diff options
author | Mohammed Suhail Munshi <MohammedSuhail.Munshi@arm.com> | 2023-01-25 11:51:50 +0000 |
---|---|---|
committer | Mohmun02 <MohammedSuhail.Munshi@arm.com> | 2023-02-01 17:10:42 +0000 |
commit | 5b9d22353d29bb3b4e6c53924564a42a6ab71050 (patch) | |
tree | 734c5260925b395a73bbbbaa7d7ab4fac392bac8 /src/cpu/kernels/CpuGemmLowpOffsetContributionOutputStageKernel.h | |
parent | ae72a46e495742863dba44fcf5fdc673c9d2afbc (diff) | |
download | ComputeLibrary-5b9d22353d29bb3b4e6c53924564a42a6ab71050.tar.gz |
Fix GEMMLowp/Batched MatMul mismatches on CPU
- Fixes Column Offset matrix is not being iterated through in y dimension
Resolves : COMPMID-5795
Signed-off-by: Mohammed Suhail Munshi <MohammedSuhail.Munshi@arm.com>
Change-Id: I0190474be404b4f0e171855739cfd0a48cbed5bc
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9020
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
Reviewed-by: SiCong Li <sicong.li@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/cpu/kernels/CpuGemmLowpOffsetContributionOutputStageKernel.h')
-rw-r--r-- | src/cpu/kernels/CpuGemmLowpOffsetContributionOutputStageKernel.h | 4 |
1 files changed, 3 insertions, 1 deletions
diff --git a/src/cpu/kernels/CpuGemmLowpOffsetContributionOutputStageKernel.h b/src/cpu/kernels/CpuGemmLowpOffsetContributionOutputStageKernel.h index ad8b05e49a..3cb99faee8 100644 --- a/src/cpu/kernels/CpuGemmLowpOffsetContributionOutputStageKernel.h +++ b/src/cpu/kernels/CpuGemmLowpOffsetContributionOutputStageKernel.h @@ -73,8 +73,10 @@ public: * * @param[in] mm_result Input tensor info containing the result of @ref CpuGemmLowpMatrixMultiplyKernel. Data type supported: S32 * @param[in] vector_sum_col Input row-vector tensor info of sums of all the entries in each column of matrix B. + * Can be a 1D or 2D tensor, in case of 2D, y dim is the batch dimension * Note: vector_sum_col can be a nullptr in case a_offset = 0. Data type supported: same as @p mm_result * @param[in] vector_sum_row Input row-vector tensor info of sums of all the entries in each row of matrix A. + * Can be a 1D or 2D tensor, in case of 2D, y dim is the batch dimension * @param[in] bias Biases tensor info. Only shared biases supported and it can be a nullptr if the addition of biases is not required. * Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p mm_result. * @param[out] dst Output tensor info containing the final quantized result. Data type supported: QASYMM8/QASYMM8_SIGNED @@ -105,7 +107,7 @@ private: int32_t _a_offset{ 0 }; int32_t _b_offset{ 0 }; int32_t _k_offset{ 0 }; - bool _slide_vector_sum_col{ true }; + bool _is_vector_sum_col_batched{ true }; GEMMLowpOutputStageInfo _output_stage{ GEMMLowpOutputStageInfo() }; }; } // namespace kernels |