megenginelite-sys 1.8.2

A safe megenginelite wrapper in Rust
Documentation 
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

/**
 * \file dnn/src/aarch64/matrix_mul/algos.h
 * MegEngine is Licensed under the Apache License, Version 2.0 (the "License")
 *
 * Copyright (c) 2014-2021 Megvii Inc. All rights reserved.
 *
 * Unless required by applicable law or agreed to in writing,
 * software distributed under the License is distributed on an
 * "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or
 * implied.
 */

#pragma once

#include "src/aarch64/matrix_mul/opr_impl.h"
#include "src/arm_common/matrix_mul/algos.h"
#include "src/fallback/matrix_mul/gemm_common.h"

namespace megdnn {
namespace aarch64 {

class MatrixMulImpl::AlgoF32K8x12x1 final : public AlgoBase {
public:
    AlgoAttribute attribute() const override { return AlgoAttribute::REPRODUCIBLE; }
    const char* name() const override { return "AARCH64_F32K8X12X1"; }
    bool usable(const KernSizeParam&) const override;
    size_t get_workspace(const KernSizeParam&) const override;
    kern_t get_kern(const KernSizeParam&) const override;
    MEGDNN_REG_GEMM_FUNC_FOR_IM2COL();
    MEGDNN_DECL_ALGO_TYPE(AARCH64_F32_K8X12X1)
};

class MatrixMulImpl::AlgoF32MK4_8x12x1 final : public AlgoBase {
public:
    AlgoAttribute attribute() const override {
        return AlgoAttribute::REPRODUCIBLE | AlgoAttribute::USABLE_DEPEND_ON_SHAPE;
    }
    const char* name() const override { return "AARCH64_F32_MK4_K8X12X1"; }
    bool usable(const KernSizeParam&) const override;
    size_t get_workspace(const KernSizeParam&) const override;
    kern_t get_kern(const KernSizeParam&) const override;
    MEGDNN_REG_GEMM_FUNC_FOR_IM2COL();
    MEGDNN_DECL_ALGO_TYPE(AARCH64_F32_MK4_K8X12X1)
};

class MatrixMulImpl::AlgoF32K4x16x1 final : public AlgoBase {
public:
    AlgoAttribute attribute() const override { return AlgoAttribute::REPRODUCIBLE; }
    const char* name() const override { return "AARCH64_F32K4X16X1"; }
    bool usable(const KernSizeParam&) const override;
    size_t get_workspace(const KernSizeParam&) const override;
    kern_t get_kern(const KernSizeParam&) const override;
    MEGDNN_REG_GEMM_FUNC_FOR_IM2COL();
    MEGDNN_DECL_ALGO_TYPE(AARCH64_F32_K4X16X1)
};

class MatrixMulImpl::AlgoF32MK4_4x16 final : public AlgoBase {
public:
    AlgoAttribute attribute() const override { return AlgoAttribute::REPRODUCIBLE; }
    const char* name() const override { return "AARCH64_F32_MK4_4x16"; }
    bool usable(const KernSizeParam&) const override;
    size_t get_workspace(const KernSizeParam&) const override;
    kern_t get_kern(const KernSizeParam&) const override;
    PackMode packmode() const override { return PackMode::NO_PACK; }
    MEGDNN_OVERRIDE_MATMUL_DESC(4, 16, 4, 4, AlgoDataType::FLOAT32, MK4)
    MEGDNN_DECL_ALGO_TYPE(AARCH64_F32_MK4_4x16)
};

class MatrixMulImpl::AlgoF32Gemv final : public arm_common::MatrixMulImpl::AlgoF32Gemv {
public:
    AlgoF32Gemv() : arm_common::MatrixMulImpl::AlgoF32Gemv() {
        m_handle_type = Handle::HandleType::AARCH64;
    }
    MEGDNN_DECL_ALGO_TYPE(AARCH64_F32_GEMV)
};

#if __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
class MatrixMulImpl::AlgoF16K8x24x1 final : public AlgoBase {
public:
    AlgoAttribute attribute() const override { return AlgoAttribute::REPRODUCIBLE; }
    const char* name() const override { return "AARCH64_F16_K8X24X1"; }
    bool usable(const KernSizeParam&) const override;
    size_t get_workspace(const KernSizeParam&) const override;
    kern_t get_kern(const KernSizeParam&) const override;
    MEGDNN_REG_GEMM_FUNC_FOR_IM2COL();
    MEGDNN_DECL_ALGO_TYPE(AARCH64_F16_K8X24X1)
};

class MatrixMulImpl::AlgoF16MK8_8x8 final : public AlgoBase {
public:
    AlgoAttribute attribute() const override { return AlgoAttribute::REPRODUCIBLE; }
    const char* name() const override { return "AARCH64_F16_MK8_8X8"; }
    bool usable(const KernSizeParam&) const override;
    size_t get_workspace(const KernSizeParam&) const override;
    kern_t get_kern(const KernSizeParam&) const override;
    PackMode packmode() const override { return PackMode::NO_PACK; }
    MEGDNN_OVERRIDE_MATMUL_DESC(8, 8, 8, 2, AlgoDataType::FLOAT16, MK8)
    MEGDNN_DECL_ALGO_TYPE(AARCH64_F16_MK8_8X8)
};

#endif

#if MGB_ENABLE_DOT
class MatrixMulImpl::AlgoInt8x8x32K8x12x4DotProd final : public AlgoBase {
public:
    AlgoAttribute attribute() const override { return AlgoAttribute::REPRODUCIBLE; }
    const char* name() const override { return "AARCH64_INT8X8X32_K8X12X4_DOTPROD"; }
    bool usable(const KernSizeParam&) const override;
    size_t get_workspace(const KernSizeParam&) const override;
    kern_t get_kern(const KernSizeParam&) const override;
    MEGDNN_REG_GEMM_FUNC_FOR_IM2COL();
    MEGDNN_DECL_ALGO_TYPE(AARCH64_INT8X8X32_K8X12X4_DOTPROD)
};

class MatrixMulImpl::AlgoInt8x8x32MK4_8x12x4DotProd final : public AlgoBase {
public:
    AlgoAttribute attribute() const override { return AlgoAttribute::REPRODUCIBLE; }
    const char* name() const override { return "AARCH64_INT8X8X32_MK4_8X12X4_DOTPROD"; }
    bool usable(const KernSizeParam&) const override;
    size_t get_workspace(const KernSizeParam&) const override;
    kern_t get_kern(const KernSizeParam&) const override;
    MEGDNN_REG_GEMM_FUNC_FOR_IM2COL();
    MEGDNN_DECL_ALGO_TYPE(AARCH64_INT8X8X32_MK4_8X12X4_DOTPROD)
};
#endif

class MatrixMulImpl::AlgoInt8x8x32MK4_4x4x16 final : public AlgoBase {
public:
    AlgoAttribute attribute() const override {
        return AlgoAttribute::REPRODUCIBLE | AlgoAttribute::USABLE_DEPEND_ON_SHAPE;
    }
    const char* name() const override { return "AARCH64_INT8X8X32_MK4_4X4X16"; }
    bool usable(const KernSizeParam&) const override;
    bool preferred(const KernSizeParam&) const override;
    size_t get_workspace(const KernSizeParam&) const override;
    kern_t get_kern(const KernSizeParam&) const override;
    PackMode packmode() const override { return PackMode::DEFAULT; }

    MEGDNN_REG_GEMM_FUNC_FOR_IM2COL();
    MEGDNN_DECL_ALGO_TYPE(AARCH64_INT8X8X32_MK4_4X4X16)
};

class MatrixMulImpl::AlgoInt8x8x32K4x4x16 final : public AlgoBase {
public:
    AlgoAttribute attribute() const override { return AlgoAttribute::REPRODUCIBLE; }
    const char* name() const override { return "AARCH64_INT8X8X32_K4X4X16"; }
    bool usable(const KernSizeParam&) const override;
    bool preferred(const KernSizeParam&) const override;
    size_t get_workspace(const KernSizeParam&) const override;
    kern_t get_kern(const KernSizeParam&) const override;

    MEGDNN_REG_GEMM_FUNC_FOR_IM2COL();
    MEGDNN_DECL_ALGO_TYPE(AARCH64_INT8X8X32_K4X4X16)
};

class MatrixMulImpl::AlgoInt8x8x32K8x8x8 final : public AlgoBase {
public:
    AlgoAttribute attribute() const override { return AlgoAttribute::REPRODUCIBLE; }
    const char* name() const override { return "AARCH64_INT8X8X32_K8X8X8"; }
    bool usable(const KernSizeParam&) const override;
    bool preferred(const KernSizeParam&) const override;
    size_t get_workspace(const KernSizeParam&) const override;
    kern_t get_kern(const KernSizeParam&) const override;
    MEGDNN_REG_GEMM_FUNC_FOR_IM2COL();
    MEGDNN_DECL_ALGO_TYPE(AARCH64_INT8X8X32_K8X8X8)
};

class MatrixMulImpl::AlgoInt8x8x16K8x8x8 final : public AlgoBase {
public:
    AlgoAttribute attribute() const override { return AlgoAttribute::REPRODUCIBLE; }
    const char* name() const override { return "AARCH64_INT8X8X16_K8X8X8"; }
    bool usable(const KernSizeParam&) const override;
    bool preferred(const KernSizeParam&) const override;
    size_t get_workspace(const KernSizeParam&) const override;
    kern_t get_kern(const KernSizeParam&) const override;

    MEGDNN_REG_GEMM_FUNC_FOR_IM2COL();
    MEGDNN_DECL_ALGO_TYPE(AARCH64_INT8X8X16_K8X8X8)
};

class MatrixMulImpl::AlgoInt8x8x16K4x4x16 final : public AlgoBase {
public:
    AlgoAttribute attribute() const override { return AlgoAttribute::REPRODUCIBLE; }
    const char* name() const override { return "AARCH64_INT8X8X16_K4X4X16"; }
    bool usable(const KernSizeParam&) const override;
    bool preferred(const KernSizeParam&) const override;
    size_t get_workspace(const KernSizeParam&) const override;
    kern_t get_kern(const KernSizeParam&) const override;
    MEGDNN_REG_GEMM_FUNC_FOR_IM2COL();
    MEGDNN_DECL_ALGO_TYPE(AARCH64_INT8X8X16_K4X4X16)
};

class MatrixMulImpl::AlgoInt4x4x16K8x8x8 final : public AlgoBase {
public:
    AlgoAttribute attribute() const override {
        return AlgoAttribute::REPRODUCIBLE | AlgoAttribute::USABLE_DEPEND_ON_SHAPE;
    }
    const char* name() const override { return "AARCH64_INT4X4X16_K8X8X8"; }
    bool usable(const KernSizeParam&) const override;
    bool preferred(const KernSizeParam&) const override;
    size_t get_workspace(const KernSizeParam&) const override;
    kern_t get_kern(const KernSizeParam&) const override;
    PackMode packmode() const override { return PackMode::DEFAULT; }
    MEGDNN_REG_GEMM_FUNC_FOR_IM2COL();
    MEGDNN_DECL_ALGO_TYPE(AARCH64_INT4X4X16_K8X8X8)
};

class MatrixMulImpl::AlgoInt8x8x16MK4_16x12x4 final : public AlgoBase {
public:
    AlgoAttribute attribute() const override {
        return AlgoAttribute::REPRODUCIBLE | AlgoAttribute::USABLE_DEPEND_ON_SHAPE;
    }
    const char* name() const override { return "AARCH64_INT8X8X16_MK4_16X12X4"; }
    bool usable(const KernSizeParam&) const override;
    bool preferred(const KernSizeParam&) const override;
    size_t get_workspace(const KernSizeParam&) const override;
    kern_t get_kern(const KernSizeParam&) const override;
    PackMode packmode() const override { return PackMode::DEFAULT; }

    MEGDNN_REG_GEMM_FUNC_FOR_IM2COL();
    MEGDNN_DECL_ALGO_TYPE(AARCH64_INT8X8X16_MK4_16X12X4)
};

class MatrixMulImpl::AlgoInt8x8x16MK4_K8x8x8 final : public AlgoBase {
public:
    AlgoAttribute attribute() const override {
        return AlgoAttribute::REPRODUCIBLE | AlgoAttribute::USABLE_DEPEND_ON_SHAPE;
    }
    const char* name() const override { return "AARCH64_INT8X8X16_MK4_K8X8X8"; }
    bool usable(const KernSizeParam&) const override;
    bool preferred(const KernSizeParam&) const override;
    size_t get_workspace(const KernSizeParam&) const override;
    kern_t get_kern(const KernSizeParam&) const override;
    PackMode packmode() const override { return PackMode::DEFAULT; }

    MEGDNN_REG_GEMM_FUNC_FOR_IM2COL();
    MEGDNN_DECL_ALGO_TYPE(AARCH64_INT8X8X16_MK4_K8X8X8)
};

class MatrixMulImpl::AlgoInt8x8x16MK4_4x4x8 final : public AlgoBase {
public:
    AlgoAttribute attribute() const override {
        return AlgoAttribute::REPRODUCIBLE | AlgoAttribute::USABLE_DEPEND_ON_SHAPE;
    }
    const char* name() const override { return "AARCH64_INT8X8X16_MK4_4X4X8"; }
    bool usable(const KernSizeParam&) const override;
    bool preferred(const KernSizeParam&) const override;
    size_t get_workspace(const KernSizeParam&) const override;
    kern_t get_kern(const KernSizeParam&) const override;
    PackMode packmode() const override { return PackMode::DEFAULT; }

    MEGDNN_REG_GEMM_FUNC_FOR_IM2COL();
    MEGDNN_DECL_ALGO_TYPE(AARCH64_INT8X8X16_MK4_4X4X8)
};

class MatrixMulImpl::AlgoInt16x16x32K12x8x1 final : public AlgoBase {
public:
    AlgoAttribute attribute() const override { return AlgoAttribute::REPRODUCIBLE; }
    const char* name() const override { return "AARCH64_INT16X16X32_K12X8X1"; }
    bool usable(const KernSizeParam&) const override;
    bool preferred(const KernSizeParam&) const override;
    size_t get_workspace(const KernSizeParam&) const override;
    kern_t get_kern(const KernSizeParam&) const override;
    MEGDNN_REG_GEMM_FUNC_FOR_IM2COL();
    MEGDNN_DECL_ALGO_TYPE(AARCH64_INT16X16X32_K12X8X1)
};

class MatrixMulImpl::AlgoInt16x16x32MK8_8x8 final : public AlgoBase {
public:
    AlgoAttribute attribute() const override { return AlgoAttribute::REPRODUCIBLE; }
    const char* name() const override { return "AARCH64_INT16X16X32_MK8_8X8"; }
    bool usable(const KernSizeParam&) const override;
    size_t get_workspace(const KernSizeParam&) const override;
    kern_t get_kern(const KernSizeParam&) const override;
    PackMode packmode() const override { return PackMode::NO_PACK; }
    MEGDNN_OVERRIDE_MATMUL_DESC(8, 8, 8, 2, AlgoDataType::INT16X16X32, MK8)
    MEGDNN_DECL_ALGO_TYPE(AARCH64_INT16X16X32_MK8_8X8)
};

#if MGB_ENABLE_DOT
class MatrixMulImpl::AlgoQuint8K8x8x4DotProd final : public AlgoBase {
public:
    AlgoAttribute attribute() const override { return AlgoAttribute::REPRODUCIBLE; }
    const char* name() const override { return "AARCH64_QUINT8_K8X8X4_DOTPROD"; }
    bool usable(const KernSizeParam&) const override;
    size_t get_workspace(const KernSizeParam&) const override;
    kern_t get_kern(const KernSizeParam&) const override;
    MEGDNN_REG_GEMM_FUNC_FOR_IM2COL();
    MEGDNN_DECL_ALGO_TYPE(AARCH64_QUINT8_K8X8X4_DOTPROD)
};
class MatrixMulImpl::AlgoQuint8GemvDotProd final : public AlgoBase {
public:
    AlgoAttribute attribute() const override {
        return AlgoAttribute::REPRODUCIBLE | AlgoAttribute::USABLE_DEPEND_ON_SHAPE;
    }
    const char* name() const override { return "AARCH64_QUINT8_GEMV_DOTPROD"; }
    bool usable(const KernSizeParam&) const override;
    bool preferred(const KernSizeParam&) const override;
    size_t get_workspace(const KernSizeParam&) const override { return 0; }
    kern_t get_kern(const KernSizeParam&) const override;
    AlgoSet algoset() const override { return AlgoSet::ALGO_TYPE_GEMV; }
    PackMode packmode() const override { return PackMode::NO_PACK; }
    MEGDNN_OVERRIDE_MATMUL_DESC(8, 16, 1, 2, AlgoDataType::QUINT8X8X32, DEFAULT)
    MEGDNN_DECL_ALGO_TYPE(AARCH64_QUINT8_GEMV_DOTPROD)
};
#endif
class MatrixMulImpl::AlgoQuint8K8x8x8 final : public AlgoBase {
public:
    AlgoAttribute attribute() const override { return AlgoAttribute::REPRODUCIBLE; }
    const char* name() const override { return "AARCH64_QUINT8_K8X8X8"; }
    bool usable(const KernSizeParam&) const override;
    size_t get_workspace(const KernSizeParam&) const override;
    kern_t get_kern(const KernSizeParam&) const override;
    MEGDNN_REG_GEMM_FUNC_FOR_IM2COL();
    MEGDNN_DECL_ALGO_TYPE(AARCH64_QUINT8_K8X8X8)
};

}  // namespace aarch64
}  // namespace megdnn

// vim: syntax=cpp.doxygen