Debian Codegraph @

modules/lsp-dsp-lib/include/private/dsp/arch/arm/neon-d32/fft.h

raw
1 /*
2 * Copyright (C) 2026 Linux Studio Plugins Project <https://lsp-plug.in/>
3 * (C) 2026 Vladimir Sadovnikov <[email protected]>
4 *
5 * This file is part of lsp-dsp-lib
6 * Created on: 31 мар. 2020 г.
7 *
8 * lsp-dsp-lib is free software: you can redistribute it and/or modify
9 * it under the terms of the GNU Lesser General Public License as published by
10 * the Free Software Foundation, either version 3 of the License, or
11 * any later version.
12 *
13 * lsp-dsp-lib is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU Lesser General Public License for more details.
17 *
18 * You should have received a copy of the GNU Lesser General Public License
19 * along with lsp-dsp-lib. If not, see <https://www.gnu.org/licenses/>.
20 */
21
22 #ifndef PRIVATE_DSP_ARCH_ARM_NEON_D32_FFT_H_
23 #define PRIVATE_DSP_ARCH_ARM_NEON_D32_FFT_H_
24
25 #ifndef PRIVATE_DSP_ARCH_ARM_NEON_D32_IMPL
26 #error "This header should not be included directly"
27 #endif /* PRIVATE_DSP_ARCH_ARM_NEON_D32_IMPL */
28
29 #include <private/dsp/arch/arm/neon-d32/fft/const.h>
30 #include <private/dsp/arch/arm/neon-d32/fft/scramble.h>
31 #include <private/dsp/arch/arm/neon-d32/fft/pscramble.h>
32 #include <private/dsp/arch/arm/neon-d32/fft/butterfly.h>
33 #include <private/dsp/arch/arm/neon-d32/fft/pbutterfly.h>
34 #include <private/dsp/arch/arm/neon-d32/fft/normalize.h>
35
36 namespace lsp
37 {
38 namespace neon_d32
39 {
40 void direct_fft(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t rank)
41 {
42 // Check bounds
43 if (rank <= 2)
44 {
45 if (rank == 2)
46 {
47 float s0_re = src_re[0] + src_re[2];
48 float s1_re = src_re[0] - src_re[2];
49 float s2_re = src_re[1] + src_re[3];
50 float s3_re = src_re[1] - src_re[3];
51
52 float s0_im = src_im[0] + src_im[2];
53 float s1_im = src_im[0] - src_im[2];
54 float s2_im = src_im[1] + src_im[3];
55 float s3_im = src_im[1] - src_im[3];
56
57 dst_re[0] = s0_re + s2_re;
58 dst_re[1] = s1_re + s3_im;
59 dst_re[2] = s0_re - s2_re;
60 dst_re[3] = s1_re - s3_im;
61
62 dst_im[0] = s0_im + s2_im;
63 dst_im[1] = s1_im - s3_re;
64 dst_im[2] = s0_im - s2_im;
65 dst_im[3] = s1_im + s3_re;
66 }
67 else if (rank == 1)
68 {
69 // s0' = s0 + s1
70 // s1' = s0 - s1
71 float s1_re = src_re[1];
72 float s1_im = src_im[1];
73 dst_re[1] = src_re[0] - s1_re;
74 dst_im[1] = src_im[0] - s1_im;
75 dst_re[0] = src_re[0] + s1_re;
76 dst_im[0] = src_im[0] + s1_im;
77 }
78 else
79 {
80 dst_re[0] = src_re[0];
81 dst_im[0] = src_im[0];
82 }
83 return;
84 }
85
86 // Scramble data
87 scramble_direct(dst_re, dst_im, src_re, src_im, rank);
88 direct_butterfly_rank3(dst_re, dst_im, 1 << (rank-3));
89
90 for (size_t i=4; i <= rank; ++i)
91 direct_butterfly_rank4p(dst_re, dst_im, i, 1 << (rank - i));
92 }
93
94 void reverse_fft(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t rank)
95 {
96 // Check bounds
97 if (rank <= 2)
98 {
99 if (rank == 2)
100 {
101 float s0_re = src_re[0] + src_re[2];
102 float s1_re = src_re[0] - src_re[2];
103 float s2_re = src_re[1] + src_re[3];
104 float s3_re = src_re[1] - src_re[3];
105
106 float s0_im = src_im[0] + src_im[2];
107 float s1_im = src_im[0] - src_im[2];
108 float s2_im = src_im[1] + src_im[3];
109 float s3_im = src_im[1] - src_im[3];
110
111 dst_re[0] = (s0_re + s2_re) * 0.25f;
112 dst_re[1] = (s1_re - s3_im) * 0.25f;
113 dst_re[2] = (s0_re - s2_re) * 0.25f;
114 dst_re[3] = (s1_re + s3_im) * 0.25f;
115
116 dst_im[0] = (s0_im + s2_im) * 0.25f;
117 dst_im[1] = (s1_im + s3_re) * 0.25f;
118 dst_im[2] = (s0_im - s2_im) * 0.25f;
119 dst_im[3] = (s1_im - s3_re) * 0.25f;
120 }
121 else if (rank == 1)
122 {
123 // s0' = s0 + s1
124 // s1' = s0 - s1
125 float s1_re = src_re[1];
126 float s1_im = src_im[1];
127 dst_re[1] = (src_re[0] - s1_re) * 0.5f;
128 dst_im[1] = (src_im[0] - s1_im) * 0.5f;
129 dst_re[0] = (src_re[0] + s1_re) * 0.5f;
130 dst_im[0] = (src_im[0] + s1_im) * 0.5f;
131 }
132 else
133 {
134 dst_re[0] = src_re[0];
135 dst_im[0] = src_im[0];
136 }
137 return;
138 }
139
140 // Scramble data
141 scramble_reverse(dst_re, dst_im, src_re, src_im, rank);
142 reverse_butterfly_rank3(dst_re, dst_im, 1 << (rank-3));
143
144 for (size_t i=4; i <= rank; ++i)
145 reverse_butterfly_rank4p(dst_re, dst_im, i, 1 << (rank - i));
146
147 dsp::normalize_fft2(dst_re, dst_im, rank);
148 }
149
150 void packed_direct_fft(float *dst, const float *src, size_t rank)
151 {
152 // Check bounds
153 if (rank <= 2)
154 {
155 if (rank == 2)
156 {
157 float s0_re = src[0] + src[4];
158 float s1_re = src[0] - src[4];
159 float s0_im = src[1] + src[5];
160 float s1_im = src[1] - src[5];
161
162 float s2_re = src[2] + src[6];
163 float s3_re = src[2] - src[6];
164 float s2_im = src[3] + src[7];
165 float s3_im = src[3] - src[7];
166
167 dst[0] = s0_re + s2_re;
168 dst[1] = s0_im + s2_im;
169 dst[2] = s1_re + s3_im;
170 dst[3] = s1_im - s3_re;
171
172 dst[4] = s0_re - s2_re;
173 dst[5] = s0_im - s2_im;
174 dst[6] = s1_re - s3_im;
175 dst[7] = s1_im + s3_re;
176 }
177 else if (rank == 1)
178 {
179 // s0' = s0 + s1
180 // s1' = s0 - s1
181 float s1_re = src[2];
182 float s1_im = src[3];
183 dst[2] = src[0] - s1_re;
184 dst[3] = src[1] - s1_im;
185 dst[0] = src[0] + s1_re;
186 dst[1] = src[1] + s1_im;
187 }
188 else
189 {
190 dst[0] = src[0];
191 dst[1] = src[1];
192 }
193 return;
194 }
195
196 packed_scramble_direct(dst, src, rank);
197 packed_direct_butterfly_rank3(dst, 1 << (rank-3));
198
199 for (size_t i=4; i <= rank; ++i)
200 packed_direct_butterfly_rank4p(dst, i, 1 << (rank - i));
201
202 packed_unscramble_direct(dst, rank);
203 }
204
205 void packed_reverse_fft(float *dst, const float *src, size_t rank)
206 {
207 // Check bounds
208 if (rank <= 2)
209 {
210 if (rank == 2)
211 {
212 float s0_re = src[0] + src[4];
213 float s1_re = src[0] - src[4];
214 float s0_im = src[1] + src[5];
215 float s1_im = src[1] - src[5];
216
217 float s2_re = src[2] + src[6];
218 float s3_re = src[2] - src[6];
219 float s2_im = src[3] + src[7];
220 float s3_im = src[3] - src[7];
221
222 dst[0] = (s0_re + s2_re)*0.25f;
223 dst[1] = (s0_im + s2_im)*0.25f;
224 dst[2] = (s1_re - s3_im)*0.25f;
225 dst[3] = (s1_im + s3_re)*0.25f;
226
227 dst[4] = (s0_re - s2_re)*0.25f;
228 dst[5] = (s0_im - s2_im)*0.25f;
229 dst[6] = (s1_re + s3_im)*0.25f;
230 dst[7] = (s1_im - s3_re)*0.25f;
231 }
232 else if (rank == 1)
233 {
234 // s0' = s0 + s1
235 // s1' = s0 - s1
236 float s1_re = src[2];
237 float s1_im = src[3];
238 dst[2] = src[0] - s1_re;
239 dst[3] = src[1] - s1_im;
240 dst[0] = src[0] + s1_re;
241 dst[1] = src[1] + s1_im;
242 }
243 else
244 {
245 dst[0] = src[0];
246 dst[1] = src[1];
247 }
248 return;
249 }
250
251 packed_scramble_reverse(dst, src, rank);
252 packed_reverse_butterfly_rank3(dst, 1 << (rank-3));
253
254 for (size_t i=4; i <= rank; ++i)
255 packed_reverse_butterfly_rank4p(dst, i, 1 << (rank - i));
256
257 packed_unscramble_reverse(dst, rank);
258 }
259
260 } /* namespace neon_d32 */
261 } /* namespace lsp */
262
263 #endif /* PRIVATE_DSP_ARCH_ARM_NEON_D32_FFT_H_ */