onnxruntime
[CUDA] Add tinygemm2 as a GEMM auto-tune candidate for small-M MatMul on SM 9.0+
#32901

Merged

[CUDA] Add tinygemm2 as a GEMM auto-tune candidate for small-M MatMul on SM 9.0+ #32901

jiafatom merged 4 commits into main from tlwu/cuda-gemm-tinygemm2

Base automatically changed from tlwu/cuda-small-m-decode-gemv to main 40 days ago

perf(cuda): add tinygemm2 as a GEMM auto-tune candidate on SM 9.0+

49745ac6

tianleiwu requested a review from

copilot-pull-request-reviewer 40 days ago

tianleiwu force pushed from cfb744f5 to 49745ac6 40 days ago

copilot-pull-request-reviewer commented on 2026-09-30

tianleiwu commented on 2026-10-03

fix(cuda): guard tinygemm2 dependent launches

d22cb1ff

fix(cuda): wrap tinygemm2 tensor map parameters

4b5cfec6

fix(cuda): align tinygemm2 maps in device ABI

bd018822

jiafatom approved these changes on 2026-10-05

jiafatom merged d2d1e8fe into main 35 days ago

jiafatom deleted the tlwu/cuda-gemm-tinygemm2 branch 35 days ago

Reviewers

jiafatom

Assignees

No one assigned

Labels

None yet

Milestone

No milestone