aboutsummaryrefslogtreecommitdiff
path: root/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_u8_4x4.hpp
diff options
context:
space:
mode:
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_u8_4x4.hpp')
-rw-r--r--src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_u8_4x4.hpp32
1 files changed, 31 insertions, 1 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_u8_4x4.hpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_u8_4x4.hpp
index 854b6751c1..b747a1cf84 100644
--- a/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_u8_4x4.hpp
+++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_u8_4x4.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2017-2020 Arm Limited.
+ * Copyright (c) 2017-2022 Arm Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -25,6 +25,7 @@
#ifdef __aarch64__
+#include "../performance_parameters.hpp"
#include "../std_transforms_fixed.hpp"
namespace arm_gemm {
@@ -66,6 +67,35 @@ public:
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, uint32_t>::value) {
+ switch (ci->get_cpu_model()) {
+ case CPUModel::A55r0:
+ case CPUModel::A55r1:
+ return { 2.25, 2.92, 1.84 };
+ case CPUModel::A510:
+ return { 2.64, 2.72, 2.64 };
+ default:
+ return { 7.95, 3.76, 7.27 };
+ }
+ }
+
+ if (std::is_same<T, uint8_t>::value) {
+ switch(ci->get_cpu_model()) {
+ case CPUModel::A55r0:
+ case CPUModel::A55r1:
+ return { 2.25, 2.18, 0.09 };
+ case CPUModel::A510:
+ return { 2.64, 1.79, 0.10 };
+ default:
+ return { 7.95, 4.09, 0.33 };
+ }
+ }
+
+ return { 0.0 };
+ }
+
kern_type kernel = a64_gemm_u8_4x4;
cls_a64_gemm_u8_4x4(const CPUInfo *) { }