image>(image, alignment);
}
+
size_t
Image::memory_used () const
{
return m;
}
-class Memory
-{
-public:
- Memory ()
- : data(0)
- , size(0)
- {}
-
- ~Memory ()
- {
- free (data);
- }
-
- uint8_t* data;
- size_t size;
-};
-
-static void
-png_write_data (png_structp png_ptr, png_bytep data, png_size_t length)
-{
- auto mem = reinterpret_cast<Memory*>(png_get_io_ptr(png_ptr));
- size_t size = mem->size + length;
-
- if (mem->data) {
- mem->data = reinterpret_cast<uint8_t*>(realloc(mem->data, size));
- } else {
- mem->data = reinterpret_cast<uint8_t*>(malloc(size));
- }
-
- if (!mem->data) {
- throw EncodeError (N_("could not allocate memory for PNG"));
- }
-
- memcpy (mem->data + mem->size, data, length);
- mem->size += length;
-}
-
-static void
-png_flush (png_structp)
-{
-
-}
-
-static void
-png_error_fn (png_structp png_ptr, char const * message)
-{
- reinterpret_cast<Image*>(png_get_error_ptr(png_ptr))->png_error (message);
-}
-
-void
-Image::png_error (char const * message)
-{
- throw EncodeError (String::compose ("Error during PNG write: %1", message));
-}
-
-dcp::ArrayData
-Image::as_png () const
-{
- DCPOMATIC_ASSERT (bytes_per_pixel(0) == 4);
- DCPOMATIC_ASSERT (planes() == 1);
- if (pixel_format() != AV_PIX_FMT_RGBA) {
- return convert_pixel_format(dcp::YUVToRGB::REC709, AV_PIX_FMT_RGBA, true, false)->as_png();
- }
-
- /* error handling? */
- png_structp png_ptr = png_create_write_struct(PNG_LIBPNG_VER_STRING, reinterpret_cast<void*>(const_cast<Image*>(this)), png_error_fn, 0);
- if (!png_ptr) {
- throw EncodeError (N_("could not create PNG write struct"));
- }
-
- Memory state;
-
- png_set_write_fn (png_ptr, &state, png_write_data, png_flush);
-
- png_infop info_ptr = png_create_info_struct(png_ptr);
- if (!info_ptr) {
- png_destroy_write_struct (&png_ptr, &info_ptr);
- throw EncodeError (N_("could not create PNG info struct"));
- }
-
- png_set_IHDR (png_ptr, info_ptr, size().width, size().height, 8, PNG_COLOR_TYPE_RGBA, PNG_INTERLACE_NONE, PNG_COMPRESSION_TYPE_DEFAULT, PNG_FILTER_TYPE_DEFAULT);
-
- png_byte ** row_pointers = reinterpret_cast<png_byte **>(png_malloc(png_ptr, size().height * sizeof(png_byte *)));
- for (int i = 0; i < size().height; ++i) {
- row_pointers[i] = (png_byte *) (data()[0] + i * stride()[0]);
- }
-
- png_write_info (png_ptr, info_ptr);
- png_write_image (png_ptr, row_pointers);
- png_write_end (png_ptr, info_ptr);
-
- png_destroy_write_struct (&png_ptr, &info_ptr);
- png_free (png_ptr, row_pointers);
-
- return dcp::ArrayData (state.data, state.size);
-}
-
void
Image::video_range_to_full_range ()
for (int y = 0; y < lines; ++y) {
uint8_t* q = p;
for (int x = 0; x < line_size()[0]; ++x) {
- *q = int((*q - 16) * factor);
+ *q = clamp(lrintf((*q - 16) * factor), 0L, 255L);
++q;
}
p += stride()[0];
}
break;
}
+ case AV_PIX_FMT_RGB48LE:
+ {
+ float const factor = 65536.0 / 56064.0;
+ uint16_t* p = reinterpret_cast<uint16_t*>(data()[0]);
+ int const lines = sample_size(0).height;
+ for (int y = 0; y < lines; ++y) {
+ uint16_t* q = p;
+ int const line_size_pixels = line_size()[0] / 2;
+ for (int x = 0; x < line_size_pixels; ++x) {
+ *q = clamp(lrintf((*q - 4096) * factor), 0L, 65535L);
+ ++q;
+ }
+ p += stride()[0] / 2;
+ }
+ break;
+ }
case AV_PIX_FMT_GBRP12LE:
{
float const factor = 4096.0 / 3504.0;
uint16_t* q = p;
int const line_size_pixels = line_size()[c] / 2;
for (int x = 0; x < line_size_pixels; ++x) {
- *q = int((*q - 256) * factor);
+ *q = clamp(lrintf((*q - 256) * factor), 0L, 4095L);
++q;
}
}