diff options
| author |  Benjamin Kramer <kramerb@google.com> | 2018-07-25 11:08:39 -0700 |
|---|---|---|
| committer |  TensorFlower Gardener <gardener@tensorflow.org> | 2018-07-25 11:11:42 -0700 |
| commit | a5285d999af961367437d72285f67c4a5a2878d4 (patch) | |
| tree | 9a5710a0f171b4aa486fcabc2e65a9ecc4238077 /tensorflow/compiler/xla/service/cpu/BUILD | |
| parent | 0bc512505957e3685305b6a850f222c6eed88c7d (diff) | |
[XLA:GPU] Use a fast approximation for tanh
Just reuse the CPU implementation, which in turn is derived from Eigen. It
claims to be accurate within +-1% which is good enough for fast math.
Refactor the CPU implementation into a common file and remove the
VectorSupportLibrary dependency (it's not needed).
PiperOrigin-RevId: 206022260
Diffstat (limited to 'tensorflow/compiler/xla/service/cpu/BUILD')
| -rw-r--r-- | tensorflow/compiler/xla/service/cpu/BUILD | 1 |
1 files changed, 1 insertions, 0 deletions
diff --git a/tensorflow/compiler/xla/service/cpu/BUILD b/tensorflow/compiler/xla/service/cpu/BUILD index ace9f96cfb..71f7f985d0 100644 --- a/tensorflow/compiler/xla/service/cpu/BUILD +++ b/tensorflow/compiler/xla/service/cpu/BUILD @@ -444,6 +444,7 @@ cc_library( deps = [ ":vector_support_library", "//tensorflow/compiler/xla/service/llvm_ir:llvm_util", + "//tensorflow/compiler/xla/service/llvm_ir:math_ops", "//tensorflow/core:lib", "@llvm//:core", "@llvm//:transform_utils", |