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

Enable the generated version of ceil.

Browse Source 
PiperOrigin-RevId: 354526985
Change-Id: I918590193279ce43efae4d8a7e088e65ff3253e9
This commit is contained in:
Stephan Herhut 2021-01-29 07:15:57 -08:00 committed by TensorFlower Gardener
parent 842c30366f
commit aea0204731
3 changed files with 4 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
tensorflow/core/kernels/cwise_op_ceil.cc
View File

@ -20,8 +20,7 @@ REGISTER4(UnaryOp, CPU, "Ceil", functor::ceil, float, Eigen::half, bfloat16,
double); double);
#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM #if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
#if !defined(MLIR_GENERATED_GPU_KERNELS_ENABLED) || \ #if !defined(MLIR_GENERATED_GPU_KERNELS_ENABLED)
!defined(MLIR_GENERATED_EXPERIMENTAL_GPU_KERNELS_ENABLED)
REGISTER3(UnaryOp, GPU, "Ceil", functor::ceil, float, Eigen::half, double); REGISTER3(UnaryOp, GPU, "Ceil", functor::ceil, float, Eigen::half, double);
#endif #endif
#endif #endif

2
tensorflow/core/kernels/cwise_op_gpu_ceil.cu.cc
View File

@ -19,7 +19,9 @@ limitations under the License.
namespace tensorflow { namespace tensorflow {
namespace functor { namespace functor {
#if !defined(MLIR_GENERATED_GPU_KERNELS_ENABLED)
DEFINE_UNARY3(ceil, Eigen::half, float, double); DEFINE_UNARY3(ceil, Eigen::half, float, double);
#endif
} // namespace functor } // namespace functor
} // namespace tensorflow } // namespace tensorflow

2
tensorflow/core/kernels/mlir_generated/BUILD
View File

@ -41,6 +41,7 @@ filegroup(
"gpu_op_abs.cc", "gpu_op_abs.cc",
"gpu_op_atan.cc", "gpu_op_atan.cc",
"gpu_op_atanh.cc", "gpu_op_atanh.cc",
"gpu_op_ceil.cc",
"gpu_op_cos.cc", "gpu_op_cos.cc",
"gpu_op_cosh.cc", "gpu_op_cosh.cc",
"gpu_op_sqrt.cc", "gpu_op_sqrt.cc",
@ -58,7 +59,6 @@ filegroup(
"gpu_op_acosh.cc", "gpu_op_acosh.cc",
"gpu_op_asin.cc", "gpu_op_asin.cc",
"gpu_op_asinh.cc", "gpu_op_asinh.cc",
"gpu_op_ceil.cc",
"gpu_op_complex.cc", "gpu_op_complex.cc",
"gpu_op_complex_abs.cc", "gpu_op_complex_abs.cc",
"gpu_op_conj.cc", "gpu_op_conj.cc",