Fix the WebRTC color bug.

media/base/simd/convert_yuv_to_rgb_c.cc

Index: media/base/simd/convert_yuv_to_rgb_c.cc
diff --git a/media/base/simd/convert_yuv_to_rgb_c.cc b/media/base/simd/convert_yuv_to_rgb_c.cc
index f04a89dbf9c7eb97825a37ff5ddf5c35164fca0f..ed7530fe0696da54844494e5772bb367bcb943a0 100644
--- a/media/base/simd/convert_yuv_to_rgb_c.cc
+++ b/media/base/simd/convert_yuv_to_rgb_c.cc
@@ -11,6 +11,22 @@ namespace media {
 #define paddsw(x, y) (((x) + (y)) < -32768 ? -32768 : \
     (((x) + (y)) > 32767 ? 32767 : ((x) + (y))))
 
+// On Android, pixel layout is RGBA
+// (see third_party/skia/include/core/SkColorPriv.h);
+// however, other Chrome platforms use BGRA
+// (see skia/config/SkUserConfig.h).
+#if defined(OS_ANDROID)
+#define SK_A32_SHIFT    24

[scherkus (not reviewing), 2013/06/14 21:32:42]

nit: can you reorder these so that it follows the pixel layout

e.g., RGBA
SK_R32_SHIFT
SK_G32_SHIFT
SK_B32_SHIFT
SK_A32_SHIFT

[hkuang1, 2013/06/15 00:24:40]

Done.

+#define SK_B32_SHIFT    16
+#define SK_G32_SHIFT    8
+#define SK_R32_SHIFT    0
+#else
+#define SK_A32_SHIFT    24
+#define SK_R32_SHIFT    16
+#define SK_G32_SHIFT    8
+#define SK_B32_SHIFT    0
+#endif
+
 static inline void ConvertYUVToRGB32_C(uint8 y,

[scherkus (not reviewing), 2013/06/14 21:17:52]

OOC does this mean we're using C routines for doing colour conversion on
Android?

[hkuang1, 2013/06/14 21:28:35]

Yes. It uses this routine. Not quite efficient, but I could not find any neon
function in this folder for android.  The reason is mainly because the painting
path for webmediaplayer_ms is the same as webmediaplayer_impl on Android.

On 2013/06/14 21:17:52, scherkus wrote:

[scherkus (not reviewing), 2013/06/14 21:32:42]

I see. Have you looked at whether libyuv has a neon optimized routine?

[hkuang1, 2013/06/14 21:50:58]

I did not check it yet. But I thought I could not use the function in libyuv
here as the YUV2RGB conversion does not use the libyuv function. Is there any
reason for that?

On 2013/06/14 21:32:42, scherkus wrote:

[scherkus (not reviewing), 2013/06/14 22:17:30]

libyuv is a separate collection of yuv routines

instead I would consider making the change higher up in the stack instead of
adding libyuv code here

[fgalligan1, 2013/06/14 23:07:18]

That is fine, but I think that is a separate issue. I think this should be fixed
here in this CL. And then open up a bug to change webrtc to use libyuv.

                                        uint8 u,
                                        uint8 v,
@@ -35,10 +51,10 @@ static inline void ConvertYUVToRGB32_C(uint8 y,
   r >>= 6;
   a >>= 6;
 
-  *reinterpret_cast<uint32*>(rgb_buf) = (packuswb(b)) |
-                                        (packuswb(g) << 8) |
-                                        (packuswb(r) << 16) |
-                                        (packuswb(a) << 24);
+  *reinterpret_cast<uint32*>(rgb_buf) = (packuswb(b) << SK_B32_SHIFT) |
+                                        (packuswb(g) << SK_G32_SHIFT) |
+                                        (packuswb(r) << SK_R32_SHIFT) |
+                                        (packuswb(a) << SK_A32_SHIFT);
 }
 
 static inline void ConvertYUVAToARGB_C(uint8 y,
@@ -66,7 +82,10 @@ static inline void ConvertYUVAToARGB_C(uint8 y,
   g = packuswb(g) * a >> 8;
   r = packuswb(r) * a >> 8;
 
-  *reinterpret_cast<uint32*>(rgb_buf) = b | (g << 8) | (r << 16) | (a << 24);
+  *reinterpret_cast<uint32*>(rgb_buf) = (b << SK_B32_SHIFT) |
+                                        (g << SK_G32_SHIFT) |
+                                        (r << SK_R32_SHIFT) |
+                                        (a << SK_A32_SHIFT);
 }
 
 void ConvertYUVToRGB32Row_C(const uint8* y_buf,