giuros01 | cd96a26 | 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/CLGenerateProposalsLayerKernel.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 *anchors, const ITensorInfo *all_anchors, const ComputeAnchorsInfo &info) |
| 43 | { |
| 44 | ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(anchors, all_anchors); |
| 45 | ARM_COMPUTE_RETURN_ERROR_ON(anchors->dimension(0) != info.values_per_roi()); |
| 46 | ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_NOT_IN(anchors, DataType::F16, DataType::F32); |
| 47 | ARM_COMPUTE_RETURN_ERROR_ON(anchors->num_dimensions() > 2); |
| 48 | if(all_anchors->total_size() > 0) |
| 49 | { |
| 50 | size_t feature_height = info.feat_height(); |
| 51 | size_t feature_width = info.feat_width(); |
| 52 | size_t num_anchors = anchors->dimension(1); |
| 53 | ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(all_anchors, anchors); |
| 54 | ARM_COMPUTE_RETURN_ERROR_ON(all_anchors->num_dimensions() > 2); |
| 55 | ARM_COMPUTE_RETURN_ERROR_ON(all_anchors->dimension(0) != info.values_per_roi()); |
| 56 | ARM_COMPUTE_RETURN_ERROR_ON(all_anchors->dimension(1) != feature_height * feature_width * num_anchors); |
| 57 | } |
| 58 | return Status{}; |
| 59 | } |
| 60 | } // namespace |
| 61 | |
| 62 | CLComputeAllAnchorsKernel::CLComputeAllAnchorsKernel() |
| 63 | : _anchors(nullptr), _all_anchors(nullptr) |
| 64 | { |
| 65 | } |
| 66 | |
| 67 | void CLComputeAllAnchorsKernel::configure(const ICLTensor *anchors, ICLTensor *all_anchors, const ComputeAnchorsInfo &info) |
| 68 | { |
| 69 | ARM_COMPUTE_ERROR_ON_NULLPTR(anchors, all_anchors); |
| 70 | ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(anchors->info(), all_anchors->info(), info)); |
| 71 | |
| 72 | // Metadata |
| 73 | const size_t num_anchors = anchors->info()->dimension(1); |
| 74 | const DataType data_type = anchors->info()->data_type(); |
| 75 | const float width = info.feat_width(); |
| 76 | const float height = info.feat_height(); |
| 77 | |
| 78 | // Initialize the output if empty |
| 79 | const TensorShape output_shape(info.values_per_roi(), width * height * num_anchors); |
| 80 | auto_init_if_empty(*all_anchors->info(), output_shape, 1, data_type); |
| 81 | |
| 82 | // Set instance variables |
| 83 | _anchors = anchors; |
| 84 | _all_anchors = all_anchors; |
| 85 | |
| 86 | // Set build options |
| 87 | CLBuildOptions build_opts; |
| 88 | build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(data_type)); |
| 89 | build_opts.add_option("-DWIDTH=" + float_to_string_with_full_precision(width)); |
| 90 | build_opts.add_option("-DHEIGHT=" + float_to_string_with_full_precision(height)); |
| 91 | build_opts.add_option("-DSTRIDE=" + float_to_string_with_full_precision(1.f / info.spatial_scale())); |
| 92 | build_opts.add_option("-DNUM_ANCHORS=" + support::cpp11::to_string(num_anchors)); |
| 93 | build_opts.add_option("-DNUM_ROI_FIELDS=" + support::cpp11::to_string(info.values_per_roi())); |
| 94 | |
| 95 | // Create kernel |
| 96 | _kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel("generate_proposals_compute_all_anchors", build_opts.options())); |
| 97 | |
| 98 | // The tensor all_anchors can be interpreted as an array of structs (each structs has values_per_roi fields). |
| 99 | // This means we don't need to pad on the X dimension, as we know in advance how many fields |
| 100 | // compose the struct. |
| 101 | Window win = calculate_max_window(*all_anchors->info(), Steps(info.values_per_roi())); |
| 102 | ICLKernel::configure_internal(win); |
| 103 | } |
| 104 | |
| 105 | Status CLComputeAllAnchorsKernel::validate(const ITensorInfo *anchors, const ITensorInfo *all_anchors, const ComputeAnchorsInfo &info) |
| 106 | { |
| 107 | ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(anchors, all_anchors, info)); |
| 108 | return Status{}; |
| 109 | } |
| 110 | |
| 111 | void CLComputeAllAnchorsKernel::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 | // Collapse everything on the first dimension |
| 117 | Window collapsed = window.collapse(ICLKernel::window(), Window::DimX); |
| 118 | |
| 119 | // Set arguments |
| 120 | unsigned int idx = 0; |
| 121 | add_1D_tensor_argument(idx, _anchors, collapsed); |
| 122 | add_1D_tensor_argument(idx, _all_anchors, collapsed); |
| 123 | |
| 124 | // Note that we don't need to loop over the slices, as we are launching exactly |
| 125 | // as many threads as all the anchors generated |
| 126 | enqueue(queue, *this, collapsed); |
| 127 | } |
| 128 | } // namespace arm_compute |