onnxruntime
WebGPU: Add indirect dispatch for flash attention graph capture to support Gemma4
#29236
Merged
Merged
WebGPU: Add indirect dispatch for flash attention graph capture to support Gemma4 #29236
justinchuby
merged
16 commits into
main
from
user/feich/gemma4_webgpu_gc_support
copilot-pull-request-reviewer
commented
on 2026-06-26
feich-ms
changed the title
add indirect dispatch support for graph capture
c5f03453
webgpu: refactor indirect dispatch normalization in flash attention
2f70dec3
webgpu: add indirect dispatch path tests for flash attention
faee205e
webgpu: clang-format and remove verbose comment in flash attention
e5f62cde
webgpu: fix RunGQASharedKV call sites for GqaTargetEp API change
4ee58bb0
webgpu: fix GQA kv_empty tests to use valid total_sequence_length
9272ceb4
webgpu: exempt kv_empty layers from graph capture share_buffer assertion
aa60c436
webgpu: fix stale kNormalizeDispatchGroupSizeFn reference after rebase
49936038
webgpu: rename AppendNormalizeDispatchShader to AppendPopulateIndirec…
9cefb106
qjia7
commented
on 2026-06-30
webgpu: inline AppendPopulateIndirectDispatchShader into its only cal…
495aac70
webgpu: switch PrepareIndirectDispatchProgram to use total_seqlen input
bc84e8d3
qjia7
commented
on 2026-06-30
Fix kv_empty WebGPU tests: reuse EP, drop unvalidated present outputs
99c41ee3
webgpu: add graph capture test for kv_empty indirect dispatch
dabdf99b
qjia7
commented
on 2026-07-01
webgpu: rewrite graph capture GQA test to exercise real ORT GC path
f99046c2
webgpu: replay graph capture test with different inputs to verify cor…
ec1d50c5
webgpu: move #ifdef USE_WEBGPU to wrap the whole GC test, not just body
053c2b98
qjia7
approved these changes
on 2026-07-02
hariharans29
approved these changes
on 2026-07-02
Assignees
No one assigned
Labels
ep:WebGPU
Milestone
No milestone
Login to write a write a comment.
Login via GitHub