2 * Copyright (c) 2018 Samsung Electronics Co., Ltd. All Rights Reserved
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
18 * Copyright (c) 2016-2018 ARM Limited.
20 * SPDX-License-Identifier: MIT
22 * Permission is hereby granted, free of charge, to any person obtaining a copy
23 * of this software and associated documentation files (the "Software"), to
24 * deal in the Software without restriction, including without limitation the
25 * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
26 * sell copies of the Software, and to permit persons to whom the Software is
27 * furnished to do so, subject to the following conditions:
29 * The above copyright notice and this permission notice shall be included in all
30 * copies or substantial portions of the Software.
32 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
33 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
34 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
35 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
36 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
37 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
41 #include "arm_compute/core/CL/kernels/CLGatherExKernel.h"
43 #include "arm_compute/core/CL/CLHelpers.h"
44 #include "arm_compute/core/CL/CLKernelLibraryEx.h"
45 #include "arm_compute/core/CL/ICLTensor.h"
46 #include "arm_compute/core/utils/misc/ShapeCalculatorEx.h"
47 #include "arm_compute/core/UtilsEx.h"
48 #include "support/StringSupport.h"
50 using namespace arm_compute;
55 inline Status validate_arguments(const ITensorInfo *input, const ITensorInfo *indices,
56 const ITensorInfo *output, int axis)
58 const uint32_t actual_axis = wrap_around(axis, static_cast<int>(input->num_dimensions()));
59 ARM_COMPUTE_RETURN_ERROR_ON(indices->num_dimensions() > 3);
60 ARM_COMPUTE_RETURN_ERROR_ON(input->num_dimensions() > 4);
61 ARM_COMPUTE_ERROR_ON(input->num_dimensions() + indices->num_dimensions() - 1 > 4);
62 ARM_COMPUTE_RETURN_ERROR_ON(actual_axis >= input->num_dimensions());
63 ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(output);
64 ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(
65 input, 1, DataType::U8, DataType::S8, DataType::QASYMM8, DataType::U16, DataType::S16,
66 DataType::U32, DataType::S32, DataType::F16, DataType::F32);
68 if (output->total_size() != 0)
70 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
71 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_QUANTIZATION_INFO(input, output);
72 TensorShape output_shape = arm_compute::misc::shape_calculator::compute_gather_shape_ex(
73 input->tensor_shape(), indices->tensor_shape(), actual_axis);
74 ARM_COMPUTE_RETURN_ERROR_ON(output_shape.total_size() != output->tensor_shape().total_size());
77 ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(indices, 1, DataType::U32, DataType::S32);
82 std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITensorInfo *indices,
83 ITensorInfo *output, int axis)
85 ARM_COMPUTE_ERROR_ON_NULLPTR(input, output, indices);
86 const uint32_t actual_axis = wrap_around(axis, static_cast<int>(input->num_dimensions()));
87 std::unique_ptr<ITensorInfo> output_info = input->clone();
88 output_info->set_tensor_shape(arm_compute::misc::shape_calculator::compute_gather_shape_ex(
89 input->tensor_shape(), indices->tensor_shape(), actual_axis));
90 // Output auto initialization if not yet initialized
91 auto_init_if_empty((*output), output_info->tensor_shape(), 1, input->data_type());
94 Window win = calculate_max_window(*output, Steps());
95 output->set_valid_region(ValidRegion(Coordinates(), output->tensor_shape()));
97 return std::make_pair(Status{}, win);
102 CLGatherExKernel::CLGatherExKernel()
103 : _input(nullptr), _indices(nullptr), _output(nullptr), _axis(0)
107 void CLGatherExKernel::configure(const ICLTensor *input, const ICLTensor *indices,
108 ICLTensor *output, int axis)
110 ARM_COMPUTE_ERROR_ON_NULLPTR(input, output, indices);
111 ARM_COMPUTE_ERROR_THROW_ON(
112 validate_arguments(input->info(), indices->info(), output->info(), axis));
114 // Configure kernel window
116 validate_and_configure_window(input->info(), indices->info(), output->info(), axis);
117 ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
122 _axis = wrap_around(axis, static_cast<int>(input->info()->num_dimensions()));
125 CLBuildOptions build_opts;
126 build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(input->info()->data_type()));
127 build_opts.add_option("-DOUTPUT_DIM_Z=" +
128 support::cpp11::to_string(output->info()->dimension(2)));
129 build_opts.add_option("-DINPUT_DIM_Z=" + support::cpp11::to_string(input->info()->dimension(2)));
130 build_opts.add_option("-DAXIS=" + support::cpp11::to_string(_axis));
131 build_opts.add_option("-DINDICES_DIM=" +
132 support::cpp11::to_string(indices->info()->num_dimensions()));
135 _kernel = static_cast<cl::Kernel>(
136 CLKernelLibraryEx::get().create_kernel("gather_ex", build_opts.options()));
137 ICLKernel::configure_internal(win_config.second);
140 Status CLGatherExKernel::validate(const ITensorInfo *input, const ITensorInfo *indices,
141 const ITensorInfo *output, int axis)
143 ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, indices, output, axis));
144 ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(),
145 indices->clone().get(),
146 output->clone().get(), axis)
151 void CLGatherExKernel::run(const Window &window, cl::CommandQueue &queue)
153 ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
154 ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(IKernel::window(), window);
156 Window window_collapsed = window.collapse_if_possible(ICLKernel::window(), Window::DimZ, 4);
157 unsigned int idx = 0;
158 add_4D_tensor_argument(idx, _input, window_collapsed);
159 add_3D_tensor_argument(idx, _indices, window_collapsed);
160 add_4D_tensor_argument(idx, _output, window_collapsed);
161 enqueue(queue, *this, window_collapsed, lws_hint());