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) 2017-2018 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/runtime/CL/functions/CLReduceOperation.h"
43 #include "arm_compute/core/TensorShape.h"
44 #include "arm_compute/runtime/CL/CLScheduler.h"
46 using namespace arm_compute;
48 CLReduceOperation::CLReduceOperation(std::shared_ptr<IMemoryManager> memory_manager)
49 : _memory_group(std::move(memory_manager)), _input(nullptr), _output(nullptr), _axis(),
50 _keep_dims(false), _interm_tensors(), _reduce_kernels(), _reshape()
54 Status CLReduceOperation::validate(const ITensorInfo *input, const ITensorInfo *output,
55 const std::set<uint32_t> &axis, bool keep_dims,
56 const ReductionOperation &op)
58 const size_t num_of_kernels = axis.size();
59 const size_t num_of_interm_tensors = num_of_kernels - (keep_dims ? 1 : 0);
61 ARM_COMPUTE_RETURN_ERROR_ON(num_of_kernels < 1);
63 // Create temporary tensor infos
64 auto interm_tensors = support::cpp14::make_unique<TensorInfo[]>(num_of_interm_tensors);
66 // Create intermediate tensor info
67 TensorShape shape{input->tensor_shape()};
69 auto it = axis.begin();
70 for (size_t i = 0; i < num_of_interm_tensors; ++i, ++it)
72 shape.set(*it, 1, false);
73 interm_tensors[i].set_data_type(input->data_type());
74 interm_tensors[i].set_tensor_shape(shape);
75 interm_tensors[i].set_num_channels(input->num_channels());
76 interm_tensors[i].set_data_layout(input->data_layout());
77 interm_tensors[i].set_quantization_info(input->quantization_info());
80 // Set a vector that is ordered ITensorInfo sequentially.
81 std::vector<const ITensorInfo *> tensors;
82 tensors.emplace_back(input);
83 for (size_t i = 0; i < num_of_interm_tensors; ++i)
85 tensors.emplace_back(interm_tensors.get() + i);
87 tensors.emplace_back(output);
89 // Validate ReduceOperation only on all kernels
91 for (size_t i = 0; i < num_of_kernels; ++i, ++it)
93 ARM_COMPUTE_RETURN_ON_ERROR(
94 CLReduceOperationKernel::validate(tensors[i], tensors[i + 1], *it, op));
99 ARM_COMPUTE_RETURN_ON_ERROR(
100 CLReshapeLayer::validate(&interm_tensors[num_of_interm_tensors - 1], output));
106 void CLReduceOperation::configure(ICLTensor *input, ICLTensor *output,
107 const std::set<uint32_t> &axis, bool keep_dims,
108 ReductionOperation op)
110 ARM_COMPUTE_ERROR_THROW_ON(validate(input->info(), output->info(), axis, keep_dims, op));
116 _keep_dims = keep_dims;
118 // NOTE The axis must have no duplication.
119 const size_t num_of_kernels = axis.size();
120 const size_t num_of_interm_tensors = num_of_kernels - (keep_dims ? 1 : 0);
122 if (num_of_kernels < 1)
124 throw std::runtime_error("CLReduceOperation: there is no axis to reduce");
127 _interm_tensors = support::cpp14::make_unique<CLTensor[]>(num_of_interm_tensors);
128 _reduce_kernels = support::cpp14::make_unique<CLReduceOperationKernel[]>(num_of_kernels);
130 // Set a vector that is ordered ICLTensors sequentially.
131 std::vector<ICLTensor *> tensors;
132 tensors.emplace_back(input);
133 for (size_t i = 0; i < num_of_interm_tensors; ++i)
135 tensors.emplace_back(_interm_tensors.get() + i);
137 tensors.emplace_back(output);
139 // Apply ReductionOperation on all kernels
140 TensorShape shape{input->info()->tensor_shape()};
141 auto it = axis.begin();
142 for (size_t i = 0; i < num_of_kernels; ++i, ++it)
144 shape.set(*it, 1, false);
145 if (!keep_dims || i != (num_of_kernels - 1))
147 _interm_tensors[i].allocator()->init(input->info()->clone()->set_tensor_shape(shape));
148 _memory_group.manage(&_interm_tensors[i]);
150 _reduce_kernels[i].configure(tensors[i], tensors[i + 1], *it, op);
153 _interm_tensors[i - 1].allocator()->allocate();
157 // Configure reshape layer if we want to drop the dimensions
160 _reshape.configure(&_interm_tensors[num_of_interm_tensors - 1], output);
161 _interm_tensors[num_of_interm_tensors - 1].allocator()->allocate();
165 void CLReduceOperation::run()
167 MemoryGroupResourceScope scope_mg(_memory_group);
169 const size_t num_of_kernels = _axis.size();
170 for (size_t i = 0; i < num_of_kernels; ++i)
172 CLScheduler::get().enqueue(_reduce_kernels[i]);