I think this is related to GpuMemoryBuffer creation.  (Note GpuProcessHost::OnGpuMemoryBufferCreated calls on the Browser IO thread)

In the test I was pinch-zooming on The Verge which makes us allocate a lot of tile resources.  We're blocking for a long time to allocate these buffers, especially in the GPU raster case the longest TileManager::ScheduleTasks is 15.159ms.

For Software raster we have to block when we allocate an IO surface since we need to map the buffer.  Could potentially avoid this with GPU raster?

Potentially. IOSurface allocation and texture-binding is very expensive, and I would advocate that we make an effort to cache them.

See UMAs at:
https://uma.googleplex.com/p/chrome/histograms/?endDate=06-13-2016&dayCount=1&histograms=GPU.IOSurface.CreateTime%2CGPU.IOSurface.TexImageTime&fixupData=true&showMax=true&filters=channel%2Ceq%2C1%2Cisofficial%2Ceq%2CTrue&implicitFilters=isofficial

The Flush is coming from GLES2Implementation::CreateImageCHROMIUMHelper()

https://cs.chromium.org/chromium/src/gpu/command_buffer/client/gles2_implementation.cc?sq=package:chromium&rcl=1465917785&l=5972

Could we use an OrderingBarrier here?

+reveman@

Answering my question about OrderingBarriers, I think it wouldn't make a difference since immediately after we send GpuCommandBufferMsg_CreateImage IPC and we synchronize with EnsureWorkVisible().

Re #2: We should be re-using resources, but we're also quite aggressive about purging resources that go unused.  Definitely needs to be looked at in more detail.

I think we could get rid of the synchronous allocation and 3 or 4 thread hops for GPU raster.

Moving this nonessential bug to the next milestone.

For more details visit https://www.chromium.org/issue-tracking/autotriage - Your friendly Sheriffbot

This issue has been moved once and is lower than Pri-1. Removing the milestone.

For more details visit https://www.chromium.org/issue-tracking/autotriage - Your friendly Sheriffbot

This issue has been Available for over a year. If it's no longer important or seems unlikely to be fixed, please consider closing it out. If it is important, please re-triage the issue.

Sorry for the inconvenience if the bug really should have been left as Available. If you change it back, also remove the "Hotlist-Recharge-Cold" label.

For more details visit https://www.chromium.org/issue-tracking/autotriage - Your friendly Sheriffbot

vmiura: is this still an issue / worth doing?

This issue has been Available for over a year. If it's no longer important or seems unlikely to be fixed, please consider closing it out. If it is important, please re-triage the issue.

Sorry for the inconvenience if the bug really should have been left as Available.

For more details visit https://www.chromium.org/issue-tracking/autotriage - Your friendly Sheriffbot

vmiura: Not sure whether this is still valid. Could you archive if this is no longer needed? Thanks.

This is probably not a bug any more because:
1. We create GMBs in GPU process instead of in the renderer for tiles via TexStorage2DImageCHROMIUM.
2. We implemented multi-flush which makes ordering barriers such as those inserted for tiles not flush when switching contexts.