[WebGPU] Implement Split-K on Conv|MatMul #26461
Implement Split-K on Conv|MatMul
46e45590
Address reviewer's comments
1f06b95f
Remove the check of `is_channels_last` in `UseSplitK`
31938158
Still require `is_channels_last` to be true
ecbc0933
Jiawei-Shao
marked this pull request as ready for review 56 days ago
Check the use of Split-K with ratio and enable Split-K on ACM
82d3d9b4
Fix incorrect ratio
0099eddf
Update ratio
05bd1f84
Update ratio
11ecdfea
Jiawei-Shao
changed the title Implement Split-K on Conv|MatMul [webgpu] Implement Split-K on Conv|MatMul 53 days ago
Jiawei-Shao
changed the title [webgpu] Implement Split-K on Conv|MatMul [WebGPU] Implement Split-K on Conv|MatMul 53 days ago
Compute FP16 values with MLFloat16
534dc2c2
qjia7
commented
on 2025-11-06
Address reviewer's comments
cfd22194
Disallow out-of-bound write
d03755b5
Use safer thresholds by now
581828e6
Merge branch 'main' into impl-splitk-matmul
2ed25de6
qjia7
commented
on 2025-11-12
Address more reviewer's comments
22f9017e
Address comments from Copilot
418d6c0c
Address more comments from Copilot
0ca5e656
Address more comments from Copilot
7a415de7
qjia7
commented
on 2025-11-14
Address reviewer's comments
13b94e86
Don't call `SetWorkgroupSize()` as we are using the default value
082d1e31
Remove a redundant declaration
6b15ede1
Merge branch 'main' into impl-splitk-matmul
04e2890a
Address comments from Copilot
fb4c7430
Remove unused declarations
4ef3ac2d
Fix another typo
fa6f2263
qjia7
dismissed these changes
on 2025-11-14
Merge branch 'main' into impl-splitk-matmul
ec8d47da
Use a higher rel_error for Linux ARM64 bots
d010d4ba
Jiawei-Shao
dismissed their stale review
via d010d4ba
42 days ago
guschmue
approved these changes
on 2025-11-17
guschmue
merged
607d5e4d
into main 38 days ago
Jiawei-Shao
deleted the impl-splitk-matmul branch 33 days ago
Assignees
No one assigned
Login to write a write a comment.
Login via GitHub