2 * Copyright (c) 2017-2018 ARM Limited.
4 * SPDX-License-Identifier: MIT
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:
13 * The above copyright notice and this permission notice shall be included in all
14 * copies or substantial portions of the Software.
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
26 #if defined(__aarch64__) && (defined(FP16_KERNELS) || defined(__ARM_FEATURE_FP16_VECTOR_ARITHMETIC))
28 #include "arm_gemm.hpp"
32 // Actual kernel implementations
33 void a64_hgemm_asimd_24x8(const __fp16 *, const __fp16 *, __fp16 *, int, int, int);
34 void a64_hgemm_asimd_24x8_a55r1(const __fp16 *, const __fp16 *, __fp16 *, int, int, int);
36 // 24x8 HGEMM "strategy" class. Describes the kernel properties.
38 // The generic "gemm_opt" function will instantiate one of these (allowing
39 // the constructor to pick a kernel implementation).
43 typedef __fp16 operand_type;
44 typedef __fp16 result_type;
46 typedef void (*kern_type)(const __fp16 *, const __fp16 *, __fp16 *, int, int, int);
48 static const int A_block = 1;
49 static const int A_interleave = 8;
50 static const bool A_transpose = false;
52 static const int B_block = 1;
53 static const int B_interleave = 24;
54 static const bool B_transpose = true;
56 static const int out_width = 24;
57 static const int out_height = 8;
58 static const int k_unroll = 1;
60 // Default to the generic kernel
61 kern_type kernel = a64_hgemm_asimd_24x8;
63 hgemm_24x8(const CPUInfo *ci)
65 if(ci->get_cpu_model() == CPUModel::A55r1)
67 kernel = a64_hgemm_asimd_24x8_a55r1;
72 } // namespace arm_gemm
74 #endif // __aarch64__ && (FP16_KERNELS || __ARM_FEATURE_FP16_VECTOR_ARITHMETIC)