This is a rather crude heuristic, but enough to recover performance without causing excessive inlining when calling functions many times. Also remove an outdated comment and a TF_RET_CHECK that's doesn't hold when not inlining everything. PiperOrigin-RevId: 291180120 Change-Id: I50434076891b69f92ded0cdbd40039a4f5858541
		
			
				
	
	
		
			53 lines
		
	
	
		
			1.8 KiB
		
	
	
	
		
			C++
		
	
	
	
	
	
			
		
		
	
	
			53 lines
		
	
	
		
			1.8 KiB
		
	
	
	
		
			C++
		
	
	
	
	
	
/* Copyright 2017 The TensorFlow Authors. All Rights Reserved.
 | 
						|
 | 
						|
Licensed under the Apache License, Version 2.0 (the "License");
 | 
						|
you may not use this file except in compliance with the License.
 | 
						|
You may obtain a copy of the License at
 | 
						|
 | 
						|
    http://www.apache.org/licenses/LICENSE-2.0
 | 
						|
 | 
						|
Unless required by applicable law or agreed to in writing, software
 | 
						|
distributed under the License is distributed on an "AS IS" BASIS,
 | 
						|
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 | 
						|
See the License for the specific language governing permissions and
 | 
						|
limitations under the License.
 | 
						|
==============================================================================*/
 | 
						|
 | 
						|
#ifndef TENSORFLOW_COMPILER_XLA_SERVICE_CALL_INLINER_H_
 | 
						|
#define TENSORFLOW_COMPILER_XLA_SERVICE_CALL_INLINER_H_
 | 
						|
 | 
						|
#include <deque>
 | 
						|
 | 
						|
#include "tensorflow/compiler/xla/service/hlo_pass_interface.h"
 | 
						|
#include "tensorflow/compiler/xla/statusor.h"
 | 
						|
 | 
						|
namespace xla {
 | 
						|
 | 
						|
// For every kCall operation in the main computation, we inline the body of the
 | 
						|
// called function, and proceed recursively.
 | 
						|
class CallInliner : public HloModulePass {
 | 
						|
 public:
 | 
						|
  using InlinedInstructionMap =
 | 
						|
      std::unordered_map<HloInstruction*, HloInstruction*>;
 | 
						|
 | 
						|
  // Inlines one call instruction.  Returns a mapping from the original
 | 
						|
  // instructions to their inlined versions.
 | 
						|
  static StatusOr<InlinedInstructionMap> Inline(HloInstruction* call);
 | 
						|
 | 
						|
  // If single_call_site is true, only functions with a single call site will be
 | 
						|
  // inlined.
 | 
						|
  explicit CallInliner(bool single_call_site = false)
 | 
						|
      : single_call_site_(single_call_site) {}
 | 
						|
  ~CallInliner() override = default;
 | 
						|
  absl::string_view name() const override { return "CallInliner"; }
 | 
						|
 | 
						|
  StatusOr<bool> Run(HloModule* module) override;
 | 
						|
 | 
						|
 private:
 | 
						|
  bool single_call_site_;
 | 
						|
};
 | 
						|
 | 
						|
}  // namespace xla
 | 
						|
 | 
						|
#endif  // TENSORFLOW_COMPILER_XLA_SERVICE_CALL_INLINER_H_
 |