experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Fix for the broken --config=rocm build

Browse Source 
The --config=rocm build was broken by the merge for PR #26840

This commit backs out the ROCm support in the file avgpooling_op.cc (added by the above PR). This is because the the template instantiations required for GPU support of the average pooling operator (which are in avgpooling_op_gpu.cu.cc) also need to be enabled for ROCm at the same time (as the code in avgpooling_op.cc) in order to avoid link errors with the `--config=rocm` build.  Enabling ROCm support for the code in avgpooling_op_gpu.cu.cc requires other PRs (the set spwaned from PR #28343) to be merged first. Once those PRs are merged, we will file another PR to re-enable ROCm support in the avgpooling*.cc files.
This commit is contained in:
Deven Desai 2019-05-13 18:23:18 +00:00
parent 6e18e7c812
commit 151ebd0bc6
1 changed files with 6 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
tensorflow/core/kernels/avgpooling_op.cc
View File

@ -36,10 +36,10 @@ limitations under the License.
#include "tensorflow/core/util/padding.h" #include "tensorflow/core/util/padding.h"
#include "tensorflow/core/util/tensor_format.h" #include "tensorflow/core/util/tensor_format.h"
#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM #if GOOGLE_CUDA
#include "tensorflow/core/kernels/maxpooling_op_gpu.h" #include "tensorflow/core/kernels/maxpooling_op_gpu.h"
#include "tensorflow/core/kernels/pooling_ops_common_gpu.h" #include "tensorflow/core/kernels/pooling_ops_common_gpu.h"
#endif // GOOGLE_CUDA || TENSORFLOW_USE_ROCM #endif // GOOGLE_CUDA
namespace tensorflow { namespace tensorflow {
@ -112,7 +112,7 @@ REGISTER_KERNEL_BUILDER(
Name("AvgPool").Device(DEVICE_CPU).TypeConstraint<Eigen::half>("T"), Name("AvgPool").Device(DEVICE_CPU).TypeConstraint<Eigen::half>("T"),
AvgPoolingOp<CPUDevice, Eigen::half>); AvgPoolingOp<CPUDevice, Eigen::half>);
#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM #if GOOGLE_CUDA
template <typename T> template <typename T>
class AvgPoolingOp<GPUDevice, T> : public UnaryOp<T> { class AvgPoolingOp<GPUDevice, T> : public UnaryOp<T> {
public: public:
@ -205,7 +205,7 @@ REGISTER_KERNEL_BUILDER(
REGISTER_KERNEL_BUILDER( REGISTER_KERNEL_BUILDER(
Name("AvgPool").Device(DEVICE_GPU).TypeConstraint<double>("T"), Name("AvgPool").Device(DEVICE_GPU).TypeConstraint<double>("T"),
AvgPoolingOp<GPUDevice, double>); AvgPoolingOp<GPUDevice, double>);
#endif // GOOGLE_CUDA || TENSORFLOW_USE_ROCM #endif // GOOGLE_CUDA
// The operation to compute AvgPool gradients. // The operation to compute AvgPool gradients.
// It takes two inputs: // It takes two inputs:
@ -368,7 +368,7 @@ TF_CALL_float(REGISTER_CPU_KERNEL);
TF_CALL_double(REGISTER_CPU_KERNEL); TF_CALL_double(REGISTER_CPU_KERNEL);
TF_CALL_half(REGISTER_CPU_KERNEL); TF_CALL_half(REGISTER_CPU_KERNEL);
#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM #if GOOGLE_CUDA
// A CUDNN based AvgPoolingGrad implementation. It includes the padding as the // A CUDNN based AvgPoolingGrad implementation. It includes the padding as the
// candidates for the pooling operation. // candidates for the pooling operation.
@ -577,6 +577,6 @@ REGISTER_KERNEL_BUILDER(Name("AvgPoolGrad")
.HostMemory("orig_input_shape"), .HostMemory("orig_input_shape"),
AvgPoolingGradOpCustomGPUKernel<Eigen::half>); AvgPoolingGradOpCustomGPUKernel<Eigen::half>);
#endif // GOOGLE_CUDA || TENSORFLOW_USE_ROCM #endif // GOOGLE_CUDA
} // namespace tensorflow } // namespace tensorflow