LeiWang1999 commented on PR #15462: URL: https://github.com/apache/tvm/pull/15462#issuecomment-1663336808
> Thank you, @LeiWang1999 for bringing cudnn backend :) Regarding find_cudnn_best_algo, there are existing API for both Python and C++ sides that you can reuse. https://github.com/apache/tvm/blob/unity/python/tvm/contrib/cudnn.py#L367 https://github.com/apache/tvm/blob/unity/src/runtime/contrib/cudnn/conv_forward.cc#L212 > > By the way, do you also plan to work on attention operators by chance? https://docs.nvidia.com/deeplearning/cudnn/api/index.html#cudnnMultiHeadAttnForward Thanks, I understand that there are existing APIs for finding the best algorithm, but my confusion lies in determining when to incorporate the find_best_algo function. In certain inference frameworks, they utilize a static flag to enable algorithm discovery during the warmup phase. but I don't no if tvm can enable this at runtime. -- 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]
