Pablo Tello | eb82fd2 | 2018-02-23 13:43:50 +0000 | [diff] [blame] | 1 | /* |
| 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 | #ifdef __ARM_FEATURE_FP16_SCALAR_ARITHMETIC |
| 25 | |
| 26 | #include "arm_gemm.hpp" |
| 27 | |
| 28 | #include "gemm_common.hpp" |
| 29 | #include "gemm_interleaved.hpp" |
| 30 | |
| 31 | #include "kernels/a32_sgemm_8x6.hpp" |
| 32 | #include "kernels/a64_hgemm_24x8.hpp" |
| 33 | #include "kernels/a64_sgemm_12x8.hpp" |
| 34 | |
| 35 | namespace arm_gemm |
| 36 | { |
| 37 | template <> |
| 38 | UniqueGemmCommon<__fp16, __fp16> gemm(const CPUInfo &ci, const unsigned int M, const unsigned int N, const unsigned int K, |
| 39 | const bool trA, const bool trB, const __fp16 alpha, const __fp16 beta, |
| 40 | const int maxthreads, const bool pretransposed_hint) |
| 41 | { |
| 42 | #ifdef __aarch64__ |
| 43 | /* If FP16 is supported, use it */ |
| 44 | if(ci.has_fp16()) |
| 45 | { |
| 46 | return UniqueGemmCommon<__fp16, __fp16>(new GemmInterleaved<hgemm_24x8, __fp16, __fp16>(&ci, M, N, K, trA, trB, alpha, beta, maxthreads, pretransposed_hint)); |
| 47 | } |
| 48 | |
| 49 | /* Fallback to using the blocked SGEMM kernel. */ |
| 50 | return UniqueGemmCommon<__fp16, __fp16>(new GemmInterleaved<sgemm_12x8, __fp16, __fp16>(&ci, M, N, K, trA, trB, alpha, beta, maxthreads, pretransposed_hint)); |
| 51 | #else |
| 52 | /* For AArch32, only support the SGEMM route. */ |
| 53 | return UniqueGemmCommon<__fp16, __fp16>(new GemmInterleaved<sgemm_8x6, __fp16, __fp16>(&ci, M, N, K, trA, trB, alpha, beta, maxthreads, pretransposed_hint)); |
| 54 | #endif |
| 55 | } |
| 56 | |
| 57 | // Instantiate static class members |
| 58 | #ifdef __aarch64__ |
| 59 | const int hgemm_24x8::out_width; |
| 60 | const int hgemm_24x8::out_height; |
| 61 | #endif |
| 62 | |
| 63 | } // namespace arm_gemm |
| 64 | |
| 65 | #endif // __ARM_FEATURE_FP16_SCALAR_ARITHMETIC |