// Initialize CUDA streams and cuDNN. stream_ = new cudaStream_t[this->group_ * CUDNN_STREAMS_PER_GROUP]; handle_ = new cudnnHandle_t[this->group_ * CUDNN_STREAMS_PER ...
V0.1_75_101 is compiling. It is cudnn 7.5 w/ cuda 10.1, There might be bugs. Send me a pull request. I made a BatchNormalD descriptor and BatchNormDEx descriptor. You will call this with a "Create" ...