blob: 1baa21fd1b01d8df49851dae43d83b71ba713164 [file] [log] [blame]
Pablo Telloeb82fd22018-02-23 13:43:50 +00001/*
2 * Copyright (c) 2017-2018 ARM Limited.
3 *
4 * SPDX-License-Identifier: MIT
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to
8 * deal in the Software without restriction, including without limitation the
9 * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
10 * sell copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in all
14 * copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
19 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22 * SOFTWARE.
23 */
24#include "arm_gemm.hpp"
25#include "gemm_common.hpp"
26#include "gemm_interleaved.hpp"
27#include "gemm_native.hpp"
28#include "gemv_native_transposed.hpp"
29#include "gemv_pretransposed.hpp"
30
31#include "kernels/a32_sgemm_8x6.hpp"
32#include "kernels/a64_sgemm_12x8.hpp"
33#include "kernels/a64_sgemm_native_16x4.hpp"
34#include "kernels/a64_sgemv_pretransposed.hpp"
35#include "kernels/a64_sgemv_trans.hpp"
36
37namespace arm_gemm
38{
39template <>
40UniqueGemmCommon<float, float> gemm<float, float>(const CPUInfo &ci, const unsigned int M, const unsigned int N, const unsigned int K,
41 const bool trA, const bool trB, const float alpha, const float beta,
42 const int maxthreads, const bool pretransposed_hint)
43{
44#ifdef __aarch64__
45 /* Cases in priority order */
46 /* GemvPretransposed: requires M=1, alpha=1, and transposed hint set */
47 if(M == 1 && alpha == 1.0f && pretransposed_hint)
48 {
49 return UniqueGemmCommon<float, float>(new GemvPretransposed<sgemv_pretransposed, float, float>(&ci, N, K, trB, beta));
50 }
51
52 /* GemvNativeTransposed: requires M=1, no trA or trB, doesn't handle beta */
53 if(M == 1 && beta == 1.0f && !trA && !trB)
54 {
55 return UniqueGemmCommon<float, float>(new GemvNativeTransposed<sgemv_trans, float, float>(&ci, N, K, alpha));
56 }
57
58 /* Native GEMM: requires M to be a multiple of 4, K a multiple of 4, N a
59 * multiple of 16, doesn't handle alpha and only makes sense for small
60 * sizes. */
61 if(N <= 128 && K <= 128 && ((M % 4) == 0) && ((K % 4) == 0) && ((N % 16) == 0) && alpha == 1.0f)
62 {
63 return UniqueGemmCommon<float, float>(new GemmNative<sgemm_native_16x4, float, float>(&ci, M, N, K, beta));
64 }
65
66 /* Blocked GEMM, handles all cases. */
67 return UniqueGemmCommon<float, float>(new GemmInterleaved<sgemm_12x8, float, float>(&ci, M, N, K, trA, trB, alpha, beta, maxthreads, pretransposed_hint));
68#else
69 return UniqueGemmCommon<float, float>(new GemmInterleaved<sgemm_8x6, float, float>(&ci, M, N, K, trA, trB, alpha, beta, maxthreads, pretransposed_hint));
70#endif
71}
72
73// Instantiate static class variables.
74#ifdef __aarch64__
75const int sgemm_12x8::out_width;
76const int sgemm_12x8::out_height;
77
78const int sgemm_native_16x4::out_width;
79const int sgemm_native_16x4::out_height;
80#else
81const int sgemm_8x6::out_width;
82const int sgemm_8x6::out_height;
83#endif
84
85} // namespace arm_gemm