diff options
| -rw-r--r-- | src/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedOnlyRHSKernel.cpp | 2 | ||||
| -rw-r--r-- | src/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.cpp | 6 |
2 files changed, 7 insertions, 1 deletions
diff --git a/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedOnlyRHSKernel.cpp b/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedOnlyRHSKernel.cpp index d39900a561..37c11000db 100644 --- a/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedOnlyRHSKernel.cpp +++ b/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedOnlyRHSKernel.cpp @@ -256,7 +256,7 @@ std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input0, ITe window_changed = window_changed || update_window_and_padding(win_out, bias_access); } - if(output_multipliers != nullptr && output_multipliers->dimension(0) > 1) + if(output_multipliers != nullptr && output_stage.is_quantized_per_channel) { AccessWindowHorizontal output_multipliers_access(output_multipliers, 0, num_elems_processed_per_iteration_x); AccessWindowHorizontal output_shifts_access(output_shifts, 0, num_elems_processed_per_iteration_x); diff --git a/src/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.cpp b/src/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.cpp index b189955c04..5a9ff7990f 100644 --- a/src/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.cpp +++ b/src/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.cpp @@ -344,6 +344,12 @@ void CLGEMMLowpMatrixMultiplyCore::configure(const CLCompileContext &compile_con GEMMLowpOutputStageInfo gemmlowp_output_stage = gemm_info.gemmlowp_output_stage(); gemmlowp_output_stage.output_data_type = _matrix_a->info()->data_type(); + if(num_filters == 1) + { + // Per-channel quantization with OFM == 1 is equivalent to uniform quantization. + // Setting this flag to false prevents the kernel from adding useless padding to the output multipliers and shifts + gemmlowp_output_stage.is_quantized_per_channel = false; + } gemm_kernel_info.output_stage = gemmlowp_output_stage; |