1 /* Copyright 2017 The TensorFlow Authors. All Rights Reserved. 2 3 Licensed under the Apache License, Version 2.0 (the "License"); 4 you may not use this file except in compliance with the License. 5 You may obtain a copy of the License at 6 7 http://www.apache.org/licenses/LICENSE-2.0 8 9 Unless required by applicable law or agreed to in writing, software 10 distributed under the License is distributed on an "AS IS" BASIS, 11 WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 12 See the License for the specific language governing permissions and 13 limitations under the License. 14 ==============================================================================*/ 15 16 #ifndef TENSORFLOW_COMPILER_XLA_SERVICE_GPU_WHILE_THUNK_H_ 17 #define TENSORFLOW_COMPILER_XLA_SERVICE_GPU_WHILE_THUNK_H_ 18 19 #include <vector> 20 21 #include "tensorflow/compiler/xla/service/gpu/buffer_allocations.h" 22 #include "tensorflow/compiler/xla/service/gpu/hlo_execution_profiler.h" 23 #include "tensorflow/compiler/xla/service/gpu/sequential_thunk.h" 24 #include "tensorflow/compiler/xla/service/gpu/thunk.h" 25 #include "tensorflow/compiler/xla/service/hlo_instruction.h" 26 #include "tensorflow/core/platform/stream_executor_no_cuda.h" 27 28 namespace xla { 29 namespace gpu { 30 31 // WhileThunk implements the while instruction on GPU by invoking a thunk 32 // sequence for the while 'condition' computation, and (conditionally) another 33 // thunk sequence for the while 'body' computation. WhileThunk assumes that 34 // buffers for the following set of while-related instructions share the same 35 // allocation: 36 // init, condition.parameter, body.parameter, body.root, while.result 37 // WhileThunk synchronizes the stream to test the result of the 'condition' 38 // computation. 39 class WhileThunk : public Thunk { 40 public: 41 // Constructs a WhileThunk to compute while instruction 'hlo'. 42 WhileThunk(const BufferAllocation::Slice& condition_result_buffer_index, 43 std::unique_ptr<ThunkSequence> condition_thunk_sequence, 44 std::unique_ptr<ThunkSequence> body_thunk_sequence, 45 const HloInstruction* hlo); 46 WhileThunk(const WhileThunk&) = delete; 47 WhileThunk& operator=(const WhileThunk&) = delete; 48 49 Status Initialize(const GpuExecutable& executable, 50 se::StreamExecutor* executor) override; 51 Status ExecuteOnStream(const BufferAllocations& buffer_allocations, 52 se::Stream* stream, 53 HloExecutionProfiler* profiler) override; 54 55 private: 56 const BufferAllocation::Slice condition_result_buffer_index_; 57 std::unique_ptr<SequentialThunk> condition_thunk_sequence_; 58 std::unique_ptr<SequentialThunk> body_thunk_sequence_; 59 }; 60 61 } // namespace gpu 62 } // namespace xla 63 64 #endif // TENSORFLOW_COMPILER_XLA_SERVICE_GPU_WHILE_THUNK_H_ 65