ROCm / rocWMMA

rocWMMA
https://rocm.docs.amd.com/projects/rocWMMA/
MIT License
91 stars 26 forks source link

Reduce count of cold and hot run #410

Closed CongMa13 closed 4 months ago

CongMa13 commented 4 months ago

Reduce count of cold and hot run since it takes 5 hrs to run the bench test on navi3x

gemm test kernel used to run 5 times, but 12 times(2 cold + 10 hot) after 4daef2129da29a4a24eaaf6ca9075d2a12e198a6

Reduce the count of (cold + hot) run to make the execution time less than 3 hours on navi3x platforms