forked from pytorch/pytorch
-
Notifications
You must be signed in to change notification settings - Fork 0
/
inline_container.cc
374 lines (330 loc) · 11.2 KB
/
inline_container.cc
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
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
#include <cstdio>
#include <cstring>
#include <cerrno>
#include <istream>
#include <ostream>
#include <fstream>
#include <c10/core/Allocator.h>
#include <c10/core/Backend.h>
#include "caffe2/core/common.h"
#include "caffe2/core/logging.h"
#include "caffe2/serialize/file_adapter.h"
#include "caffe2/serialize/inline_container.h"
#include "caffe2/serialize/istream_adapter.h"
#include "caffe2/serialize/read_adapter_interface.h"
#include "miniz.h"
namespace caffe2 {
namespace serialize {
size_t istream_read_func(void *pOpaque, mz_uint64 file_ofs, void *pBuf, size_t n) {
auto self = static_cast<PyTorchStreamReader*>(pOpaque);
return self->read(file_ofs, static_cast<char*>(pBuf), n);
}
static std::string basename(const std::string& name) {
size_t start = 0;
for(size_t i = 0; i < name.size(); ++i) {
if (name[i] == '\\' || name[i] == '/') {
start = i + 1;
}
}
if (start >= name.size())
return "";
size_t end = name.size();
for(size_t i = end; i > start; --i) {
if (name[i - 1] == '.') {
end = i - 1;
break;
}
}
return name.substr(start, end - start);
}
size_t PyTorchStreamReader::read(uint64_t pos, char* buf, size_t n) {
return in_->read(pos, buf, n, "reading file");
}
PyTorchStreamReader::PyTorchStreamReader(const std::string& file_name)
: ar_(std::make_unique<mz_zip_archive>()),
in_(std::make_unique<FileAdapter>(file_name)) {
init();
}
PyTorchStreamReader::PyTorchStreamReader(std::istream* in)
: ar_(std::make_unique<mz_zip_archive>()),
in_(std::make_unique<IStreamAdapter>(in)) {
init();
}
PyTorchStreamReader::PyTorchStreamReader(
std::unique_ptr<ReadAdapterInterface> in)
: ar_(std::make_unique<mz_zip_archive>()), in_(std::move(in)) {
init();
}
void PyTorchStreamReader::init() {
AT_ASSERT(in_ != nullptr);
AT_ASSERT(ar_ != nullptr);
memset(ar_.get(), 0, sizeof(mz_zip_archive));
size_t size = in_->size();
// check for the old magic number,
constexpr size_t kMagicValueLength = 8;
if (size > kMagicValueLength) {
char buf[kMagicValueLength];
read(0, buf, kMagicValueLength);
valid("checking magic number");
AT_ASSERTM(
memcmp("PYTORCH1", buf, kMagicValueLength) != 0,
"File is an unsupported archive format from the preview release.");
}
ar_->m_pIO_opaque = this;
ar_->m_pRead = istream_read_func;
mz_zip_reader_init(ar_.get(), size, 0);
valid("reading zip archive");
// figure out the archive_name (i.e. the zip folder all the other files are in)
// all lookups to getRecord will be prefixed by this folder
int n = mz_zip_reader_get_num_files(ar_.get());
if (n == 0) {
CAFFE_THROW("archive does not contain any files");
}
size_t name_size = mz_zip_reader_get_filename(ar_.get(), 0, nullptr, 0);
valid("getting filename");
std::string buf(name_size, '\0');
mz_zip_reader_get_filename(ar_.get(), 0, &buf[0], name_size);
valid("getting filename");
auto pos = buf.find_first_of('/');
if (pos == std::string::npos) {
CAFFE_THROW("file in archive is not in a subdirectory: ", buf);
}
archive_name_ = buf.substr(0, pos);
archive_name_plus_slash_ = archive_name_ + "/";
// version check
at::DataPtr version_ptr;
size_t version_size;
std::tie(version_ptr, version_size) = getRecord("version");
std::string version(static_cast<const char*>(version_ptr.get()), version_size);
version_ = caffe2::stoull(version);
AT_ASSERTM(
version_ >= kMinSupportedFileFormatVersion,
"Attempted to read a PyTorch file with version ",
c10::to_string(version_),
", but the minimum supported version for reading is ",
c10::to_string(kMinSupportedFileFormatVersion),
". Your PyTorch script module file is too old. Please re-export it again.");
AT_ASSERTM(
version_ <= kMaxSupportedFileFormatVersion,
"Attempted to read a PyTorch file with version ",
version_,
", but the maximum supported version for reading is ",
kMaxSupportedFileFormatVersion,
". Your PyTorch installation may be too old.");
}
void PyTorchStreamReader::valid(const char* what, const char* info) {
auto err = mz_zip_get_last_error(ar_.get());
if (err != MZ_ZIP_NO_ERROR) {
CAFFE_THROW(
"PytorchStreamReader failed ",
what,
info,
": ",
mz_zip_get_error_string(err));
}
}
constexpr int MZ_ZIP_LOCAL_DIR_HEADER_SIZE = 30;
constexpr int MZ_ZIP_LDH_FILENAME_LEN_OFS = 26;
constexpr int MZ_ZIP_LDH_EXTRA_LEN_OFS = 28;
static size_t getPadding(
size_t cursor,
size_t filename_size,
size_t size,
std::string& padding_buf) {
size_t start = cursor + MZ_ZIP_LOCAL_DIR_HEADER_SIZE + filename_size +
sizeof(mz_uint16) * 2;
if (size >= MZ_UINT32_MAX || cursor >= MZ_UINT32_MAX) {
start += sizeof(mz_uint16) * 2;
if (size >= MZ_UINT32_MAX) {
start += 2 * sizeof(mz_uint64);
}
if (cursor >= MZ_UINT32_MAX) {
start += sizeof(mz_uint64);
}
}
size_t mod = start % kFieldAlignment;
size_t next_offset = (mod == 0) ? start : (start + kFieldAlignment - mod);
size_t padding_size = next_offset - start;
size_t padding_size_plus_fbxx = padding_size + 4;
if (padding_buf.size() < padding_size_plus_fbxx) {
padding_buf.append(padding_size_plus_fbxx - padding_buf.size(), 'Z');
}
// zip extra encoding (key, size_of_extra_bytes)
padding_buf[0] = 'F';
padding_buf[1] = 'B';
padding_buf[2] = (uint8_t)padding_size;
padding_buf[3] = (uint8_t)(padding_size >> 8);
return padding_size_plus_fbxx;
}
bool PyTorchStreamReader::hasRecord(const std::string& name) {
std::string ss = archive_name_plus_slash_ + name;
mz_zip_reader_locate_file(ar_.get(), ss.c_str(), nullptr, 0);
bool result = ar_->m_last_error != MZ_ZIP_FILE_NOT_FOUND;
if (!result) {
ar_->m_last_error = MZ_ZIP_NO_ERROR;
}
valid("attempting to locate file ", name.c_str());
return result;
}
std::vector<std::string> PyTorchStreamReader::getAllRecords() {
mz_uint num_files = mz_zip_reader_get_num_files(ar_.get());
std::vector<std::string> out;
char buf[MZ_ZIP_MAX_ARCHIVE_FILENAME_SIZE];
for (size_t i = 0; i < num_files; i++) {
mz_zip_reader_get_filename(ar_.get(), i, buf, MZ_ZIP_MAX_ARCHIVE_FILENAME_SIZE);
out.push_back(buf);
}
return out;
}
size_t PyTorchStreamReader::getRecordID(const std::string& name) {
std::string ss = archive_name_plus_slash_ + name;
size_t result = mz_zip_reader_locate_file(ar_.get(), ss.c_str(), nullptr, 0);
if (ar_->m_last_error == MZ_ZIP_FILE_NOT_FOUND) {
CAFFE_THROW("file not found: ", ss);
}
valid("locating file ", name.c_str());
return result;
}
// return dataptr, size
std::tuple<at::DataPtr, size_t> PyTorchStreamReader::getRecord(const std::string& name) {
size_t key = getRecordID(name);
mz_zip_archive_file_stat stat;
mz_zip_reader_file_stat(ar_.get(), key, &stat);
valid("retrieving file meta-data for ", name.c_str());
void * ptr = malloc(stat.m_uncomp_size);
mz_zip_reader_extract_to_mem(ar_.get(), key, ptr, stat.m_uncomp_size, 0);
valid("reading file ", name.c_str());
at::DataPtr retval(ptr, ptr, free, at::kCPU);
return std::make_tuple(std::move(retval), stat.m_uncomp_size);
}
static int64_t read_le_16(uint8_t* buf) {
return buf[0] + (buf[1] << 8);
}
size_t PyTorchStreamReader::getRecordOffset(const std::string& name) {
mz_zip_archive_file_stat stat;
mz_zip_reader_file_stat(ar_.get(), getRecordID(name), &stat);
valid("retrieving file meta-data for ", name.c_str());
uint8_t local_header[MZ_ZIP_LOCAL_DIR_HEADER_SIZE];
in_->read(
stat.m_local_header_ofs,
local_header,
MZ_ZIP_LOCAL_DIR_HEADER_SIZE,
"reading file header");
size_t filename_len = read_le_16(local_header + MZ_ZIP_LDH_FILENAME_LEN_OFS);
size_t extra_len = read_le_16(local_header + MZ_ZIP_LDH_EXTRA_LEN_OFS);
return stat.m_local_header_ofs + MZ_ZIP_LOCAL_DIR_HEADER_SIZE + filename_len + extra_len;
}
PyTorchStreamReader::~PyTorchStreamReader() {
mz_zip_clear_last_error(ar_.get());
mz_zip_reader_end(ar_.get());
valid("closing reader for archive ", archive_name_.c_str());
}
size_t ostream_write_func(
void* pOpaque,
mz_uint64 file_ofs,
const void* pBuf,
size_t n) {
auto self = static_cast<PyTorchStreamWriter*>(pOpaque);
if (self->current_pos_ != file_ofs) {
CAFFE_THROW("unexpected pos ", self->current_pos_, " vs ", file_ofs);
}
size_t ret = self->writer_func_(pBuf, n);
if (n != ret) {
self->err_seen_ = true;
}
self->current_pos_ += ret;
return ret;
}
PyTorchStreamWriter::PyTorchStreamWriter(std::string file_name)
: archive_name_(basename(file_name)) {
setup(file_name);
}
PyTorchStreamWriter::PyTorchStreamWriter(
const std::function<size_t(const void*, size_t)>& writer_func)
: archive_name_("archive"), writer_func_(writer_func) {
setup(archive_name_);
}
void PyTorchStreamWriter::setup(const string& file_name) {
ar_ = std::make_unique<mz_zip_archive>();
memset(ar_.get(), 0, sizeof(mz_zip_archive));
archive_name_plus_slash_ = archive_name_ + "/"; // for writeRecord().
if (archive_name_.size() == 0) {
CAFFE_THROW("invalid file name: ", file_name);
}
if (!writer_func_) {
file_stream_.open(
file_name,
std::ofstream::out | std::ofstream::trunc | std::ofstream::binary);
valid("opening archive ", file_name.c_str());
writer_func_ = [this](const void* buf, size_t nbytes) -> size_t {
file_stream_.write(static_cast<const char*>(buf), nbytes);
return !file_stream_ ? 0 : nbytes;
};
}
ar_->m_pIO_opaque = this;
ar_->m_pWrite = ostream_write_func;
mz_zip_writer_init_v2(ar_.get(), 0, MZ_ZIP_FLAG_WRITE_ZIP64);
valid("initializing archive ", file_name.c_str());
std::string version = c10::to_string(kProducedFileFormatVersion);
version.push_back('\n');
writeRecord("version", version.c_str(), version.size());
}
void PyTorchStreamWriter::writeRecord(
const std::string& name,
const void* data,
size_t size,
bool compress) {
AT_ASSERT(!finalized_);
AT_ASSERT(!archive_name_plus_slash_.empty());
std::string full_name = archive_name_plus_slash_ + name;
size_t padding_size =
getPadding(ar_->m_archive_size, full_name.size(), size, padding_);
uint32_t flags = compress ? MZ_BEST_COMPRESSION : 0;
mz_zip_writer_add_mem_ex_v2(
ar_.get(),
full_name.c_str(),
data,
size,
nullptr,
0,
flags,
0,
0,
nullptr,
padding_.c_str(),
padding_size,
nullptr,
0);
valid("writing file ", name.c_str());
}
void PyTorchStreamWriter::writeEndOfFile() {
AT_ASSERT(!finalized_);
finalized_ = true;
mz_zip_writer_finalize_archive(ar_.get());
mz_zip_writer_end(ar_.get());
valid("writing central directory for archive ", archive_name_.c_str());
if (file_stream_.is_open()) {
file_stream_.close();
}
}
void PyTorchStreamWriter::valid(const char* what, const char* info) {
auto err = mz_zip_get_last_error(ar_.get());
if (err != MZ_ZIP_NO_ERROR) {
CAFFE_THROW(
"PytorchStreamWriter failed ",
what,
info,
": ",
mz_zip_get_error_string(err));
}
if (err_seen_) {
CAFFE_THROW("PytorchStreamWriter failed ", what, info, ".");
}
}
PyTorchStreamWriter::~PyTorchStreamWriter() {
if (!finalized_) {
writeEndOfFile();
}
}
} // namespace serialize
} // namespace caffe2