decode_altivec.c

mpg123 是 MPEG 1.0/2.0/2.5 的实时播放软件

/*
	decode.c: decoding samples...

	copyright 1995-2006 by the mpg123 project - free software under the terms of the LGPL 2.1
	see COPYING and AUTHORS files in distribution or http://mpg123.de
	initially written by Michael Hipp
	altivec optimization by tmkk
*/

#include <stdlib.h>
#include <math.h>
#include <string.h>

#include "config.h"
#include "mpg123.h"

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

#define WRITE_SAMPLE(samples,sum,clip) \
  if( (sum) > REAL_PLUS_32767) { *(samples) = 0x7fff; (clip)++; } \
  else if( (sum) < REAL_MINUS_32768) { *(samples) = -0x8000; (clip)++; } \
  else { *(samples) = REAL_TO_SHORT(sum); }

int synth_1to1_8bit(real *bandPtr,int channel,unsigned char *samples,int *pnt)
{
  short samples_tmp[64];
  short *tmp1 = samples_tmp + channel;
  int i,ret;
  int pnt1=0;

  ret = synth_1to1(bandPtr,channel,(unsigned char *) samples_tmp,&pnt1);
  samples += channel + *pnt;

  for(i=0;i<32;i++) {
    *samples = conv16to8[*tmp1>>AUSHIFT];
    samples += 2;
    tmp1 += 2;
  }
  *pnt += 64;

  return ret;
}

int synth_1to1_8bit_mono(real *bandPtr,unsigned char *samples,int *pnt)
{
  short samples_tmp[64];
  short *tmp1 = samples_tmp;
  int i,ret;
  int pnt1 = 0;

  ret = synth_1to1(bandPtr,0,(unsigned char *) samples_tmp,&pnt1);
  samples += *pnt;

  for(i=0;i<32;i++) {
    *samples++ = conv16to8[*tmp1>>AUSHIFT];
    tmp1 += 2;
  }
  *pnt += 32;
  
  return ret;
}

int synth_1to1_8bit_mono2stereo(real *bandPtr,unsigned char *samples,int *pnt)
{
  short samples_tmp[64];
  short *tmp1 = samples_tmp;
  int i,ret;
  int pnt1 = 0;

  ret = synth_1to1(bandPtr,0,(unsigned char *) samples_tmp,&pnt1);
  samples += *pnt;

  for(i=0;i<32;i++) {
    *samples++ = conv16to8[*tmp1>>AUSHIFT];
    *samples++ = conv16to8[*tmp1>>AUSHIFT];
    tmp1 += 2;
  }
  *pnt += 64;

  return ret;
}

int synth_1to1_mono(real *bandPtr,unsigned char *samples,int *pnt)
{
  short samples_tmp[64];
  short *tmp1 = samples_tmp;
  int i,ret;
  int pnt1 = 0;

  ret = synth_1to1(bandPtr,0,(unsigned char *) samples_tmp,&pnt1);
  samples += *pnt;

  for(i=0;i<32;i++) {
    *( (short *)samples) = *tmp1;
    samples += 2;
    tmp1 += 2;
  }
  *pnt += 64;

  return ret;
}


int synth_1to1_mono2stereo(real *bandPtr,unsigned char *samples,int *pnt)
{
  int i,ret;

  ret = synth_1to1(bandPtr,0,samples,pnt);
  samples = samples + *pnt - 128;

  for(i=0;i<32;i++) {
    ((short *)samples)[1] = ((short *)samples)[0];
    samples+=4;
  }

  return ret;
}


int synth_1to1(real *bandPtr,int channel,unsigned char *out,int *pnt)
{
  static real __attribute__ ((aligned (16))) buffs[4][4][0x110];
  static const int step = 2;
  static int bo = 1;
  short *samples = (short *) (out+*pnt);

  real *b0,(*buf)[0x110];
  int clip = 0; 
  int bo1;

  if(have_eq_settings)
	do_equalizer(bandPtr,channel);

  if(!channel) {
    bo--;
    bo &= 0xf;
    buf = buffs[0];
  }
  else {
    samples++;
    buf = buffs[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 = decwin + 16 - bo1;
		
		int __attribute__ ((aligned (16))) clip_tmp[4];
		vector float v1,v2,v3,v4,v5,v6,v7,v8,v9;
		vector unsigned char vperm1,vperm2,vperm3,vperm4,vperm5;
		vector float vsum,vsum2,vsum3,vsum4,vmin,vmax;
		vector signed int vclip;
		vector signed short vsample1,vsample2;
		vclip = vec_xor(vclip,vclip);
#ifdef __APPLE__
		vmax = (vector float)(32767.0f);
		vmin = (vector float)(-32768.0f);
		vperm5 = (vector unsigned char)(0,1,18,19,2,3,22,23,4,5,26,27,6,7,30,31);
#else
		vmax = (vector float){32767.0f,32767.0f,32767.0f,32767.0f};
		vmin = (vector float){-32768.0f,-32768.0f,-32768.0f,-32768.0f};
		vperm5 = (vector unsigned char){0,1,18,19,2,3,22,23,4,5,26,27,6,7,30,31};
#endif
		
		vperm1 = vec_lvsl(0,window);
		vperm3 = vec_lvsl(0,samples);
		vperm4 = vec_lvsr(0,samples);
		for (j=4;j;j--)
		{
			vsum = vec_xor(vsum,vsum);
			vsum2 = vec_xor(vsum2,vsum2);
			vsum3 = vec_xor(vsum3,vsum3);
			vsum4 = vec_xor(vsum4,vsum4);
			v1 = vec_ld(0,window);
			v2 = vec_ld(16,window);
			v3 = vec_ld(32,window);
			v4 = vec_ld(48,window);
			v5 = vec_ld(64,window);
			v1 = vec_perm(v1,v2,vperm1);
			v6 = vec_ld(0,b0);
			v2 = vec_perm(v2,v3,vperm1);
			v7 = vec_ld(16,b0);
			v3 = vec_perm(v3,v4,vperm1);
			v8 = vec_ld(32,b0);
			v4 = vec_perm(v4,v5,vperm1);
			v9 = vec_ld(48,b0);
			
			vsum = vec_madd(v1,v6,vsum);
			vsum = vec_madd(v2,v7,vsum);
			vsum = vec_madd(v3,v8,vsum);
			vsum = vec_madd(v4,v9,vsum);
			
			window += 32;
			b0 += 16;
			
			v1 = vec_ld(0,window);
			v2 = vec_ld(16,window);
			v3 = vec_ld(32,window);
			v4 = vec_ld(48,window);
			v5 = vec_ld(64,window);
			v1 = vec_perm(v1,v2,vperm1);
			v6 = vec_ld(0,b0);
			v2 = vec_perm(v2,v3,vperm1);
			v7 = vec_ld(16,b0);
			v3 = vec_perm(v3,v4,vperm1);
			v8 = vec_ld(32,b0);
			v4 = vec_perm(v4,v5,vperm1);
			v9 = vec_ld(48,b0);
			
			vsum2 = vec_madd(v1,v6,vsum2);
			vsum2 = vec_madd(v2,v7,vsum2);
			vsum2 = vec_madd(v3,v8,vsum2);
			vsum2 = vec_madd(v4,v9,vsum2);
			
			window += 32;
			b0 += 16;
			
			v1 = vec_ld(0,window);
			v2 = vec_ld(16,window);
			v3 = vec_ld(32,window);
			v4 = vec_ld(48,window);
			v5 = vec_ld(64,window);
			v1 = vec_perm(v1,v2,vperm1);
			v6 = vec_ld(0,b0);
			v2 = vec_perm(v2,v3,vperm1);
			v7 = vec_ld(16,b0);
			v3 = vec_perm(v3,v4,vperm1);
			v8 = vec_ld(32,b0);
			v4 = vec_perm(v4,v5,vperm1);
			v9 = vec_ld(48,b0);
			
			vsum3 = vec_madd(v1,v6,vsum3);
			vsum3 = vec_madd(v2,v7,vsum3);
			vsum3 = vec_madd(v3,v8,vsum3);
			vsum3 = vec_madd(v4,v9,vsum3);
			
			window += 32;
			b0 += 16;
			
			v1 = vec_ld(0,window);
			v2 = vec_ld(16,window);
			v3 = vec_ld(32,window);
			v4 = vec_ld(48,window);
			v5 = vec_ld(64,window);
			v1 = vec_perm(v1,v2,vperm1);
			v6 = vec_ld(0,b0);
			v2 = vec_perm(v2,v3,vperm1);
			v7 = vec_ld(16,b0);
			v3 = vec_perm(v3,v4,vperm1);
			v8 = vec_ld(32,b0);
			v4 = vec_perm(v4,v5,vperm1);
			v9 = vec_ld(48,b0);
			
			vsum4 = vec_madd(v1,v6,vsum4);
			vsum4 = vec_madd(v2,v7,vsum4);
			vsum4 = vec_madd(v3,v8,vsum4);
			vsum4 = vec_madd(v4,v9,vsum4);
			
			window += 32;
			b0 += 16;
			
			v1 = vec_mergeh(vsum,vsum3);
			v2 = vec_mergeh(vsum2,vsum4);
			v3 = vec_mergel(vsum,vsum3);
			v4 = vec_mergel(vsum2,vsum4);
			v5 = vec_mergeh(v1,v2);
			v6 = vec_mergel(v1,v2);
			v7 = vec_mergeh(v3,v4);
			v8 = vec_mergel(v3,v4);
			
			vsum = vec_sub(v5,v6);
			v9 = vec_sub(v7,v8);
			vsum = vec_add(vsum,v9);
			
			v3 = (vector float)vec_cts(vsum,0);
			v1 = (vector float)vec_cmpgt(vsum,vmax);
			v2 = (vector float)vec_cmplt(vsum,vmin);
			vsample1 = vec_ld(0,samples);
			vsample2 = vec_ld(15,samples);