[GitHub] anirudh2290 closed pull request #11270: Enable CUDNN for conv1D (#11194)

[GitBox <gi...@apache.org>]

anirudh2290 closed pull request #11270: Enable CUDNN for conv1D (#11194)
URL: https://github.com/apache/incubator-mxnet/pull/11270
 
 
   

This is a PR merged from a forked repository.
As GitHub hides the original diff on merge, it is displayed below for
the sake of provenance:

As this is a foreign pull request (from a fork), the diff is supplied
below (as it won't show otherwise due to GitHub magic):

diff --git a/src/operator/nn/convolution.cu b/src/operator/nn/convolution.cu
index 65a320ded16..9f61212d5c7 100644
--- a/src/operator/nn/convolution.cu
+++ b/src/operator/nn/convolution.cu
@@ -89,8 +89,11 @@ void ConvolutionCompute<gpu>(const nnvm::NodeAttrs& attrs,
   const ConvolutionParam& param = nnvm::get<ConvolutionParam>(attrs.parsed);
   int dtype = inputs[conv::kData].type_flag_;
 
-  // If 1D convolution, use MXNet implementation
-  if (param.kernel.ndim() == 1) {
+#if CUDNN_MAJOR < 5
+  if (param_.layout.value() != kNCW &&
+      param_.layout.value() != kNCHW &&
+      param_.layout.value() != kNCDHW) {
+    // Need CuDNN > 5.0 for layout support. use MXNet implementation
     MSHADOW_REAL_TYPE_SWITCH(dtype, DType, {
       ConvolutionOp<gpu, DType> op;
       op.Init(param);
@@ -98,6 +101,8 @@ void ConvolutionCompute<gpu>(const nnvm::NodeAttrs& attrs,
     })
     return;
   }
+#endif
+
 #if MXNET_USE_CUDNN == 0 || CUDNN_MAJOR < 7
   if (param.num_filter == param.num_group &&
       param.layout.value() == mshadow::kNCHW &&
@@ -162,8 +167,11 @@ void ConvolutionGradCompute<gpu>(const nnvm::NodeAttrs& attrs,
   const std::vector<TBlob> &in_grad = outputs;
   int dtype = out_grad.type_flag_;
 
-  // If 1D convolution, use MXNet implementation
-  if (param.kernel.ndim() == 1) {
+#if CUDNN_MAJOR < 5
+  if (param_.layout.value() != kNCW &&
+      param_.layout.value() != kNCHW &&
+      param_.layout.value() != kNCDHW) {
+    // Need CuDNN > 5.0 for layout support. use MXNet implementation
     MSHADOW_REAL_TYPE_SWITCH(dtype, DType, {
       ConvolutionOp<gpu, DType> op;
       op.Init(param);
@@ -171,6 +179,7 @@ void ConvolutionGradCompute<gpu>(const nnvm::NodeAttrs& attrs,
     })
     return;
   }
+#endif
 #if MXNET_USE_CUDNN == 0 || CUDNN_MAJOR < 7
   if (param.num_filter == param.num_group &&
       param.layout.value() == mshadow::kNCHW &&


 

----------------------------------------------------------------
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:
users@infra.apache.org


With regards,
Apache Git Services