[ExecuTorch][WebGPU] Add f16-multiply variant of the steel q4gsw prefill GEMM - #20817

Merged

JCNTH merged 9 commits into

gh/JCNTH/5/origfrom

gh/JCNTH/6/orig

Jul 9, 2026

Commits on Jul 9, 2026

[ExecuTorch][WebGPU] Add f16-multiply variant of the steel q4gsw prefill GEMM
JCNTH
committed
e1a58b2
[ExecuTorch][WebGPU] Test coverage for the f16-multiply steel q4gsw GEMM
JCNTH
committed
049d2ca
[ExecuTorch][WebGPU] Add opt-in native f16 KV cache (−280 MiB, default-OFF)
JCNTH
committed
6f8b480
[ExecuTorch][WebGPU] Test coverage for the f16 KV cache
JCNTH
committed
1ee483f
[ExecuTorch][WebGPU] Reuse pool for single-op scratch (-86.7 MiB)
JCNTH
committed
e2329b8
[ExecuTorch][WebGPU] Packed-word-dequant f16 steel q4gsw prefill GEMM
JCNTH
committed
ac021a8
[ExecuTorch][WebGPU] Test coverage for the packed-word-dequant f16 steel GEMM
JCNTH
committed
3424b1d
[ExecuTorch][WebGPU] f16-accumulate (pwdqf16acc) steel q4gsw prefill GEMM
JCNTH
committed
0cc2d41
[ExecuTorch][WebGPU] Test coverage for the f16-accumulate steel GEMM
JCNTH
committed
9147926