diff options
| author |  A. Unique TensorFlower <gardener@tensorflow.org> | 2018-10-09 16:53:29 -0700 |
|---|---|---|
| committer | TensorFlower Gardener <gardener@tensorflow.org> | 2018-10-09 17:06:10 -0700 |
| commit | bb5fc614a4a358b350ef8dd19cb7010760fa9b29 (patch) | |
| tree | 43a745ffdc409d0ff4660342d6a62735ac366a13 /tensorflow/compiler/xla/service/service.cc | |
| parent | 65b7d0b2f84c334327a295bf41bc06c7f6b8ffe5 (diff) | |
[XLA] Cleanup: Make AllocationTracker::Resolve const.
So that when resolving some global data, we don't have to worry whether
"Resolve" is going to mutate the real data.
PiperOrigin-RevId: 216448145
Diffstat (limited to 'tensorflow/compiler/xla/service/service.cc')
| -rw-r--r-- | tensorflow/compiler/xla/service/service.cc | 4 |
1 files changed, 2 insertions, 2 deletions
diff --git a/tensorflow/compiler/xla/service/service.cc b/tensorflow/compiler/xla/service/service.cc index b27a92f2a0..084df17951 100644 --- a/tensorflow/compiler/xla/service/service.cc +++ b/tensorflow/compiler/xla/service/service.cc @@ -207,7 +207,7 @@ Status Service::ValidateResultShape(const Shape& client_shape, StatusOr<std::vector<std::vector<const ShapedBuffer*>>> Service::ResolveAndValidateArguments( absl::Span<const GlobalDataHandle* const> arguments, - absl::Span<se::StreamExecutor* const> stream_executors) { + absl::Span<se::StreamExecutor* const> stream_executors) const { CHECK_EQ(options_.number_of_replicas(), stream_executors.size()); std::vector<std::vector<const ShapedBuffer*>> replicated_arguments; replicated_arguments.resize(options_.number_of_replicas()); @@ -590,7 +590,7 @@ StatusOr<std::vector<se::StreamExecutor*>> Service::GetExecutors( StatusOr<std::vector<std::vector<const ShapedBuffer*>>> Service::GetArguments( const ExecutionOptions& execution_options, - absl::Span<const GlobalDataHandle* const> arguments) { + absl::Span<const GlobalDataHandle* const> arguments) const { // Resolve the allocations for the arguments of the computation, and create // a vector of device memory offsets for the arguments from the allocations. // In the case of partitioned computations, assume all arguments go on the |