Skip to content

[ExecuTorch][WebGPU] Remove the orphan q4gsw_linear_coop4 shader #21007

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
JCNTH merged 2 commits into from
Jul 17, 2026
+235 −202
Merged

[ExecuTorch][WebGPU] Remove the orphan q4gsw_linear_coop4 shader #21007

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
5b4ee97
[ExecuTorch][WebGPU] Remove the orphan q4gsw_linear_coop4 shader
JCNTH Jul 15, 2026
b418824
[ExecuTorch][WebGPU] Test: rms_norm workgroup size is runtime-configu…
JCNTH Jul 15, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
87 changes: 0 additions & 87 deletions backends/webgpu/runtime/ops/quantized_linear/q4gsw_linear_coop4.wgsl
View file
Open in desktop

This file was deleted.

111 changes: 0 additions & 111 deletions backends/webgpu/runtime/ops/quantized_linear/q4gsw_linear_coop4_wgsl.h
View file
Open in desktop

This file was deleted.

4 changes: 2 additions & 2 deletions backends/webgpu/test/ops/test_quantized_linear.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -57,8 +57,8 @@ class Q4gswConfig:
# decode GEMV: the handler routes M==1 -> bicol, so each reads its own per-
# column scale (col0/col1) across many K-groups (down_proj: 256 groups). q4gsw
# requires N % 8 == 0 (torchao pads N for the scale layout), so odd-N / N=1 are
# not exportable -- bicol's has1 odd-N guard is defensive (mirrors coop4's
# general-N robustness) and unreachable through this op.
# not exportable -- bicol's has1 odd-N guard is defensive and unreachable
# through this op.
# M>1 prefill: prefer the steel GEMM (K%16==0) on a >=256-invocation device
# (e.g. lvp); else shmem (K>=4096 or N>=2048) or register-tiled (SwiftShader
# caps at 128). Same fp64 golden regardless of which kernel runs.
Expand Down
Loading
Loading