ptrendx commented on a change in pull request #13362: Add NHWC layout support
to Pooling (cuDNN only)
URL: https://github.com/apache/incubator-mxnet/pull/13362#discussion_r239279756
##########
File path: src/operator/nn/pooling.cu
##########
@@ -116,10 +118,12 @@ void PoolingGradCompute<gpu>(const nnvm::NodeAttrs&
attrs,
switch (param.pool_type) {
case pool_enum::kMaxPooling:
case pool_enum::kAvgPooling:
- GetCuDNNPoolingOp<DType>(param).Backward(ctx, inputs[ograd_idx],
+ if (GetCuDNNPoolingOp<DType>(param).Backward(ctx, inputs[ograd_idx],
inputs[in_data_idx],
inputs[out_data_idx],
- req[0], outputs[0]);
- return;
+ req[0], outputs[0])) {
+ return;
+ }
+ break;
Review comment:
No - if Backward returns false, program will exit the switch statement and
proceed to non-cuDNN implementation.
----------------------------------------------------------------
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:
[email protected]
With regards,
Apache Git Services
