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

Use the correct device ordinal to check whether the device the executable was

Browse Source 
built for is equivalent to the device the it will run on.

Before this patch, if the device to run on was provided via a stream without
setting the device ordinal in the ExecutableRunOptions, we would check the
default device against the device the executable was built for.

PiperOrigin-RevId: 206892902
This commit is contained in:
A. Unique TensorFlower 2018-08-01 01:02:45 -07:00 committed by TensorFlower Gardener
parent 26ba623dcc
commit abd645085b
1 changed files with 8 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
tensorflow/compiler/xla/client/local_client.cc
View File

@ -101,11 +101,14 @@ Status LocalExecutable::ValidateExecutionOptions(
} }
} }
// Verify that the device the executable was built for is equivalent to the // Verify that the device the executable was built for is equivalent
// device it will run on. // to the device it will run on.
int run_device_ordinal = run_options.device_ordinal() == -1 int run_device_ordinal = run_options.device_ordinal();
? backend_->default_device_ordinal() if (run_device_ordinal == -1) {
: run_options.device_ordinal(); run_device_ordinal = run_options.stream() != nullptr
? run_options.stream()->parent()->device_ordinal()
: backend_->default_device_ordinal();
}
TF_ASSIGN_OR_RETURN(bool devices_equivalent, TF_ASSIGN_OR_RETURN(bool devices_equivalent,
backend_->devices_equivalent( backend_->devices_equivalent(
run_device_ordinal, build_options_.device_ordinal())); run_device_ordinal, build_options_.device_ordinal()));