|
12046
|
1 /*
|
|
|
2 * FFT transform with Altivec optimizations
|
|
|
3 * Copyright (c) 2009 Loren Merritt
|
|
|
4 *
|
|
|
5 * This file is part of FFmpeg.
|
|
|
6 *
|
|
|
7 * FFmpeg is free software; you can redistribute it and/or
|
|
|
8 * modify it under the terms of the GNU Lesser General Public
|
|
|
9 * License as published by the Free Software Foundation; either
|
|
|
10 * version 2.1 of the License, or (at your option) any later version.
|
|
|
11 *
|
|
|
12 * FFmpeg is distributed in the hope that it will be useful,
|
|
|
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
15 * Lesser General Public License for more details.
|
|
|
16 *
|
|
|
17 * You should have received a copy of the GNU Lesser General Public
|
|
|
18 * License along with FFmpeg; if not, write to the Free Software
|
|
|
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
|
|
20 */
|
|
|
21
|
|
|
22 /*
|
|
|
23 * These functions are not individually interchangeable with the C versions.
|
|
|
24 * While C takes arrays of FFTComplex, Altivec leaves intermediate results
|
|
|
25 * in blocks as convenient to the vector size.
|
|
|
26 * i.e. {4x real, 4x imaginary, 4x real, ...}
|
|
|
27 *
|
|
|
28 * I ignore standard calling convention.
|
|
|
29 * Instead, the following registers are treated as global constants:
|
|
|
30 * v14: zero
|
|
|
31 * v15..v18: cosines
|
|
|
32 * v19..v29: permutations
|
|
|
33 * r9: 16
|
|
|
34 * r12: ff_cos_tabs
|
|
|
35 * and the rest are free for local use.
|
|
|
36 */
|
|
|
37
|
|
|
38 #include "config.h"
|
|
|
39 #include "asm.S"
|
|
|
40
|
|
|
41 .text
|
|
|
42
|
|
|
43 /* Apple gas doesn't support this shorthand */
|
|
|
44 .macro mtctr rx
|
|
|
45 mtspr 9, \rx
|
|
|
46 .endm
|
|
|
47
|
|
|
48 .macro addi2 ra, imm // add 32-bit immediate
|
|
|
49 .if \imm & 0xffff
|
|
|
50 addi \ra, \ra, \imm@l
|
|
|
51 .endif
|
|
|
52 .if (\imm+0x8000)>>16
|
|
|
53 addis \ra, \ra, \imm@ha
|
|
|
54 .endif
|
|
|
55 .endm
|
|
|
56
|
|
|
57 #if ARCH_PPC64
|
|
|
58 #define PTR .quad
|
|
|
59 .macro LOAD_PTR ra, rbase, offset
|
|
|
60 ld \ra,(\offset)*8(\rbase)
|
|
|
61 .endm
|
|
|
62 .macro STORE_PTR ra, rbase, offset
|
|
|
63 std \ra,(\offset)*8(\rbase)
|
|
|
64 .endm
|
|
|
65 #else
|
|
|
66 #define PTR .int
|
|
|
67 .macro LOAD_PTR ra, rbase, offset
|
|
|
68 lwz \ra,(\offset)*4(\rbase)
|
|
|
69 .endm
|
|
|
70 .macro STORE_PTR ra, rbase, offset
|
|
|
71 stw \ra,(\offset)*4(\rbase)
|
|
|
72 .endm
|
|
|
73 #endif
|
|
|
74
|
|
|
75 .macro FFT4 a0, a1, a2, a3 // in:0-1 out:2-3
|
|
|
76 vperm \a2,\a0,\a1,v20 // vcprm(0,1,s2,s1) // {r0,i0,r3,i2}
|
|
|
77 vperm \a3,\a0,\a1,v21 // vcprm(2,3,s0,s3) // {r1,i1,r2,i3}
|
|
|
78 vaddfp \a0,\a2,\a3 // {t1,t2,t6,t5}
|
|
|
79 vsubfp \a1,\a2,\a3 // {t3,t4,t8,t7}
|
|
|
80 vmrghw \a2,\a0,\a1 // vcprm(0,s0,1,s1) // {t1,t3,t2,t4}
|
|
|
81 vperm \a3,\a0,\a1,v22 // vcprm(2,s3,3,s2) // {t6,t7,t5,t8}
|
|
|
82 vaddfp \a0,\a2,\a3 // {r0,r1,i0,i1}
|
|
|
83 vsubfp \a1,\a2,\a3 // {r2,r3,i2,i3}
|
|
|
84 vperm \a2,\a0,\a1,v23 // vcprm(0,1,s0,s1) // {r0,r1,r2,r3}
|
|
|
85 vperm \a3,\a0,\a1,v24 // vcprm(2,3,s2,s3) // {i0,i1,i2,i3}
|
|
|
86 .endm
|
|
|
87
|
|
|
88 .macro FFT4x2 a0, a1, b0, b1, a2, a3, b2, b3
|
|
|
89 vperm \a2,\a0,\a1,v20 // vcprm(0,1,s2,s1) // {r0,i0,r3,i2}
|
|
|
90 vperm \a3,\a0,\a1,v21 // vcprm(2,3,s0,s3) // {r1,i1,r2,i3}
|
|
|
91 vperm \b2,\b0,\b1,v20
|
|
|
92 vperm \b3,\b0,\b1,v21
|
|
|
93 vaddfp \a0,\a2,\a3 // {t1,t2,t6,t5}
|
|
|
94 vsubfp \a1,\a2,\a3 // {t3,t4,t8,t7}
|
|
|
95 vaddfp \b0,\b2,\b3
|
|
|
96 vsubfp \b1,\b2,\b3
|
|
|
97 vmrghw \a2,\a0,\a1 // vcprm(0,s0,1,s1) // {t1,t3,t2,t4}
|
|
|
98 vperm \a3,\a0,\a1,v22 // vcprm(2,s3,3,s2) // {t6,t7,t5,t8}
|
|
|
99 vmrghw \b2,\b0,\b1
|
|
|
100 vperm \b3,\b0,\b1,v22
|
|
|
101 vaddfp \a0,\a2,\a3 // {r0,r1,i0,i1}
|
|
|
102 vsubfp \a1,\a2,\a3 // {r2,r3,i2,i3}
|
|
|
103 vaddfp \b0,\b2,\b3
|
|
|
104 vsubfp \b1,\b2,\b3
|
|
|
105 vperm \a2,\a0,\a1,v23 // vcprm(0,1,s0,s1) // {r0,r1,r2,r3}
|
|
|
106 vperm \a3,\a0,\a1,v24 // vcprm(2,3,s2,s3) // {i0,i1,i2,i3}
|
|
|
107 vperm \b2,\b0,\b1,v23
|
|
|
108 vperm \b3,\b0,\b1,v24
|
|
|
109 .endm
|
|
|
110
|
|
|
111 .macro FFT8 a0, a1, b0, b1, a2, a3, b2, b3, b4 // in,out:a0-b1
|
|
|
112 vmrghw \b2,\b0,\b1 // vcprm(0,s0,1,s1) // {r4,r6,i4,i6}
|
|
|
113 vmrglw \b3,\b0,\b1 // vcprm(2,s2,3,s3) // {r5,r7,i5,i7}
|
|
|
114 vperm \a2,\a0,\a1,v20 // FFT4 ...
|
|
|
115 vperm \a3,\a0,\a1,v21
|
|
|
116 vaddfp \b0,\b2,\b3 // {t1,t3,t2,t4}
|
|
|
117 vsubfp \b1,\b2,\b3 // {r5,r7,i5,i7}
|
|
|
118 vperm \b4,\b1,\b1,v25 // vcprm(2,3,0,1) // {i5,i7,r5,r7}
|
|
|
119 vaddfp \a0,\a2,\a3
|
|
|
120 vsubfp \a1,\a2,\a3
|
|
|
121 vmaddfp \b1,\b1,v17,v14 // * {-1,1,1,-1}/sqrt(2)
|
|
|
122 vmaddfp \b1,\b4,v18,\b1 // * { 1,1,1,1 }/sqrt(2) // {t8,ta,t7,t9}
|
|
|
123 vmrghw \a2,\a0,\a1
|
|
|
124 vperm \a3,\a0,\a1,v22
|
|
|
125 vperm \b2,\b0,\b1,v26 // vcprm(1,2,s3,s0) // {t3,t2,t9,t8}
|
|
|
126 vperm \b3,\b0,\b1,v27 // vcprm(0,3,s2,s1) // {t1,t4,t7,ta}
|
|
|
127 vaddfp \a0,\a2,\a3
|
|
|
128 vsubfp \a1,\a2,\a3
|
|
|
129 vaddfp \b0,\b2,\b3 // {t1,t2,t9,ta}
|
|
|
130 vsubfp \b1,\b2,\b3 // {t6,t5,tc,tb}
|
|
|
131 vperm \a2,\a0,\a1,v23
|
|
|
132 vperm \a3,\a0,\a1,v24
|
|
|
133 vperm \b2,\b0,\b1,v28 // vcprm(0,2,s1,s3) // {t1,t9,t5,tb}
|
|
|
134 vperm \b3,\b0,\b1,v29 // vcprm(1,3,s0,s2) // {t2,ta,t6,tc}
|
|
|
135 vsubfp \b0,\a2,\b2 // {r4,r5,r6,r7}
|
|
|
136 vsubfp \b1,\a3,\b3 // {i4,i5,i6,i7}
|
|
|
137 vaddfp \a0,\a2,\b2 // {r0,r1,r2,r3}
|
|
|
138 vaddfp \a1,\a3,\b3 // {i0,i1,i2,i3}
|
|
|
139 .endm
|
|
|
140
|
|
|
141 .macro BF d0,d1,s0,s1
|
|
|
142 vsubfp \d1,\s0,\s1
|
|
|
143 vaddfp \d0,\s0,\s1
|
|
|
144 .endm
|
|
|
145
|
|
12085
|
146 .macro zip d0,d1,s0,s1
|
|
|
147 vmrghw \d0,\s0,\s1
|
|
|
148 vmrglw \d1,\s0,\s1
|
|
|
149 .endm
|
|
|
150
|
|
|
151 .macro def_fft4 interleave
|
|
|
152 fft4\interleave\()_altivec:
|
|
12046
|
153 lvx v0, 0,r3
|
|
|
154 lvx v1,r9,r3
|
|
|
155 FFT4 v0,v1,v2,v3
|
|
12085
|
156 .ifnb \interleave
|
|
|
157 zip v0,v1,v2,v3
|
|
|
158 stvx v0, 0,r3
|
|
|
159 stvx v1,r9,r3
|
|
|
160 .else
|
|
12046
|
161 stvx v2, 0,r3
|
|
|
162 stvx v3,r9,r3
|
|
12085
|
163 .endif
|
|
12046
|
164 blr
|
|
12085
|
165 .endm
|
|
12046
|
166
|
|
12085
|
167 .macro def_fft8 interleave
|
|
|
168 fft8\interleave\()_altivec:
|
|
12046
|
169 addi r4,r3,32
|
|
|
170 lvx v0, 0,r3
|
|
|
171 lvx v1,r9,r3
|
|
|
172 lvx v2, 0,r4
|
|
|
173 lvx v3,r9,r4
|
|
|
174 FFT8 v0,v1,v2,v3,v4,v5,v6,v7,v8
|
|
12085
|
175 .ifnb \interleave
|
|
|
176 zip v4,v5,v0,v1
|
|
|
177 zip v6,v7,v2,v3
|
|
|
178 stvx v4, 0,r3
|
|
|
179 stvx v5,r9,r3
|
|
|
180 stvx v6, 0,r4
|
|
|
181 stvx v7,r9,r4
|
|
|
182 .else
|
|
12046
|
183 stvx v0, 0,r3
|
|
|
184 stvx v1,r9,r3
|
|
|
185 stvx v2, 0,r4
|
|
|
186 stvx v3,r9,r4
|
|
12085
|
187 .endif
|
|
12046
|
188 blr
|
|
12085
|
189 .endm
|
|
12046
|
190
|
|
12085
|
191 .macro def_fft16 interleave
|
|
|
192 fft16\interleave\()_altivec:
|
|
12046
|
193 addi r5,r3,64
|
|
|
194 addi r6,r3,96
|
|
|
195 addi r4,r3,32
|
|
|
196 lvx v0, 0,r5
|
|
|
197 lvx v1,r9,r5
|
|
|
198 lvx v2, 0,r6
|
|
|
199 lvx v3,r9,r6
|
|
|
200 FFT4x2 v0,v1,v2,v3,v4,v5,v6,v7
|
|
|
201 lvx v0, 0,r3
|
|
|
202 lvx v1,r9,r3
|
|
|
203 lvx v2, 0,r4
|
|
|
204 lvx v3,r9,r4
|
|
|
205 FFT8 v0,v1,v2,v3,v8,v9,v10,v11,v12
|
|
|
206 vmaddfp v8,v4,v15,v14 // r2*wre
|
|
|
207 vmaddfp v9,v5,v15,v14 // i2*wre
|
|
|
208 vmaddfp v10,v6,v15,v14 // r3*wre
|
|
|
209 vmaddfp v11,v7,v15,v14 // i3*wre
|
|
|
210 vmaddfp v8,v5,v16,v8 // i2*wim
|
|
|
211 vnmsubfp v9,v4,v16,v9 // r2*wim
|
|
|
212 vnmsubfp v10,v7,v16,v10 // i3*wim
|
|
|
213 vmaddfp v11,v6,v16,v11 // r3*wim
|
|
|
214 BF v10,v12,v10,v8
|
|
|
215 BF v11,v13,v9,v11
|
|
|
216 BF v0,v4,v0,v10
|
|
|
217 BF v3,v7,v3,v12
|
|
12085
|
218 BF v1,v5,v1,v11
|
|
|
219 BF v2,v6,v2,v13
|
|
|
220 .ifnb \interleave
|
|
|
221 zip v8, v9,v0,v1
|
|
|
222 zip v10,v11,v2,v3
|
|
|
223 zip v12,v13,v4,v5
|
|
|
224 zip v14,v15,v6,v7
|
|
|
225 stvx v8, 0,r3
|
|
|
226 stvx v9,r9,r3
|
|
|
227 stvx v10, 0,r4
|
|
|
228 stvx v11,r9,r4
|
|
|
229 stvx v12, 0,r5
|
|
|
230 stvx v13,r9,r5
|
|
|
231 stvx v14, 0,r6
|
|
|
232 stvx v15,r9,r6
|
|
|
233 .else
|
|
12046
|
234 stvx v0, 0,r3
|
|
|
235 stvx v4, 0,r5
|
|
|
236 stvx v3,r9,r4
|
|
|
237 stvx v7,r9,r6
|
|
|
238 stvx v1,r9,r3
|
|
|
239 stvx v5,r9,r5
|
|
|
240 stvx v2, 0,r4
|
|
|
241 stvx v6, 0,r6
|
|
12085
|
242 .endif
|
|
12046
|
243 blr
|
|
12085
|
244 .endm
|
|
12046
|
245
|
|
|
246 // void pass(float *z, float *wre, int n)
|
|
|
247 .macro PASS interleave, suffix
|
|
|
248 fft_pass\suffix\()_altivec:
|
|
|
249 mtctr r5
|
|
|
250 slwi r0,r5,4
|
|
|
251 slwi r7,r5,6 // o2
|
|
|
252 slwi r5,r5,5 // o1
|
|
|
253 add r10,r5,r7 // o3
|
|
|
254 add r0,r4,r0 // wim
|
|
|
255 addi r6,r5,16 // o1+16
|
|
|
256 addi r8,r7,16 // o2+16
|
|
|
257 addi r11,r10,16 // o3+16
|
|
|
258 1:
|
|
|
259 lvx v8, 0,r4 // wre
|
|
|
260 lvx v10, 0,r0 // wim
|
|
|
261 sub r0,r0,r9
|
|
|
262 lvx v9, 0,r0
|
|
|
263 vperm v9,v9,v10,v19 // vcprm(s0,3,2,1) => wim[0 .. -3]
|
|
|
264 lvx v4,r3,r7 // r2 = z[o2]
|
|
|
265 lvx v5,r3,r8 // i2 = z[o2+16]
|
|
|
266 lvx v6,r3,r10 // r3 = z[o3]
|
|
|
267 lvx v7,r3,r11 // i3 = z[o3+16]
|
|
|
268 vmaddfp v10,v4,v8,v14 // r2*wre
|
|
|
269 vmaddfp v11,v5,v8,v14 // i2*wre
|
|
|
270 vmaddfp v12,v6,v8,v14 // r3*wre
|
|
|
271 vmaddfp v13,v7,v8,v14 // i3*wre
|
|
|
272 lvx v0, 0,r3 // r0 = z[0]
|
|
|
273 lvx v3,r3,r6 // i1 = z[o1+16]
|
|
|
274 vmaddfp v10,v5,v9,v10 // i2*wim
|
|
|
275 vnmsubfp v11,v4,v9,v11 // r2*wim
|
|
|
276 vnmsubfp v12,v7,v9,v12 // i3*wim
|
|
|
277 vmaddfp v13,v6,v9,v13 // r3*wim
|
|
|
278 lvx v1,r3,r9 // i0 = z[16]
|
|
|
279 lvx v2,r3,r5 // r1 = z[o1]
|
|
|
280 BF v12,v8,v12,v10
|
|
|
281 BF v13,v9,v11,v13
|
|
|
282 BF v0,v4,v0,v12
|
|
|
283 BF v3,v7,v3,v8
|
|
|
284 .if !\interleave
|
|
|
285 stvx v0, 0,r3
|
|
|
286 stvx v4,r3,r7
|
|
|
287 stvx v3,r3,r6
|
|
|
288 stvx v7,r3,r11
|
|
|
289 .endif
|
|
|
290 BF v1,v5,v1,v13
|
|
|
291 BF v2,v6,v2,v9
|
|
|
292 .if !\interleave
|
|
|
293 stvx v1,r3,r9
|
|
|
294 stvx v2,r3,r5
|
|
|
295 stvx v5,r3,r8
|
|
|
296 stvx v6,r3,r10
|
|
|
297 .else
|
|
|
298 vmrghw v8,v0,v1
|
|
|
299 vmrglw v9,v0,v1
|
|
|
300 stvx v8, 0,r3
|
|
|
301 stvx v9,r3,r9
|
|
|
302 vmrghw v8,v2,v3
|
|
|
303 vmrglw v9,v2,v3
|
|
|
304 stvx v8,r3,r5
|
|
|
305 stvx v9,r3,r6
|
|
|
306 vmrghw v8,v4,v5
|
|
|
307 vmrglw v9,v4,v5
|
|
|
308 stvx v8,r3,r7
|
|
|
309 stvx v9,r3,r8
|
|
|
310 vmrghw v8,v6,v7
|
|
|
311 vmrglw v9,v6,v7
|
|
|
312 stvx v8,r3,r10
|
|
|
313 stvx v9,r3,r11
|
|
|
314 .endif
|
|
|
315 addi r3,r3,32
|
|
|
316 addi r4,r4,16
|
|
|
317 bdnz 1b
|
|
|
318 sub r3,r3,r5
|
|
|
319 blr
|
|
|
320 .endm
|
|
|
321
|
|
|
322 .macro DECL_FFT suffix, bits, n, n2, n4
|
|
|
323 fft\n\suffix\()_altivec:
|
|
|
324 mflr r0
|
|
|
325 STORE_PTR r0,r1,\bits-5
|
|
|
326 bl fft\n2\()_altivec
|
|
|
327 addi2 r3,\n*4
|
|
|
328 bl fft\n4\()_altivec
|
|
|
329 addi2 r3,\n*2
|
|
|
330 bl fft\n4\()_altivec
|
|
|
331 addi2 r3,\n*-6
|
|
|
332 LOAD_PTR r0,r1,\bits-5
|
|
|
333 LOAD_PTR r4,r12,\bits
|
|
|
334 mtlr r0
|
|
|
335 li r5,\n/16
|
|
|
336 b fft_pass\suffix\()_altivec
|
|
|
337 .endm
|
|
|
338
|
|
|
339 .macro DECL_FFTS interleave, suffix
|
|
|
340 .text
|
|
12085
|
341 def_fft4 \suffix
|
|
|
342 def_fft8 \suffix
|
|
|
343 def_fft16 \suffix
|
|
12046
|
344 PASS \interleave, \suffix
|
|
|
345 DECL_FFT \suffix, 5, 32, 16, 8
|
|
|
346 DECL_FFT \suffix, 6, 64, 32, 16
|
|
|
347 DECL_FFT \suffix, 7, 128, 64, 32
|
|
|
348 DECL_FFT \suffix, 8, 256, 128, 64
|
|
|
349 DECL_FFT \suffix, 9, 512, 256, 128
|
|
|
350 DECL_FFT \suffix,10, 1024, 512, 256
|
|
|
351 DECL_FFT \suffix,11, 2048, 1024, 512
|
|
|
352 DECL_FFT \suffix,12, 4096, 2048, 1024
|
|
|
353 DECL_FFT \suffix,13, 8192, 4096, 2048
|
|
|
354 DECL_FFT \suffix,14,16384, 8192, 4096
|
|
|
355 DECL_FFT \suffix,15,32768,16384, 8192
|
|
|
356 DECL_FFT \suffix,16,65536,32768,16384
|
|
|
357
|
|
|
358 .rodata
|
|
|
359 .global EXTERN_ASM\()ff_fft_dispatch\suffix\()_altivec
|
|
|
360 EXTERN_ASM\()ff_fft_dispatch\suffix\()_altivec:
|
|
12085
|
361 PTR fft4\suffix\()_altivec
|
|
|
362 PTR fft8\suffix\()_altivec
|
|
|
363 PTR fft16\suffix\()_altivec
|
|
12046
|
364 PTR fft32\suffix\()_altivec
|
|
|
365 PTR fft64\suffix\()_altivec
|
|
|
366 PTR fft128\suffix\()_altivec
|
|
|
367 PTR fft256\suffix\()_altivec
|
|
|
368 PTR fft512\suffix\()_altivec
|
|
|
369 PTR fft1024\suffix\()_altivec
|
|
|
370 PTR fft2048\suffix\()_altivec
|
|
|
371 PTR fft4096\suffix\()_altivec
|
|
|
372 PTR fft8192\suffix\()_altivec
|
|
|
373 PTR fft16384\suffix\()_altivec
|
|
|
374 PTR fft32768\suffix\()_altivec
|
|
|
375 PTR fft65536\suffix\()_altivec
|
|
|
376 .endm
|
|
|
377
|
|
|
378 DECL_FFTS 0
|
|
|
379 DECL_FFTS 1, _interleave
|
