Fixed bug in Tensor Cores V100 (1. Desc in Batch norm, 2. Manually selected algo).

Also fixed time measure on Linux for multi-threading.
This commit is contained in:
AlexeyAB
2018-04-15 01:51:21 +03:00
parent 16cfff811f
commit eb9c88ef73
6 changed files with 74 additions and 26 deletions

View File

@ -281,7 +281,7 @@ struct layer{
#ifdef CUDNN
cudnnTensorDescriptor_t srcTensorDesc, dstTensorDesc;
cudnnTensorDescriptor_t dsrcTensorDesc, ddstTensorDesc;
cudnnTensorDescriptor_t normTensorDesc;
cudnnTensorDescriptor_t normTensorDesc, normDstTensorDesc;
cudnnFilterDescriptor_t weightDesc;
cudnnFilterDescriptor_t dweightDesc;
cudnnConvolutionDescriptor_t convDesc;