Remove PPC perf counter support
[libav.git] / libavcodec / ppc / fft_altivec.c
1 /*
2 * FFT/IFFT transforms
3 * AltiVec-enabled
4 * Copyright (c) 2003 Romain Dolbeau <romain@dolbeau.org>
5 * Based on code Copyright (c) 2002 Fabrice Bellard
6 *
7 * This file is part of FFmpeg.
8 *
9 * FFmpeg is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; either
12 * version 2.1 of the License, or (at your option) any later version.
13 *
14 * FFmpeg is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
18 *
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with FFmpeg; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 */
23 #include "libavcodec/fft.h"
24 #include "util_altivec.h"
25 #include "dsputil_altivec.h"
26
27 /**
28 * Do a complex FFT with the parameters defined in ff_fft_init(). The
29 * input data must be permuted before with s->revtab table. No
30 * 1.0/sqrt(n) normalization is done.
31 * AltiVec-enabled
32 * This code assumes that the 'z' pointer is 16 bytes-aligned
33 * It also assumes all FFTComplex are 8 bytes-aligned pair of float
34 * The code is exactly the same as the SSE version, except
35 * that successive MUL + ADD/SUB have been merged into
36 * fused multiply-add ('vec_madd' in altivec)
37 */
38 static void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z)
39 {
40 register const vector float vczero = (const vector float)vec_splat_u32(0.);
41
42 int ln = s->nbits;
43 int j, np, np2;
44 int nblocks, nloops;
45 register FFTComplex *p, *q;
46 FFTComplex *cptr, *cptr1;
47 int k;
48
49 np = 1 << ln;
50
51 {
52 vector float *r, a, b, a1, c1, c2;
53
54 r = (vector float *)&z[0];
55
56 c1 = vcii(p,p,n,n);
57
58 if (s->inverse) {
59 c2 = vcii(p,p,n,p);
60 } else {
61 c2 = vcii(p,p,p,n);
62 }
63
64 j = (np >> 2);
65 do {
66 a = vec_ld(0, r);
67 a1 = vec_ld(sizeof(vector float), r);
68
69 b = vec_perm(a,a,vcprmle(1,0,3,2));
70 a = vec_madd(a,c1,b);
71 /* do the pass 0 butterfly */
72
73 b = vec_perm(a1,a1,vcprmle(1,0,3,2));
74 b = vec_madd(a1,c1,b);
75 /* do the pass 0 butterfly */
76
77 /* multiply third by -i */
78 b = vec_perm(b,b,vcprmle(2,3,1,0));
79
80 /* do the pass 1 butterfly */
81 vec_st(vec_madd(b,c2,a), 0, r);
82 vec_st(vec_nmsub(b,c2,a), sizeof(vector float), r);
83
84 r += 2;
85 } while (--j != 0);
86 }
87 /* pass 2 .. ln-1 */
88
89 nblocks = np >> 3;
90 nloops = 1 << 2;
91 np2 = np >> 1;
92
93 cptr1 = s->exptab1;
94 do {
95 p = z;
96 q = z + nloops;
97 j = nblocks;
98 do {
99 cptr = cptr1;
100 k = nloops >> 1;
101 do {
102 vector float a,b,c,t1;
103
104 a = vec_ld(0, (float*)p);
105 b = vec_ld(0, (float*)q);
106
107 /* complex mul */
108 c = vec_ld(0, (float*)cptr);
109 /* cre*re cim*re */
110 t1 = vec_madd(c, vec_perm(b,b,vcprmle(2,2,0,0)),vczero);
111 c = vec_ld(sizeof(vector float), (float*)cptr);
112 /* -cim*im cre*im */
113 b = vec_madd(c, vec_perm(b,b,vcprmle(3,3,1,1)),t1);
114
115 /* butterfly */
116 vec_st(vec_add(a,b), 0, (float*)p);
117 vec_st(vec_sub(a,b), 0, (float*)q);
118
119 p += 2;
120 q += 2;
121 cptr += 4;
122 } while (--k);
123
124 p += nloops;
125 q += nloops;
126 } while (--j);
127 cptr1 += nloops * 2;
128 nblocks = nblocks >> 1;
129 nloops = nloops << 1;
130 } while (nblocks != 0);
131 }
132
133 av_cold void ff_fft_init_altivec(FFTContext *s)
134 {
135 s->fft_calc = ff_fft_calc_altivec;
136 s->split_radix = 0;
137 }