Bug 198978 – [WHLSL] Create a shading language test harness

• Component: WebGPU
• Status: RESOLVED FIXED

[Justin Fan]

[WHLSL] Create a shading language test harness

[Justin Fan]

Created attachment 372399 [details]
Patch

[EWS Watchlist]

Comment on attachment 372399 [details]
Patch

Attachment 372399 [details] did not pass mac-wk2-ews (mac-wk2):
Output: https://webkit-queues.webkit.org/results/12513922

New failing tests:
webgpu/whlsl-harness-test.html

[EWS Watchlist]

Created attachment 372407 [details]
Archive of layout-test-results from ews105 for mac-highsierra-wk2

The attached test failures were seen while running run-webkit-tests on the mac-wk2-ews.
Bot: ews105  Port: mac-highsierra-wk2  Platform: Mac OS X 10.13.6

[Myles C. Maxfield]

Comment on attachment 372399 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=372399&action=review

I don’t see any WHLSL yet. What would that look like using this harness?

> LayoutTests/webgpu/whlsl-harness-test.html:27
> +let tests = {};

Can the harness be used in many different tests? I think we will want many files rather than dumping all the tests in a single file.

[Justin Fan]

(In reply to Myles C. Maxfield from comment #4)
> Comment on attachment 372399 [details]
> Patch
> 
> View in context:
> https://bugs.webkit.org/attachment.cgi?id=372399&action=review
> 
> I don’t see any WHLSL yet. What would that look like using this harness?
> 
> > LayoutTests/webgpu/whlsl-harness-test.html:27
> > +let tests = {};
> 
> Can the harness be used in many different tests? I think we will want many
> files rather than dumping all the tests in a single file.

Yes; whlsl-test-harness.js is designed to be used by itself or in combo with the WPT test harness as shown here. whlsl-harness-test.html is just to test the harness itself.

I plan on replacing the MSL with WHLSL, or adding a WHLSL mode to the harness immediately. This is just to make sure that the WebGPU bits will work properly, and to serve as a proof of concept.

[Justin Fan]

Created attachment 372420 [details]
Patch

[EWS Watchlist]

Comment on attachment 372420 [details]
Patch

Attachment 372420 [details] did not pass mac-wk2-ews (mac-wk2):
Output: https://webkit-queues.webkit.org/results/12516121

New failing tests:
webgpu/whlsl-harness-test.html

[EWS Watchlist]

Created attachment 372425 [details]
Archive of layout-test-results from ews107 for mac-highsierra-wk2

The attached test failures were seen while running run-webkit-tests on the mac-wk2-ews.
Bot: ews107  Port: mac-highsierra-wk2  Platform: Mac OS X 10.13.6

[Justin Fan]

Investigating failure on High Sierra.

[Justin Fan]

Figured out the issue. Some types (e.g. uint) are restricted to the metal namespace (e.g. metal::uint) on the version of the MSL compiler on High Sierra. Adding "using namespace metal" to all Metal shaders generated by the harness.

[Justin Fan]

Created attachment 372519 [details]
Patch for landing

[WebKit Commit Bot]

Comment on attachment 372519 [details]
Patch for landing

Clearing flags on attachment: 372519

Committed r246628: <https://trac.webkit.org/changeset/246628>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.

[Radar WebKit Bug Importer]

<rdar://problem/51926676>

[Truitt Savell]

Looks like the new test webgpu/whlsl-harness-test.html
added in https://trac.webkit.org/changeset/246628/webkit

is a constant crash on High Sierra WK2

History:
https://webkit-test-results.webkit.org/dashboards/flakiness_dashboard.html#showAllRuns=true&tests=webgpu%2Fwhlsl-harness-test.html

Crash:
VM Regions Near 0x44:
--> 
    __TEXT                 0000000104a67000-0000000104a68000 [    4K] r-x/rwx SM=COW  /Volumes/VOLUME/*/*.Development

Application Specific Information:
CRASHING TEST: webgpu/whlsl-harness-test.html

Thread 0 Crashed:: Dispatch queue: com.apple.main-thread
0   com.apple.driver.AppleIntelHD4000GraphicsMTLDriver	0x00007fff2ee03733 IGAccelComputeCommandEncoder::setEmulationBuffer(MTLIGAccelBuffer*, unsigned int, unsigned int) + 43
1   com.apple.driver.AppleIntelHD4000GraphicsMTLDriver	0x00007fff2ee0edcf -[MTLIGAccelComputeCommandEncoder setEmulationBuffers:offsets:withRange:] + 96
2   com.apple.Metal               	0x00007fff399c8a2d __57-[MTLIOAccelComputeCommandEncoder bindEmulationArguments]_block_invoke.8 + 54
3   com.apple.Metal               	0x00007fff399d93e3 -[MTLIndirectArgumentBufferEmulationData bindIndirectArgumentsWithFunctions:] + 166
4   com.apple.Metal               	0x00007fff399c89a5 -[MTLIOAccelComputeCommandEncoder bindEmulationArguments] + 354
5   com.apple.driver.AppleIntelHD4000GraphicsMTLDriver	0x00007fff2edfabc3 -[MTLIGAccelComputeCommandEncoder dispatchThreadgroups:threadsPerThreadgroup:] + 211
6   com.apple.WebCore             	0x000000010662297c WebCore::GPUComputePassEncoder::dispatch(unsigned int, unsigned int, unsigned int) + 140

[Jon Lee]

Justin, Myles, I suggest for now to skip webgpu on High Sierra and we file a separate bug to re-enable for follow-up.

[Justin Fan]

Filed https://bugs.webkit.org/show_bug.cgi?id=199076.