apps/codecs/lib/fft-ffmpeg_cf.h

   1 /***************************************************************************
   2  *             __________               __   ___.
   3  *   Open      \______   \ ____   ____ |  | _\_ |__   _______  ___
   4  *   Source     |       _//  _ \_/ ___\|  |/ /| __ \ /  _ \  \/  /
   5  *   Jukebox    |    |   (  <_> )  \___|    < | \_\ (  <_> > <  <
   6  *   Firmware   |____|_  /\____/ \___  >__|_ \|___  /\____/__/\_ \
   7  *                     \/            \/     \/    \/            \/
   8  * $Id$
   9  *
  10  * Copyright (C) 2010 Nils Wallménius
  11  *
  12  * Coldfire v2 optimisations for ffmpeg's fft (used in fft-ffmpeg.c)
  13  *
  14  * This program is free software; you can redistribute it and/or
  15  * modify it under the terms of the GNU General Public License
  16  * as published by the Free Software Foundation; either version 2
  17  * of the License, or (at your option) any later version.
  18  *
  19  * This software is distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY
  20  * KIND, either express or implied.
  21  *
  22  ****************************************************************************/
  23
  24 #ifdef CPU_COLDFIRE
  25 #define FFT_FFMPEG_INCL_OPTIMISED_FFT4
  26 static inline void fft4(FFTComplex * z)
  27 {
  28     asm volatile ("movem.l (%[z]), %%d0-%%d7\n\t"
  29                   "move.l  %%d0, %%a0\n\t"
  30                   "add.l   %%d2, %%d0\n\t" /* d0 == t1 */
  31                   "neg.l   %%d2\n\t"
  32                   "add.l   %%a0, %%d2\n\t" /* d2 == t3, a0 free */
  33                   "move.l  %%d6, %%a0\n\t"
  34                   "sub.l   %%d4, %%d6\n\t" /* d6 == t8 */
  35                   "add.l   %%d4, %%a0\n\t" /* a0 == t6 */
  36
  37                   "move.l  %%d0, %%d4\n\t"
  38                   "sub.l   %%a0, %%d4\n\t" /* z[2].re done */
  39                   "add.l   %%a0, %%d0\n\t" /* z[0].re done, a0 free */
  40
  41                   "move.l  %%d5, %%a0\n\t"
  42                   "sub.l   %%d7, %%d5\n\t" /* d5 == t7 */
  43                   "add.l   %%d7, %%a0\n\t" /* a0 == t5 */
  44
  45                   "move.l  %%d1, %%d7\n\t"
  46                   "sub.l   %%d3, %%d7\n\t" /* d7 == t4 */
  47                   "add.l   %%d3, %%d1\n\t" /* d1 == t2 */
  48
  49                   "move.l  %%d7, %%d3\n\t"
  50                   "sub.l   %%d6, %%d7\n\t" /* z[3].im done */
  51                   "add.l   %%d6, %%d3\n\t" /* z[1].im done */
  52
  53                   "move.l  %%d2, %%d6\n\t"
  54                   "sub.l   %%d5, %%d6\n\t" /* z[3].re done */
  55                   "add.l   %%d5, %%d2\n\t" /* z[1].re done */
  56
  57                   "move.l  %%d1, %%d5\n\t"
  58                   "sub.l   %%a0, %%d5\n\t" /* z[2].im done */
  59                   "add.l   %%a0, %%d1\n\t" /* z[0].im done */
  60
  61                   "movem.l %%d0-%%d7, (%[z])\n\t"
  62                   : :[z] "a" (z)
  63                   : "d0", "d1", "d2", "d3", "d4", "d5", "d6", "d7",
  64                     "a0", "cc", "memory");
  65
  66 }
  67
  68 #define FFT_FFMPEG_INCL_OPTIMISED_FFT8
  69 static inline void fft8(FFTComplex *z)
  70 {
  71     asm volatile ("movem.l (4*8, %[z]), %%d0-%%d7\n\t"
  72                   "move.l  %%d0, %%a1\n\t"
  73                   "add.l   %%d2, %%a1\n\t" /* a1 == t1 */
  74                   "sub.l   %%d2, %%d0\n\t" /* d0 == z[5].re */
  75
  76                   "move.l  %%d1, %%a2\n\t"
  77                   "add.l   %%d3, %%a2\n\t" /* a2 == t2 */
  78                   "sub.l   %%d3, %%d1\n\t" /* d1 == z[5].im */
  79
  80                   "move.l  %%d4, %%d2\n\t"
  81                   "add.l   %%d6, %%d2\n\t" /* d2 == t3 */
  82                   "sub.l   %%d6, %%d4\n\t" /* d4 == z[7].re */
  83
  84                   "move.l  %%d5, %%d3\n\t"
  85                   "add.l   %%d7, %%d3\n\t" /* d3 == t4 */
  86                   "sub.l   %%d7, %%d5\n\t" /* d5 == z[7].im */
  87
  88                   "move.l  %%d2, %%a4\n\t"
  89                   "sub.l   %%a1, %%a4\n\t" /* a4 == t8 */
  90                   "add.l   %%d2, %%a1\n\t" /* a1 == t1, d2 free */
  91
  92                   "move.l  %%a2, %%a3\n\t"
  93                   "sub.l   %%d3, %%a3\n\t" /* a3 == t7 */
  94                   "add.l   %%d3, %%a2\n\t" /* a2 == t2, d3 free */
  95
  96                   /* emac block from TRANSFORM_EQUAL, do this now
  97                      so we don't need to store and load z[5] and z[7] */
  98                   "move.l  %[_cPI2_8], %%d2\n\t"
  99                   "mac.l   %%d2, %%d0, %%acc0\n\t"
 100                   "mac.l   %%d2, %%d1, %%acc1\n\t"
 101                   "mac.l   %%d2, %%d4, %%acc2\n\t"
 102                   "mac.l   %%d2, %%d5, %%acc3\n\t"
 103
 104                   /* fft4, clobbers all d regs and a0 */
 105                   "movem.l (%[z]), %%d0-%%d7\n\t"
 106                   "move.l  %%d0, %%a0\n\t"
 107                   "add.l   %%d2, %%d0\n\t" /* d0 == t1 */
 108                   "neg.l   %%d2\n\t"
 109                   "add.l   %%a0, %%d2\n\t" /* d2 == t3, a0 free */
 110                   "move.l  %%d6, %%a0\n\t"
 111                   "sub.l   %%d4, %%d6\n\t" /* d6 == t8 */
 112                   "add.l   %%d4, %%a0\n\t" /* a0 == t6 */
 113
 114                   "move.l  %%d0, %%d4\n\t"
 115                   "sub.l   %%a0, %%d4\n\t" /* z[2].re done */
 116                   "add.l   %%a0, %%d0\n\t" /* z[0].re done, a0 free */
 117
 118                   "move.l  %%d5, %%a0\n\t"
 119                   "sub.l   %%d7, %%d5\n\t" /* d5 == t7 */
 120                   "add.l   %%d7, %%a0\n\t" /* a0 == t5 */
 121
 122                   "move.l  %%d1, %%d7\n\t"
 123                   "sub.l   %%d3, %%d7\n\t" /* d7 == t4 */
 124                   "add.l   %%d3, %%d1\n\t" /* d1 == t2 */
 125
 126                   "move.l  %%d7, %%d3\n\t"
 127                   "sub.l   %%d6, %%d7\n\t" /* z[3].im done */
 128                   "add.l   %%d6, %%d3\n\t" /* z[1].im done */
 129
 130                   "move.l  %%d2, %%d6\n\t"
 131                   "sub.l   %%d5, %%d6\n\t" /* z[3].re done */
 132                   "add.l   %%d5, %%d2\n\t" /* z[1].re done */
 133
 134                   "move.l  %%d1, %%d5\n\t"
 135                   "sub.l   %%a0, %%d5\n\t" /* z[2].im done */
 136                   "add.l   %%a0, %%d1\n\t" /* z[0].im done */
 137                   /* end of fft4, but don't store yet */
 138
 139                   "move.l  %%d0, %%a0\n\t"
 140                   "add.l   %%a1, %%d0\n\t"
 141                   "sub.l   %%a1, %%a0\n\t" /* z[4].re, z[0].re done, a1 free */
 142
 143                   "move.l  %%d1, %%a1\n\t"
 144                   "add.l   %%a2, %%d1\n\t"
 145                   "sub.l   %%a2, %%a1\n\t" /* z[4].im, z[0].im done, a2 free */
 146
 147                   "move.l  %%d4, %%a2\n\t"
 148                   "add.l   %%a3, %%d4\n\t"
 149                   "sub.l   %%a3, %%a2\n\t" /* z[6].re, z[2].re done, a3 free */
 150
 151                   "move.l  %%d5, %%a3\n\t"
 152                   "add.l   %%a4, %%d5\n\t"
 153                   "sub.l   %%a4, %%a3\n\t" /* z[6].im, z[2].im done, a4 free */
 154
 155                   "movem.l %%d0-%%d1, (%[z])\n\t"      /* save z[0] */
 156                   "movem.l %%d4-%%d5, (2*8, %[z])\n\t" /* save z[2] */
 157                   "movem.l %%a0-%%a1, (4*8, %[z])\n\t" /* save z[4] */
 158                   "movem.l %%a2-%%a3, (6*8, %[z])\n\t" /* save z[6] */
 159
 160                   /* TRANSFORM_EQUAL */
 161                   "movclr.l %%acc0, %%d0\n\t"
 162                   "movclr.l %%acc1, %%d1\n\t"
 163                   "movclr.l %%acc2, %%d4\n\t"
 164                   "movclr.l %%acc3, %%d5\n\t"
 165
 166                   "move.l  %%d1, %%a0\n\t"
 167                   "add.l   %%d0, %%a0\n\t" /* a0 == t1 */
 168                   "sub.l   %%d0, %%d1\n\t" /* d1 == t2 */
 169
 170                   "move.l  %%d4, %%d0\n\t"
 171                   "add.l   %%d5, %%d0\n\t" /* d0 == t6 */
 172                   "sub.l   %%d5, %%d4\n\t" /* d4 == t5 */
 173
 174                   "move.l  %%d4, %%a1\n\t"
 175                   "sub.l   %%a0, %%a1\n\t" /* a1 == temp1 */
 176                   "add.l   %%a0, %%d4\n\t" /* d4 == temp2 */
 177
 178                   "move.l  %%d2, %%a2\n\t"
 179                   "sub.l   %%d4, %%a2\n\t" /* a2 == z[5].re */
 180                   "add.l   %%d4, %%d2\n\t" /* z[1].re done */
 181
 182                   "move.l  %%d7, %%d5\n\t"
 183                   "sub.l   %%a1, %%d5\n\t" /* d5 == z[7].im */
 184                   "add.l   %%a1, %%d7\n\t" /* z[3].im done */
 185
 186                   "move.l  %%d1, %%a0\n\t"
 187                   "sub.l   %%d0, %%a0\n\t" /* a0 == temp1 */
 188                   "add.l   %%d0, %%d1\n\t" /* d1 == temp2 */
 189
 190                   "move.l  %%d6, %%d4\n\t"
 191                   "sub.l   %%a0, %%d4\n\t" /* d4 == z[7].re */
 192                   "add.l   %%a0, %%d6\n\t" /* z[3].re done */
 193
 194                   "move.l  %%d3, %%a3\n\t"
 195                   "sub.l   %%d1, %%a3\n\t" /* a3 == z[5].im */
 196                   "add.l   %%d1, %%d3\n\t" /* z[1].im done */
 197
 198                   "movem.l %%d2-%%d3, (1*8, %[z])\n\t" /* save z[1] */
 199                   "movem.l %%d6-%%d7, (3*8, %[z])\n\t" /* save z[3] */
 200                   "movem.l %%a2-%%a3, (5*8, %[z])\n\t" /* save z[5] */
 201                   "movem.l %%d4-%%d5, (7*8, %[z])\n\t" /* save z[7] */
 202                   : :[z] "a" (z), [_cPI2_8] "i" (cPI2_8)
 203                   : "d0", "d1", "d2", "d3", "d4", "d5", "d6", "d7",
 204                     "a0", "a1", "a2", "a3", "a4", "cc", "memory");
 205 }
 206 #endif /* CPU_COLDIFRE */