MIPS: dspr2: added optimization for function MakeARGB32

inline function MakeARGB32 calls changed to call
via pointers to functions which make (a)rgb for
entire row

Change-Id: Ia4bd4be171a46c1e1821e408b073ff5791c587a9

Android.mk

@@ -35,6 +35,8 @@ LOCAL_SRC_FILES := \
     src/dsp/alpha_processing.c \
     src/dsp/alpha_processing_mips_dsp_r2.c \
     src/dsp/alpha_processing_sse2.c \
+    src/dsp/argb.c \
+    src/dsp/argb_mips_dsp_r2.c \
     src/dsp/cpu.c \
     src/dsp/dec.c \
     src/dsp/dec_clip_tables.c \

Makefile.vc

@@ -180,6 +180,8 @@ DSP_DEC_OBJS = \
     $(DIROBJ)\dsp\alpha_processing.obj \
     $(DIROBJ)\dsp\alpha_processing_mips_dsp_r2.obj \
     $(DIROBJ)\dsp\alpha_processing_sse2.obj \
+    $(DIROBJ)\dsp\argb.obj \
+    $(DIROBJ)\dsp\argb_mips_dsp_r2.obj \
     $(DIROBJ)\dsp\cpu.obj \
     $(DIROBJ)\dsp\dec.obj \
     $(DIROBJ)\dsp\dec_clip_tables.obj \

makefile.unix

@@ -110,6 +110,8 @@ DSP_DEC_OBJS = \
     src/dsp/alpha_processing.o \
     src/dsp/alpha_processing_mips_dsp_r2.o \
     src/dsp/alpha_processing_sse2.o \
+    src/dsp/argb.o \
+    src/dsp/argb_mips_dsp_r2.o \
     src/dsp/cpu.o \
     src/dsp/dec.o \
     src/dsp/dec_clip_tables.o \

src/dsp/Makefile.am

@@ -11,6 +11,8 @@ commondir = $(includedir)/webp
 COMMON_SOURCES =
 COMMON_SOURCES += alpha_processing.c
 COMMON_SOURCES += alpha_processing_mips_dsp_r2.c
+COMMON_SOURCES += argb.c
+COMMON_SOURCES += argb_mips_dsp_r2.c
 COMMON_SOURCES += cpu.c
 COMMON_SOURCES += dec.c
 COMMON_SOURCES += dec_clip_tables.c

src/dsp/argb.c

@@ -0,0 +1,57 @@
+// Copyright 2014 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+//   ARGB making functions.
+//
+// Author: Djordje Pesut (djordje.pesut@imgtec.com)
+
+#include "./dsp.h"
+
+static WEBP_INLINE uint32_t MakeARGB32(int a, int r, int g, int b) {
+  return (((uint32_t)a << 24) | (r << 16) | (g << 8) | b);
+}
+
+static void PackARGB(const uint8_t* a, const uint8_t* r, const uint8_t* g,
+                     const uint8_t* b, int len, int step, uint32_t* out) {
+  int i, offset = 0;
+  for (i = 0; i < len; ++i) {
+    out[i] = MakeARGB32(a[offset], r[offset], g[offset], b[offset]);
+    offset += step;
+  }
+}
+
+static void PackRGB(const uint8_t* r, const uint8_t* g, const uint8_t* b,
+                    int len, int step, uint32_t* out) {
+  int i, offset = 0;
+  for (i = 0; i < len; ++i) {
+    out[i] = MakeARGB32(0xff, r[offset], g[offset], b[offset]);
+    offset += step;
+  }
+}
+
+void (*VP8PackARGB)(const uint8_t*, const uint8_t*, const uint8_t*,
+                    const uint8_t*, int, int, uint32_t*);
+void (*VP8PackRGB)(const uint8_t*, const uint8_t*, const uint8_t*,
+                   int, int, uint32_t*);
+
+extern void VP8EncDspARGBInitMIPSdspR2(void);
+
+WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspARGBInit(void) {
+  VP8PackARGB = PackARGB;
+  VP8PackRGB = PackRGB;
+
+  // If defined, use CPUInfo() to overwrite some pointers with faster versions.
+  if (VP8GetCPUInfo != NULL) {
+#if defined(WEBP_USE_MIPS_DSP_R2)
+    if (VP8GetCPUInfo(kMIPSdspR2)) {
+      VP8EncDspARGBInitMIPSdspR2();
+    }
+#endif
+  }
+}

src/dsp/argb_mips_dsp_r2.c

@@ -0,0 +1,107 @@
+// Copyright 2014 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+//   ARGB making functions (mips version).
+//
+// Author: Djordje Pesut (djordje.pesut@imgtec.com)
+
+#include "./dsp.h"
+
+#if defined(WEBP_USE_MIPS_DSP_R2)
+
+static void PackARGB(const uint8_t* a, const uint8_t* r, const uint8_t* g,
+                     const uint8_t* b, int len, int step, uint32_t* out) {
+  int temp0, temp1, temp2, temp3, offset;
+  const int rest = len & 1;
+  const uint32_t* const loop_end = out + len - rest;
+  __asm__ volatile (
+    "xor          %[offset],   %[offset], %[offset]    \n\t"
+    "beq          %[loop_end], %[out],    0f           \n\t"
+  "2:                                                  \n\t"
+    "lbux         %[temp0],    %[offset](%[a])         \n\t"
+    "lbux         %[temp1],    %[offset](%[r])         \n\t"
+    "lbux         %[temp2],    %[offset](%[g])         \n\t"
+    "lbux         %[temp3],    %[offset](%[b])         \n\t"
+    "ins          %[temp1],    %[temp0],  16,     16   \n\t"
+    "ins          %[temp3],    %[temp2],  16,     16   \n\t"
+    "addiu        %[out],      %[out],    4            \n\t"
+    "precr.qb.ph  %[temp0],    %[temp1],  %[temp3]     \n\t"
+    "sw           %[temp0],    -4(%[out])              \n\t"
+    "addu         %[offset],   %[offset], %[step]      \n\t"
+    "bne          %[loop_end], %[out],    2b           \n\t"
+  "0:                                                  \n\t"
+    "beq          %[rest],     $zero,     1f           \n\t"
+    "lbux         %[temp0],    %[offset](%[a])         \n\t"
+    "lbux         %[temp1],    %[offset](%[r])         \n\t"
+    "lbux         %[temp2],    %[offset](%[g])         \n\t"
+    "lbux         %[temp3],    %[offset](%[b])         \n\t"
+    "ins          %[temp1],    %[temp0],  16,     16   \n\t"
+    "ins          %[temp3],    %[temp2],  16,     16   \n\t"
+    "precr.qb.ph  %[temp0],    %[temp1],  %[temp3]     \n\t"
+    "sw           %[temp0],    0(%[out])               \n\t"
+  "1:                                                  \n\t"
+    : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
+      [temp3]"=&r"(temp3), [offset]"=&r"(offset), [out]"+&r"(out)
+    : [a]"r"(a), [r]"r"(r), [g]"r"(g), [b]"r"(b), [step]"r"(step),
+      [loop_end]"r"(loop_end), [rest]"r"(rest)
+    : "memory"
+  );
+}
+
+static void PackRGB(const uint8_t* r, const uint8_t* g, const uint8_t* b,
+                    int len, int step, uint32_t* out) {
+  int temp0, temp1, temp2, offset;
+  const int rest = len & 1;
+  const int a = 0xff;
+  const uint32_t* const loop_end = out + len - rest;
+  __asm__ volatile (
+    "xor          %[offset],   %[offset], %[offset]    \n\t"
+    "beq          %[loop_end], %[out],    0f           \n\t"
+  "2:                                                  \n\t"
+    "lbux         %[temp0],    %[offset](%[r])         \n\t"
+    "lbux         %[temp1],    %[offset](%[g])         \n\t"
+    "lbux         %[temp2],    %[offset](%[b])         \n\t"
+    "ins          %[temp0],    %[a],      16,     16   \n\t"
+    "ins          %[temp2],    %[temp1],  16,     16   \n\t"
+    "addiu        %[out],      %[out],    4            \n\t"
+    "precr.qb.ph  %[temp0],    %[temp0],  %[temp2]     \n\t"
+    "sw           %[temp0],    -4(%[out])              \n\t"
+    "addu         %[offset],   %[offset], %[step]      \n\t"
+    "bne          %[loop_end], %[out],    2b           \n\t"
+  "0:                                                  \n\t"
+    "beq          %[rest],     $zero,     1f           \n\t"
+    "lbux         %[temp0],    %[offset](%[r])         \n\t"
+    "lbux         %[temp1],    %[offset](%[g])         \n\t"
+    "lbux         %[temp2],    %[offset](%[b])         \n\t"
+    "ins          %[temp0],    %[a],      16,     16   \n\t"
+    "ins          %[temp2],    %[temp1],  16,     16   \n\t"
+    "precr.qb.ph  %[temp0],    %[temp0],  %[temp2]     \n\t"
+    "sw           %[temp0],    0(%[out])               \n\t"
+  "1:                                                  \n\t"
+    : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
+      [offset]"=&r"(offset), [out]"+&r"(out)
+    : [a]"r"(a), [r]"r"(r), [g]"r"(g), [b]"r"(b), [step]"r"(step),
+      [loop_end]"r"(loop_end), [rest]"r"(rest)
+    : "memory"
+  );
+}
+
+#endif  // WEBP_USE_MIPS_DSP_R2
+
+//------------------------------------------------------------------------------
+// Entry point
+
+extern WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspARGBInitMIPSdspR2(void);
+
+WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspARGBInitMIPSdspR2(void) {
+#if defined(WEBP_USE_MIPS_DSP_R2)
+  VP8PackARGB = PackARGB;
+  VP8PackRGB = PackRGB;
+#endif  // WEBP_USE_MIPS_DSP_R2
+}

src/dsp/dsp.h

@@ -337,6 +337,16 @@ void WebPMultARGBRowC(uint32_t* const ptr, int width, int inverse);
 // To be called first before using the above.
 WEBP_TSAN_IGNORE_FUNCTION void WebPInitAlphaProcessing(void);
 
+// ARGB making functions.
+extern void (*VP8PackARGB)(const uint8_t* a, const uint8_t* r,
+                           const uint8_t* g, const uint8_t* b, int len,
+                           int step, uint32_t* out);
+extern void (*VP8PackRGB)(const uint8_t* r, const uint8_t* g, const uint8_t* b,
+                          int len, int step, uint32_t* out);
+
+// To be called first before using the above.
+WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspARGBInit(void);
+
 #ifdef __cplusplus
 }    // extern "C"
 #endif

src/enc/picture.c

@@ -15,6 +15,7 @@
 #include <stdlib.h>
 
 #include "./vp8enci.h"
+#include "../dsp/dsp.h"
 #include "../utils/utils.h"
 
 //------------------------------------------------------------------------------
@@ -38,6 +39,7 @@ int WebPPictureInitInternal(WebPPicture* picture, int version) {
     memset(picture, 0, sizeof(*picture));
     picture->writer = DummyWriter;
     WebPEncodingSetError(picture, VP8_ENC_OK);
+    VP8EncDspARGBInit();
   }
   return 1;
 }

src/enc/picture_csp.c

@@ -32,10 +32,6 @@ static const union {
 } test_endian = { 0xff000000u };
 #define ALPHA_IS_LAST (test_endian.bytes[3] == 0xff)
 
-static WEBP_INLINE uint32_t MakeARGB32(int a, int r, int g, int b) {
-  return (((uint32_t)a << 24) | (r << 16) | (g << 8) | b);
-}
-
 //------------------------------------------------------------------------------
 // Detection of non-trivial transparency
 
@@ -1065,13 +1061,19 @@ static int Import(WebPPicture* const picture,
   if (!WebPPictureAlloc(picture)) return 0;
 
   assert(step >= (import_alpha ? 4 : 3));
-  for (y = 0; y < height; ++y) {
+  if (import_alpha) {
-    uint32_t* const dst = &picture->argb[y * picture->argb_stride];
+    for (y = 0; y < height; ++y) {
-    int x;
+      uint32_t* const dst = &picture->argb[y * picture->argb_stride];
-    for (x = 0; x < width; ++x) {
+      const int offset = y * rgb_stride;
-      const int offset = step * x + y * rgb_stride;
+      VP8PackARGB(a_ptr + offset, r_ptr + offset, g_ptr + offset,
-      dst[x] = MakeARGB32(import_alpha ? a_ptr[offset] : 0xff,
+                  b_ptr + offset, width, step, dst);
-                          r_ptr[offset], g_ptr[offset], b_ptr[offset]);
+    }
+  } else {
+    for (y = 0; y < height; ++y) {
+      uint32_t* const dst = &picture->argb[y * picture->argb_stride];
+      const int offset = y * rgb_stride;
+      VP8PackRGB(r_ptr + offset, g_ptr + offset, b_ptr + offset,
+                 width, step, dst);
     }
   }
   return 1;