diff options
Diffstat (limited to 'src')
6 files changed, 10 insertions, 9 deletions
diff --git a/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NCHWKernel.cpp b/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NCHWKernel.cpp index 615327a7cc..fe796e96b8 100644 --- a/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NCHWKernel.cpp +++ b/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NCHWKernel.cpp @@ -88,7 +88,7 @@ std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITen { // Output auto inizialitation if not yet initialized const TensorShape output_shape = compute_depthwise_convolution_shape(*input, *weights, conv_info, depth_multiplier, dilation); - auto_init_if_empty(*output, input->clone()->set_tensor_shape(output_shape)); + auto_init_if_empty(*output, input->clone()->set_tensor_shape(output_shape).set_quantization_info(output->quantization_info())); const unsigned int conv_stride_x = conv_info.stride().first; const unsigned int conv_stride_y = conv_info.stride().second; diff --git a/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NHWCKernel.cpp b/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NHWCKernel.cpp index f489e82cb9..779cf25fdf 100644 --- a/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NHWCKernel.cpp +++ b/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NHWCKernel.cpp @@ -109,12 +109,7 @@ std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITen const TensorShape output_shape = arm_compute::misc::shape_calculator::compute_depthwise_convolution_shape( *input, TensorInfo(TensorShape(weights_width, weights_height), 1, weights->data_type()).set_data_layout(DataLayout::NCHW), conv_info, depth_multiplier, dilation); - // Output auto inizialitation if not yet initialized - auto_init_if_empty(*output, - output_shape, - 1, - input->data_type(), - input->quantization_info()); + auto_init_if_empty(*output, input->clone()->set_tensor_shape(output_shape).set_quantization_info(output->quantization_info())); const bool is_qasymm = is_data_type_quantized_asymmetric(input->data_type()); const bool is_stride_1_dilation_1 = ((conv_info.stride().first == conv_info.stride().second) && (conv_info.stride().first == 1) && dilation.x() == 1 && dilation.y() == 1); diff --git a/src/core/NEON/kernels/NEDepthwiseConvolutionLayer3x3Kernel.cpp b/src/core/NEON/kernels/NEDepthwiseConvolutionLayer3x3Kernel.cpp index 385be04e4a..e47786525e 100644 --- a/src/core/NEON/kernels/NEDepthwiseConvolutionLayer3x3Kernel.cpp +++ b/src/core/NEON/kernels/NEDepthwiseConvolutionLayer3x3Kernel.cpp @@ -195,7 +195,7 @@ std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITen const DataType output_dt = (input->data_type() == DataType::QASYMM8) ? DataType::S32 : input->data_type(); // Output auto inizialitation if not yet initialized - auto_init_if_empty(*output, input->clone()->set_is_resizable(true).reset_padding().set_tensor_shape(output_shape).set_data_type(output_dt)); + auto_init_if_empty(*output, input->clone()->set_is_resizable(true).reset_padding().set_tensor_shape(output_shape).set_data_type(output_dt).set_quantization_info(output->quantization_info())); // Configure kernel window (generic) const unsigned int conv_stride_x = conv_info.stride().first; diff --git a/src/runtime/CL/functions/CLDepthwiseConvolutionLayer.cpp b/src/runtime/CL/functions/CLDepthwiseConvolutionLayer.cpp index e912740d69..16082775c3 100644 --- a/src/runtime/CL/functions/CLDepthwiseConvolutionLayer.cpp +++ b/src/runtime/CL/functions/CLDepthwiseConvolutionLayer.cpp @@ -90,6 +90,7 @@ void CLDepthwiseConvolutionLayer3x3::configure(ICLTensor *input, const ICLTensor // Configure the function to transform the weights tensor from HWI -> IHW _permute_weights_to_nchw.configure(weights, &_permuted_weights, PermutationVector(1U, 2U, 0U)); _permuted_weights.info()->set_data_layout(DataLayout::NCHW); + _permuted_output.info()->set_quantization_info(output->info()->quantization_info()); input_to_use = &_permuted_input; weights_to_use = &_permuted_weights; diff --git a/src/runtime/NEON/functions/NEDepthwiseConvolutionLayer.cpp b/src/runtime/NEON/functions/NEDepthwiseConvolutionLayer.cpp index 4bc8439d93..43288ec4c6 100644 --- a/src/runtime/NEON/functions/NEDepthwiseConvolutionLayer.cpp +++ b/src/runtime/NEON/functions/NEDepthwiseConvolutionLayer.cpp @@ -87,6 +87,7 @@ void NEDepthwiseConvolutionLayer3x3::configure_generic(ITensor // Configure the function to transform the weights tensor from HWI -> IHW _permute_weights.configure(weights, &_permuted_weights, PermutationVector(1U, 2U, 0U)); _permuted_weights.info()->set_data_layout(DataLayout::NCHW); + _permuted_output.info()->set_quantization_info(output->info()->quantization_info()); // Configure depthwise _dwc_kernel.configure(&_permuted_input, &_permuted_weights, (_is_quantized) ? &_accumulator : &_permuted_output, conv_info, depth_multiplier, dilation); @@ -164,6 +165,9 @@ void NEDepthwiseConvolutionLayer3x3::configure_optimized(const ITensor _permute_weights.configure(weights, &_permuted_weights, PermutationVector(2U, 0U, 1U)); _permuted_weights.info()->set_data_layout(DataLayout::NHWC); + _permuted_output.info()->set_data_layout(DataLayout::NHWC); + _permuted_output.info()->set_quantization_info(output->info()->quantization_info()); + // Configure optimized depthwise _dwc_optimized_func.configure(&_permuted_input, &_permuted_weights, biases, &_permuted_output, conv_info, depth_multiplier, act_info_to_use); @@ -423,6 +427,7 @@ void NEDepthwiseConvolutionLayer::configure(ITensor *input, const ITensor *weigh permute(output_shape, PermutationVector(1U, 2U, 0U)); _permuted_output.allocator()->init(output->info()->clone()->set_is_resizable(true).reset_padding().set_tensor_shape(output_shape)); _permuted_output.info()->set_data_layout(DataLayout::NCHW); + _permuted_output.info()->set_quantization_info(output->info()->quantization_info()); output_to_use = &_permuted_output; } diff --git a/src/runtime/NEON/functions/assembly/NEDepthwiseConvolutionAssemblyDispatch.cpp b/src/runtime/NEON/functions/assembly/NEDepthwiseConvolutionAssemblyDispatch.cpp index 0499d9930f..5f57bbfe23 100644 --- a/src/runtime/NEON/functions/assembly/NEDepthwiseConvolutionAssemblyDispatch.cpp +++ b/src/runtime/NEON/functions/assembly/NEDepthwiseConvolutionAssemblyDispatch.cpp @@ -176,7 +176,7 @@ void NEDepthwiseConvolutionAssemblyDispatch::configure(const ITensor // Output auto inizialitation if not yet initialized const TensorShape output_shape = misc::shape_calculator::compute_depthwise_convolution_shape(*input->info(), *weights->info(), conv_info, depth_multiplier); - auto_init_if_empty(*output->info(), input->info()->clone()->set_is_resizable(true).reset_padding().set_tensor_shape(output_shape)); + auto_init_if_empty(*output->info(), input->info()->clone()->set_is_resizable(true).reset_padding().set_tensor_shape(output_shape).set_quantization_info(output->info()->quantization_info())); _input = input; _weights = weights; |