From 597a85666a84c9a9414264966651551564b79299 Mon Sep 17 00:00:00 2001 From: Gian Marco Iodice Date: Wed, 1 Aug 2018 15:06:06 +0100 Subject: COMPMID-872 - Rework NEGEMMConvolutionLayer to use NEGEMM Change-Id: I55f0018ac7214775ebbca63f58a3bf5c93732fec Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/142632 Tested-by: Jenkins Reviewed-by: Anthony Barbier --- .../NEON/functions/NEGEMMConvolutionLayer.cpp | 736 ++++++++------------- 1 file changed, 293 insertions(+), 443 deletions(-) (limited to 'src/runtime/NEON/functions/NEGEMMConvolutionLayer.cpp') diff --git a/src/runtime/NEON/functions/NEGEMMConvolutionLayer.cpp b/src/runtime/NEON/functions/NEGEMMConvolutionLayer.cpp index aace261e32..c0a5d0a436 100644 --- a/src/runtime/NEON/functions/NEGEMMConvolutionLayer.cpp +++ b/src/runtime/NEON/functions/NEGEMMConvolutionLayer.cpp @@ -23,10 +23,10 @@ */ #include "arm_compute/runtime/NEON/functions/NEGEMMConvolutionLayer.h" -#include "arm_compute/core/PixelValue.h" #include "arm_compute/core/Size2D.h" #include "arm_compute/core/Utils.h" #include "arm_compute/core/Validate.h" +#include "arm_compute/core/utils/misc/ShapeCalculator.h" #include "arm_compute/core/utils/quantization/AsymmHelpers.h" #include "arm_compute/runtime/NEON/NEScheduler.h" #include "support/ToolchainSupport.h" @@ -34,96 +34,50 @@ #include #include -namespace -{ -arm_compute::TensorShape get_reshaped_weights_shape(const arm_compute::ITensorInfo *weights, bool append_bias) -{ - const unsigned int mat_weights_cols = weights->dimension(3); - const unsigned int mat_weights_rows = weights->dimension(0) * weights->dimension(1) * weights->dimension(2) + (append_bias ? 1 : 0); - return arm_compute::TensorShape(mat_weights_cols, mat_weights_rows); -} -} // namespace +using namespace arm_compute; +using namespace arm_compute::misc::shape_calculator; -namespace arm_compute -{ -NEConvolutionLayerReshapeWeights::NEConvolutionLayerReshapeWeights(std::shared_ptr memory_manager) - : _memory_group(std::move(memory_manager)), _weights_reshape_kernel(), _weights_transposed_kernel(), _weights_reshaped(), _transpose1xW(false) +NEConvolutionLayerReshapeWeights::NEConvolutionLayerReshapeWeights() + : _weights_reshape_kernel() { } -void NEConvolutionLayerReshapeWeights::configure(const ITensor *weights, const ITensor *biases, ITensor *output, bool transpose1xW) +void NEConvolutionLayerReshapeWeights::configure(const ITensor *weights, const ITensor *biases, ITensor *output) { // Perform validation step ARM_COMPUTE_ERROR_ON_NULLPTR(weights, output); ARM_COMPUTE_ERROR_THROW_ON(NEConvolutionLayerReshapeWeights::validate(weights->info(), (biases != nullptr) ? biases->info() : nullptr, - output->info(), - transpose1xW)); + output->info())); - // Check if bias are present, if yes they will be embedded to the weights matrix - const bool append_biases = (biases != nullptr) && !is_data_type_quantized_asymmetric(weights->info()->data_type()); - //const unsigned bias_element = (append_biases) ? 1 : 0; + const bool append_biases = (biases != nullptr) && !is_data_type_quantized_asymmetric(weights->info()->data_type()); const ITensor *biases_to_use = (append_biases) ? biases : nullptr; - _transpose1xW = transpose1xW; - - if(transpose1xW) - { - // Create tensor to store the reshaped weights - TensorInfo info_wr = weights->info()->clone()->set_is_resizable(true).reset_padding().set_tensor_shape(get_reshaped_weights_shape(weights->info(), append_biases)); - - _weights_reshaped.allocator()->init(info_wr); - _memory_group.manage(&_weights_reshaped); - - _weights_reshape_kernel.configure(weights, biases, &_weights_reshaped); - _weights_transposed_kernel.configure(&_weights_reshaped, output); - - _weights_reshaped.allocator()->allocate(); - } - else - { - _weights_reshape_kernel.configure(weights, biases_to_use, output); - } + _weights_reshape_kernel.configure(weights, biases_to_use, output); output->info()->set_quantization_info(weights->info()->quantization_info()); } -Status NEConvolutionLayerReshapeWeights::validate(const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, bool transpose1xW) +Status NEConvolutionLayerReshapeWeights::validate(const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output) { + ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(weights); ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(weights, 1, DataType::QASYMM8, DataType::F16, DataType::F32); ARM_COMPUTE_RETURN_ERROR_ON(weights->num_dimensions() > 4); - if(!is_data_type_quantized_asymmetric(weights->data_type())) - { - ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(weights, output); - } - // Check if bias are present, if yes they will be embedded to the weights matrix - const bool append_bias = (biases != nullptr); - if(append_bias) + if(biases != nullptr) { + const int idx_kernels = get_data_layout_dimension_index(weights->data_layout(), DataLayoutDimension::BATCHES); ARM_COMPUTE_RETURN_ERROR_ON(is_data_type_quantized_asymmetric(weights->data_type())); ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(weights, biases); - ARM_COMPUTE_RETURN_ERROR_ON(biases->dimension(0) != weights->dimension(3)); + ARM_COMPUTE_RETURN_ERROR_ON(biases->dimension(0) != weights->dimension(idx_kernels)); ARM_COMPUTE_RETURN_ERROR_ON(biases->num_dimensions() > 1); } - // Checks performed when biases are present - if(append_bias) + if((output != nullptr) && (output->total_size() != 0)) { - ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(weights, biases); - ARM_COMPUTE_RETURN_ERROR_ON(biases->dimension(0) != weights->dimension(3)); - ARM_COMPUTE_RETURN_ERROR_ON(biases->num_dimensions() > 1); - } + ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(weights, output); - if(transpose1xW) - { - TensorInfo weights_reshaped = weights->clone()->set_tensor_shape(get_reshaped_weights_shape(weights, append_bias)); - ARM_COMPUTE_RETURN_ON_ERROR(NEWeightsReshapeKernel::validate(weights, biases, &weights_reshaped)); - ARM_COMPUTE_RETURN_ON_ERROR(NEGEMMTranspose1xWKernel::validate(&weights_reshaped, output)); - } - else - { - ARM_COMPUTE_RETURN_ON_ERROR(NEWeightsReshapeKernel::validate(weights, biases, output)); + NEWeightsReshapeKernel::validate(weights, biases, output); } return Status{}; @@ -131,108 +85,21 @@ Status NEConvolutionLayerReshapeWeights::validate(const ITensorInfo *weights, co void NEConvolutionLayerReshapeWeights::run() { - _memory_group.acquire(); - NEScheduler::get().schedule(&_weights_reshape_kernel, 3); - - if(_transpose1xW) - { - NEScheduler::get().schedule(&_weights_transposed_kernel, Window::DimY); - } - - _memory_group.release(); -} - -namespace -{ -TensorShape get_reshaped_weights_shape_conv(const ITensorInfo *weights, bool append_bias, bool is_fully_connected_convolution) -{ - unsigned int mat_weights_cols = weights->dimension(3); - unsigned int mat_weights_rows = weights->dimension(0) * weights->dimension(1) * weights->dimension(2) + (append_bias ? 1 : 0); - - if(is_fully_connected_convolution) - { - // Create tensor to store the reshaped weights - return TensorShape(mat_weights_cols, mat_weights_rows); - } - else - { - // Create tensor to store transposed weights - const float transpose_width = 16.0f / weights->element_size(); - return TensorShape(mat_weights_rows * static_cast(transpose_width), static_cast(std::ceil(mat_weights_cols / transpose_width))); - } -} - -Status validate_and_initialize_values(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const PadStrideInfo &conv_info, const WeightsInfo &weights_info, - const ActivationLayerInfo &act_info, DataType &dt, - bool &append_bias, bool &skip_im2col, - bool &are_weights_reshaped, unsigned int &kernel_width, unsigned int &kernel_height, - bool &is_fully_connected_convolution, bool &is_interleaved, bool &is_quantized, bool &is_activationlayer_enabled, - unsigned int &mat_weights_cols, unsigned int &mat_weights_rows, - unsigned int &conv_w, unsigned int &conv_h, const Size2D &dilation) -{ - ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::F16, DataType::F32); - ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, weights); - ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_LAYOUT(input, weights); - - DataLayout data_layout = input->data_layout(); - const int idx_width = get_data_layout_dimension_index(data_layout, DataLayoutDimension::WIDTH); - const int idx_height = get_data_layout_dimension_index(data_layout, DataLayoutDimension::HEIGHT); - const int idx_channel = get_data_layout_dimension_index(data_layout, DataLayoutDimension::CHANNEL); - - ARM_COMPUTE_RETURN_ERROR_ON(!weights_info.are_reshaped() && weights->dimension(idx_channel) != input->dimension(idx_channel)); - ARM_COMPUTE_RETURN_ERROR_ON(weights->num_dimensions() > 4); - ARM_COMPUTE_RETURN_ERROR_ON(weights_info.are_reshaped() && is_data_type_quantized_asymmetric(input->data_type())); - ARM_COMPUTE_RETURN_ERROR_ON_MSG(data_layout == DataLayout::NHWC && input->data_type() != DataType::F32, "NHWC is only supported for FP32 data type."); - - dt = input->data_type(); - is_quantized = is_data_type_quantized_asymmetric(dt); - - if(biases != nullptr) - { - if(is_quantized) - { - ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(biases, 1, DataType::S32); - } - else - { - ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, biases); - } - ARM_COMPUTE_RETURN_ERROR_ON(!weights_info.are_reshaped() && biases->dimension(0) != weights->dimension(3)); - ARM_COMPUTE_RETURN_ERROR_ON(biases->num_dimensions() > 1); - } - - // If we have 1x1 convolution and data layout is NHWC we can disable im2col - append_bias = (biases != nullptr) && (!is_quantized); - are_weights_reshaped = weights_info.are_reshaped(); - kernel_width = (are_weights_reshaped) ? weights_info.kernel_size().first : weights->dimension(idx_width); - kernel_height = (are_weights_reshaped) ? weights_info.kernel_size().second : weights->dimension(idx_height); - mat_weights_cols = weights->dimension(3); - mat_weights_rows = weights->dimension(idx_width) * weights->dimension(idx_height) * weights->dimension(idx_channel) + ((append_bias && !skip_im2col) ? 1 : 0); - skip_im2col = (data_layout == DataLayout::NHWC && kernel_width == 1 && kernel_height == 1 && conv_info.stride().first == 1 && conv_info.stride().second == 1); - - std::tie(conv_w, conv_h) = scaled_dimensions(input->dimension(idx_width), input->dimension(idx_height), kernel_width, kernel_height, - conv_info, dilation); - - // Check if its a "fully connected" convolution - is_fully_connected_convolution = ((conv_w == 1) && (conv_h == 1)); - is_interleaved = (!is_fully_connected_convolution && !is_quantized); - is_activationlayer_enabled = act_info.enabled(); - - return Status{}; } -} // namespace NEGEMMConvolutionLayer::NEGEMMConvolutionLayer(const std::shared_ptr &memory_manager) - : _memory_group(memory_manager), _asm_glue(memory_manager), _input_im2col_kernel(), _input_interleave_kernel(), _reshape_weights(), _mm_kernel(), _mm_gemmlowp(memory_manager), - _gemmlowp_output_stage(), _output_col2im_kernel(), _activationlayer_function(), _add_bias_kernel(), _original_weights(nullptr), _input_im2col_reshaped(), _input_interleaved_reshaped(), - _weights_reshaped(), _gemm_output(), _tmp_output(), _data_layout(DataLayout::NCHW), _append_bias(false), _is_fully_connected_convolution(false), _are_weights_reshaped(false), _is_quantized(false), - _is_interleaved(false), _is_activationlayer_enabled(false), _skip_im2col(false), _is_prepared(false) + : _memory_group(memory_manager), _reshape_weights(), _im2col_kernel(), _mm_gemm(), _mm_gemmlowp(memory_manager), _gemmlowp_output_stage(), _col2im_kernel(), _activationlayer_function(), + _add_bias_kernel(), _original_weights(nullptr), _im2col_output(), _weights_reshaped(), _gemm_output(), _tmp_output(), _data_layout(DataLayout::NCHW), _append_bias(false), _skip_im2col(false), + _skip_col2im(false), _is_quantized(false), _is_activationlayer_enabled(false), _is_prepared(false) { } -void NEGEMMConvolutionLayer::configure_mm(const ITensor *input, const ITensor *weights, ITensor *output, bool is_interleaved, const GEMMReshapeInfo &reshape_info) +void NEGEMMConvolutionLayer::configure_mm(const ITensor *input, const ITensor *weights, ITensor *output, int gemm_3d_depth) { + ARM_COMPUTE_ERROR_ON_NULLPTR(input, weights); + ARM_COMPUTE_ERROR_THROW_ON(validate_mm(input->info(), weights->info(), output->info(), gemm_3d_depth, _skip_im2col)); + if(_is_quantized) { // Since we need negative offsets for computing convolution, we need to change QuantizationInfo() @@ -251,129 +118,125 @@ void NEGEMMConvolutionLayer::configure_mm(const ITensor *input, const ITensor *w } else { - _mm_kernel.configure(input, weights, output, 1.f, is_interleaved, reshape_info); + // Configure matrix multiply function + _mm_gemm.configure(input, weights, nullptr, output, 1.0f, 0.0f, GEMMInfo(false, false, true /* Reshape weights only for the first run*/, gemm_3d_depth, + _skip_im2col /* Reinterpret the input as 3D if im2col is skipped */)); } } -void NEGEMMConvolutionLayer::configure(const ITensor *input, const ITensor *weights, const ITensor *biases, ITensor *output, const PadStrideInfo &conv_info, const WeightsInfo &weights_info, - const Size2D &dilation, const ActivationLayerInfo &act_info) +Status NEGEMMConvolutionLayer::validate_mm(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *output, int gemm_3d_depth, bool skip_im2col) { - // Perform validate step - ARM_COMPUTE_ERROR_ON_NULLPTR(input, weights, output); - - DataType dt{}; - unsigned int kernel_width = 0; - unsigned int kernel_height = 0; - unsigned int mat_weights_cols = 0; - unsigned int mat_weights_rows = 0; - unsigned int conv_w = 0; - unsigned int conv_h = 0; - - _data_layout = input->info()->data_layout(); - const bool is_nhwc = _data_layout == DataLayout::NHWC; - const int idx_width = get_data_layout_dimension_index(_data_layout, DataLayoutDimension::WIDTH); - const int idx_height = get_data_layout_dimension_index(_data_layout, DataLayoutDimension::HEIGHT); - const int idx_channel = get_data_layout_dimension_index(_data_layout, DataLayoutDimension::CHANNEL); - - Status status = validate_and_initialize_values(input->info(), weights->info(), (biases == nullptr) ? nullptr : biases->info(), conv_info, weights_info, act_info, dt, _append_bias, _skip_im2col, - _are_weights_reshaped, - kernel_width, kernel_height, - _is_fully_connected_convolution, _is_interleaved, _is_quantized, _is_activationlayer_enabled, - mat_weights_cols, mat_weights_rows, conv_w, conv_h, dilation); - - ARM_COMPUTE_ERROR_THROW_ON(status); - - _is_prepared = false; - _original_weights = weights; - const ITensor *biases_to_use = (_append_bias) ? biases : nullptr; + const bool is_quantized = is_data_type_quantized_asymmetric(input->data_type()); - bool run_optimised = dt == DataType::F32; - - // Reshape weights if needed - if(run_optimised) + const GEMMInfo &gemm_info = GEMMInfo(false, false, true /* Reshape weights only for the first run */, gemm_3d_depth, skip_im2col); + if(is_quantized) { - TensorShape reshaped_weights_shape{ mat_weights_cols, mat_weights_rows }; + // Since we need negative offsets for computing convolution, we need to change QuantizationInfo() + // Extract and negate input and weights offset + const QuantizationInfo input_quantization_info = input->quantization_info(); + const QuantizationInfo weights_quantization_info = weights->quantization_info(); + + std::unique_ptr input_qa = input->clone(); + std::unique_ptr weights_qa = weights->clone(); + input_qa->set_quantization_info(QuantizationInfo(input_quantization_info.scale, -input_quantization_info.offset)); + weights_qa->set_quantization_info(QuantizationInfo(weights_quantization_info.scale, -weights_quantization_info.offset)); - // Create tensor to store the reshaped weights - _weights_reshaped.allocator()->init(TensorInfo(reshaped_weights_shape, 1, dt)); - _reshape_weights.configure(weights, biases, &_weights_reshaped, false /* 1xW transpose */); - weights = &_weights_reshaped; + // Perform validation step on GEMMLowp + NEGEMMLowpMatrixMultiplyCore::validate(input_qa.get(), weights_qa.get(), output, gemm_info); } else { - if(_are_weights_reshaped) - { - if(_is_fully_connected_convolution || _is_quantized) - { - mat_weights_cols = weights_info.num_kernels(); - mat_weights_rows = weights->info()->dimension(idx_height); - } - else - { - mat_weights_cols = weights_info.num_kernels(); - mat_weights_rows = weights_info.kernel_size().first * weights_info.kernel_size().second * input->info()->dimension(idx_channel) + (_append_bias ? 1 : 0); - } - } - else - { - TensorShape reshaped_weights_shape; - - if(_is_fully_connected_convolution || _is_quantized) - { - reshaped_weights_shape = TensorShape{ mat_weights_cols, mat_weights_rows }; - } - else - { - // Create tensor to store transposed weights - const float transpose_width = 16.0f / input->info()->element_size(); - reshaped_weights_shape = TensorShape{ mat_weights_rows *static_cast(transpose_width), - static_cast(std::ceil(mat_weights_cols / transpose_width)) }; - } - - // Create tensor to store the reshaped weights - _weights_reshaped.allocator()->init(TensorInfo(reshaped_weights_shape, 1, dt)); - _reshape_weights.configure(weights, biases_to_use, &_weights_reshaped, _is_interleaved /* 1xW transpose */); - weights = &_weights_reshaped; - } + // Perform validation step on Matrix multiply function + NEGEMM::validate(input, weights, nullptr, output, 1.0f, 0.0f, gemm_info); } + return Status{}; +} + +void NEGEMMConvolutionLayer::configure(const ITensor *input, const ITensor *weights, const ITensor *biases, ITensor *output, const PadStrideInfo &conv_info, const WeightsInfo &weights_info, + const Size2D &dilation, const ActivationLayerInfo &act_info) +{ + ARM_COMPUTE_ERROR_ON_NULLPTR(input, weights, output); - // In case we skip im2col we have to add bias + ARM_COMPUTE_ERROR_THROW_ON(NEGEMMConvolutionLayer::validate(input->info(), + weights->info(), + biases != nullptr ? biases->info() : nullptr, + output->info(), + conv_info, + weights_info, + dilation, + act_info)); + + const DataType data_type = input->info()->data_type(); + const DataLayout data_layout = input->info()->data_layout(); + const int idx_width = get_data_layout_dimension_index(data_layout, DataLayoutDimension::WIDTH); + const int idx_height = get_data_layout_dimension_index(data_layout, DataLayoutDimension::HEIGHT); + const int idx_channel = get_data_layout_dimension_index(data_layout, DataLayoutDimension::CHANNEL); + const int idx_kernels = get_data_layout_dimension_index(data_layout, DataLayoutDimension::BATCHES); + + const unsigned int kernel_width = weights->info()->dimension(idx_width); + const unsigned int kernel_height = weights->info()->dimension(idx_height); + + _is_prepared = weights_info.retain_internal_weights(); + _original_weights = weights; + _is_quantized = is_data_type_quantized_asymmetric(input->info()->data_type()); + _data_layout = data_layout; + _skip_im2col = (data_layout == DataLayout::NHWC && kernel_width == 1 && kernel_height == 1 && conv_info.stride().first == 1 && conv_info.stride().second == 1) && !_is_quantized; + _skip_col2im = (data_layout == DataLayout::NHWC) && !_is_quantized; + _append_bias = (biases != nullptr) && (!_is_quantized); + + // TODO (giaiod01): Validate GEMM3D + + const bool is_nhwc = _data_layout == DataLayout::NHWC; + const ITensor *gemm_input_to_use = input; + ITensor *gemm_output_to_use = output; + ITensor *gemm_output_staged_to_use = output; + + const unsigned bias_element = (_append_bias && !_skip_im2col) ? 1 : 0; + const ITensor *biases_to_use = (_append_bias && !_skip_im2col) ? biases : nullptr; + + // Get parameters from conv_info + unsigned int stride_x = 0; + unsigned int stride_y = 0; + std::tie(stride_x, stride_y) = conv_info.stride(); + + // Get convolved dimensions + unsigned int conv_w = 0; + unsigned int conv_h = 0; + std::tie(conv_w, conv_h) = scaled_dimensions(input->info()->dimension(idx_width), + input->info()->dimension(idx_height), + kernel_width, + kernel_height, + conv_info, + dilation); + + unsigned int mat_weights_cols = weights->info()->dimension(idx_kernels); + unsigned int mat_weights_rows = weights->info()->dimension(idx_width) * weights->info()->dimension(idx_height) * weights->info()->dimension(idx_channel) + bias_element; + + // _weights_reshaped will be auto configured in the kernel. + // Just append biases and do not transpose 1xW as it will be reshaped in NEGEMM + _reshape_weights.configure(weights, biases_to_use, &_weights_reshaped); + + weights = &_weights_reshaped; + + // Create tensor to store im2col reshaped inputs if(!_skip_im2col) { - const unsigned int mat_input_cols = mat_weights_rows; - const unsigned int mat_input_rows = conv_w * conv_h; - - // Create tensor to store im2col reshaped inputs - TensorShape shape_im2col(input->info()->tensor_shape()); - shape_im2col.set(0, mat_input_cols); - shape_im2col.set(1, mat_input_rows); + // Calculate im2col shape + // For NEON the batch size is on the fourth dimension + // TODO (giaiod01): Use auto-init COMPMID-1277 + TensorShape shape_im2col = input->info()->tensor_shape(); + shape_im2col.set(0, mat_weights_rows); + shape_im2col.set(1, conv_w * conv_h); shape_im2col.set(2, 1); - _input_im2col_reshaped.allocator()->init(input->info()->clone()->set_is_resizable(true).reset_padding().set_tensor_shape(shape_im2col)); - _memory_group.manage(&_input_im2col_reshaped); - // Create tensor (interleave) to prepare input tensor for GEMM - if(!_is_fully_connected_convolution && !run_optimised && _is_interleaved) - { - TensorShape shape_interleaved(shape_im2col); - shape_interleaved.set(idx_width, shape_interleaved.x() * 4); - shape_interleaved.set(idx_height, std::ceil(shape_interleaved[idx_height] / 4.f)); - _input_interleaved_reshaped.allocator()->init(input->info()->clone()->set_is_resizable(true).reset_padding().set_tensor_shape(shape_interleaved)); - _memory_group.manage(&_input_interleaved_reshaped); - } + _im2col_output.allocator()->init(input->info()->clone()->set_is_resizable(true).reset_padding().set_tensor_shape(shape_im2col)); + _memory_group.manage(&_im2col_output); - // Create GEMM output tensor - TensorShape shape_gemm(_input_im2col_reshaped.info()->tensor_shape()); - shape_gemm.set(0, mat_weights_cols); - shape_gemm.set(1, mat_input_rows); - const DataType gemm_data_type = _is_quantized ? DataType::S32 : dt; - // GEMM output should be S32 for acquiring raw integer accumulator without quantized postprocessing for quantized asymmetric input. - TensorInfo info_gemm(shape_gemm, 1, gemm_data_type); - info_gemm.set_quantization_info(output->info()->quantization_info()); - _gemm_output.allocator()->init(info_gemm); - _memory_group.manage(&_gemm_output); + // Configure and tune im2col + _im2col_kernel.configure(input, &_im2col_output, Size2D(kernel_width, kernel_height), conv_info, _append_bias, false, false, dilation); - // Configure im2col - _input_im2col_kernel.configure(input, &_input_im2col_reshaped, Size2D(kernel_width, kernel_height), conv_info, _append_bias, false, false, dilation); + // Update GEMM input + gemm_input_to_use = &_im2col_output; } else if(_append_bias) { @@ -381,124 +244,165 @@ void NEGEMMConvolutionLayer::configure(const ITensor *input, const ITensor *weig _add_bias_kernel.configure(output, biases, output, ConvertPolicy::SATURATE); } - // Configure matrix multiply - if(run_optimised) - { - _asm_glue.configure(_skip_im2col ? input : &_input_im2col_reshaped, weights, is_nhwc ? output : &_gemm_output, 1.f, 0.f, true); - if(!_asm_glue.is_configured()) - { - ARM_COMPUTE_ERROR("setup_assembly_kernel failed."); - } - } - else + // Create GEMM output tensor + if(!is_nhwc || _is_quantized) { - if(_is_interleaved) - { - // Configure GEMMInterleave4x4. _input_interleaved_reshaped will be auto configured in the kernel - _input_interleave_kernel.configure(&_input_im2col_reshaped, &_input_interleaved_reshaped); + // Calculate GEMM output shape + TensorShape shape_gemm = _im2col_output.info()->tensor_shape(); + shape_gemm.set(0, mat_weights_cols); + shape_gemm.set(1, conv_w * conv_h); - // Configure GEMM - configure_mm(&_input_interleaved_reshaped, weights, &_gemm_output, _is_interleaved, GEMMReshapeInfo(_input_im2col_reshaped.info()->dimension(idx_height), 0 /* no transpose */, - _input_im2col_reshaped.info()->dimension(idx_width))); - _input_interleaved_reshaped.allocator()->allocate(); - } - else - { - configure_mm(&_input_im2col_reshaped, weights, &_gemm_output, _is_interleaved); - } + // GEMM output should be S32 for acquiring raw integer accumulator without quantized postprocessing for quantized asymmetric input. + const DataType gemm_data_type = _is_quantized ? DataType::S32 : data_type; + // FIXME: input->clone() doesn't work with subtensors for grouped convolutions. + TensorInfo info_gemm(shape_gemm, 1, gemm_data_type); + info_gemm.set_quantization_info(output->info()->quantization_info()); + _gemm_output.allocator()->init(info_gemm); + _memory_group.manage(&_gemm_output); + + // Update GEMM output + gemm_output_to_use = &_gemm_output; } + // Configure and tune GEMM + configure_mm(gemm_input_to_use, weights, gemm_output_to_use, (data_layout == DataLayout::NHWC) ? conv_h : 1); + if(!_skip_im2col) { - _input_im2col_reshaped.allocator()->allocate(); + _im2col_output.allocator()->allocate(); + } - // Configure output stage for quantized case - if(_is_quantized) - { - const QuantizationInfo output_quant_info = (output->info()->total_size() == 0) ? input->info()->quantization_info() : output->info()->quantization_info(); + // Configure output stage for quantized case + if(_is_quantized) + { + const QuantizationInfo output_quant_info = (output->info()->total_size() == 0) ? input->info()->quantization_info() : output->info()->quantization_info(); - float multiplier = input->info()->quantization_info().scale * weights->info()->quantization_info().scale / output_quant_info.scale; - int output_multiplier, output_shift; - quantization::calculate_quantized_multiplier_less_than_one(multiplier, &output_multiplier, &output_shift); - _memory_group.manage(&_tmp_output); - _gemmlowp_output_stage.configure(&_gemm_output, biases, &_tmp_output, output_multiplier, output_shift, output_quant_info.offset); - } + float multiplier = input->info()->quantization_info().scale * weights->info()->quantization_info().scale / output_quant_info.scale; + int output_multiplier, output_shift; + quantization::calculate_quantized_multiplier_less_than_one(multiplier, &output_multiplier, &output_shift); - // Configure Col2Im - if(!is_nhwc) - { - _output_col2im_kernel.configure(_is_quantized ? &_tmp_output : &_gemm_output, output, Size2D(conv_w, conv_h)); - } + _memory_group.manage(&_tmp_output); + gemm_output_staged_to_use = &_tmp_output; - if(_is_quantized) - { - _tmp_output.allocator()->allocate(); - } + _gemmlowp_output_stage.configure(gemm_output_to_use, biases, gemm_output_staged_to_use, output_multiplier, output_shift, output_quant_info.offset); + } + + if(!_skip_col2im) + { + // Configure and tune Col2Im + _col2im_kernel.configure(_is_quantized ? gemm_output_staged_to_use : gemm_output_to_use, output, Size2D(conv_w, conv_h)); + } + + if(!is_nhwc || _is_quantized) + { + _tmp_output.allocator()->allocate(); _gemm_output.allocator()->allocate(); } - ARM_COMPUTE_ERROR_ON_MSG((output->info()->dimension(idx_width) != conv_w) || (output->info()->dimension(idx_height) != conv_h), "Output shape does not match the expected one"); + ARM_COMPUTE_ERROR_ON_MSG((output->info()->dimension(idx_width) != conv_w) || (output->info()->dimension(idx_height) != conv_h), + "Output shape does not match the expected one"); //Configure Activation Layer + _is_activationlayer_enabled = act_info.enabled(); + if(_is_activationlayer_enabled) { _activationlayer_function.configure(output, nullptr, act_info); } + + ARM_COMPUTE_UNUSED(weights_info); } Status NEGEMMConvolutionLayer::validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const PadStrideInfo &conv_info, const WeightsInfo &weights_info, const Size2D &dilation, const ActivationLayerInfo &act_info) { - ARM_COMPUTE_UNUSED(output); - - DataType dt{}; - bool append_bias{}; - bool skip_im2col{}; - bool are_weights_reshaped{}; - bool is_fully_connected_convolution{}; - bool is_interleaved{}; - bool is_quantized{}; - bool is_activationlayer_enabled{}; - unsigned int kernel_width = 0; - unsigned int kernel_height = 0; - unsigned int mat_weights_cols = 0; - unsigned int mat_weights_rows = 0; - unsigned int conv_w = 0; - unsigned int conv_h = 0; + ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, weights, output); + ARM_COMPUTE_RETURN_ERROR_ON_MSG(weights_info.are_reshaped(), "Weights already reshaped are not supported!"); + ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::F16, DataType::F32); + ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, weights); + ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_LAYOUT(input, weights); const DataLayout data_layout = input->data_layout(); - const bool is_nhwc = data_layout == DataLayout::NHWC; + const DataType data_type = input->data_type(); const int idx_width = get_data_layout_dimension_index(data_layout, DataLayoutDimension::WIDTH); const int idx_height = get_data_layout_dimension_index(data_layout, DataLayoutDimension::HEIGHT); + const int idx_channel = get_data_layout_dimension_index(data_layout, DataLayoutDimension::CHANNEL); + const int idx_kernels = get_data_layout_dimension_index(data_layout, DataLayoutDimension::BATCHES); - Status status = validate_and_initialize_values(input, weights, biases, conv_info, weights_info, act_info, dt, append_bias, skip_im2col, are_weights_reshaped, kernel_width, kernel_height, - is_fully_connected_convolution, is_interleaved, is_quantized, is_activationlayer_enabled, mat_weights_cols, mat_weights_rows, - conv_w, conv_h, dilation); + const unsigned int kernel_width = weights->dimension(idx_width); + const unsigned int kernel_height = weights->dimension(idx_height); - const Size2D kernel_weights = Size2D(kernel_width, kernel_height); + TensorInfo im2col_reshaped_info, info_gemm, tmp_info, weights_reshaped_info; + const ITensorInfo *gemm_input_to_use = input; + const ITensorInfo *gemm_output_to_use = output; + const ITensorInfo *gemm_output_staged_to_use = output; + const ITensorInfo *weights_to_use = weights; - ARM_COMPUTE_RETURN_ON_ERROR(status); + const bool is_nhwc = data_layout == DataLayout::NHWC; + const bool is_quantized = is_data_type_quantized_asymmetric(data_type); + bool skip_im2col = (data_layout == DataLayout::NHWC && kernel_width == 1 && kernel_height == 1) && !is_quantized; + const bool append_bias = (biases != nullptr) && (!is_quantized); + const unsigned bias_element = (append_bias && !skip_im2col) ? 1 : 0; - std::unique_ptr reshaped_weights = weights->clone(); - bool optimised_kernel = false; + ARM_COMPUTE_RETURN_ERROR_ON(weights->dimension(idx_channel) != input->dimension(idx_channel)); + ARM_COMPUTE_RETURN_ERROR_ON(weights->num_dimensions() > 4); - if(dt == DataType::F32) + // Validate biases + if(biases != nullptr) { - optimised_kernel = true; + if(is_quantized) + { + ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(biases, 1, DataType::S32); + } + else + { + ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, biases); + } + ARM_COMPUTE_RETURN_ERROR_ON(biases->dimension(0) != weights->dimension(idx_kernels)); + ARM_COMPUTE_RETURN_ERROR_ON(biases->num_dimensions() > 1); } - const unsigned int mat_input_cols = mat_weights_rows; - const unsigned int mat_input_rows = conv_w * conv_h; - TensorShape shape_im2col = input->tensor_shape(); - shape_im2col.set(0, mat_input_cols); - shape_im2col.set(1, mat_input_rows); - shape_im2col.set(2, 1); - TensorInfo im2_col_info = input->clone()->set_tensor_shape(shape_im2col); + if(act_info.enabled()) + { + ARM_COMPUTE_ERROR_ON(act_info.b() > act_info.a()); + } + + // Get convolved dimensions + unsigned int conv_w = 0; + unsigned int conv_h = 0; + + std::tie(conv_w, conv_h) = scaled_dimensions(input->dimension(idx_width), + input->dimension(idx_height), + kernel_width, + kernel_height, + conv_info, + dilation); + + unsigned int mat_weights_cols = weights->dimension(idx_kernels); + unsigned int mat_weights_rows = weights->dimension(idx_width) * weights->dimension(idx_height) * weights->dimension(idx_channel) + bias_element; + + // Output tensor auto inizialization if not yet initialized + ARM_COMPUTE_RETURN_ON_ERROR(NEConvolutionLayerReshapeWeights::validate(weights, is_quantized ? nullptr : biases, nullptr)); + weights_reshaped_info = TensorInfo(compute_weights_reshaped_shape(*weights, append_bias), 1, data_type); + weights_to_use = &weights_reshaped_info; + + // TODO (giaiod01): Validate GEMM3D if(!skip_im2col) { - // Validate im2col - ARM_COMPUTE_RETURN_ON_ERROR(NEIm2ColKernel::validate(input, &im2_col_info, kernel_weights, conv_info, append_bias, false, false, dilation)); + // Create tensor info for im2col reshaped inputs + // For NEON the batch size is on the fourth dimension + // TODO (giaiod01): Use auto-init COMPMID-1277 + TensorShape shape_im2col = input->tensor_shape(); + shape_im2col.set(0, mat_weights_rows); + shape_im2col.set(1, conv_w * conv_h); + shape_im2col.set(2, 1); + + im2col_reshaped_info = TensorInfo(shape_im2col, 1, data_type); + im2col_reshaped_info.set_quantization_info(input->quantization_info()); + + ARM_COMPUTE_RETURN_ON_ERROR(NEIm2ColKernel::validate(input, &im2col_reshaped_info, Size2D(kernel_width, kernel_height), conv_info, append_bias, false, false, dilation)); + gemm_input_to_use = &im2col_reshaped_info; } else if(append_bias) { @@ -507,65 +411,44 @@ Status NEGEMMConvolutionLayer::validate(const ITensorInfo *input, const ITensorI } // Create GEMM output tensor - TensorShape shape_gemm(im2_col_info.tensor_shape()); - shape_gemm.set(0, mat_weights_cols); - shape_gemm.set(1, mat_input_rows); - TensorInfo gemm_output_info = input->clone()->set_tensor_shape(shape_gemm); - - // Reshape weights if needed - if(optimised_kernel) + if(!is_nhwc || is_quantized) { - ARM_COMPUTE_RETURN_ERROR_ON(are_weights_reshaped); + TensorShape shape_gemm = gemm_input_to_use->tensor_shape(); + shape_gemm.set(0, mat_weights_cols); + shape_gemm.set(1, conv_w * conv_h); + const DataType gemm_data_type = is_quantized ? DataType::S32 : data_type; + // GEMM output should be S32 for acquiring raw integer accumulator without quantized postprocessing for quantized asymmetric input. + info_gemm = TensorInfo(shape_gemm, 1, gemm_data_type); + info_gemm.set_quantization_info(output->quantization_info()); - // Create tensor to store the reshaped weights - reshaped_weights->set_tensor_shape(get_reshaped_weights_shape_conv(weights, append_bias, is_fully_connected_convolution)); - ARM_COMPUTE_RETURN_ON_ERROR(NEConvolutionLayerReshapeWeights::validate(weights, biases, reshaped_weights.get(), !is_fully_connected_convolution /* 1xW transpose */)); + gemm_output_to_use = &info_gemm; } - else if(!is_quantized) - { - TensorShape reshaped_weights_shape; - if(is_fully_connected_convolution || is_quantized) - { - reshaped_weights_shape = TensorShape{ mat_weights_cols, mat_weights_rows }; - } - else - { - // Create tensor to store transposed weights - const float transpose_width = 16.0f / input->element_size(); - reshaped_weights_shape = TensorShape{ mat_weights_rows *static_cast(transpose_width), - static_cast(std::ceil(mat_weights_cols / transpose_width)) }; - } + ARM_COMPUTE_RETURN_ON_ERROR(validate_mm(gemm_input_to_use, weights_to_use, gemm_output_to_use, (data_layout == DataLayout::NHWC) ? conv_h : 1, skip_im2col)); - // Create tensor to store the reshaped weights - reshaped_weights->set_tensor_shape(get_reshaped_weights_shape_conv(weights, append_bias, is_fully_connected_convolution)); - ARM_COMPUTE_RETURN_ON_ERROR(NEConvolutionLayerReshapeWeights::validate(weights, biases, reshaped_weights.get(), !is_fully_connected_convolution /* 1xW transpose */)); - weights = reshaped_weights.get(); + if(is_quantized) + { + float multiplier = input->quantization_info().scale * weights_to_use->quantization_info().scale / output->quantization_info().scale; + int output_multiplier, output_shift; + quantization::calculate_quantized_multiplier_less_than_one(multiplier, &output_multiplier, &output_shift); - // Validate GEMM interleave and multiply - if(is_interleaved) - { - TensorShape shape_interleaved = shape_im2col; - shape_interleaved.set(idx_width, shape_interleaved.x() * 4); - shape_interleaved.set(idx_height, std::ceil(shape_interleaved.y() / 4.f)); - TensorInfo input_interleaved_info = input->clone()->set_tensor_shape(shape_interleaved); - ARM_COMPUTE_RETURN_ON_ERROR(NEGEMMInterleave4x4Kernel::validate(&im2_col_info, &input_interleaved_info)); - ARM_COMPUTE_RETURN_ON_ERROR(NEGEMMMatrixMultiplyKernel::validate(&input_interleaved_info, weights, &gemm_output_info, 1.f, is_interleaved, GEMMReshapeInfo(shape_im2col[1], // m - weights->tensor_shape()[0], // n - shape_im2col[0]) /* k */)); - } - else - { - ARM_COMPUTE_RETURN_ON_ERROR(NEGEMMMatrixMultiplyKernel::validate(&im2_col_info, weights, &gemm_output_info, 1.f, is_interleaved, GEMMReshapeInfo())); - } + tmp_info = TensorInfo(gemm_output_to_use->tensor_shape(), 1, DataType::QASYMM8); + tmp_info.set_quantization_info(output->quantization_info()); + gemm_output_staged_to_use = &tmp_info; + + // Validate output stage for quantized case + NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPoint::validate(gemm_output_to_use, biases, gemm_output_staged_to_use, output->quantization_info().offset); } - if(!is_nhwc) + + // Validate Col2Im + if(!is_nhwc || is_quantized) { - ARM_COMPUTE_RETURN_ON_ERROR(NECol2ImKernel::validate(&gemm_output_info, output, Size2D(conv_w, conv_h))); + ARM_COMPUTE_RETURN_ON_ERROR(NECol2ImKernel::validate(is_quantized ? gemm_output_staged_to_use : gemm_output_to_use, + output, + Size2D(conv_w, conv_h))); } - ARM_COMPUTE_RETURN_ERROR_ON_MSG((output->dimension(idx_width) != conv_w) || (output->dimension(idx_height) != conv_h), "Output shape does not match the expected one"); - + //Validate Activation Layer if(act_info.enabled()) { ARM_COMPUTE_RETURN_ON_ERROR(NEActivationLayer::validate(output, nullptr, act_info)); @@ -583,32 +466,23 @@ void NEGEMMConvolutionLayer::run() if(!_skip_im2col) { // Run input reshaping - unsigned int _y_dim = get_data_layout_dimension_index(_data_layout, DataLayoutDimension::HEIGHT); - NEScheduler::get().schedule(&_input_im2col_kernel, _y_dim); + unsigned int y_dim = get_data_layout_dimension_index(_data_layout, DataLayoutDimension::HEIGHT); + NEScheduler::get().schedule(&_im2col_kernel, y_dim); } - // Runs matrix multiply on reshaped matrices - if(_asm_glue.is_configured()) + // Runs NEGEMM or NEGEMMLowpMatrixMultiplyCore functions + if(_is_quantized) { - _asm_glue.run(); + // Run gemmlowp + _mm_gemmlowp.run(); + + // Run output stage + _gemmlowp_output_stage.run(); } else { - if(_is_interleaved) - { - // Run interleave - NEScheduler::get().schedule(&_input_interleave_kernel, Window::DimY); - } - - // Runs matrix multiply on reshaped matrices - if(_is_quantized) - { - _mm_gemmlowp.run(); - } - else - { - NEScheduler::get().schedule(&_mm_kernel, Window::DimY); - } + // Run gemm + _mm_gemm.run(); } if(_skip_im2col && _append_bias) @@ -616,16 +490,10 @@ void NEGEMMConvolutionLayer::run() NEScheduler::get().schedule(&_add_bias_kernel, Window::DimY); } - // Run output stage for quantized case - if(_is_quantized) - { - _gemmlowp_output_stage.run(); - } - // Reshape output matrix - if(_data_layout == DataLayout::NCHW) + if(!_skip_col2im) { - NEScheduler::get().schedule(&_output_col2im_kernel, Window::DimY); + NEScheduler::get().schedule(&_col2im_kernel, Window::DimY); } if(_is_activationlayer_enabled) @@ -640,32 +508,15 @@ void NEGEMMConvolutionLayer::prepare() { if(!_is_prepared) { - // Run weights reshaping (Runs once for every configure) - if(!_are_weights_reshaped) - { - ARM_COMPUTE_ERROR_ON(!_original_weights->is_used()); - - _weights_reshaped.allocator()->allocate(); - _reshape_weights.run(); - _reshape_weights = NEConvolutionLayerReshapeWeights(); - _original_weights->mark_as_unused(); - _are_weights_reshaped = true; - } + ARM_COMPUTE_ERROR_ON(!_original_weights->is_used()); - // Run GEMM prepare stage - if(_asm_glue.is_configured()) - { - _asm_glue.prepare(); - } - else - { - if(_is_quantized) - { - _mm_gemmlowp.prepare(); - } - } + // Run weights reshaping and mark original weights tensor as unused + _weights_reshaped.allocator()->allocate(); + _reshape_weights.run(); + _original_weights->mark_as_unused(); - // Release weights in case buffer is pretransposed + // Prepare GEMM + _is_quantized ? _mm_gemmlowp.prepare() : _mm_gemm.prepare(); if(!_weights_reshaped.is_used()) { _weights_reshaped.allocator()->free(); @@ -674,4 +525,3 @@ void NEGEMMConvolutionLayer::prepare() _is_prepared = true; } } -} // namespace arm_compute -- cgit v1.2.1