1 /*
2 * Copyright (c) 2017-2020 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 "arm_compute/core/CPP/kernels/CPPPermuteKernel.h"
25
26 #include "arm_compute/core/Helpers.h"
27 #include "arm_compute/core/utils/misc/ShapeCalculator.h"
28 #include "src/core/helpers/AutoConfiguration.h"
29 #include "src/core/helpers/WindowHelpers.h"
30
31 #include <cstddef>
32 #include <cstdint>
33
34 using namespace arm_compute;
35
36 namespace
37 {
validate_arguments(const ITensorInfo * input,const ITensorInfo * output,const PermutationVector & perm)38 Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, const PermutationVector &perm)
39 {
40 ARM_COMPUTE_RETURN_ERROR_ON(input->data_type() == DataType::UNKNOWN);
41 ARM_COMPUTE_RETURN_ERROR_ON_MSG(perm.num_dimensions() > 4, "Only up to 4D permutation vectors are supported");
42
43 const TensorShape output_shape = misc::shape_calculator::compute_permutation_output_shape(*input, perm);
44
45 // Validate configured output
46 if(output->total_size() != 0)
47 {
48 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(output->tensor_shape(), output_shape);
49 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
50 }
51
52 return Status{};
53 }
54
55 } // namespace
56
57 template <typename T>
run_permute(const Window & window)58 void CPPPermuteKernel::run_permute(const Window &window)
59 {
60 // Permute strides
61 Strides strides = _output->info()->strides_in_bytes();
62 Strides perm_strides = strides;
63 permute_strides(perm_strides, _perm);
64
65 // Create output window
66 Window window_out(window);
67 const Window::Dimension zero_window = Window::Dimension(0, 0, 0);
68 for(size_t d = 0; d <= _perm.num_dimensions(); ++d)
69 {
70 window_out.set(d, zero_window);
71 }
72
73 // Create iterators
74 Iterator in(_input, window);
75 Iterator out(_output, window_out);
76
77 if(_input->info()->num_dimensions() <= 3)
78 {
79 execute_window_loop(window, [&](const Coordinates & id)
80 {
81 const int idx = id[0] * perm_strides[0] + id[1] * perm_strides[1] + id[2] * perm_strides[2];
82 *(reinterpret_cast<T *>(out.ptr() + idx)) = *(reinterpret_cast<const T *>(in.ptr()));
83 },
84 in, out);
85 }
86 else if(_input->info()->num_dimensions() >= 4)
87 {
88 execute_window_loop(window, [&](const Coordinates & id)
89 {
90 const int idx = id[0] * perm_strides[0] + id[1] * perm_strides[1] + id[2] * perm_strides[2] + id[3] * perm_strides[3];
91 *(reinterpret_cast<T *>(out.ptr() + idx)) = *(reinterpret_cast<const T *>(in.ptr()));
92 },
93 in, out);
94 }
95 }
96
CPPPermuteKernel()97 CPPPermuteKernel::CPPPermuteKernel()
98 : _func(), _input(nullptr), _output(nullptr), _perm()
99 {
100 }
101
configure(const ITensor * input,ITensor * output,const PermutationVector & perm)102 void CPPPermuteKernel::configure(const ITensor *input, ITensor *output, const PermutationVector &perm)
103 {
104 ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
105 const TensorShape output_shape = misc::shape_calculator::compute_permutation_output_shape(*input->info(), perm);
106 // Output auto inizialitation if not yet initialized
107 auto_init_if_empty(*output->info(), input->info()->clone()->set_tensor_shape(output_shape));
108
109 // Perform validation step
110 ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), perm));
111
112 _input = input;
113 _output = output;
114 _perm = perm;
115
116 switch(input->info()->element_size())
117 {
118 case 1:
119 _func = &CPPPermuteKernel::run_permute<uint8_t>;
120 break;
121 case 2:
122 _func = &CPPPermuteKernel::run_permute<uint16_t>;
123 break;
124 case 4:
125 _func = &CPPPermuteKernel::run_permute<uint32_t>;
126 break;
127 default:
128 ARM_COMPUTE_ERROR("Element size not supported");
129 break;
130 }
131
132 // Configure kernel window
133 Window win = calculate_max_window(*input->info(), Steps());
134
135 // The CPPPermute doesn't need padding so update_window_and_padding() can be skipped
136 Coordinates coord;
137 coord.set_num_dimensions(output->info()->num_dimensions());
138 output->info()->set_valid_region(ValidRegion(coord, output->info()->tensor_shape()));
139
140 ICPPKernel::configure(win);
141 }
142
validate(const ITensorInfo * input,const ITensorInfo * output,const PermutationVector & perm)143 Status CPPPermuteKernel::validate(const ITensorInfo *input, const ITensorInfo *output, const PermutationVector &perm)
144 {
145 ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, perm));
146 return Status{};
147 }
148
run(const Window & window,const ThreadInfo & info)149 void CPPPermuteKernel::run(const Window &window, const ThreadInfo &info)
150 {
151 ARM_COMPUTE_UNUSED(info);
152 ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
153 ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICPPKernel::window(), window);
154
155 if(_func != nullptr)
156 {
157 (this->*_func)(window);
158 }
159 }
160