xref: /aosp_15_r20/external/ComputeLibrary/src/core/NEON/kernels/arm_gemm/kernels/a64_sgemm_8x6.hpp (revision c217d954acce2dbc11938adb493fc0abd69584f3)
1 /*
2  * Copyright (c) 2017-2021 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 #pragma once
25 
26 #ifdef __aarch64__
27 
28 #include "../std_transforms_fixed.hpp"
29 
30 namespace arm_gemm {
31 
32 // Actual kernel implementations
33 void a64_sgemm_asimd_8x6(const float *, const float *, float *, int, int, int);
34 
35 // 8x6 SGEMM "strategy" class.
36 //
37 // This describes the characteristics of a family of kernels, in terms of
38 // the required interleave properties and the output block size.
39 //
40 // All kernels in the family must share these characteristics.  The actual
41 // kernel to be used can be chosen at runtime, based on the CPU_type
42 // structure.
43 class cls_a64_sgemm_8x6 {
44 public:
45     typedef float operand_type;
46     typedef float result_type;
47 
48     typedef void (*kern_type)(const float *, const float *, float *, int, int, int);
49 
50     /* Kernel blocking parameters */
out_width()51     static unsigned int out_width() {
52         return 6;
53     }
54 
out_height()55     static unsigned int out_height() {
56         return 8;
57     }
58 
k_unroll()59     static unsigned int k_unroll() {
60         return 1;
61     }
62 
63     // Use the standard fixed size transforms.
64     StdTransformsFixed<operand_type, result_type, 8, 6, 1> transforms = {};
65 
66     kern_type kernel=a64_sgemm_asimd_8x6;
67 
cls_a64_sgemm_8x6(const CPUInfo *)68     cls_a64_sgemm_8x6(const CPUInfo *) {
69 
70     }
71 };
72 
73 } // namespace arm_gemm
74 
75 #endif // __aarch64__
76