[WebGPU] Widen MatMulNBits SubgroupMatrix-8x16x16 tiling to 128x256 #32605
daijh
marked this pull request as draft 36 days ago
daijh
force pushed
from
9c3af0c3
to
473f0a1e
34 days ago
daijh
changed the title [WebGPU] Widen SubgroupMatrixMatMulNBits tile size to 128x256 [WebGPU] Widen SubgroupMatrix 8x16x16 tiling to 128x256 34 days ago
daijh
force pushed
from
473f0a1e
to
181a16da
34 days ago
daijh
force pushed
from
181a16da
to
9b8ba9aa
34 days ago
daijh
marked this pull request as ready for review 34 days ago
daijh
force pushed
from
6c3edc0f
to
20246354
30 days ago
daijh
force pushed
from
20246354
to
0892a255
29 days ago
[WebGPU] Widen SubgroupMatrix 8x16x16 tiling to 128x256
124a0408
Update comments
16754008
Optimize for small M
66749141
Fix js build: avoid weight_index_indirect.getByOffset
707bad83
webgpu: tune subgroup matrix nbits matmul
4878bd5b
webgpu: tune subgroup matrix nbits matmul
c3a8bb2e
webgpu: reconcile subgroup matrix tuning with main
cf6480b7
daijh
force pushed
from
9cedcf41
to
cf6480b7
13 days ago
daijh
changed the title [WebGPU] Widen SubgroupMatrix 8x16x16 tiling to 128x256 [WebGPU] Widen MatMulNBits SubgroupMatrix-8x16x16 tiling to 128x256 13 days ago
Update
3b3ccaf5
jchen10
approved these changes
on 2026-10-09
webgpu: clarify pack naming and inline offsets helper
d8b9a63f
qjia7
commented
on 2026-10-09
WebGPU: revert 16x16x16 MatMulNBits N alignment to previous restriction
4e86cd72
WebGPU: enforce MatMulNBits workgroup device limits
08fcf984
Add TODO for 16x16x16
244e151f
Update
fa30dd14
Update
40093670
Add back comments
0279c705
qjia7
approved these changes
on 2026-10-10
qjia7
enabled auto-merge (squash) 11 days ago
qjia7
merged
bb47315d
into main 11 days ago
daijh
deleted the pr/matmulnbits-sgm branch 11 days ago
Assignees
No one assigned
Login to write a write a comment.
Login via GitHub