-
Notifications
You must be signed in to change notification settings - Fork 5.7k
/
Copy pathkernel_utils.h
419 lines (380 loc) · 23.5 KB
/
kernel_utils.h
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
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
// Copyright (c) 2021 PaddlePaddle 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.
#pragma once
#include "paddle/phi/backends/all_context.h"
#include "paddle/phi/common/int_array.h"
#include "paddle/phi/common/scalar.h"
#include "paddle/phi/common/tensor_ref.h"
#include "paddle/phi/core/dense_tensor.h"
#include "paddle/phi/core/enforce.h"
#include "paddle/phi/core/extended_tensor.h"
#include "paddle/phi/core/framework/feed_fetch_type.h"
#include "paddle/phi/core/kernel_context.h"
#include "paddle/phi/core/selected_rows.h"
#include "paddle/phi/core/sparse_coo_tensor.h"
#include "paddle/phi/core/sparse_csr_tensor.h"
#include "paddle/phi/core/string_tensor.h"
#include "paddle/phi/core/tensor_array.h"
#include "paddle/phi/core/vocab/string_array.h"
namespace phi {
// PD_KERNEL has been used by custom op api
#define PHI_KERNEL(...) \
::phi::KernelImpl<decltype(&__VA_ARGS__), &__VA_ARGS__>::Compute
#define PHI_VARIADIC_KERNEL(...) \
reinterpret_cast<void*>(&::phi::KernelImpl<decltype(&__VA_ARGS__), \
&__VA_ARGS__>::VariadicCompute)
#define PD_SPECIALIZE_KernelCallHelper_FOR_DEVICE_CONTEXT(dev_ctx) \
template <typename... Tail> \
struct KernelCallHelper<const dev_ctx&, Tail...> { \
template <int dev_ctx_idx, \
int in_idx, \
int attr_idx, \
int out_idx, \
typename... PreviousArgs> \
static void Compute(KernelContext* ctx, PreviousArgs&... pargs) { \
static_assert(in_idx == 0, \
"Kernel's DeviceContext should appear before Inputs."); \
static_assert( \
attr_idx == 0, \
"Kernel's DeviceContext should appear before Attributes."); \
static_assert(out_idx == 0, \
"Kernel's DeviceContext should appear before Outputs."); \
const dev_ctx& arg = ctx->GetDeviceContext<dev_ctx>(); \
KernelCallHelper<Tail...>:: \
template Compute<dev_ctx_idx + 1, in_idx, attr_idx, out_idx>( \
ctx, pargs..., arg); \
} \
}
#define PD_SPECIALIZE_KernelCallHelper_FOR_INPUT(tensor_type) \
template <typename... Tail> \
struct KernelCallHelper<const tensor_type&, Tail...> { \
template <int dev_ctx_idx, \
int in_idx, \
int attr_idx, \
int out_idx, \
typename... PreviousArgs> \
static void Compute(KernelContext* ctx, PreviousArgs&... pargs) { \
static_assert(attr_idx == 0, \
"Kernel's Input should appear before Attributes."); \
static_assert(out_idx == 0, \
"Kernel's Input should appear before Outputs."); \
const std::pair<int, int>& range = ctx->InputRangeAt(in_idx); \
const tensor_type& arg = ctx->InputAt<tensor_type>(range.first); \
KernelCallHelper<Tail...>:: \
template Compute<dev_ctx_idx, in_idx + 1, attr_idx, out_idx>( \
ctx, pargs..., arg); \
} \
}
#define PD_SPECIALIZE_KernelCallHelper_FOR_OPTIONAL_INPUT(tensor_type) \
template <typename... Tail> \
struct KernelCallHelper<const paddle::optional<tensor_type>&, Tail...> { \
template <int dev_ctx_idx, \
int in_idx, \
int attr_idx, \
int out_idx, \
typename... PreviousArgs> \
static void Compute(KernelContext* ctx, PreviousArgs&... pargs) { \
static_assert(attr_idx == 0, \
"Kernel's Input should appear before Attributes."); \
static_assert(out_idx == 0, \
"Kernel's Input should appear before Outputs."); \
const std::pair<int, int>& range = ctx->InputRangeAt(in_idx); \
auto arg = ctx->OptionalInputAt<tensor_type>(range.first); \
KernelCallHelper<Tail...>:: \
template Compute<dev_ctx_idx, in_idx + 1, attr_idx, out_idx>( \
ctx, pargs..., arg); \
} \
}
#define PD_SPECIALIZE_KernelCallHelper_FOR_MULTI_INPUT(tensor_type) \
template <typename... Tail> \
struct KernelCallHelper<const std::vector<const tensor_type*>&, Tail...> { \
template <int dev_ctx_idx, \
int in_idx, \
int attr_idx, \
int out_idx, \
typename... PreviousArgs> \
static void Compute(KernelContext* ctx, PreviousArgs&... pargs) { \
static_assert(attr_idx == 0, \
"Kernel's Input should appear before Attributes."); \
static_assert(out_idx == 0, \
"Kernel's Input should appear before Outputs."); \
const std::pair<int, int>& range = ctx->InputRangeAt(in_idx); \
std::vector<const tensor_type*> arg = \
ctx->InputsBetween<tensor_type>(range.first, range.second); \
KernelCallHelper<Tail...>:: \
template Compute<dev_ctx_idx, in_idx + 1, attr_idx, out_idx>( \
ctx, pargs..., arg); \
} \
}
#define PD_SPECIALIZE_KernelCallHelper_FOR_OPTIONAL_MULTI_INPUT(tensor_type) \
template <typename... Tail> \
struct KernelCallHelper< \
const paddle::optional<std::vector<const tensor_type*>>&, \
Tail...> { \
template <int dev_ctx_idx, \
int in_idx, \
int attr_idx, \
int out_idx, \
typename... PreviousArgs> \
static void Compute(KernelContext* ctx, PreviousArgs&... pargs) { \
static_assert(attr_idx == 0, \
"Kernel's Input should appear before Attributes."); \
static_assert(out_idx == 0, \
"Kernel's Input should appear before Outputs."); \
const std::pair<int, int>& range = ctx->InputRangeAt(in_idx); \
paddle::optional<std::vector<const tensor_type*>> arg = \
ctx->OptionalInputsBetween<tensor_type>(range.first, range.second); \
KernelCallHelper<Tail...>:: \
template Compute<dev_ctx_idx, in_idx + 1, attr_idx, out_idx>( \
ctx, pargs..., arg); \
} \
}
#define PD_SPECIALIZE_KernelCallHelper_FOR_ATTRIBUTE(attr_type) \
template <typename... Tail> \
struct KernelCallHelper<attr_type, Tail...> { \
template <int dev_ctx_idx, \
int in_idx, \
int attr_idx, \
int out_idx, \
typename... PreviousArgs> \
static void Compute(KernelContext* ctx, PreviousArgs&... pargs) { \
static_assert(out_idx == 0, \
"Kernel's Attributes should appear before Outputs."); \
attr_type arg = ctx->AttrAt<attr_type>(attr_idx); \
KernelCallHelper<Tail...>:: \
template Compute<dev_ctx_idx, in_idx, attr_idx + 1, out_idx>( \
ctx, pargs..., arg); \
} \
}
#define PD_SPECIALIZE_KernelCallHelper_FOR_CONST_ATTRIBUTE_REF(attr_type) \
template <typename... Tail> \
struct KernelCallHelper<const attr_type&, Tail...> { \
template <int dev_ctx_idx, \
int in_idx, \
int attr_idx, \
int out_idx, \
typename... PreviousArgs> \
static void Compute(KernelContext* ctx, PreviousArgs&... pargs) { \
static_assert(out_idx == 0, \
"Kernel's Attributes should appear before Outputs."); \
const attr_type& arg = ctx->AttrAt<attr_type>(attr_idx); \
KernelCallHelper<Tail...>:: \
template Compute<dev_ctx_idx, in_idx, attr_idx + 1, out_idx>( \
ctx, pargs..., arg); \
} \
}
#define PD_SPECIALIZE_KernelCallHelper_FOR_OUTPUT(tensor_type) \
template <typename... Tail> \
struct KernelCallHelper<tensor_type*, Tail...> { \
template <int dev_ctx_idx, \
int in_idx, \
int attr_idx, \
int out_idx, \
typename... PreviousArgs> \
static void Compute(KernelContext* ctx, PreviousArgs&... pargs) { \
const std::pair<int, int>& range = ctx->OutputRangeAt(out_idx); \
tensor_type* arg = ctx->MutableOutputAt<tensor_type>(range.first); \
KernelCallHelper<Tail...>:: \
template Compute<dev_ctx_idx, in_idx, attr_idx, out_idx + 1>( \
ctx, pargs..., arg); \
} \
}
#define PD_SPECIALIZE_KernelCallHelper_FOR_MULTI_OUTPUT(tensor_type) \
template <typename... Tail> \
struct KernelCallHelper<std::vector<tensor_type*>, Tail...> { \
template <int dev_ctx_idx, \
int in_idx, \
int attr_idx, \
int out_idx, \
typename... PreviousArgs> \
static void Compute(KernelContext* ctx, PreviousArgs&... pargs) { \
const std::pair<int, int>& range = ctx->OutputRangeAt(out_idx); \
std::vector<tensor_type*> arg = \
ctx->MutableOutputBetween<tensor_type>(range.first, range.second); \
KernelCallHelper<Tail...>:: \
template Compute<dev_ctx_idx, in_idx, attr_idx, out_idx + 1>( \
ctx, pargs..., arg); \
} \
}
#define PD_SPECIALIZE_KernelCallHelper_FOR_TENSOR_SCALAR(attr_type) \
template <typename... Tail> \
struct KernelCallHelper<const attr_type&, Tail...> { \
template <int dev_ctx_idx, \
int in_idx, \
int attr_idx, \
int out_idx, \
typename... PreviousArgs> \
static void Compute(KernelContext* ctx, PreviousArgs&... pargs) { \
static_assert(out_idx == 0, \
"Kernel's Attributes should appear before Outputs."); \
const Attribute& t = ctx->AttrAt(attr_idx); \
static Attribute cmp_t = phi::TensorRef(nullptr); \
attr_type attr1; \
if (cmp_t.index() == t.index()) { \
attr1 = attr_type(*paddle::get<phi::TensorRef>(t).Get()); \
} else { \
attr1 = paddle::get<attr_type>(t); \
} \
KernelCallHelper<Tail...>:: \
template Compute<dev_ctx_idx, in_idx, attr_idx + 1, out_idx>( \
ctx, pargs..., attr1); \
} \
}
#define PD_SPECIALIZE_KernelCallHelper_FOR_TENSOR_INTARRAY(attr_type) \
template <typename... Tail> \
struct KernelCallHelper<const attr_type&, Tail...> { \
template <int dev_ctx_idx, \
int in_idx, \
int attr_idx, \
int out_idx, \
typename... PreviousArgs> \
static void Compute(KernelContext* ctx, PreviousArgs&... pargs) { \
static_assert(out_idx == 0, \
"Kernel's Attributes should appear before Outputs."); \
const Attribute& t = ctx->AttrAt(attr_idx); \
static Attribute cmp_t = phi::TensorRef(nullptr); \
static Attribute vec_ref = \
std::vector<phi::TensorRef>({phi::TensorRef(nullptr)}); \
attr_type attr1; \
if (cmp_t.index() == t.index()) { \
attr1 = attr_type(*paddle::get<phi::TensorRef>(t).Get()); \
} else if (vec_ref.index() == t.index()) { \
attr1 = attr_type(paddle::get<std::vector<phi::TensorRef>>(t)); \
} else { \
attr1 = paddle::get<attr_type>(t); \
} \
KernelCallHelper<Tail...>:: \
template Compute<dev_ctx_idx, in_idx, attr_idx + 1, out_idx>( \
ctx, pargs..., attr1); \
} \
}
template <typename T>
struct TypeTag {};
template <typename Fn, Fn fn>
struct KernelImpl;
template <typename Return,
typename DevCtx,
typename... Args,
Return (*kernel_fn)(DevCtx, Args...)>
struct KernelImpl<Return (*)(DevCtx, Args...), kernel_fn> {
static void Compute(KernelContext* ctx) {
KernelCallHelper<DevCtx, Args..., TypeTag<int>>::
template Compute<0, 0, 0, 0>(ctx);
}
static void VariadicCompute(const DeviceContext& dev_ctx, Args... args) {
return kernel_fn(static_cast<DevCtx>(dev_ctx), std::forward<Args>(args)...);
}
private:
template <typename... RemainingArgs>
struct KernelCallHelper;
/* DeviceContext Helpers */
PD_SPECIALIZE_KernelCallHelper_FOR_DEVICE_CONTEXT(CPUContext);
#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
PD_SPECIALIZE_KernelCallHelper_FOR_DEVICE_CONTEXT(GPUContext);
#endif
#ifdef PADDLE_WITH_XPU
PD_SPECIALIZE_KernelCallHelper_FOR_DEVICE_CONTEXT(XPUContext);
#endif
#ifdef PADDLE_WITH_CUSTOM_DEVICE
PD_SPECIALIZE_KernelCallHelper_FOR_DEVICE_CONTEXT(CustomContext);
#endif
#ifdef PADDLE_WITH_DNNL
PD_SPECIALIZE_KernelCallHelper_FOR_DEVICE_CONTEXT(OneDNNContext);
#endif
/* Input Helpers */
PD_SPECIALIZE_KernelCallHelper_FOR_INPUT(DenseTensor);
PD_SPECIALIZE_KernelCallHelper_FOR_OPTIONAL_INPUT(DenseTensor);
PD_SPECIALIZE_KernelCallHelper_FOR_OPTIONAL_INPUT(SelectedRows);
PD_SPECIALIZE_KernelCallHelper_FOR_MULTI_INPUT(DenseTensor);
PD_SPECIALIZE_KernelCallHelper_FOR_INPUT(ExtendedTensor);
PD_SPECIALIZE_KernelCallHelper_FOR_OPTIONAL_INPUT(ExtendedTensor);
PD_SPECIALIZE_KernelCallHelper_FOR_MULTI_INPUT(ExtendedTensor);
PD_SPECIALIZE_KernelCallHelper_FOR_MULTI_INPUT(TensorBase);
PD_SPECIALIZE_KernelCallHelper_FOR_MULTI_INPUT(SelectedRows);
PD_SPECIALIZE_KernelCallHelper_FOR_INPUT(SelectedRows);
PD_SPECIALIZE_KernelCallHelper_FOR_OPTIONAL_MULTI_INPUT(DenseTensor);
PD_SPECIALIZE_KernelCallHelper_FOR_INPUT(SparseCooTensor);
PD_SPECIALIZE_KernelCallHelper_FOR_OPTIONAL_INPUT(SparseCooTensor);
PD_SPECIALIZE_KernelCallHelper_FOR_MULTI_INPUT(SparseCooTensor);
PD_SPECIALIZE_KernelCallHelper_FOR_INPUT(SparseCsrTensor);
PD_SPECIALIZE_KernelCallHelper_FOR_OPTIONAL_INPUT(SparseCsrTensor);
PD_SPECIALIZE_KernelCallHelper_FOR_MULTI_INPUT(SparseCsrTensor);
PD_SPECIALIZE_KernelCallHelper_FOR_INPUT(StringTensor);
PD_SPECIALIZE_KernelCallHelper_FOR_OPTIONAL_INPUT(StringTensor);
PD_SPECIALIZE_KernelCallHelper_FOR_MULTI_INPUT(StringTensor);
PD_SPECIALIZE_KernelCallHelper_FOR_INPUT(TensorArray);
PD_SPECIALIZE_KernelCallHelper_FOR_MULTI_INPUT(TensorArray);
PD_SPECIALIZE_KernelCallHelper_FOR_INPUT(phi::Strings);
PD_SPECIALIZE_KernelCallHelper_FOR_OPTIONAL_INPUT(phi::Strings);
PD_SPECIALIZE_KernelCallHelper_FOR_INPUT(phi::FeedList);
/* Attribute Helpers */
PD_SPECIALIZE_KernelCallHelper_FOR_ATTRIBUTE(bool);
PD_SPECIALIZE_KernelCallHelper_FOR_ATTRIBUTE(float);
PD_SPECIALIZE_KernelCallHelper_FOR_ATTRIBUTE(double);
PD_SPECIALIZE_KernelCallHelper_FOR_ATTRIBUTE(int);
PD_SPECIALIZE_KernelCallHelper_FOR_ATTRIBUTE(int64_t);
PD_SPECIALIZE_KernelCallHelper_FOR_ATTRIBUTE(phi::dtype::float16);
PD_SPECIALIZE_KernelCallHelper_FOR_ATTRIBUTE(DataType);
PD_SPECIALIZE_KernelCallHelper_FOR_ATTRIBUTE(DataLayout);
PD_SPECIALIZE_KernelCallHelper_FOR_ATTRIBUTE(Place);
PD_SPECIALIZE_KernelCallHelper_FOR_CONST_ATTRIBUTE_REF(std::string);
PD_SPECIALIZE_KernelCallHelper_FOR_TENSOR_SCALAR(Scalar);
PD_SPECIALIZE_KernelCallHelper_FOR_TENSOR_INTARRAY(IntArray);
PD_SPECIALIZE_KernelCallHelper_FOR_CONST_ATTRIBUTE_REF(std::vector<bool>);
PD_SPECIALIZE_KernelCallHelper_FOR_CONST_ATTRIBUTE_REF(std::vector<int>);
PD_SPECIALIZE_KernelCallHelper_FOR_CONST_ATTRIBUTE_REF(std::vector<int64_t>);
PD_SPECIALIZE_KernelCallHelper_FOR_CONST_ATTRIBUTE_REF(std::vector<float>);
PD_SPECIALIZE_KernelCallHelper_FOR_CONST_ATTRIBUTE_REF(std::vector<double>);
PD_SPECIALIZE_KernelCallHelper_FOR_CONST_ATTRIBUTE_REF(
std::vector<std::string>);
PD_SPECIALIZE_KernelCallHelper_FOR_CONST_ATTRIBUTE_REF(std::vector<Scalar>);
/* Output Helpers */
PD_SPECIALIZE_KernelCallHelper_FOR_OUTPUT(DenseTensor);
PD_SPECIALIZE_KernelCallHelper_FOR_MULTI_OUTPUT(DenseTensor);
PD_SPECIALIZE_KernelCallHelper_FOR_OUTPUT(SelectedRows);
PD_SPECIALIZE_KernelCallHelper_FOR_OUTPUT(SparseCooTensor);
PD_SPECIALIZE_KernelCallHelper_FOR_MULTI_OUTPUT(SparseCooTensor);
PD_SPECIALIZE_KernelCallHelper_FOR_OUTPUT(SparseCsrTensor);
PD_SPECIALIZE_KernelCallHelper_FOR_MULTI_OUTPUT(SparseCsrTensor);
PD_SPECIALIZE_KernelCallHelper_FOR_OUTPUT(StringTensor);
PD_SPECIALIZE_KernelCallHelper_FOR_MULTI_OUTPUT(StringTensor);
PD_SPECIALIZE_KernelCallHelper_FOR_OUTPUT(TensorArray);
PD_SPECIALIZE_KernelCallHelper_FOR_OUTPUT(ExtendedTensor);
PD_SPECIALIZE_KernelCallHelper_FOR_MULTI_OUTPUT(ExtendedTensor);
PD_SPECIALIZE_KernelCallHelper_FOR_MULTI_OUTPUT(Vocab);
/* End case */
template <typename T>
struct KernelCallHelper<TypeTag<T>> {
template <int dev_ctx_idx, int in_idx, int attr_idx, int out_idx>
static void Compute(KernelContext* ctx UNUSED,
DevCtx dev_ctx,
Args&... args) {
static_assert(dev_ctx_idx > 0,
"Kernel should pass DeviceContext as argument.");
return kernel_fn(dev_ctx, args...);
}
};
};
inline bool recompute_reduce_all(const DenseTensor& x,
const IntArray& dims,
bool reduce_all = false) {
if (dims.size() == 0 || x.dims().size() == 0 ||
static_cast<int>(dims.size()) == x.dims().size() || reduce_all) {
// when input 0D, it can only reduce_all
return true;
} else {
return false;
}
}
} // namespace phi