From 0cdbda5e51e6ef9e03017231e56ee85ede69bb9a Mon Sep 17 00:00:00 2001 From: Sheri Zhang Date: Tue, 25 Feb 2020 15:57:21 +0000 Subject: COMPMID-2789: Add support for QASYMM8_SIGNED in CLGEMMDeconvolutionLayer Change-Id: I7e3bcb01025e827f6f62491749c691c205ee7481 Signed-off-by: Sheri Zhang Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/2844 Tested-by: Arm Jenkins Reviewed-by: Michele Di Giorgio Comments-Addressed: Arm Jenkins --- .../CL/functions/CLGEMMDeconvolutionLayer.cpp | 61 ++++++++++++++-------- 1 file changed, 38 insertions(+), 23 deletions(-) (limited to 'src/runtime/CL/functions/CLGEMMDeconvolutionLayer.cpp') diff --git a/src/runtime/CL/functions/CLGEMMDeconvolutionLayer.cpp b/src/runtime/CL/functions/CLGEMMDeconvolutionLayer.cpp index 14bda11f5f..3298858215 100644 --- a/src/runtime/CL/functions/CLGEMMDeconvolutionLayer.cpp +++ b/src/runtime/CL/functions/CLGEMMDeconvolutionLayer.cpp @@ -62,6 +62,33 @@ std::pair compute_start_end_slice_coordinates(const IT return { start, end }; } +Status construct_gemmlowp_output_stage(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *output, GEMMLowpOutputStageInfo &output_stage_info) +{ + const auto data_type = input->data_type(); + + if(is_data_type_quantized_asymmetric(data_type)) + { + const UniformQuantizationInfo iq_info = input->quantization_info().uniform(); + const UniformQuantizationInfo wq_info = weights->quantization_info().uniform(); + const UniformQuantizationInfo oq_info = output->quantization_info().uniform(); + + float multiplier = iq_info.scale * wq_info.scale / oq_info.scale; + int output_multiplier(0); + int output_shift(0); + ARM_COMPUTE_RETURN_ON_ERROR(quantization::calculate_quantized_multiplier(multiplier, &output_multiplier, &output_shift)); + + output_stage_info.type = GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT; + output_stage_info.gemmlowp_multiplier = output_multiplier; + output_stage_info.gemmlowp_shift = output_shift; + output_stage_info.gemmlowp_offset = oq_info.offset; + const auto min_max_bound = get_min_max(data_type); + output_stage_info.gemmlowp_min_bound = (std::get<0>(min_max_bound)).get(); + output_stage_info.gemmlowp_max_bound = (std::get<1>(min_max_bound)).get(); + output_stage_info.output_data_type = data_type; + } + return Status{}; +} + } // namespace CLGEMMDeconvolutionLayer::CLGEMMDeconvolutionLayer(std::shared_ptr memory_manager) // NOLINT @@ -93,7 +120,7 @@ CLGEMMDeconvolutionLayer::CLGEMMDeconvolutionLayer(std::shared_ptrset_tensor_shape(gemm_output_shape).set_is_resizable(true); GEMMInfo gemm_info(false, false, true, input->dimension(idx_h), true); + GEMMLowpOutputStageInfo output_stage_info; + if(is_quantized) { ARM_COMPUTE_RETURN_ON_ERROR(CLGEMMLowpMatrixMultiplyCore::validate(&input->clone()->set_tensor_shape(nhwc_input_shape), &reshaped_t_info, nullptr, &gemm_output_info.set_data_type(DataType::S32), gemm_info)); + ARM_COMPUTE_RETURN_ON_ERROR(construct_gemmlowp_output_stage(input, weights, output, output_stage_info)); + } else { @@ -160,9 +191,8 @@ Status CLGEMMDeconvolutionLayer::validate(const ITensorInfo *input, const ITenso { const auto start_end = compute_start_end_slice_coordinates(col2im_output_info, deconv_info, is_nchw); ARM_COMPUTE_RETURN_ON_ERROR(CLDeconvolutionReshapeOutputKernel::validate(&gemm_output_info, bias, &col2im_output_info, input, weights, deconv_info)); - ARM_COMPUTE_RETURN_ON_ERROR(CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPoint::validate(&col2im_output_info, nullptr, - &col2im_output_info.clone()->set_is_resizable(true).set_data_type(DataType::QASYMM8))); - ARM_COMPUTE_RETURN_ON_ERROR(CLSlice::validate(&col2im_output_info.clone()->set_is_resizable(true).set_data_type(DataType::QASYMM8), output, start_end.first, start_end.second)); + ARM_COMPUTE_RETURN_ON_ERROR(CLGEMMLowpOutputStage::validate(&col2im_output_info, nullptr, &col2im_output_info.clone()->set_is_resizable(true).set_data_type(input->data_type()), output_stage_info)); + ARM_COMPUTE_RETURN_ON_ERROR(CLSlice::validate(&col2im_output_info.clone()->set_is_resizable(true).set_data_type(input->data_type()), output, start_end.first, start_end.second)); } else if(padded_input) { @@ -173,16 +203,7 @@ Status CLGEMMDeconvolutionLayer::validate(const ITensorInfo *input, const ITenso else if(is_quantized) { ARM_COMPUTE_RETURN_ON_ERROR(CLDeconvolutionReshapeOutputKernel::validate(&gemm_output_info, bias, &col2im_output_info, input, weights, deconv_info)); - - const UniformQuantizationInfo iq_info = input->quantization_info().uniform(); - const UniformQuantizationInfo wq_info = weights->quantization_info().uniform(); - const UniformQuantizationInfo oq_info = output->quantization_info().uniform(); - - float multiplier = iq_info.scale * wq_info.scale / oq_info.scale; - int output_multiplier(0); - int output_shift(0); - ARM_COMPUTE_RETURN_ON_ERROR(quantization::calculate_quantized_multiplier(multiplier, &output_multiplier, &output_shift)); - ARM_COMPUTE_RETURN_ON_ERROR(CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPoint::validate(&col2im_output_info, nullptr, output)); + ARM_COMPUTE_RETURN_ON_ERROR(CLGEMMLowpOutputStage::validate(&col2im_output_info, nullptr, output, output_stage_info)); } else { @@ -297,15 +318,9 @@ void CLGEMMDeconvolutionLayer::configure(const ICLTensor *input, const ICLTensor if(_is_quantized) { - const UniformQuantizationInfo iq_info = input->info()->quantization_info().uniform(); - const UniformQuantizationInfo wq_info = weights->info()->quantization_info().uniform(); - const UniformQuantizationInfo oq_info = output->info()->quantization_info().uniform(); - - float multiplier = iq_info.scale * wq_info.scale / oq_info.scale; - int output_multiplier(0); - int output_shift(0); - quantization::calculate_quantized_multiplier(multiplier, &output_multiplier, &output_shift); - _gemmlowp_output_stage.configure(&_gemmlowp_final, nullptr, output_stage_output, output_multiplier, output_shift, oq_info.offset); + GEMMLowpOutputStageInfo output_stage_info; + construct_gemmlowp_output_stage(input->info(), weights->info(), output->info(), output_stage_info); + _gemmlowp_output_stage.configure(&_gemmlowp_final, nullptr, output_stage_output, output_stage_info); _gemmlowp_final.allocator()->allocate(); } -- cgit v1.2.1