diff options
author | Manjunath Kudlur <keveman@gmail.com> | 2015-11-06 16:27:58 -0800 |
---|---|---|
committer | Manjunath Kudlur <keveman@gmail.com> | 2015-11-06 16:27:58 -0800 |
commit | f41959ccb2d9d4c722fe8fc3351401d53bcf4900 (patch) | |
tree | ef0ca22cb2a5ac4bdec9d080d8e0788a53ed496d /tensorflow/core/kernels/reduction_ops_sum.cc |
TensorFlow: Initial commit of TensorFlow library.
TensorFlow is an open source software library for numerical computation
using data flow graphs.
Base CL: 107276108
Diffstat (limited to 'tensorflow/core/kernels/reduction_ops_sum.cc')
-rw-r--r-- | tensorflow/core/kernels/reduction_ops_sum.cc | 37 |
1 files changed, 37 insertions, 0 deletions
diff --git a/tensorflow/core/kernels/reduction_ops_sum.cc b/tensorflow/core/kernels/reduction_ops_sum.cc new file mode 100644 index 0000000000..82d685e225 --- /dev/null +++ b/tensorflow/core/kernels/reduction_ops_sum.cc @@ -0,0 +1,37 @@ +#include "tensorflow/core/kernels/reduction_ops_common.h" + +namespace tensorflow { + +#define REGISTER_CPU_KERNELS(type) \ + REGISTER_KERNEL_BUILDER( \ + Name("Sum").Device(DEVICE_CPU).TypeConstraint<type>("T"), \ + ReductionOp<CPUDevice, type, Eigen::internal::SumReducer<type>>); +TF_CALL_REAL_NUMBER_TYPES(REGISTER_CPU_KERNELS); +#undef REGISTER_CPU_KERNELS + +// NOTE: We should have mean(complex64,int32), too. But that needs to +// change Eigen::internal::MeanReducer to cast int to complex<float>. +// We don't see immediate need of mean(complex64,int32) anyway. +REGISTER_KERNEL_BUILDER( + Name("Sum").Device(DEVICE_CPU).TypeConstraint<complex64>("T"), + ReductionOp<CPUDevice, complex64, Eigen::internal::SumReducer<complex64>>); + +#if GOOGLE_CUDA + +#define REGISTER_GPU_KERNELS(type) \ + REGISTER_KERNEL_BUILDER( \ + Name("Sum") \ + .Device(DEVICE_GPU) \ + .TypeConstraint<type>("T") \ + .HostMemory("reduction_indices"), \ + ReductionOp<GPUDevice, type, Eigen::internal::SumReducer<type>>); +REGISTER_GPU_KERNELS(float); +#undef REGISTER_GPU_KERNELS + +REGISTER_KERNEL_BUILDER( + Name("Sum").Device(DEVICE_GPU).TypeConstraint<complex64>("T"), + ReductionOp<GPUDevice, complex64, Eigen::internal::SumReducer<complex64>>); + +#endif + +} // namespace tensorflow |