aboutsummaryrefslogtreecommitdiff
path: root/src/core/NEON/kernels/arm_gemm/kernels/a64_interleaved_bf16fp32_mmla_8x12.hpp
diff options
context:
space:
mode:
authorramelg01 <ramy.elgammal@arm.com>2021-12-08 02:00:16 +0000
committerSheri Zhang <sheri.zhang@arm.com>2021-12-14 10:53:54 +0000
commitb6e49ebb5b83edbf45513978c0c77a8871c6b36c (patch)
tree411724910a6db7ae8d3451b457796ea6ec4a86a2 /src/core/NEON/kernels/arm_gemm/kernels/a64_interleaved_bf16fp32_mmla_8x12.hpp
parent6f3a9f5f4ef6ec7aa8e91df3c1f373d95931dd7b (diff)
downloadComputeLibrary-b6e49ebb5b83edbf45513978c0c77a8871c6b36c.tar.gz
Update A510 arm_gemm cpu Kernels
Resolves: COMPMID-4910 Signed-off-by: Ramy Elgammal <ramy.elgammal@arm.com> Change-Id: I79b4aa51e07ad1fe81d9218ed8a8f34f0ec5ab06 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/6803 Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Sheri Zhang <sheri.zhang@arm.com>
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/kernels/a64_interleaved_bf16fp32_mmla_8x12.hpp')
-rw-r--r--src/core/NEON/kernels/arm_gemm/kernels/a64_interleaved_bf16fp32_mmla_8x12.hpp12
1 files changed, 10 insertions, 2 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_interleaved_bf16fp32_mmla_8x12.hpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_interleaved_bf16fp32_mmla_8x12.hpp
index 9b3517a802..6ec6bd2ed8 100644
--- a/src/core/NEON/kernels/arm_gemm/kernels/a64_interleaved_bf16fp32_mmla_8x12.hpp
+++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_interleaved_bf16fp32_mmla_8x12.hpp
@@ -36,6 +36,7 @@ namespace arm_gemm
{
// Actual kernel implementations
void a64_interleaved_bf16fp32_mmla_8x12( ARGLIST );
+void a64_interleaved_bf16fp32_mmla_8x12_a510( ARGLIST );
class cls_a64_interleaved_bf16fp32_mmla_8x12
{
@@ -78,7 +79,7 @@ public:
default:
return { 31.54, 4.30, 7.33 };
case CPUModel::V1:
- return { 59.94, 5.08, 9.83 };
+ return { 41.44, 5.01, 5.64 };
case CPUModel::A510:
return { 7.82, 4.05, 3.07 };
}
@@ -101,8 +102,15 @@ public:
// Default to the generic kernel
kern_type kernel=a64_interleaved_bf16fp32_mmla_8x12;
- cls_a64_interleaved_bf16fp32_mmla_8x12(const CPUInfo *)
+ cls_a64_interleaved_bf16fp32_mmla_8x12(const CPUInfo *ci)
{
+ switch(ci->get_cpu_model()) {
+ default:
+ break;
+ case CPUModel::A510:
+ kernel=a64_interleaved_bf16fp32_mmla_8x12_a510;
+ break;
+ }
}
};