Implement 1D Adaptive Workload Splitting in CPPScheduler
Resolves COMPMID-4649
Change-Id: I941d2f8a40737ff05c49f6695a42884731ef2dc9
Signed-off-by: Dana Zlotnik <dana.zlotnik@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/6656
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: SiCong Li <sicong.li@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
diff --git a/src/cpu/kernels/CpuAddKernel.cpp b/src/cpu/kernels/CpuAddKernel.cpp
index 77aa1ff..73c1fda 100644
--- a/src/cpu/kernels/CpuAddKernel.cpp
+++ b/src/cpu/kernels/CpuAddKernel.cpp
@@ -294,9 +294,16 @@
size_t CpuAddKernel::get_mws(const CPUInfo &platform, size_t thread_count) const
{
- ARM_COMPUTE_UNUSED(platform, thread_count);
-
- return ICPPKernel::small_network_mws;
+ ARM_COMPUTE_UNUSED(thread_count);
+ // Tuning results that gave optimized results in performance investigation
+ if (platform.get_cpu_model() == CPUModel::A73 )
+ {
+ return 10240;
+ }
+ else
+ {
+ return 9216;
+ }
}
} // namespace kernels