Remove --tab-capture-upscale/downscale-quality.

content/browser/compositor/gl_helper.cc

Index: content/browser/compositor/gl_helper.cc
diff --git a/content/browser/compositor/gl_helper.cc b/content/browser/compositor/gl_helper.cc
index 55493e900689ba6bcf421056e29fefb1645ab875..17d500bd36c96352d352b73b26816ea95f9498d1 100644
--- a/content/browser/compositor/gl_helper.cc
+++ b/content/browser/compositor/gl_helper.cc
@@ -182,7 +182,6 @@ class GLHelper::CopyTextureToImpl
                              GLHelper::ScalerQuality quality);
 
   ReadbackYUVInterface* CreateReadbackPipelineYUV(
-      GLHelper::ScalerQuality quality,
       const gfx::Size& src_size,
       const gfx::Rect& src_subrect,
       const gfx::Size& dst_size,
@@ -1181,7 +1180,6 @@ bool GLHelper::IsReadbackConfigSupported(SkColorType color_type) {
 }
 
 ReadbackYUVInterface* GLHelper::CopyTextureToImpl::CreateReadbackPipelineYUV(
-    GLHelper::ScalerQuality quality,
     const gfx::Size& src_size,
     const gfx::Rect& src_subrect,
     const gfx::Size& dst_size,
@@ -1201,6 +1199,19 @@ ReadbackYUVInterface* GLHelper::CopyTextureToImpl::CreateReadbackPipelineYUV(
   if (supported == GLHelperReadbackSupport::SWIZZLE)
     swizzle = kSwizzleBGRA;
 
+  // Tests show that the two up-scaling methods perform comparable speeds on the

[miu, 2016/03/29 20:08:52]

This comment should provide a little more context about the use case, and the
numbers are probably too much detail.  Suggest you replace this comment with:

"The YUV readback pipeline is used by real-time video capture of
tabs/windows/desktops.  Therefore, the scaler chosen here is based on
performance measurements of full end-to-end systems.  When down-scaling, always
use the "fast" scaler because it performs well on both low- and high- end
machines, provides decent image quality, and doesn't overwhelm downstream video
encoders with too much entropy (which can drastically increase CPU utilization).
 When up-scaling, always use "best" because the quality improvement is huge with
insignificant performance penalty.  Note that this strategy differs from
single-frame snapshot capture (i.e., using the Skia/RGB code path), where "best"
is always preferred."

[xjz, 2016/03/29 21:14:50]

Yuri: Thanks for the suggestion. But do you want to change the default scaler
for single-frame snapshot capture path from "good" to "best"? It is "good" in
current code:
https://code.google.com/p/chromium/codesearch#chromium/src/content/browser/co....
As in the case of multiple parses down-scaling, "good" is almost as good as
"fast" in quality but is much faster, maybe just keep using "good" as the
default scaler for the single-frame capture path?

[miu, 2016/03/31 21:44:05]

Ah, no.  :)  Actually, we should just leave out that last part of the comment,
as discussed.

[xjz, 2016/03/31 23:03:39]

Done.

+  // tested machines. So we choose "best" up-scaling method by default to get
+  // the best quality.
+  // The three down-scaling methods perform close speeds on high end machines,
+  // but not on other machines. The "best" method could be up to 96% slower than
+  // the "good" one, and the "good" method could be up to 250% slower than the
+  // "fast" one. As we don't have criteria to determine the proper scaling
+  // method in flight, and generally the down-scaling is more often needed on
+  // low end machines, we choose the "fast" down-scaling method by default.
+  GLHelper::ScalerQuality quality = ((src_size.width() < dst_size.width()) &&

[miu, 2016/03/29 20:08:52]

Should this be:

  GLHelper::ScalerQuality quality = ((src_size.GetArea() < dst_size.GetArea())
      ? GLHelper::SCALER_QUALITY_BEST
      : GLHelper::SCALER_QUALITY_FAST;

[xjz, 2016/03/29 21:14:50]

The scaling is done in two separate dimensions. My intention was to use "best"
only when we need do upscaling in both dimensions. If using this condition check
on area, we may use "best" when one dimension is down-scaling, and results in a
performance penalty. 

[miu, 2016/03/31 21:44:05]

Sounds good.

+                                     (src_size.height() < dst_size.height()))
+                                        ? GLHelper::SCALER_QUALITY_BEST
+                                        : GLHelper::SCALER_QUALITY_FAST;
   if (max_draw_buffers_ >= 2 && use_mrt) {
     return new ReadbackYUV_MRT(gl_, this, helper_->scaler_impl_.get(), quality,
                                src_size, src_subrect, dst_size, flip_vertically,
@@ -1212,7 +1223,6 @@ ReadbackYUVInterface* GLHelper::CopyTextureToImpl::CreateReadbackPipelineYUV(
 }
 
 ReadbackYUVInterface* GLHelper::CreateReadbackPipelineYUV(
-    ScalerQuality quality,
     const gfx::Size& src_size,
     const gfx::Rect& src_subrect,
     const gfx::Size& dst_size,
@@ -1220,7 +1230,7 @@ ReadbackYUVInterface* GLHelper::CreateReadbackPipelineYUV(
     bool use_mrt) {
   InitCopyTextToImpl();
   return copy_texture_to_impl_->CreateReadbackPipelineYUV(
-      quality, src_size, src_subrect, dst_size, flip_vertically, use_mrt);
+      src_size, src_subrect, dst_size, flip_vertically, use_mrt);
 }
 
 }  // namespace content