WebGPU CTS fails to initialize DirectX debug layer(s)
Categories
(Core :: Graphics: WebGPU, defect, P2)
Tracking
()
Tracking | Status | |
---|---|---|
firefox-esr102 | --- | unaffected |
firefox111 | --- | unaffected |
firefox112 | --- | unaffected |
firefox113 | --- | disabled |
firefox114 | --- | fix-optional |
People
(Reporter: ErichDonGubler, Unassigned)
References
(Blocks 1 open bug, Regression)
Details
(Keywords: regression)
WebGPU CTS currently fails to initialize debugging layers for its dx12
back end. This is a potentially significant problem for debugging failures in Windows builds; clear diagnostics are often key to resolving impactful issues such as portability hazards and undefined behavior. We'd like to ensure that this runs correctly in at least our CI runners on Windows machines.
Steps to reproduce
-
Configure WebGPU to run on Firefox. ATOW, this means setting the
dom.webgpu.enabled
pref. totrue
, but this may not be necessary soon. -
Ensure that the
dom.webgpu.wgpu-backend
pref. is set todx12
. -
Ensure that the
RUST_LOG
environment variable is set to at leastwgpu_hal=info
. -
Invoke Firefox in a context where WebGPU will be exercised, and observe log messages like the following:
[WARN wgpu_hal::dx12::instance] Unable to enable D3D12 debug interface: 0x887A002D [WARN wgpu_hal::auxil::dxgi::factory] Unable to enable DXGI debug interface: 0x887A002D
To give a concrete example of the above, you can apply the following patch, which sets the correct prefs. for WebGPU's current CTS tests:
--- a/testing/web-platform/mozilla/meta/webgpu/__dir__.ini
+++ b/testing/web-platform/mozilla/meta/webgpu/__dir__.ini
@@ -1,1 +1,2 @@
-prefs: [dom.webgpu.enabled:true]
+prefs: [dom.webgpu.enabled:true,
+ dom.webgpu.wgpu-backend:dx12]
…and run one of the reftests:
RUST_LOG=wgpu_hal=info ./mach wpt /_mozilla/webgpu/webgpu/web_platform/reftests/canvas_colorspace_rgba8unorm.https.html
Observations
This issue has been observed only on Windows 11 until now. It's possible that other versions of Windows also have this issue, and we haven't witnessed it and recorded it here yet.
Reporter | ||
Comment 1•2 years ago
|
||
Reporter | ||
Updated•2 years ago
|
Reporter | ||
Updated•2 years ago
|
Comment 2•2 years ago
|
||
Set release status flags based on info from the regressing bug 1720941
Updated•2 years ago
|
Updated•2 years ago
|
Reporter | ||
Updated•2 years ago
|
Reporter | ||
Updated•2 years ago
|
Reporter | ||
Updated•2 years ago
|
Comment hidden (obsolete) |
Reporter | ||
Updated•1 year ago
|
Reporter | ||
Comment 4•1 year ago
•
|
||
The entry in DXGI_ERROR
for the error code I listed in the description:
DXGI_ERROR_SDK_COMPONENT_MISSING
0x887A002D
The operation depends on an SDK component that is missing or mismatched.
Reporter | ||
Updated•1 year ago
|
Reporter | ||
Updated•1 year ago
|
Updated•11 months ago
|
Description
•