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