mmx for linux yuv convert function.

BUG=18449,20718
TEST=play a video on linux without scaling and it should go 4 times faster than with scaling.

Review URL: http://codereview.chromium.org/174442

git-svn-id: svn://svn.chromium.org/chrome/trunk/src@25001 0039d316-1c4b-4281-b951-d872f2087c98

5 files changed, 903 insertions, 88 deletions

diff --git a/media/base/yuv_convert.cc b/media/base/yuv_convert.cc
index 7e4da5e..c73dfe4 100644
--- a/media/base/yuv_convert.cc
+++ b/media/base/yuv_convert.cc
@@ -131,7 +131,7 @@ void ScaleYUVToRGB32(const uint8* y_buf,
     const uint8* u_ptr = u_buf + (scaled_y >> y_shift) * uv_pitch;
     const uint8* v_ptr = v_buf + (scaled_y >> y_shift) * uv_pitch;
 
-#if USE_MMX
+#if USE_MMX && defined(_MSC_VER)
     if (scaled_width == (width * 2)) {
       DoubleYUVToRGB32Row(y_ptr, u_ptr, v_ptr,
                           dest_pixel, scaled_width);
diff --git a/media/base/yuv_row.h b/media/base/yuv_row.h
index b8651f5..bb06722 100644
--- a/media/base/yuv_row.h
+++ b/media/base/yuv_row.h
@@ -5,13 +5,14 @@
 // yuv_row internal functions to handle YUV conversion and scaling to RGB.
 // These functions are used from both yuv_convert.cc and yuv_scale.cc.
 
+// TODO(fbarchard): Write function that can handle rotation and scaling.
+
 #ifndef MEDIA_BASE_YUV_ROW_H_
 #define MEDIA_BASE_YUV_ROW_H_
 
 #include "base/basictypes.h"
 
-namespace media {
-
+extern "C" {
 // Can only do 1x.
 // This is the second fastest of the scalers.
 void FastConvertYUVToRGB32Row(const uint8* y_buf,
@@ -59,11 +60,13 @@ void ScaleYUVToRGB32Row(const uint8* y_buf,
                         uint8* rgb_buf,
                         int width,
                         int scaled_dx);
+}  // extern "C"
 
-// MMX for Windows; C++ for other platforms.
-#ifndef USE_MMX
+#if !defined(USE_MMX)
 #if defined(_MSC_VER)
 #define USE_MMX 1
+#elif defined(OS_LINUX) && !defined(ARCH_CPU_X86_64)
+#define USE_MMX 1
 #else
 #define USE_MMX 0
 #endif
@@ -79,7 +82,5 @@ void ScaleYUVToRGB32Row(const uint8* y_buf,
 #define EMMS()
 #endif
 
-}  // namespace media
-
 #endif  // MEDIA_BASE_YUV_ROW_H_
 
diff --git a/media/base/yuv_row_linux.cc b/media/base/yuv_row_linux.cc
index eb3d07e..9d567ad 100644
--- a/media/base/yuv_row_linux.cc
+++ b/media/base/yuv_row_linux.cc
@@ -4,24 +4,435 @@
 
 #include "media/base/yuv_row.h"
 
-// Enable bilinear filtering by turning on the following macro.
-//  #define MEDIA_BILINEAR_FILTER 1
-
 #ifdef _DEBUG
 #include "base/logging.h"
 #else
 #define DCHECK(a)
 #endif
 
-namespace media {
+// TODO(fbarchard): Move this to yuv_row_posix.cc to share with Mac.
+// TODO(fbarchard): Do 64 bit version.
+
+extern "C" {
+#if USE_MMX
+
+#define RGBY(i) { \
+  static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \
+  static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \
+  static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \
+  0 \
+}
+
+#define RGBU(i) { \
+  static_cast<int16>(2.018 * 64 * (i - 128) + 0.5), \
+  static_cast<int16>(-0.391 * 64 * (i - 128) + 0.5), \
+  0, \
+  static_cast<int16>(256 * 64 - 1) \
+}
+
+#define RGBV(i) { \
+  0, \
+  static_cast<int16>(-0.813 * 64 * (i - 128) + 0.5), \
+  static_cast<int16>(1.596 * 64 * (i - 128) + 0.5), \
+  0 \
+}
+
+#define MMX_ALIGNED(var) var __attribute__((aligned(16)))
+
+MMX_ALIGNED(int16 kCoefficientsRgbY[256][4]) = {
+  RGBY(0x00), RGBY(0x01), RGBY(0x02), RGBY(0x03),
+  RGBY(0x04), RGBY(0x05), RGBY(0x06), RGBY(0x07),
+  RGBY(0x08), RGBY(0x09), RGBY(0x0A), RGBY(0x0B),
+  RGBY(0x0C), RGBY(0x0D), RGBY(0x0E), RGBY(0x0F),
+  RGBY(0x10), RGBY(0x11), RGBY(0x12), RGBY(0x13),
+  RGBY(0x14), RGBY(0x15), RGBY(0x16), RGBY(0x17),
+  RGBY(0x18), RGBY(0x19), RGBY(0x1A), RGBY(0x1B),
+  RGBY(0x1C), RGBY(0x1D), RGBY(0x1E), RGBY(0x1F),
+  RGBY(0x20), RGBY(0x21), RGBY(0x22), RGBY(0x23),
+  RGBY(0x24), RGBY(0x25), RGBY(0x26), RGBY(0x27),
+  RGBY(0x28), RGBY(0x29), RGBY(0x2A), RGBY(0x2B),
+  RGBY(0x2C), RGBY(0x2D), RGBY(0x2E), RGBY(0x2F),
+  RGBY(0x30), RGBY(0x31), RGBY(0x32), RGBY(0x33),
+  RGBY(0x34), RGBY(0x35), RGBY(0x36), RGBY(0x37),
+  RGBY(0x38), RGBY(0x39), RGBY(0x3A), RGBY(0x3B),
+  RGBY(0x3C), RGBY(0x3D), RGBY(0x3E), RGBY(0x3F),
+  RGBY(0x40), RGBY(0x41), RGBY(0x42), RGBY(0x43),
+  RGBY(0x44), RGBY(0x45), RGBY(0x46), RGBY(0x47),
+  RGBY(0x48), RGBY(0x49), RGBY(0x4A), RGBY(0x4B),
+  RGBY(0x4C), RGBY(0x4D), RGBY(0x4E), RGBY(0x4F),
+  RGBY(0x50), RGBY(0x51), RGBY(0x52), RGBY(0x53),
+  RGBY(0x54), RGBY(0x55), RGBY(0x56), RGBY(0x57),
+  RGBY(0x58), RGBY(0x59), RGBY(0x5A), RGBY(0x5B),
+  RGBY(0x5C), RGBY(0x5D), RGBY(0x5E), RGBY(0x5F),
+  RGBY(0x60), RGBY(0x61), RGBY(0x62), RGBY(0x63),
+  RGBY(0x64), RGBY(0x65), RGBY(0x66), RGBY(0x67),
+  RGBY(0x68), RGBY(0x69), RGBY(0x6A), RGBY(0x6B),
+  RGBY(0x6C), RGBY(0x6D), RGBY(0x6E), RGBY(0x6F),
+  RGBY(0x70), RGBY(0x71), RGBY(0x72), RGBY(0x73),
+  RGBY(0x74), RGBY(0x75), RGBY(0x76), RGBY(0x77),
+  RGBY(0x78), RGBY(0x79), RGBY(0x7A), RGBY(0x7B),
+  RGBY(0x7C), RGBY(0x7D), RGBY(0x7E), RGBY(0x7F),
+  RGBY(0x80), RGBY(0x81), RGBY(0x82), RGBY(0x83),
+  RGBY(0x84), RGBY(0x85), RGBY(0x86), RGBY(0x87),
+  RGBY(0x88), RGBY(0x89), RGBY(0x8A), RGBY(0x8B),
+  RGBY(0x8C), RGBY(0x8D), RGBY(0x8E), RGBY(0x8F),
+  RGBY(0x90), RGBY(0x91), RGBY(0x92), RGBY(0x93),
+  RGBY(0x94), RGBY(0x95), RGBY(0x96), RGBY(0x97),
+  RGBY(0x98), RGBY(0x99), RGBY(0x9A), RGBY(0x9B),
+  RGBY(0x9C), RGBY(0x9D), RGBY(0x9E), RGBY(0x9F),
+  RGBY(0xA0), RGBY(0xA1), RGBY(0xA2), RGBY(0xA3),
+  RGBY(0xA4), RGBY(0xA5), RGBY(0xA6), RGBY(0xA7),
+  RGBY(0xA8), RGBY(0xA9), RGBY(0xAA), RGBY(0xAB),
+  RGBY(0xAC), RGBY(0xAD), RGBY(0xAE), RGBY(0xAF),
+  RGBY(0xB0), RGBY(0xB1), RGBY(0xB2), RGBY(0xB3),
+  RGBY(0xB4), RGBY(0xB5), RGBY(0xB6), RGBY(0xB7),
+  RGBY(0xB8), RGBY(0xB9), RGBY(0xBA), RGBY(0xBB),
+  RGBY(0xBC), RGBY(0xBD), RGBY(0xBE), RGBY(0xBF),
+  RGBY(0xC0), RGBY(0xC1), RGBY(0xC2), RGBY(0xC3),
+  RGBY(0xC4), RGBY(0xC5), RGBY(0xC6), RGBY(0xC7),
+  RGBY(0xC8), RGBY(0xC9), RGBY(0xCA), RGBY(0xCB),
+  RGBY(0xCC), RGBY(0xCD), RGBY(0xCE), RGBY(0xCF),
+  RGBY(0xD0), RGBY(0xD1), RGBY(0xD2), RGBY(0xD3),
+  RGBY(0xD4), RGBY(0xD5), RGBY(0xD6), RGBY(0xD7),
+  RGBY(0xD8), RGBY(0xD9), RGBY(0xDA), RGBY(0xDB),
+  RGBY(0xDC), RGBY(0xDD), RGBY(0xDE), RGBY(0xDF),
+  RGBY(0xE0), RGBY(0xE1), RGBY(0xE2), RGBY(0xE3),
+  RGBY(0xE4), RGBY(0xE5), RGBY(0xE6), RGBY(0xE7),
+  RGBY(0xE8), RGBY(0xE9), RGBY(0xEA), RGBY(0xEB),
+  RGBY(0xEC), RGBY(0xED), RGBY(0xEE), RGBY(0xEF),
+  RGBY(0xF0), RGBY(0xF1), RGBY(0xF2), RGBY(0xF3),
+  RGBY(0xF4), RGBY(0xF5), RGBY(0xF6), RGBY(0xF7),
+  RGBY(0xF8), RGBY(0xF9), RGBY(0xFA), RGBY(0xFB),
+  RGBY(0xFC), RGBY(0xFD), RGBY(0xFE), RGBY(0xFF),
+};
+
+MMX_ALIGNED(int16 kCoefficientsRgbU[256][4]) = {
+  RGBU(0x00), RGBU(0x01), RGBU(0x02), RGBU(0x03),
+  RGBU(0x04), RGBU(0x05), RGBU(0x06), RGBU(0x07),
+  RGBU(0x08), RGBU(0x09), RGBU(0x0A), RGBU(0x0B),
+  RGBU(0x0C), RGBU(0x0D), RGBU(0x0E), RGBU(0x0F),
+  RGBU(0x10), RGBU(0x11), RGBU(0x12), RGBU(0x13),
+  RGBU(0x14), RGBU(0x15), RGBU(0x16), RGBU(0x17),
+  RGBU(0x18), RGBU(0x19), RGBU(0x1A), RGBU(0x1B),
+  RGBU(0x1C), RGBU(0x1D), RGBU(0x1E), RGBU(0x1F),
+  RGBU(0x20), RGBU(0x21), RGBU(0x22), RGBU(0x23),
+  RGBU(0x24), RGBU(0x25), RGBU(0x26), RGBU(0x27),
+  RGBU(0x28), RGBU(0x29), RGBU(0x2A), RGBU(0x2B),
+  RGBU(0x2C), RGBU(0x2D), RGBU(0x2E), RGBU(0x2F),
+  RGBU(0x30), RGBU(0x31), RGBU(0x32), RGBU(0x33),
+  RGBU(0x34), RGBU(0x35), RGBU(0x36), RGBU(0x37),
+  RGBU(0x38), RGBU(0x39), RGBU(0x3A), RGBU(0x3B),
+  RGBU(0x3C), RGBU(0x3D), RGBU(0x3E), RGBU(0x3F),
+  RGBU(0x40), RGBU(0x41), RGBU(0x42), RGBU(0x43),
+  RGBU(0x44), RGBU(0x45), RGBU(0x46), RGBU(0x47),
+  RGBU(0x48), RGBU(0x49), RGBU(0x4A), RGBU(0x4B),
+  RGBU(0x4C), RGBU(0x4D), RGBU(0x4E), RGBU(0x4F),
+  RGBU(0x50), RGBU(0x51), RGBU(0x52), RGBU(0x53),
+  RGBU(0x54), RGBU(0x55), RGBU(0x56), RGBU(0x57),
+  RGBU(0x58), RGBU(0x59), RGBU(0x5A), RGBU(0x5B),
+  RGBU(0x5C), RGBU(0x5D), RGBU(0x5E), RGBU(0x5F),
+  RGBU(0x60), RGBU(0x61), RGBU(0x62), RGBU(0x63),
+  RGBU(0x64), RGBU(0x65), RGBU(0x66), RGBU(0x67),
+  RGBU(0x68), RGBU(0x69), RGBU(0x6A), RGBU(0x6B),
+  RGBU(0x6C), RGBU(0x6D), RGBU(0x6E), RGBU(0x6F),
+  RGBU(0x70), RGBU(0x71), RGBU(0x72), RGBU(0x73),
+  RGBU(0x74), RGBU(0x75), RGBU(0x76), RGBU(0x77),
+  RGBU(0x78), RGBU(0x79), RGBU(0x7A), RGBU(0x7B),
+  RGBU(0x7C), RGBU(0x7D), RGBU(0x7E), RGBU(0x7F),
+  RGBU(0x80), RGBU(0x81), RGBU(0x82), RGBU(0x83),
+  RGBU(0x84), RGBU(0x85), RGBU(0x86), RGBU(0x87),
+  RGBU(0x88), RGBU(0x89), RGBU(0x8A), RGBU(0x8B),
+  RGBU(0x8C), RGBU(0x8D), RGBU(0x8E), RGBU(0x8F),
+  RGBU(0x90), RGBU(0x91), RGBU(0x92), RGBU(0x93),
+  RGBU(0x94), RGBU(0x95), RGBU(0x96), RGBU(0x97),
+  RGBU(0x98), RGBU(0x99), RGBU(0x9A), RGBU(0x9B),
+  RGBU(0x9C), RGBU(0x9D), RGBU(0x9E), RGBU(0x9F),
+  RGBU(0xA0), RGBU(0xA1), RGBU(0xA2), RGBU(0xA3),
+  RGBU(0xA4), RGBU(0xA5), RGBU(0xA6), RGBU(0xA7),
+  RGBU(0xA8), RGBU(0xA9), RGBU(0xAA), RGBU(0xAB),
+  RGBU(0xAC), RGBU(0xAD), RGBU(0xAE), RGBU(0xAF),
+  RGBU(0xB0), RGBU(0xB1), RGBU(0xB2), RGBU(0xB3),
+  RGBU(0xB4), RGBU(0xB5), RGBU(0xB6), RGBU(0xB7),
+  RGBU(0xB8), RGBU(0xB9), RGBU(0xBA), RGBU(0xBB),
+  RGBU(0xBC), RGBU(0xBD), RGBU(0xBE), RGBU(0xBF),
+  RGBU(0xC0), RGBU(0xC1), RGBU(0xC2), RGBU(0xC3),
+  RGBU(0xC4), RGBU(0xC5), RGBU(0xC6), RGBU(0xC7),
+  RGBU(0xC8), RGBU(0xC9), RGBU(0xCA), RGBU(0xCB),
+  RGBU(0xCC), RGBU(0xCD), RGBU(0xCE), RGBU(0xCF),
+  RGBU(0xD0), RGBU(0xD1), RGBU(0xD2), RGBU(0xD3),
+  RGBU(0xD4), RGBU(0xD5), RGBU(0xD6), RGBU(0xD7),
+  RGBU(0xD8), RGBU(0xD9), RGBU(0xDA), RGBU(0xDB),
+  RGBU(0xDC), RGBU(0xDD), RGBU(0xDE), RGBU(0xDF),
+  RGBU(0xE0), RGBU(0xE1), RGBU(0xE2), RGBU(0xE3),
+  RGBU(0xE4), RGBU(0xE5), RGBU(0xE6), RGBU(0xE7),
+  RGBU(0xE8), RGBU(0xE9), RGBU(0xEA), RGBU(0xEB),
+  RGBU(0xEC), RGBU(0xED), RGBU(0xEE), RGBU(0xEF),
+  RGBU(0xF0), RGBU(0xF1), RGBU(0xF2), RGBU(0xF3),
+  RGBU(0xF4), RGBU(0xF5), RGBU(0xF6), RGBU(0xF7),
+  RGBU(0xF8), RGBU(0xF9), RGBU(0xFA), RGBU(0xFB),
+  RGBU(0xFC), RGBU(0xFD), RGBU(0xFE), RGBU(0xFF),
+};
+
+MMX_ALIGNED(int16 kCoefficientsRgbV[256][4]) = {
+  RGBV(0x00), RGBV(0x01), RGBV(0x02), RGBV(0x03),
+  RGBV(0x04), RGBV(0x05), RGBV(0x06), RGBV(0x07),
+  RGBV(0x08), RGBV(0x09), RGBV(0x0A), RGBV(0x0B),
+  RGBV(0x0C), RGBV(0x0D), RGBV(0x0E), RGBV(0x0F),
+  RGBV(0x10), RGBV(0x11), RGBV(0x12), RGBV(0x13),
+  RGBV(0x14), RGBV(0x15), RGBV(0x16), RGBV(0x17),
+  RGBV(0x18), RGBV(0x19), RGBV(0x1A), RGBV(0x1B),
+  RGBV(0x1C), RGBV(0x1D), RGBV(0x1E), RGBV(0x1F),
+  RGBV(0x20), RGBV(0x21), RGBV(0x22), RGBV(0x23),
+  RGBV(0x24), RGBV(0x25), RGBV(0x26), RGBV(0x27),
+  RGBV(0x28), RGBV(0x29), RGBV(0x2A), RGBV(0x2B),
+  RGBV(0x2C), RGBV(0x2D), RGBV(0x2E), RGBV(0x2F),
+  RGBV(0x30), RGBV(0x31), RGBV(0x32), RGBV(0x33),
+  RGBV(0x34), RGBV(0x35), RGBV(0x36), RGBV(0x37),
+  RGBV(0x38), RGBV(0x39), RGBV(0x3A), RGBV(0x3B),
+  RGBV(0x3C), RGBV(0x3D), RGBV(0x3E), RGBV(0x3F),
+  RGBV(0x40), RGBV(0x41), RGBV(0x42), RGBV(0x43),
+  RGBV(0x44), RGBV(0x45), RGBV(0x46), RGBV(0x47),
+  RGBV(0x48), RGBV(0x49), RGBV(0x4A), RGBV(0x4B),
+  RGBV(0x4C), RGBV(0x4D), RGBV(0x4E), RGBV(0x4F),
+  RGBV(0x50), RGBV(0x51), RGBV(0x52), RGBV(0x53),
+  RGBV(0x54), RGBV(0x55), RGBV(0x56), RGBV(0x57),
+  RGBV(0x58), RGBV(0x59), RGBV(0x5A), RGBV(0x5B),
+  RGBV(0x5C), RGBV(0x5D), RGBV(0x5E), RGBV(0x5F),
+  RGBV(0x60), RGBV(0x61), RGBV(0x62), RGBV(0x63),
+  RGBV(0x64), RGBV(0x65), RGBV(0x66), RGBV(0x67),
+  RGBV(0x68), RGBV(0x69), RGBV(0x6A), RGBV(0x6B),
+  RGBV(0x6C), RGBV(0x6D), RGBV(0x6E), RGBV(0x6F),
+  RGBV(0x70), RGBV(0x71), RGBV(0x72), RGBV(0x73),
+  RGBV(0x74), RGBV(0x75), RGBV(0x76), RGBV(0x77),
+  RGBV(0x78), RGBV(0x79), RGBV(0x7A), RGBV(0x7B),
+  RGBV(0x7C), RGBV(0x7D), RGBV(0x7E), RGBV(0x7F),
+  RGBV(0x80), RGBV(0x81), RGBV(0x82), RGBV(0x83),
+  RGBV(0x84), RGBV(0x85), RGBV(0x86), RGBV(0x87),
+  RGBV(0x88), RGBV(0x89), RGBV(0x8A), RGBV(0x8B),
+  RGBV(0x8C), RGBV(0x8D), RGBV(0x8E), RGBV(0x8F),
+  RGBV(0x90), RGBV(0x91), RGBV(0x92), RGBV(0x93),
+  RGBV(0x94), RGBV(0x95), RGBV(0x96), RGBV(0x97),
+  RGBV(0x98), RGBV(0x99), RGBV(0x9A), RGBV(0x9B),
+  RGBV(0x9C), RGBV(0x9D), RGBV(0x9E), RGBV(0x9F),
+  RGBV(0xA0), RGBV(0xA1), RGBV(0xA2), RGBV(0xA3),
+  RGBV(0xA4), RGBV(0xA5), RGBV(0xA6), RGBV(0xA7),
+  RGBV(0xA8), RGBV(0xA9), RGBV(0xAA), RGBV(0xAB),
+  RGBV(0xAC), RGBV(0xAD), RGBV(0xAE), RGBV(0xAF),
+  RGBV(0xB0), RGBV(0xB1), RGBV(0xB2), RGBV(0xB3),
+  RGBV(0xB4), RGBV(0xB5), RGBV(0xB6), RGBV(0xB7),
+  RGBV(0xB8), RGBV(0xB9), RGBV(0xBA), RGBV(0xBB),
+  RGBV(0xBC), RGBV(0xBD), RGBV(0xBE), RGBV(0xBF),
+  RGBV(0xC0), RGBV(0xC1), RGBV(0xC2), RGBV(0xC3),
+  RGBV(0xC4), RGBV(0xC5), RGBV(0xC6), RGBV(0xC7),
+  RGBV(0xC8), RGBV(0xC9), RGBV(0xCA), RGBV(0xCB),
+  RGBV(0xCC), RGBV(0xCD), RGBV(0xCE), RGBV(0xCF),
+  RGBV(0xD0), RGBV(0xD1), RGBV(0xD2), RGBV(0xD3),
+  RGBV(0xD4), RGBV(0xD5), RGBV(0xD6), RGBV(0xD7),
+  RGBV(0xD8), RGBV(0xD9), RGBV(0xDA), RGBV(0xDB),
+  RGBV(0xDC), RGBV(0xDD), RGBV(0xDE), RGBV(0xDF),
+  RGBV(0xE0), RGBV(0xE1), RGBV(0xE2), RGBV(0xE3),
+  RGBV(0xE4), RGBV(0xE5), RGBV(0xE6), RGBV(0xE7),
+  RGBV(0xE8), RGBV(0xE9), RGBV(0xEA), RGBV(0xEB),
+  RGBV(0xEC), RGBV(0xED), RGBV(0xEE), RGBV(0xEF),
+  RGBV(0xF0), RGBV(0xF1), RGBV(0xF2), RGBV(0xF3),
+  RGBV(0xF4), RGBV(0xF5), RGBV(0xF6), RGBV(0xF7),
+  RGBV(0xF8), RGBV(0xF9), RGBV(0xFA), RGBV(0xFB),
+  RGBV(0xFC), RGBV(0xFD), RGBV(0xFE), RGBV(0xFF),
+};
+
+#undef RGBY
+#undef RGBU
+#undef RGBV
+#undef MMX_ALIGNED
+
+// TODO(fbarchard): Use the following function instead of
+// pure assembly to help make code more portable to 64 bit
+// and Mac, which has different labels.
+// no-gcse eliminates the frame pointer, freeing up ebp.
+
+#if defined(FUTURE_64BIT_VERSION)
+void __attribute__((optimize("O2", "no-gcse")))
+    NewFastConvertYUVToRGB32Row(const uint8* y_buf,
+                                const uint8* u_buf,
+                                const uint8* v_buf,
+                                uint8* rgb_buf,
+                                int width)  {
+  asm(
+  "shr    %4\n"
+"1:\n"
+  "movzb  (%1),%%eax\n"
+  "add    $0x1,%1\n"
+  "movzb  (%2),%%ebx\n"
+  "add    $0x1,%2\n"
+  "movq   kCoefficientsRgbU(,%%eax,8),%%mm0\n"
+  "movzb  (%0),%%eax\n"
+  "paddsw kCoefficientsRgbV(,%%ebx,8),%%mm0\n"
+  "movzb  0x1(%0),%%ebx\n"
+  "movq   kCoefficientsRgbY(,%%eax,8),%%mm1\n"
+  "add    $0x2,%0\n"
+  "movq   kCoefficientsRgbY(,%%ebx,8),%%mm2\n"
+  "paddsw %%mm0,%%mm1\n"
+  "paddsw %%mm0,%%mm2\n"
+  "psraw  $0x6,%%mm1\n"
+  "psraw  $0x6,%%mm2\n"
+  "packuswb %%mm2,%%mm1\n"
+  "movntq %%mm1,0x0(%3)\n"
+  "add    $0x8,%3\n"
+  "sub    $0x1,%4\n"
+  "jne    1b\n"
+      : : "r"(y_buf),"r"(u_buf),"r"(v_buf),"r"(rgb_buf),"r"(width)
+      : "eax","ebx");
+}
+#endif
+
+void FastConvertYUVToRGB32Row(const uint8* y_buf,
+                              const uint8* u_buf,
+                              const uint8* v_buf,
+                              uint8* rgb_buf,
+                              int width);
+
+  asm(
+  ".global FastConvertYUVToRGB32Row\n"
+"FastConvertYUVToRGB32Row:\n"
+  "pusha\n"
+  "mov    0x24(%esp),%edx\n"
+  "mov    0x28(%esp),%edi\n"
+  "mov    0x2c(%esp),%esi\n"
+  "mov    0x30(%esp),%ebp\n"
+  "mov    0x34(%esp),%ecx\n"
+  "jmp    convertend\n"
+
+"convertloop:"
+  "movzbl (%edi),%eax\n"
+  "add    $0x1,%edi\n"
+  "movzbl (%esi),%ebx\n"
+  "add    $0x1,%esi\n"
+  "movq   kCoefficientsRgbU(,%eax,8),%mm0\n"
+  "movzbl (%edx),%eax\n"
+  "paddsw kCoefficientsRgbV(,%ebx,8),%mm0\n"
+  "movzbl 0x1(%edx),%ebx\n"
+  "movq   kCoefficientsRgbY(,%eax,8),%mm1\n"
+  "add    $0x2,%edx\n"
+  "movq   kCoefficientsRgbY(,%ebx,8),%mm2\n"
+  "paddsw %mm0,%mm1\n"
+  "paddsw %mm0,%mm2\n"
+  "psraw  $0x6,%mm1\n"
+  "psraw  $0x6,%mm2\n"
+  "packuswb %mm2,%mm1\n"
+  "movntq %mm1,0x0(%ebp)\n"
+  "add    $0x8,%ebp\n"
+"convertend:"
+  "sub    $0x2,%ecx\n"
+  "jns    convertloop\n"
+
+  "and    $0x1,%ecx\n"
+  "je     convertdone\n"
+
+  "movzbl (%edi),%eax\n"
+  "movq   kCoefficientsRgbU(,%eax,8),%mm0\n"
+  "movzbl (%esi),%eax\n"
+  "paddsw kCoefficientsRgbV(,%eax,8),%mm0\n"
+  "movzbl (%edx),%eax\n"
+  "movq   kCoefficientsRgbY(,%eax,8),%mm1\n"
+  "paddsw %mm0,%mm1\n"
+  "psraw  $0x6,%mm1\n"
+  "packuswb %mm1,%mm1\n"
+  "movd   %mm1,0x0(%ebp)\n"
+"convertdone:"
+  "popa\n"
+  "ret\n"
+);
+
+
+void ScaleYUVToRGB32Row(const uint8* y_buf,
+                        const uint8* u_buf,
+                        const uint8* v_buf,
+                        uint8* rgb_buf,
+                        int width,
+                        int scaled_dx);
+
+  asm(
+  ".global ScaleYUVToRGB32Row\n"
+"ScaleYUVToRGB32Row:\n"
+  "pusha\n"
+  "mov    0x24(%esp),%edx\n"
+  "mov    0x28(%esp),%edi\n"
+  "mov    0x2c(%esp),%esi\n"
+  "mov    0x30(%esp),%ebp\n"
+  "mov    0x34(%esp),%ecx\n"
+  "xor    %ebx,%ebx\n"
+  "jmp    scaleend\n"
+
+"scaleloop:"
+  "mov    %ebx,%eax\n"
+  "sar    $0x5,%eax\n"
+  "movzbl (%edi,%eax,1),%eax\n"
+  "movq   kCoefficientsRgbU(,%eax,8),%mm0\n"
+  "mov    %ebx,%eax\n"
+  "sar    $0x5,%eax\n"
+  "movzbl (%esi,%eax,1),%eax\n"
+  "paddsw kCoefficientsRgbV(,%eax,8),%mm0\n"
+  "mov    %ebx,%eax\n"
+  "add    0x38(%esp),%ebx\n"
+  "sar    $0x4,%eax\n"
+  "movzbl (%edx,%eax,1),%eax\n"
+  "movq   kCoefficientsRgbY(,%eax,8),%mm1\n"
+  "mov    %ebx,%eax\n"
+  "add    0x38(%esp),%ebx\n"
+  "sar    $0x4,%eax\n"
+  "movzbl (%edx,%eax,1),%eax\n"
+  "movq   kCoefficientsRgbY(,%eax,8),%mm2\n"
+  "paddsw %mm0,%mm1\n"
+  "paddsw %mm0,%mm2\n"
+  "psraw  $0x6,%mm1\n"
+  "psraw  $0x6,%mm2\n"
+  "packuswb %mm2,%mm1\n"
+  "movntq %mm1,0x0(%ebp)\n"
+  "add    $0x8,%ebp\n"
+"scaleend:"
+  "sub    $0x2,%ecx\n"
+  "jns    scaleloop\n"
+
+  "and    $0x1,%ecx\n"
+  "je     scaledone\n"
+
+  "mov    %ebx,%eax\n"
+  "sar    $0x5,%eax\n"
+  "movzbl (%edi,%eax,1),%eax\n"
+  "movq   kCoefficientsRgbU(,%eax,8),%mm0\n"
+  "mov    %ebx,%eax\n"
+  "sar    $0x5,%eax\n"
+  "movzbl (%esi,%eax,1),%eax\n"
+  "paddsw kCoefficientsRgbV(,%eax,8),%mm0\n"
+  "mov    %ebx,%eax\n"
+  "sar    $0x4,%eax\n"
+  "movzbl (%edx,%eax,1),%eax\n"
+  "movq   kCoefficientsRgbY(,%eax,8),%mm1\n"
+  "mov    %ebx,%eax\n"
+  "sar    $0x4,%eax\n"
+  "movzbl (%edx,%eax,1),%eax\n"
+  "movq   kCoefficientsRgbY(,%eax,8),%mm2\n"
+  "paddsw %mm0,%mm1\n"
+  "paddsw %mm0,%mm2\n"
+  "psraw  $0x6,%mm1\n"
+  "psraw  $0x6,%mm2\n"
+  "packuswb %mm2,%mm1\n"
+  "movd   %mm1,0x0(%ebp)\n"
+
+"scaledone:"
+  "popa\n"
+  "ret\n"
+);
+
+#else  // USE_MMX
 
 // Reference version of YUV converter.
 static const int kClipTableSize = 256;
 static const int kClipOverflow = 288;  // Cb max is 535.
 
-static uint8 g_rgb_clip_table[kClipOverflow +
-                              kClipTableSize +
-                              kClipOverflow] = {
+static uint8 kRgbClipTable[kClipOverflow +
+                           kClipTableSize +
+                           kClipOverflow] = {
   0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,  // 288 underflow values
   0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,  // clipped to 0.
   0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
@@ -137,7 +548,7 @@ static inline uint32 clip(int32 value) {
   DCHECK(((value >> 8) + kClipOverflow) >= 0);
   DCHECK(((value >> 8) + kClipOverflow) <
          (kClipOverflow + kClipTableSize + kClipOverflow));
-  return static_cast<uint32>(g_rgb_clip_table[((value) >> 8) + kClipOverflow]);
+  return static_cast<uint32>(kRgbClipTable[((value) >> 8) + kClipOverflow]);
 }
 
 static inline void YuvPixel(uint8 y,
@@ -196,6 +607,6 @@ void ScaleYUVToRGB32Row(const uint8* y_buf,
     scaled_x += scaled_dx;
   }
 }
-
-}  // namespace media
+#endif // USE_MMX
+}  // extern "C"
 
diff --git a/media/base/yuv_row_mac.cc b/media/base/yuv_row_mac.cc
index eb3d07e..c80b213 100644
--- a/media/base/yuv_row_mac.cc
+++ b/media/base/yuv_row_mac.cc
@@ -4,24 +4,435 @@
 
 #include "media/base/yuv_row.h"
 
-// Enable bilinear filtering by turning on the following macro.
-//  #define MEDIA_BILINEAR_FILTER 1
-
 #ifdef _DEBUG
 #include "base/logging.h"
 #else
 #define DCHECK(a)
 #endif
 
-namespace media {
+// TODO(fbarchard): Make MMX work in DLLs.  Currently only works in unittests.
+// TODO(fbarchard): Do 64 bit version.
+
+extern "C" {
+#if USE_MMX
+
+#define RGBY(i) { \
+  static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \
+  static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \
+  static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \
+  0 \
+}
+
+#define RGBU(i) { \
+  static_cast<int16>(2.018 * 64 * (i - 128) + 0.5), \
+  static_cast<int16>(-0.391 * 64 * (i - 128) + 0.5), \
+  0, \
+  static_cast<int16>(256 * 64 - 1) \
+}
+
+#define RGBV(i) { \
+  0, \
+  static_cast<int16>(-0.813 * 64 * (i - 128) + 0.5), \
+  static_cast<int16>(1.596 * 64 * (i - 128) + 0.5), \
+  0 \
+}
+
+#define MMX_ALIGNED(var) var __attribute__((aligned(16)))
+
+MMX_ALIGNED(int16 kCoefficientsRgbY[256][4]) = {
+  RGBY(0x00), RGBY(0x01), RGBY(0x02), RGBY(0x03),
+  RGBY(0x04), RGBY(0x05), RGBY(0x06), RGBY(0x07),
+  RGBY(0x08), RGBY(0x09), RGBY(0x0A), RGBY(0x0B),
+  RGBY(0x0C), RGBY(0x0D), RGBY(0x0E), RGBY(0x0F),
+  RGBY(0x10), RGBY(0x11), RGBY(0x12), RGBY(0x13),
+  RGBY(0x14), RGBY(0x15), RGBY(0x16), RGBY(0x17),
+  RGBY(0x18), RGBY(0x19), RGBY(0x1A), RGBY(0x1B),
+  RGBY(0x1C), RGBY(0x1D), RGBY(0x1E), RGBY(0x1F),
+  RGBY(0x20), RGBY(0x21), RGBY(0x22), RGBY(0x23),
+  RGBY(0x24), RGBY(0x25), RGBY(0x26), RGBY(0x27),
+  RGBY(0x28), RGBY(0x29), RGBY(0x2A), RGBY(0x2B),
+  RGBY(0x2C), RGBY(0x2D), RGBY(0x2E), RGBY(0x2F),
+  RGBY(0x30), RGBY(0x31), RGBY(0x32), RGBY(0x33),
+  RGBY(0x34), RGBY(0x35), RGBY(0x36), RGBY(0x37),
+  RGBY(0x38), RGBY(0x39), RGBY(0x3A), RGBY(0x3B),
+  RGBY(0x3C), RGBY(0x3D), RGBY(0x3E), RGBY(0x3F),
+  RGBY(0x40), RGBY(0x41), RGBY(0x42), RGBY(0x43),
+  RGBY(0x44), RGBY(0x45), RGBY(0x46), RGBY(0x47),
+  RGBY(0x48), RGBY(0x49), RGBY(0x4A), RGBY(0x4B),
+  RGBY(0x4C), RGBY(0x4D), RGBY(0x4E), RGBY(0x4F),
+  RGBY(0x50), RGBY(0x51), RGBY(0x52), RGBY(0x53),
+  RGBY(0x54), RGBY(0x55), RGBY(0x56), RGBY(0x57),
+  RGBY(0x58), RGBY(0x59), RGBY(0x5A), RGBY(0x5B),
+  RGBY(0x5C), RGBY(0x5D), RGBY(0x5E), RGBY(0x5F),
+  RGBY(0x60), RGBY(0x61), RGBY(0x62), RGBY(0x63),
+  RGBY(0x64), RGBY(0x65), RGBY(0x66), RGBY(0x67),
+  RGBY(0x68), RGBY(0x69), RGBY(0x6A), RGBY(0x6B),
+  RGBY(0x6C), RGBY(0x6D), RGBY(0x6E), RGBY(0x6F),
+  RGBY(0x70), RGBY(0x71), RGBY(0x72), RGBY(0x73),
+  RGBY(0x74), RGBY(0x75), RGBY(0x76), RGBY(0x77),
+  RGBY(0x78), RGBY(0x79), RGBY(0x7A), RGBY(0x7B),
+  RGBY(0x7C), RGBY(0x7D), RGBY(0x7E), RGBY(0x7F),
+  RGBY(0x80), RGBY(0x81), RGBY(0x82), RGBY(0x83),
+  RGBY(0x84), RGBY(0x85), RGBY(0x86), RGBY(0x87),
+  RGBY(0x88), RGBY(0x89), RGBY(0x8A), RGBY(0x8B),
+  RGBY(0x8C), RGBY(0x8D), RGBY(0x8E), RGBY(0x8F),
+  RGBY(0x90), RGBY(0x91), RGBY(0x92), RGBY(0x93),
+  RGBY(0x94), RGBY(0x95), RGBY(0x96), RGBY(0x97),
+  RGBY(0x98), RGBY(0x99), RGBY(0x9A), RGBY(0x9B),
+  RGBY(0x9C), RGBY(0x9D), RGBY(0x9E), RGBY(0x9F),
+  RGBY(0xA0), RGBY(0xA1), RGBY(0xA2), RGBY(0xA3),
+  RGBY(0xA4), RGBY(0xA5), RGBY(0xA6), RGBY(0xA7),
+  RGBY(0xA8), RGBY(0xA9), RGBY(0xAA), RGBY(0xAB),
+  RGBY(0xAC), RGBY(0xAD), RGBY(0xAE), RGBY(0xAF),
+  RGBY(0xB0), RGBY(0xB1), RGBY(0xB2), RGBY(0xB3),
+  RGBY(0xB4), RGBY(0xB5), RGBY(0xB6), RGBY(0xB7),
+  RGBY(0xB8), RGBY(0xB9), RGBY(0xBA), RGBY(0xBB),
+  RGBY(0xBC), RGBY(0xBD), RGBY(0xBE), RGBY(0xBF),
+  RGBY(0xC0), RGBY(0xC1), RGBY(0xC2), RGBY(0xC3),
+  RGBY(0xC4), RGBY(0xC5), RGBY(0xC6), RGBY(0xC7),
+  RGBY(0xC8), RGBY(0xC9), RGBY(0xCA), RGBY(0xCB),
+  RGBY(0xCC), RGBY(0xCD), RGBY(0xCE), RGBY(0xCF),
+  RGBY(0xD0), RGBY(0xD1), RGBY(0xD2), RGBY(0xD3),
+  RGBY(0xD4), RGBY(0xD5), RGBY(0xD6), RGBY(0xD7),
+  RGBY(0xD8), RGBY(0xD9), RGBY(0xDA), RGBY(0xDB),
+  RGBY(0xDC), RGBY(0xDD), RGBY(0xDE), RGBY(0xDF),
+  RGBY(0xE0), RGBY(0xE1), RGBY(0xE2), RGBY(0xE3),
+  RGBY(0xE4), RGBY(0xE5), RGBY(0xE6), RGBY(0xE7),
+  RGBY(0xE8), RGBY(0xE9), RGBY(0xEA), RGBY(0xEB),
+  RGBY(0xEC), RGBY(0xED), RGBY(0xEE), RGBY(0xEF),
+  RGBY(0xF0), RGBY(0xF1), RGBY(0xF2), RGBY(0xF3),
+  RGBY(0xF4), RGBY(0xF5), RGBY(0xF6), RGBY(0xF7),
+  RGBY(0xF8), RGBY(0xF9), RGBY(0xFA), RGBY(0xFB),
+  RGBY(0xFC), RGBY(0xFD), RGBY(0xFE), RGBY(0xFF),
+};
+
+MMX_ALIGNED(int16 kCoefficientsRgbU[256][4]) = {
+  RGBU(0x00), RGBU(0x01), RGBU(0x02), RGBU(0x03),
+  RGBU(0x04), RGBU(0x05), RGBU(0x06), RGBU(0x07),
+  RGBU(0x08), RGBU(0x09), RGBU(0x0A), RGBU(0x0B),
+  RGBU(0x0C), RGBU(0x0D), RGBU(0x0E), RGBU(0x0F),
+  RGBU(0x10), RGBU(0x11), RGBU(0x12), RGBU(0x13),
+  RGBU(0x14), RGBU(0x15), RGBU(0x16), RGBU(0x17),
+  RGBU(0x18), RGBU(0x19), RGBU(0x1A), RGBU(0x1B),
+  RGBU(0x1C), RGBU(0x1D), RGBU(0x1E), RGBU(0x1F),
+  RGBU(0x20), RGBU(0x21), RGBU(0x22), RGBU(0x23),
+  RGBU(0x24), RGBU(0x25), RGBU(0x26), RGBU(0x27),
+  RGBU(0x28), RGBU(0x29), RGBU(0x2A), RGBU(0x2B),
+  RGBU(0x2C), RGBU(0x2D), RGBU(0x2E), RGBU(0x2F),
+  RGBU(0x30), RGBU(0x31), RGBU(0x32), RGBU(0x33),
+  RGBU(0x34), RGBU(0x35), RGBU(0x36), RGBU(0x37),
+  RGBU(0x38), RGBU(0x39), RGBU(0x3A), RGBU(0x3B),
+  RGBU(0x3C), RGBU(0x3D), RGBU(0x3E), RGBU(0x3F),
+  RGBU(0x40), RGBU(0x41), RGBU(0x42), RGBU(0x43),
+  RGBU(0x44), RGBU(0x45), RGBU(0x46), RGBU(0x47),
+  RGBU(0x48), RGBU(0x49), RGBU(0x4A), RGBU(0x4B),
+  RGBU(0x4C), RGBU(0x4D), RGBU(0x4E), RGBU(0x4F),
+  RGBU(0x50), RGBU(0x51), RGBU(0x52), RGBU(0x53),
+  RGBU(0x54), RGBU(0x55), RGBU(0x56), RGBU(0x57),
+  RGBU(0x58), RGBU(0x59), RGBU(0x5A), RGBU(0x5B),
+  RGBU(0x5C), RGBU(0x5D), RGBU(0x5E), RGBU(0x5F),
+  RGBU(0x60), RGBU(0x61), RGBU(0x62), RGBU(0x63),
+  RGBU(0x64), RGBU(0x65), RGBU(0x66), RGBU(0x67),
+  RGBU(0x68), RGBU(0x69), RGBU(0x6A), RGBU(0x6B),
+  RGBU(0x6C), RGBU(0x6D), RGBU(0x6E), RGBU(0x6F),
+  RGBU(0x70), RGBU(0x71), RGBU(0x72), RGBU(0x73),
+  RGBU(0x74), RGBU(0x75), RGBU(0x76), RGBU(0x77),
+  RGBU(0x78), RGBU(0x79), RGBU(0x7A), RGBU(0x7B),
+  RGBU(0x7C), RGBU(0x7D), RGBU(0x7E), RGBU(0x7F),
+  RGBU(0x80), RGBU(0x81), RGBU(0x82), RGBU(0x83),
+  RGBU(0x84), RGBU(0x85), RGBU(0x86), RGBU(0x87),
+  RGBU(0x88), RGBU(0x89), RGBU(0x8A), RGBU(0x8B),
+  RGBU(0x8C), RGBU(0x8D), RGBU(0x8E), RGBU(0x8F),
+  RGBU(0x90), RGBU(0x91), RGBU(0x92), RGBU(0x93),
+  RGBU(0x94), RGBU(0x95), RGBU(0x96), RGBU(0x97),
+  RGBU(0x98), RGBU(0x99), RGBU(0x9A), RGBU(0x9B),
+  RGBU(0x9C), RGBU(0x9D), RGBU(0x9E), RGBU(0x9F),
+  RGBU(0xA0), RGBU(0xA1), RGBU(0xA2), RGBU(0xA3),
+  RGBU(0xA4), RGBU(0xA5), RGBU(0xA6), RGBU(0xA7),
+  RGBU(0xA8), RGBU(0xA9), RGBU(0xAA), RGBU(0xAB),
+  RGBU(0xAC), RGBU(0xAD), RGBU(0xAE), RGBU(0xAF),
+  RGBU(0xB0), RGBU(0xB1), RGBU(0xB2), RGBU(0xB3),
+  RGBU(0xB4), RGBU(0xB5), RGBU(0xB6), RGBU(0xB7),
+  RGBU(0xB8), RGBU(0xB9), RGBU(0xBA), RGBU(0xBB),
+  RGBU(0xBC), RGBU(0xBD), RGBU(0xBE), RGBU(0xBF),
+  RGBU(0xC0), RGBU(0xC1), RGBU(0xC2), RGBU(0xC3),
+  RGBU(0xC4), RGBU(0xC5), RGBU(0xC6), RGBU(0xC7),
+  RGBU(0xC8), RGBU(0xC9), RGBU(0xCA), RGBU(0xCB),
+  RGBU(0xCC), RGBU(0xCD), RGBU(0xCE), RGBU(0xCF),
+  RGBU(0xD0), RGBU(0xD1), RGBU(0xD2), RGBU(0xD3),
+  RGBU(0xD4), RGBU(0xD5), RGBU(0xD6), RGBU(0xD7),
+  RGBU(0xD8), RGBU(0xD9), RGBU(0xDA), RGBU(0xDB),
+  RGBU(0xDC), RGBU(0xDD), RGBU(0xDE), RGBU(0xDF),
+  RGBU(0xE0), RGBU(0xE1), RGBU(0xE2), RGBU(0xE3),
+  RGBU(0xE4), RGBU(0xE5), RGBU(0xE6), RGBU(0xE7),
+  RGBU(0xE8), RGBU(0xE9), RGBU(0xEA), RGBU(0xEB),
+  RGBU(0xEC), RGBU(0xED), RGBU(0xEE), RGBU(0xEF),
+  RGBU(0xF0), RGBU(0xF1), RGBU(0xF2), RGBU(0xF3),
+  RGBU(0xF4), RGBU(0xF5), RGBU(0xF6), RGBU(0xF7),
+  RGBU(0xF8), RGBU(0xF9), RGBU(0xFA), RGBU(0xFB),
+  RGBU(0xFC), RGBU(0xFD), RGBU(0xFE), RGBU(0xFF),
+};
+
+MMX_ALIGNED(int16 kCoefficientsRgbV[256][4]) = {
+  RGBV(0x00), RGBV(0x01), RGBV(0x02), RGBV(0x03),
+  RGBV(0x04), RGBV(0x05), RGBV(0x06), RGBV(0x07),
+  RGBV(0x08), RGBV(0x09), RGBV(0x0A), RGBV(0x0B),
+  RGBV(0x0C), RGBV(0x0D), RGBV(0x0E), RGBV(0x0F),
+  RGBV(0x10), RGBV(0x11), RGBV(0x12), RGBV(0x13),
+  RGBV(0x14), RGBV(0x15), RGBV(0x16), RGBV(0x17),
+  RGBV(0x18), RGBV(0x19), RGBV(0x1A), RGBV(0x1B),
+  RGBV(0x1C), RGBV(0x1D), RGBV(0x1E), RGBV(0x1F),
+  RGBV(0x20), RGBV(0x21), RGBV(0x22), RGBV(0x23),
+  RGBV(0x24), RGBV(0x25), RGBV(0x26), RGBV(0x27),
+  RGBV(0x28), RGBV(0x29), RGBV(0x2A), RGBV(0x2B),
+  RGBV(0x2C), RGBV(0x2D), RGBV(0x2E), RGBV(0x2F),
+  RGBV(0x30), RGBV(0x31), RGBV(0x32), RGBV(0x33),
+  RGBV(0x34), RGBV(0x35), RGBV(0x36), RGBV(0x37),
+  RGBV(0x38), RGBV(0x39), RGBV(0x3A), RGBV(0x3B),
+  RGBV(0x3C), RGBV(0x3D), RGBV(0x3E), RGBV(0x3F),
+  RGBV(0x40), RGBV(0x41), RGBV(0x42), RGBV(0x43),
+  RGBV(0x44), RGBV(0x45), RGBV(0x46), RGBV(0x47),
+  RGBV(0x48), RGBV(0x49), RGBV(0x4A), RGBV(0x4B),
+  RGBV(0x4C), RGBV(0x4D), RGBV(0x4E), RGBV(0x4F),
+  RGBV(0x50), RGBV(0x51), RGBV(0x52), RGBV(0x53),
+  RGBV(0x54), RGBV(0x55), RGBV(0x56), RGBV(0x57),
+  RGBV(0x58), RGBV(0x59), RGBV(0x5A), RGBV(0x5B),
+  RGBV(0x5C), RGBV(0x5D), RGBV(0x5E), RGBV(0x5F),
+  RGBV(0x60), RGBV(0x61), RGBV(0x62), RGBV(0x63),
+  RGBV(0x64), RGBV(0x65), RGBV(0x66), RGBV(0x67),
+  RGBV(0x68), RGBV(0x69), RGBV(0x6A), RGBV(0x6B),
+  RGBV(0x6C), RGBV(0x6D), RGBV(0x6E), RGBV(0x6F),
+  RGBV(0x70), RGBV(0x71), RGBV(0x72), RGBV(0x73),
+  RGBV(0x74), RGBV(0x75), RGBV(0x76), RGBV(0x77),
+  RGBV(0x78), RGBV(0x79), RGBV(0x7A), RGBV(0x7B),
+  RGBV(0x7C), RGBV(0x7D), RGBV(0x7E), RGBV(0x7F),
+  RGBV(0x80), RGBV(0x81), RGBV(0x82), RGBV(0x83),
+  RGBV(0x84), RGBV(0x85), RGBV(0x86), RGBV(0x87),
+  RGBV(0x88), RGBV(0x89), RGBV(0x8A), RGBV(0x8B),
+  RGBV(0x8C), RGBV(0x8D), RGBV(0x8E), RGBV(0x8F),
+  RGBV(0x90), RGBV(0x91), RGBV(0x92), RGBV(0x93),
+  RGBV(0x94), RGBV(0x95), RGBV(0x96), RGBV(0x97),
+  RGBV(0x98), RGBV(0x99), RGBV(0x9A), RGBV(0x9B),
+  RGBV(0x9C), RGBV(0x9D), RGBV(0x9E), RGBV(0x9F),
+  RGBV(0xA0), RGBV(0xA1), RGBV(0xA2), RGBV(0xA3),
+  RGBV(0xA4), RGBV(0xA5), RGBV(0xA6), RGBV(0xA7),
+  RGBV(0xA8), RGBV(0xA9), RGBV(0xAA), RGBV(0xAB),
+  RGBV(0xAC), RGBV(0xAD), RGBV(0xAE), RGBV(0xAF),
+  RGBV(0xB0), RGBV(0xB1), RGBV(0xB2), RGBV(0xB3),
+  RGBV(0xB4), RGBV(0xB5), RGBV(0xB6), RGBV(0xB7),
+  RGBV(0xB8), RGBV(0xB9), RGBV(0xBA), RGBV(0xBB),
+  RGBV(0xBC), RGBV(0xBD), RGBV(0xBE), RGBV(0xBF),
+  RGBV(0xC0), RGBV(0xC1), RGBV(0xC2), RGBV(0xC3),
+  RGBV(0xC4), RGBV(0xC5), RGBV(0xC6), RGBV(0xC7),
+  RGBV(0xC8), RGBV(0xC9), RGBV(0xCA), RGBV(0xCB),
+  RGBV(0xCC), RGBV(0xCD), RGBV(0xCE), RGBV(0xCF),
+  RGBV(0xD0), RGBV(0xD1), RGBV(0xD2), RGBV(0xD3),
+  RGBV(0xD4), RGBV(0xD5), RGBV(0xD6), RGBV(0xD7),
+  RGBV(0xD8), RGBV(0xD9), RGBV(0xDA), RGBV(0xDB),
+  RGBV(0xDC), RGBV(0xDD), RGBV(0xDE), RGBV(0xDF),
+  RGBV(0xE0), RGBV(0xE1), RGBV(0xE2), RGBV(0xE3),
+  RGBV(0xE4), RGBV(0xE5), RGBV(0xE6), RGBV(0xE7),
+  RGBV(0xE8), RGBV(0xE9), RGBV(0xEA), RGBV(0xEB),
+  RGBV(0xEC), RGBV(0xED), RGBV(0xEE), RGBV(0xEF),
+  RGBV(0xF0), RGBV(0xF1), RGBV(0xF2), RGBV(0xF3),
+  RGBV(0xF4), RGBV(0xF5), RGBV(0xF6), RGBV(0xF7),
+  RGBV(0xF8), RGBV(0xF9), RGBV(0xFA), RGBV(0xFB),
+  RGBV(0xFC), RGBV(0xFD), RGBV(0xFE), RGBV(0xFF),
+};
+
+#undef RGBY
+#undef RGBU
+#undef RGBV
+#undef MMX_ALIGNED
+
+// TODO(fbarchard): Use the following function instead of
+// pure assembly to help make code more portable to 64 bit
+// and Mac, which has different labels.
+// no-gcse eliminates the frame pointer, freeing up ebp.
+
+#if defined(FUTURE_64BIT_VERSION)
+void __attribute__((optimize("O2", "no-gcse")))
+    NewFastConvertYUVToRGB32Row(const uint8* y_buf,
+                                const uint8* u_buf,
+                                const uint8* v_buf,
+                                uint8* rgb_buf,
+                                int width)  {
+  asm(
+  "shr    %4\n"
+"1:\n"
+  "movzb  (%1),%%eax\n"
+  "add    $0x1,%1\n"
+  "movzb  (%2),%%ebx\n"
+  "add    $0x1,%2\n"
+  "movq   kCoefficientsRgbU(,%%eax,8),%%mm0\n"
+  "movzb  (%0),%%eax\n"
+  "paddsw kCoefficientsRgbV(,%%ebx,8),%%mm0\n"
+  "movzb  0x1(%0),%%ebx\n"
+  "movq   kCoefficientsRgbY(,%%eax,8),%%mm1\n"
+  "add    $0x2,%0\n"
+  "movq   kCoefficientsRgbY(,%%ebx,8),%%mm2\n"
+  "paddsw %%mm0,%%mm1\n"
+  "paddsw %%mm0,%%mm2\n"
+  "psraw  $0x6,%%mm1\n"
+  "psraw  $0x6,%%mm2\n"
+  "packuswb %%mm2,%%mm1\n"
+  "movntq %%mm1,0x0(%3)\n"
+  "add    $0x8,%3\n"
+  "sub    $0x1,%4\n"
+  "jne    1b\n"
+      : : "r"(y_buf),"r"(u_buf),"r"(v_buf),"r"(rgb_buf),"r"(width)
+      : "eax","ebx");
+}
+#endif
+
+extern void FastConvertYUVToRGB32Row(const uint8* y_buf,
+                                     const uint8* u_buf,
+                                     const uint8* v_buf,
+                                     uint8* rgb_buf,
+                                     int width);
+
+  __asm__(
+" .globl _FastConvertYUVToRGB32Row\n"
+"_FastConvertYUVToRGB32Row:\n"
+  "pusha\n"
+  "mov    0x24(%esp),%edx\n"
+  "mov    0x28(%esp),%edi\n"
+  "mov    0x2c(%esp),%esi\n"
+  "mov    0x30(%esp),%ebp\n"
+  "mov    0x34(%esp),%ecx\n"
+  "jmp    convertend\n"
+
+"convertloop:"
+  "movzbl (%edi),%eax\n"
+  "add    $0x1,%edi\n"
+  "movzbl (%esi),%ebx\n"
+  "add    $0x1,%esi\n"
+  "movq   _kCoefficientsRgbU(,%eax,8),%mm0\n"
+  "movzbl (%edx),%eax\n"
+  "paddsw _kCoefficientsRgbV(,%ebx,8),%mm0\n"
+  "movzbl 0x1(%edx),%ebx\n"
+  "movq   _kCoefficientsRgbY(,%eax,8),%mm1\n"
+  "add    $0x2,%edx\n"
+  "movq   _kCoefficientsRgbY(,%ebx,8),%mm2\n"
+  "paddsw %mm0,%mm1\n"
+  "paddsw %mm0,%mm2\n"
+  "psraw  $0x6,%mm1\n"
+  "psraw  $0x6,%mm2\n"
+  "packuswb %mm2,%mm1\n"
+  "movntq %mm1,0x0(%ebp)\n"
+  "add    $0x8,%ebp\n"
+"convertend:"
+  "sub    $0x2,%ecx\n"
+  "jns    convertloop\n"
+
+  "and    $0x1,%ecx\n"
+  "je     convertdone\n"
+
+  "movzbl (%edi),%eax\n"
+  "movq   _kCoefficientsRgbU(,%eax,8),%mm0\n"
+  "movzbl (%esi),%eax\n"
+  "paddsw _kCoefficientsRgbV(,%eax,8),%mm0\n"
+  "movzbl (%edx),%eax\n"
+  "movq   _kCoefficientsRgbY(,%eax,8),%mm1\n"
+  "paddsw %mm0,%mm1\n"
+  "psraw  $0x6,%mm1\n"
+  "packuswb %mm1,%mm1\n"
+  "movd   %mm1,0x0(%ebp)\n"
+"convertdone:"
+  "popa\n"
+  "ret\n"
+);
+
+
+extern void ScaleYUVToRGB32Row(const uint8* y_buf,
+                               const uint8* u_buf,
+                               const uint8* v_buf,
+                               uint8* rgb_buf,
+                               int width,
+                               int scaled_dx);
+
+  __asm__(
+" .globl _ScaleYUVToRGB32Row\n"
+"_ScaleYUVToRGB32Row:\n"
+  "pusha\n"
+  "mov    0x24(%esp),%edx\n"
+  "mov    0x28(%esp),%edi\n"
+  "mov    0x2c(%esp),%esi\n"
+  "mov    0x30(%esp),%ebp\n"
+  "mov    0x34(%esp),%ecx\n"
+  "xor    %ebx,%ebx\n"
+  "jmp    scaleend\n"
+
+"scaleloop:"
+  "mov    %ebx,%eax\n"
+  "sar    $0x5,%eax\n"
+  "movzbl (%edi,%eax,1),%eax\n"
+  "movq   kCoefficientsRgbU(,%eax,8),%mm0\n"
+  "mov    %ebx,%eax\n"
+  "sar    $0x5,%eax\n"
+  "movzbl (%esi,%eax,1),%eax\n"
+  "paddsw kCoefficientsRgbV(,%eax,8),%mm0\n"
+  "mov    %ebx,%eax\n"
+  "add    0x38(%esp),%ebx\n"
+  "sar    $0x4,%eax\n"
+  "movzbl (%edx,%eax,1),%eax\n"
+  "movq   kCoefficientsRgbY(,%eax,8),%mm1\n"
+  "mov    %ebx,%eax\n"
+  "add    0x38(%esp),%ebx\n"
+  "sar    $0x4,%eax\n"
+  "movzbl (%edx,%eax,1),%eax\n"
+  "movq   kCoefficientsRgbY(,%eax,8),%mm2\n"
+  "paddsw %mm0,%mm1\n"
+  "paddsw %mm0,%mm2\n"
+  "psraw  $0x6,%mm1\n"
+  "psraw  $0x6,%mm2\n"
+  "packuswb %mm2,%mm1\n"
+  "movntq %mm1,0x0(%ebp)\n"
+  "add    $0x8,%ebp\n"
+"scaleend:"
+  "sub    $0x2,%ecx\n"
+  "jns    scaleloop\n"
+
+  "and    $0x1,%ecx\n"
+  "je     scaledone\n"
+
+  "mov    %ebx,%eax\n"
+  "sar    $0x5,%eax\n"
+  "movzbl (%edi,%eax,1),%eax\n"
+  "movq   kCoefficientsRgbU(,%eax,8),%mm0\n"
+  "mov    %ebx,%eax\n"
+  "sar    $0x5,%eax\n"
+  "movzbl (%esi,%eax,1),%eax\n"
+  "paddsw kCoefficientsRgbV(,%eax,8),%mm0\n"
+  "mov    %ebx,%eax\n"
+  "sar    $0x4,%eax\n"
+  "movzbl (%edx,%eax,1),%eax\n"
+  "movq   kCoefficientsRgbY(,%eax,8),%mm1\n"
+  "mov    %ebx,%eax\n"
+  "sar    $0x4,%eax\n"
+  "movzbl (%edx,%eax,1),%eax\n"
+  "movq   kCoefficientsRgbY(,%eax,8),%mm2\n"
+  "paddsw %mm0,%mm1\n"
+  "paddsw %mm0,%mm2\n"
+  "psraw  $0x6,%mm1\n"
+  "psraw  $0x6,%mm2\n"
+  "packuswb %mm2,%mm1\n"
+  "movd   %mm1,0x0(%ebp)\n"
+
+"scaledone:"
+  "popa\n"
+  "ret\n"
+);
+
+#else  // USE_MMX
 
 // Reference version of YUV converter.
 static const int kClipTableSize = 256;
 static const int kClipOverflow = 288;  // Cb max is 535.
 
-static uint8 g_rgb_clip_table[kClipOverflow +
-                              kClipTableSize +
-                              kClipOverflow] = {
+static uint8 kRgbClipTable[kClipOverflow +
+                           kClipTableSize +
+                           kClipOverflow] = {
   0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,  // 288 underflow values
   0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,  // clipped to 0.
   0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
@@ -137,7 +548,7 @@ static inline uint32 clip(int32 value) {
   DCHECK(((value >> 8) + kClipOverflow) >= 0);
   DCHECK(((value >> 8) + kClipOverflow) <
          (kClipOverflow + kClipTableSize + kClipOverflow));
-  return static_cast<uint32>(g_rgb_clip_table[((value) >> 8) + kClipOverflow]);
+  return static_cast<uint32>(kRgbClipTable[((value) >> 8) + kClipOverflow]);
 }
 
 static inline void YuvPixel(uint8 y,
@@ -196,6 +607,6 @@ void ScaleYUVToRGB32Row(const uint8* y_buf,
     scaled_x += scaled_dx;
   }
 }
-
-}  // namespace media
+#endif // USE_MMX
+}  // extern "C"
 
diff --git a/media/base/yuv_row_win.cc b/media/base/yuv_row_win.cc
index 53dadc4..e6f15bb 100644
--- a/media/base/yuv_row_win.cc
+++ b/media/base/yuv_row_win.cc
@@ -4,11 +4,7 @@
 
 #include "media/base/yuv_row.h"
 
-// Enable bilinear filtering by turning on the following macro.
-//  #define MEDIA_BILINEAR_FILTER 1
-
-namespace media {
-
+extern "C" {
 #define RGBY(i) { \
   static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \
   static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \
@@ -32,8 +28,7 @@ namespace media {
 
 #define MMX_ALIGNED(var) __declspec(align(16)) var
 
-extern "C" {
-MMX_ALIGNED(int16 coefficients_RGB_Y[256][4]) = {
+MMX_ALIGNED(int16 kCoefficientsRgbY[256][4]) = {
   RGBY(0x00), RGBY(0x01), RGBY(0x02), RGBY(0x03),
   RGBY(0x04), RGBY(0x05), RGBY(0x06), RGBY(0x07),
   RGBY(0x08), RGBY(0x09), RGBY(0x0A), RGBY(0x0B),
@@ -100,7 +95,7 @@ MMX_ALIGNED(int16 coefficients_RGB_Y[256][4]) = {
   RGBY(0xFC), RGBY(0xFD), RGBY(0xFE), RGBY(0xFF),
 };
 
-MMX_ALIGNED(int16 coefficients_RGB_U[256][4]) = {
+MMX_ALIGNED(int16 kCoefficientsRgbU[256][4]) = {
   RGBU(0x00), RGBU(0x01), RGBU(0x02), RGBU(0x03),
   RGBU(0x04), RGBU(0x05), RGBU(0x06), RGBU(0x07),
   RGBU(0x08), RGBU(0x09), RGBU(0x0A), RGBU(0x0B),
@@ -167,7 +162,7 @@ MMX_ALIGNED(int16 coefficients_RGB_U[256][4]) = {
   RGBU(0xFC), RGBU(0xFD), RGBU(0xFE), RGBU(0xFF),
 };
 
-MMX_ALIGNED(int16 coefficients_RGB_V[256][4]) = {
+MMX_ALIGNED(int16 kCoefficientsRgbV[256][4]) = {
   RGBV(0x00), RGBV(0x01), RGBV(0x02), RGBV(0x03),
   RGBV(0x04), RGBV(0x05), RGBV(0x06), RGBV(0x07),
   RGBV(0x08), RGBV(0x09), RGBV(0x0A), RGBV(0x0B),
@@ -233,7 +228,6 @@ MMX_ALIGNED(int16 coefficients_RGB_V[256][4]) = {
   RGBV(0xF8), RGBV(0xF9), RGBV(0xFA), RGBV(0xFB),
   RGBV(0xFC), RGBV(0xFD), RGBV(0xFE), RGBV(0xFF),
 };
-}  // extern "C"
 
 #undef RGBHY
 #undef RGBY
@@ -258,20 +252,20 @@ void FastConvertYUVToRGB32Row(const uint8* y_buf,
     mov       esi, [esp + 32 + 12]  // V
     mov       ebp, [esp + 32 + 16]  // rgb
     mov       ecx, [esp + 32 + 20]  // width
-    jmp       wend
+    jmp       convertend
 
- wloop :
+ convertloop :
     movzx     eax, byte ptr [edi]
     add       edi, 1
     movzx     ebx, byte ptr [esi]
     add       esi, 1
-    movq      mm0, [coefficients_RGB_U + 8 * eax]
+    movq      mm0, [kCoefficientsRgbU + 8 * eax]
     movzx     eax, byte ptr [edx]
-    paddsw    mm0, [coefficients_RGB_V + 8 * ebx]
+    paddsw    mm0, [kCoefficientsRgbV + 8 * ebx]
     movzx     ebx, byte ptr [edx + 1]
-    movq      mm1, [coefficients_RGB_Y + 8 * eax]
+    movq      mm1, [kCoefficientsRgbY + 8 * eax]
     add       edx, 2
-    movq      mm2, [coefficients_RGB_Y + 8 * ebx]
+    movq      mm2, [kCoefficientsRgbY + 8 * ebx]
     paddsw    mm1, mm0
     paddsw    mm2, mm0
     psraw     mm1, 6
@@ -279,24 +273,24 @@ void FastConvertYUVToRGB32Row(const uint8* y_buf,
     packuswb  mm1, mm2
     movntq    [ebp], mm1
     add       ebp, 8
- wend :
+ convertend :
     sub       ecx, 2
-    jns       wloop
+    jns       convertloop
 
     and       ecx, 1  // odd number of pixels?
-    jz        wdone
+    jz        convertdone
 
     movzx     eax, byte ptr [edi]
-    movq      mm0, [coefficients_RGB_U + 8 * eax]
+    movq      mm0, [kCoefficientsRgbU + 8 * eax]
     movzx     eax, byte ptr [esi]
-    paddsw    mm0, [coefficients_RGB_V + 8 * eax]
+    paddsw    mm0, [kCoefficientsRgbV + 8 * eax]
     movzx     eax, byte ptr [edx]
-    movq      mm1, [coefficients_RGB_Y + 8 * eax]
+    movq      mm1, [kCoefficientsRgbY + 8 * eax]
     paddsw    mm1, mm0
     psraw     mm1, 6
     packuswb  mm1, mm1
     movd      [ebp], mm1
- wdone :
+ convertdone :
 
     popad
     ret
@@ -323,16 +317,16 @@ void ConvertYUVToRGB32Row(const uint8* y_buf,
  wloop :
     movzx     eax, byte ptr [edi]
     add       edi, ebx
-    movq      mm0, [coefficients_RGB_U + 8 * eax]
+    movq      mm0, [kCoefficientsRgbU + 8 * eax]
     movzx     eax, byte ptr [esi]
     add       esi, ebx
-    paddsw    mm0, [coefficients_RGB_V + 8 * eax]
+    paddsw    mm0, [kCoefficientsRgbV + 8 * eax]
     movzx     eax, byte ptr [edx]
     add       edx, ebx
-    movq      mm1, [coefficients_RGB_Y + 8 * eax]
+    movq      mm1, [kCoefficientsRgbY + 8 * eax]
     movzx     eax, byte ptr [edx]
     add       edx, ebx
-    movq      mm2, [coefficients_RGB_Y + 8 * eax]
+    movq      mm2, [kCoefficientsRgbY + 8 * eax]
     paddsw    mm1, mm0
     paddsw    mm2, mm0
     psraw     mm1, 6
@@ -348,11 +342,11 @@ void ConvertYUVToRGB32Row(const uint8* y_buf,
     jz        wdone
 
     movzx     eax, byte ptr [edi]
-    movq      mm0, [coefficients_RGB_U + 8 * eax]
+    movq      mm0, [kCoefficientsRgbU + 8 * eax]
     movzx     eax, byte ptr [esi]
-    paddsw    mm0, [coefficients_RGB_V + 8 * eax]
+    paddsw    mm0, [kCoefficientsRgbV + 8 * eax]
     movzx     eax, byte ptr [edx]
-    movq      mm1, [coefficients_RGB_Y + 8 * eax]
+    movq      mm1, [kCoefficientsRgbY + 8 * eax]
     paddsw    mm1, mm0
     psraw     mm1, 6
     packuswb  mm1, mm1
@@ -385,17 +379,17 @@ void RotateConvertYUVToRGB32Row(const uint8* y_buf,
     movzx     eax, byte ptr [edi]
     mov       ebx, [esp + 32 + 28]  // uvstep
     add       edi, ebx
-    movq      mm0, [coefficients_RGB_U + 8 * eax]
+    movq      mm0, [kCoefficientsRgbU + 8 * eax]
     movzx     eax, byte ptr [esi]
     add       esi, ebx
-    paddsw    mm0, [coefficients_RGB_V + 8 * eax]
+    paddsw    mm0, [kCoefficientsRgbV + 8 * eax]
     movzx     eax, byte ptr [edx]
     mov       ebx, [esp + 32 + 24]  // ystep
     add       edx, ebx
-    movq      mm1, [coefficients_RGB_Y + 8 * eax]
+    movq      mm1, [kCoefficientsRgbY + 8 * eax]
     movzx     eax, byte ptr [edx]
     add       edx, ebx
-    movq      mm2, [coefficients_RGB_Y + 8 * eax]
+    movq      mm2, [kCoefficientsRgbY + 8 * eax]
     paddsw    mm1, mm0
     paddsw    mm2, mm0
     psraw     mm1, 6
@@ -411,11 +405,11 @@ void RotateConvertYUVToRGB32Row(const uint8* y_buf,
     jz        wdone
 
     movzx     eax, byte ptr [edi]
-    movq      mm0, [coefficients_RGB_U + 8 * eax]
+    movq      mm0, [kCoefficientsRgbU + 8 * eax]
     movzx     eax, byte ptr [esi]
-    paddsw    mm0, [coefficients_RGB_V + 8 * eax]
+    paddsw    mm0, [kCoefficientsRgbV + 8 * eax]
     movzx     eax, byte ptr [edx]
-    movq      mm1, [coefficients_RGB_Y + 8 * eax]
+    movq      mm1, [kCoefficientsRgbY + 8 * eax]
     paddsw    mm1, mm0
     psraw     mm1, 6
     packuswb  mm1, mm1
@@ -447,10 +441,10 @@ void DoubleYUVToRGB32Row(const uint8* y_buf,
     add       edi, 1
     movzx     ebx, byte ptr [esi]
     add       esi, 1
-    movq      mm0, [coefficients_RGB_U + 8 * eax]
+    movq      mm0, [kCoefficientsRgbU + 8 * eax]
     movzx     eax, byte ptr [edx]
-    paddsw    mm0, [coefficients_RGB_V + 8 * ebx]
-    movq      mm1, [coefficients_RGB_Y + 8 * eax]
+    paddsw    mm0, [kCoefficientsRgbV + 8 * ebx]
+    movq      mm1, [kCoefficientsRgbY + 8 * eax]
     paddsw    mm1, mm0
     psraw     mm1, 6
     packuswb  mm1, mm1
@@ -459,7 +453,7 @@ void DoubleYUVToRGB32Row(const uint8* y_buf,
 
     movzx     ebx, byte ptr [edx + 1]
     add       edx, 2
-    paddsw    mm0, [coefficients_RGB_Y + 8 * ebx]
+    paddsw    mm0, [kCoefficientsRgbY + 8 * ebx]
     psraw     mm0, 6
     packuswb  mm0, mm0
     punpckldq mm0, mm0
@@ -473,11 +467,11 @@ void DoubleYUVToRGB32Row(const uint8* y_buf,
     jz        wdone
 
     movzx     eax, byte ptr [edi]
-    movq      mm0, [coefficients_RGB_U + 8 * eax]
+    movq      mm0, [kCoefficientsRgbU + 8 * eax]
     movzx     eax, byte ptr [esi]
-    paddsw    mm0, [coefficients_RGB_V + 8 * eax]
+    paddsw    mm0, [kCoefficientsRgbV + 8 * eax]
     movzx     eax, byte ptr [edx]
-    movq      mm1, [coefficients_RGB_Y + 8 * eax]
+    movq      mm1, [kCoefficientsRgbY + 8 * eax]
     paddsw    mm1, mm0
     psraw     mm1, 6
     packuswb  mm1, mm1
@@ -514,27 +508,27 @@ void ScaleYUVToRGB32Row(const uint8* y_buf,
     mov       ebp, [esp + 32 + 16]  // rgb
     mov       ecx, [esp + 32 + 20]  // width
     xor       ebx, ebx              // x
-    jmp       wend
+    jmp       scaleend
 
- wloop :
+ scaleloop :
     mov       eax, ebx
     sar       eax, 5
     movzx     eax, byte ptr [edi + eax]
-    movq      mm0, [coefficients_RGB_U + 8 * eax]
+    movq      mm0, [kCoefficientsRgbU + 8 * eax]
     mov       eax, ebx
     sar       eax, 5
     movzx     eax, byte ptr [esi + eax]
-    paddsw    mm0, [coefficients_RGB_V + 8 * eax]
+    paddsw    mm0, [kCoefficientsRgbV + 8 * eax]
     mov       eax, ebx
     add       ebx, [esp + 32 + 24]  // x += dx
     sar       eax, 4
     movzx     eax, byte ptr [edx + eax]
-    movq      mm1, [coefficients_RGB_Y + 8 * eax]
+    movq      mm1, [kCoefficientsRgbY + 8 * eax]
     mov       eax, ebx
     add       ebx, [esp + 32 + 24]  // x += dx
     sar       eax, 4
     movzx     eax, byte ptr [edx + eax]
-    movq      mm2, [coefficients_RGB_Y + 8 * eax]
+    movq      mm2, [kCoefficientsRgbY + 8 * eax]
     paddsw    mm1, mm0
     paddsw    mm2, mm0
     psraw     mm1, 6
@@ -542,29 +536,29 @@ void ScaleYUVToRGB32Row(const uint8* y_buf,
     packuswb  mm1, mm2
     movntq    [ebp], mm1
     add       ebp, 8
- wend :
+ scaleend :
     sub       ecx, 2
-    jns       wloop
+    jns       scaleloop
 
     and       ecx, 1  // odd number of pixels?
-    jz        wdone
+    jz        scaledone
 
     mov       eax, ebx
     sar       eax, 5
     movzx     eax, byte ptr [edi + eax]
-    movq      mm0, [coefficients_RGB_U + 8 * eax]
+    movq      mm0, [kCoefficientsRgbU + 8 * eax]
     mov       eax, ebx
     sar       eax, 5
     movzx     eax, byte ptr [esi + eax]
-    paddsw    mm0, [coefficients_RGB_V + 8 * eax]
+    paddsw    mm0, [kCoefficientsRgbV + 8 * eax]
     mov       eax, ebx
     sar       eax, 4
     movzx     eax, byte ptr [edx + eax]
-    movq      mm1, [coefficients_RGB_Y + 8 * eax]
+    movq      mm1, [kCoefficientsRgbY + 8 * eax]
     mov       eax, ebx
     sar       eax, 4
     movzx     eax, byte ptr [edx + eax]
-    movq      mm2, [coefficients_RGB_Y + 8 * eax]
+    movq      mm2, [kCoefficientsRgbY + 8 * eax]
     paddsw    mm1, mm0
     paddsw    mm2, mm0
     psraw     mm1, 6
@@ -572,12 +566,10 @@ void ScaleYUVToRGB32Row(const uint8* y_buf,
     packuswb  mm1, mm2
     movd      [ebp], mm1
 
- wdone :
-
+ scaledone :
     popad
     ret
   }
 }
-
-}  // namespace media
+}  // extern "C"