aboutsummaryrefslogtreecommitdiff
path: root/src/core/NEON/kernels/arm_gemm/kernels/a64_interleaved_s8s32_mmla_8x12.hpp
diff options
context:
space:
mode:
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/kernels/a64_interleaved_s8s32_mmla_8x12.hpp')
-rw-r--r--src/core/NEON/kernels/arm_gemm/kernels/a64_interleaved_s8s32_mmla_8x12.hpp13
1 files changed, 11 insertions, 2 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_interleaved_s8s32_mmla_8x12.hpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_interleaved_s8s32_mmla_8x12.hpp
index ff69bc8f53..4cc3ed040a 100644
--- a/src/core/NEON/kernels/arm_gemm/kernels/a64_interleaved_s8s32_mmla_8x12.hpp
+++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_interleaved_s8s32_mmla_8x12.hpp
@@ -35,6 +35,7 @@ namespace arm_gemm
{
// Actual kernel implementations
void a64_interleaved_s8s32_mmla_8x12( ARGLIST );
+void a64_interleaved_s8s32_mmla_8x12_a510( ARGLIST );
class cls_a64_interleaved_s8s32_mmla_8x12
{
@@ -91,7 +92,7 @@ public:
case CPUModel::A510:
return { 48.22, 2.49, 0.29 };
case CPUModel::V1:
- return { 116.76, 4.67, 0.60 };
+ return { 75.54, 8.06, 0.63 };
}
}
@@ -100,9 +101,17 @@ public:
// Default to the generic kernel
kern_type kernel=a64_interleaved_s8s32_mmla_8x12;
- cls_a64_interleaved_s8s32_mmla_8x12(const CPUInfo *)
+ cls_a64_interleaved_s8s32_mmla_8x12(const CPUInfo *ci)
{
+ switch(ci->get_cpu_model()) {
+ default:
+ break;
+ case CPUModel::A510:
+ kernel=a64_interleaved_s8s32_mmla_8x12_a510;
+ break;
+ }
}
+
};
} // namespace arm_gemm