• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (c) 2018-2020 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 <complex>
25 
26 #include "arm_compute/runtime/CL/functions/CLStackLayer.h"
27 
28 #include "arm_compute/core/CL/ICLTensor.h"
29 #include "arm_compute/core/Error.h"
30 #include "arm_compute/core/Helpers.h"
31 #include "arm_compute/core/TensorInfo.h"
32 #include "arm_compute/core/Types.h"
33 #include "arm_compute/core/utils/misc/ShapeCalculator.h"
34 #include "arm_compute/runtime/CL/CLScheduler.h"
35 #include "src/core/CL/kernels/CLStackLayerKernel.h"
36 #include "support/MemorySupport.h"
37 
38 namespace arm_compute
39 {
CLStackLayer()40 CLStackLayer::CLStackLayer() // NOLINT
41     : _input(),
42       _stack_kernels(),
43       _num_inputs(0)
44 {
45 }
46 
47 CLStackLayer::~CLStackLayer() = default;
48 
configure(const std::vector<ICLTensor * > & input,int axis,ICLTensor * output)49 void CLStackLayer::configure(const std::vector<ICLTensor *> &input, int axis, ICLTensor *output)
50 {
51     configure(CLKernelLibrary::get().get_compile_context(), input, axis, output);
52 }
53 
configure(const CLCompileContext & compile_context,const std::vector<ICLTensor * > & input,int axis,ICLTensor * output)54 void CLStackLayer::configure(const CLCompileContext &compile_context, const std::vector<ICLTensor *> &input, int axis, ICLTensor *output)
55 {
56     _num_inputs = input.size();
57     _stack_kernels.reserve(_num_inputs);
58 
59     // Wrap around negative values
60     const unsigned int axis_u = wrap_around(axis, static_cast<int>(input[0]->info()->num_dimensions() + 1));
61 
62     for(unsigned int i = 0; i < _num_inputs; i++)
63     {
64         _stack_kernels.emplace_back(support::cpp14::make_unique<CLStackLayerKernel>());
65         _stack_kernels.back()->configure(compile_context, input[i], axis_u, i, _num_inputs, output);
66     }
67 }
68 
validate(const std::vector<ITensorInfo * > & input,int axis,const ITensorInfo * output)69 Status CLStackLayer::validate(const std::vector<ITensorInfo *> &input, int axis, const ITensorInfo *output)
70 {
71     ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(output);
72     ARM_COMPUTE_RETURN_ERROR_ON(input.empty());
73 
74     // Wrap around negative values
75     const size_t       rank   = input[0]->num_dimensions();
76     const unsigned int axis_u = wrap_around(axis, static_cast<int>(rank + 1));
77 
78     const unsigned int num_inputs = input.size();
79 
80     for(unsigned int i = 0; i < num_inputs; i++)
81     {
82         // All the tensors must have the same rank
83         ARM_COMPUTE_RETURN_ERROR_ON(input[i]->num_dimensions() != rank);
84         // Validate Kernel
85         ARM_COMPUTE_RETURN_ON_ERROR(CLStackLayerKernel::validate(input[i], axis_u, i, num_inputs, output));
86     }
87 
88     return Status{};
89 }
90 
run()91 void CLStackLayer::run()
92 {
93     for(unsigned i = 0; i < _num_inputs; i++)
94     {
95         CLScheduler::get().enqueue(*_stack_kernels[i], false);
96     }
97 }
98 } // namespace arm_compute
99