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

Remove canonicalizer and extra breakup-islands passes at the end of the TPU bridge.

Browse Source 
As the end of the TPU bridge passes is preparing the IR for exporting back to Graph, having the canonicalizer run again is not be necessary as we want to preserve tf_executor.graph with single op tf_executor.islands.

PiperOrigin-RevId: 291248163
Change-Id: I8ac71d190d94c1515a1c76c43b6d11020bc1b2c2
This commit is contained in:
Andy Ly 2020-01-23 14:42:10 -08:00 committed by TensorFlower Gardener
parent 1c1a90e2ae
commit e4d2552523
1 changed files with 0 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
tensorflow/compiler/mlir/tensorflow/transforms/bridge.cc
View File

@ -59,8 +59,6 @@ void CreateTPUBridge(OpPassManager &pm) {
pm.addNestedPass<FuncOp>(CreateBreakUpIslandsPass());
pm.addNestedPass<FuncOp>(TFDevice::CreateReplicateToIslandPass());
pm.addNestedPass<FuncOp>(CreateBreakUpIslandsPass());
pm.addNestedPass<FuncOp>(createCanonicalizerPass());
pm.addNestedPass<FuncOp>(CreateBreakUpIslandsPass());
}
tensorflow::Status TPUBridge(ModuleOp module, bool enable_logging) {