diff options
| author | giuros01 <giuseppe.rossini@arm.com> | 2019-10-01 12:17:49 +0100 |
|---|---|---|
| committer | Giuseppe Rossini <giuseppe.rossini@arm.com> | 2019-10-01 14:01:22 +0000 |
| commit | 3d8fe497fc76ec6ad265c03fe02e29ed2ddf2d93 (patch) | |
| tree | 1b5274641d01819b7f59c1834df9cfffb81629d6 /arm_compute | |
| parent | 2aa7fd011a4baff52dceb00a71b3674f819df8fc (diff) | |
| download | ComputeLibrary-3d8fe497fc76ec6ad265c03fe02e29ed2ddf2d93.tar.gz | |
INFPRF-609:Performance Issue of the Latest ArmCL
We were creating too many small GEMM workloads. It affects performance when
the number of thread is small and the matrices are bigger (especially
when more single-threaded process are running on the same machine)
Change-Id: I807019a7b2d043ca72b4bca11eb0b1960da00694
Signed-off-by: giuros01 <giuseppe.rossini@arm.com>
Reviewed-on: https://review.mlplatform.org/c/2012
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'arm_compute')
| -rw-r--r-- | arm_compute/runtime/NEON/functions/assembly/NEGEMMInterleavedWrapper.h | 1 |
1 files changed, 0 insertions, 1 deletions
diff --git a/arm_compute/runtime/NEON/functions/assembly/NEGEMMInterleavedWrapper.h b/arm_compute/runtime/NEON/functions/assembly/NEGEMMInterleavedWrapper.h index d3dda9a95f..eeea0babf1 100644 --- a/arm_compute/runtime/NEON/functions/assembly/NEGEMMInterleavedWrapper.h +++ b/arm_compute/runtime/NEON/functions/assembly/NEGEMMInterleavedWrapper.h @@ -141,7 +141,6 @@ private: std::vector<MatrixMultiplyWorkload> _mm_workloads{}; std::vector<IScheduler::Workload> _workloads{}; std::string _tag{}; - unsigned int _num_windows{ 1 }; }; } // namespace arm_compute #endif /* __ARM_COMPUTE_NEGEMMINTERLEAVEDWRAPPER_H__ */ |