T1: (failed) optimization attempt to use signed magnitude representation (#931)

Instead of using two-complement integer representation of data samples during
code-block decoding, use a signed magnitude representation to avoid comparison
branches in passes.

However the (best) timings show that it is actually slightly slower, or at
least no better. 51649 ms with this attempt vs 51536 before on MAPA_005.jp2
reencoded with default options.

Two variants with sign in msb (the default) or in lsb lead to similar
performances

diff --git a/src/lib/openjp2/t1.c b/src/lib/openjp2/t1.c
index 0cc677235fa2ef179bc93c2468ba974034d044bb..fc7ad9b5fbfbaafa6fe01be6b29d406b56f5f2dd 100644 (file)
--- a/src/lib/openjp2/t1.c
+++ b/src/lib/openjp2/t1.c
@@ -12,6 +12,7 @@
  * Copyright (c) 2003-2014, Antonin Descampe
  * Copyright (c) 2005, Herve Drolon, FreeImage Team
  * Copyright (c) 2007, Callum Lerwick <seg@haxxed.com>
+ * Copyright (c) 2017, IntoPIX SA
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -466,7 +467,7 @@ static INLINE void opj_t1_dec_sigpass_step_raw(
                             (T1_COLFLAG_RBS * row)))) {
         if (opj_raw_decode(raw)) {
             v = opj_raw_decode(raw);    /* ESSAI */
-            *datap = v ? -oneplushalf : oneplushalf;
+            *datap = BUILD_SMR(v, oneplushalf);
             opj_t1_updateflagscolflags(flagsp, colflagsp, v, t1->flags_stride, row);
         }
 #ifdef CONSISTENCY_CHECK
@@ -502,7 +503,7 @@ static INLINE void opj_t1_dec_sigpass_step_mqc(
         if (opj_mqc_decode(mqc)) {
             opj_mqc_setcurctx(mqc, opj_t1_getctxno_sc((OPJ_UINT32)flag));
             v = opj_mqc_decode(mqc) ^ opj_t1_getspb((OPJ_UINT32)flag);
-            *datap = v ? -oneplushalf : oneplushalf;
+            *datap = BUILD_SMR(v, oneplushalf);
             opj_t1_updateflagscolflags(flagsp, colflagsp, (OPJ_UINT32)v, flags_stride, row);
         }
 #ifdef CONSISTENCY_CHECK
@@ -534,7 +535,7 @@ static INLINE void opj_t1_dec_sigpass_step_mqc_vsc(
         if (opj_mqc_decode(mqc)) {
             opj_mqc_setcurctx(mqc, opj_t1_getctxno_sc(flag));
             v = (OPJ_UINT32)opj_mqc_decode(mqc) ^ opj_t1_getspb(flag);
-            *datap = v ? -oneplushalf : oneplushalf;
+            *datap = BUILD_SMR(v, oneplushalf);
             opj_t1_updateflagscolflags(flagsp, colflagsp, v, t1->flags_stride, row);
         }
 #ifdef CONSISTENCY_CHECK
@@ -771,7 +772,8 @@ static INLINE void opj_t1_dec_refpass_step_raw(
             ((T1_COLFLAG_SIG_ROW_0) << (T1_COLFLAG_RBS * row))) {
         v = (OPJ_INT32)opj_raw_decode(raw);
         t = v ? poshalf : neghalf;
-        *datap += *datap < 0 ? -t : t;
+        *datap = UPDATE_MANTISSA_FROM_SMR(*datap,
+                                          GET_MANTISSA_FROM_SMR(*datap) + t);
         *colflagsp |= (opj_colflag_t)(T1_COLFLAG_REFINE_ROW_0 <<
                                       (T1_COLFLAG_RBS * row));
     }
@@ -811,7 +813,8 @@ static INLINE void opj_t1_dec_refpass_step_mqc(
         opj_mqc_setcurctx(mqc, tmp2);      /* ESSAI */
         v = opj_mqc_decode(mqc);
         t = v ? poshalf : neghalf;
-        *datap += *datap < 0 ? -t : t;
+        *datap = UPDATE_MANTISSA_FROM_SMR(*datap,
+                                          GET_MANTISSA_FROM_SMR(*datap) + t);
         *colflagsp |= (opj_colflag_t)(T1_COLFLAG_REFINE_ROW_0 <<
                                       (T1_COLFLAG_RBS * row));
     }
@@ -843,7 +846,8 @@ static INLINE void opj_t1_dec_refpass_step_mqc_vsc(
         opj_mqc_setcurctx(mqc, tmp2);      /* ESSAI */
         v = (OPJ_UINT32)opj_mqc_decode(mqc);
         t = v ? poshalf : neghalf;
-        *datap += *datap < 0 ? -t : t;
+        *datap = UPDATE_MANTISSA_FROM_SMR(*datap,
+                                          GET_MANTISSA_FROM_SMR(*datap) + t);
         *colflagsp |= (opj_colflag_t)(T1_COLFLAG_REFINE_ROW_0 <<
                                       (T1_COLFLAG_RBS * row));
     }
@@ -1074,7 +1078,7 @@ static void opj_t1_dec_clnpass_step_partial(
     flag = *flagsp;
     opj_mqc_setcurctx(mqc, opj_t1_getctxno_sc((OPJ_UINT32)flag));
     v = opj_mqc_decode(mqc) ^ opj_t1_getspb((OPJ_UINT32)flag);
-    *datap = v ? -oneplushalf : oneplushalf;
+    *datap = BUILD_SMR(v, oneplushalf);
     opj_t1_updateflagscolflags(flagsp, colflagsp, (OPJ_UINT32)v, t1->flags_stride,
                                row);
 #ifdef CONSISTENCY_CHECK
@@ -1104,7 +1108,7 @@ static void opj_t1_dec_clnpass_step(
         if (opj_mqc_decode(mqc)) {
             opj_mqc_setcurctx(mqc, opj_t1_getctxno_sc((OPJ_UINT32)flag));
             v = opj_mqc_decode(mqc) ^ opj_t1_getspb((OPJ_UINT32)flag);
-            *datap = v ? -oneplushalf : oneplushalf;
+            *datap = BUILD_SMR(v, oneplushalf);
             opj_t1_updateflagscolflags(flagsp, colflagsp, (OPJ_UINT32)v, t1->flags_stride,
                                        row);
         }
@@ -1135,7 +1139,7 @@ static void opj_t1_dec_clnpass_step_only_if_flag_not_sig_visit(
         if (opj_mqc_decode(mqc)) {
             opj_mqc_setcurctx(mqc, opj_t1_getctxno_sc((OPJ_UINT32)flag));
             v = opj_mqc_decode(mqc) ^ opj_t1_getspb((OPJ_UINT32)flag);
-            *datap = v ? -oneplushalf : oneplushalf;
+            *datap = BUILD_SMR(v, oneplushalf);
             opj_t1_updateflagscolflags(flagsp, colflagsp, (OPJ_UINT32)v, flags_stride, row);
         }
     }
@@ -1168,7 +1172,7 @@ static void opj_t1_dec_clnpass_step_vsc(
 LABEL_PARTIAL:
             opj_mqc_setcurctx(mqc, opj_t1_getctxno_sc(flag));
             v = (OPJ_UINT32)opj_mqc_decode(mqc) ^ opj_t1_getspb(flag);
-            *datap = v ? -oneplushalf : oneplushalf;
+            *datap = BUILD_SMR(v, oneplushalf);
             opj_t1_updateflagscolflags(flagsp, colflagsp, v, t1->flags_stride, row);
         }
     }
@@ -1748,11 +1752,11 @@ static void opj_t1_clbl_decode_processor(void* user_data, opj_tls_t* tls)
         OPJ_INT32 thresh = 1 << tccp->roishift;
         for (j = 0; j < cblk_h; ++j) {
             for (i = 0; i < cblk_w; ++i) {
-                OPJ_INT32 val = datap[(j * cblk_w) + i];
-                OPJ_INT32 mag = abs(val);
+                OPJ_INT32 mag = GET_MANTISSA_FROM_SMR(datap[(j * cblk_w) + i]);
                 if (mag >= thresh) {
                     mag >>= tccp->roishift;
-                    datap[(j * cblk_w) + i] = val < 0 ? -mag : mag;
+                    datap[(j * cblk_w) + i] =
+                        UPDATE_MANTISSA_FROM_SMR(datap[(j * cblk_w) + i], mag);
                 }
             }
         }
@@ -1763,18 +1767,18 @@ static void opj_t1_clbl_decode_processor(void* user_data, opj_tls_t* tls)
         for (j = 0; j < cblk_h; ++j) {
             i = 0;
             for (; i < (cblk_w & ~(OPJ_UINT32)3U); i += 4U) {
-                OPJ_INT32 tmp0 = datap[(j * cblk_w) + i + 0U];
-                OPJ_INT32 tmp1 = datap[(j * cblk_w) + i + 1U];
-                OPJ_INT32 tmp2 = datap[(j * cblk_w) + i + 2U];
-                OPJ_INT32 tmp3 = datap[(j * cblk_w) + i + 3U];
-                ((OPJ_INT32*)tiledp)[(j * tile_w) + i + 0U] = tmp0 / 2;
-                ((OPJ_INT32*)tiledp)[(j * tile_w) + i + 1U] = tmp1 / 2;
-                ((OPJ_INT32*)tiledp)[(j * tile_w) + i + 2U] = tmp2 / 2;
-                ((OPJ_INT32*)tiledp)[(j * tile_w) + i + 3U] = tmp3 / 2;
+                OPJ_UINT32 tmp0 = datap[(j * cblk_w) + i + 0U];
+                OPJ_UINT32 tmp1 = datap[(j * cblk_w) + i + 1U];
+                OPJ_UINT32 tmp2 = datap[(j * cblk_w) + i + 2U];
+                OPJ_UINT32 tmp3 = datap[(j * cblk_w) + i + 3U];
+                ((OPJ_INT32*)tiledp)[(j * tile_w) + i + 0U] = GET_FROM_SMR(tmp0) / 2;
+                ((OPJ_INT32*)tiledp)[(j * tile_w) + i + 1U] = GET_FROM_SMR(tmp1) / 2;
+                ((OPJ_INT32*)tiledp)[(j * tile_w) + i + 2U] = GET_FROM_SMR(tmp2) / 2;
+                ((OPJ_INT32*)tiledp)[(j * tile_w) + i + 3U] = GET_FROM_SMR(tmp3) / 2;
             }
             for (; i < cblk_w; ++i) {
-                OPJ_INT32 tmp = datap[(j * cblk_w) + i];
-                ((OPJ_INT32*)tiledp)[(j * tile_w) + i] = tmp / 2;
+                OPJ_UINT32 tmp = datap[(j * cblk_w) + i];
+                ((OPJ_INT32*)tiledp)[(j * tile_w) + i] = GET_FROM_SMR(tmp) / 2;
             }
         }
     } else {        /* if (tccp->qmfbid == 0) */
@@ -1783,7 +1787,7 @@ static void opj_t1_clbl_decode_processor(void* user_data, opj_tls_t* tls)
         for (j = 0; j < cblk_h; ++j) {
             OPJ_FLOAT32* OPJ_RESTRICT tiledp2 = tiledp;
             for (i = 0; i < cblk_w; ++i) {
-                OPJ_FLOAT32 tmp = (OPJ_FLOAT32) * datap * band->stepsize;
+                OPJ_FLOAT32 tmp = (OPJ_FLOAT32) GET_FROM_SMR(*datap) * band->stepsize;
                 *tiledp2 = tmp;
                 datap++;
                 tiledp2++;
@@ -2264,7 +2268,8 @@ static void opj_t1_dec_refpass_step(opj_t1_t *t1,
             v = opj_mqc_decode(mqc);
         }
         t = v ? poshalf : neghalf;
-        *datap += *datap < 0 ? -t : t;
+        *datap = UPDATE_MANTISSA_FROM_SMR(*datap,
+                                          GET_MANTISSA_FROM_SMR(*datap) + t);
         *flagsp |= T1_REFINE;
     }
 }               /* VSC and  BYPASS by Antonin  */

diff --git a/src/lib/openjp2/t1.h b/src/lib/openjp2/t1.h
index 31a2d9f39dd893a8ebd41a505945dc5f6c0ec5d6..e547e9293dc073cf45a0be66689dc8b4cdee5b54 100644 (file)
--- 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) 2017, IntoPIX SA
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -148,6 +149,23 @@ typedef struct opj_t1 {
 
 #define MACRO_t1_flags(x,y) t1->flags[((x)*(t1->flags_stride))+(y)]
 
+/* SMR: Signed magnitude representation */
+#define SMR_SIGN_IN_MSB
+#ifdef SMR_SIGN_IN_MSB
+#define BUILD_SMR(s,mantissa)       ((OPJ_INT32)((((OPJ_UINT32)(s)) << 31) | (mantissa)))
+#define UPDATE_MANTISSA_FROM_SMR(smr, mantissa) \
+                                    ((((OPJ_UINT32)(smr)) & 0x80000000) | (mantissa))
+#define GET_SIGN_FROM_SMR(smr)      (((OPJ_UINT32)(smr)) >> 31)
+#define GET_MANTISSA_FROM_SMR(smr)  (((OPJ_UINT32)(smr)) & 0x7FFFFFFF)
+#else
+#define BUILD_SMR(s,mantissa)       ((OPJ_INT32)((s) | ((OPJ_UINT32)(mantissa) << 1)))
+#define UPDATE_MANTISSA_FROM_SMR(smr, mantissa) \
+                                    ((((OPJ_UINT32)(smr)) & 0x1) | ((OPJ_UINT32)(mantissa) << 1))
+#define GET_SIGN_FROM_SMR(smr)      (((OPJ_UINT32)(smr)) & 1)
+#define GET_MANTISSA_FROM_SMR(smr)  (((OPJ_UINT32)(smr)) >> 1)
+#endif
+#define GET_FROM_SMR(smr)           (GET_SIGN_FROM_SMR(smr) ? -(OPJ_INT32)GET_MANTISSA_FROM_SMR(smr) : (OPJ_INT32)GET_MANTISSA_FROM_SMR(smr))
+
 /** @name Exported functions */
 /*@{*/
 /* ----------------------------------------------------------------------- */