aboutsummaryrefslogtreecommitdiffhomepage
path: root/tensorflow/core/kernels/cwise_op_sub.cc
diff options
context:
space:
mode:
Diffstat (limited to 'tensorflow/core/kernels/cwise_op_sub.cc')
-rw-r--r--tensorflow/core/kernels/cwise_op_sub.cc21
1 files changed, 21 insertions, 0 deletions
diff --git a/tensorflow/core/kernels/cwise_op_sub.cc b/tensorflow/core/kernels/cwise_op_sub.cc
new file mode 100644
index 0000000000..c3c5952f8d
--- /dev/null
+++ b/tensorflow/core/kernels/cwise_op_sub.cc
@@ -0,0 +1,21 @@
+#include "tensorflow/core/kernels/cwise_ops_common.h"
+
+namespace tensorflow {
+REGISTER5(BinaryOp, CPU, "Sub", functor::sub, float, double, int32, int64,
+ complex64);
+#if GOOGLE_CUDA
+REGISTER3(BinaryOp, GPU, "Sub", functor::sub, float, double, int64);
+#endif
+
+// 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_GPU)
+ .HostMemory("x")
+ .HostMemory("y")
+ .HostMemory("z")
+ .TypeConstraint<int32>("T"),
+ BinaryOp<CPUDevice, functor::sub<int32>>);
+
+} // namespace tensorflow