WebGPU: Add indirect dispatch for flash attention graph capture to support Gemma4#29236

Merged

justinchuby merged 16 commits into

mainfrom

user/feich/gemma4_webgpu_gc_support

Jul 2, 2026

Commits on Jun 30, 2026

add indirect dispatch support for graph capture
feich-ms
committed
webgpu: refactor indirect dispatch normalization in flash attention

feich-ms
and
claude
committed
webgpu: add indirect dispatch path tests for flash attention

feich-ms
and
claude
committed
webgpu: clang-format and remove verbose comment in flash attention

feich-ms
and
claude
committed
webgpu: fix RunGQASharedKV call sites for GqaTargetEp API change

feich-ms
and
claude
committed
webgpu: fix GQA kv_empty tests to use valid total_sequence_length

feich-ms
and
claude
committed
webgpu: exempt kv_empty layers from graph capture share_buffer assertion

feich-ms
and
claude
committed
webgpu: fix stale kNormalizeDispatchGroupSizeFn reference after rebase

feich-ms
and
claude
committed
webgpu: rename AppendNormalizeDispatchShader to AppendPopulateIndirectDispatchShader

feich-ms
and
claude
committed
webgpu: inline AppendPopulateIndirectDispatchShader into its only call site

feich-ms
and
claude
committed
webgpu: switch PrepareIndirectDispatchProgram to use total_seqlen input

feich-ms
and
claude
committed
Fix kv_empty WebGPU tests: reuse EP, drop unvalidated present outputs
Copilot
authored

Commits on Jul 1, 2026

Commits on Jul 2, 2026