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

Merge pull request #39231 from yongtang:mlir

Browse Source 
PiperOrigin-RevId: 310466095
Change-Id: I86255cdcb4e8ffe0694dd68516ca582c31a4de42
This commit is contained in:
TensorFlower Gardener 2020-05-07 17:20:13 -07:00
commit bd17cdd0d7
1 changed files with 16 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
tensorflow/compiler/mlir/python/BUILD
View File

@ -12,6 +12,22 @@ cc_library(
"//tensorflow/c:tf_status_helper",
"//tensorflow/compiler/mlir/tensorflow:convert_graphdef",
"//tensorflow/compiler/mlir/tensorflow:error_util",
# (yongtang) The graph_optimization_pass_registration needs to be part
# of a shared object that will be loaded whenever `import tensorflow`
# is run. The natural place is libtensorflow_framework.so.
# While adding graph_optimization_pass_registration to
# libtensorflow_framework.so is possible with some modification in
# dependency, many tests will fail due to multiple copies of LLVM.
# See https://github.com/tensorflow/tensorflow/pull/39231 for details.
# Alternatively, we place graph_optimization_pass_registration here
# because:
# - tensorflow/python/_pywrap_mlir.so already depends on LLVM anyway
# - tensorflow/python/_pywrap_mlir.so always loaded as part of python
# binding
# TODO: It might be still preferrable to place graph_optimization_pass
# as part of the libtensorflow_framework.so, as it is the central
# place for core related components.
"//tensorflow/compiler/mlir/tensorflow:graph_optimization_pass_registration",
"//tensorflow/compiler/mlir/tensorflow:import_utils",
"@llvm-project//llvm:support",
"@llvm-project//mlir:IR",