recon.cc

ac3的解码程序

/*
  File: recon.hh

  */

#include "all.hh"

// #define HAVE_3Dnow

#ifdef HAVE_MMX
#ifndef HAVE_3Dnow
	static long long ADD_1	=	0x0101010101010101LL;
	static long long MASK_AND = 0x7f7f7f7f7f7f7f7fLL;
#endif
#endif

#ifdef HAVE_MMX
#ifndef HAVE_3Dnow
extern "C" void recva(unsigned char *s, unsigned char *d, int lx, int lx2, int h);
extern "C" void recvac(unsigned char *s, unsigned char *d, int lx, int lx2, int h);
extern "C" void rech(unsigned char *s, unsigned char *d, int lx2, int h);
extern "C" void rechc(unsigned char *s, unsigned char *d, int lx2, int h);
#else
	inline void recva(unsigned char *s, unsigned char *d, int lx, int lx2, int h);
	inline void recvac(unsigned char *s, unsigned char *d, int lx, int lx2, int h);
	inline void rech(unsigned char *s, unsigned char *d, int lx2, int h);
	inline void rechc(unsigned char *s, unsigned char *d, int lx2, int h);
#endif
#else
	inline void recva(unsigned char *s, unsigned char *d, int lx, int lx2, int h);
	inline void recvac(unsigned char *s, unsigned char *d, int lx, int lx2, int h);
	inline void rech(unsigned char *s, unsigned char *d, int lx2, int h);
	inline void rechc(unsigned char *s, unsigned char *d, int lx2, int h);
#endif

inline void rec(unsigned char *s, unsigned char *d, int lx2, int h);
inline void recc(unsigned char *s, unsigned char *d, int lx2, int h);
inline void reca(unsigned char *s, unsigned char *d, int lx2, int h);
inline void recac(unsigned char *s, unsigned char *d, int lx2, int h);
inline void recv(unsigned char *s, unsigned char *d, int lx, int lx2, int h);
inline void recvc(unsigned char *s, unsigned char *d, int lx, int lx2, int h);

inline void recha(unsigned char *s, unsigned char *d, int lx2, int h);
inline void rechac(unsigned char *s, unsigned char *d, int lx2, int h);
inline void rec4(unsigned char *s, unsigned char *d, int lx, int lx2, int h);
inline void rec4c(unsigned char *s, unsigned char *d, int lx, int lx2, int h);
inline void rec4a(unsigned char *s, unsigned char *d, int lx, int lx2, int h);
inline void rec4ac(unsigned char *s, unsigned char *d, int lx, int lx2, int h);

void recon_comp(unsigned char *src, unsigned char *dst,
                  int lx, int lx2, int w, int h, int x, int y, int dx, int dy, int addflag);



void recon(unsigned char *src[], int sfield, 
	   unsigned char *dst[], int dfield, int lx,int lx2,
	   int w, int h, int x, int y, int dx, int dy, int addflag){
  /*
    unsigned char *src[]; * prediction source buffer *
    int sfield;           * prediction source field number (0 or 1) *
    unsigned char *dst[]; * prediction destination buffer *
    int dfield;           * prediction destination field number (0 or 1)*
    int lx,lx2;           * horizontal offsets *
    int w,h;              * prediction block/sub-block width, height *
    int x,y;              * pixel co-ordinates of top-left sample in current MB *
    int dx,dy;            * horizontal, vertical motion vector *
    int addflag;          * add prediction error to prediction ? *
    */
  /* Y */
  recon_comp((src[0]+(sfield?lx2>>1:0)), 
	     dst[0]+(dfield?lx2>>1:0),
             lx,lx2,w,h,x,y,dx,dy,addflag);
  
  if (chroma_format!=CHROMA444){
    lx>>=1; dx/=2;lx2>>=1; w = 0; x>>=1; 
  }

  if (chroma_format==CHROMA420){
    h>>=1; dy/=2;y>>=1; 
  }
  
  /* Cb */
  recon_comp((src[1]+(sfield?lx2>>1:0)), 
	     dst[1]+(dfield?lx2>>1:0),
	     lx,lx2,w,h,x,y,dx,dy,addflag);
  /* Cr */
  recon_comp((src[2]+(sfield?lx2>>1:0)),
	     dst[2]+(dfield?lx2>>1:0),
             lx,lx2,w,h,x,y,dx,dy,addflag);
}

//#define WIDTH 16
#define WIDTH 1
void LayerData::reconstruct(int bx, int by, int mb_type, int motion_type,
			    int PMV[2][2][2], int mv_field_sel[2][2], 
			    int dmvector[2], int stwtype){
  int currentfield;
  unsigned char **predframe;
  int DMV[2][2];
  int stwtop, stwbot;

  stwtop = stwtype%3; /* 0:temporal, 1:(spat+temp)/2, 2:spatial */
  stwbot = stwtype/3;

  if ((mb_type & MB_FORWARD) || (pict_type==P_TYPE)){
    if (pict_struct==FRAME_PICTURE){
      if ((motion_type==MC_FRAME) || !(mb_type & MB_FORWARD)){
        /* frame-based prediction */
		{
        if (stwtop<2)
          recon(oldrefframe,0,newframe,0,
        	    coded_picture_width,coded_picture_width<<1,WIDTH,8,bx,by,
            PMV[0][0][0],PMV[0][0][1],stwtop);

        if (stwbot<2)
          recon(oldrefframe,1,newframe,1,
        	    coded_picture_width,coded_picture_width<<1,WIDTH,8,bx,by,
            PMV[0][0][0],PMV[0][0][1],stwbot);
      }
      }
      else if (motion_type==MC_FIELD) /* field-based prediction */
      {
        /* top field prediction */
        if (stwtop<2)
          recon(oldrefframe,mv_field_sel[0][0],newframe,0,
            coded_picture_width<<1,coded_picture_width<<1,WIDTH,8,bx,by>>1,
            PMV[0][0][0],PMV[0][0][1]>>1,stwtop);

        /* bottom field prediction */
        if (stwbot<2)
          recon(oldrefframe,mv_field_sel[1][0],newframe,1,
            coded_picture_width<<1,coded_picture_width<<1,WIDTH,8,bx,by>>1,
            PMV[1][0][0],PMV[1][0][1]>>1,stwbot);
      }
      else if (motion_type==MC_DMV){ /* dual prime prediction */
        /* calculate derived motion vectors */
        ld->calc_DMV(DMV,dmvector,PMV[0][0][0],PMV[0][0][1]>>1);

        if (stwtop<2){
          /* predict top field from top field */
          recon(oldrefframe,0,newframe,0,
            coded_picture_width<<1,coded_picture_width<<1,WIDTH,8,bx,by>>1,
            PMV[0][0][0],PMV[0][0][1]>>1,0);

          /* predict and add to top field from bottom field */
          recon(oldrefframe,1,newframe,0,
            coded_picture_width<<1,coded_picture_width<<1,WIDTH,8,bx,by>>1,
            DMV[0][0],DMV[0][1],1);
        }

        if (stwbot<2)
        {
          /* predict bottom field from bottom field */
          recon(oldrefframe,1,newframe,1,
            coded_picture_width<<1,coded_picture_width<<1,WIDTH,8,bx,by>>1,
            PMV[0][0][0],PMV[0][0][1]>>1,0);

          /* predict and add to bottom field from top field */
          recon(oldrefframe,0,newframe,1,
            coded_picture_width<<1,coded_picture_width<<1,WIDTH,8,bx,by>>1,
            DMV[1][0],DMV[1][1],1);
        }
      }
      else
        /* invalid motion_type */
        if (!quiet) printf("invalid motion_type\n");
    }
    else /* TOP_FIELD or BOTTOM_FIELD */
    {
      /* field picture */
      currentfield = (pict_struct==BOTTOM_FIELD);

      /* determine which frame to use for prediction */
      if ((pict_type==P_TYPE) && secondfield
         && (currentfield!=mv_field_sel[0][0]))
        predframe = refframe; /* same frame */
      else
        predframe = oldrefframe; /* previous frame */

      if ((motion_type==MC_FIELD) || !(mb_type & MB_FORWARD))
      {
        /* field-based prediction */
        if (stwtop<2)
          recon(predframe,mv_field_sel[0][0],newframe,0,
            coded_picture_width<<1,coded_picture_width<<1,WIDTH,16,bx,by,
            PMV[0][0][0],PMV[0][0][1],stwtop);
      }
      else if (motion_type==MC_16X8)
      {
        if (stwtop<2)
        {
          recon(predframe,mv_field_sel[0][0],newframe,0,
            coded_picture_width<<1,coded_picture_width<<1,WIDTH,8,bx,by,
            PMV[0][0][0],PMV[0][0][1],stwtop);

          /* determine which frame to use for lower half prediction */
          if ((pict_type==P_TYPE) && secondfield
             && (currentfield!=mv_field_sel[1][0]))
            predframe = refframe; /* same frame */
          else
            predframe = oldrefframe; /* previous frame */

          recon(predframe,mv_field_sel[1][0],newframe,0,
            coded_picture_width<<1,coded_picture_width<<1,WIDTH,8,bx,by+8,
            PMV[1][0][0],PMV[1][0][1],stwtop);
        }
      }
      else if (motion_type==MC_DMV) /* dual prime prediction */
      {
        if (secondfield)
          predframe = refframe; /* same frame */
        else
          predframe = oldrefframe; /* previous frame */

        /* calculate derived motion vectors */
        calc_DMV(DMV,dmvector,PMV[0][0][0],PMV[0][0][1]);

        /* predict from field of same parity */
        recon(oldrefframe,currentfield,newframe,0,
          coded_picture_width<<1,coded_picture_width<<1,WIDTH,16,bx,by,
          PMV[0][0][0],PMV[0][0][1],0);

        /* predict from field of opposite parity */
        recon(predframe,!currentfield,newframe,0,
          coded_picture_width<<1,coded_picture_width<<1,WIDTH,16,bx,by,
          DMV[0][0],DMV[0][1],1);
      }
      else
        /* invalid motion_type */
        printf("invalid motion_type\n");
    }
    stwtop = stwbot = 1;
  }

  if (mb_type & MB_BACKWARD)
  {
    if (pict_struct==FRAME_PICTURE)
    {
      if (motion_type==MC_FRAME)
      {
        /* frame-based prediction */
        if (stwtop<2)
          recon(refframe,0,newframe,0,
            	coded_picture_width,coded_picture_width<<1,WIDTH,8,bx,by,
            PMV[0][1][0],PMV[0][1][1],stwtop);

        if (stwbot<2)
          recon(refframe,1,newframe,1,
				coded_picture_width,coded_picture_width<<1,WIDTH,8,bx,by,
		PMV[0][1][0],PMV[0][1][1],stwbot);
      }
      else {           /* field-based prediction */
	/* top field prediction */
	if (stwtop<2){
	  recon(refframe,mv_field_sel[0][1],newframe,0,
				(coded_picture_width<<1), (coded_picture_width<<1),WIDTH,8,bx, (by>>1),
		PMV[0][1][0], (PMV[0][1][1]>>1), stwtop);
	}
	
        /* bottom field prediction */
        if (stwbot<2){
          recon(refframe, mv_field_sel[1][1], newframe, 1, (coded_picture_width<<1),
				(coded_picture_width<<1), WIDTH, 8, bx, (by>>1),
		PMV[1][1][0], (PMV[1][1][1]>>1), stwbot);
	}
      }
    }
    else {                       /* TOP_FIELD or BOTTOM_FIELD */
      /* field picture */
      if (motion_type==MC_FIELD){
        /* field-based prediction */
        recon(refframe,mv_field_sel[0][1],newframe,0,
	      coded_picture_width<<1,coded_picture_width<<1,WIDTH,16,bx,by,
	      PMV[0][1][0],PMV[0][1][1],stwtop);
      }
      else if (motion_type==MC_16X8)
      {
        recon(refframe,mv_field_sel[0][1],newframe,0,
          coded_picture_width<<1,coded_picture_width<<1,WIDTH,8,bx,by,
          PMV[0][1][0],PMV[0][1][1],stwtop);

        recon(refframe,mv_field_sel[1][1],newframe,0,
          coded_picture_width<<1,coded_picture_width<<1,WIDTH,8,bx,by+8,
          PMV[1][1][0],PMV[1][1][1],stwtop);
      }
      else
        /* invalid motion_type */
        printf("invalid motion_type\n");
    }
  } /* mb_type & MB_BACKWARD */
}

inline void rec(unsigned char *s, unsigned char *d, int lx2, int h)
{
#ifdef HAVE_MMX
	__asm__ __volatile__(
		".align 8\n"
		"1:\t"
			"movq ( %1 ),     	%%mm0\n"			/* 8 s */
		    "movq 8( %1 ),    	%%mm2\n"			/* 16 s */
		    "movq %%mm0,  		( %2 )\n"   
			"addl %3,			%1\n"
		    "movq %%mm2,  		8( %2 )\n"   
			"decl %0\n"		
		    "leal (%2, %3), %2\n"
	 	"jnz    1b"            
		: 
		: "c" (h), "r" (s), "r" (d), "r" (lx2)
	);   
#else
  for (int j=0; j<h; j++, s+=lx2, d+=lx2){
    d[0] = s[0];d[1] = s[1];d[2] = s[2];d[3] = s[3];
    d[4] = s[4];d[5] = s[5];d[6] = s[6];d[7] = s[7];
    d[8] = s[8];d[9] = s[9];d[10] = s[10];d[11] = s[11];
    d[12] = s[12];d[13] = s[13];d[14] = s[14];d[15] = s[15];
  }
#endif
}

inline void recc(unsigned char *s, unsigned char *d, int lx2, int h){
#ifdef HAVE_MMX
	__asm__ __volatile__(
		".align 8\n"
	    "1:\t"
	    	"movq ( %1 ),     %%mm0\n"
	      	"addl %3,   	  %1\n"
	      	"movq %%mm0,  	  ( %2 )\n"   
			"decl %0\n"
	      	"leal (%2, %3), %2\n"
	   	"jnz    1b"            
		: 
		: "c" (h), "r" (s), "r" (d), "r" (lx2)
	);   
#else
  for (int j=0; j<h; j++, s+=lx2, d+=lx2){
    d[0] = s[0];d[1] = s[1];d[2] = s[2];d[3] = s[3];
    d[4] = s[4];d[5] = s[5];d[6] = s[6];d[7] = s[7];
  }
#endif
}

inline void reca(unsigned char *s,unsigned char  *d, int lx2, int h){
#ifdef HAVE_MMX
	#ifdef HAVE_3Dnow
		__asm__ (
			".align	8\n"
			"1:"
		   		"movq		(%1),	%%mm0\n"      /* 8 s */
			   	"movq		(%2),	%%mm2\n"      /* 8 d */
		   		"movq		8(%1),	%%mm1\n"      /* 8 s */
		   		"movq		8(%2),	%%mm3\n"      /* 8 d */
				"pavgusb 	%%mm2, 	%%mm0\n"
				"addl		%3,		%1\n"
				"pavgusb 	%%mm3, 	%%mm1\n"

				"movq		%%mm0,	(%2)\n"
				"movq		%%mm1,	8(%2)\n"
				"addl		%3,		%2\n"
			"loop		1b\n"
			:
			: "c" (h), "r" (s), "r" (d), "r" (lx2)
		);
	#else
	__asm__ (
  		"movq	     MASK_AND, 	%%mm5\n"
  		"movq	     ADD_1, 	%%mm6\n"
		"1:\t"
  			"movq        (%1),%%mm0\n"
  			"movq        (%2),%%mm1\n"
  			"movq       8(%1),%%mm2\n"
  			"movq       8(%2),%%mm3\n"
  			"psrlw		$1,%%mm0\n"
  			"psrlw	    $1,%%mm1\n"
  			"psrlw	    $1,%%mm2\n"
  			"psrlw	    $1,%%mm3\n"
  			"pand	    %%mm5,%%mm0\n"
  			"pand	    %%mm5,%%mm1\n"
  			"pand	    %%mm5,%%mm2\n"
  			"pand	    %%mm5,%%mm3\n"
  			"paddusb    %%mm1,%%mm0\n"
  			"paddusb    %%mm3,%%mm2\n"
  			"paddusb    %%mm6,%%mm0\n"
  			"paddusb    %%mm6,%%mm2\n"
  			"movq       %%mm0,(%2)\n"
  			"addl       %3,%1\n"
  			"movq	    %%mm2, 8(%2)\n"
  			"decl       %0\n"
  			"leal		(%2, %3), %2\n"
  		"jnz        1b\n"
		:
		: "c" (h), "r" (s), "r" (d), "r" (lx2)
		);
	#endif
#else
  for (int j=0; j<h; j++, s+=lx2, d+=lx2){
    d[0] = (unsigned int)(d[0] + s[0] + 1)>>1;
    d[1] = (unsigned int)(d[1] + s[1] + 1)>>1;
    d[2] = (unsigned int)(d[2] + s[2] + 1)>>1;
    d[3] = (unsigned int)(d[3] + s[3] + 1)>>1;
    d[4] = (unsigned int)(d[4] + s[4] + 1)>>1;
    d[5] = (unsigned int)(d[5] + s[5] + 1)>>1;
    d[6] = (unsigned int)(d[6] + s[6] + 1)>>1;
    d[7] = (unsigned int)(d[7] + s[7] + 1)>>1;
    d[8] = (unsigned int)(d[8] + s[8] + 1)>>1;
    d[9] = (unsigned int)(d[9] + s[9] + 1)>>1;
    d[10] = (unsigned int)(d[10] + s[10] + 1)>>1;
    d[11] = (unsigned int)(d[11] + s[11] + 1)>>1;
    d[12] = (unsigned int)(d[12] + s[12] + 1)>>1;
    d[13] = (unsigned int)(d[13] + s[13] + 1)>>1;
    d[14] = (unsigned int)(d[14] + s[14] + 1)>>1;
    d[15] = (unsigned int)(d[15] + s[15] + 1)>>1;
  }
#endif
}

inline void recac(unsigned char *s, unsigned char *d, int lx2, int h){
#ifdef HAVE_MMX
	#ifdef HAVE_3Dnow