diff options
| author | Radu Salavat <radu.salavat@arm.com> | 2024-02-27 18:32:26 +0000 |
|---|---|---|
| committer | Radu Salavat <radu.salavat@arm.com> | 2024-04-11 08:47:50 +0000 |
| commit | f1f1f87132690a8061801ef1a4638d637c780df7 (patch) | |
| tree | 8ad4c3739217b3bc6281f4e0b9a7a63fe6c3f9bb /src/cpu/kernels/assembly/arm_gemm.hpp | |
| parent | 1322065a3fbd15b00dbfb0969d6b438b5ba15530 (diff) | |
| download | ComputeLibrary-f1f1f87132690a8061801ef1a4638d637c780df7.tar.gz | |
Add in place summation to CPU GEMM kernels
Instead of dispatching the sum postop for GEMM kernels to a
separate kernel + add, that requires an extra destination sized
allocation, plus 3 extra load/stores per element,
just do it in the GEMM kernel.
Resolves: ONCPUML-1442
Signed-off-by: Radu Salavat <radu.salavat@arm.com>
Co-authored-by: Milos Puzovic <milos.puzovic@arm.com>
Change-Id: I7a1f2da3300875fa1ac88b705a34390969518077
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/11298
Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/cpu/kernels/assembly/arm_gemm.hpp')
| -rw-r--r-- | src/cpu/kernels/assembly/arm_gemm.hpp | 11 |
1 files changed, 10 insertions, 1 deletions
diff --git a/src/cpu/kernels/assembly/arm_gemm.hpp b/src/cpu/kernels/assembly/arm_gemm.hpp index 9a913c5c58..5d7cf79857 100644 --- a/src/cpu/kernels/assembly/arm_gemm.hpp +++ b/src/cpu/kernels/assembly/arm_gemm.hpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2018-2022 Arm Limited. + * Copyright (c) 2018-2022, 2024 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -21,6 +21,10 @@ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE * SOFTWARE. */ + +#ifndef ACL_SRC_CPU_KERNELS_ASSEMBLY_ARM_GEMM_HPP +#define ACL_SRC_CPU_KERNELS_ASSEMBLY_ARM_GEMM_HPP + #pragma once #include "arm_gemm_local.hpp" @@ -151,6 +155,7 @@ public: int _maxthreads; bool _fixed_format; bool _fast_mode; + bool _accumulate; const GemmConfig *_cfg; GemmArgs(const CPUInfo *ci, @@ -165,6 +170,7 @@ public: const int maxthreads, bool fixed_format = false, bool fast_mode = false, + bool accumulate = false, const GemmConfig *cfg = nullptr) : _ci(ci), _Msize(M), @@ -178,6 +184,7 @@ public: _maxthreads(maxthreads), _fixed_format(fixed_format), _fast_mode(fast_mode), + _accumulate(accumulate), _cfg(cfg) { } @@ -278,3 +285,5 @@ template <typename Top, typename Tret, class OutputStage = Nothing> bool has_opt_gemm(WeightFormat &weight_format, const GemmArgs &args, const OutputStage & = {}); } // namespace arm_gemm + +#endif // ACL_SRC_CPU_KERNELS_ASSEMBLY_ARM_GEMM_HPP |