xref: /aosp_15_r20/external/ComputeLibrary/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp32_mla_8x4.hpp (revision c217d954acce2dbc11938adb493fc0abd69584f3)
1 /*
2  * Copyright (c) 2019-2021 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
14  * all 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
21  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
22  * IN THE SOFTWARE.
23  */
24 #pragma once
25 
26 #ifdef __aarch64__
27 #include "../std_transforms_fixed.hpp"
28 
29 #define ARGLIST  \
30     unsigned int, const unsigned int *, \
31     IndirectInputArg<float>, \
32     size_t, size_t, \
33     const float *, \
34     IndirectOutputArg<float>, \
35     const float *, Activation, bool
36 
37 namespace arm_gemm
38 {
39 // Actual kernel implementations
40 void a64_hybrid_fp32_mla_8x4( ARGLIST );
41 void a64_hybrid_fp32_mla_8x4_a55( ARGLIST );
42 
43 class cls_a64_hybrid_fp32_mla_8x4
44 {
45 public:
46     typedef float lhs_operand_type;
47     typedef float rhs_operand_type;
48     typedef float result_type;
49 
50     typedef void (*kern_type)( ARGLIST );
51 
52     /* Kernel blocking parameters */
out_height()53     static constexpr unsigned int out_height()
54     {
55         return 8;
56     }
57 
out_width()58     static unsigned int out_width()
59     {
60         return 4;
61     }
62 
k_unroll()63     static constexpr unsigned int k_unroll()
64     {
65         return 1;
66     }
67 
supports_accumulate()68     static constexpr bool supports_accumulate()
69     {
70         return true;
71     }
72 
73     StdTransformsFixed<rhs_operand_type, result_type, 8, 4, 1> transforms = {};
74 
75     // Default to the generic kernel
76     kern_type kernel=a64_hybrid_fp32_mla_8x4;
cls_a64_hybrid_fp32_mla_8x4(const CPUInfo * ci)77     cls_a64_hybrid_fp32_mla_8x4(const CPUInfo *ci)
78     {
79         switch(ci->get_cpu_model()) {
80             default:
81                 break;
82             case CPUModel::A55r1:
83             case CPUModel::A53:
84                 kernel=a64_hybrid_fp32_mla_8x4_a55;
85                 break;
86         }
87     }
88 };
89 
90 } // namespace arm_gemm
91 
92 #undef ARGLIST
93 
94 #endif // __aarch64__
95