1 /*
2 * Copyright (c) 2019-2021 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 "arm_compute/runtime/NEON/functions/NEFFT1D.h"
25
26 #include "arm_compute/core/ITensor.h"
27 #include "arm_compute/core/Validate.h"
28 #include "arm_compute/runtime/NEON/NEScheduler.h"
29 #include "src/common/utils/Log.h"
30 #include "src/core/NEON/kernels/NEFFTDigitReverseKernel.h"
31 #include "src/core/NEON/kernels/NEFFTRadixStageKernel.h"
32 #include "src/core/NEON/kernels/NEFFTScaleKernel.h"
33 #include "src/core/utils/helpers/fft.h"
34
35 namespace arm_compute
36 {
37 NEFFT1D::~NEFFT1D() = default;
38
NEFFT1D(std::shared_ptr<IMemoryManager> memory_manager)39 NEFFT1D::NEFFT1D(std::shared_ptr<IMemoryManager> memory_manager)
40 : _memory_group(std::move(memory_manager)), _digit_reverse_kernel(), _fft_kernels(), _scale_kernel(), _digit_reversed_input(), _digit_reverse_indices(), _num_ffts(0), _axis(0), _run_scale(false)
41 {
42 }
43
configure(const ITensor * input,ITensor * output,const FFT1DInfo & config)44 void NEFFT1D::configure(const ITensor *input, ITensor *output, const FFT1DInfo &config)
45 {
46 ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
47 ARM_COMPUTE_ERROR_THROW_ON(NEFFT1D::validate(input->info(), output->info(), config));
48 ARM_COMPUTE_LOG_PARAMS(input, output, config);
49
50 // Decompose size to radix factors
51 const auto supported_radix = NEFFTRadixStageKernel::supported_radix();
52 const unsigned int N = input->info()->tensor_shape()[config.axis];
53 const auto decomposed_vector = arm_compute::helpers::fft::decompose_stages(N, supported_radix);
54 ARM_COMPUTE_ERROR_ON(decomposed_vector.empty());
55
56 // Flags
57 _run_scale = config.direction == FFTDirection::Inverse;
58
59 const bool is_c2r = input->info()->num_channels() == 2 && output->info()->num_channels() == 1;
60
61 // Configure digit reverse
62 FFTDigitReverseKernelInfo digit_reverse_config;
63 digit_reverse_config.axis = config.axis;
64 digit_reverse_config.conjugate = config.direction == FFTDirection::Inverse;
65 TensorInfo digit_reverse_indices_info(TensorShape(input->info()->tensor_shape()[config.axis]), 1, DataType::U32);
66 _digit_reverse_indices.allocator()->init(digit_reverse_indices_info);
67 _memory_group.manage(&_digit_reversed_input);
68 _digit_reverse_kernel = std::make_unique<NEFFTDigitReverseKernel>();
69 _digit_reverse_kernel->configure(input, &_digit_reversed_input, &_digit_reverse_indices, digit_reverse_config);
70
71 // Create and configure FFT kernels
72 unsigned int Nx = 1;
73 _num_ffts = decomposed_vector.size();
74 _fft_kernels.resize(_num_ffts);
75 _axis = config.axis;
76
77 for(unsigned int i = 0; i < _num_ffts; ++i)
78 {
79 const unsigned int radix_for_stage = decomposed_vector.at(i);
80
81 FFTRadixStageKernelInfo fft_kernel_info;
82 fft_kernel_info.axis = config.axis;
83 fft_kernel_info.radix = radix_for_stage;
84 fft_kernel_info.Nx = Nx;
85 fft_kernel_info.is_first_stage = (i == 0);
86 _fft_kernels[i] = std::make_unique<NEFFTRadixStageKernel>();
87 _fft_kernels[i]->configure(&_digit_reversed_input, ((i == (_num_ffts - 1)) && !is_c2r) ? output : nullptr, fft_kernel_info);
88
89 Nx *= radix_for_stage;
90 }
91
92 // Configure scale kernel
93 if(_run_scale)
94 {
95 FFTScaleKernelInfo scale_config;
96 scale_config.scale = static_cast<float>(N);
97 scale_config.conjugate = config.direction == FFTDirection::Inverse;
98 _scale_kernel = std::make_unique<NEFFTScaleKernel>();
99 is_c2r ? _scale_kernel->configure(&_digit_reversed_input, output, scale_config) : _scale_kernel->configure(output, nullptr, scale_config);
100 }
101
102 // Allocate tensors
103 _digit_reversed_input.allocator()->allocate();
104 _digit_reverse_indices.allocator()->allocate();
105
106 // Init digit reverse indices
107 const auto digit_reverse_cpu = arm_compute::helpers::fft::digit_reverse_indices(N, decomposed_vector);
108 std::copy_n(digit_reverse_cpu.data(), N, reinterpret_cast<unsigned int *>(_digit_reverse_indices.buffer()));
109 }
110
validate(const ITensorInfo * input,const ITensorInfo * output,const FFT1DInfo & config)111 Status NEFFT1D::validate(const ITensorInfo *input, const ITensorInfo *output, const FFT1DInfo &config)
112 {
113 ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
114 ARM_COMPUTE_RETURN_ERROR_ON(input->data_type() != DataType::F32);
115 ARM_COMPUTE_RETURN_ERROR_ON(input->num_channels() > 2);
116 ARM_COMPUTE_RETURN_ERROR_ON(std::set<unsigned int>({ 0, 1 }).count(config.axis) == 0);
117
118 // Check if FFT is decomposable
119 const auto supported_radix = NEFFTRadixStageKernel::supported_radix();
120 const unsigned int N = input->tensor_shape()[config.axis];
121 const auto decomposed_vector = arm_compute::helpers::fft::decompose_stages(N, supported_radix);
122 ARM_COMPUTE_RETURN_ERROR_ON(decomposed_vector.empty());
123
124 // Checks performed when output is configured
125 if((output != nullptr) && (output->total_size() != 0))
126 {
127 // All combinations are supported except real input with real output (i.e., both input channels set to 1)
128 ARM_COMPUTE_RETURN_ERROR_ON(output->num_channels() == 1 && input->num_channels() == 1);
129 ARM_COMPUTE_RETURN_ERROR_ON(output->num_channels() > 2);
130 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input, output);
131 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
132 }
133
134 return Status{};
135 }
136
run()137 void NEFFT1D::run()
138 {
139 MemoryGroupResourceScope scope_mg(_memory_group);
140
141 NEScheduler::get().schedule(_digit_reverse_kernel.get(), (_axis == 0 ? Window::DimY : Window::DimZ));
142
143 for(unsigned int i = 0; i < _num_ffts; ++i)
144 {
145 NEScheduler::get().schedule(_fft_kernels[i].get(), (_axis == 0 ? Window::DimY : Window::DimX));
146 }
147
148 // Run output scaling
149 if(_run_scale)
150 {
151 NEScheduler::get().schedule(_scale_kernel.get(), Window::DimY);
152 }
153 }
154 } // namespace arm_compute
155