giuros01 | c04a0e8 | 2018-10-03 12:44:35 +0100 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (c) 2018 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/CLBoundingBoxTransformKernel.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/CLValidate.h" |
| 30 | #include "arm_compute/core/CL/ICLArray.h" |
| 31 | #include "arm_compute/core/CL/ICLTensor.h" |
| 32 | #include "arm_compute/core/CL/OpenCL.h" |
| 33 | #include "arm_compute/core/Helpers.h" |
| 34 | #include "arm_compute/core/TensorInfo.h" |
| 35 | #include "arm_compute/core/Utils.h" |
| 36 | #include "arm_compute/core/Window.h" |
| 37 | |
| 38 | namespace arm_compute |
| 39 | { |
| 40 | namespace |
| 41 | { |
| 42 | Status validate_arguments(const ITensorInfo *boxes, const ITensorInfo *pred_boxes, const ITensorInfo *deltas) |
| 43 | { |
| 44 | ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(boxes, pred_boxes, deltas); |
| 45 | ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_NOT_IN(boxes, DataType::F32, DataType::F16); |
| 46 | ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_NOT_IN(deltas, DataType::F32, DataType::F16); |
| 47 | ARM_COMPUTE_RETURN_ERROR_ON(deltas->tensor_shape()[1] != boxes->tensor_shape()[1]); |
| 48 | ARM_COMPUTE_RETURN_ERROR_ON(deltas->tensor_shape()[0] % 4 != 0); |
| 49 | ARM_COMPUTE_RETURN_ERROR_ON(boxes->tensor_shape()[0] != 4); |
| 50 | ARM_COMPUTE_RETURN_ERROR_ON(deltas->num_dimensions() > 2); |
| 51 | ARM_COMPUTE_RETURN_ERROR_ON(boxes->num_dimensions() > 2); |
| 52 | |
| 53 | if(pred_boxes->total_size() > 0) |
| 54 | { |
| 55 | ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(pred_boxes->tensor_shape(), deltas->tensor_shape()); |
| 56 | ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(pred_boxes, deltas); |
| 57 | ARM_COMPUTE_RETURN_ERROR_ON(pred_boxes->num_dimensions() > 2); |
| 58 | } |
| 59 | return Status{}; |
| 60 | } |
| 61 | } // namespace |
| 62 | |
| 63 | CLBoundingBoxTransformKernel::CLBoundingBoxTransformKernel() |
| 64 | : _boxes(nullptr), _pred_boxes(nullptr), _deltas(nullptr) |
| 65 | { |
| 66 | } |
| 67 | |
| 68 | void CLBoundingBoxTransformKernel::configure(const ICLTensor *boxes, ICLTensor *pred_boxes, const ICLTensor *deltas, const BoundingBoxTransformInfo &info) |
| 69 | { |
| 70 | ARM_COMPUTE_ERROR_ON_NULLPTR(boxes, pred_boxes, deltas); |
| 71 | auto_init_if_empty(*pred_boxes->info(), *deltas->info()); |
| 72 | |
| 73 | // Set instance variables |
| 74 | _boxes = boxes; |
| 75 | _pred_boxes = pred_boxes; |
| 76 | _deltas = deltas; |
| 77 | |
| 78 | // Get image height and widht (rescaled) |
| 79 | const int img_h = floor(info.img_height() / info.scale() + 0.5f); |
| 80 | const int img_w = floor(info.img_width() / info.scale() + 0.5f); |
| 81 | |
| 82 | // Set build options |
| 83 | CLBuildOptions build_opts; |
| 84 | build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(deltas->info()->data_type())); |
| 85 | build_opts.add_option("-DWEIGHT_X=" + float_to_string_with_full_precision(info.weights()[0])); |
| 86 | build_opts.add_option("-DWEIGHT_Y=" + float_to_string_with_full_precision(info.weights()[1])); |
| 87 | build_opts.add_option("-DWEIGHT_W=" + float_to_string_with_full_precision(info.weights()[2])); |
| 88 | build_opts.add_option("-DWEIGHT_H=" + float_to_string_with_full_precision(info.weights()[3])); |
| 89 | build_opts.add_option("-DBBOX_XFORM_CLIP=" + float_to_string_with_full_precision(info.bbox_xform_clip())); |
| 90 | build_opts.add_option("-DIMG_WIDTH=" + support::cpp11::to_string(img_w)); |
| 91 | build_opts.add_option("-DIMG_HEIGHT=" + support::cpp11::to_string(img_h)); |
| 92 | build_opts.add_option("-DBOX_FIELDS=" + support::cpp11::to_string(4)); |
| 93 | build_opts.add_option_if(info.apply_scale(), "-DSCALE=" + float_to_string_with_full_precision(info.scale())); |
| 94 | |
| 95 | // Create kernel |
| 96 | _kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel("bounding_box_transform", build_opts.options())); |
| 97 | |
| 98 | // Since the number of columns is a multiple of 4 by definition, we don't need to pad the tensor |
| 99 | const unsigned int num_elems_processed_per_iteration = 4; |
| 100 | Window win = calculate_max_window(*deltas->info(), Steps(num_elems_processed_per_iteration)); |
| 101 | ICLKernel::configure_internal(win); |
| 102 | } |
| 103 | |
| 104 | Status CLBoundingBoxTransformKernel::validate(const ITensorInfo *boxes, const ITensorInfo *pred_boxes, const ITensorInfo *deltas, const BoundingBoxTransformInfo &info) |
| 105 | { |
| 106 | ARM_COMPUTE_UNUSED(info); |
| 107 | ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(boxes, pred_boxes, deltas)); |
| 108 | return Status{}; |
| 109 | } |
| 110 | |
| 111 | void CLBoundingBoxTransformKernel::run(const Window &window, cl::CommandQueue &queue) |
| 112 | { |
| 113 | ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this); |
| 114 | ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(IKernel::window(), window); |
| 115 | |
| 116 | Window slice = window.first_slice_window_2D(); |
| 117 | |
| 118 | // Set arguments |
| 119 | unsigned int idx = 0; |
| 120 | add_1D_tensor_argument(idx, _boxes, slice); |
| 121 | add_2D_tensor_argument(idx, _pred_boxes, slice); |
| 122 | add_2D_tensor_argument(idx, _deltas, slice); |
| 123 | |
| 124 | // Note that we don't need to loop over the slices, as we are sure that we are dealing with all 2D tensors |
| 125 | enqueue(queue, *this, slice); |
| 126 | } |
| 127 | } // namespace arm_compute |