Implements RGB to YV12 conversion in YASM.

media/base/yuv_convert.cc

 // Copyright (c) 2011 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 // This webpage shows layout of YV12 and other YUV formats
 // http://www.fourcc.org/yuv.php
 // The actual conversion is best described here
 // http://en.wikipedia.org/wiki/YUV
 // An article on optimizing YUV conversion using tables instead of multiplies
 // http://lestourtereaux.free.fr/papers/data/yuvrgb.pdf
 //
 // YV12 is a full plane of Y and a half height, half width chroma planes
 // YV16 is a full plane of Y and a full height, half width chroma planes
 //
 // ARGB pixel format is output, which on little endian is stored as BGRA.
 // The alpha is set to 255, allowing the application to use RGBA or RGB32.
 
 #include "media/base/yuv_convert.h"
 
 #include "build/build_config.h"
 #include "media/base/cpu_features.h"
+#include "media/base/simd/yuv_convert.h"
 #include "media/base/yuv_convert_internal.h"
 #include "media/base/yuv_row.h"
 
 #if USE_MMX
 #if defined(_MSC_VER)
 #include <intrin.h>
 #else
 #include <mmintrin.h>
 #endif
 #endif
 
 #if USE_SSE2
 #include <emmintrin.h>
 #endif
 
+#define USE_YASM

[Alpha Left Google, 2011/08/22 14:16:59]

It's better not to use this #ifdef.

+
 namespace media {
 
 // 16.16 fixed point arithmetic
 const int kFractionBits = 16;
 const int kFractionMax = 1 << kFractionBits;
 const int kFractionMask = ((1 << kFractionBits) - 1);
 
 // Convert a frame of YUV to 32 bit ARGB.
 void ConvertYUVToRGB32(const uint8* y_buf,
                        const uint8* u_buf,
                        const uint8* v_buf,
                        uint8* rgb_buf,
                        int width,
                        int height,
                        int y_pitch,
                        int uv_pitch,
                        int rgb_pitch,
                        YUVType yuv_type) {
+#ifdef USE_YASM

[Alpha Left Google, 2011/08/22 14:16:59]

There seems to be a slow down using this new YUV->RGB32 function, I would like
to verify this.

+  if (hasSSE2()) {
+    simd::ConvertYUVtoRGB32(y_buf, u_buf, v_buf, rgb_buf, width, height,
+                            y_pitch, uv_pitch, rgb_pitch, yuv_type);
+    return;
+  }
+#endif
+
   unsigned int y_shift = yuv_type;
   for (int y = 0; y < height; ++y) {
     uint8* rgb_row = rgb_buf + y * rgb_pitch;
     const uint8* y_ptr = y_buf + y * y_pitch;
     const uint8* u_ptr = u_buf + (y >> y_shift) * uv_pitch;
     const uint8* v_ptr = v_buf + (y >> y_shift) * uv_pitch;
 
     FastConvertYUVToRGB32Row(y_ptr,
                              u_ptr,
                              v_ptr,
@@ skipping 295 matching lines @@
                        int uvstride) {
   static void (*convert_proc)(const uint8*, uint8*, uint8*, uint8*,
                               int, int, int, int, int) = NULL;
   if (!convert_proc) {
 #if defined(ARCH_CPU_ARM_FAMILY)
     // For ARM processors, always use C version.
     // TODO(hclam): Implement a NEON version.
     convert_proc = &ConvertRGB32ToYUV_C;
 #else
     // For x86 processors, check if SSE2 is supported.
     if (hasSSE2())

[Alpha Left Google, 2011/08/22 14:16:59]

if hasSSE2() compiles on all platform we can get rid of the above ifdef check
for ARM.

+#ifdef USE_YASM

[Alpha Left Google, 2011/08/22 14:16:59]

Don't use an ifdef, just apply the new function.

+      convert_proc = &simd::ConvertRGB32toYUV;
+#else
       convert_proc = &ConvertRGB32ToYUV_SSE2;

[Alpha Left Google, 2011/08/22 14:16:59]

Remove this line.

+#endif
     else
       convert_proc = &ConvertRGB32ToYUV_C;
 #endif
   }
 
   convert_proc(rgbframe, yplane, uplane, vplane, width, height,
                rgbstride, ystride, uvstride);
 }
 
 void ConvertRGB24ToYUV(const uint8* rgbframe,
                        uint8* yplane,
                        uint8* uplane,
                        uint8* vplane,
                        int width,
                        int height,
                        int rgbstride,
                        int ystride,
                        int uvstride) {
+#ifdef USE_YASM

[Alpha Left Google, 2011/08/22 14:16:59]

Same here use if without ifdef. Also because with this code, hasSSE2 cannot be
compiled on ARM. I think it's better to have hasSSE2 compiles on all platforms
and return false on ARM. 

+  if (hasSSE2()) {
+    simd::ConvertRGB24toYUV(rgbframe, yplane, uplane, vplane, width, height,
+                            rgbstride, ystride, uvstride);
+    return;
+  }
+#endif
   ConvertRGB24ToYUV_C(rgbframe, yplane, uplane, vplane, width, height,
                       rgbstride, ystride, uvstride);
 }
 
 void ConvertYUY2ToYUV(const uint8* src,
                       uint8* yplane,
                       uint8* uplane,
                       uint8* vplane,
                       int width,
                       int height) {
   ConvertYUY2ToYUV_C(src, yplane, uplane, vplane, width, height);
 }
 }  // namespace media