onnxruntime
[WebGPU] LinearAttention: increase tile_v when subgroups are available
#28519
Merged

[WebGPU] LinearAttention: increase tile_v when subgroups are available #28519

guschmue merged 6 commits into microsoft:main from daijh:linear-attn-dev
daijh
daijh daijh marked this pull request as draft 180 days ago
daijh [WebGPU] LinearAttention: increase tile_v when subgroups are available
5a8762d3
daijh Add comments
e84d6752
guschmue guschmue added ep:WebGPU
daijh Gate the tile_v expansion on seq_length >= 16
4cb94d53
daijh daijh force pushed from 58a09c74 to 4cb94d53 176 days ago
daijh daijh marked this pull request as ready for review 176 days ago
daijh
qjia7 qjia7 requested a review from copilot-pull-request-reviewer copilot-pull-request-reviewer 174 days ago
copilot-pull-request-reviewer
copilot-pull-request-reviewer commented on 2026-05-21
daijh Clamp tile_v to workgroup_size
315460be
daijh
qjia7
qjia7 commented on 2026-05-22
daijh Add TODO to support vec2 (components==2) in linear attention
c8f87882
daijh Simplify components check: remove redundant head_dim_v >= 4 guard
c7520c0c
guschmue
guschmue approved these changes on 2026-05-22
guschmue guschmue merged 1053327e into main 173 days ago
daijh daijh deleted the linear-attn-dev branch 170 days ago

Login to write a write a comment.

Login via GitHub

Assignees
No one assigned
Labels
Milestone