xla::Service Class Reference

#include <service.h>

Inheritance diagram for xla::Service:

Collaboration diagram for xla::Service:

Public Member Functions
tensorflow::Status	Computation (const ComputationRequest *arg, ComputationResponse *result) override
	Populates computation_ with a valid object. Used before any given operation is enqueued. More...
tensorflow::Status	Op (const OpRequest *arg, OpResponse *result) override
	Enqueue an Op on the computation. More...

Detailed Description

Google Docs:

The XLA service object, which is the same across all platforms. It maintains the service state of computations and allocations, and delegates target-specific requests to the target-specific infrastructure (target-specific compiler, StreamExecutor).

Member Function Documentation

Computation()

tensorflow::Status xla::Service::Computation ( const ComputationRequest *  arg, ComputationResponse *  result  )

override

Populates computation_ with a valid object. Used before any given operation is enqueued.

1. Check existence of the name of given request
2. Use member computation_tracker_'s method NewComputation() to create a new computation and store the ComputationHandle in the given ComputationResponse object result.

Here is the call graph for this function:

Op()

tensorflow::Status xla::Service::Op ( const OpRequest *  arg, OpResponse *  result  )

override

Enqueue an Op on the computation.

1. Get the UserComputation object.
2. Based on different cases of operation, call the right method of UserComputation object.
3. Set result and add some common setting (metadata, sharding) in computation.

Here is the call graph for this function:

---

The documentation for this class was generated from the following files:

• tensorflow/tensorflow/compiler/xla/service/service.h
• tensorflow/tensorflow/compiler/xla/service/service.cc