SSE2 implementation of VP8PackARGB

Change-Id: I40c0e26a6a2701216e4ddebcf793aa535677f437

Android.mk

@@ -37,6 +37,7 @@ LOCAL_SRC_FILES := \
     src/dsp/alpha_processing_sse2.c \
     src/dsp/argb.c \
     src/dsp/argb_mips_dsp_r2.c \
+    src/dsp/argb_sse2.c \
     src/dsp/cpu.c \
     src/dsp/dec.c \
     src/dsp/dec_clip_tables.c \

Makefile.vc

@@ -182,6 +182,7 @@ DSP_DEC_OBJS = \
     $(DIROBJ)\dsp\alpha_processing_sse2.obj \
     $(DIROBJ)\dsp\argb.obj \
     $(DIROBJ)\dsp\argb_mips_dsp_r2.obj \
+    $(DIROBJ)\dsp\argb_sse2.obj \
     $(DIROBJ)\dsp\cpu.obj \
     $(DIROBJ)\dsp\dec.obj \
     $(DIROBJ)\dsp\dec_clip_tables.obj \

makefile.unix

@@ -112,6 +112,7 @@ DSP_DEC_OBJS = \
     src/dsp/alpha_processing_sse2.o \
     src/dsp/argb.o \
     src/dsp/argb_mips_dsp_r2.o \
+    src/dsp/argb_sse2.o \
     src/dsp/cpu.o \
     src/dsp/dec.o \
     src/dsp/dec_clip_tables.o \

src/dsp/Makefile.am

@@ -11,8 +11,6 @@ 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
@@ -37,6 +35,8 @@ COMMON_SOURCES += yuv_mips32.c
 COMMON_SOURCES += yuv_mips_dsp_r2.c
 
 ENC_SOURCES =
+ENC_SOURCES += argb.c
+ENC_SOURCES += argb_mips_dsp_r2.c
 ENC_SOURCES += enc.c
 ENC_SOURCES += enc_mips32.c
 ENC_SOURCES += enc_mips_dsp_r2.c
@@ -58,6 +58,7 @@ libwebpdspdecode_sse2_la_CPPFLAGS = $(libwebpdsp_sse2_la_CPPFLAGS)
 libwebpdspdecode_sse2_la_CFLAGS = $(libwebpdsp_sse2_la_CFLAGS)
 
 libwebpdsp_sse2_la_SOURCES =
+libwebpdsp_sse2_la_SOURCES += argb_sse2.c
 libwebpdsp_sse2_la_SOURCES += enc_sse2.c
 libwebpdsp_sse2_la_CPPFLAGS = $(libwebpdsp_la_CPPFLAGS)
 libwebpdsp_sse2_la_CFLAGS = $(AM_CFLAGS) $(SSE2_FLAGS)

src/dsp/argb.c

@@ -40,6 +40,7 @@ void (*VP8PackRGB)(const uint8_t*, const uint8_t*, const uint8_t*,
                    int, int, uint32_t*);
 
 extern void VP8EncDspARGBInitMIPSdspR2(void);
+extern void VP8EncDspARGBInitSSE2(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspARGBInit(void) {
   VP8PackARGB = PackARGB;
@@ -47,6 +48,11 @@ WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspARGBInit(void) {
 
   // If defined, use CPUInfo() to overwrite some pointers with faster versions.
   if (VP8GetCPUInfo != NULL) {
+#if defined(WEBP_USE_SSE2)
+    if (VP8GetCPUInfo(kSSE2)) {
+      VP8EncDspARGBInitSSE2();
+    }
+#endif
 #if defined(WEBP_USE_MIPS_DSP_R2)
     if (VP8GetCPUInfo(kMIPSdspR2)) {
       VP8EncDspARGBInitMIPSdspR2();

src/dsp/argb_sse2.c

@@ -0,0 +1,62 @@
+// 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 (SSE2 version).
+//
+// Author: Skal (pascal.massimino@gmail.com)
+
+#include "./dsp.h"
+
+#if defined(WEBP_USE_SSE2)
+
+#include <assert.h>
+#include <emmintrin.h>
+#include <string.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, uint32_t* out) {
+  if (g == r + 1) {  // RGBA input order. Need to swap R and B.
+    int i = 0;
+    const int len4 = len & ~3;
+    const __m128i mask = _mm_set1_epi32(0x00ff00ffu);
+    assert(b == r + 2);
+    assert(a == r + 3);
+    for (; i < len4; i += 4) {
+      const __m128i A = _mm_loadu_si128((__m128i*)(r + 4 * i));
+      const __m128i B = _mm_and_si128(A, mask);     // R 0 B 0
+      const __m128i C = _mm_andnot_si128(mask, A);  // 0 G 0 A
+      const __m128i D = _mm_shufflelo_epi16(B, _MM_SHUFFLE(2, 3, 0, 1));
+      const __m128i E = _mm_shufflehi_epi16(D, _MM_SHUFFLE(2, 3, 0, 1));
+      const __m128i F = _mm_or_si128(E, C);
+      _mm_storeu_si128((__m128i*)(out + i), F);
+    }
+    for (; i < len; ++i) {
+      out[i] = MakeARGB32(a[4 * i], r[4 * i], g[4 * i], b[4 * i]);
+    }
+  } else {
+    assert(g == b + 1);
+    assert(r == b + 2);
+    assert(a == b + 3);
+    memcpy(out, b, len * 4);
+  }
+}
+
+#endif    // WEBP_USE_SSE2
+
+extern void VP8EncDspARGBInitSSE2(void);
+
+WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspARGBInitSSE2(void) {
+#if defined(WEBP_USE_SSE2)
+  VP8PackARGB = PackARGB;
+#endif
+}