zhreshold commented on a change in pull request #14139: Performance improvement in Normalize GPU Kernel URL: https://github.com/apache/incubator-mxnet/pull/14139#discussion_r256639710
########## File path: src/operator/image/image_random.cu ########## @@ -99,18 +111,134 @@ void ToTensorImplCUDA(mshadow::Stream<gpu> *s, W = input.size(2); C = input.size(3); blocks = N > 0 ? N : 1; - blocks = N; } - // One block per image. - // Number of threads = (32, 32) is optimal, because, - // computation is minimal and overhead of CUDA preparing - // all threads is minimal. + ToTensorCudaKernel<gpu, DType> - <<<blocks, dim3(32, 32), 0, stream>>>(input, output, + <<<blocks, dim3(H, cuda::kMaxThreadsPerBlock / H), 0, stream>>>(input, output, Review comment: please fix ToTensor similarly in a separate PR. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services