Make C, MMX and SSE2 all output the same results

BUG=19113,41045
TEST=none.  just cleanup and making all platforms produce the same results

Review URL: http://codereview.chromium.org/1638008

git-svn-id: svn://svn.chromium.org/chrome/trunk/src@44591 0039d316-1c4b-4281-b951-d872f2087c98

1 files changed, 156 insertions, 246 deletions

diff --git a/media/base/yuv_row_win.cc b/media/base/yuv_row_win.cc
index c7cf8d3..e5354a8 100644
--- a/media/base/yuv_row_win.cc
+++ b/media/base/yuv_row_win.cc
@@ -4,254 +4,25 @@
 
 #include "media/base/yuv_row.h"
 
-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), \
-  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) __declspec(align(16)) var
-
-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),
-};
+#define kCoefficientsRgbU kCoefficientsRgbY + 2048
+#define kCoefficientsRgbV kCoefficientsRgbY + 4096
 
-#undef RGBHY
-#undef RGBY
-#undef RGBU
-#undef RGBV
-#undef MMX_ALIGNED
-
-// Warning C4799: function has no EMMS instruction.
-// EMMS() is slow and should be called by the calling function once per image.
-#pragma warning(disable: 4799)
+extern "C" {
 
+#if USE_MMX
 __declspec(naked)
 void FastConvertYUVToRGB32Row(const uint8* y_buf,
                               const uint8* u_buf,
                               const uint8* v_buf,
                               uint8* rgb_buf,
-                              int width) {
+                              int source_width) {
   __asm {
     pushad
     mov       edx, [esp + 32 + 4]   // Y
     mov       edi, [esp + 32 + 8]   // U
     mov       esi, [esp + 32 + 12]  // V
     mov       ebp, [esp + 32 + 16]  // rgb
-    mov       ecx, [esp + 32 + 20]  // width
+    mov       ecx, [esp + 32 + 20]  // source_width
     jmp       convertend
 
  convertloop :
@@ -302,7 +73,7 @@ void ConvertYUVToRGB32Row(const uint8* y_buf,
                           const uint8* u_buf,
                           const uint8* v_buf,
                           uint8* rgb_buf,
-                          int width,
+                          int source_width,
                           int step) {
   __asm {
     pushad
@@ -310,7 +81,7 @@ void ConvertYUVToRGB32Row(const uint8* y_buf,
     mov       edi, [esp + 32 + 8]   // U
     mov       esi, [esp + 32 + 12]  // V
     mov       ebp, [esp + 32 + 16]  // rgb
-    mov       ecx, [esp + 32 + 20]  // width
+    mov       ecx, [esp + 32 + 20]  // source_width
     mov       ebx, [esp + 32 + 24]  // step
     jmp       wend
 
@@ -363,7 +134,7 @@ void RotateConvertYUVToRGB32Row(const uint8* y_buf,
                                 const uint8* u_buf,
                                 const uint8* v_buf,
                                 uint8* rgb_buf,
-                                int width,
+                                int source_width,
                                 int ystep,
                                 int uvstep) {
   __asm {
@@ -372,7 +143,7 @@ void RotateConvertYUVToRGB32Row(const uint8* y_buf,
     mov       edi, [esp + 32 + 8]   // U
     mov       esi, [esp + 32 + 12]  // V
     mov       ebp, [esp + 32 + 16]  // rgb
-    mov       ecx, [esp + 32 + 20]  // width
+    mov       ecx, [esp + 32 + 20]  // source_width
     jmp       wend
 
  wloop :
@@ -426,14 +197,14 @@ void DoubleYUVToRGB32Row(const uint8* y_buf,
                          const uint8* u_buf,
                          const uint8* v_buf,
                          uint8* rgb_buf,
-                         int width) {
+                         int source_width) {
   __asm {
     pushad
     mov       edx, [esp + 32 + 4]   // Y
     mov       edi, [esp + 32 + 8]   // U
     mov       esi, [esp + 32 + 12]  // V
     mov       ebp, [esp + 32 + 16]  // rgb
-    mov       ecx, [esp + 32 + 20]  // width
+    mov       ecx, [esp + 32 + 20]  // source_width
     jmp       wend
 
  wloop :
@@ -498,7 +269,7 @@ void ScaleYUVToRGB32Row(const uint8* y_buf,
                         const uint8* u_buf,
                         const uint8* v_buf,
                         uint8* rgb_buf,
-                        int width,
+                        int source_width,
                         int dx) {
   __asm {
     pushad
@@ -506,7 +277,7 @@ void ScaleYUVToRGB32Row(const uint8* y_buf,
     mov       edi, [esp + 32 + 8]   // U
     mov       esi, [esp + 32 + 12]  // V
     mov       ebp, [esp + 32 + 16]  // rgb
-    mov       ecx, [esp + 32 + 20]  // width
+    mov       ecx, [esp + 32 + 20]  // source_width
     xor       ebx, ebx              // x
     jmp       scaleend
 
@@ -571,14 +342,14 @@ void LinearScaleYUVToRGB32Row(const uint8* y_buf,
                               const uint8* u_buf,
                               const uint8* v_buf,
                               uint8* rgb_buf,
-                              int width,
-                              int scaled_dx) {
+                              int source_width,
+                              int source_dx) {
 __asm {
     pushad
     mov edx, [esp + 32 + 4]
     mov edi, [esp + 32 + 8]
     mov ebp, [esp + 32 + 16]
-    mov ecx, [esp + 32 + 20]    // width = width * scaled_dx + ebx
+    mov ecx, [esp + 32 + 20]    // source_width = source_width * source_dx + ebx
     imul ecx, [esp + 32 + 24]
     xor ebx, ebx
     add ecx, ebx
@@ -668,5 +439,144 @@ lscalelastpixel:
     ret
 };
 }
+#else  // USE_MMX
+
+// C reference code that mimic the YUV assembly.
+#define packuswb(x) ((x) < 0 ? 0 : ((x) > 255 ? 255 : (x)))
+#define paddsw(x, y) (((x) + (y)) < -32768 ? -32768 : \\
+    (((x) + (y)) > 32767 ? 32767 : ((x) + (y))))
+
+static inline void YuvPixel(uint8 y,
+                            uint8 u,
+                            uint8 v,
+                            uint8* rgb_buf) {
+
+  int b = kCoefficientsRgbY[256+u][0];
+  int g = kCoefficientsRgbY[256+u][1];
+  int r = kCoefficientsRgbY[256+u][2];
+  int a = kCoefficientsRgbY[256+u][3];
+
+  b = paddsw(b, kCoefficientsRgbY[512+v][0]);
+  g = paddsw(g, kCoefficientsRgbY[512+v][1]);
+  r = paddsw(r, kCoefficientsRgbY[512+v][2]);
+  a = paddsw(a, kCoefficientsRgbY[512+v][3]);
+
+  b = paddsw(b, kCoefficientsRgbY[y][0]);
+  g = paddsw(g, kCoefficientsRgbY[y][1]);
+  r = paddsw(r, kCoefficientsRgbY[y][2]);
+  a = paddsw(a, kCoefficientsRgbY[y][3]);
+
+  b >>= 6;
+  g >>= 6;
+  r >>= 6;
+  a >>= 6;
+
+  *reinterpret_cast<uint32*>(rgb_buf) = (packuswb(b)) |
+                                        (packuswb(g) << 8) |
+                                        (packuswb(r) << 16) |
+                                        (packuswb(a) << 24);
+}
+
+#if TEST_MMX_YUV
+static inline void YuvPixel(uint8 y,
+                            uint8 u,
+                            uint8 v,
+                            uint8* rgb_buf) {
+
+  __asm {
+    movzx     eax, u
+    movq      mm0, [kCoefficientsRgbY+2048 + 8 * eax]
+    movzx     eax, v
+    paddsw    mm0, [kCoefficientsRgbY+4096 + 8 * eax]
+    movzx     eax, y
+    movq      mm1, [kCoefficientsRgbY + 8 * eax]
+    paddsw    mm1, mm0
+    psraw     mm1, 6
+    packuswb  mm1, mm1
+    mov       eax, rgb_buf
+    movd      [eax], mm1
+    emms
+  }
+}
+#endif
 
+void FastConvertYUVToRGB32Row(const uint8* y_buf,
+                              const uint8* u_buf,
+                              const uint8* v_buf,
+                              uint8* rgb_buf,
+                              int source_width) {
+  for (int x = 0; x < source_width; x += 2) {
+    uint8 u = u_buf[x >> 1];
+    uint8 v = v_buf[x >> 1];
+    uint8 y0 = y_buf[x];
+    YuvPixel(y0, u, v, rgb_buf);
+    if ((x + 1) < source_width) {
+      uint8 y1 = y_buf[x + 1];
+      YuvPixel(y1, u, v, rgb_buf + 4);
+    }
+    rgb_buf += 8;  // Advance 2 pixels.
+  }
+}
+
+// 16.16 fixed point is used.  A shift by 16 isolates the integer.
+// A shift by 17 is used to further subsample the chrominence channels.
+// & 0xffff isolates the fixed point fraction.  >> 2 to get the upper 2 bits,
+// for 1/65536 pixel accurate interpolation.
+void ScaleYUVToRGB32Row(const uint8* y_buf,
+                        const uint8* u_buf,
+                        const uint8* v_buf,
+                        uint8* rgb_buf,
+                        int source_width,
+                        int dx) {
+  int x = 0;
+  for (int i = 0; i < source_width; i += 2) {
+    int y = y_buf[x >> 16];
+    int u = u_buf[(x >> 17)];
+    int v = v_buf[(x >> 17)];
+    YuvPixel(y, u, v, rgb_buf);
+    x += dx;
+    if ((i + 1) < source_width) {
+      y = y_buf[x >> 16];
+      YuvPixel(y, u, v, rgb_buf+4);
+      x += dx;
+    }
+    rgb_buf += 8;
+  }
+}
+
+void LinearScaleYUVToRGB32Row(const uint8* y_buf,
+                              const uint8* u_buf,
+                              const uint8* v_buf,
+                              uint8* rgb_buf,
+                              int source_width,
+                              int dx) {
+  int x = 0;
+  for (int i = 0; i < source_width; i += 2) {
+    int y0 = y_buf[x >> 16];
+    int y1 = y_buf[(x >> 16) + 1];
+    int u0 = u_buf[(x >> 17)];
+    int u1 = u_buf[(x >> 17) + 1];
+    int v0 = v_buf[(x >> 17)];
+    int v1 = v_buf[(x >> 17) + 1];
+    int y_frac = (x & 65535);
+    int uv_frac = ((x >> 1) & 65535);
+    int y = (y_frac * y1 + (y_frac ^ 65535) * y0) >> 16;
+    int u = (uv_frac * u1 + (uv_frac ^ 65535) * u0) >> 16;
+    int v = (uv_frac * v1 + (uv_frac ^ 65535) * v0) >> 16;
+    YuvPixel(y, u, v, rgb_buf);
+    x += dx;
+    if ((i + 1) < source_width) {
+      y0 = y_buf[x >> 16];
+      y1 = y_buf[(x >> 16) + 1];
+      y_frac = (x & 65535);
+      y = (y_frac * y1 + (y_frac ^ 65535) * y0) >> 16;
+      YuvPixel(y, u, v, rgb_buf+4);
+      x += dx;
+    }
+    rgb_buf += 8;
+  }
+}
+
+#endif  // USE_MMX
 }  // extern "C"
+