1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
|
/* 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.
==============================================================================*/
// Declares the ExecutorExecutor class, which is a CPU-only implementation of
// the StreamExecutor interface. For now, this is used for testing and to
// examine the performance of host-based StreamExecutor code.
#ifndef TENSORFLOW_COMPILER_EXECUTOR_STREAM_EXECUTOR_EXECUTOR_EXECUTOR_H_
#define TENSORFLOW_COMPILER_EXECUTOR_STREAM_EXECUTOR_EXECUTOR_EXECUTOR_H_
#include "tensorflow/stream_executor/host/host_stream.h"
#include "tensorflow/stream_executor/host/host_timer.h"
#include "tensorflow/compiler/xla/shape_util.h"
#include "tensorflow/stream_executor/blas.h"
#include "tensorflow/stream_executor/lib/error.h"
#include "tensorflow/stream_executor/lib/status.h"
#include "tensorflow/stream_executor/lib/statusor.h"
#include "tensorflow/stream_executor/rng.h"
#include "tensorflow/stream_executor/stream_executor.h"
#include "tensorflow/stream_executor/stream_executor_internal.h"
#include <list>
#include <mutex>
namespace perftools {
namespace gputools {
namespace executorplugin {
using Args = tensorflow::gtl::ArraySlice<DeviceMemoryBase>;
class ExecutorExecutor : public internal::StreamExecutorInterface {
public:
explicit ExecutorExecutor(const PluginConfig &plugin_config);
~ExecutorExecutor() override;
port::Status Init(int device_ordinal, DeviceOptions device_options) override {
return port::Status::OK();
}
bool GetKernel(const MultiKernelLoaderSpec &spec,
KernelBase *kernel) override {
return false;
}
bool Launch(Stream *stream, const ThreadDim &thread_dims,
const BlockDim &block_dims, const KernelBase &kernel,
const KernelArgsArrayBase &args) override {
return false;
}
void *Allocate(uint64 size) override;
void *AllocateSubBuffer(DeviceMemoryBase *mem, uint64 offset_bytes,
uint64 size_bytes) override;
void Deallocate(DeviceMemoryBase *mem) override;
void *HostMemoryAllocate(uint64 size) override { return new char[size]; }
void HostMemoryDeallocate(void *mem) override {
delete[] static_cast<char *>(mem);
}
bool HostMemoryRegister(void *mem, uint64 size) override { return true; }
bool HostMemoryUnregister(void *mem) override { return true; }
bool Memcpy(Stream *stream, void *host_dst, const DeviceMemoryBase &pop_src,
uint64 size) override;
bool Memcpy(Stream *stream, DeviceMemoryBase *pop_dst, const void *host_src,
uint64 size) override;
bool MemcpyDeviceToDevice(Stream *stream, DeviceMemoryBase *pop_dst,
const DeviceMemoryBase &host_src,
uint64 size) override {
return false;
}
bool MemZero(Stream *stream, DeviceMemoryBase *location,
uint64 size) override {
return false;
}
bool Memset(Stream *stream, DeviceMemoryBase *location, uint8 pattern,
uint64 size) override {
return false;
}
bool Memset32(Stream *stream, DeviceMemoryBase *location, uint32 pattern,
uint64 size) override {
return false;
}
// No "synchronize all activity" implemented for this platform at the moment.
bool SynchronizeAllActivity() override { return false; }
bool SynchronousMemZero(DeviceMemoryBase *location, uint64 size) override {
return false;
}
bool SynchronousMemSet(DeviceMemoryBase *location, int value,
uint64 size) override {
return false;
}
port::Status SynchronousMemcpy(DeviceMemoryBase *pop_dst,
const void *host_src, uint64 size) override;
port::Status SynchronousMemcpy(void *host_dst,
const DeviceMemoryBase &pop_src,
uint64 size) override;
port::Status SynchronousMemcpyDeviceToDevice(DeviceMemoryBase *pop_dst,
const DeviceMemoryBase &pop_src,
uint64 size) override {
return port::Status{port::error::UNIMPLEMENTED, ""};
}
bool HostCallback(Stream *stream, std::function<void()> callback) override;
port::Status AllocateEvent(Event *event) override {
return port::Status{port::error::UNIMPLEMENTED, ""};
}
port::Status DeallocateEvent(Event *event) override {
return port::Status{port::error::UNIMPLEMENTED, ""};
}
port::Status RecordEvent(Stream *stream, Event *event) override {
return port::Status{port::error::UNIMPLEMENTED, ""};
}
port::Status WaitForEvent(Stream *stream, Event *event) override {
return port::Status{port::error::UNIMPLEMENTED, ""};
}
Event::Status PollForEventStatus(Event *event) override {
return Event::Status::kError;
}
bool AllocateStream(Stream *stream) override { return true; }
void DeallocateStream(Stream *stream) override {}
bool CreateStreamDependency(Stream *dependent, Stream *other) override;
bool AllocateTimer(Timer *timer) override { return true; }
void DeallocateTimer(Timer *timer) override {}
bool StartTimer(Stream *stream, Timer *timer) override;
bool StopTimer(Stream *stream, Timer *timer) override;
bool BlockHostUntilDone(Stream *stream) override;
int PlatformDeviceCount() override { return 1; }
bool DeviceMemoryUsage(int64 *free, int64 *total) const override {
return false;
}
DeviceDescription *PopulateDeviceDescription() const override;
port::Status EnablePeerAccessTo(StreamExecutorInterface *other) override {
return port::Status::OK();
}
bool CanEnablePeerAccessTo(StreamExecutorInterface *other) override {
return true;
}
SharedMemoryConfig GetDeviceSharedMemoryConfig() override {
return SharedMemoryConfig::kDefault;
}
port::Status SetDeviceSharedMemoryConfig(SharedMemoryConfig config) override {
return port::Status{port::error::UNIMPLEMENTED,
"Shared memory not supported"};
}
std::unique_ptr<internal::EventInterface> CreateEventImplementation()
override {
return nullptr;
}
std::unique_ptr<internal::KernelInterface> CreateKernelImplementation()
override {
return nullptr;
}
std::unique_ptr<internal::StreamInterface> GetStreamImplementation()
override {
return std::unique_ptr<internal::StreamInterface>(new host::HostStream());
}
std::unique_ptr<internal::TimerInterface> GetTimerImplementation() override {
return std::unique_ptr<internal::TimerInterface>(new host::HostTimer());
}
port::StatusOr<DeviceMemoryBase> ExecuteGraph(const xla::Shape &shape,
Args args);
private:
DeviceMemoryBase AllocateSingleOutput(const xla::Shape &shape);
port::StatusOr<DeviceMemoryBase> AllocateOutputBuffer(
const xla::Shape &shape);
const PluginConfig plugin_config_;
};
} // namespace executorplugin
} // namespace gputools
} // namespace perftools
#endif // TENSORFLOW_COMPILER_EXECUTOR_STREAM_EXECUTOR_EXECUTOR_EXECUTOR_H_
|
