diff options
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/kernels')
-rw-r--r-- | src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp32_mla_16x4.hpp | 18 | ||||
-rw-r--r-- | src/core/NEON/kernels/arm_gemm/kernels/a64_sgemm_12x8.hpp | 16 |
2 files changed, 33 insertions, 1 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp32_mla_16x4.hpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp32_mla_16x4.hpp index 8d8ede8137..4147ab60dc 100644 --- a/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp32_mla_16x4.hpp +++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp32_mla_16x4.hpp @@ -25,7 +25,7 @@ #ifdef __aarch64__ - +#include "../performance_parameters.hpp" #include "../std_transforms_fixed.hpp" namespace arm_gemm @@ -75,6 +75,22 @@ public: return true; } + static PerformanceParameters get_performance_parameters(const CPUInfo *ci) { + switch (ci->get_cpu_model()) { + case CPUModel::A55r1: + return { 2.866 }; + + case CPUModel::A53: + return { 1.419 }; + + case CPUModel::A73: + return { 2.551 }; + + default: + return { 6.25 }; + } + } + StdTransformsFixed<operand_type, result_type, 4, 16, 1> transforms = {}; // Default to the generic kernel diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_sgemm_12x8.hpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_sgemm_12x8.hpp index 5c3d6409b9..981ce34b49 100644 --- a/src/core/NEON/kernels/arm_gemm/kernels/a64_sgemm_12x8.hpp +++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_sgemm_12x8.hpp @@ -67,6 +67,22 @@ public: // Use the standard fixed size transforms. StdTransformsFixed<operand_type, result_type, 8, 12> transforms = {}; + static PerformanceParameters get_performance_parameters(const CPUInfo *ci) { + switch (ci->get_cpu_model()) { + case CPUModel::A55r1: + return { 3.724, 1.416, 1.113 }; + + case CPUModel::A53: + return { 2.777, 0.987, 0.898 }; + + case CPUModel::A73: + return { 2.885, 1.429, 1.163 }; + + default: + return { 6.949, 4.149, 2.826 }; + } + } + kern_type kernel=a64_sgemm_asimd_12x8; sgemm_12x8(const CPUInfo *ci) { |