diff options
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s8_4x4.hpp')
| -rw-r--r-- | src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s8_4x4.hpp | 35 |
1 files changed, 32 insertions, 3 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s8_4x4.hpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s8_4x4.hpp index fb21bfc863..1363b939ab 100644 --- a/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s8_4x4.hpp +++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s8_4x4.hpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2017-2018 ARM Limited. + * Copyright (c) 2017-2022 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -26,6 +26,7 @@ #ifdef __aarch64__ #include "../std_transforms_fixed.hpp" +#include "../performance_parameters.hpp" namespace arm_gemm { @@ -34,7 +35,7 @@ void a64_gemm_s8_4x4(const int8_t *, const int8_t *, int32_t *, int, int, int); #include "arm_gemm.hpp" -class gemm_s8_4x4 { +class cls_a64_gemm_s8_4x4 { public: typedef int8_t operand_type; typedef int32_t result_type; @@ -56,10 +57,38 @@ public: // Use the standard fixed size transforms. StdTransformsFixed<operand_type, result_type, 4, 4, 16> transforms = {}; + StdTransformsFixed<operand_type, result_type, 4, 4, 16, true> transforms_quantized = {}; + + template<typename T> + static PerformanceParameters get_performance_parameters(const CPUInfo *ci) { + if (std::is_same<T, int32_t>::value) { + switch (ci->get_cpu_model()) { + case CPUModel::A55r0: + case CPUModel::A55r1: + return { 3.12, 2.93, 1.84 }; + case CPUModel::A510: + return { 3.32, 2.56, 2.63 }; + default: + return { 7.97, 3.72, 7.31 }; + } + } + + if (std::is_same<T, int8_t>::value) { + switch(ci->get_cpu_model()) { + case CPUModel::A55r0: + case CPUModel::A55r1: + return { 3.12, 2.18, 0.09 }; + case CPUModel::A510: + return { 3.33, 2.89, 0.09 }; + default: + return { 7.97, 3.74, 0.34 }; + } + } + } kern_type kernel=a64_gemm_s8_4x4; - gemm_s8_4x4(const CPUInfo *ci) { UNUSED(ci); } + cls_a64_gemm_s8_4x4(const CPUInfo *) { } }; } // namespace arm_gemm |