diff options
author | Georgios Pinitas <georgios.pinitas@arm.com> | 2020-11-17 18:46:40 +0000 |
---|---|---|
committer | Michele Di Giorgio <michele.digiorgio@arm.com> | 2020-11-23 13:50:40 +0000 |
commit | 573c7bbb15c8f205e93397105e66fbc5102882e4 (patch) | |
tree | 00f34dfe387f3a264e6eb74d0665b2190ab4d3bb /src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp16_mla_6x32.hpp | |
parent | 34c0c8a7d841111a0147ef84e29627f3e610d273 (diff) | |
download | ComputeLibrary-573c7bbb15c8f205e93397105e66fbc5102882e4.tar.gz |
Update tuning numbers for A55 for both fp16 and fp32
Resolves: COMPMID-3974
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Change-Id: I6d5189e44ebeda1575a80dd14ec3a09c75f68e03
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4521
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
(cherry picked from commit 40943df83026b66356f24e30f31f78c8b9e59c92)
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4169
Tested-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp16_mla_6x32.hpp')
-rw-r--r-- | src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp16_mla_6x32.hpp | 11 |
1 files changed, 11 insertions, 0 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp16_mla_6x32.hpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp16_mla_6x32.hpp index 46de98504e..876b63c811 100644 --- a/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp16_mla_6x32.hpp +++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp16_mla_6x32.hpp @@ -71,6 +71,17 @@ public: StdTransformsFixed<operand_type, result_type, 6, 32, 1> transforms = {}; + static PerformanceParameters get_performance_parameters(const CPUInfo *ci) + { + switch (ci->get_cpu_model()) { + case CPUModel::A55r1: + return { 5.22 }; + + default: + return { 14.53 }; + } + } + // Default to the generic kernel kern_type kernel=a64_hybrid_fp16_mla_6x32; |