[ExecuTorch][WebGPU] Request shader-f16 device feature (enablement, fail-open)#20814

Merged

JCNTH merged 12 commits into

gh/JCNTH/2/origfrom

gh/JCNTH/3/orig

Jul 9, 2026

Commits on Jul 9, 2026

[ExecuTorch][WebGPU] Request shader-f16 device feature (enablement, fail-open)
JCNTH
committed
[ExecuTorch][WebGPU] Add 256-thread "steel" q4gsw prefill GEMM
JCNTH
committed
[ExecuTorch][WebGPU] Steel q4gsw prefill GEMM — test coverage
JCNTH
committed
[ExecuTorch][WebGPU] Add f16-multiply variant of the steel q4gsw prefill GEMM
JCNTH
committed
[ExecuTorch][WebGPU] Test coverage for the f16-multiply steel q4gsw GEMM
JCNTH
committed
[ExecuTorch][WebGPU] Add opt-in native f16 KV cache (−280 MiB, default-OFF)
JCNTH
committed
[ExecuTorch][WebGPU] Test coverage for the f16 KV cache
JCNTH
committed
[ExecuTorch][WebGPU] Reuse pool for single-op scratch (-86.7 MiB)
JCNTH
committed
[ExecuTorch][WebGPU] Packed-word-dequant f16 steel q4gsw prefill GEMM
JCNTH
committed
[ExecuTorch][WebGPU] Test coverage for the packed-word-dequant f16 steel GEMM
JCNTH
committed
[ExecuTorch][WebGPU] f16-accumulate (pwdqf16acc) steel q4gsw prefill GEMM
JCNTH
committed
[ExecuTorch][WebGPU] Test coverage for the f16-accumulate steel GEMM
JCNTH
committed