diff options
| author |  Justin Lebar <jlebar@google.com> | 2018-08-16 15:33:49 -0700 |
|---|---|---|
| committer |  TensorFlower Gardener <gardener@tensorflow.org> | 2018-08-16 15:38:17 -0700 |
| commit | 4a5a37c350cd5c036d97b7aaf0ad3734443ff764 (patch) | |
| tree | 2ce0d6b105886f734103ab5982f93fe68357aa52 /tensorflow/compiler/xla/service/cpu | |
| parent | e45de7cbe36cd5f72a9abb5c828bde9dac700bd8 (diff) | |
Run ZeroSizedHloElimination in the right place in the CPU pass pipeline.
Fix apparent typo that was causing ZeroSizedHloElimination to be run
much later than expected.
PiperOrigin-RevId: 209061101
Diffstat (limited to 'tensorflow/compiler/xla/service/cpu')
| -rw-r--r-- | tensorflow/compiler/xla/service/cpu/cpu_compiler.cc | 2 |
1 files changed, 1 insertions, 1 deletions
diff --git a/tensorflow/compiler/xla/service/cpu/cpu_compiler.cc b/tensorflow/compiler/xla/service/cpu/cpu_compiler.cc index c7e766ca6b..fde8fbd486 100644 --- a/tensorflow/compiler/xla/service/cpu/cpu_compiler.cc +++ b/tensorflow/compiler/xla/service/cpu/cpu_compiler.cc @@ -278,7 +278,7 @@ Status CpuCompiler::RunHloPasses(HloModule* module, bool is_aot_compile, // BatchNormExpander can create zero-sized ops, so zero-sized HLO // elimination has to come after that pass. - pipeline.AddPass<ZeroSizedHloElimination>(); + pass.AddPass<ZeroSizedHloElimination>(); pass.AddPass<WhileLoopInvariantCodeMotion>(); pass.AddPass<TupleSimplifier>(); |