YUV software decode path stride fixes.

cc/texture_uploader.cc

Index: cc/texture_uploader.cc
diff --git a/cc/texture_uploader.cc b/cc/texture_uploader.cc
index a1c812a855a9e0a1482f4728bb28a51973ca439a..1de6417f986d58ffd434d2e026bf34c10a577fa8 100644
--- a/cc/texture_uploader.cc
+++ b/cc/texture_uploader.cc
@@ -7,6 +7,7 @@
 #include <algorithm>
 #include <vector>
 
+#include "base/bits.h"
 #include "base/debug/alias.h"
 #include "base/debug/trace_event.h"
 #include "base/metrics/histogram.h"
@@ -229,11 +230,15 @@ void TextureUploader::uploadWithTexSubImage(const uint8* image,
 
     const uint8* pixel_source;
     unsigned int bytes_per_pixel = Resource::bytesPerPixel(format);
+    // Use 4-byte row alignment (OpenGL default) for upload performance.
+    unsigned int upload_image_stride =
+        base::bits::RoundUp(bytes_per_pixel * source_rect.width(), 4);
+    m_context->pixelStorei(GL_UNPACK_ALIGNMENT, 4);

[danakj, 2012/11/20 01:53:08]

It works the same without this pixelStorei() call right? We don't want to make
redundant calls through the context, so this can go away then.

Same below.

 
-    if (image_rect.width() == source_rect.width() && !offset.x()) {
-        pixel_source = &image[bytes_per_pixel * offset.y() * image_rect.width()];
+    if (upload_image_stride == image_rect.width() * bytes_per_pixel && !offset.x()) {
+        pixel_source = &image[image_rect.width() * bytes_per_pixel * offset.y()];
     } else {
-        size_t needed_size = source_rect.width() * source_rect.height() * bytes_per_pixel;
+        size_t needed_size = upload_image_stride * source_rect.height();
         if (m_subImageSize < needed_size) {
             m_subImage.reset(new uint8[needed_size]);
             m_subImageSize = needed_size;
@@ -241,7 +246,7 @@ void TextureUploader::uploadWithTexSubImage(const uint8* image,
         // Strides not equal, so do a row-by-row memcpy from the
         // paint results into a temp buffer for uploading.
         for (int row = 0; row < source_rect.height(); ++row)
-            memcpy(&m_subImage[source_rect.width() * bytes_per_pixel * row],
+            memcpy(&m_subImage[upload_image_stride * row],
                    &image[bytes_per_pixel * (offset.x() +
                           (offset.y() + row) * image_rect.width())],
                    source_rect.width() * bytes_per_pixel);
@@ -294,6 +299,12 @@ void TextureUploader::uploadWithMapTexSubImage(const uint8* image,
     // Offset from image-rect to source-rect.
     gfx::Vector2d offset(source_rect.origin() - image_rect.origin());
 
+    unsigned int bytes_per_pixel = Resource::bytesPerPixel(format);
+    // Use 4-byte row alignment (OpenGL default) for upload performance.
+    unsigned int upload_image_stride =
+        base::bits::RoundUp(bytes_per_pixel * source_rect.width(), 4);
+    m_context->pixelStorei(GL_UNPACK_ALIGNMENT, 4);
+
     // Upload tile data via a mapped transfer buffer
     uint8* pixel_dest = static_cast<uint8*>(
         m_context->mapTexSubImage2DCHROMIUM(GL_TEXTURE_2D,
@@ -312,17 +323,15 @@ void TextureUploader::uploadWithMapTexSubImage(const uint8* image,
         return;
     }
 
-    unsigned int bytes_per_pixel = Resource::bytesPerPixel(format);
-
-    if (image_rect.width() == source_rect.width() && !offset.x()) {
+    if (upload_image_stride == image_rect.width() * bytes_per_pixel && !offset.x()) {
         memcpy(pixel_dest,
-               &image[offset.y() * image_rect.width() * bytes_per_pixel],
-               image_rect.width() * source_rect.height() * bytes_per_pixel);
+               &image[image_rect.width() * bytes_per_pixel * offset.y()],
+               source_rect.height() * image_rect.width() * bytes_per_pixel);
     } else {
         // Strides not equal, so do a row-by-row memcpy from the
         // paint results into the pixelDest
         for (int row = 0; row < source_rect.height(); ++row)
-            memcpy(&pixel_dest[source_rect.width() * row * bytes_per_pixel],
+            memcpy(&pixel_dest[upload_image_stride * row],
                    &image[bytes_per_pixel * (offset.x() +
                           (offset.y() + row) * image_rect.width())],
                    source_rect.width() * bytes_per_pixel);