onnxruntime
[WebGPU] Handle empty-rank MatMul and DequantizeLinear correctly
#33171
Open
Open
[WebGPU] Handle empty-rank MatMul and DequantizeLinear correctly #33171
mustjab
wants to merge
14 commits into
main
from
fix/webgpu-matmul-empty-k
[WebGPU] Handle MatMul with zero inner dimension
efd1da4f
[WebGPU] Cover empty Conv weights and require GPU execution in zero-K…
914a07fc
copilot-pull-request-reviewer
commented
on 2026-10-08
[WebGPU] Keep zero-K Conv test within contrib guard
86106176
copilot-pull-request-reviewer
commented
on 2026-10-08
[WebGPU] Cover shared MatMul helper empty output through Conv
90851f1d
copilot-pull-request-reviewer
commented
on 2026-10-08
[WebGPU] Consolidate empty-rank correctness and repair capture regres…
cbd4c542
mustjab
changed the title
copilot-pull-request-reviewer
commented
on 2026-10-08
[WebGPU] Mark custom capture tester as executed
47a84a3c
copilot-pull-request-reviewer
commented
on 2026-10-09
Fix CPU zero-K MatMul broadcast output initialization
46f363d1
copilot-pull-request-reviewer
commented
on 2026-10-09
Preserve zero-sized MatMul broadcast batches and offsets
938581b9
copilot-pull-request-reviewer
commented
on 2026-10-09
Resolve WebGPU capture test models before serialization
811b27be
Make MatMul batch-offset dimension conversion explicit on 32-bit targets
0273308e
copilot-pull-request-reviewer
commented
on 2026-10-09
Test full CPU zero-K output for transposed MatMul with a right vector
3b55d147
sushraja-msft
dismissed these changes
on 2026-10-12
Merge main and preserve empty-rank correctness in the MatMul dispatcher
db084a54
copilot-pull-request-reviewer
commented
on 2026-10-12
Disable CPU fallback in WebGPU MatMul execution regressions
bf6821b1
copilot-pull-request-reviewer
commented
on 2026-10-12
Disambiguate capture-test tensor access for GCC
f40b6403
Reviewers
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Login to write a write a comment.
Login via GitHub