xref: /third_party/ffmpeg/libavcodec/ppc/fft_vsx.h (revision cabdff1a)
1#ifndef AVCODEC_PPC_FFT_VSX_H
2#define AVCODEC_PPC_FFT_VSX_H
3/*
4 * FFT  transform, optimized with VSX built-in functions
5 * Copyright (c) 2014 Rong Yan  Copyright (c) 2009 Loren Merritt
6 *
7 * This algorithm (though not any of the implementation details) is
8 * based on libdjbfft by D. J. Bernstein, and fft_altivec_s.S.
9 *
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
28#include "config.h"
29#include "libavutil/cpu.h"
30#include "libavutil/ppc/util_altivec.h"
31#include "libavcodec/fft.h"
32#include "libavcodec/fft-internal.h"
33
34#if HAVE_VSX
35
36void ff_fft_calc_interleave_vsx(FFTContext *s, FFTComplex *z);
37void ff_fft_calc_vsx(FFTContext *s, FFTComplex *z);
38
39
40#define byte_2complex (2*sizeof(FFTComplex))
41#define byte_4complex (4*sizeof(FFTComplex))
42#define byte_6complex (6*sizeof(FFTComplex))
43#define byte_8complex (8*sizeof(FFTComplex))
44#define byte_10complex (10*sizeof(FFTComplex))
45#define byte_12complex (12*sizeof(FFTComplex))
46#define byte_14complex (14*sizeof(FFTComplex))
47
48inline static void pass_vsx_interleave(FFTComplex *z, const FFTSample *wre, unsigned int n)
49{
50    int o1 = n<<1;
51    int o2 = n<<2;
52    int o3 = o1+o2;
53    int i1, i2, i3;
54    FFTSample* out = (FFTSample*)z;
55    const FFTSample *wim = wre+o1;
56    vec_f vz0, vzo1, vzo2, vzo3;
57    vec_f x0, x1, x2, x3;
58    vec_f x4, x5, x6, x7;
59    vec_f x8, x9, x10, x11;
60    vec_f x12, x13, x14, x15;
61    vec_f x16, x17, x18, x19;
62    vec_f x20, x21, x22, x23;
63    vec_f vz0plus1, vzo1plus1, vzo2plus1, vzo3plus1;
64    vec_f y0, y1, y2, y3;
65    vec_f y4, y5, y8, y9;
66    vec_f y10, y13, y14, y15;
67    vec_f y16, y17, y18, y19;
68    vec_f y20, y21, y22, y23;
69    vec_f wr1, wi1, wr0, wi0;
70    vec_f wr2, wi2, wr3, wi3;
71    vec_f xmulwi0, xmulwi1, ymulwi2, ymulwi3;
72
73    n = n-2;
74    i1 = o1*sizeof(FFTComplex);
75    i2 = o2*sizeof(FFTComplex);
76    i3 = o3*sizeof(FFTComplex);
77    vzo2 = vec_ld(i2, &(out[0]));  // zo2.r  zo2.i  z(o2+1).r  z(o2+1).i
78    vzo2plus1 = vec_ld(i2+16, &(out[0]));
79    vzo3 = vec_ld(i3, &(out[0]));  // zo3.r  zo3.i  z(o3+1).r  z(o3+1).i
80    vzo3plus1 = vec_ld(i3+16, &(out[0]));
81    vz0 = vec_ld(0, &(out[0]));    // z0.r  z0.i  z1.r  z1.i
82    vz0plus1 = vec_ld(16, &(out[0]));
83    vzo1 = vec_ld(i1, &(out[0]));  // zo1.r  zo1.i  z(o1+1).r  z(o1+1).i
84    vzo1plus1 = vec_ld(i1+16, &(out[0]));
85
86    x0 = vec_add(vzo2, vzo3);
87    x1 = vec_sub(vzo2, vzo3);
88    y0 = vec_add(vzo2plus1, vzo3plus1);
89    y1 = vec_sub(vzo2plus1, vzo3plus1);
90
91    wr1 = vec_splats(wre[1]);
92    wi1 = vec_splats(wim[-1]);
93    wi2 = vec_splats(wim[-2]);
94    wi3 = vec_splats(wim[-3]);
95    wr2 = vec_splats(wre[2]);
96    wr3 = vec_splats(wre[3]);
97
98    x2 = vec_perm(x0, x1, vcprm(2,s2,3,s3));
99    x3 = vec_perm(x0, x1, vcprm(s3,3,s2,2));
100
101    y4 = vec_perm(y0, y1, vcprm(s1,1,s0,0));
102    y5 = vec_perm(y0, y1, vcprm(s3,3,s2,2));
103    y2 = vec_perm(y0, y1, vcprm(0,s0,1,s1));
104    y3 = vec_perm(y0, y1, vcprm(2,s2,3,s3));
105
106    ymulwi2 = vec_mul(y4, wi2);
107    ymulwi3 = vec_mul(y5, wi3);
108    x4 = vec_mul(x2, wr1);
109    x5 = vec_mul(x3, wi1);
110    y8 = vec_madd(y2, wr2, ymulwi2);
111    y9 = vec_msub(y2, wr2, ymulwi2);
112    x6 = vec_add(x4, x5);
113    x7 = vec_sub(x4, x5);
114    y13 = vec_madd(y3, wr3, ymulwi3);
115    y14 = vec_msub(y3, wr3, ymulwi3);
116
117    x8 = vec_perm(x6, x7, vcprm(0,1,s2,s3));
118    y10 = vec_perm(y8, y9, vcprm(0,1,s2,s3));
119    y15 = vec_perm(y13, y14, vcprm(0,1,s2,s3));
120
121    x9 = vec_perm(x0, x8, vcprm(0,1,s0,s2));
122    x10 = vec_perm(x1, x8, vcprm(1,0,s3,s1));
123
124    y16 = vec_perm(y10, y15, vcprm(0,2,s0,s2));
125    y17 = vec_perm(y10, y15, vcprm(3,1,s3,s1));
126
127    x11 = vec_add(vz0, x9);
128    x12 = vec_sub(vz0, x9);
129    x13 = vec_add(vzo1, x10);
130    x14 = vec_sub(vzo1, x10);
131
132    y18 = vec_add(vz0plus1, y16);
133    y19 = vec_sub(vz0plus1, y16);
134    y20 = vec_add(vzo1plus1, y17);
135    y21 = vec_sub(vzo1plus1, y17);
136
137    x15 = vec_perm(x13, x14, vcprm(0,s1,2,s3));
138    x16 = vec_perm(x13, x14, vcprm(s0,1,s2,3));
139    y22 = vec_perm(y20, y21, vcprm(0,s1,2,s3));
140    y23 = vec_perm(y20, y21, vcprm(s0,1,s2,3));
141
142
143    vec_st(x11, 0, &(out[0]));
144    vec_st(y18, 16, &(out[0]));
145    vec_st(x15, i1, &(out[0]));
146    vec_st(y22, i1+16, &(out[0]));
147    vec_st(x12, i2, &(out[0]));
148    vec_st(y19, i2+16, &(out[0]));
149    vec_st(x16, i3, &(out[0]));
150    vec_st(y23, i3+16, &(out[0]));
151
152    do {
153        out += 8;
154        wre += 4;
155        wim -= 4;
156        wr0 = vec_splats(wre[0]);
157        wr1 = vec_splats(wre[1]);
158        wi0 = vec_splats(wim[0]);
159        wi1 = vec_splats(wim[-1]);
160
161        wr2 = vec_splats(wre[2]);
162        wr3 = vec_splats(wre[3]);
163        wi2 = vec_splats(wim[-2]);
164        wi3 = vec_splats(wim[-3]);
165
166        vzo2 = vec_ld(i2, &(out[0]));  // zo2.r  zo2.i  z(o2+1).r  z(o2+1).i
167        vzo2plus1 = vec_ld(i2+16, &(out[0]));
168        vzo3 = vec_ld(i3, &(out[0]));  // zo3.r  zo3.i  z(o3+1).r  z(o3+1).i
169        vzo3plus1 = vec_ld(i3+16, &(out[0]));
170        vz0 = vec_ld(0, &(out[0]));    // z0.r  z0.i  z1.r  z1.i
171        vz0plus1 = vec_ld(16, &(out[0]));
172        vzo1 = vec_ld(i1, &(out[0])); // zo1.r  zo1.i  z(o1+1).r  z(o1+1).i
173        vzo1plus1 = vec_ld(i1+16, &(out[0]));
174
175        x0 = vec_add(vzo2, vzo3);
176        x1 = vec_sub(vzo2, vzo3);
177
178        y0 = vec_add(vzo2plus1, vzo3plus1);
179        y1 = vec_sub(vzo2plus1, vzo3plus1);
180
181        x4 = vec_perm(x0, x1, vcprm(s1,1,s0,0));
182        x5 = vec_perm(x0, x1, vcprm(s3,3,s2,2));
183        x2 = vec_perm(x0, x1, vcprm(0,s0,1,s1));
184        x3 = vec_perm(x0, x1, vcprm(2,s2,3,s3));
185
186        y2 = vec_perm(y0, y1, vcprm(0,s0,1,s1));
187        y3 = vec_perm(y0, y1, vcprm(2,s2,3,s3));
188        xmulwi0 = vec_mul(x4, wi0);
189        xmulwi1 = vec_mul(x5, wi1);
190
191        y4 = vec_perm(y0, y1, vcprm(s1,1,s0,0));
192        y5 = vec_perm(y0, y1, vcprm(s3,3,s2,2));
193
194        x8 = vec_madd(x2, wr0, xmulwi0);
195        x9 = vec_msub(x2, wr0, xmulwi0);
196        ymulwi2 = vec_mul(y4, wi2);
197        ymulwi3 = vec_mul(y5, wi3);
198
199        x13 = vec_madd(x3, wr1, xmulwi1);
200        x14 = vec_msub(x3, wr1, xmulwi1);
201
202        y8 = vec_madd(y2, wr2, ymulwi2);
203        y9 = vec_msub(y2, wr2, ymulwi2);
204        y13 = vec_madd(y3, wr3, ymulwi3);
205        y14 = vec_msub(y3, wr3, ymulwi3);
206
207        x10 = vec_perm(x8, x9, vcprm(0,1,s2,s3));
208        x15 = vec_perm(x13, x14, vcprm(0,1,s2,s3));
209
210        y10 = vec_perm(y8, y9, vcprm(0,1,s2,s3));
211        y15 = vec_perm(y13, y14, vcprm(0,1,s2,s3));
212
213        x16 = vec_perm(x10, x15, vcprm(0,2,s0,s2));
214        x17 = vec_perm(x10, x15, vcprm(3,1,s3,s1));
215
216        y16 = vec_perm(y10, y15, vcprm(0,2,s0,s2));
217        y17 = vec_perm(y10, y15, vcprm(3,1,s3,s1));
218
219        x18 = vec_add(vz0, x16);
220        x19 = vec_sub(vz0, x16);
221        x20 = vec_add(vzo1, x17);
222        x21 = vec_sub(vzo1, x17);
223
224        y18 = vec_add(vz0plus1, y16);
225        y19 = vec_sub(vz0plus1, y16);
226        y20 = vec_add(vzo1plus1, y17);
227        y21 = vec_sub(vzo1plus1, y17);
228
229        x22 = vec_perm(x20, x21, vcprm(0,s1,2,s3));
230        x23 = vec_perm(x20, x21, vcprm(s0,1,s2,3));
231
232        y22 = vec_perm(y20, y21, vcprm(0,s1,2,s3));
233        y23 = vec_perm(y20, y21, vcprm(s0,1,s2,3));
234
235        vec_st(x18, 0, &(out[0]));
236        vec_st(y18, 16, &(out[0]));
237        vec_st(x22, i1, &(out[0]));
238        vec_st(y22, i1+16, &(out[0]));
239        vec_st(x19, i2, &(out[0]));
240        vec_st(y19, i2+16, &(out[0]));
241        vec_st(x23, i3, &(out[0]));
242        vec_st(y23, i3+16, &(out[0]));
243    } while (n-=2);
244}
245
246inline static void fft2_vsx_interleave(FFTComplex *z)
247{
248    FFTSample r1, i1;
249
250    r1 = z[0].re - z[1].re;
251    z[0].re += z[1].re;
252    z[1].re = r1;
253
254    i1 = z[0].im - z[1].im;
255    z[0].im += z[1].im;
256    z[1].im = i1;
257 }
258
259inline static void fft4_vsx_interleave(FFTComplex *z)
260{
261    vec_f a, b, c, d;
262    float* out=  (float*)z;
263    a = vec_ld(0, &(out[0]));
264    b = vec_ld(byte_2complex, &(out[0]));
265
266    c = vec_perm(a, b, vcprm(0,1,s2,s1));
267    d = vec_perm(a, b, vcprm(2,3,s0,s3));
268    a = vec_add(c, d);
269    b = vec_sub(c, d);
270
271    c = vec_perm(a, b, vcprm(0,1,s0,s1));
272    d = vec_perm(a, b, vcprm(2,3,s3,s2));
273
274    a = vec_add(c, d);
275    b = vec_sub(c, d);
276    vec_st(a, 0, &(out[0]));
277    vec_st(b, byte_2complex, &(out[0]));
278}
279
280inline static void fft8_vsx_interleave(FFTComplex *z)
281{
282    vec_f vz0, vz1, vz2, vz3;
283    vec_f x0, x1, x2, x3;
284    vec_f x4, x5, x6, x7;
285    vec_f x8, x9, x10, x11;
286    vec_f x12, x13, x14, x15;
287    vec_f x16, x17, x18, x19;
288    vec_f x20, x21, x22, x23;
289    vec_f x24, x25, x26, x27;
290    vec_f x28, x29, x30, x31;
291    vec_f x32, x33, x34;
292
293    float* out=  (float*)z;
294    vec_f vc1 = {sqrthalf, sqrthalf, sqrthalf, sqrthalf};
295
296    vz0 = vec_ld(0, &(out[0]));
297    vz1 = vec_ld(byte_2complex, &(out[0]));
298    vz2 = vec_ld(byte_4complex, &(out[0]));
299    vz3 = vec_ld(byte_6complex, &(out[0]));
300
301    x0 = vec_perm(vz0, vz1, vcprm(0,1,s2,s1));
302    x1 = vec_perm(vz0, vz1, vcprm(2,3,s0,s3));
303    x2 = vec_perm(vz2, vz3, vcprm(2,1,s0,s1));
304    x3 = vec_perm(vz2, vz3, vcprm(0,3,s2,s3));
305
306    x4 = vec_add(x0, x1);
307    x5 = vec_sub(x0, x1);
308    x6 = vec_add(x2, x3);
309    x7 = vec_sub(x2, x3);
310
311    x8 = vec_perm(x4, x5, vcprm(0,1,s0,s1));
312    x9 = vec_perm(x4, x5, vcprm(2,3,s3,s2));
313    x10 = vec_perm(x6, x7, vcprm(2,1,s2,s1));
314    x11 = vec_perm(x6, x7, vcprm(0,3,s0,s3));
315
316    x12 = vec_add(x8, x9);
317    x13 = vec_sub(x8, x9);
318    x14 = vec_add(x10, x11);
319    x15 = vec_sub(x10, x11);
320    x16 = vec_perm(x12, x13, vcprm(0,s0,1,s1));
321    x17 = vec_perm(x14, x15, vcprm(0,s0,1,s1));
322    x18 = vec_perm(x16, x17, vcprm(s0,s3,s2,s1));
323    x19 = vec_add(x16, x18); // z0.r  z2.r  z0.i  z2.i
324    x20 = vec_sub(x16, x18); // z4.r  z6.r  z4.i  z6.i
325
326    x21 = vec_perm(x12, x13, vcprm(2,s2,3,s3));
327    x22 = vec_perm(x14, x15, vcprm(2,3,s2,s3));
328    x23 = vec_perm(x14, x15, vcprm(3,2,s3,s2));
329    x24 = vec_add(x22, x23);
330    x25 = vec_sub(x22, x23);
331    x26 = vec_mul( vec_perm(x24, x25, vcprm(2,s2,0,s0)), vc1);
332
333    x27 = vec_add(x21, x26); // z1.r  z7.r z1.i z3.i
334    x28 = vec_sub(x21, x26); //z5.r  z3.r z5.i z7.i
335
336    x29 = vec_perm(x19, x27, vcprm(0,2,s0,s2)); // z0.r  z0.i  z1.r  z1.i
337    x30 = vec_perm(x19, x27, vcprm(1,3,s1,s3)); // z2.r  z2.i  z7.r  z3.i
338    x31 = vec_perm(x20, x28, vcprm(0,2,s0,s2)); // z4.r  z4.i  z5.r  z5.i
339    x32 = vec_perm(x20, x28, vcprm(1,3,s1,s3)); // z6.r  z6.i  z3.r  z7.i
340    x33 = vec_perm(x30, x32, vcprm(0,1,s2,3));  // z2.r  z2.i  z3.r  z3.i
341    x34 = vec_perm(x30, x32, vcprm(s0,s1,2,s3)); // z6.r  z6.i  z7.r  z7.i
342
343    vec_st(x29, 0, &(out[0]));
344    vec_st(x33, byte_2complex, &(out[0]));
345    vec_st(x31, byte_4complex, &(out[0]));
346    vec_st(x34, byte_6complex, &(out[0]));
347}
348
349inline static void fft16_vsx_interleave(FFTComplex *z)
350{
351    float* out=  (float*)z;
352    vec_f vc0 = {sqrthalf, sqrthalf, sqrthalf, sqrthalf};
353    vec_f vc1 = {ff_cos_16[1], ff_cos_16[1], ff_cos_16[1], ff_cos_16[1]};
354    vec_f vc2 = {ff_cos_16[3], ff_cos_16[3], ff_cos_16[3], ff_cos_16[3]};
355    vec_f vz0, vz1, vz2, vz3;
356    vec_f vz4, vz5, vz6, vz7;
357    vec_f x0, x1, x2, x3;
358    vec_f x4, x5, x6, x7;
359    vec_f x8, x9, x10, x11;
360    vec_f x12, x13, x14, x15;
361    vec_f x16, x17, x18, x19;
362    vec_f x20, x21, x22, x23;
363    vec_f x24, x25, x26, x27;
364    vec_f x28, x29, x30, x31;
365    vec_f x32, x33, x34, x35;
366    vec_f x36, x37, x38, x39;
367    vec_f x40, x41, x42, x43;
368    vec_f x44, x45, x46, x47;
369    vec_f x48, x49, x50, x51;
370    vec_f x52, x53, x54, x55;
371    vec_f x56, x57, x58, x59;
372    vec_f x60, x61, x62, x63;
373    vec_f x64, x65, x66, x67;
374    vec_f x68, x69, x70, x71;
375    vec_f x72, x73, x74, x75;
376    vec_f x76, x77, x78, x79;
377    vec_f x80, x81, x82, x83;
378    vec_f x84, x85, x86;
379
380    vz0 = vec_ld(0, &(out[0]));
381    vz1 = vec_ld(byte_2complex, &(out[0]));
382    vz2 = vec_ld(byte_4complex, &(out[0]));
383    vz3 = vec_ld(byte_6complex, &(out[0]));
384    vz4 = vec_ld(byte_8complex, &(out[0]));
385    vz5 = vec_ld(byte_10complex, &(out[0]));
386    vz6 = vec_ld(byte_12complex, &(out[0]));
387    vz7 = vec_ld(byte_14complex, &(out[0]));
388
389    x0 = vec_perm(vz0, vz1, vcprm(0,1,s2,s1));
390    x1 = vec_perm(vz0, vz1, vcprm(2,3,s0,s3));
391    x2 = vec_perm(vz2, vz3, vcprm(0,1,s0,s1));
392    x3 = vec_perm(vz2, vz3, vcprm(2,3,s2,s3));
393
394    x4 = vec_perm(vz4, vz5, vcprm(0,1,s2,s1));
395    x5 = vec_perm(vz4, vz5, vcprm(2,3,s0,s3));
396    x6 = vec_perm(vz6, vz7, vcprm(0,1,s2,s1));
397    x7 = vec_perm(vz6, vz7, vcprm(2,3,s0,s3));
398
399    x8 = vec_add(x0, x1);
400    x9 = vec_sub(x0, x1);
401    x10 = vec_add(x2, x3);
402    x11 = vec_sub(x2, x3);
403
404    x12 = vec_add(x4, x5);
405    x13 = vec_sub(x4, x5);
406    x14 = vec_add(x6, x7);
407    x15 = vec_sub(x6, x7);
408
409    x16 = vec_perm(x8, x9, vcprm(0,1,s0,s1));
410    x17 = vec_perm(x8, x9, vcprm(2,3,s3,s2));
411    x18 = vec_perm(x10, x11, vcprm(2,1,s1,s2));
412    x19 = vec_perm(x10, x11, vcprm(0,3,s0,s3));
413    x20 = vec_perm(x12, x14, vcprm(0,1,s0, s1));
414    x21 = vec_perm(x12, x14, vcprm(2,3,s2,s3));
415    x22 = vec_perm(x13, x15, vcprm(0,1,s0,s1));
416    x23 = vec_perm(x13, x15, vcprm(3,2,s3,s2));
417
418    x24 = vec_add(x16, x17);
419    x25 = vec_sub(x16, x17);
420    x26 = vec_add(x18, x19);
421    x27 = vec_sub(x18, x19);
422    x28 = vec_add(x20, x21);
423    x29 = vec_sub(x20, x21);
424    x30 = vec_add(x22, x23);
425    x31 = vec_sub(x22, x23);
426
427    x32 = vec_add(x24, x26);
428    x33 = vec_sub(x24, x26);
429    x34 = vec_perm(x32, x33, vcprm(0,1,s0,s1));
430
431    x35 = vec_perm(x28, x29, vcprm(2,1,s1,s2));
432    x36 = vec_perm(x28, x29, vcprm(0,3,s0,s3));
433    x37 = vec_add(x35, x36);
434    x38 = vec_sub(x35, x36);
435    x39 = vec_perm(x37, x38, vcprm(0,1,s1,s0));
436
437    x40 = vec_perm(x27, x38, vcprm(3,2,s2,s3));
438    x41 = vec_perm(x26,  x37, vcprm(2,3,s3,s2));
439    x42 = vec_add(x40, x41);
440    x43 = vec_sub(x40, x41);
441    x44 = vec_mul(x42, vc0);
442    x45 = vec_mul(x43, vc0);
443
444    x46 = vec_add(x34, x39);  // z0.r  z0.i  z4.r  z4.i
445    x47 = vec_sub(x34, x39);  // z8.r  z8.i  z12.r  z12.i
446
447    x48 = vec_perm(x30, x31, vcprm(2,1,s1,s2));
448    x49 = vec_perm(x30, x31, vcprm(0,3,s3,s0));
449    x50 = vec_add(x48, x49);
450    x51 = vec_sub(x48, x49);
451    x52 = vec_mul(x50, vc1);
452    x53 = vec_mul(x50, vc2);
453    x54 = vec_mul(x51, vc1);
454    x55 = vec_mul(x51, vc2);
455
456    x56 = vec_perm(x24, x25, vcprm(2,3,s2,s3));
457    x57 = vec_perm(x44, x45, vcprm(0,1,s1,s0));
458    x58 = vec_add(x56, x57);
459    x59 = vec_sub(x56, x57);
460
461    x60 = vec_perm(x54, x55, vcprm(1,0,3,2));
462    x61 = vec_perm(x54, x55, vcprm(s1,s0,s3,s2));
463    x62 = vec_add(x52, x61);
464    x63 = vec_sub(x52, x61);
465    x64 = vec_add(x60, x53);
466    x65 = vec_sub(x60, x53);
467    x66 = vec_perm(x62, x64, vcprm(0,1,s3,s2));
468    x67 = vec_perm(x63, x65, vcprm(s0,s1,3,2));
469
470    x68 = vec_add(x58, x66); // z1.r    z1.i  z3.r    z3.i
471    x69 = vec_sub(x58, x66); // z9.r    z9.i  z11.r  z11.i
472    x70 = vec_add(x59, x67); // z5.r    z5.i  z15.r  z15.i
473    x71 = vec_sub(x59, x67); // z13.r  z13.i z7.r   z7.i
474
475    x72 = vec_perm(x25, x27, vcprm(s1,s0,s2,s3));
476    x73 = vec_add(x25, x72);
477    x74 = vec_sub(x25, x72);
478    x75 = vec_perm(x73, x74, vcprm(0,1,s0,s1));
479    x76 = vec_perm(x44, x45, vcprm(3,2,s2,s3));
480    x77 = vec_add(x75, x76); // z2.r   z2.i    z6.r    z6.i
481    x78 = vec_sub(x75, x76); // z10.r  z10.i  z14.r  z14.i
482
483    x79 = vec_perm(x46, x68, vcprm(0,1,s0,s1)); // z0.r  z0.i  z1.r  z1.i
484    x80 = vec_perm(x77, x68, vcprm(0,1,s2,s3)); // z2.r  z2.i  z3.r  z3.i
485    x81 = vec_perm(x46, x70, vcprm(2,3,s0,s1)); // z4.r  z4.i  z5.r  z5.i
486    x82 = vec_perm(x71, x77, vcprm(s2,s3,2,3)); // z6.r  z6.i  z7.r  z7.i
487    vec_st(x79, 0, &(out[0]));
488    vec_st(x80, byte_2complex, &(out[0]));
489    vec_st(x81, byte_4complex, &(out[0]));
490    vec_st(x82, byte_6complex, &(out[0]));
491    x83 = vec_perm(x47, x69, vcprm(0,1,s0,s1)); // z8.r  z8.i  z9.r  z9.i
492    x84 = vec_perm(x78, x69, vcprm(0,1,s2,s3)); // z10.r  z10.i  z11.r  z11.i
493    x85 = vec_perm(x47, x71, vcprm(2,3,s0,s1)); // z12.r  z12.i  z13.r  z13.i
494    x86 = vec_perm(x70, x78, vcprm(s2,s3,2,3)); // z14.r  z14.i  z15.r  z15.i
495    vec_st(x83, byte_8complex, &(out[0]));
496    vec_st(x84, byte_10complex, &(out[0]));
497    vec_st(x85, byte_12complex, &(out[0]));
498    vec_st(x86, byte_14complex, &(out[0]));
499}
500
501inline static void fft4_vsx(FFTComplex *z)
502{
503    vec_f a, b, c, d;
504    float* out=  (float*)z;
505    a = vec_ld(0, &(out[0]));
506    b = vec_ld(byte_2complex, &(out[0]));
507
508    c = vec_perm(a, b, vcprm(0,1,s2,s1));
509    d = vec_perm(a, b, vcprm(2,3,s0,s3));
510    a = vec_add(c, d);
511    b = vec_sub(c, d);
512
513    c = vec_perm(a,b, vcprm(0,s0,1,s1));
514    d = vec_perm(a, b, vcprm(2,s3,3,s2));
515
516    a = vec_add(c, d);
517    b = vec_sub(c, d);
518
519    c = vec_perm(a, b, vcprm(0,1,s0,s1));
520    d = vec_perm(a, b, vcprm(2,3,s2,s3));
521
522    vec_st(c, 0, &(out[0]));
523    vec_st(d, byte_2complex, &(out[0]));
524    return;
525}
526
527inline static void fft8_vsx(FFTComplex *z)
528{
529    vec_f vz0, vz1, vz2, vz3;
530    vec_f vz4, vz5, vz6, vz7, vz8;
531
532    float* out=  (float*)z;
533    vec_f vc0 = {0.0, 0.0, 0.0, 0.0};
534    vec_f vc1 = {-sqrthalf, sqrthalf, sqrthalf, -sqrthalf};
535    vec_f vc2 = {sqrthalf, sqrthalf, sqrthalf, sqrthalf};
536
537    vz0 = vec_ld(0, &(out[0]));
538    vz1 = vec_ld(byte_2complex, &(out[0]));
539    vz2 = vec_ld(byte_4complex, &(out[0]));
540    vz3 = vec_ld(byte_6complex, &(out[0]));
541
542    vz6 = vec_perm(vz2, vz3, vcprm(0,s0,1,s1));
543    vz7 = vec_perm(vz2, vz3, vcprm(2,s2,3,s3));
544    vz4 = vec_perm(vz0, vz1, vcprm(0,1,s2,s1));
545    vz5 = vec_perm(vz0, vz1, vcprm(2,3,s0,s3));
546
547    vz2 = vec_add(vz6, vz7);
548    vz3 = vec_sub(vz6, vz7);
549    vz8 = vec_perm(vz3, vz3, vcprm(2,3,0,1));
550
551    vz0 = vec_add(vz4, vz5);
552    vz1 = vec_sub(vz4, vz5);
553
554    vz3 = vec_madd(vz3, vc1, vc0);
555    vz3 = vec_madd(vz8, vc2, vz3);
556
557    vz4 = vec_perm(vz0, vz1, vcprm(0,s0,1,s1));
558    vz5 = vec_perm(vz0, vz1, vcprm(2,s3,3,s2));
559    vz6 = vec_perm(vz2, vz3, vcprm(1,2,s3,s0));
560    vz7 = vec_perm(vz2, vz3, vcprm(0,3,s2,s1));
561
562    vz0 = vec_add(vz4, vz5);
563    vz1 = vec_sub(vz4, vz5);
564    vz2 = vec_add(vz6, vz7);
565    vz3 = vec_sub(vz6, vz7);
566
567    vz4 = vec_perm(vz0, vz1, vcprm(0,1,s0,s1));
568    vz5 = vec_perm(vz0, vz1, vcprm(2,3,s2,s3));
569    vz6 = vec_perm(vz2, vz3, vcprm(0,2,s1,s3));
570    vz7 = vec_perm(vz2, vz3, vcprm(1,3,s0,s2));
571
572
573    vz2 = vec_sub(vz4, vz6);
574    vz3 = vec_sub(vz5, vz7);
575
576    vz0 = vec_add(vz4, vz6);
577    vz1 = vec_add(vz5, vz7);
578
579    vec_st(vz0, 0, &(out[0]));
580    vec_st(vz1, byte_2complex, &(out[0]));
581    vec_st(vz2, byte_4complex, &(out[0]));
582    vec_st(vz3, byte_6complex, &(out[0]));
583    return;
584}
585
586inline static void fft16_vsx(FFTComplex *z)
587{
588    float* out=  (float*)z;
589    vec_f vc0 = {0.0, 0.0, 0.0, 0.0};
590    vec_f vc1 = {-sqrthalf, sqrthalf, sqrthalf, -sqrthalf};
591    vec_f vc2 = {sqrthalf, sqrthalf, sqrthalf, sqrthalf};
592    vec_f vc3 = {1.0, 0.92387953, sqrthalf, 0.38268343};
593    vec_f vc4 = {0.0, 0.38268343, sqrthalf, 0.92387953};
594    vec_f vc5 = {-0.0, -0.38268343, -sqrthalf, -0.92387953};
595
596    vec_f vz0, vz1, vz2, vz3;
597    vec_f vz4, vz5, vz6, vz7;
598    vec_f vz8, vz9, vz10, vz11;
599    vec_f vz12, vz13;
600
601    vz0 = vec_ld(byte_8complex, &(out[0]));
602    vz1 = vec_ld(byte_10complex, &(out[0]));
603    vz2 = vec_ld(byte_12complex, &(out[0]));
604    vz3 = vec_ld(byte_14complex, &(out[0]));
605
606    vz4 = vec_perm(vz0, vz1, vcprm(0,1,s2,s1));
607    vz5 = vec_perm(vz0, vz1, vcprm(2,3,s0,s3));
608    vz6 = vec_perm(vz2, vz3, vcprm(0,1,s2,s1));
609    vz7 = vec_perm(vz2, vz3, vcprm(2,3,s0,s3));
610
611    vz0 = vec_add(vz4, vz5);
612    vz1= vec_sub(vz4, vz5);
613    vz2 = vec_add(vz6, vz7);
614    vz3 = vec_sub(vz6, vz7);
615
616    vz4 = vec_perm(vz0, vz1, vcprm(0,s0,1,s1));
617    vz5 = vec_perm(vz0, vz1, vcprm(2,s3,3,s2));
618    vz6 = vec_perm(vz2, vz3, vcprm(0,s0,1,s1));
619    vz7 = vec_perm(vz2, vz3, vcprm(2,s3,3,s2));
620
621    vz0 = vec_add(vz4, vz5);
622    vz1 = vec_sub(vz4, vz5);
623    vz2 = vec_add(vz6, vz7);
624    vz3 = vec_sub(vz6, vz7);
625
626    vz4 = vec_perm(vz0, vz1, vcprm(0,1,s0,s1));
627    vz5 = vec_perm(vz0, vz1, vcprm(2,3,s2,s3));
628
629    vz6 = vec_perm(vz2, vz3, vcprm(0,1,s0,s1));
630    vz7 = vec_perm(vz2, vz3, vcprm(2,3,s2,s3));
631
632    vz0 = vec_ld(0, &(out[0]));
633    vz1 = vec_ld(byte_2complex, &(out[0]));
634    vz2 = vec_ld(byte_4complex, &(out[0]));
635    vz3 = vec_ld(byte_6complex, &(out[0]));
636    vz10 = vec_perm(vz2, vz3, vcprm(0,s0,1,s1));
637    vz11 = vec_perm(vz2, vz3, vcprm(2,s2,3,s3));
638    vz8 = vec_perm(vz0, vz1, vcprm(0,1,s2,s1));
639    vz9 = vec_perm(vz0, vz1, vcprm(2,3,s0,s3));
640
641    vz2 = vec_add(vz10, vz11);
642    vz3 = vec_sub(vz10, vz11);
643    vz12 = vec_perm(vz3, vz3, vcprm(2,3,0,1));
644    vz0 = vec_add(vz8, vz9);
645    vz1 = vec_sub(vz8, vz9);
646
647    vz3 = vec_madd(vz3, vc1, vc0);
648    vz3 = vec_madd(vz12, vc2, vz3);
649    vz8 = vec_perm(vz0, vz1, vcprm(0,s0,1,s1));
650    vz9 = vec_perm(vz0, vz1, vcprm(2,s3,3,s2));
651    vz10 = vec_perm(vz2, vz3, vcprm(1,2,s3,s0));
652    vz11 = vec_perm(vz2, vz3, vcprm(0,3,s2,s1));
653
654    vz0 = vec_add(vz8, vz9);
655    vz1 = vec_sub(vz8, vz9);
656    vz2 = vec_add(vz10, vz11);
657    vz3 = vec_sub(vz10, vz11);
658
659    vz8 = vec_perm(vz0, vz1, vcprm(0,1,s0,s1));
660    vz9 = vec_perm(vz0, vz1, vcprm(2,3,s2,s3));
661    vz10 = vec_perm(vz2, vz3, vcprm(0,2,s1,s3));
662    vz11 = vec_perm(vz2, vz3, vcprm(1,3,s0,s2));
663
664    vz2 = vec_sub(vz8, vz10);
665    vz3 = vec_sub(vz9, vz11);
666    vz0 = vec_add(vz8, vz10);
667    vz1 = vec_add(vz9, vz11);
668
669    vz8 = vec_madd(vz4, vc3, vc0);
670    vz9 = vec_madd(vz5, vc3, vc0);
671    vz10 = vec_madd(vz6, vc3, vc0);
672    vz11 = vec_madd(vz7, vc3, vc0);
673
674    vz8 = vec_madd(vz5, vc4, vz8);
675    vz9 = vec_madd(vz4, vc5, vz9);
676    vz10 = vec_madd(vz7, vc5, vz10);
677    vz11 = vec_madd(vz6, vc4, vz11);
678
679    vz12 = vec_sub(vz10, vz8);
680    vz10 = vec_add(vz10, vz8);
681
682    vz13 = vec_sub(vz9, vz11);
683    vz11 = vec_add(vz9, vz11);
684
685    vz4 = vec_sub(vz0, vz10);
686    vz0 = vec_add(vz0, vz10);
687
688    vz7= vec_sub(vz3, vz12);
689    vz3= vec_add(vz3, vz12);
690
691    vz5 = vec_sub(vz1, vz11);
692    vz1 = vec_add(vz1, vz11);
693
694    vz6 = vec_sub(vz2, vz13);
695    vz2 = vec_add(vz2, vz13);
696
697    vec_st(vz0, 0, &(out[0]));
698    vec_st(vz1, byte_2complex, &(out[0]));
699    vec_st(vz2, byte_4complex, &(out[0]));
700    vec_st(vz3, byte_6complex, &(out[0]));
701    vec_st(vz4, byte_8complex, &(out[0]));
702    vec_st(vz5, byte_10complex, &(out[0]));
703    vec_st(vz6, byte_12complex, &(out[0]));
704    vec_st(vz7, byte_14complex, &(out[0]));
705    return;
706
707}
708inline static void pass_vsx(FFTComplex * z, const FFTSample * wre, unsigned int n)
709{
710    int o1 = n<<1;
711    int o2 = n<<2;
712    int o3 = o1+o2;
713    int i1, i2, i3;
714    FFTSample* out = (FFTSample*)z;
715    const FFTSample *wim = wre+o1;
716    vec_f v0, v1, v2, v3;
717    vec_f v4, v5, v6, v7;
718    vec_f v8, v9, v10, v11;
719    vec_f v12, v13;
720
721    n = n-2;
722    i1 = o1*sizeof(FFTComplex);
723    i2 = o2*sizeof(FFTComplex);
724    i3 = o3*sizeof(FFTComplex);
725
726    v8 = vec_ld(0, &(wre[0]));
727    v10 = vec_ld(0, &(wim[0]));
728    v9 = vec_ld(0, &(wim[-4]));
729    v9 = vec_perm(v9, v10, vcprm(s0,3,2,1));
730
731    v4 = vec_ld(i2, &(out[0]));
732    v5 = vec_ld(i2+16, &(out[0]));
733    v6 = vec_ld(i3, &(out[0]));
734    v7 = vec_ld(i3+16, &(out[0]));
735    v10 = vec_mul(v4, v8); // r2*wre
736    v11 = vec_mul(v5, v8); // i2*wre
737    v12 = vec_mul(v6, v8); // r3*wre
738    v13 = vec_mul(v7, v8); // i3*wre
739
740    v0 = vec_ld(0, &(out[0])); // r0
741    v3 = vec_ld(i1+16, &(out[0])); // i1
742    v10 = vec_madd(v5, v9, v10); // r2*wim
743    v11 = vec_nmsub(v4, v9, v11); // i2*wim
744    v12 = vec_nmsub(v7, v9, v12); // r3*wim
745    v13 = vec_madd(v6, v9, v13); // i3*wim
746
747    v1 = vec_ld(16, &(out[0])); // i0
748    v2 = vec_ld(i1, &(out[0])); // r1
749    v8 = vec_sub(v12, v10);
750    v12 = vec_add(v12, v10);
751    v9 = vec_sub(v11, v13);
752    v13 = vec_add(v11, v13);
753    v4 = vec_sub(v0, v12);
754    v0 = vec_add(v0, v12);
755    v7 = vec_sub(v3, v8);
756    v3 = vec_add(v3, v8);
757
758    vec_st(v0, 0, &(out[0])); // r0
759    vec_st(v3, i1+16, &(out[0])); // i1
760    vec_st(v4, i2, &(out[0])); // r2
761    vec_st(v7, i3+16, &(out[0]));// i3
762
763    v5 = vec_sub(v1, v13);
764    v1 = vec_add(v1, v13);
765    v6 = vec_sub(v2, v9);
766    v2 = vec_add(v2, v9);
767
768    vec_st(v1, 16, &(out[0])); // i0
769    vec_st(v2, i1, &(out[0])); // r1
770    vec_st(v5, i2+16, &(out[0])); // i2
771    vec_st(v6, i3, &(out[0])); // r3
772
773    do {
774        out += 8;
775        wre += 4;
776        wim -= 4;
777
778        v8 = vec_ld(0, &(wre[0]));
779        v10 = vec_ld(0, &(wim[0]));
780        v9 = vec_ld(0, &(wim[-4]));
781        v9 = vec_perm(v9, v10, vcprm(s0,3,2,1));
782
783        v4 = vec_ld(i2, &(out[0])); // r2
784        v5 = vec_ld(i2+16, &(out[0])); // i2
785        v6 = vec_ld(i3, &(out[0])); // r3
786        v7 = vec_ld(i3+16, &(out[0]));// i3
787        v10 = vec_mul(v4, v8); // r2*wre
788        v11 = vec_mul(v5, v8); // i2*wre
789        v12 = vec_mul(v6, v8); // r3*wre
790        v13 = vec_mul(v7, v8); // i3*wre
791
792        v0 = vec_ld(0, &(out[0])); // r0
793        v3 = vec_ld(i1+16, &(out[0])); // i1
794        v10 = vec_madd(v5, v9, v10); // r2*wim
795        v11 = vec_nmsub(v4, v9, v11); // i2*wim
796        v12 = vec_nmsub(v7, v9, v12); // r3*wim
797        v13 = vec_madd(v6, v9, v13); // i3*wim
798
799        v1 = vec_ld(16, &(out[0])); // i0
800        v2 = vec_ld(i1, &(out[0])); // r1
801        v8 = vec_sub(v12, v10);
802        v12 = vec_add(v12, v10);
803        v9 = vec_sub(v11, v13);
804        v13 = vec_add(v11, v13);
805        v4 = vec_sub(v0, v12);
806        v0 = vec_add(v0, v12);
807        v7 = vec_sub(v3, v8);
808        v3 = vec_add(v3, v8);
809
810        vec_st(v0, 0, &(out[0])); // r0
811        vec_st(v3, i1+16, &(out[0])); // i1
812        vec_st(v4, i2, &(out[0])); // r2
813        vec_st(v7, i3+16, &(out[0])); // i3
814
815        v5 = vec_sub(v1, v13);
816        v1 = vec_add(v1, v13);
817        v6 = vec_sub(v2, v9);
818        v2 = vec_add(v2, v9);
819
820        vec_st(v1, 16, &(out[0])); // i0
821        vec_st(v2, i1, &(out[0])); // r1
822        vec_st(v5, i2+16, &(out[0])); // i2
823        vec_st(v6, i3, &(out[0])); // r3
824    } while (n-=2);
825}
826
827#endif
828
829#endif /* AVCODEC_PPC_FFT_VSX_H */
830