comparison mp3lib/decod386.c @ 9002:60d144a16088

An altivec-optimized DCT64 for mp3lib (partially, it seems roughly three times as fast as the C code according to quick-n-dirty gprof tests) This one is bit-perfect. patch by Romain Dolbeau <[e-mail]>

9001:01a9cf43074c

    static int bo = 1;
    synth_1to1_MMX_s(bandPtr, channel, samples, (short *) buffs, &bo); 
    return 0;
} 
#endif
static int synth_1to1(real *bandPtr,int channel,unsigned char *out,int *pnt)
{
  static real buffs[2][2][0x110];
  static const int step = 2;
  static int bo = 1;

  }

  if(bo & 0x1) {
    b0 = buf[0];
    bo1 = bo;
    dct64(buf[1]+((bo+1)&0xf),buf[0]+bo,bandPtr);
  }
  else {
    b0 = buf[1];
    bo1 = bo+1;
    dct64(buf[0]+bo,buf[1]+bo+1,bandPtr);
  }

  {
    register int j;
    real *window = mp3lib_decwin + 16 - bo1;

9002:60d144a16088

    static int bo = 1;
    synth_1to1_MMX_s(bandPtr, channel, samples, (short *) buffs, &bo); 
    return 0;
} 
#endif

#ifdef HAVE_ALTIVEC
#define dct64_base(a,b,c) if(gCpuCaps.hasAltiVec) dct64_altivec(a,b,c); else dct64(a,b,c)
#else /* HAVE_ALTIVEC */
#define dct64_base(a,b,c) dct64(a,b,c)
#endif /* HAVE_ALTIVEC */
 
static int synth_1to1(real *bandPtr,int channel,unsigned char *out,int *pnt)
{
  static real buffs[2][2][0x110];
  static const int step = 2;
  static int bo = 1;

  }

  if(bo & 0x1) {
    b0 = buf[0];
    bo1 = bo;
    dct64_base(buf[1]+((bo+1)&0xf),buf[0]+bo,bandPtr);
  }
  else {
    b0 = buf[1];
    bo1 = bo+1;
    dct64_base(buf[0]+bo,buf[1]+bo+1,bandPtr);
  }

  {
    register int j;
    real *window = mp3lib_decwin + 16 - bo1;