diff options
author | Georgios Pinitas <georgios.pinitas@arm.com> | 2021-07-16 16:16:43 +0100 |
---|---|---|
committer | Georgios Pinitas <georgios.pinitas@arm.com> | 2021-07-22 02:25:50 +0000 |
commit | 4ee8b1599dbaf7634d25607fa5ac96ba3dc6b0f2 (patch) | |
tree | 2f8362d33cdad4212f4b96995681c68184c759e1 /src/runtime/cpu/operators/CpuGemmLowpMatrixMultiplyCore.cpp | |
parent | 59fd7a722e5bc7e85309d6200bc37a772721a719 (diff) | |
download | ComputeLibrary-4ee8b1599dbaf7634d25607fa5ac96ba3dc6b0f2.tar.gz |
Update GEMM assembly kernels
- Introduce Fp32 kernels with internal calculations in Bfloat16 when
fast_mode is enabled
- Improve kernel selection heuristics
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Change-Id: I68a9e7e862b6fd2721b46e0d7cc791091c4ab279
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5965
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/runtime/cpu/operators/CpuGemmLowpMatrixMultiplyCore.cpp')
-rw-r--r-- | src/runtime/cpu/operators/CpuGemmLowpMatrixMultiplyCore.cpp | 1 |
1 files changed, 1 insertions, 0 deletions
diff --git a/src/runtime/cpu/operators/CpuGemmLowpMatrixMultiplyCore.cpp b/src/runtime/cpu/operators/CpuGemmLowpMatrixMultiplyCore.cpp index 651ce436a0..56eb4fbb87 100644 --- a/src/runtime/cpu/operators/CpuGemmLowpMatrixMultiplyCore.cpp +++ b/src/runtime/cpu/operators/CpuGemmLowpMatrixMultiplyCore.cpp @@ -63,6 +63,7 @@ cpu::AsmGemmInfo init_assembly_metadata(const GEMMInfo &info) asm_info.depth_output_gemm3d = info.depth_output_gemm3d(); asm_info.activation_info = info.activation_info(); asm_info.output_stage = info.gemmlowp_output_stage(); + asm_info.fast_mode = info.fast_math(); return asm_info; } |