Hzfengsy opened a new pull request, #13891: URL: https://github.com/apache/tvm/pull/13891
This PR adds Tensor Core intrinsics with `shared.dyn` scope and changes the default rules to use `shared.dyn`. Here are the performance improvement of GEMM 1024x1024x1024 on my device (RTX-3080) | | Use `shared` | Use `shared.dyn`. | Speedup | | ----------- | --------------------- | -------------------- | ---------- | fp 16-16-16 | 66399.8766 GFLOPs | 71778.3808 GFLOPs | 8.1% | fp 16-16-32 | 44292.5893 GFLOPs | 49070.2514 GFLOPS | 10.8% | cc @vinx13 @junrushao -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
