2 * Copyright (c) 2018 ARM Limited.
4 * SPDX-License-Identifier: MIT
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:
13 * The above copyright notice and this permission notice shall be included in all
14 * copies or substantial portions of the Software.
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
24 #include "arm_compute/core/CL/kernels/CLPermuteKernel.h"
26 #include "arm_compute/core/CL/CLHelpers.h"
27 #include "arm_compute/core/CL/CLKernelLibrary.h"
28 #include "arm_compute/core/CL/ICLTensor.h"
29 #include "arm_compute/core/CL/OpenCL.h"
30 #include "arm_compute/core/Error.h"
31 #include "arm_compute/core/Helpers.h"
32 #include "arm_compute/core/Types.h"
33 #include "arm_compute/core/utils/misc/ShapeCalculator.h"
34 #include "support/ToolchainSupport.h"
36 using namespace arm_compute;
38 CLPermuteKernel::CLPermuteKernel()
39 : _input(nullptr), _output(nullptr), _perm()
44 TensorShape get_output_shape(const ITensorInfo *input, const PermutationVector &perm)
46 TensorShape output_shape = input->tensor_shape();
47 permute(output_shape, perm);
51 Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, const PermutationVector &perm)
53 ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::U8, DataType::S8, DataType::QS8, DataType::QASYMM8,
54 DataType::U16, DataType::S16, DataType::QS16,
55 DataType::U32, DataType::S32,
56 DataType::F16, DataType::F32);
57 ARM_COMPUTE_RETURN_ERROR_ON_MSG((input->num_dimensions() < 3), "Invalid input size!");
58 ARM_COMPUTE_RETURN_ERROR_ON_MSG(((perm.num_dimensions() == 3 && !(perm[0] == 2 && perm[1] == 0 && perm[2] == 1) && !(perm[0] == 1 && perm[1] == 2 && perm[2] == 0)) || (perm.num_dimensions() == 4
59 && !(perm[0] == 3 && perm[1] == 2 && perm[2] == 0 && perm[3] == 1))),
60 "Only [2, 0, 1],[1, 2, 0] and [3, 2, 0, 1] permutation is supported");
62 const TensorShape output_shape = misc::shape_calculator::compute_permutation_output_shape(*input, perm);
64 // Validate configured output
65 if(output->total_size() != 0)
67 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(output->tensor_shape(), output_shape);
68 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
69 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_FIXED_POINT(input, output);
75 void CLPermuteKernel::configure(const ICLTensor *input, ICLTensor *output, const PermutationVector &perm)
77 ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
78 ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), perm));
84 const TensorShape output_shape = get_output_shape(input->info(), perm);
85 // Output auto inizialitation if not yet initialized
86 auto_init_if_empty(*output->info(), input->info()->clone()->set_tensor_shape(output_shape));
89 std::set<std::string> build_opts;
91 build_opts.emplace("-DDATA_TYPE=" + get_cl_type_from_data_type(input->info()->data_type()));
92 build_opts.emplace("-DDEPTH_IN=" + support::cpp11::to_string(input->info()->dimension(2)));
94 // Run [2, 0, 1] permute
95 if(_perm[0] == 2 && _perm[1] == 0 && _perm[2] == 1)
97 _kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel("permute_201", build_opts));
99 // Run [1, 2, 0] permute
100 else if(_perm[0] == 1 && _perm[1] == 2 && _perm[2] == 0)
102 _kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel("permute_120", build_opts));
104 // Run [3, 2, 0, 1] permute
105 else if(_perm[0] == 3 && _perm[1] == 2 && _perm[2] == 0 && _perm[3] == 1)
107 _kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel("permute_3201", build_opts));
111 ARM_COMPUTE_ERROR("Not supported.");
114 // Configure kernel window
115 Window win = calculate_max_window(*input->info(), Steps());
117 ICLKernel::configure(win);
120 Status CLPermuteKernel::validate(const ITensorInfo *input, const ITensorInfo *output, const PermutationVector &perm)
122 ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
123 ARM_COMPUTE_RETURN_ERROR_ON(validate_arguments(input, output, perm));
128 void CLPermuteKernel::run(const Window &window, cl::CommandQueue &queue)
130 ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
131 ARM_COMPUTE_ERROR_ON_MISMATCHING_WINDOWS(ICLKernel::window(), window);
133 Window slice_in = window.first_slice_window_4D().collapse(ICLKernel::window(), 2, 4);
135 // Setup output slice
136 Window slice_out(slice_in);
137 slice_out.set(Window::DimX, Window::Dimension(0, 0, 0));
138 slice_out.set(Window::DimY, Window::Dimension(0, 0, 0));
139 slice_out.set(Window::DimZ, Window::Dimension(0, 0, 0));
140 slice_out.set(3, Window::Dimension(0, 0, 0));
144 unsigned int idx = 0;
145 add_4D_tensor_argument(idx, _input, slice_in);
146 add_4D_tensor_argument(idx, _output, slice_out);
147 enqueue(queue, *this, slice_in);
149 while(window.slide_window_slice_4D(slice_in) && window.slide_window_slice_4D(slice_out));