From fedefc3a8d76b9dea5945414324427ef5a01835d Mon Sep 17 00:00:00 2001 From: Luca Foschiani Date: Mon, 17 Feb 2020 17:02:49 +0000 Subject: COMPMID-2765 Add support for QASYMM8_SIGNED in NEDeconvolutionLayer Signed-off-by: Luca Foschiani Change-Id: I8295fadee15311a9ab846aa24c031b82c0b799eb Signed-off-by: Michalis Spyrou Signed-off-by: Michele Di Giorgio Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/2952 Comments-Addressed: Arm Jenkins Tested-by: Arm Jenkins Reviewed-by: Sheri Zhang --- src/core/CPP/kernels/CPPFlipWeightsKernel.cpp | 113 -------------------------- src/core/NEON/kernels/NEReverseKernel.cpp | 69 ++++++---------- 2 files changed, 23 insertions(+), 159 deletions(-) delete mode 100644 src/core/CPP/kernels/CPPFlipWeightsKernel.cpp (limited to 'src/core') diff --git a/src/core/CPP/kernels/CPPFlipWeightsKernel.cpp b/src/core/CPP/kernels/CPPFlipWeightsKernel.cpp deleted file mode 100644 index 2d4c0ce5c8..0000000000 --- a/src/core/CPP/kernels/CPPFlipWeightsKernel.cpp +++ /dev/null @@ -1,113 +0,0 @@ -/* - * Copyright (c) 2018 ARM Limited. - * - * SPDX-License-Identifier: MIT - * - * Permission is hereby granted, free of charge, to any person obtaining a copy - * of this software and associated documentation files (the "Software"), to - * deal in the Software without restriction, including without limitation the - * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or - * sell copies of the Software, and to permit persons to whom the Software is - * furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice shall be included in all - * copies or substantial portions of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE - * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER - * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, - * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE - * SOFTWARE. - */ -#include "arm_compute/core/CPP/kernels/CPPFlipWeightsKernel.h" - -#include "arm_compute/core/Error.h" -#include "arm_compute/core/Helpers.h" -#include "arm_compute/core/ITensor.h" -#include "arm_compute/core/TensorInfo.h" -#include "arm_compute/core/Types.h" -#include "arm_compute/core/Validate.h" -#include "arm_compute/core/utils/misc/ShapeCalculator.h" - -#include -#include - -using namespace arm_compute; - -CPPFlipWeightsKernel::CPPFlipWeightsKernel() - : _input(nullptr), _output(nullptr), _func(nullptr) -{ -} - -template -void CPPFlipWeightsKernel::flip_weights(const Window &window_input) -{ - // Create iterators - Iterator in(_input, window_input); - - const DataLayout data_layout = _input->info()->data_layout(); - const size_t idx_w = get_data_layout_dimension_index(data_layout, DataLayoutDimension::WIDTH); - const size_t idx_h = get_data_layout_dimension_index(data_layout, DataLayoutDimension::HEIGHT); - - const int kernel_width = _input->info()->dimension(idx_w); - const int kernel_height = _input->info()->dimension(idx_h); - - execute_window_loop(window_input, [&](const Coordinates & id) - { - const unsigned int x = kernel_width - id[idx_w] - 1; - const unsigned int y = kernel_height - id[idx_h] - 1; - Coordinates output_coord(id); - output_coord.set(idx_w, x); - output_coord.set(idx_h, y); - *(reinterpret_cast(_output->ptr_to_element(output_coord))) = *(reinterpret_cast(in.ptr())); - }, - in); -} - -void CPPFlipWeightsKernel::configure(const ITensor *input, ITensor *output) -{ - ARM_COMPUTE_ERROR_ON_NULLPTR(input, output); - ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::F16, DataType::F32); - ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_LAYOUT(input, output); - ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input, output); - - _input = input; - _output = output; - - // Configure kernel window - Window win = calculate_max_window(*input->info(), Steps()); - - // The CPPFlipWeightsKernel doesn't need padding so update_window_and_padding() can be skipped - Coordinates coord; - coord.set_num_dimensions(output->info()->num_dimensions()); - output->info()->set_valid_region(ValidRegion(coord, output->info()->tensor_shape())); - - ICPPKernel::configure(win); - - switch(input->info()->data_type()) - { - case DataType::F32: - _func = &CPPFlipWeightsKernel::flip_weights; - break; - case DataType::F16: - _func = &CPPFlipWeightsKernel::flip_weights; - break; - case DataType::QASYMM8: - _func = &CPPFlipWeightsKernel::flip_weights; - break; - default: - ARM_COMPUTE_ERROR("Not supported"); - } -} - -void CPPFlipWeightsKernel::run(const Window &window, const ThreadInfo &info) -{ - ARM_COMPUTE_UNUSED(info); - ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this); - ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICPPKernel::window(), window); - ARM_COMPUTE_ERROR_ON(_func == nullptr); - - (this->*_func)(window); -} diff --git a/src/core/NEON/kernels/NEReverseKernel.cpp b/src/core/NEON/kernels/NEReverseKernel.cpp index 2f584164dc..5a8c446ddd 100644 --- a/src/core/NEON/kernels/NEReverseKernel.cpp +++ b/src/core/NEON/kernels/NEReverseKernel.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2018-2019 ARM Limited. + * Copyright (c) 2018-2020 ARM Limited. * * SPDX-License-Identifier: MIT * @@ -106,33 +106,20 @@ void run_reverse(const Window &window, const ITensor *input, const ITensor *axis } // Check if we need a left-over loop for the y dimension - const int window_step_x = 16 / input->info()->element_size(); - const int window_start_x = window.x().start(); - const int window_end_x = std::min(window.x().end(), static_cast(input->info()->dimension(0))); - const int window_end_x_multiple_of = ((window_end_x - window_start_x) / window_step_x) * window_step_x; - bool left_over_loop_x = (((window_end_x - window_start_x) % window_step_x) != 0); + const int window_step_x = 16 / input->info()->element_size(); + const int window_start_x = window.x().start(); + const int window_end_x = window.x().end(); - Window slice = window.first_slice_window_4D(); + Window win(window); + win.set(Window::DimX, Window::Dimension(0, 1, 1)); - if(left_over_loop_x) + Iterator input_it(input, win); + execute_window_loop(win, [&](const Coordinates & id) { - // Check if window_end_y_multiple_of is greater than window_start_y - if(window_end_x_multiple_of > window_start_x) + int x = window_start_x; + for(; x <= (window_end_x - window_step_x); x += window_step_x) { - slice.set(Window::DimX, Window::Dimension(window_start_x, window_end_x_multiple_of, window_step_x)); - } - else - { - slice.set(Window::DimX, Window::Dimension(0, 0, 1)); - } - } - - do - { - Iterator input_it(input, slice); - execute_window_loop(slice, [&](const Coordinates & id) - { - auto in = wrapper::vloadq(reinterpret_cast(input_it.ptr())); + auto in = wrapper::vloadq(reinterpret_cast(input_it.ptr()) + x); // Reverse 0 axis if(axis_bit & 0x1) @@ -141,39 +128,29 @@ void run_reverse(const Window &window, const ITensor *input, const ITensor *axis in = wrapper::vcombine(wrapper::vgethigh(in), wrapper::vgetlow(in)); } - const int offset_x = (axis_bit & 0x1) ? output->info()->dimension(0) - id.x() - window_step_x : id.x(); + const int offset_x = (axis_bit & 0x1) ? output->info()->dimension(0) - x - window_step_x : x; const int offset_y = (axis_bit & 0x2) ? output->info()->dimension(1) - id.y() - 1 : id.y(); const int offset_z = (axis_bit & 0x4) ? output->info()->dimension(2) - id.z() - 1 : id.z(); const int offset_w = (axis_bit & 0x8) ? output->info()->dimension(3) - id[3] - 1 : id[3]; auto out_ptr = reinterpret_cast(output->ptr_to_element(Coordinates(offset_x, offset_y, offset_z, offset_w))); wrapper::vstore(out_ptr, in); - }, - input_it); + } - if(left_over_loop_x) + // Compute left-over elements + for(; x < window_end_x; ++x) { - slice.set(Window::DimX, Window::Dimension(window_end_x_multiple_of, window_end_x, 1)); + const auto in = *(reinterpret_cast(input_it.ptr()) + x); - Iterator input_it(input, slice); - - // Compute left-over elements along the y dimension (1x1) - execute_window_loop(slice, [&](const Coordinates & id) - { - const auto in = *reinterpret_cast(input_it.ptr()); - - const int offset_x = (axis_bit & 0x1) ? output->info()->dimension(0) - id.x() - 1 : id.x(); - const int offset_y = (axis_bit & 0x2) ? output->info()->dimension(1) - id.y() - 1 : id.y(); - const int offset_z = (axis_bit & 0x4) ? output->info()->dimension(2) - id.z() - 1 : id.z(); - const int offset_w = (axis_bit & 0x8) ? output->info()->dimension(3) - id[3] - 1 : id[3]; + const int offset_x = (axis_bit & 0x1) ? output->info()->dimension(0) - x - 1 : x; + const int offset_y = (axis_bit & 0x2) ? output->info()->dimension(1) - id.y() - 1 : id.y(); + const int offset_z = (axis_bit & 0x4) ? output->info()->dimension(2) - id.z() - 1 : id.z(); + const int offset_w = (axis_bit & 0x8) ? output->info()->dimension(3) - id[3] - 1 : id[3]; - *reinterpret_cast(output->ptr_to_element(Coordinates(offset_x, offset_y, offset_z, offset_w))) = in; - }, - input_it); + *reinterpret_cast(output->ptr_to_element(Coordinates(offset_x, offset_y, offset_z, offset_w))) = in; } - - } - while(window.slide_window_slice_4D(slice)); + }, + input_it); } void NEReverseKernel::run(const Window &window, const ThreadInfo &info) -- cgit v1.2.1