Some users still affected by AMD switchable blacklist entry falling back to D3D9 |
|||||||||||
Issue descriptionChrome Version: 60.0.3112.90 OS: Windows 10 What steps will reproduce the problem? (1) Go to figma.com with a Windows laptop with AMD Switchable graphics (2) Open any file (3) Zoom in and out What is the expected result? Expect the content to render correctly. What happens instead? The content renders incorrectly -- tile copies going to the wrong place, clears not working -- because of the fallback to the D3D9 rendering path from Issue 451420 . about:gpu from an affected laptop is attached. I'm not sure what exact model this is but it contains an AMD R5 M240 which is fairly recent (~2013). It's disappointing that the D3D11 hardware accelerated video decoding paths were broken on this configuration and we should revisit it to see if this blacklist entry can be removed in favor of another one.
,
Aug 16 2017
Newer AMD drivers fixed bugs in handling of NV12 videos. I've tested on a Dell Inspiron 15" 5000 Series with dual Intel HD 620 and AMD R7 M445 with --disable_d3d11=0 and the H.264 videos from Issue 451420 all play correctly. I'm going to restrict the driver bug workaround to older AMD drivers on these dual-GPU systems.
,
Aug 16 2017
,
Aug 16 2017
Awesome. Is --disable_d3d11=0 something we can tell our Win8/Win10 users to tryout if they have a system where DX11 isn't taken, but could be.
,
Aug 16 2017
Yes, you can recommend that to customers. It won't enable WebGL 2.0 but I can confirm it fixes the rendering artifacts seen in Figma with the D3D9 backend on this particular AMD switchable laptop.
,
Aug 16 2017
Here's about:gpu from the Dell Inspiron laptop. Note that the driver version is 21.19.128.4 dated 9-3-2016. The driver version that was known to fix DXGI videos on AMD GPUs, from Issue 644293, was 21.19.519.2. Since I've been able to test with 21.19.128.4 and assuming that it's the driver more than the specific AMD or Intel GPU in these switchable systems, I'll restrict the disabling of the D3D11 backend to driver versions earlier than this one.
,
Aug 16 2017
In older Issue 641967 some users reported that they successfully turned on the D3D11 path on AMD switchable GPUs with driver version 20.19.0.32832 dated 6-30-2016 . I'll restrict the workarounds based on driver date instead of specific driver version.
,
Aug 16 2017
Issue 641967 has been merged into this issue.
,
Aug 16 2017
https://chromium-review.googlesource.com/617387 is up for review, re-enabling the D3D11 path and WebGL 2.0 on these machines with newer drivers (June 2016 and newer). We only have one such machine in house but the MPEG-4 video from Issue 451420 renders correctly with this change. www.figma.com also renders more correctly, and simple WebGL 2.0 content runs.
,
Aug 16 2017
Forgot to actually attach about:gpu from the Inspiron above.
,
Aug 17 2017
The following revision refers to this bug: https://chromium.googlesource.com/chromium/src.git/+/51a69c87c1012ed072d5a47dab7059361474db37 commit 51a69c87c1012ed072d5a47dab7059361474db37 Author: Ken Russell <kbr@chromium.org> Date: Thu Aug 17 19:36:40 2017 Re-enable D3D11 on Win/AMD switchable machines with newer drivers. Testing on a Dell Inspiron 5000 Series laptop running Windows 10 and with dual Intel HD 620 and AMD R7 M445 GPUs, the problem with hardware-accelerated video decoding seems to be fixed, and the Direct3D 11 path is faster and fixes rendering bugs with some customers' applications. Restrict the workarounds disabling D3D11 and WebGL 2.0 to drivers older than 2016-06. BUG= 755722 Cq-Include-Trybots: master.tryserver.chromium.android:android_optional_gpu_tests_rel;master.tryserver.chromium.linux:linux_optional_gpu_tests_rel;master.tryserver.chromium.mac:mac_optional_gpu_tests_rel;master.tryserver.chromium.win:win_optional_gpu_tests_rel Change-Id: I76ae209bf91db980a779a76e4ddb05e4970c699a Reviewed-on: https://chromium-review.googlesource.com/617387 Commit-Queue: Kenneth Russell <kbr@chromium.org> Reviewed-by: Antoine Labour <piman@chromium.org> Reviewed-by: Victor Miura <vmiura@chromium.org> Cr-Commit-Position: refs/heads/master@{#495281} [modify] https://crrev.com/51a69c87c1012ed072d5a47dab7059361474db37/gpu/config/gpu_driver_bug_list.json [modify] https://crrev.com/51a69c87c1012ed072d5a47dab7059361474db37/gpu/config/software_rendering_list.json
,
Aug 17 2017
kbr, thanks for the fix! Would we be able to cherry-pick this to M61 (or even M60)?
,
Aug 17 2017
birunthan@: I think we'd better let this go through the Canary / Dev channels as normal. There's a risk that we may regress video playback on some systems, and we need to let end users test this out. Please feel free to encourage your customers affected by this issue to test with Canary / Dev and see how it behaves. If any bugs are encountered, please file them and email me the bug ID. Thanks. Closing as fixed.
,
Oct 17 2017
,
Oct 17 2017
Issue 775443 has been merged into this issue.
,
Jun 6 2018
|
|||||||||||
►
Sign in to add a comment |
|||||||||||
Comment 1 by kbr@chromium.org
, Aug 15 201713.5 KB
13.5 KB View Download