[GitHub] TaoLv commented on a change in pull request #10116: [MXNET-105] Fix CuDNN performance after code refactor

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

TaoLv commented on a change in pull request #10116: [MXNET-105] Fix CuDNN performance after code refactor
URL: https://github.com/apache/incubator-mxnet/pull/10116#discussion_r177113032
 
 

 ##########
 File path: src/operator/nn/batch_norm.cu
 ##########
 @@ -705,19 +700,18 @@ void BatchNormGradCompute<gpu>(const nnvm::NodeAttrs& attrs,
   if (!param.use_global_stats && !param.cudnn_off && shape.ndim() <= 4
       && param.axis == mxnet::op::batchnorm::DEFAULT_AXIS) {
     MSHADOW_REAL_TYPE_SWITCH(dtype, DType, {
-      GetCuDNNOp<DType>(param).Backward(ctx, out_grad, in_data, out_data,
-        req, in_grad, aux_states);
+      GetCuDNNOp<DType>(param).Backward(ctx, inputs, req, outputs);
     })
   } else {
     MSHADOW_REAL_TYPE_SWITCH_EX(dtype, DType, AccReal, {
-      BatchNormBackward<gpu, DType, AccReal>(ctx, param, out_grad,
-          in_data, out_data, req, in_grad, aux_states);
+      BatchNormBackward<gpu, DType, AccReal>(ctx, param, inputs, req, outputs);
     })
   }
 #else
+  aux_states[batchnorm::kMovingMean] = inputs[6];
+  aux_states[batchnorm::kMovingVar] = inputs[7];
 
 Review comment:
   @zheng-da aux_states is not defined if USE_CUDNN is not enabled. @marcoabreu seems there is no pure cuda ci environment which is not built with cudnn.

----------------------------------------------------------------
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