-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathbasic.cpp
More file actions
360 lines (339 loc) · 16.5 KB
/
Copy pathbasic.cpp
File metadata and controls
360 lines (339 loc) · 16.5 KB
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
/******************************************************************************
* Copyright (c) 2025 AISS Group at Harbin Institute of Technology. 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 "asnumpy/array/basic.hpp"
#include <aclnnop/aclnn_fill_scalar.h>
#include <aclnnop/aclnn_ones.h>
#include <aclnnop/aclnn_zero.h>
#include <aclnnop/aclnn_eye.h>
#include <aclnnop/aclnn_arange.h>
#include <fmt/core.h>
#include <fmt/format.h>
NPUArray Empty(const std::vector<int64_t>& shape, py::dtype dtype) {
try {
return NPUArray(shape, dtype);
} catch (const std::exception& e) {
throw std::runtime_error(fmt::format("[creation.cpp](empty) NPUArray construction error = {}", e.what()));
}
}
NPUArray EmptyLike(const NPUArray& prototype, py::dtype dtype) {
try {
// 若未指定dtype，使用原型数组的dtype
py::dtype target_dtype = dtype.is_none() ? prototype.dtype() : dtype;
// 基于原型的形状和目标dtype创建空数组
return NPUArray(prototype.shape, target_dtype);
} catch (const std::exception& e) {
throw std::runtime_error(fmt::format("[creation.cpp](empty_like) NPUArray construction error = {}", e.what()));
}
}
NPUArray Zeros(const std::vector<int64_t>& shape, py::dtype dtype) {
auto array = NPUArray(shape, dtype);
uint64_t workspaceSize = 0;
aclOpExecutor *executor;
auto error = aclnnInplaceZeroGetWorkspaceSize(array.tensorPtr, &workspaceSize, &executor);
if(error != ACL_SUCCESS) throw std::runtime_error(fmt::format("[creation.cpp](zeros) aclnnInplaceZeroGetWorkspaceSize error = {}",error));
// 检查workspaceSize是否有效
if(workspaceSize < 0) throw std::runtime_error(fmt::format("[creation.cpp](zeros) Invalid workspaceSize: {}", workspaceSize));
// 申请工作空间
void *workspaceAddr = nullptr;
if(workspaceSize > 0) {
error = aclrtMalloc(&workspaceAddr, workspaceSize, ACL_MEM_MALLOC_HUGE_FIRST);
if(error != ACL_SUCCESS) throw std::runtime_error(fmt::format("[creation.cpp](zeros) aclrtMalloc error = {}",error));
}
error = aclnnInplaceZero(workspaceAddr, workspaceSize, executor, nullptr);
if(error != ACL_SUCCESS) throw std::runtime_error(fmt::format("[creation.cpp](zeros) aclnnInplaceZero error = {}",error));
error = aclrtSynchronizeDevice();
if(error != ACL_SUCCESS) throw std::runtime_error(fmt::format("[creation.cpp](zeros) aclrtSynchronizeDevice error = {}",error));
// 执行结束后释放工作空间
if(workspaceAddr != nullptr) {
aclrtFree(workspaceAddr);
}
return array;
}
NPUArray Zeros_like(const NPUArray& other, py::dtype dtype) {
auto array = NPUArray(other.shape, dtype);
uint64_t workspaceSize = 0;
aclOpExecutor *executor;
auto error = aclnnInplaceZeroGetWorkspaceSize(array.tensorPtr, &workspaceSize, &executor);
if(error != ACL_SUCCESS) throw std::runtime_error(fmt::format("[creation.cpp](zeros) aclnnInplaceZeroGetWorkspaceSize error = {}",error));
// 检查workspaceSize是否有效
if(workspaceSize < 0) throw std::runtime_error(fmt::format("[creation.cpp](zeros) Invalid workspaceSize: {}", workspaceSize));
// 申请工作空间
void *workspaceAddr = nullptr;
if(workspaceSize > 0) {
error = aclrtMalloc(&workspaceAddr, workspaceSize, ACL_MEM_MALLOC_HUGE_FIRST);
if(error != ACL_SUCCESS) throw std::runtime_error(fmt::format("[creation.cpp](zeros) aclrtMalloc error = {}",error));
}
error = aclnnInplaceZero(workspaceAddr, workspaceSize, executor, nullptr);
if(error != ACL_SUCCESS) throw std::runtime_error(fmt::format("[creation.cpp](zeros) aclnnInplaceZero error = {}",error));
error = aclrtSynchronizeDevice();
if(error != ACL_SUCCESS) throw std::runtime_error(fmt::format("[creation.cpp](zeros) aclrtSynchronizeDevice error = {}",error));
// 执行结束后释放工作空间
if(workspaceAddr != nullptr) {
aclrtFree(workspaceAddr);
}
return array;
}
NPUArray Full(const std::vector<int64_t>& shape, const py::object& value, py::dtype dtype) {
auto array = NPUArray(shape, dtype);
double valueDouble = 0;
if (value.is_none()) {
throw std::runtime_error("[creation.cpp](full) Input is None");
}
try {
valueDouble = py::cast<double>(value);
} catch (const py::cast_error& e) {
throw std::runtime_error("[creation.cpp](full) Conversion error: " + std::string(e.what()));
}
aclScalar* scalar = CreateScalar(valueDouble, array.aclDtype);
uint64_t workspaceSize = 0;
aclOpExecutor *executor;
auto error = aclnnInplaceFillScalarGetWorkspaceSize(array.tensorPtr, scalar, &workspaceSize, &executor);
if(error != ACL_SUCCESS) throw std::runtime_error(fmt::format("[creation.cpp](full) aclnnInplaceFillScalarGetWorkspaceSize error = {}",error));
// 检查workspaceSize是否有效
if(workspaceSize < 0) throw std::runtime_error(fmt::format("[creation.cpp](full) Invalid workspaceSize: {}", workspaceSize));
// 3. 申请工作空间
void *workspaceAddr = nullptr;
if(workspaceSize > 0) {
error = aclrtMalloc(&workspaceAddr, workspaceSize, ACL_MEM_MALLOC_HUGE_FIRST);
if(error != ACL_SUCCESS) throw std::runtime_error(fmt::format("[creation.cpp](full) aclrtMalloc error = {}",error));
}
error = aclnnInplaceFillScalar(workspaceAddr, workspaceSize, executor, nullptr);
if(error != ACL_SUCCESS) throw std::runtime_error(fmt::format("[creation.cpp](full) aclnnInplaceFillScalar error = {}", error));
error = aclrtSynchronizeDevice();
if(error != ACL_SUCCESS) throw std::runtime_error(fmt::format("[creation.cpp](full) aclrtSynchronizeDevice error = {}",error));
// 6. 释放
if(workspaceAddr != nullptr) {
aclrtFree(workspaceAddr);
}
aclDestroyScalar(scalar);
return array;
}
NPUArray Full_like(const NPUArray& other, const py::object& value, py::dtype dtype) {
auto array = NPUArray(other.shape, dtype);
double valueDouble = 0;
if (value.is_none()) {
throw std::runtime_error("[creation.cpp](full) Input is None");
}
try {
valueDouble = py::cast<double>(value);
} catch (const py::cast_error& e) {
throw std::runtime_error("[creation.cpp](full) Conversion error: " + std::string(e.what()));
}
aclScalar* scalar = CreateScalar(valueDouble, array.aclDtype);
uint64_t workspaceSize = 0;
aclOpExecutor *executor;
auto error = aclnnInplaceFillScalarGetWorkspaceSize(array.tensorPtr, scalar, &workspaceSize, &executor);
if(error != ACL_SUCCESS) throw std::runtime_error(fmt::format("[creation.cpp](full) aclnnInplaceFillScalarGetWorkspaceSize error = {}",error));
// 检查workspaceSize是否有效
if(workspaceSize < 0) throw std::runtime_error(fmt::format("[creation.cpp](full) Invalid workspaceSize: {}", workspaceSize));
// 3. 申请工作空间
void *workspaceAddr = nullptr;
if(workspaceSize > 0) {
error = aclrtMalloc(&workspaceAddr, workspaceSize, ACL_MEM_MALLOC_HUGE_FIRST);
if(error != ACL_SUCCESS) throw std::runtime_error(fmt::format("[creation.cpp](full) aclrtMalloc error = {}",error));
}
error = aclnnInplaceFillScalar(workspaceAddr, workspaceSize, executor, nullptr);
if(error != ACL_SUCCESS) throw std::runtime_error(fmt::format("[creation.cpp](full) aclnnInplaceFillScalar error = {}", error));
error = aclrtSynchronizeDevice();
if(error != ACL_SUCCESS) throw std::runtime_error(fmt::format("[creation.cpp](full) aclrtSynchronizeDevice error = {}",error));
// 6. 释放
if(workspaceAddr != nullptr) {
aclrtFree(workspaceAddr);
}
aclDestroyScalar(scalar);
return array;
}
NPUArray Eye(int64_t n, py::dtype dtype) {
auto array = NPUArray({n, n}, dtype);
uint64_t workspaceSize = 0;
aclOpExecutor *executor;
auto error = aclnnEyeGetWorkspaceSize(n, n, array.tensorPtr, &workspaceSize, &executor);
if(error != ACL_SUCCESS) throw std::runtime_error(fmt::format("[creation.cpp](eye) aclnnEyeGetWorkspaceSize error = {}",error));
// 检查workspaceSize是否有效
if(workspaceSize < 0) throw std::runtime_error(fmt::format("[creation.cpp](eye) Invalid workspaceSize: {}", workspaceSize));
// 申请工作空间
void *workspaceAddr = nullptr;
if(workspaceSize > 0) {
error = aclrtMalloc(&workspaceAddr, workspaceSize, ACL_MEM_MALLOC_HUGE_FIRST);
if(error != ACL_SUCCESS) throw std::runtime_error(fmt::format("[creation.cpp](eye) aclrtMalloc error = {}",error));
}
error = aclnnEye(workspaceAddr, workspaceSize, executor, nullptr);
if(error != ACL_SUCCESS) throw std::runtime_error(fmt::format("[creation.cpp](eye) aclnnEye error = {}",error));
error = aclrtSynchronizeDevice();
if(error != ACL_SUCCESS) throw std::runtime_error(fmt::format("[creation.cpp](eye) aclrtSynchronizeDevice error = {}",error));
// 执行结束后释放工作空间
if(workspaceAddr != nullptr) {
aclrtFree(workspaceAddr);
}
return array;
}
NPUArray Ones(const std::vector<int64_t>& shape, py::dtype dtype) {
auto array = NPUArray(shape, dtype);
uint64_t workspaceSize = 0;
aclOpExecutor *executor;
auto error = aclnnInplaceOneGetWorkspaceSize(array.tensorPtr, &workspaceSize, &executor);
if(error != ACL_SUCCESS) {
std::string error_msg = fmt::format("[basic.cpp](ones) aclnnInplaceOneGetWorkspaceSize error = {}", error);
const char* detailed_msg = aclGetRecentErrMsg();
if (detailed_msg != nullptr && strlen(detailed_msg) > 0) {
error_msg += std::string(" - ") + detailed_msg;
}
throw std::runtime_error(error_msg);
}
// 检查workspaceSize是否有效
if(workspaceSize < 0) throw std::runtime_error(fmt::format("[basic.cpp](ones) Invalid workspaceSize: {}", workspaceSize));
// 申请工作空间
void *workspaceAddr = nullptr;
if(workspaceSize > 0) {
error = aclrtMalloc(&workspaceAddr, workspaceSize, ACL_MEM_MALLOC_HUGE_FIRST);
if(error != ACL_SUCCESS) {
std::string error_msg = fmt::format("[basic.cpp](ones) aclrtMalloc error = {}", error);
const char* detailed_msg = aclGetRecentErrMsg();
if (detailed_msg != nullptr && strlen(detailed_msg) > 0) {
error_msg += std::string(" - ") + detailed_msg;
}
throw std::runtime_error(error_msg);
}
}
error = aclnnInplaceOne(workspaceAddr, workspaceSize, executor, nullptr);
if(error != ACL_SUCCESS) {
std::string error_msg = fmt::format("[basic.cpp](ones) aclnnInplaceOne error = {}", error);
const char* detailed_msg = aclGetRecentErrMsg();
if (detailed_msg != nullptr && strlen(detailed_msg) > 0) {
error_msg += std::string(" - ") + detailed_msg;
}
throw std::runtime_error(error_msg);
}
error = aclrtSynchronizeDevice();
if(error != ACL_SUCCESS) {
std::string error_msg = fmt::format("[basic.cpp](ones) aclrtSynchronizeDevice error = {}", error);
const char* detailed_msg = aclGetRecentErrMsg();
if (detailed_msg != nullptr && strlen(detailed_msg) > 0) {
error_msg += std::string(" - ") + detailed_msg;
}
throw std::runtime_error(error_msg);
}
// 执行结束后释放工作空间
if(workspaceAddr != nullptr) {
aclrtFree(workspaceAddr);
}
return array;
}
NPUArray Identity(int64_t n, py::dtype dtype) {
auto array = NPUArray({n, n}, dtype);
uint64_t workspaceSize = 0;
aclOpExecutor *executor;
auto error = aclnnEyeGetWorkspaceSize(n, n, // 行, 列, 对角线偏移(k=0)
array.tensorPtr,
&workspaceSize,
&executor);
if (error != ACL_SUCCESS) {
std::string error_msg = fmt::format("[basic.cpp](identity) aclnnEyeGetWorkspaceSize error = {}", error);
const char* detailed_msg = aclGetRecentErrMsg();
if (detailed_msg != nullptr && strlen(detailed_msg) > 0) {
error_msg += std::string(" - ") + detailed_msg;
}
throw std::runtime_error(error_msg);
}
if (workspaceSize < 0) {
throw std::runtime_error(fmt::format("[basic.cpp](identity) Invalid workspaceSize: {}", workspaceSize));
}
void* workspaceAddr = nullptr;
if (workspaceSize > 0) {
error = aclrtMalloc(&workspaceAddr, workspaceSize, ACL_MEM_MALLOC_HUGE_FIRST);
if (error != ACL_SUCCESS) {
std::string error_msg = fmt::format("[basic.cpp](identity) aclrtMalloc error = {}", error);
const char* detailed_msg = aclGetRecentErrMsg();
if (detailed_msg != nullptr && strlen(detailed_msg) > 0) {
error_msg += std::string(" - ") + detailed_msg;
}
throw std::runtime_error(error_msg);
}
}
error = aclnnEye(workspaceAddr, workspaceSize, executor, nullptr);
if (error != ACL_SUCCESS) {
std::string error_msg = fmt::format("[basic.cpp](identity) aclnnEye error = {}", error);
const char* detailed_msg = aclGetRecentErrMsg();
if (detailed_msg != nullptr && strlen(detailed_msg) > 0) {
error_msg += std::string(" - ") + detailed_msg;
}
throw std::runtime_error(error_msg);
}
error = aclrtSynchronizeDevice();
if (error != ACL_SUCCESS) {
std::string error_msg = fmt::format("[basic.cpp](identity) aclrtSynchronizeDevice error = {}", error);
const char* detailed_msg = aclGetRecentErrMsg();
if (detailed_msg != nullptr && strlen(detailed_msg) > 0) {
error_msg += std::string(" - ") + detailed_msg;
}
throw std::runtime_error(error_msg);
}
if (workspaceAddr != nullptr) {
aclrtFree(workspaceAddr);
}
return array;
}
NPUArray ones_like(const NPUArray& other, py::dtype dtype) {
auto array = NPUArray(other.shape, dtype);
uint64_t workspaceSize = 0;
aclOpExecutor *executor;
auto error = aclnnInplaceOneGetWorkspaceSize(array.tensorPtr, &workspaceSize, &executor);
if (error != ACL_SUCCESS) {
std::string error_msg = fmt::format("[basic.cpp](ones_like) aclnnInplaceOneGetWorkspaceSize error = {}", error);
const char* detailed_msg = aclGetRecentErrMsg();
if (detailed_msg && strlen(detailed_msg) > 0) {
error_msg += std::string(" - ") + detailed_msg;
}
throw std::runtime_error(error_msg);
}
if (workspaceSize < 0) {
throw std::runtime_error(fmt::format("[basic.cpp](ones_like) Invalid workspaceSize: {}", workspaceSize));
}
void* workspaceAddr = nullptr;
if (workspaceSize > 0) {
error = aclrtMalloc(&workspaceAddr, workspaceSize, ACL_MEM_MALLOC_HUGE_FIRST);
if (error != ACL_SUCCESS) {
std::string error_msg = fmt::format("[basic.cpp](ones_like) aclrtMalloc error = {}", error);
const char* detailed_msg = aclGetRecentErrMsg();
if (detailed_msg && strlen(detailed_msg) > 0) {
error_msg += std::string(" - ") + detailed_msg;
}
throw std::runtime_error(error_msg);
}
}
error = aclnnInplaceOne(workspaceAddr, workspaceSize, executor, nullptr);
if (error != ACL_SUCCESS) {
std::string error_msg = fmt::format("[basic.cpp](ones_like) aclnnInplaceOne error = {}", error);
const char* detailed_msg = aclGetRecentErrMsg();
if (detailed_msg && strlen(detailed_msg) > 0) {
error_msg += std::string(" - ") + detailed_msg;
}
throw std::runtime_error(error_msg);
}
error = aclrtSynchronizeDevice();
if (error != ACL_SUCCESS) {
std::string error_msg = fmt::format("[basic.cpp](ones_like) aclrtSynchronizeDevice error = {}", error);
const char* detailed_msg = aclGetRecentErrMsg();
if (detailed_msg && strlen(detailed_msg) > 0) {
error_msg += std::string(" - ") + detailed_msg;
}
throw std::runtime_error(error_msg);
}
if (workspaceAddr) {
aclrtFree(workspaceAddr);
}
return array;
}