view liba52/resample_altivec.c @ 13568:1cb0e1833515

Currently vbeGetProtModeInfo call the 0x4f0a function of int 10h the get a simple 32 bits protected mode interface to some VESA functions. This protected mode interface is interesting because it's quicker than the raw int 10h interface. Unfortunatly, begining with VBE 3.0, the 0x4f0a function is optional, and some video cards don't implement it (3dfx, intel 845/855/865...). This protected mode interface is then only used in vbeSetWindow and vbeSetDisplayStart : ?- vbeSetWindow already implement an alternative methode if protected mode interface is not available. ?- vbeSetDisplayStart also contain an alternative implementation, but this one is disabled with a #if 0. I don't exactly know why because it works well ! So currently, cards which don't have the 0x4f0a function are not supported. This patch correct this. ?- vbeGetProtModeInfo failure is not fatal. ?- vbeSetDisplayStart has it's alternative implementation reenabled. ? ?it's used only with cards which don't have the 0x4f0a function ? ?so this won't make any difference for cards which were already ? ?working. This patch also make the failure of vbeGetModeInfo not fatal. The VBE 3.0 standard state that GetModeInfo can fail with some mode which are listed as supported if the mode can't be used in the current situation (not enough video memory for example). So a failure of vbeGetModeInfo don't mean that other modes won't work and should really not be fatal. patch by Aurelien Jacobs <[e-mail]>

// this code is based on a52dec/libao/audio_out_oss.c
// AltiVec support Copyright (c) 2004 Romain Dolbeau <[e-mail]>

#ifndef SYS_DARWIN
#include <altivec.h>
#endif

const vector signed int magic = {0x43c00000,0x43c00000,0x43c00000,0x43c00000};

static inline vector signed short convert16_altivec(vector signed int v1, vector signed int v2)
{
  register vector signed short result;
  v1 = vec_subs(v1, magic);
  v2 = vec_subs(v2, magic);
  result = vec_packs(v1, v2);

  return result;
}

static int a52_resample_STEREO_to_2_altivec(float * _f, int16_t * s16){
#if 0
  int i;
  int32_t * f = (int32_t *) _f;
  for (i = 0; i < 256; i++) {
    s16[2*i] = convert (f[i]);
    s16[2*i+1] = convert (f[i+256]);
  }
  return 2*256;
#else
  int i = 0;
  int32_t * f = (int32_t *) _f;
  register vector signed int f0, f4, f256, f260;
  register vector signed short reven, rodd, r0, r1;

  for (i = 0; i < 256; i+= 8) {
    f0 = vec_ld(0, f);
    f4 = vec_ld(16, f);
    
    f256 = vec_ld(1024, f);
    f260 = vec_ld(1040, f);

    reven = convert16_altivec(f0, f4);
    rodd = convert16_altivec(f256, f260);

    r0 = vec_mergeh(reven, rodd);
    r1 = vec_mergel(reven, rodd);
    
    vec_st(r0, 0, s16);
    vec_st(r1, 16, s16);

    f += 8;
    s16 += 16;
  }
  return(2*256);
#endif
}

static void* a52_resample_altivec(int flags, int ch){
fprintf(stderr, "Checking for AltiVec resampler : 0x%08x, %d\n", flags, ch);

  switch (flags) {
  case A52_CHANNEL:
  case A52_STEREO:
  case A52_DOLBY:
    if(ch==2) return a52_resample_STEREO_to_2_altivec;
    break;

  default:
	fprintf(stderr, "Unsupported flags: 0x%08x (%d channels)\n", flags, ch);
	break;
  }
  return NULL;
}