aboutsummaryrefslogtreecommitdiff
path: root/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s8_8x12.hpp
diff options
context:
space:
mode:
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s8_8x12.hpp')
-rw-r--r--src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s8_8x12.hpp35
1 files changed, 30 insertions, 5 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s8_8x12.hpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s8_8x12.hpp
index 7c7b894b08..9af1b4df12 100644
--- a/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s8_8x12.hpp
+++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s8_8x12.hpp
@@ -61,13 +61,38 @@ public:
StdTransformsFixed<operand_type, result_type, 8, 12, 4> transforms = {};
StdTransformsFixed<operand_type, result_type, 8, 12, 4, true> transforms_quantized = {};
+ template<typename T>
static PerformanceParameters get_performance_parameters(const CPUInfo *ci) {
- switch (ci->get_cpu_model()) {
- case CPUModel::A55r1:
- return { 15.361, 0.9341, 0.1636 };
+ if (std::is_same<T, int8_t>::value) {
+ switch (ci->get_cpu_model()) {
+ case CPUModel::A510:
+ return { 19.73, 3.38, 0.27 };
- default:
- return { 29.0698, 3.9793, 0.4003 };
+ case CPUModel::A55r1:
+ return { 15.361, 0.9341, 0.1636 };
+
+ case CPUModel::V1:
+ return { 51.14, 7.38, 0.65 };
+
+ default:
+ return { 29.0698, 3.9793, 0.4003 };
+ }
+ }
+
+ if (std::is_same<T, int32_t>::value) {
+ switch (ci->get_cpu_model()) {
+ case CPUModel::A510:
+ return { 19.73, 3.38, 3.70 };
+
+ case CPUModel::A55r1:
+ return { 14.286, 1.171, 1.209 };
+
+ case CPUModel::V1:
+ return { 61.58, 4.78, 10.83 };
+
+ default:
+ return { 31.82, 3.51, 8.03 };
+ }
}
}