onnxruntime
[webgpu] Optimize MatMulNBits for f16 Block32 prefill performance
#23908
Merged
Merged
[webgpu] Optimize MatMulNBits for f16 Block32 prefill performance #23908
guschmue
merged
12 commits into
microsoft:main
from
daijh:matmul-f16-block32-prefill
sushraja-msft
commented
on 2025-03-06
sushraja-msft
commented
on 2025-03-06
sushraja-msft
commented
on 2025-03-06
sushanthr
commented
on 2025-03-25
sushanthr
commented
on 2025-03-25
sushanthr
commented
on 2025-03-25
sushanthr
commented
on 2025-03-25
sushanthr
approved these changes
on 2025-03-25
sushanthr
commented
on 2025-03-26
sushanthr
approved these changes
on 2025-03-26
[webgpu] Optimize MatMulNBits for f16 Block32 prefill performance
1be49b21
Resolve comments
4ead0043
Fix variable naming
14bbe9df
Add comment on f32 accumulator
0f55827f
Improve comment
695d9d05
More comment and avoid magic number
fd751cbd
Improve variable naming
58d76f62
Add tile_m and tile_n into constructor
ae482a2a
Rename to MatMulNBitsWideTileProgram
287be7e3
Improve comment to reflect new naming
ca1710a9
Fix lint
17c0b1f3
guschmue
dismissed these changes
on 2025-04-02
Prefer onnxruntime::narrow
e7f8bb43
guschmue
approved these changes
on 2025-04-04
Login to write a write a comment.
Reviewers
Assignees
No one assigned
Labels
ep:WebGPU
Milestone
No milestone