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 "dsputil_ppc.h"
25#include "util_altivec.h"
26#include "dsputil_altivec.h"
27
28/**
29 * Do a complex FFT with the parameters defined in ff_fft_init(). The
30 * input data must be permuted before with s->revtab table. No
31 * 1.0/sqrt(n) normalization is done.
32 * AltiVec-enabled
33 * This code assumes that the 'z' pointer is 16 bytes-aligned
34 * It also assumes all FFTComplex are 8 bytes-aligned pair of float
35 * The code is exactly the same as the SSE version, except
36 * that successive MUL + ADD/SUB have been merged into
37 * fused multiply-add ('vec_madd' in altivec)
38 */
39static void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z)
40{
41POWERPC_PERF_DECLARE(altivec_fft_num, s->nbits >= 6);
42    register const vector float vczero = (const vector float)vec_splat_u32(0.);
43
44    int ln = s->nbits;
45    int j, np, np2;
46    int nblocks, nloops;
47    register FFTComplex *p, *q;
48    FFTComplex *cptr, *cptr1;
49    int k;
50
51POWERPC_PERF_START_COUNT(altivec_fft_num, s->nbits >= 6);
52
53    np = 1 << ln;
54
55    {
56        vector float *r, a, b, a1, c1, c2;
57
58        r = (vector float *)&z[0];
59
60        c1 = vcii(p,p,n,n);
61
62        if (s->inverse) {
63            c2 = vcii(p,p,n,p);
64        } else {
65            c2 = vcii(p,p,p,n);
66        }
67
68        j = (np >> 2);
69        do {
70            a = vec_ld(0, r);
71            a1 = vec_ld(sizeof(vector float), r);
72
73            b = vec_perm(a,a,vcprmle(1,0,3,2));
74            a = vec_madd(a,c1,b);
75            /* do the pass 0 butterfly */
76
77            b = vec_perm(a1,a1,vcprmle(1,0,3,2));
78            b = vec_madd(a1,c1,b);
79            /* do the pass 0 butterfly */
80
81            /* multiply third by -i */
82            b = vec_perm(b,b,vcprmle(2,3,1,0));
83
84            /* do the pass 1 butterfly */
85            vec_st(vec_madd(b,c2,a), 0, r);
86            vec_st(vec_nmsub(b,c2,a), sizeof(vector float), r);
87
88            r += 2;
89        } while (--j != 0);
90    }
91    /* pass 2 .. ln-1 */
92
93    nblocks = np >> 3;
94    nloops = 1 << 2;
95    np2 = np >> 1;
96
97    cptr1 = s->exptab1;
98    do {
99        p = z;
100        q = z + nloops;
101        j = nblocks;
102        do {
103            cptr = cptr1;
104            k = nloops >> 1;
105            do {
106                vector float a,b,c,t1;
107
108                a = vec_ld(0, (float*)p);
109                b = vec_ld(0, (float*)q);
110
111                /* complex mul */
112                c = vec_ld(0, (float*)cptr);
113                /*  cre*re cim*re */
114                t1 = vec_madd(c, vec_perm(b,b,vcprmle(2,2,0,0)),vczero);
115                c = vec_ld(sizeof(vector float), (float*)cptr);
116                /*  -cim*im cre*im */
117                b = vec_madd(c, vec_perm(b,b,vcprmle(3,3,1,1)),t1);
118
119                /* butterfly */
120                vec_st(vec_add(a,b), 0, (float*)p);
121                vec_st(vec_sub(a,b), 0, (float*)q);
122
123                p += 2;
124                q += 2;
125                cptr += 4;
126            } while (--k);
127
128            p += nloops;
129            q += nloops;
130        } while (--j);
131        cptr1 += nloops * 2;
132        nblocks = nblocks >> 1;
133        nloops = nloops << 1;
134    } while (nblocks != 0);
135
136POWERPC_PERF_STOP_COUNT(altivec_fft_num, s->nbits >= 6);
137}
138
139av_cold void ff_fft_init_altivec(FFTContext *s)
140{
141    s->fft_calc = ff_fft_calc_altivec;
142    s->split_radix = 0;
143}
144