Use glDiscardFramebuffer to save memory bandwidth

Leverage glDiscardFramebuffer for tiled-based GPU to improve preserveDrawingBuffer=false WebGL
performance and also save memory bandwidth. In case of glDiscardFramebuffer, there is no need
to call glClear again.

BUG=418272
Committed: https://src.chromium.org/viewvc/blink?view=rev&revision=183217

[Hongbo Min, 2014-09-30 06:02:01]

hongbo.min@intel.com changed reviewers:
+ kbr@chromium.org, zmo@chromium.org

[Hongbo Min, 2014-09-30 06:03:43]

kbr@, this is the first patch to improve Discard-mode WebGL performance. It
seems that the clearing on a discarded framebuffer won't need to call glClear
explicitly, and this would greatly improve discard-mode WebGL performance, ~30%
on ASUS Nexus 7.

[Ken Russell (switch to Gerrit), 2014-10-01 20:40:06]

Excellent. Thank you for looking into this. One comment, then LGTM.

I assume you've run the WebGL conformance suite with this change? What about
layout tests? Even though printing from platforms with tile-based GPUs isn't
that common, I'd like to ensure tests like
virtual/threaded/printing/webgl-repeated-printing.html continue to work with
this change.

https://codereview.chromium.org/614883002/diff/1/Source/platform/graphics/gpu...
File Source/platform/graphics/gpu/DrawingBuffer.cpp (right):

https://codereview.chromium.org/614883002/diff/1/Source/platform/graphics/gpu...
Source/platform/graphics/gpu/DrawingBuffer.cpp:268: // Explicitly discard
framebuffer to save GPU memory bandwidth for tile-based GPU arch.
Please test for the presence of this extension in DrawingBuffer::create, pass a
flag for it into the DrawingBuffer's constructor, and test the flag here before
making the call.

[Hongbo Min, 2014-10-02 00:51:07]

On 2014/10/01 20:40:06, Ken Russell wrote:
> Excellent. Thank you for looking into this. One comment, then LGTM.
> 
> I assume you've run the WebGL conformance suite with this change? What about
> layout tests? Even though printing from platforms with tile-based GPUs isn't
> that common, I'd like to ensure tests like
> virtual/threaded/printing/webgl-repeated-printing.html continue to work with
> this change.
> 
>
https://codereview.chromium.org/614883002/diff/1/Source/platform/graphics/gpu...
> File Source/platform/graphics/gpu/DrawingBuffer.cpp (right):
> 
>
https://codereview.chromium.org/614883002/diff/1/Source/platform/graphics/gpu...
> Source/platform/graphics/gpu/DrawingBuffer.cpp:268: // Explicitly discard
> framebuffer to save GPU memory bandwidth for tile-based GPU arch.
> Please test for the presence of this extension in DrawingBuffer::create, pass
a
> flag for it into the DrawingBuffer's constructor, and test the flag here
before
> making the call.

thanks for your comments. I will keep improving this CL.

[Hongbo Min, 2014-10-02 01:27:06]

Hi, this CL has the following changes:
1. Check if GL_EXT_discardframebuffer is supported in DrawingBuffer::create;
2. Bypass WebGLRenderingContextBase::clearIfComposited if the framebuffer is
discarded by glDiscardFramebuffer, so that glClear on framebuffer is a
performance bottleneck, to reduce the calling of glClear could greatly improve
fps (over 30% on ASUS Nexus 7)

The trybots is building now and let's wait for the results to see if it is fine.

Please kindly review. Thanks.

[Hongbo Min, 2014-10-02 01:31:08]

Patchset #2 (id:20001) has been deleted

[Ken Russell (switch to Gerrit), 2014-10-03 23:19:56]

LGTM again. Thank you for picking this up.

[Ken Russell (switch to Gerrit), 2014-10-03 23:19:59]

The CQ bit was checked by kbr@chromium.org

[commit-bot: I haz the power, 2014-10-03 23:21:10]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/614883002/40001

[commit-bot: I haz the power, 2014-10-04 00:07:45]

Committed patchset #2 (id:40001) as 183217