diff options
Diffstat (limited to 'tensorflow/core/kernels/cwise_op_sub.cc')
| -rw-r--r-- | tensorflow/core/kernels/cwise_op_sub.cc | 12 |
1 files changed, 12 insertions, 0 deletions
diff --git a/tensorflow/core/kernels/cwise_op_sub.cc b/tensorflow/core/kernels/cwise_op_sub.cc index e1326dbed1..eab1e2a09c 100644 --- a/tensorflow/core/kernels/cwise_op_sub.cc +++ b/tensorflow/core/kernels/cwise_op_sub.cc @@ -32,6 +32,18 @@ REGISTER(BinaryOp, CPU, "Sub", functor::sub, int32); .TypeConstraint<TYPE>("T"), \ BinaryOp<SYCLDevice, functor::sub<TYPE>>); REGISTER_SYCL_KERNEL(float); + REGISTER_SYCL_KERNEL(double); + +// A special GPU kernel for int32. +// TODO(b/25387198): Also enable int32 in device memory. This kernel +// registration requires all int32 inputs and outputs to be in host memory. +REGISTER_KERNEL_BUILDER(Name("Sub") + .Device(DEVICE_SYCL) + .HostMemory("x") + .HostMemory("y") + .HostMemory("z") + .TypeConstraint<int32>("T"), + BinaryOp<CPUDevice, functor::sub<int32>>); #undef REGISTER_SYCL_KERNEL #endif // TENSORFLOW_USE_SYCL #if GOOGLE_CUDA |