Add ROCm support for most cwise ops

tensorflow/core/kernels/cwise_op_abs.cc

@@ -20,7 +20,7 @@ REGISTER5(UnaryOp, CPU, "Abs", functor::abs, float, Eigen::half, double, int32,
           int64);
 REGISTER2(UnaryOp, CPU, "ComplexAbs", functor::abs, complex64, complex128);
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 REGISTER4(UnaryOp, GPU, "Abs", functor::abs, float, Eigen::half, double, int64);
 REGISTER2(UnaryOp, GPU, "ComplexAbs", functor::abs, complex64, complex128);
 

tensorflow/core/kernels/cwise_op_acos.cc

@@ -18,7 +18,7 @@ limitations under the License.
 namespace tensorflow {
 REGISTER2(UnaryOp, CPU, "Acos", functor::acos, float, double);
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 REGISTER2(UnaryOp, GPU, "Acos", functor::acos, float, double);
 #endif
 

tensorflow/core/kernels/cwise_op_acosh.cc

@@ -24,7 +24,7 @@ REGISTER4(UnaryOp, CPU, "Acosh", functor::acosh, float, double, complex64,
 REGISTER2(UnaryOp, SYCL, "Acosh", functor::acosh, float, double);
 #endif  // TENSORFLOW_USE_SYCL
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 REGISTER2(UnaryOp, GPU, "Acosh", functor::acosh, float, double);
 #endif
 }  // namespace tensorflow

tensorflow/core/kernels/cwise_op_add_1.cc

@@ -21,7 +21,7 @@ REGISTER6(BinaryOp, CPU, "Add", functor::add, float, Eigen::half, double, int32,
 REGISTER6(BinaryOp, CPU, "AddV2", functor::add, float, Eigen::half, double,
           int32, int64, bfloat16);
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 REGISTER3(BinaryOp, GPU, "Add", functor::add, float, Eigen::half, double);
 REGISTER3(BinaryOp, GPU, "AddV2", functor::add, float, Eigen::half, double);
 

tensorflow/core/kernels/cwise_op_add_2.cc

@@ -28,12 +28,12 @@ REGISTER6(BinaryOp, CPU, "Add", functor::add, int8, int16, complex64, uint8,
 // is_aggregate.
 REGISTER5(BinaryOp, CPU, "AddV2", functor::add, int8, int16, complex64, uint8,
           complex128);
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 REGISTER4(BinaryOp, GPU, "Add", functor::add, uint8, int64, complex64,
           complex128);
 REGISTER4(BinaryOp, GPU, "AddV2", functor::add, uint8, int64, complex64,
           complex128);
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #endif  // !defined(__ANDROID_TYPES_SLIM__)
 

tensorflow/core/kernels/cwise_op_asin.cc

@@ -18,7 +18,7 @@ limitations under the License.
 namespace tensorflow {
 REGISTER2(UnaryOp, CPU, "Asin", functor::asin, float, double);
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 REGISTER2(UnaryOp, GPU, "Asin", functor::asin, float, double);
 #endif
 

tensorflow/core/kernels/cwise_op_asinh.cc

@@ -24,7 +24,7 @@ REGISTER4(UnaryOp, CPU, "Asinh", functor::asinh, float, double, complex64,
 REGISTER2(UnaryOp, SYCL, "Asinh", functor::asinh, float, double);
 #endif  // TENSORFLOW_USE_SYCL
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 REGISTER2(UnaryOp, GPU, "Asinh", functor::asinh, float, double);
 #endif
 }  // namespace tensorflow

tensorflow/core/kernels/cwise_op_atan.cc

@@ -18,7 +18,7 @@ limitations under the License.
 namespace tensorflow {
 REGISTER2(UnaryOp, CPU, "Atan", functor::atan, float, double);
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 REGISTER2(UnaryOp, GPU, "Atan", functor::atan, float, double);
 #endif
 

tensorflow/core/kernels/cwise_op_atan2.cc

@@ -17,7 +17,7 @@ limitations under the License.
 
 namespace tensorflow {
 REGISTER2(BinaryOp, CPU, "Atan2", functor::atan2, float, double);
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 REGISTER2(BinaryOp, GPU, "Atan2", functor::atan2, float, double);
 #endif
 }  // namespace tensorflow

tensorflow/core/kernels/cwise_op_atanh.cc

@@ -24,7 +24,7 @@ REGISTER4(UnaryOp, CPU, "Atanh", functor::atanh, float, double, complex64,
 REGISTER2(UnaryOp, SYCL, "Atanh", functor::atanh, float, double);
 #endif  // TENSORFLOW_USE_SYCL
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 REGISTER2(UnaryOp, GPU, "Atanh", functor::atanh, float, double);
 #endif
 }  // namespace tensorflow

tensorflow/core/kernels/cwise_op_bessel.cc

@@ -20,7 +20,7 @@ REGISTER3(UnaryOp, CPU, "BesselI0e", functor::bessel_i0e, Eigen::half, float,
           double);
 REGISTER3(UnaryOp, CPU, "BesselI1e", functor::bessel_i1e, Eigen::half, float,
           double);
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 REGISTER3(UnaryOp, GPU, "BesselI0e", functor::bessel_i0e, Eigen::half, float,
           double);
 REGISTER3(UnaryOp, GPU, "BesselI1e", functor::bessel_i1e, Eigen::half, float,

tensorflow/core/kernels/cwise_op_bessel.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -24,4 +24,4 @@ DEFINE_UNARY3(bessel_i1e, Eigen::half, float, double);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_bitwise_and.cc

@@ -36,9 +36,9 @@ REGISTER_SYCL_KERNEL(uint64);
 
 #endif  // TENSORFLOW_USE_SYCL
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 REGISTER8(BinaryOp, GPU, "BitwiseAnd", functor::bitwise_and, int8, int16, int32,
           int64, uint8, uint16, uint32, uint64);
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 }  // namespace tensorflow

tensorflow/core/kernels/cwise_op_bitwise_or.cc

@@ -36,9 +36,9 @@ REGISTER_SYCL_KERNEL(uint64);
 
 #endif  // TENSORFLOW_USE_SYCL
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 REGISTER8(BinaryOp, GPU, "BitwiseOr", functor::bitwise_or, int8, int16, int32,
           int64, uint8, uint16, uint32, uint64);
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 }  // namespace tensorflow

tensorflow/core/kernels/cwise_op_bitwise_xor.cc

@@ -36,9 +36,9 @@ REGISTER_SYCL_KERNEL(uint64);
 
 #endif  // TENSORFLOW_USE_SYCL
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 REGISTER8(BinaryOp, GPU, "BitwiseXor", functor::bitwise_xor, int8, int16, int32,
           int64, uint8, uint16, uint32, uint64);
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 }  // namespace tensorflow

tensorflow/core/kernels/cwise_op_ceil.cc

@@ -18,7 +18,7 @@ limitations under the License.
 namespace tensorflow {
 REGISTER3(UnaryOp, CPU, "Ceil", functor::ceil, float, Eigen::half, double);
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 REGISTER3(UnaryOp, GPU, "Ceil", functor::ceil, float, Eigen::half, double);
 #endif
 

tensorflow/core/kernels/cwise_op_clip.cc

@@ -181,7 +181,7 @@ REGISTER_CPU_KERNEL(uint8);
 REGISTER_CPU_KERNEL(uint16);
 #undef REGISTER_CPU_KERNEL
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #define REGISTER_GPU_KERNEL(type)                                       \
   REGISTER_KERNEL_BUILDER(                                              \

tensorflow/core/kernels/cwise_op_complex.cc

@@ -26,7 +26,7 @@ namespace tensorflow {
 REGISTER_COMPLEX(CPU, float, complex64);
 REGISTER_COMPLEX(CPU, double, complex128);
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 REGISTER_COMPLEX(GPU, float, complex64);
 REGISTER_COMPLEX(GPU, double, complex128);
 #endif

tensorflow/core/kernels/cwise_op_conj.cc

@@ -22,7 +22,7 @@ REGISTER2(UnaryOp, CPU, "Conj", functor::conj, complex64, complex128);
 
 REGISTER_VARIANT(UnaryVariantOp, CPU, "Conj", CONJ_VARIANT_UNARY_OP);
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 REGISTER_KERNEL_BUILDER(
     Name("Conj").Device(DEVICE_GPU).TypeConstraint<Variant>("T"),
     UnaryVariantOp<GPUDevice, CONJ_VARIANT_UNARY_OP>);

tensorflow/core/kernels/cwise_op_cos.cc

@@ -19,7 +19,7 @@ namespace tensorflow {
 REGISTER5(UnaryOp, CPU, "Cos", functor::cos, float, Eigen::half, double,
           complex64, complex128);
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 REGISTER3(UnaryOp, GPU, "Cos", functor::cos, float, Eigen::half, double);
 #endif
 

tensorflow/core/kernels/cwise_op_cosh.cc

@@ -29,7 +29,7 @@ REGISTER_SYCL_KERNEL(double);
 #undef REGISTER_SYCL_KERNEL
 #endif  // TENSORFLOW_USE_SYCL
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 REGISTER2(UnaryOp, GPU, "Cosh", functor::cosh, float, double);
 #endif
 }  // namespace tensorflow

tensorflow/core/kernels/cwise_op_digamma.cc

@@ -18,8 +18,8 @@ limitations under the License.
 namespace tensorflow {
 REGISTER3(UnaryOp, CPU, "Digamma", functor::digamma, float, Eigen::half,
           double);
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 REGISTER3(UnaryOp, GPU, "Digamma", functor::digamma, float, Eigen::half,
           double);
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 }  // namespace tensorflow

tensorflow/core/kernels/cwise_op_equal_to_1.cc

@@ -24,7 +24,7 @@ REGISTER_KERNEL_BUILDER(
 REGISTER_KERNEL_BUILDER(
     Name("ApproximateEqual").Device(DEVICE_CPU).TypeConstraint<double>("T"),
     ApproximateEqualOp<CPUDevice, double>);
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 REGISTER4(BinaryOp, GPU, "Equal", functor::equal_to, float, Eigen::half, double,
           uint8);
 REGISTER_KERNEL_BUILDER(

tensorflow/core/kernels/cwise_op_equal_to_2.cc

@@ -24,10 +24,10 @@ namespace tensorflow {
 
 REGISTER6(BinaryOp, CPU, "Equal", functor::equal_to, int32, int64, complex64,
           complex128, string, bool);
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 REGISTER6(BinaryOp, GPU, "Equal", functor::equal_to, int8, int16, int64,
           complex64, complex128, bool);
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #endif  // !defined(__ANDROID_TYPES_SLIM__)
 

tensorflow/core/kernels/cwise_op_erf.cc

@@ -17,7 +17,7 @@ limitations under the License.
 
 namespace tensorflow {
 REGISTER3(UnaryOp, CPU, "Erf", functor::erf, float, Eigen::half, double);
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 REGISTER3(UnaryOp, GPU, "Erf", functor::erf, float, Eigen::half, double);
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 }  // namespace tensorflow

tensorflow/core/kernels/cwise_op_erfc.cc

@@ -17,7 +17,7 @@ limitations under the License.
 
 namespace tensorflow {
 REGISTER3(UnaryOp, CPU, "Erfc", functor::erfc, float, Eigen::half, double);
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 REGISTER3(UnaryOp, GPU, "Erfc", functor::erfc, float, Eigen::half, double);
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 }  // namespace tensorflow

tensorflow/core/kernels/cwise_op_exp.cc

@@ -19,7 +19,7 @@ namespace tensorflow {
 REGISTER5(UnaryOp, CPU, "Exp", functor::exp, float, Eigen::half, double,
           complex64, complex128);
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 REGISTER5(UnaryOp, GPU, "Exp", functor::exp, float, Eigen::half, double,
           complex64, complex128);
 #endif

tensorflow/core/kernels/cwise_op_expm1.cc

@@ -18,7 +18,7 @@ limitations under the License.
 namespace tensorflow {
 REGISTER5(UnaryOp, CPU, "Expm1", functor::expm1, float, Eigen::half, double,
           complex64, complex128);
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 REGISTER3(UnaryOp, GPU, "Expm1", functor::expm1, float, Eigen::half, double);
 #endif
 #ifdef TENSORFLOW_USE_SYCL

tensorflow/core/kernels/cwise_op_floor.cc

@@ -18,7 +18,7 @@ limitations under the License.
 namespace tensorflow {
 REGISTER3(UnaryOp, CPU, "Floor", functor::floor, float, Eigen::half, double);
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 REGISTER3(UnaryOp, GPU, "Floor", functor::floor, float, Eigen::half, double);
 #endif
 #ifdef TENSORFLOW_USE_SYCL

tensorflow/core/kernels/cwise_op_floor_div.cc

@@ -21,14 +21,14 @@ REGISTER6(BinaryOp, CPU, "FloorDiv", functor::safe_floor_div, uint8, uint16,
 REGISTER3(BinaryOp, CPU, "FloorDiv", functor::floor_div_real, float,
           Eigen::half, double);
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 REGISTER4(BinaryOp, GPU, "FloorDiv", functor::floor_div, uint8, uint16, int16,
           int64);
 REGISTER3(BinaryOp, GPU, "FloorDiv", functor::floor_div_real, float,
           Eigen::half, double);
 #endif
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 // 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.

tensorflow/core/kernels/cwise_op_floor_mod.cc

@@ -19,7 +19,7 @@ namespace tensorflow {
 REGISTER2(BinaryOp, CPU, "FloorMod", functor::safe_floor_mod, int32, int64);
 REGISTER2(BinaryOp, CPU, "FloorMod", functor::floor_fmod, float, double);
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 // 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.

tensorflow/core/kernels/cwise_op_gpu_abs.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_UNARY6(abs, Eigen::half, float, double, int64, complex64, complex128);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_acos.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_UNARY2(acos, float, double);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_acosh.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 #include "tensorflow/core/kernels/cwise_ops_gpu_gradients.cu.h"
@@ -24,4 +24,4 @@ DEFINE_UNARY2(acosh, float, double);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_add.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -24,4 +24,4 @@ DEFINE_BINARY7(add, Eigen::half, float, double, uint8, int64, complex64,
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_asin.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_UNARY2(asin, float, double);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_asinh.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 #include "tensorflow/core/kernels/cwise_ops_gpu_gradients.cu.h"
@@ -24,4 +24,4 @@ DEFINE_UNARY2(asinh, float, double);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_atan.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_UNARY2(atan, float, double);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_atan2.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_BINARY2(atan2, float, double);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_atanh.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 #include "tensorflow/core/kernels/cwise_ops_gpu_gradients.cu.h"
@@ -24,4 +24,4 @@ DEFINE_UNARY2(atanh, float, double);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_bitwise_and.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -24,4 +24,4 @@ DEFINE_BINARY8(bitwise_and, int8, int16, int32, int64, uint8, uint16, uint32,
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_bitwise_or.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -24,4 +24,4 @@ DEFINE_BINARY8(bitwise_or, int8, int16, int32, int64, uint8, uint16, uint32,
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_bitwise_xor.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -24,4 +24,4 @@ DEFINE_BINARY8(bitwise_xor, int8, int16, int32, int64, uint8, uint16, uint32,
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_ceil.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_UNARY3(ceil, Eigen::half, float, double);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_complex.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_BINARY2(make_complex, float, double);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_conj.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -24,4 +24,4 @@ DEFINE_UNARY1(conj, complex128);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_cos.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_UNARY3(cos, Eigen::half, float, double);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_cosh.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_UNARY2(cosh, float, double);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_digamma.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_UNARY3(digamma, Eigen::half, float, double);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_equal_to.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -25,4 +25,4 @@ DEFINE_APPROXIMATE_EQUAL2(float, double);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_erf.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_UNARY3(erf, Eigen::half, float, double);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_erfc.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_UNARY3(erfc, Eigen::half, float, double);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_exp.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_UNARY5(exp, Eigen::half, float, double, complex64, complex128);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_expm1.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_UNARY3(expm1, Eigen::half, float, double);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_floor.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_UNARY3(floor, Eigen::half, float, double);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_floor_div.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -24,4 +24,4 @@ DEFINE_BINARY3(floor_div_real, Eigen::half, float, double);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_floor_mod.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ namespace functor {
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_greater.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_BINARY7(greater, Eigen::half, float, double, int64, uint8, int8, int16);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_greater_equal.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -24,4 +24,4 @@ DEFINE_BINARY7(greater_equal, Eigen::half, float, double, int64, uint8, int8,
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_igammas_double.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 #include "tensorflow/core/kernels/cwise_ops_gpu_gradients.cu.h"
@@ -26,4 +26,4 @@ DEFINE_BINARY1(igammac, double);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_igammas_float.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 #include "tensorflow/core/kernels/cwise_ops_gpu_gradients.cu.h"
@@ -26,4 +26,4 @@ DEFINE_BINARY1(igammac, float);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_imag.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_UNARY2(get_imag, complex64, complex128);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_inverse.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 #include "tensorflow/core/kernels/cwise_ops_gpu_gradients.cu.h"
@@ -25,4 +25,4 @@ DEFINE_SIMPLE_BINARY3(inverse_grad, Eigen::half, float, double);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_invert.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_UNARY8(invert, int8, int16, int32, int64, uint8, uint16, uint32, uint64);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_isfinite.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_UNARY3(isfinite, Eigen::half, float, double);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_isinf.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_UNARY3(isinf, Eigen::half, float, double);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_isnan.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_UNARY3(isnan, Eigen::half, float, double);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_left_shift.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -24,4 +24,4 @@ DEFINE_BINARY8(left_shift, int8, int16, int32, int64, uint8, uint16, uint32,
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_less.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_BINARY7(less, Eigen::half, float, double, int64, uint8, int8, int16);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_less_equal.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -24,4 +24,4 @@ DEFINE_BINARY7(less_equal, Eigen::half, float, double, int64, uint8, int8,
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_lgamma.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_UNARY3(lgamma, Eigen::half, float, double);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_log.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_UNARY3(log, Eigen::half, float, double);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_log1p.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_UNARY3(log1p, Eigen::half, float, double);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_logical_and.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -27,4 +27,4 @@ template struct BinaryFunctor<GPUDevice, logical_and, 5>;
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_logical_not.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ template struct UnaryFunctor<GPUDevice, logical_not>;
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_logical_or.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -27,4 +27,4 @@ template struct BinaryFunctor<GPUDevice, logical_or, 5>;
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_maximum.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_BINARY4(maximum, Eigen::half, float, double, int64);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_minimum.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_BINARY4(minimum, Eigen::half, float, double, int64);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_mod.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ namespace functor {
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_mul.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -25,4 +25,4 @@ DEFINE_BINARY5(mul_no_nan, Eigen::half, float, double, complex64, complex128);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_neg.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -24,4 +24,4 @@ DEFINE_UNARY7(neg, Eigen::half, float, double, int32, int64, complex64,
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_not_equal_to.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -24,4 +24,4 @@ DEFINE_BINARY10(not_equal_to, float, Eigen::half, double, uint8, int8, int16,
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_pow.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_BINARY4(pow, Eigen::half, float, double, int64);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_random_grad.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_BINARY2(random_gamma_grad, float, double);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_real.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_UNARY2(get_real, complex64, complex128);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_right_shift.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -24,4 +24,4 @@ DEFINE_BINARY8(right_shift, int8, int16, int32, int64, uint8, uint16, uint32,
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_rint.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_UNARY2(rint, float, double);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_round.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_UNARY5(round, Eigen::half, float, double, int32, int64);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_rsqrt.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 #include "tensorflow/core/kernels/cwise_ops_gpu_gradients.cu.h"
@@ -25,4 +25,4 @@ DEFINE_SIMPLE_BINARY3(rsqrt_grad, Eigen::half, float, double);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_select.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #define EIGEN_USE_GPU
 
@@ -129,4 +129,4 @@ SELECT_FUNCTOR(complex128);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_sigmoid.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 #include "tensorflow/core/kernels/cwise_ops_gpu_gradients.cu.h"
@@ -25,4 +25,4 @@ DEFINE_SIMPLE_BINARY3(sigmoid_grad, Eigen::half, float, double);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_sign.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_UNARY6(sign, Eigen::half, float, double, int64, complex64, complex128);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_sin.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_UNARY3(sin, Eigen::half, float, double);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_sinh.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_UNARY2(sinh, float, double);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_sqrt.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 #include "tensorflow/core/kernels/cwise_ops_gpu_gradients.cu.h"
@@ -25,4 +25,4 @@ DEFINE_SIMPLE_BINARY3(sqrt_grad, Eigen::half, float, double);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_square.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_UNARY4(square, Eigen::half, float, double, int64);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_squared_difference.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_BINARY4(squared_difference, float, Eigen::half, double, int64);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_sub.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_BINARY6(sub, Eigen::half, float, double, int64, complex64, complex128);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_tan.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 
@@ -23,4 +23,4 @@ DEFINE_UNARY3(tan, Eigen::half, float, double);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM

tensorflow/core/kernels/cwise_op_gpu_tanh.cu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#if GOOGLE_CUDA
+#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
 
 #include "tensorflow/core/kernels/cwise_ops_gpu_common.cu.h"
 #include "tensorflow/core/kernels/cwise_ops_gpu_gradients.cu.h"
@@ -25,4 +25,4 @@ DEFINE_SIMPLE_BINARY3(tanh_grad, Eigen::half, float, double);
 }  // namespace functor
 }  // namespace tensorflow
 
-#endif  // GOOGLE_CUDA
+#endif  // GOOGLE_CUDA || TENSORFLOW_USE_ROCM