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
|
/*
* Copyright (c) 2017-2020 Arm Limited.
*
* SPDX-License-Identifier: MIT
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to
* deal in the Software without restriction, including without limitation the
* rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
* sell copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in all
* copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*/
// This can only be built if the target/compiler supports FP16 arguments.
#ifdef __ARM_FP16_ARGS
#include "arm_gemm.hpp"
#include "gemm_common.hpp"
#include "gemm_hybrid.hpp"
#include "gemm_implementation.hpp"
#include "gemm_interleaved.hpp"
#include "kernels/a32_sgemm_8x6.hpp"
#include "kernels/a64_hgemm_24x8.hpp"
#include "kernels/a64_sgemm_12x8.hpp"
#include "kernels/sve_hybrid_fp16_mla_4VLx4.hpp"
#include "kernels/sve_interleaved_fp16_mla_3VLx8.hpp"
namespace arm_gemm {
static const GemmImplementation<__fp16, __fp16> gemm_fp16_methods[] = {
#if defined(__ARM_FEATURE_SVE)
{
GemmMethod::GEMM_HYBRID,
"hybrid_fp16_mla_4VLx4",
[](const GemmArgs &args) { return (args._Ksize >= 8); },
[](const GemmArgs &args) { return ((args._Ksize <= 256) && (args._Nsize <= 256)) || ((args._nmulti > 1) && ((args._Msize / args._maxthreads) < 8)); },
[](const GemmArgs &args) { return new GemmHybrid<hybrid_fp16_mla_4VLx4, __fp16, __fp16>(args); }
},
{
GemmMethod::GEMM_INTERLEAVED,
"interleaved_fp16_mla_3VLx8",
[](const GemmArgs &args) { return (args._Ksize > 4); },
nullptr,
[](const GemmArgs &args) { return new GemmInterleaved<interleaved_fp16_mla_3VLx8, __fp16, __fp16>(args); }
},
#endif
#if defined(__aarch64__) && (defined(__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) || defined(FP16_KERNELS))
{
GemmMethod::GEMM_INTERLEAVED,
"hgemm_24x8",
#ifndef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
[](const GemmArgs &args) { return args._ci->has_fp16(); },
#else
nullptr,
#endif
nullptr,
[](const GemmArgs &args) { return new GemmInterleaved<hgemm_24x8, __fp16, __fp16>(args); }
},
#endif // aarch64 && FP16
#ifdef __aarch64__
{
GemmMethod::GEMM_INTERLEAVED,
"sgemm_12x8",
nullptr,
nullptr,
[](const GemmArgs &args) { return new GemmInterleaved<sgemm_12x8, __fp16, __fp16>(args); }
},
#elif defined(__arm__)
{
GemmMethod::GEMM_INTERLEAVED,
"sgemm_8x6",
nullptr,
nullptr,
[](const GemmArgs &args) { return new GemmInterleaved<sgemm_8x6, __fp16, __fp16>(args); }
},
#else // not AArch64 or AArch32
# error Unknown Architecture
#endif
{
GemmMethod::DEFAULT,
"",
nullptr,
nullptr,
nullptr,
}
};
template<>
const GemmImplementation<__fp16, __fp16> *gemm_implementation_list<__fp16, __fp16>() {
return gemm_fp16_methods;
}
/* Explicitly instantiate the external functions for these types. */
template UniqueGemmCommon<__fp16, __fp16> gemm<__fp16, __fp16, Nothing>(const GemmArgs &args, const Nothing &);
template KernelDescription get_gemm_method<__fp16, __fp16, Nothing>(const GemmArgs &args, const Nothing &);
template std::vector<KernelDescription> get_compatible_kernels<__fp16, __fp16, Nothing>(const GemmArgs &args, const Nothing &);
} // namespace arm_gemm
#endif // __ARM_FP16_ARGS
|
