-
Notifications
You must be signed in to change notification settings - Fork 92
image_copy:mip_levels tests fail for bc compressed formats with AMD Radeon RX 5500 XT on macOS #4378
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Comments
@mwyrzykowski Same as #4377, does this look familiar to you? |
I don't have access to my AMD Mac. I'm assuming it doesn't fail on the same device on Intel? |
That's my understanding when looking at https://dawn-review.googlesource.com/c/dawn/+/239483/3/webgpu-cts/expectations.txt?checksPatchset=2 |
Yea well, looking here, they passed on intel (you need to click "load more") |
@greggman Did you have a chance to try it on your AMD Mac by any chance? |
unfortunately I will not be near my AMD Mac until June If anyone else wants to try it go to https://gpuweb.github.io/cts/standalone/?debug=1&power_preference=high-performance&runnow=1&q=webgpu:api,operation,command_buffer,image_copy:mip_levels:* in Chrome canary on AMD Mac. The bc format should look like this. 2 tests per format, one for '2d', on for '3d'. Both green. This screenshot is from M1 mac. ![]() Here's a link for low-power. Should be Intel on a dual GPU mac |
So I tried these tests on Swiftshader. They mostly pass though I think they eventually run out of memory. It looks like they make some pretty big textures. The astc-12x12-unorm 3d ones especially, the largest texture they make is 1728,1344,64 which compressed is 16meg. But to compare they render to 1728x1344*64 rgba8unorm which is 594meg x 2! I could make them compare one layer at a time, or, I can just make them not choose such large sizes (no idea why those particular numbers) |
@greggman Could https://issues.chromium.org/issues/415303046#comment5 and this CTS bug related? |
Ugh, Okay, so, I got this idea that I would add a high-water mark to the webgpu-memory tracking library, then put that into the webgpu-dev-extension so I could easily apply it to the CTS and see if other places are using lots of memory. I haven't tested all the CTS yet, but with regards to this test, in core it uses 235meg, in compat it uses 638meg because of the need to render. But, it's not using lots of memory because of the texture size as it's only rendering mip level 4 which is (108x84x4). Instead the issue is that subcases run in parallel and the test is async since it has to I'm not exactly sure the best way to fix this. We can:
I'm kind of leaning on exploring (2) (@kainino0x - thougts?) |
FYI Some tests from |
2 would be cool, it might even help with some other tests, but maybe it's unnecessarily complicated? |
I can do that. Just FYI to others following this thread. This memory issue is not the cause of the failures. 690meg is not that much on an AMD Mac and the tests passed on the same Mac on Intel |
IIUC 2 would be nice but would not necessarily fix the root issue of those image_copy:mip_levels tests that fail for bc compressed formats with AMD Radeon RX 5500 XT on macOS right? |
Correct, it will not fix the failures. It's just something I noticed when looking at the tests. |
Shall we wait until June so that you can investigate? |
According to https://ci.chromium.org/ui/p/chromium/builders/try/mac-dawn-rel/45372/overview, it seems like AMD Radeon RX 5500 XT is having issues on macOS for the following
image_copy:mip_levels
tests for BC compressed formats:Here's the cause of a failure for
webgpu:api,operation,command_buffer,image_copy:mip_levels:initMethod="CopyB2T";checkMethod="FullCopyT2B";format="bc1-rgba-unorm";dimension="3d"
Any idea if it's a CTS issue or a AMD Radeon RX 5500 XT driver bug?
FYI @greggman since 1a47862 introduced this
The text was updated successfully, but these errors were encountered: