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
|
#include "caffe/common.hpp"
#include "caffe/syncedmem.hpp"
#include "caffe/util/math_functions.hpp"
namespace caffe {
SyncedMemory::SyncedMemory()
: cpu_ptr_(NULL), gpu_ptr_(NULL), size_(0), head_(UNINITIALIZED),
own_cpu_data_(false), cpu_malloc_use_cuda_(false), own_gpu_data_(false) {
#ifndef CPU_ONLY
#ifdef DEBUG
CUDA_CHECK(cudaGetDevice(&device_));
#endif
#endif
}
SyncedMemory::SyncedMemory(size_t size)
: cpu_ptr_(NULL), gpu_ptr_(NULL), size_(size), head_(UNINITIALIZED),
own_cpu_data_(false), cpu_malloc_use_cuda_(false), own_gpu_data_(false) {
#ifndef CPU_ONLY
#ifdef DEBUG
CUDA_CHECK(cudaGetDevice(&device_));
#endif
#endif
}
SyncedMemory::~SyncedMemory() {
check_device();
if (cpu_ptr_ && own_cpu_data_) {
CaffeFreeHost(cpu_ptr_, cpu_malloc_use_cuda_);
}
#ifndef CPU_ONLY
if (gpu_ptr_ && own_gpu_data_) {
CUDA_CHECK(cudaFree(gpu_ptr_));
}
#endif // CPU_ONLY
}
inline void SyncedMemory::to_cpu() {
check_device();
switch (head_) {
case UNINITIALIZED:
CaffeMallocHost(&cpu_ptr_, size_, &cpu_malloc_use_cuda_);
caffe_memset(size_, 0, cpu_ptr_);
head_ = HEAD_AT_CPU;
own_cpu_data_ = true;
break;
case HEAD_AT_GPU:
#ifndef CPU_ONLY
if (cpu_ptr_ == NULL) {
CaffeMallocHost(&cpu_ptr_, size_, &cpu_malloc_use_cuda_);
own_cpu_data_ = true;
}
caffe_gpu_memcpy(size_, gpu_ptr_, cpu_ptr_);
head_ = SYNCED;
#else
#ifndef USE_ACL
NO_GPU;
#endif
#endif
break;
case HEAD_AT_CPU:
case SYNCED:
break;
}
}
inline void SyncedMemory::to_gpu() {
check_device();
#ifndef CPU_ONLY
switch (head_) {
case UNINITIALIZED:
CUDA_CHECK(cudaMalloc(&gpu_ptr_, size_));
caffe_gpu_memset(size_, 0, gpu_ptr_);
head_ = HEAD_AT_GPU;
own_gpu_data_ = true;
break;
case HEAD_AT_CPU:
if (gpu_ptr_ == NULL) {
CUDA_CHECK(cudaMalloc(&gpu_ptr_, size_));
own_gpu_data_ = true;
}
caffe_gpu_memcpy(size_, cpu_ptr_, gpu_ptr_);
head_ = SYNCED;
break;
case HEAD_AT_GPU:
case SYNCED:
break;
}
#else
NO_GPU;
#endif
}
const void* SyncedMemory::cpu_data() {
check_device();
to_cpu();
return (const void*)cpu_ptr_;
}
void SyncedMemory::set_cpu_data(void* data) {
check_device();
CHECK(data);
if (own_cpu_data_) {
CaffeFreeHost(cpu_ptr_, cpu_malloc_use_cuda_);
}
cpu_ptr_ = data;
head_ = HEAD_AT_CPU;
own_cpu_data_ = false;
}
const void* SyncedMemory::gpu_data() {
check_device();
#ifndef CPU_ONLY
to_gpu();
return (const void*)gpu_ptr_;
#else
#ifdef USE_ACL
to_cpu();
return (const void*)cpu_ptr_;
#else
NO_GPU;
return NULL;
#endif
#endif
}
void SyncedMemory::set_gpu_data(void* data) {
check_device();
#ifndef CPU_ONLY
CHECK(data);
if (own_gpu_data_) {
CUDA_CHECK(cudaFree(gpu_ptr_));
}
gpu_ptr_ = data;
head_ = HEAD_AT_GPU;
own_gpu_data_ = false;
#else
#ifdef USE_ACL
gpu_ptr_ = data;
head_ = HEAD_AT_GPU;
own_gpu_data_ = false;
#else
NO_GPU;
#endif
#endif
}
void* SyncedMemory::mutable_cpu_data() {
check_device();
to_cpu();
head_ = HEAD_AT_CPU;
return cpu_ptr_;
}
void* SyncedMemory::mutable_gpu_data() {
check_device();
#ifndef CPU_ONLY
to_gpu();
head_ = HEAD_AT_GPU;
return gpu_ptr_;
#else
#ifdef USE_ACL
to_cpu();
head_ = HEAD_AT_GPU;
return cpu_ptr_;
#else
NO_GPU;
return NULL;
#endif
#endif
}
#ifndef CPU_ONLY
void SyncedMemory::async_gpu_push(const cudaStream_t& stream) {
check_device();
CHECK(head_ == HEAD_AT_CPU);
if (gpu_ptr_ == NULL) {
CUDA_CHECK(cudaMalloc(&gpu_ptr_, size_));
own_gpu_data_ = true;
}
const cudaMemcpyKind put = cudaMemcpyHostToDevice;
CUDA_CHECK(cudaMemcpyAsync(gpu_ptr_, cpu_ptr_, size_, put, stream));
// Assume caller will synchronize on the stream before use
head_ = SYNCED;
}
#endif
void SyncedMemory::check_device() {
#ifndef CPU_ONLY
#ifdef DEBUG
int device;
cudaGetDevice(&device);
CHECK(device == device_);
if (gpu_ptr_ && own_gpu_data_) {
cudaPointerAttributes attributes;
CUDA_CHECK(cudaPointerGetAttributes(&attributes, gpu_ptr_));
CHECK(attributes.device == device_);
}
#endif
#endif
}
} // namespace caffe
|
