reconstruct.c

jpeg and mpeg 编解码技术源代码

#include "mpeg3video.h"
#include <stdio.h>

#ifdef HAVE_MMX

#ifdef HAVE_3Dnow
static inline void recva_mmx(unsigned char *s, unsigned char *d, int lx, int lx2, int h)
{
	__asm__(
		".align	8\n"
		"1:"
	   		"movq		(%1),	%%mm0\n"      /* 8 s */
	   		"movq		8(%1),	%%mm1\n"      /* 8 s */
	   		"movq		(%4),	%%mm2\n"      /* 8 s +lx */
	   		"movq		8(%4),	%%mm3\n"      /* 8 s +lx **/

			"pavgusb %%mm2, 	%%mm0\n"
			"addl		%3,		%1\n"
			"pavgusb %%mm3, 	%%mm1\n"

	   		"movq		(%2),	%%mm2\n"      /* 8 d */
	   		"movq		8(%2),	%%mm3\n"      /* 8 d */
			"pavgusb %%mm2, 	%%mm0\n"
			"addl		%3,		%4\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), "r" (s +lx)
	);
}

static inline void recvac_mmx(unsigned char *s, unsigned char *d, int lx,int lx2, int h)
{
	__asm__(
		".align	8\n"
		"1:"
	   		"movq		(%1),	%%mm0\n"      /* 8 s */
	   		"movq		(%4),	%%mm2\n"      /* 8 s +lx */
			"addl		%3,		%1\n"
			"pavgusb %%mm2, 	%%mm0\n"
	   		"movq		(%2),	%%mm3\n"      /* 8 d */
			"addl		%3,		%4\n"
			"pavgusb %%mm3, 	%%mm0\n"
			"movq		%%mm0,	(%2)\n"
			"addl		%3,		%2\n"
		"loop		1b\n"
      :
      : "c" (h), "r" (s), "r" (d), "r" (lx2), "r" (s +lx)
	);
}

static inline void rech_mmx(unsigned char *s, unsigned char *d, int lx2, int h)
{
	__asm__ (
		".align	8\n"
		"1:"
	   		"movq		(%1),	%%mm0\n"      /* 8 s */
	   		"movq		8(%1),	%%mm1\n"      /* 8 s */
	   		"movq		1(%1),	%%mm2\n"      /* 8 s */
	   		"movq		9(%1),	%%mm3\n"      /* 8 s */

			"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)
	);
}

static inline void rechc_mmx(unsigned char *s, unsigned char *d, int lx2, int h)
{
	__asm__ (
		".align	8\n"
		"1:"
	   		"movq		(%1),	%%mm0\n"      /* 8 s */
	   		"movq		1(%1),	%%mm2\n"      /* 8 s +1 */
			"addl		%3,		%1\n"
			"pavgusb 	%%mm2, 	%%mm0\n"
			"movq		%%mm0,	(%2)\n"
			"addl		%3,		%2\n"
		"loop		1b\n"
      :
      : "c" (h), "r" (s), "r" (d), "r" (lx2)
	);
}

static inline void recha_mmx(unsigned char *s, unsigned char *d,int lx2, int h)
{
	__asm__ (
		".align	8\n"
		"1:"
	   		"movq		(%1),	%%mm0\n"      /* 8 s */
	   		"movq		8(%1),	%%mm1\n"      /* 8 s */
	   		"movq		1(%1),	%%mm2\n"      /* 8 s */
	   		"movq		9(%1),	%%mm3\n"      /* 8 s */

			"pavgusb 	%%mm2, 	%%mm0\n"
			"addl		%3,		%1\n"
			"pavgusb 	%%mm3, 	%%mm1\n"

	   		"movq		(%2),	%%mm2\n"      /* 8 d */
	   		"movq		8(%2),	%%mm3\n"      /* 8 d */
			"pavgusb 	%%mm2, 	%%mm0\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)
	);
}

static inline void rechac_mmx(unsigned char *s,unsigned char  *d, int lx2, int h)
{
	__asm__ (
		".align	8\n"
		"1:"
	   		"movq		(%1),	%%mm0\n"      /* 8 s */
	   		"movq		1(%1),	%%mm2\n"      /* 8 s */

			"addl		%3,		%1\n"
			"pavgusb 	%%mm2, 	%%mm0\n"

	   		"movq		(%2),	%%mm1\n"      /* 8 d */
			"pavgusb 	%%mm1, 	%%mm0\n"

			"movq		%%mm0,	(%2)\n"
			"addl		%3,		%2\n"
		"loop		1b\n"
      :
      : "c" (h), "r" (s), "r" (d), "r" (lx2)
	);
}

static inline void rec4_mmx(unsigned char *s, unsigned char *d, int lx, int lx2, int h)
{
	__asm__ __volatile__(
	   	"movq		(%1),	%%mm0\n"  /* 8 s */
	   	"movq		8(%1),	%%mm1\n"  /* 8 s */
	   	"movq		1(%1),	%%mm2\n"  /* 8 s +1*/
	   	"movq		9(%1),	%%mm3\n"  /* 8 s +1*/
		".align 8\n"
		"1:"
	   		"movq		(%4),	%%mm4\n"  /* 8 s+lx */
			"pavgusb 	%%mm2, 	%%mm0\n"
	   		"movq		8(%4),	%%mm5\n"  /* 8 s+lx */
			"pavgusb 	%%mm3, 	%%mm1\n"

	   		"movq		1(%4),	%%mm6\n"  /* 8 s+lx +1*/
			"pavgusb 	%%mm4, 	%%mm0\n"
	   		"movq		9(%4),	%%mm7\n"  /* 8 s+lx +1*/
			"pavgusb 	%%mm5, 	%%mm1\n"

			"pavgusb 	%%mm6, 	%%mm0\n"
			"addl		%3,		%4\n"
			"pavgusb 	%%mm7, 	%%mm1\n"
			"movq		%%mm0,	(%2)\n"
			"movq		%%mm6,	%%mm2\n"
			"movq		%%mm7,	%%mm3\n"
			"movq		%%mm1,	8(%2)\n"
			"movq		%%mm4,	%%mm0\n"
			"movq		%%mm5,	%%mm1\n"
			"addl		%3,		%2\n"
		"loop		1b\n"
      	:
		: "c" (h), "r" (s), "r" (d), "r" (lx2), "r" (s +lx)
	);
}

static inline void rec4c_mmx(unsigned char *s, unsigned char *d, int lx, int lx2, int h)
{
	__asm__ __volatile__(
	   	"movq		(%1),	%%mm0\n"  /* 8 s */
	   	"movq		1(%1),	%%mm2\n"  /* 8 s +1*/
		".align 8\n"
		"1:"
	   		"movq		(%4),	%%mm4\n"  /* 8 s+lx */
			"pavgusb 	%%mm2, 	%%mm0\n"

	   		"movq		1(%4),	%%mm6\n"  /* 8 s+lx +1*/
			"pavgusb 	%%mm4, 	%%mm0\n"

			"addl		%3,		%4\n"
			"pavgusb 	%%mm6, 	%%mm0\n"
			"movq		%%mm0,	(%2)\n"
			"movq		%%mm6,	%%mm2\n"
			"movq		%%mm4,	%%mm0\n"
			"addl		%3,		%2\n"
		"loop		1b\n"
      	:
		: "c" (h), "r" (s), "r" (d), "r" (lx2), "r" (s +lx)
	);
}

static inline void rec4a_mmx(unsigned char *s, unsigned char *d, int lx, int lx2, int h)
{
	__asm__ __volatile__(
	   	"movq		(%1),	%%mm0\n"  /* 8 s */
	   	"movq		8(%1),	%%mm1\n"  /* 8 s */
	   	"movq		1(%1),	%%mm2\n"  /* 8 s +1*/
	   	"movq		9(%1),	%%mm3\n"  /* 8 s +1*/
		".align 8\n"
		"1:"
	   		"movq		(%4),	%%mm4\n"  /* 8 s+lx */
			"pavgusb 	%%mm2, 	%%mm0\n"
	   		"movq		8(%4),	%%mm5\n"  /* 8 s+lx */
			"pavgusb 	%%mm3, 	%%mm1\n"

	   		"movq		1(%4),	%%mm6\n"  /* 8 s+lx +1*/
			"pavgusb 	%%mm4, 	%%mm0\n"
	   		"movq		9(%4),	%%mm7\n"  /* 8 s+lx +1*/
			"pavgusb 	%%mm5, 	%%mm1\n"
			"movq		(%2),	%%mm2\n"
			"pavgusb 	%%mm6, 	%%mm0\n"
			"movq		8(%2),	%%mm3\n"

			"pavgusb 	%%mm2, 	%%mm0\n"
			"addl		%3,		%4\n"
			"pavgusb 	%%mm3, 	%%mm1\n"
			"movq		%%mm0,	(%2)\n"

			"pavgusb 	%%mm7, 	%%mm1\n"
			"movq		%%mm6,	%%mm2\n"
			"movq		%%mm7,	%%mm3\n"
			"movq		%%mm1,	8(%2)\n"
			"movq		%%mm4,	%%mm0\n"
			"movq		%%mm5,	%%mm1\n"
			"addl		%3,		%2\n"
		"loop		1b\n"
      	:
		: "c" (h), "r" (s), "r" (d), "r" (lx2), "r" (s +lx)
	);
}

static inline void rec4ac_mmx(unsigned char *s, unsigned char  *d, int lx, int lx2, int h)
{
	__asm__ __volatile__(
	   	"movq		(%1),	%%mm0\n"  /* 8 s */
	   	"movq		1(%1),	%%mm2\n"  /* 8 s +1*/
		".align 8\n"
		"1:"
	   		"movq		(%4),	%%mm4\n"  /* 8 s+lx */
			"pavgusb 	%%mm2, 	%%mm0\n"

	   		"movq		1(%4),	%%mm6\n"  /* 8 s+lx +1*/
			"pavgusb 	%%mm4, 	%%mm0\n"
			"movq		(%2),	%%mm1\n"  /* 8 d */
			"pavgusb 	%%mm6, 	%%mm0\n"
			"addl		%3,		%4\n"
			"pavgusb 	%%mm1, 	%%mm0\n"
			"movq		%%mm6,	%%mm2\n"
			"movq		%%mm0,	(%2)\n"
			"movq		%%mm4,	%%mm0\n"
			"addl		%3,		%2\n"
		"loop		1b\n"
      	:
		: "c" (h), "r" (s), "r" (d), "r" (lx2), "r" (s +lx)
	);
}

#else  // HAVE_3DNOW
	static long long ADD_1	=	0x0101010101010101LL;
	static long long MASK_AND = 0x7f7f7f7f7f7f7f7fLL;
#endif

static inline void rec_mmx(unsigned char *s, unsigned char *d, int lx2, int h)
{
	__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)
	);   
}


static inline void recc_mmx(unsigned char *s, unsigned char *d, int lx2, int h)
{
	__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)
	);   
}


static inline void reca_mmx(unsigned char *s, unsigned char  *d, int lx2, int h)
{
#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   /* No 3dnow */
	__asm__ (
  		"movq	     MASK_AND, 	%%mm5\n"
  		"movq	     ADD_1, 	%%mm6\n"
		"1:\t"
  			"movq        (%1),%%mm0\n"        /* Load 16 pixels from each row */
  			"movq        (%2),%%mm1\n"
  			"movq       8(%1),%%mm2\n"
  			"movq       8(%2),%%mm3\n"
  			"psrlw		$1,%%mm0\n"           /* Shift pixels down */
  			"psrlw	    $1,%%mm1\n"
  			"pand	    %%mm5,%%mm0\n"        /* Zero out significant bit */
  			"psrlw	    $1,%%mm2\n"
  			"pand	    %%mm5,%%mm1\n"
  			"psrlw	    $1,%%mm3\n"
  			"pand	    %%mm5,%%mm2\n"
  			"paddusb    %%mm1,%%mm0\n"        /* Add pixels */
  			"pand	    %%mm5,%%mm3\n"
  			"paddusb    %%mm3,%%mm2\n"
  			"paddusb    %%mm6,%%mm0\n"        /* Add 1 to results */
  			"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
}


static inline void recac_mmx(unsigned char *s, unsigned char *d, int lx2, int h)
{
#ifdef HAVE_3Dnow
	__asm__ (
		".align	8\n"
		"1:"
		   	"movq		(%1),	%%mm0\n"      /* 8 s */
	   		"movq		(%2),	%%mm2\n"      /* 8 d */
			"pavgusb 	%%mm2, 	%%mm0\n"
			"addl		%3,		%1\n"
			"movq		%%mm0,	(%2)\n"
			"addl		%3,		%2\n"
		"loop		1b\n"
		:
		: "c" (h), "r" (s), "r" (d), "r" (lx2)
	);
#else /* No 3dnow */
	__asm__ (
  		"movq	     MASK_AND, 	%%mm5\n"
  		"movq	     ADD_1, 	%%mm6\n"
		"1:\t"
  			"movq       (%1),%%mm0\n"
  			"movq       (%2),%%mm1\n"
  			"psrlw		$1,%%mm0\n"
  			"psrlw	    $1,%%mm1\n"
  			"pand	    %%mm5,%%mm0\n"
  			"pand	    %%mm5,%%mm1\n"
  			"paddusb    %%mm1,%%mm0\n"
  			"paddusb    %%mm6,%%mm0\n"
  			"addl       %3,%1\n"
  			"movq       %%mm0,(%2)\n"
  			"decl       %0\n"
  			"leal		(%2, %3), %2\n"
  		"jnz        1b\n"
		:
		: "c" (h), "r" (s), "r" (d), "r" (lx2)
		);
#endif
}


static inline void recv_mmx(unsigned char *s, unsigned char *d, int lx, int lx2, int h)
{
#ifdef HAVE_3Dnow
	__asm__(
		".align	8\n"
		"1:"
	   		"movq		(%1),	%%mm0\n"      /* 8 s */
	   		"movq		(%4),	%%mm2\n"      /* 8 s +lx */
	   		"movq		8(%1),	%%mm1\n"      /* 8 s */
	   		"movq		8(%4),	%%mm3\n"      /* 8 s +lx **/

			"pavgusb %%mm2, 	%%mm0\n"
			"addl		%3,		%1\n"
			"pavgusb %%mm3, 	%%mm1\n"

			"movq		%%mm0,	(%2)\n"
			"addl		%3,		%4\n"
			"movq		%%mm1,	8(%2)\n"
			"addl		%3,		%2\n"
		"loop		1b\n"
      :
      : "c" (h), "r" (s), "r" (d), "r" (lx2), "r" (s +lx)