Regression in WebglConformance_conformance_canvas_canvas_test on Linux AMD |
||||
Issue descriptionSee https://build.chromium.org/p/chromium.gpu.fyi/builders/Linux%20Release%20%28AMD%20R7%20240%29/builds/1591 Suspecting: Make disabling 2d canvas GPU acceleration for getImageData less aggressive. xiangze.zhang@intel.com https://codereview.chromium.org/2687073003 Kai as pixel wrangler can you triage and revert or suppress as necessary?
,
Feb 23 2017
Looking into it. Hard to see what's going on with the other unrelated failures on that bot as well...
,
Feb 24 2017
,
Feb 24 2017
Yes, I'm going to wait until then to continue looking into this. xiangze: if you think this may be caused by your change, please look into it. (Please revert if you can't fix it very quickly.)
,
Feb 24 2017
,
Feb 24 2017
The failed test case is related to my change. I'll revert my CL if this issue is urgent. But I can't reproduce this regression in my local Linux Haswell machine so I can't do further investigations. This test case draws image from a webgl canvas to a 40*40 2d canvas. Before my change, the backend of the 2d canvas is switched from CPU to GPU(due to EnableAccelerationToAvoidReadbacks heuristic) and switch back from GPU to CPU(due to GetImageDataForcesNoAcceleration heuristic) 3 times. After my change, the backend will only switch to GPU once and stay there.
,
Feb 24 2017
Xiangze: I'm sorry, but the failure on this configuration (Linux AMD) may indicate a deeper problem with your change. Since you indicate that you can't do further investigation, we can't leave the potentially buggy change in the source tree. I'm going to revert it.
,
Feb 24 2017
kbr: OK, thanks. Let's see if it will recover after revert. I'll have a look at what's causing this problem later.
,
Feb 27 2017
FYI: there are other failures on this bot now, but it seems that the canvas_canvas_test has cleared up.
,
Feb 27 2017
I'm going to close this since the bot is no longer failing this test. But please investigate this failure before attempting to re-land the patch. |
||||
►
Sign in to add a comment |
||||
Comment 1 by jmad...@chromium.org
, Feb 23 2017