1 /*
2 * Copyright (c) 2018-2021 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 "src/cpu/operators/CpuConcatenate.h"
25
26 #include "src/cpu/kernels/CpuConcatenateBatchKernel.h"
27 #include "src/cpu/kernels/CpuConcatenateDepthKernel.h"
28 #include "src/cpu/kernels/CpuConcatenateHeightKernel.h"
29 #include "src/cpu/kernels/CpuConcatenateWidthKernel.h"
30
31 #include "arm_compute/core/utils/misc/ShapeCalculator.h"
32 #include "arm_compute/runtime/NEON/NEScheduler.h"
33
34 #include "arm_compute/core/Error.h"
35 #include "arm_compute/core/ITensor.h"
36 #include "arm_compute/core/TensorInfo.h"
37 #include "arm_compute/core/Types.h"
38 #include "arm_compute/core/Validate.h"
39 #include "src/common/utils/Log.h"
40 #include "src/core/helpers/AutoConfiguration.h"
41
42 namespace arm_compute
43 {
44 namespace cpu
45 {
configure(const std::vector<const ITensorInfo * > & srcs_vector,ITensorInfo * dst,size_t axis)46 void CpuConcatenate::configure(const std::vector<const ITensorInfo *> &srcs_vector, ITensorInfo *dst, size_t axis)
47 {
48 ARM_COMPUTE_ERROR_ON(dst == nullptr);
49 ARM_COMPUTE_LOG_PARAMS(srcs_vector, dst, axis);
50
51 _axis = axis;
52 _num_srcs = srcs_vector.size();
53
54 TensorShape dst_shape = arm_compute::misc::shape_calculator::calculate_concatenate_shape(srcs_vector, axis);
55
56 // Output auto inizialitation if not yet initialized
57 auto_init_if_empty(*dst, dst_shape, 1, srcs_vector[0]->data_type());
58 ARM_COMPUTE_ERROR_THROW_ON(CpuConcatenate::validate(srcs_vector, dst, axis));
59
60 unsigned int offset = 0;
61
62 for(unsigned int i = 0; i < _num_srcs; ++i)
63 {
64 switch(axis)
65 {
66 case Window::DimX:
67 {
68 auto kernel = std::make_unique<kernels::CpuConcatenateWidthKernel>();
69 kernel->configure(srcs_vector.at(i), offset, dst);
70 _concat_kernels.emplace_back(std::move(kernel));
71 break;
72 }
73 case Window::DimY:
74 {
75 auto kernel = std::make_unique<kernels::CpuConcatenateHeightKernel>();
76 kernel->configure(srcs_vector.at(i), offset, dst);
77 _concat_kernels.emplace_back(std::move(kernel));
78 break;
79 }
80 case Window::DimZ:
81 {
82 auto kernel = std::make_unique<kernels::CpuConcatenateDepthKernel>();
83 kernel->configure(srcs_vector.at(i), offset, dst);
84 _concat_kernels.emplace_back(std::move(kernel));
85 break;
86 }
87 case 3:
88 {
89 auto kernel = std::make_unique<kernels::CpuConcatenateBatchKernel>();
90 kernel->configure(srcs_vector.at(i), offset, dst);
91 _concat_kernels.emplace_back(std::move(kernel));
92 break;
93 }
94 default:
95 ARM_COMPUTE_ERROR("Axis not supported");
96 }
97 offset += srcs_vector.at(i)->dimension(axis);
98 }
99 }
100
validate(const std::vector<const ITensorInfo * > & srcs_vector,const ITensorInfo * dst,size_t axis)101 Status CpuConcatenate::validate(const std::vector<const ITensorInfo *> &srcs_vector, const ITensorInfo *dst, size_t axis)
102 {
103 ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(dst);
104 ARM_COMPUTE_RETURN_ERROR_ON(srcs_vector.size() < 2);
105
106 unsigned int offset = 0;
107 for(const auto &src : srcs_vector)
108 {
109 ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src);
110 switch(axis)
111 {
112 case Window::DimX:
113 {
114 ARM_COMPUTE_RETURN_ON_ERROR(kernels::CpuConcatenateWidthKernel::validate(src, offset, dst));
115 break;
116 }
117 case Window::DimY:
118 {
119 ARM_COMPUTE_RETURN_ON_ERROR(kernels::CpuConcatenateHeightKernel::validate(src, offset, dst));
120 break;
121 }
122 case Window::DimZ:
123 {
124 ARM_COMPUTE_RETURN_ON_ERROR(kernels::CpuConcatenateDepthKernel::validate(src, offset, dst));
125 break;
126 }
127 case 3:
128 {
129 ARM_COMPUTE_RETURN_ON_ERROR(kernels::CpuConcatenateBatchKernel::validate(src, offset, dst));
130 break;
131 }
132 default:
133 ARM_COMPUTE_ERROR("Axis not supported");
134 }
135 offset += src->dimension(axis);
136 }
137
138 if(dst->total_size() != 0)
139 {
140 TensorShape dst_shape = arm_compute::misc::shape_calculator::calculate_concatenate_shape(srcs_vector, axis);
141 ARM_COMPUTE_RETURN_ERROR_ON(dst_shape.total_size() != dst->tensor_shape().total_size());
142 }
143
144 return Status{};
145 }
146
run(ITensorPack & tensors)147 void CpuConcatenate::run(ITensorPack &tensors)
148 {
149 if(tensors.empty())
150 {
151 ARM_COMPUTE_ERROR("No inputs provided");
152 }
153
154 if(static_cast<int>(tensors.size() - 1) != static_cast<int>(_num_srcs))
155 {
156 ARM_COMPUTE_ERROR("Configured with different number of inputs");
157 }
158
159 int i = 0;
160 for(auto &k : _concat_kernels)
161 {
162 ITensorPack pack;
163 pack.add_tensor(TensorType::ACL_SRC, tensors.get_const_tensor(ACL_SRC_VEC + i));
164 pack.add_tensor(TensorType::ACL_DST, tensors.get_tensor(ACL_DST));
165 NEScheduler::get().schedule_op(k.get(), Window::DimY, k->window(), pack);
166 ++i;
167 }
168 }
169 } // namespace cpu
170 } // namespace arm_compute
171