T1: use more compact flags to optimize cache usage in encoder passes. (#172)

Ported from Carl Hetherington work (actually through Matthieu Darbois's port
on top of OpenJPEG 2.1.0)

Can reduce total encoding time by 10-15%

WARNING: VSC mode is not implemented, and so is a temporary regression
that must be fixed.

1 files changed, 101 insertions, 0 deletions

diff --git a/src/lib/openjp2/t1.h b/src/lib/openjp2/t1.h
index 31a2d9f3..f5ce9108 100644
--- a/src/lib/openjp2/t1.h
+++ b/src/lib/openjp2/t1.h
@@ -11,6 +11,7 @@
  * Copyright (c) 2003-2007, Francois-Olivier Devaux
  * Copyright (c) 2003-2014, Antonin Descampe
  * Copyright (c) 2005, Herve Drolon, FreeImage Team
+ * Copyright (c) 2012, Carl Hetherington
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -112,6 +113,101 @@ in T1.C are used by some function in TCD.C.
 #define T1_COLFLAG_VISIT_ROW_3     (T1_COLFLAG_VISIT_ROW_0     << (3U * T1_COLFLAG_RBS))
 #define T1_COLFLAG_REFINE_ROW_3    (T1_COLFLAG_REFINE_ROW_0    << (3U * T1_COLFLAG_RBS))
 
+
+/* BEGINNING of flags that apply to opj_flag_enc_t */
+/** We hold the state of individual data points for the T1 encoder using
+ *  a single 32-bit flags word to hold the state of 4 data points.  This corresponds
+ *  to the 4-point-high columns that the data is processed in.
+ *
+ *  These #defines declare the layout of a 32-bit flags word.
+ *
+ *  This is currently done for encoding only.
+ */
+
+#define T1_SIGMA_0  (1U << 0)
+#define T1_SIGMA_1  (1U << 1)
+#define T1_SIGMA_2  (1U << 2)
+#define T1_SIGMA_3  (1U << 3)
+#define T1_SIGMA_4  (1U << 4)
+#define T1_SIGMA_5  (1U << 5)
+#define T1_SIGMA_6  (1U << 6)
+#define T1_SIGMA_7  (1U << 7)
+#define T1_SIGMA_8  (1U << 8)
+#define T1_SIGMA_9  (1U << 9)
+#define T1_SIGMA_10 (1U << 10)
+#define T1_SIGMA_11 (1U << 11)
+#define T1_SIGMA_12 (1U << 12)
+#define T1_SIGMA_13 (1U << 13)
+#define T1_SIGMA_14 (1U << 14)
+#define T1_SIGMA_15 (1U << 15)
+#define T1_SIGMA_16 (1U << 16)
+#define T1_SIGMA_17 (1U << 17)
+
+#define T1_CHI_0    (1U << 18)
+#define T1_CHI_0_I  18
+#define T1_CHI_1    (1U << 19)
+#define T1_CHI_1_I  19
+#define T1_MU_0     (1U << 20)
+#define T1_PI_0     (1U << 21)
+#define T1_CHI_2    (1U << 22)
+#define T1_CHI_2_I  22
+#define T1_MU_1     (1U << 23)
+#define T1_PI_1     (1U << 24)
+#define T1_CHI_3    (1U << 25)
+#define T1_MU_2     (1U << 26)
+#define T1_PI_2     (1U << 27)
+#define T1_CHI_4    (1U << 28)
+#define T1_MU_3     (1U << 29)
+#define T1_PI_3     (1U << 30)
+#define T1_CHI_5    (1U << 31)
+
+
+/** As an example, the bits T1_SIGMA_3, T1_SIGMA_4 and T1_SIGMA_5
+ *  indicate the significance state of the west neighbour of data point zero
+ *  of our four, the point itself, and its east neighbour respectively.
+ *  Many of the bits are arranged so that given a flags word, you can
+ *  look at the values for the data point 0, then shift the flags
+ *  word right by 3 bits and look at the same bit positions to see the
+ *  values for data point 1.
+ *
+ *  The #defines below help a bit with this; say you have a flags word
+ *  f, you can do things like
+ *
+ *  (f & T1_SIGMA_THIS)
+ *
+ *  to see the significance bit of data point 0, then do
+ *
+ *  ((f >> 3) & T1_SIGMA_THIS)
+ *
+ *  to see the significance bit of data point 1.
+ */
+
+#define T1_SIGMA_NW   T1_SIGMA_0
+#define T1_SIGMA_N    T1_SIGMA_1
+#define T1_SIGMA_NE   T1_SIGMA_2
+#define T1_SIGMA_W    T1_SIGMA_3
+#define T1_SIGMA_THIS T1_SIGMA_4
+#define T1_SIGMA_E    T1_SIGMA_5
+#define T1_SIGMA_SW   T1_SIGMA_6
+#define T1_SIGMA_S    T1_SIGMA_7
+#define T1_SIGMA_SE   T1_SIGMA_8
+#define T1_SIGMA_NEIGHBOURS (T1_SIGMA_NW | T1_SIGMA_N | T1_SIGMA_NE | T1_SIGMA_W | T1_SIGMA_E | T1_SIGMA_SW | T1_SIGMA_S | T1_SIGMA_SE)
+
+#define T1_CHI_THIS   T1_CHI_1
+#define T1_CHI_THIS_I T1_CHI_1_I
+#define T1_MU_THIS    T1_MU_0
+#define T1_PI_THIS    T1_PI_0
+
+#define T1_LUT_SGN_W (1U << 0)
+#define T1_LUT_SIG_N (1U << 1)
+#define T1_LUT_SGN_E (1U << 2)
+#define T1_LUT_SIG_W (1U << 3)
+#define T1_LUT_SGN_N (1U << 4)
+#define T1_LUT_SIG_E (1U << 5)
+#define T1_LUT_SGN_S (1U << 6)
+#define T1_LUT_SIG_S (1U << 7)
+/* END of flags that apply to opj_flag_enc_t */
+
 /* ----------------------------------------------------------------------- */
 
 typedef OPJ_UINT16 opj_flag_t;
@@ -119,6 +215,8 @@ typedef OPJ_UINT16 opj_flag_t;
 /** Flags for 4 consecutive rows of a column */
 typedef OPJ_UINT16 opj_colflag_t;
 
+typedef OPJ_UINT32 opj_flag_enc_t;
+
 /**
 Tier-1 coding (coding of code-block coefficients)
 */
@@ -130,12 +228,15 @@ typedef struct opj_t1 {
     opj_raw_t *raw;
 
     OPJ_INT32  *data;
+    /** Flags used by decoder */
     opj_flag_t *flags;
     /** Addition flag array such that colflags[1+0] is for state of col=0,row=0..3,
        colflags[1+1] for col=1, row=0..3, colflags[1+flags_stride] for col=0,row=4..7, ...
        This array avoids too much cache trashing when processing by 4 vertical samples
        as done in the various decoding steps. */
     opj_colflag_t* colflags;
+    /** Flags used by encoder */
+    opj_flag_enc_t *enc_flags;
     OPJ_UINT32 w;
     OPJ_UINT32 h;
     OPJ_UINT32 datasize;