Add CUDA implementation of GatherBlockQuantized operator #25575
kunal-vaishnavi
merged
24 commits into
microsoft:main
from
xiaomsft:xiaoh/gather_block_quantized_cuda
kunal-vaishnavi
commented
on 2025-07-29
kunal-vaishnavi
commented
on 2025-07-29
kunal-vaishnavi
commented
on 2025-07-29
kunal-vaishnavi
commented
on 2025-07-29
kunal-vaishnavi
commented
on 2025-07-29
kunal-vaishnavi
commented
on 2025-07-29
tianleiwu
commented
on 2025-07-29
tianleiwu
commented
on 2025-07-30
tianleiwu
commented
on 2025-07-30
tianleiwu
dismissed these changes
on 2025-08-01
Add CUDA implementation of GatherBlockQuantized operator
7a63bfaf
Clang format
4256d1bc
Run with lint runner
ec4c2d3d
Add indices shape check
5d159c26
Fix webgpu tests
5bf1c918
Disable webgpu tests which are failing
d992d72e
remove unused formal parameter
8f16bac0
void unused parameter
fc31dc40
void unused parameter
cbad96f2
Fix nodiscard status
20ed595f
Fix integer precision
9a395218
cast raw data instead of using template for data fetching
92da3cda
Fix CUDA_MS_OP_THREE_TYPED_CLASS_NAME
941dd734
Fix mismatched marcro name
eb4522d3
Fix BuildKernelCreateInfo
ef55de87
Use raw data
7167c2dd
Fix typo
75e497b1
Fix axis check
b023cd8a
Fix kernel
15c32198
Support both signed and unsigned data type
2f73aac7
Fix templates
a7020376
Fix sign type
eab73b58
Disable more tests
0f67c22e
Update autogenerated doc
0c7938e0
kunal-vaishnavi
approved these changes
on 2025-08-01
tianleiwu
approved these changes
on 2025-08-01
Reviewers
Assignees
No one assigned
Labels
cherry-picked
Milestone
No milestone
Login to write a write a comment.
Login via GitHub