Michele Di Giorgio | 56dd726 | 2017-07-27 09:53:49 +0100 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (c) 2017 ARM Limited. |
| 3 | * |
| 4 | * SPDX-License-Identifier: MIT |
| 5 | * |
| 6 | * Permission is hereby granted, free of charge, to any person obtaining a copy |
| 7 | * of this software and associated documentation files (the "Software"), to |
| 8 | * deal in the Software without restriction, including without limitation the |
| 9 | * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or |
| 10 | * sell copies of the Software, and to permit persons to whom the Software is |
| 11 | * furnished to do so, subject to the following conditions: |
| 12 | * |
| 13 | * The above copyright notice and this permission notice shall be included in all |
| 14 | * copies or substantial portions of the Software. |
| 15 | * |
| 16 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
| 17 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
| 18 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE |
| 19 | * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
| 20 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, |
| 21 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE |
| 22 | * SOFTWARE. |
| 23 | */ |
| 24 | #include "arm_compute/core/CL/kernels/CLDequantizationLayerKernel.h" |
| 25 | |
| 26 | #include "arm_compute/core/AccessWindowStatic.h" |
| 27 | #include "arm_compute/core/CL/CLHelpers.h" |
| 28 | #include "arm_compute/core/CL/CLKernelLibrary.h" |
| 29 | #include "arm_compute/core/CL/ICLTensor.h" |
| 30 | #include "arm_compute/core/TensorInfo.h" |
| 31 | #include "arm_compute/core/Utils.h" |
| 32 | #include "arm_compute/core/Validate.h" |
| 33 | #include "arm_compute/core/Window.h" |
| 34 | |
| 35 | using namespace arm_compute; |
| 36 | |
| 37 | CLDequantizationLayerKernel::CLDequantizationLayerKernel() |
| 38 | : _input(nullptr), _output(nullptr), _min_max(nullptr) |
| 39 | { |
| 40 | } |
| 41 | |
| 42 | void CLDequantizationLayerKernel::configure(const ICLTensor *input, ICLTensor *output, const ICLTensor *min_max) |
| 43 | { |
| 44 | ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::U8); |
| 45 | ARM_COMPUTE_ERROR_ON_NULLPTR(output, min_max); |
| 46 | ARM_COMPUTE_ERROR_ON(input->info()->num_dimensions() < 3); |
| 47 | |
| 48 | // Output tensor auto initialization if not yet initialized |
| 49 | auto_init_if_empty(*output->info(), input->info()->tensor_shape(), 1, DataType::F32, 0); |
| 50 | |
| 51 | ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::F32); |
| 52 | ARM_COMPUTE_ERROR_ON_MISMATCHING_SHAPES(input, output); |
| 53 | |
| 54 | _input = input; |
| 55 | _output = output; |
| 56 | _min_max = min_max; |
| 57 | |
| 58 | constexpr unsigned int num_elems_processed_per_iteration = 4; |
| 59 | |
| 60 | // Create kernel |
| 61 | _kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel("dequantization_layer")); |
| 62 | |
| 63 | // Configure window |
| 64 | Window win = calculate_max_window(*input->info(), Steps(num_elems_processed_per_iteration)); |
| 65 | AccessWindowHorizontal input_access(input->info(), 0, num_elems_processed_per_iteration); |
| 66 | AccessWindowHorizontal output_access(output->info(), 0, num_elems_processed_per_iteration); |
| 67 | AccessWindowStatic min_max_access(min_max->info(), 0, 0, 2, min_max->info()->dimension(1)); |
| 68 | |
| 69 | // Update window and padding |
| 70 | update_window_and_padding(win, input_access, output_access, min_max_access); |
| 71 | |
| 72 | output_access.set_valid_region(win, input->info()->valid_region()); |
| 73 | |
| 74 | ICLKernel::configure(win); |
| 75 | } |
| 76 | |
| 77 | void CLDequantizationLayerKernel::run(const Window &window, cl::CommandQueue &queue) |
| 78 | { |
| 79 | ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this); |
| 80 | ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window); |
| 81 | |
| 82 | Window window_collapsed = window.collapse_if_possible(ICLKernel::window(), 3); |
| 83 | Window slice = window_collapsed.first_slice_window_3D(); |
| 84 | |
| 85 | Window min_max_window = window; |
| 86 | min_max_window.set(Window::DimX, Window::Dimension(0, 0, 0)); |
| 87 | min_max_window.set(Window::DimY, Window::Dimension(0, _min_max->info()->dimension(1), 1)); |
| 88 | min_max_window.set(Window::DimZ, Window::Dimension(0, 0, 0)); |
| 89 | |
| 90 | Window min_max_slice = min_max_window.first_slice_window_1D(); |
| 91 | |
| 92 | do |
| 93 | { |
| 94 | unsigned int idx = 0; |
| 95 | add_3D_tensor_argument(idx, _input, slice); |
| 96 | add_3D_tensor_argument(idx, _output, slice); |
| 97 | add_1D_tensor_argument(idx, _min_max, min_max_slice); |
| 98 | enqueue(queue, *this, slice); |
| 99 | } |
| 100 | while(window.slide_window_slice_3D(slice) && min_max_window.slide_window_slice_1D(min_max_slice)); |
| 101 | } |