aboutsummaryrefslogtreecommitdiff
path: root/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_s8s32_dot_6x16.hpp
diff options
context:
space:
mode:
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_s8s32_dot_6x16.hpp')
-rw-r--r--src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_s8s32_dot_6x16.hpp41
1 files changed, 31 insertions, 10 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_s8s32_dot_6x16.hpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_s8s32_dot_6x16.hpp
index 759a78a413..d91c69b8a0 100644
--- a/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_s8s32_dot_6x16.hpp
+++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_s8s32_dot_6x16.hpp
@@ -22,8 +22,8 @@
* IN THE SOFTWARE.
*/
#pragma once
-#ifdef __aarch64__
+#ifdef __aarch64__
#include "../std_transforms_fixed.hpp"
#include "../performance_parameters.hpp"
@@ -44,7 +44,8 @@ void a64_hybrid_s8s32_dot_6x16_a55( ARGLIST );
class cls_a64_hybrid_s8s32_dot_6x16
{
public:
- typedef int8_t operand_type;
+ typedef int8_t lhs_operand_type;
+ typedef int8_t rhs_operand_type;
typedef int32_t result_type;
typedef void (*kern_type)( ARGLIST );
@@ -70,16 +71,35 @@ public:
return true;
}
- StdTransformsFixed<operand_type, result_type, 6, 16, 4> transforms = {};
-
- static PerformanceParameters get_performance_parameters(const CPUInfo *ci)
+ StdTransformsFixed<rhs_operand_type, result_type, 6, 16, 4> transforms = {};
+ template<typename T>
+ static inline PerformanceParameters get_performance_parameters(const CPUInfo *ci)
{
- switch (ci->get_cpu_model()) {
- case CPUModel::A55r1:
- return { 12.667, 2.0799, 0.2279 };
- default:
- return { 29.6736, 11.4025, 0.5591 };
+ if (std::is_same<T, int32_t>::value) {
+ switch (ci->get_cpu_model()) {
+ default:
+ return { 31.65 };
+ case CPUModel::A510:
+ return { 15.87 };
+ case CPUModel::V1:
+ return { 54.50 };
+ }
}
+
+ if (std::is_same<T, int8_t>::value) {
+ switch (ci->get_cpu_model()) {
+ case CPUModel::A55r1:
+ return { 9.5238, 2.0799, 0.2279 };
+ default:
+ return { 29.6736, 11.4025, 0.5591 };
+ case CPUModel::A510:
+ return { 16.66, 3.92, 0.48 };
+ case CPUModel::V1:
+ return { 55.40, 19.21, 0.93 };
+ }
+ }
+
+ return { 1.0 };
}
// Default to the generic kernel
@@ -99,4 +119,5 @@ public:
} // namespace arm_gemm
#undef ARGLIST
+
#endif // __aarch64__