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
|
/* Copyright 2015 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.
==============================================================================*/
#include "tensorflow/core/framework/tracking_allocator.h"
#include "tensorflow/core/platform/env.h"
#include "tensorflow/core/platform/logging.h"
namespace tensorflow {
TrackingAllocator::TrackingAllocator(Allocator* allocator, bool track_sizes)
: allocator_(allocator),
ref_(1),
allocated_(0),
high_watermark_(0),
total_bytes_(0),
track_sizes_locally_(track_sizes && !allocator_->TracksAllocationSizes()),
next_allocation_id_(0) {}
void* TrackingAllocator::AllocateRaw(
size_t alignment, size_t num_bytes,
const AllocationAttributes& allocation_attr) {
void* ptr = allocator_->AllocateRaw(alignment, num_bytes, allocation_attr);
// If memory is exhausted AllocateRaw returns nullptr, and we should
// pass this through to the caller
if (nullptr == ptr) {
return ptr;
}
if (allocator_->TracksAllocationSizes()) {
size_t allocated_bytes = allocator_->AllocatedSize(ptr);
{
mutex_lock lock(mu_);
allocated_ += allocated_bytes;
high_watermark_ = std::max(high_watermark_, allocated_);
total_bytes_ += allocated_bytes;
allocations_.emplace_back(allocated_bytes, Env::Default()->NowMicros());
++ref_;
}
} else if (track_sizes_locally_) {
// Call the underlying allocator to try to get the allocated size
// whenever possible, even when it might be slow. If this fails,
// use the requested size as an approximation.
size_t allocated_bytes = allocator_->AllocatedSizeSlow(ptr);
allocated_bytes = std::max(num_bytes, allocated_bytes);
mutex_lock lock(mu_);
next_allocation_id_ += 1;
Chunk chunk = {num_bytes, allocated_bytes, next_allocation_id_};
in_use_.emplace(std::make_pair(ptr, chunk));
allocated_ += allocated_bytes;
high_watermark_ = std::max(high_watermark_, allocated_);
total_bytes_ += allocated_bytes;
allocations_.emplace_back(allocated_bytes, Env::Default()->NowMicros());
++ref_;
} else {
mutex_lock lock(mu_);
total_bytes_ += num_bytes;
allocations_.emplace_back(num_bytes, Env::Default()->NowMicros());
++ref_;
}
return ptr;
}
void TrackingAllocator::DeallocateRaw(void* ptr) {
// freeing a null ptr is a no-op
if (nullptr == ptr) {
return;
}
bool should_delete;
// fetch the following outside the lock in case the call to
// AllocatedSize is slow
bool tracks_allocation_sizes = allocator_->TracksAllocationSizes();
size_t allocated_bytes = 0;
if (tracks_allocation_sizes) {
allocated_bytes = allocator_->AllocatedSize(ptr);
} else if (track_sizes_locally_) {
mutex_lock lock(mu_);
auto itr = in_use_.find(ptr);
if (itr != in_use_.end()) {
tracks_allocation_sizes = true;
allocated_bytes = (*itr).second.allocated_size;
in_use_.erase(itr);
}
}
Allocator* allocator = allocator_;
{
mutex_lock lock(mu_);
if (tracks_allocation_sizes) {
CHECK_GE(allocated_, allocated_bytes);
allocated_ -= allocated_bytes;
allocations_.emplace_back(-allocated_bytes, Env::Default()->NowMicros());
}
should_delete = UnRef();
}
allocator->DeallocateRaw(ptr);
if (should_delete) {
delete this;
}
}
bool TrackingAllocator::TracksAllocationSizes() {
return track_sizes_locally_ || allocator_->TracksAllocationSizes();
}
size_t TrackingAllocator::RequestedSize(const void* ptr) {
if (track_sizes_locally_) {
mutex_lock lock(mu_);
auto it = in_use_.find(ptr);
if (it != in_use_.end()) {
return (*it).second.requested_size;
}
return 0;
} else {
return allocator_->RequestedSize(ptr);
}
}
size_t TrackingAllocator::AllocatedSize(const void* ptr) {
if (track_sizes_locally_) {
mutex_lock lock(mu_);
auto it = in_use_.find(ptr);
if (it != in_use_.end()) {
return (*it).second.allocated_size;
}
return 0;
} else {
return allocator_->AllocatedSize(ptr);
}
}
int64 TrackingAllocator::AllocationId(const void* ptr) {
if (track_sizes_locally_) {
mutex_lock lock(mu_);
auto it = in_use_.find(ptr);
if (it != in_use_.end()) {
return (*it).second.allocation_id;
}
return 0;
} else {
return allocator_->AllocationId(ptr);
}
}
void TrackingAllocator::GetStats(AllocatorStats* stats) {
allocator_->GetStats(stats);
}
void TrackingAllocator::ClearStats() { allocator_->ClearStats(); }
std::tuple<size_t, size_t, size_t> TrackingAllocator::GetSizes() {
size_t high_watermark;
size_t total_bytes;
size_t still_live_bytes;
{
mutex_lock lock(mu_);
high_watermark = high_watermark_;
total_bytes = total_bytes_;
still_live_bytes = allocated_;
}
return std::make_tuple(total_bytes, high_watermark, still_live_bytes);
}
gtl::InlinedVector<AllocRecord, 4> TrackingAllocator::GetRecordsAndUnRef() {
bool should_delete;
gtl::InlinedVector<AllocRecord, 4> allocations;
{
mutex_lock lock(mu_);
allocations.swap(allocations_);
should_delete = UnRef();
}
if (should_delete) {
delete this;
}
return allocations;
}
gtl::InlinedVector<AllocRecord, 4> TrackingAllocator::GetCurrentRecords() {
gtl::InlinedVector<AllocRecord, 4> allocations;
{
mutex_lock lock(mu_);
for (const AllocRecord& alloc : allocations_) {
allocations.push_back(alloc);
}
}
return allocations;
}
bool TrackingAllocator::UnRef() {
CHECK_GE(ref_, 1);
--ref_;
return (ref_ == 0);
}
} // end namespace tensorflow
|
