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_COPY_THUNK_H_
17 #define TENSORFLOW_COMPILER_XLA_SERVICE_GPU_COPY_THUNK_H_
18 
19 #include "tensorflow/compiler/xla/service/buffer_assignment.h"
20 #include "tensorflow/compiler/xla/service/gpu/buffer_allocations.h"
21 #include "tensorflow/compiler/xla/service/gpu/hlo_execution_profiler.h"
22 #include "tensorflow/compiler/xla/service/gpu/thunk.h"
23 #include "tensorflow/compiler/xla/service/hlo_instruction.h"
24 #include "tensorflow/core/platform/stream_executor_no_cuda.h"
25 #include "tensorflow/core/platform/types.h"
26 
27 namespace xla {
28 namespace gpu {
29 
30 // A thunk that copies data from a host buffer to a device buffer.
31 class HostToDeviceCopyThunk : public Thunk {
32  public:
33   // Constructs a CopyThunk that copies host data from `source_address` to the
34   // device buffer `destination_buffer`. `mem_size` is the size of the data in
35   // bytes.
36   HostToDeviceCopyThunk(const void* source_address,
37                         const BufferAllocation::Slice& destination_buffer,
38                         uint64 mem_size, const HloInstruction* hlo_instruction);
39 
40   HostToDeviceCopyThunk(const HostToDeviceCopyThunk&) = delete;
41   HostToDeviceCopyThunk& operator=(const HostToDeviceCopyThunk&) = delete;
42 
43   Status ExecuteOnStream(const BufferAllocations& buffer_allocations,
44                          se::Stream* stream,
45                          HloExecutionProfiler* profiler) override;
46 
47  private:
48   const void* source_address_;
49   const BufferAllocation::Slice destination_buffer_;
50   const uint64 mem_size_;
51 };
52 
53 // A thunk that copies data from a device buffer to another device buffer.
54 class DeviceToDeviceCopyThunk : public Thunk {
55  public:
56   // Constructs a CopyThunk that copies host data from `source_buffer` to the
57   // device buffer `destination_buffer`. `mem_size` is the size of the data in
58   // bytes.
59   DeviceToDeviceCopyThunk(const BufferAllocation::Slice& source_buffer,
60                           const BufferAllocation::Slice& destination_buffer,
61                           uint64 mem_size,
62                           const HloInstruction* hlo_instruction);
63 
64   DeviceToDeviceCopyThunk(const DeviceToDeviceCopyThunk&) = delete;
65   DeviceToDeviceCopyThunk& operator=(const DeviceToDeviceCopyThunk&) = delete;
66 
67   Status ExecuteOnStream(const BufferAllocations& buffer_allocations,
68                          se::Stream* stream,
69                          HloExecutionProfiler* profiler) override;
70 
71  private:
72   const BufferAllocation::Slice source_buffer_;
73   const BufferAllocation::Slice destination_buffer_;
74   const uint64 mem_size_;
75 };
76 
77 }  // namespace gpu
78 }  // namespace xla
79 
80 #endif  // TENSORFLOW_COMPILER_XLA_SERVICE_GPU_COPY_THUNK_H_
81