|
10153
|
1 /*
|
|
|
2 * ARM NEON optimised FFT
|
|
|
3 *
|
|
|
4 * Copyright (c) 2009 Mans Rullgard <mans@mansr.com>
|
|
|
5 * Copyright (c) 2009 Naotoshi Nojiri
|
|
|
6 *
|
|
12188
|
7 * This algorithm (though not any of the implementation details) is
|
|
|
8 * based on libdjbfft by D. J. Bernstein.
|
|
|
9 *
|
|
10153
|
10 * This file is part of FFmpeg.
|
|
|
11 *
|
|
|
12 * FFmpeg is free software; you can redistribute it and/or
|
|
|
13 * modify it under the terms of the GNU Lesser General Public
|
|
|
14 * License as published by the Free Software Foundation; either
|
|
|
15 * version 2.1 of the License, or (at your option) any later version.
|
|
|
16 *
|
|
|
17 * FFmpeg is distributed in the hope that it will be useful,
|
|
|
18 * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
20 * Lesser General Public License for more details.
|
|
|
21 *
|
|
|
22 * You should have received a copy of the GNU Lesser General Public
|
|
|
23 * License along with FFmpeg; if not, write to the Free Software
|
|
|
24 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
|
|
25 */
|
|
|
26
|
|
|
27 #include "asm.S"
|
|
|
28
|
|
|
29 #define M_SQRT1_2 0.70710678118654752440
|
|
|
30
|
|
|
31 .text
|
|
|
32
|
|
|
33 function fft4_neon
|
|
|
34 vld1.32 {d0-d3}, [r0,:128]
|
|
|
35
|
|
|
36 vext.32 q8, q1, q1, #1 @ i2,r3 d3=i3,r2
|
|
|
37 vsub.f32 d6, d0, d1 @ r0-r1,i0-i1
|
|
|
38 vsub.f32 d7, d16, d17 @ r3-r2,i2-i3
|
|
|
39 vadd.f32 d4, d0, d1 @ r0+r1,i0+i1
|
|
|
40 vadd.f32 d5, d2, d3 @ i2+i3,r2+r3
|
|
|
41 vadd.f32 d1, d6, d7
|
|
|
42 vsub.f32 d3, d6, d7
|
|
|
43 vadd.f32 d0, d4, d5
|
|
|
44 vsub.f32 d2, d4, d5
|
|
|
45
|
|
|
46 vst1.32 {d0-d3}, [r0,:128]
|
|
|
47
|
|
|
48 bx lr
|
|
11443
|
49 endfunc
|
|
10153
|
50
|
|
|
51 function fft8_neon
|
|
|
52 mov r1, r0
|
|
|
53 vld1.32 {d0-d3}, [r1,:128]!
|
|
|
54 vld1.32 {d16-d19}, [r1,:128]
|
|
|
55
|
|
|
56 movw r2, #0x04f3 @ sqrt(1/2)
|
|
|
57 movt r2, #0x3f35
|
|
|
58 eor r3, r2, #1<<31
|
|
|
59 vdup.32 d31, r2
|
|
|
60
|
|
|
61 vext.32 q11, q1, q1, #1 @ i2,r3,i3,r2
|
|
|
62 vadd.f32 d4, d16, d17 @ r4+r5,i4+i5
|
|
|
63 vmov d28, r3, r2
|
|
|
64 vadd.f32 d5, d18, d19 @ r6+r7,i6+i7
|
|
|
65 vsub.f32 d17, d16, d17 @ r4-r5,i4-i5
|
|
|
66 vsub.f32 d19, d18, d19 @ r6-r7,i6-i7
|
|
|
67 vrev64.32 d29, d28
|
|
|
68 vadd.f32 d20, d0, d1 @ r0+r1,i0+i1
|
|
|
69 vadd.f32 d21, d2, d3 @ r2+r3,i2+i3
|
|
|
70 vmul.f32 d26, d17, d28 @ -a2r*w,a2i*w
|
|
|
71 vext.32 q3, q2, q2, #1
|
|
|
72 vmul.f32 d27, d19, d29 @ a3r*w,-a3i*w
|
|
|
73 vsub.f32 d23, d22, d23 @ i2-i3,r3-r2
|
|
|
74 vsub.f32 d22, d0, d1 @ r0-r1,i0-i1
|
|
|
75 vmul.f32 d24, d17, d31 @ a2r*w,a2i*w
|
|
|
76 vmul.f32 d25, d19, d31 @ a3r*w,a3i*w
|
|
|
77 vadd.f32 d0, d20, d21
|
|
|
78 vsub.f32 d2, d20, d21
|
|
|
79 vadd.f32 d1, d22, d23
|
|
|
80 vrev64.32 q13, q13
|
|
|
81 vsub.f32 d3, d22, d23
|
|
|
82 vsub.f32 d6, d6, d7
|
|
|
83 vadd.f32 d24, d24, d26 @ a2r+a2i,a2i-a2r t1,t2
|
|
|
84 vadd.f32 d25, d25, d27 @ a3r-a3i,a3i+a3r t5,t6
|
|
|
85 vadd.f32 d7, d4, d5
|
|
|
86 vsub.f32 d18, d2, d6
|
|
|
87 vext.32 q13, q12, q12, #1
|
|
|
88 vadd.f32 d2, d2, d6
|
|
|
89 vsub.f32 d16, d0, d7
|
|
|
90 vadd.f32 d5, d25, d24
|
|
|
91 vsub.f32 d4, d26, d27
|
|
|
92 vadd.f32 d0, d0, d7
|
|
|
93 vsub.f32 d17, d1, d5
|
|
|
94 vsub.f32 d19, d3, d4
|
|
|
95 vadd.f32 d3, d3, d4
|
|
|
96 vadd.f32 d1, d1, d5
|
|
|
97
|
|
|
98 vst1.32 {d16-d19}, [r1,:128]
|
|
|
99 vst1.32 {d0-d3}, [r0,:128]
|
|
|
100
|
|
|
101 bx lr
|
|
11443
|
102 endfunc
|
|
10153
|
103
|
|
|
104 function fft16_neon
|
|
|
105 movrel r1, mppm
|
|
|
106 vld1.32 {d16-d19}, [r0,:128]! @ q8{r0,i0,r1,i1} q9{r2,i2,r3,i3}
|
|
|
107 pld [r0, #32]
|
|
|
108 vld1.32 {d2-d3}, [r1,:128]
|
|
|
109 vext.32 q13, q9, q9, #1
|
|
|
110 vld1.32 {d22-d25}, [r0,:128]! @ q11{r4,i4,r5,i5} q12{r6,i5,r7,i7}
|
|
|
111 vadd.f32 d4, d16, d17
|
|
|
112 vsub.f32 d5, d16, d17
|
|
|
113 vadd.f32 d18, d18, d19
|
|
|
114 vsub.f32 d19, d26, d27
|
|
|
115
|
|
|
116 vadd.f32 d20, d22, d23
|
|
|
117 vsub.f32 d22, d22, d23
|
|
|
118 vsub.f32 d23, d24, d25
|
|
|
119 vadd.f32 q8, q2, q9 @ {r0,i0,r1,i1}
|
|
|
120 vadd.f32 d21, d24, d25
|
|
|
121 vmul.f32 d24, d22, d2
|
|
|
122 vsub.f32 q9, q2, q9 @ {r2,i2,r3,i3}
|
|
|
123 vmul.f32 d25, d23, d3
|
|
|
124 vuzp.32 d16, d17 @ {r0,r1,i0,i1}
|
|
|
125 vmul.f32 q1, q11, d2[1]
|
|
|
126 vuzp.32 d18, d19 @ {r2,r3,i2,i3}
|
|
|
127 vrev64.32 q12, q12
|
|
|
128 vadd.f32 q11, q12, q1 @ {t1a,t2a,t5,t6}
|
|
|
129 vld1.32 {d24-d27}, [r0,:128]! @ q12{r8,i8,r9,i9} q13{r10,i10,r11,i11}
|
|
|
130 vzip.32 q10, q11
|
|
|
131 vld1.32 {d28-d31}, [r0,:128] @ q14{r12,i12,r13,i13} q15{r14,i14,r15,i15}
|
|
|
132 vadd.f32 d0, d22, d20
|
|
|
133 vadd.f32 d1, d21, d23
|
|
|
134 vsub.f32 d2, d21, d23
|
|
|
135 vsub.f32 d3, d22, d20
|
|
|
136 sub r0, r0, #96
|
|
|
137 vext.32 q13, q13, q13, #1
|
|
|
138 vsub.f32 q10, q8, q0 @ {r4,r5,i4,i5}
|
|
|
139 vadd.f32 q8, q8, q0 @ {r0,r1,i0,i1}
|
|
|
140 vext.32 q15, q15, q15, #1
|
|
|
141 vsub.f32 q11, q9, q1 @ {r6,r7,i6,i7}
|
|
|
142 vswp d25, d26 @ q12{r8,i8,i10,r11} q13{r9,i9,i11,r10}
|
|
|
143 vadd.f32 q9, q9, q1 @ {r2,r3,i2,i3}
|
|
|
144 vswp d29, d30 @ q14{r12,i12,i14,r15} q15{r13,i13,i15,r14}
|
|
|
145 vadd.f32 q0, q12, q13 @ {t1,t2,t5,t6}
|
|
|
146 vadd.f32 q1, q14, q15 @ {t1a,t2a,t5a,t6a}
|
|
10346
|
147 movrel r2, X(ff_cos_16)
|
|
10153
|
148 vsub.f32 q13, q12, q13 @ {t3,t4,t7,t8}
|
|
|
149 vrev64.32 d1, d1
|
|
|
150 vsub.f32 q15, q14, q15 @ {t3a,t4a,t7a,t8a}
|
|
|
151 vrev64.32 d3, d3
|
|
|
152 movrel r3, pmmp
|
|
|
153 vswp d1, d26 @ q0{t1,t2,t3,t4} q13{t6,t5,t7,t8}
|
|
|
154 vswp d3, d30 @ q1{t1a,t2a,t3a,t4a} q15{t6a,t5a,t7a,t8a}
|
|
|
155 vadd.f32 q12, q0, q13 @ {r8,i8,r9,i9}
|
|
|
156 vadd.f32 q14, q1, q15 @ {r12,i12,r13,i13}
|
|
|
157 vld1.32 {d4-d5}, [r2,:64]
|
|
|
158 vsub.f32 q13, q0, q13 @ {r10,i10,r11,i11}
|
|
|
159 vsub.f32 q15, q1, q15 @ {r14,i14,r15,i15}
|
|
|
160 vswp d25, d28 @ q12{r8,i8,r12,i12} q14{r9,i9,r13,i13}
|
|
|
161 vld1.32 {d6-d7}, [r3,:128]
|
|
|
162 vrev64.32 q1, q14
|
|
|
163 vmul.f32 q14, q14, d4[1]
|
|
|
164 vmul.f32 q1, q1, q3
|
|
|
165 vmla.f32 q14, q1, d5[1] @ {t1a,t2a,t5a,t6a}
|
|
|
166 vswp d27, d30 @ q13{r10,i10,r14,i14} q15{r11,i11,r15,i15}
|
|
|
167 vzip.32 q12, q14
|
|
|
168 vadd.f32 d0, d28, d24
|
|
|
169 vadd.f32 d1, d25, d29
|
|
|
170 vsub.f32 d2, d25, d29
|
|
|
171 vsub.f32 d3, d28, d24
|
|
|
172 vsub.f32 q12, q8, q0 @ {r8,r9,i8,i9}
|
|
|
173 vadd.f32 q8, q8, q0 @ {r0,r1,i0,i1}
|
|
|
174 vsub.f32 q14, q10, q1 @ {r12,r13,i12,i13}
|
|
|
175 mov r1, #32
|
|
|
176 vadd.f32 q10, q10, q1 @ {r4,r5,i4,i5}
|
|
|
177 vrev64.32 q0, q13
|
|
|
178 vmul.f32 q13, q13, d5[0]
|
|
|
179 vrev64.32 q1, q15
|
|
|
180 vmul.f32 q15, q15, d5[1]
|
|
|
181 vst2.32 {d16-d17},[r0,:128], r1
|
|
|
182 vmul.f32 q0, q0, q3
|
|
|
183 vst2.32 {d20-d21},[r0,:128], r1
|
|
|
184 vmul.f32 q1, q1, q3
|
|
|
185 vmla.f32 q13, q0, d5[0] @ {t1,t2,t5,t6}
|
|
|
186 vmla.f32 q15, q1, d4[1] @ {t1a,t2a,t5a,t6a}
|
|
|
187 vst2.32 {d24-d25},[r0,:128], r1
|
|
|
188 vst2.32 {d28-d29},[r0,:128]
|
|
|
189 vzip.32 q13, q15
|
|
|
190 sub r0, r0, #80
|
|
|
191 vadd.f32 d0, d30, d26
|
|
|
192 vadd.f32 d1, d27, d31
|
|
|
193 vsub.f32 d2, d27, d31
|
|
|
194 vsub.f32 d3, d30, d26
|
|
|
195 vsub.f32 q13, q9, q0 @ {r10,r11,i10,i11}
|
|
|
196 vadd.f32 q9, q9, q0 @ {r2,r3,i2,i3}
|
|
|
197 vsub.f32 q15, q11, q1 @ {r14,r15,i14,i15}
|
|
|
198 vadd.f32 q11, q11, q1 @ {r6,r7,i6,i7}
|
|
|
199 vst2.32 {d18-d19},[r0,:128], r1
|
|
|
200 vst2.32 {d22-d23},[r0,:128], r1
|
|
|
201 vst2.32 {d26-d27},[r0,:128], r1
|
|
|
202 vst2.32 {d30-d31},[r0,:128]
|
|
|
203 bx lr
|
|
11443
|
204 endfunc
|
|
10153
|
205
|
|
|
206 function fft_pass_neon
|
|
|
207 push {r4-r6,lr}
|
|
|
208 mov r6, r2 @ n
|
|
|
209 lsl r5, r2, #3 @ 2 * n * sizeof FFTSample
|
|
|
210 lsl r4, r2, #4 @ 2 * n * sizeof FFTComplex
|
|
|
211 lsl r2, r2, #5 @ 4 * n * sizeof FFTComplex
|
|
|
212 add r3, r2, r4
|
|
|
213 add r4, r4, r0 @ &z[o1]
|
|
|
214 add r2, r2, r0 @ &z[o2]
|
|
|
215 add r3, r3, r0 @ &z[o3]
|
|
|
216 vld1.32 {d20-d21},[r2,:128] @ {z[o2],z[o2+1]}
|
|
|
217 movrel r12, pmmp
|
|
|
218 vld1.32 {d22-d23},[r3,:128] @ {z[o3],z[o3+1]}
|
|
|
219 add r5, r5, r1 @ wim
|
|
|
220 vld1.32 {d6-d7}, [r12,:128] @ pmmp
|
|
|
221 vswp d21, d22
|
|
|
222 vld1.32 {d4}, [r1,:64]! @ {wre[0],wre[1]}
|
|
|
223 sub r5, r5, #4 @ wim--
|
|
|
224 vrev64.32 q1, q11
|
|
|
225 vmul.f32 q11, q11, d4[1]
|
|
|
226 vmul.f32 q1, q1, q3
|
|
|
227 vld1.32 {d5[0]}, [r5,:32] @ d5[0] = wim[-1]
|
|
|
228 vmla.f32 q11, q1, d5[0] @ {t1a,t2a,t5a,t6a}
|
|
|
229 vld2.32 {d16-d17},[r0,:128] @ {z[0],z[1]}
|
|
|
230 sub r6, r6, #1 @ n--
|
|
|
231 vld2.32 {d18-d19},[r4,:128] @ {z[o1],z[o1+1]}
|
|
|
232 vzip.32 q10, q11
|
|
|
233 vadd.f32 d0, d22, d20
|
|
|
234 vadd.f32 d1, d21, d23
|
|
|
235 vsub.f32 d2, d21, d23
|
|
|
236 vsub.f32 d3, d22, d20
|
|
|
237 vsub.f32 q10, q8, q0
|
|
|
238 vadd.f32 q8, q8, q0
|
|
|
239 vsub.f32 q11, q9, q1
|
|
|
240 vadd.f32 q9, q9, q1
|
|
|
241 vst2.32 {d20-d21},[r2,:128]! @ {z[o2],z[o2+1]}
|
|
|
242 vst2.32 {d16-d17},[r0,:128]! @ {z[0],z[1]}
|
|
|
243 vst2.32 {d22-d23},[r3,:128]! @ {z[o3],z[o3+1]}
|
|
|
244 vst2.32 {d18-d19},[r4,:128]! @ {z[o1],z[o1+1]}
|
|
|
245 sub r5, r5, #8 @ wim -= 2
|
|
|
246 1:
|
|
|
247 vld1.32 {d20-d21},[r2,:128] @ {z[o2],z[o2+1]}
|
|
|
248 vld1.32 {d22-d23},[r3,:128] @ {z[o3],z[o3+1]}
|
|
|
249 vswp d21, d22
|
|
|
250 vld1.32 {d4}, [r1]! @ {wre[0],wre[1]}
|
|
|
251 vrev64.32 q0, q10
|
|
|
252 vmul.f32 q10, q10, d4[0]
|
|
|
253 vrev64.32 q1, q11
|
|
|
254 vmul.f32 q11, q11, d4[1]
|
|
|
255 vld1.32 {d5}, [r5] @ {wim[-1],wim[0]}
|
|
|
256 vmul.f32 q0, q0, q3
|
|
|
257 sub r5, r5, #8 @ wim -= 2
|
|
|
258 vmul.f32 q1, q1, q3
|
|
|
259 vmla.f32 q10, q0, d5[1] @ {t1,t2,t5,t6}
|
|
|
260 vmla.f32 q11, q1, d5[0] @ {t1a,t2a,t5a,t6a}
|
|
|
261 vld2.32 {d16-d17},[r0,:128] @ {z[0],z[1]}
|
|
|
262 subs r6, r6, #1 @ n--
|
|
|
263 vld2.32 {d18-d19},[r4,:128] @ {z[o1],z[o1+1]}
|
|
|
264 vzip.32 q10, q11
|
|
|
265 vadd.f32 d0, d22, d20
|
|
|
266 vadd.f32 d1, d21, d23
|
|
|
267 vsub.f32 d2, d21, d23
|
|
|
268 vsub.f32 d3, d22, d20
|
|
|
269 vsub.f32 q10, q8, q0
|
|
|
270 vadd.f32 q8, q8, q0
|
|
|
271 vsub.f32 q11, q9, q1
|
|
|
272 vadd.f32 q9, q9, q1
|
|
|
273 vst2.32 {d20-d21}, [r2,:128]! @ {z[o2],z[o2+1]}
|
|
|
274 vst2.32 {d16-d17}, [r0,:128]! @ {z[0],z[1]}
|
|
|
275 vst2.32 {d22-d23}, [r3,:128]! @ {z[o3],z[o3+1]}
|
|
|
276 vst2.32 {d18-d19}, [r4,:128]! @ {z[o1],z[o1+1]}
|
|
|
277 bne 1b
|
|
|
278
|
|
|
279 pop {r4-r6,pc}
|
|
11443
|
280 endfunc
|
|
10153
|
281
|
|
|
282 .macro def_fft n, n2, n4
|
|
|
283 .align 6
|
|
|
284 function fft\n\()_neon
|
|
|
285 push {r4, lr}
|
|
|
286 mov r4, r0
|
|
|
287 bl fft\n2\()_neon
|
|
|
288 add r0, r4, #\n4*2*8
|
|
|
289 bl fft\n4\()_neon
|
|
|
290 add r0, r4, #\n4*3*8
|
|
|
291 bl fft\n4\()_neon
|
|
|
292 mov r0, r4
|
|
|
293 pop {r4, lr}
|
|
10346
|
294 movrel r1, X(ff_cos_\n)
|
|
10153
|
295 mov r2, #\n4/2
|
|
|
296 b fft_pass_neon
|
|
11443
|
297 endfunc
|
|
10153
|
298 .endm
|
|
|
299
|
|
|
300 def_fft 32, 16, 8
|
|
|
301 def_fft 64, 32, 16
|
|
|
302 def_fft 128, 64, 32
|
|
|
303 def_fft 256, 128, 64
|
|
|
304 def_fft 512, 256, 128
|
|
|
305 def_fft 1024, 512, 256
|
|
|
306 def_fft 2048, 1024, 512
|
|
|
307 def_fft 4096, 2048, 1024
|
|
|
308 def_fft 8192, 4096, 2048
|
|
|
309 def_fft 16384, 8192, 4096
|
|
|
310 def_fft 32768, 16384, 8192
|
|
|
311 def_fft 65536, 32768, 16384
|
|
|
312
|
|
|
313 function ff_fft_calc_neon, export=1
|
|
|
314 ldr r2, [r0]
|
|
|
315 sub r2, r2, #2
|
|
|
316 movrel r3, fft_tab_neon
|
|
|
317 ldr r3, [r3, r2, lsl #2]
|
|
|
318 mov r0, r1
|
|
|
319 bx r3
|
|
11443
|
320 endfunc
|
|
10153
|
321
|
|
|
322 function ff_fft_permute_neon, export=1
|
|
|
323 push {r4,lr}
|
|
|
324 mov r12, #1
|
|
|
325 ldr r2, [r0] @ nbits
|
|
12047
|
326 ldr r3, [r0, #12] @ tmp_buf
|
|
10153
|
327 ldr r0, [r0, #8] @ revtab
|
|
|
328 lsl r12, r12, r2
|
|
|
329 mov r2, r12
|
|
|
330 1:
|
|
|
331 vld1.32 {d0-d1}, [r1,:128]!
|
|
|
332 ldr r4, [r0], #4
|
|
10172
|
333 uxth lr, r4
|
|
|
334 uxth r4, r4, ror #16
|
|
|
335 add lr, r3, lr, lsl #3
|
|
|
336 add r4, r3, r4, lsl #3
|
|
10153
|
337 vst1.32 {d0}, [lr,:64]
|
|
|
338 vst1.32 {d1}, [r4,:64]
|
|
|
339 subs r12, r12, #2
|
|
|
340 bgt 1b
|
|
|
341
|
|
|
342 sub r1, r1, r2, lsl #3
|
|
|
343 1:
|
|
|
344 vld1.32 {d0-d3}, [r3,:128]!
|
|
|
345 vst1.32 {d0-d3}, [r1,:128]!
|
|
|
346 subs r2, r2, #4
|
|
|
347 bgt 1b
|
|
|
348
|
|
|
349 pop {r4,pc}
|
|
11443
|
350 endfunc
|
|
10153
|
351
|
|
|
352 .section .rodata
|
|
|
353 .align 4
|
|
|
354 fft_tab_neon:
|
|
|
355 .word fft4_neon
|
|
|
356 .word fft8_neon
|
|
|
357 .word fft16_neon
|
|
|
358 .word fft32_neon
|
|
|
359 .word fft64_neon
|
|
|
360 .word fft128_neon
|
|
|
361 .word fft256_neon
|
|
|
362 .word fft512_neon
|
|
|
363 .word fft1024_neon
|
|
|
364 .word fft2048_neon
|
|
|
365 .word fft4096_neon
|
|
|
366 .word fft8192_neon
|
|
|
367 .word fft16384_neon
|
|
|
368 .word fft32768_neon
|
|
|
369 .word fft65536_neon
|
|
12104
|
370 ELF .size fft_tab_neon, . - fft_tab_neon
|
|
10153
|
371
|
|
|
372 .align 4
|
|
|
373 pmmp: .float +1.0, -1.0, -1.0, +1.0
|
|
|
374 mppm: .float -M_SQRT1_2, M_SQRT1_2, M_SQRT1_2, -M_SQRT1_2
|
