diff options
| author |  Eli Bendersky <eliben@google.com> | 2017-06-19 16:32:34 -0700 |
|---|---|---|
| committer |  TensorFlower Gardener <gardener@tensorflow.org> | 2017-06-19 16:36:16 -0700 |
| commit | ac47dc166f290d631c156846039ac78f30f362af (patch) | |
| tree | eea90e685f428759400b05533322f0b698c5ac3a /tensorflow/compiler/xla/BUILD | |
| parent | 76ee2cf584269c782961a7d835e9febd15522188 (diff) | |
[XLA] Properly propagate backend optimization options to XLA in more places.
With the flag changes, backend optimization options become critical because
without optimization the CPU code runs very slow and GPU code won't compile
PTX->AS properly in many cases.
Therefore in this CL we add a more helpful fatal CHECK in the GPU backend,
along with a convenient way to get default execution options from flags, which
clients of XLA should use.
PiperOrigin-RevId: 159496597
Diffstat (limited to 'tensorflow/compiler/xla/BUILD')
| -rw-r--r-- | tensorflow/compiler/xla/BUILD | 15 |
1 files changed, 15 insertions, 0 deletions
diff --git a/tensorflow/compiler/xla/BUILD b/tensorflow/compiler/xla/BUILD index cde790c0ae..c508071f8c 100644 --- a/tensorflow/compiler/xla/BUILD +++ b/tensorflow/compiler/xla/BUILD @@ -47,6 +47,21 @@ xla_proto_library( ) cc_library( + name = "execution_options_util", + srcs = [ + "execution_options_util.cc", + ], + hdrs = [ + "execution_options_util.h", + ], + visibility = [":friends"], + deps = [ + ":xla_proto", + "//tensorflow/compiler/xla/legacy_flags:debug_options_flags", + ], +) + +cc_library( name = "test", testonly = 1, hdrs = ["test.h"], |