dsp_mmx.c

mediastreamer2是开源的网络传输媒体流的库

/********************************************************************
 *                                                                  *
 * THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE.   *
 * USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS     *
 * GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
 * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING.       *
 *                                                                  *
 * THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2003                *
 * by the Xiph.Org Foundation http://www.xiph.org/                  *
 *                                                                  *
 ********************************************************************

  function:
  last mod: $Id: dsp_mmx.c 11582 2006-06-16 10:57:19Z msmith $

 ********************************************************************/

#include <stdlib.h>

#include "codec_internal.h"
#include "dsp.h"

static const __attribute__ ((aligned(8),used)) ogg_int64_t V128 = 0x0080008000800080LL;

#define DSP_OP_AVG(a,b) ((((int)(a)) + ((int)(b)))/2)
#define DSP_OP_DIFF(a,b) (((int)(a)) - ((int)(b)))
#define DSP_OP_ABS_DIFF(a,b) abs((((int)(a)) - ((int)(b))))

static void sub8x8__mmx (unsigned char *FiltPtr, unsigned char *ReconPtr,
                  ogg_int16_t *DctInputPtr, ogg_uint32_t PixelsPerLine,
                  ogg_uint32_t ReconPixelsPerLine) 
{
  __asm__ __volatile__ (
    "  .balign 16                   \n\t"

    "  pxor        %%mm7, %%mm7     \n\t" 

    ".rept 8                        \n\t"
    "  movq        (%0), %%mm0      \n\t" /* mm0 = FiltPtr */
    "  movq        (%1), %%mm1      \n\t" /* mm1 = ReconPtr */
    "  movq        %%mm0, %%mm2     \n\t" /* dup to prepare for up conversion */
    "  movq        %%mm1, %%mm3     \n\t" /* dup to prepare for up conversion */
    /* convert from UINT8 to INT16 */
    "  punpcklbw   %%mm7, %%mm0     \n\t" /* mm0 = INT16(FiltPtr) */
    "  punpcklbw   %%mm7, %%mm1     \n\t" /* mm1 = INT16(ReconPtr) */
    "  punpckhbw   %%mm7, %%mm2     \n\t" /* mm2 = INT16(FiltPtr) */
    "  punpckhbw   %%mm7, %%mm3     \n\t" /* mm3 = INT16(ReconPtr) */
    /* start calculation */
    "  psubw       %%mm1, %%mm0     \n\t" /* mm0 = FiltPtr - ReconPtr */
    "  psubw       %%mm3, %%mm2     \n\t" /* mm2 = FiltPtr - ReconPtr */
    "  movq        %%mm0,  (%2)     \n\t" /* write answer out */
    "  movq        %%mm2, 8(%2)     \n\t" /* write answer out */
    /* Increment pointers */
    "  add         $16, %2           \n\t"
    "  add         %3, %0           \n\t"
    "  add         %4, %1           \n\t"
    ".endr                          \n\t"

     : "+r" (FiltPtr),
       "+r" (ReconPtr),
       "+r" (DctInputPtr)
     : "m" (PixelsPerLine),
       "m" (ReconPixelsPerLine) 
     : "memory"
  );
}

static void sub8x8_128__mmx (unsigned char *FiltPtr, ogg_int16_t *DctInputPtr,
                      ogg_uint32_t PixelsPerLine) 
{
  __asm__ __volatile__ (
    "  .balign 16                   \n\t"

    "  pxor        %%mm7, %%mm7     \n\t" 
    "  movq        %[V128], %%mm1   \n\t"

    ".rept 8                        \n\t"
    "  movq        (%0), %%mm0      \n\t" /* mm0 = FiltPtr */
    "  movq        %%mm0, %%mm2     \n\t" /* dup to prepare for up conversion */
    /* convert from UINT8 to INT16 */
    "  punpcklbw   %%mm7, %%mm0     \n\t" /* mm0 = INT16(FiltPtr) */
    "  punpckhbw   %%mm7, %%mm2     \n\t" /* mm2 = INT16(FiltPtr) */
    /* start calculation */
    "  psubw       %%mm1, %%mm0     \n\t" /* mm0 = FiltPtr - 128 */
    "  psubw       %%mm1, %%mm2     \n\t" /* mm2 = FiltPtr - 128 */
    "  movq        %%mm0,  (%1)     \n\t" /* write answer out */
    "  movq        %%mm2, 8(%1)     \n\t" /* write answer out */
    /* Increment pointers */
    "  add         $16, %1           \n\t"
    "  add         %2, %0           \n\t"
    ".endr                          \n\t"

     : "+r" (FiltPtr),
       "+r" (DctInputPtr)
     : "m" (PixelsPerLine),
       [V128] "m" (V128)
     : "memory"
  );
}

static void sub8x8avg2__mmx (unsigned char *FiltPtr, unsigned char *ReconPtr1,
                     unsigned char *ReconPtr2, ogg_int16_t *DctInputPtr,
                     ogg_uint32_t PixelsPerLine,
                     ogg_uint32_t ReconPixelsPerLine) 
{
  __asm__ __volatile__ (
    "  .balign 16                   \n\t"

    "  pxor        %%mm7, %%mm7     \n\t" 

    ".rept 8                        \n\t"
    "  movq        (%0), %%mm0      \n\t" /* mm0 = FiltPtr */
    "  movq        (%1), %%mm1      \n\t" /* mm1 = ReconPtr1 */
    "  movq        (%2), %%mm4      \n\t" /* mm1 = ReconPtr2 */
    "  movq        %%mm0, %%mm2     \n\t" /* dup to prepare for up conversion */
    "  movq        %%mm1, %%mm3     \n\t" /* dup to prepare for up conversion */
    "  movq        %%mm4, %%mm5     \n\t" /* dup to prepare for up conversion */
    /* convert from UINT8 to INT16 */
    "  punpcklbw   %%mm7, %%mm0     \n\t" /* mm0 = INT16(FiltPtr) */
    "  punpcklbw   %%mm7, %%mm1     \n\t" /* mm1 = INT16(ReconPtr1) */
    "  punpcklbw   %%mm7, %%mm4     \n\t" /* mm1 = INT16(ReconPtr2) */
    "  punpckhbw   %%mm7, %%mm2     \n\t" /* mm2 = INT16(FiltPtr) */
    "  punpckhbw   %%mm7, %%mm3     \n\t" /* mm3 = INT16(ReconPtr1) */
    "  punpckhbw   %%mm7, %%mm5     \n\t" /* mm3 = INT16(ReconPtr2) */
    /* average ReconPtr1 and ReconPtr2 */
    "  paddw       %%mm4, %%mm1     \n\t" /* mm1 = ReconPtr1 + ReconPtr2 */
    "  paddw       %%mm5, %%mm3     \n\t" /* mm3 = ReconPtr1 + ReconPtr2 */
    "  psrlw       $1, %%mm1        \n\t" /* mm1 = (ReconPtr1 + ReconPtr2) / 2 */
    "  psrlw       $1, %%mm3        \n\t" /* mm3 = (ReconPtr1 + ReconPtr2) / 2 */
    "  psubw       %%mm1, %%mm0     \n\t" /* mm0 = FiltPtr - ((ReconPtr1 + ReconPtr2) / 2) */
    "  psubw       %%mm3, %%mm2     \n\t" /* mm2 = FiltPtr - ((ReconPtr1 + ReconPtr2) / 2) */
    "  movq        %%mm0,  (%3)     \n\t" /* write answer out */
    "  movq        %%mm2, 8(%3)     \n\t" /* write answer out */
    /* Increment pointers */
    "  add         $16, %3           \n\t"
    "  add         %4, %0           \n\t"
    "  add         %5, %1           \n\t"
    "  add         %5, %2           \n\t"
    ".endr                          \n\t"

     : "+r" (FiltPtr),
       "+r" (ReconPtr1),
       "+r" (ReconPtr2),
       "+r" (DctInputPtr)
     : "m" (PixelsPerLine),
       "m" (ReconPixelsPerLine) 
     : "memory"
  );
}

static ogg_uint32_t row_sad8__mmx (unsigned char *Src1, unsigned char *Src2)
{
  ogg_uint32_t MaxSad;

  __asm__ __volatile__ (
    "  .balign 16                   \n\t"

    "  pxor        %%mm6, %%mm6     \n\t"	/* zero out mm6 for unpack */
    "  pxor        %%mm7, %%mm7     \n\t" 	/* zero out mm7 for unpack */
    "  movq        (%1), %%mm0      \n\t"	/* take 8 bytes */
    "  movq        (%2), %%mm1      \n\t"

    "  movq        %%mm0, %%mm2     \n\t"
    "  psubusb     %%mm1, %%mm0     \n\t" 	/* A - B */
    "  psubusb     %%mm2, %%mm1     \n\t"	/* B - A */
    "  por         %%mm1, %%mm0     \n\t"      	/* and or gives abs difference */

    "  movq        %%mm0, %%mm1     \n\t"

    "  punpcklbw   %%mm6, %%mm0     \n\t"       /* ; unpack low four bytes to higher precision */
    "  punpckhbw   %%mm7, %%mm1     \n\t"       /* ; unpack high four bytes to higher precision */

    "  movq        %%mm0, %%mm2     \n\t"
    "  movq        %%mm1, %%mm3     \n\t"
    "  psrlq       $32, %%mm2       \n\t"	/* fold and add */
    "  psrlq       $32, %%mm3       \n\t"
    "  paddw       %%mm2, %%mm0     \n\t"
    "  paddw       %%mm3, %%mm1     \n\t"
    "  movq        %%mm0, %%mm2     \n\t"
    "  movq        %%mm1, %%mm3     \n\t"
    "  psrlq       $16, %%mm2       \n\t"
    "  psrlq       $16, %%mm3       \n\t"
    "  paddw       %%mm2, %%mm0     \n\t"
    "  paddw       %%mm3, %%mm1     \n\t"

    "  psubusw     %%mm0, %%mm1     \n\t"
    "  paddw       %%mm0, %%mm1     \n\t" 	/* mm1 = max(mm1, mm0) */
    "  movd        %%mm1, %0        \n\t"
    "  andl        $0xffff, %0      \n\t"

     : "=m" (MaxSad),
       "+r" (Src1), 
       "+r" (Src2) 
     :
     : "memory"
  );
  return MaxSad;
}

static ogg_uint32_t col_sad8x8__mmx (unsigned char *Src1, unsigned char *Src2,
		                    ogg_uint32_t stride)
{
  ogg_uint32_t MaxSad;

  __asm__ __volatile__ (
    "  .balign 16                   \n\t"

    "  pxor        %%mm3, %%mm3     \n\t"	/* zero out mm3 for unpack */
    "  pxor        %%mm4, %%mm4     \n\t"	/* mm4 low sum */
    "  pxor        %%mm5, %%mm5     \n\t" 	/* mm5 high sum */
    "  pxor        %%mm6, %%mm6     \n\t"	/* mm6 low sum */
    "  pxor        %%mm7, %%mm7     \n\t" 	/* mm7 high sum */
    "  mov         $4, %%edi        \n\t"	/* 4 rows */
    "1:                             \n\t"
    "  movq        (%1), %%mm0      \n\t"	/* take 8 bytes */
    "  movq        (%2), %%mm1      \n\t"	/* take 8 bytes */

    "  movq        %%mm0, %%mm2     \n\t"
    "  psubusb     %%mm1, %%mm0     \n\t" 	/* A - B */
    "  psubusb     %%mm2, %%mm1     \n\t"	/* B - A */
    "  por         %%mm1, %%mm0     \n\t"      	/* and or gives abs difference */
    "  movq        %%mm0, %%mm1     \n\t"

    "  punpcklbw   %%mm3, %%mm0     \n\t"	/* unpack to higher precision for accumulation */
    "  paddw       %%mm0, %%mm4     \n\t"	/* accumulate difference... */
    "  punpckhbw   %%mm3, %%mm1     \n\t"	/* unpack high four bytes to higher precision */
    "  paddw       %%mm1, %%mm5     \n\t"	/* accumulate difference... */
    "  add         %3, %1           \n\t"	/* Inc pointer into the new data */
    "  add         %3, %2           \n\t"	/* Inc pointer into the new data */

    "  dec         %%edi            \n\t"
    "  jnz 1b                       \n\t"

    "  mov         $4, %%edi        \n\t"	/* 4 rows */
    "2:                             \n\t"
    "  movq        (%1), %%mm0      \n\t"	/* take 8 bytes */
    "  movq        (%2), %%mm1      \n\t"	/* take 8 bytes */

    "  movq        %%mm0, %%mm2     \n\t"
    "  psubusb     %%mm1, %%mm0     \n\t" 	/* A - B */
    "  psubusb     %%mm2, %%mm1     \n\t"	/* B - A */
    "  por         %%mm1, %%mm0     \n\t"      	/* and or gives abs difference */
    "  movq        %%mm0, %%mm1     \n\t"

    "  punpcklbw   %%mm3, %%mm0     \n\t"	/* unpack to higher precision for accumulation */
    "  paddw       %%mm0, %%mm6     \n\t"	/* accumulate difference... */
    "  punpckhbw   %%mm3, %%mm1     \n\t"	/* unpack high four bytes to higher precision */
    "  paddw       %%mm1, %%mm7     \n\t"	/* accumulate difference... */
    "  add         %3, %1           \n\t"	/* Inc pointer into the new data */
    "  add         %3, %2           \n\t"	/* Inc pointer into the new data */

    "  dec         %%edi            \n\t"
    "  jnz 2b                       \n\t"

    "  psubusw     %%mm6, %%mm7     \n\t"
    "  paddw       %%mm6, %%mm7     \n\t" 	/* mm7 = max(mm7, mm6) */
    "  psubusw     %%mm4, %%mm5     \n\t" 	
    "  paddw       %%mm4, %%mm5     \n\t" 	/* mm5 = max(mm5, mm4) */
    "  psubusw     %%mm5, %%mm7     \n\t" 	
    "  paddw       %%mm5, %%mm7     \n\t" 	/* mm7 = max(mm5, mm7) */
    "  movq        %%mm7, %%mm6     \n\t"
    "  psrlq       $32, %%mm6       \n\t"
    "  psubusw     %%mm6, %%mm7     \n\t" 	
    "  paddw       %%mm6, %%mm7     \n\t" 	/* mm7 = max(mm5, mm7) */
    "  movq        %%mm7, %%mm6     \n\t"
    "  psrlq       $16, %%mm6       \n\t"
    "  psubusw     %%mm6, %%mm7     \n\t" 	
    "  paddw       %%mm6, %%mm7     \n\t" 	/* mm7 = max(mm5, mm7) */
    "  movd        %%mm7, %0        \n\t"
    "  andl        $0xffff, %0      \n\t"

     : "=r" (MaxSad),
       "+r" (Src1), 
       "+r" (Src2) 
     : "r" (stride)
     : "memory", "edi"
  );

  return MaxSad;
}

static ogg_uint32_t sad8x8__mmx (unsigned char *ptr1, ogg_uint32_t stride1,
		       	    unsigned char *ptr2, ogg_uint32_t stride2)
{
  ogg_uint32_t  DiffVal;

  __asm__ __volatile__ (
    "  .balign 16                   \n\t"
    "  pxor        %%mm6, %%mm6     \n\t"	/* zero out mm6 for unpack */
    "  pxor        %%mm7, %%mm7     \n\t" 	/* mm7 contains the result */
    ".rept 8                         \n\t"
    "  movq        (%1), %%mm0      \n\t"	/* take 8 bytes */
    "  movq        (%2), %%mm1      \n\t"
    "  movq        %%mm0, %%mm2     \n\t"

    "  psubusb     %%mm1, %%mm0     \n\t" 	/* A - B */
    "  psubusb     %%mm2, %%mm1     \n\t"	/* B - A */
    "  por         %%mm1, %%mm0     \n\t"      	/* and or gives abs difference */
    "  movq        %%mm0, %%mm1     \n\t"

    "  punpcklbw   %%mm6, %%mm0     \n\t"	/* unpack to higher precision for accumulation */
    "  paddw       %%mm0, %%mm7     \n\t"	/* accumulate difference... */
    "  punpckhbw   %%mm6, %%mm1     \n\t"	/* unpack high four bytes to higher precision */
    "  add         %3, %1           \n\t"	/* Inc pointer into the new data */
    "  paddw       %%mm1, %%mm7     \n\t"	/* accumulate difference... */
    "  add         %4, %2           \n\t"	/* Inc pointer into ref data */
    ".endr                          \n\t"

    "  movq        %%mm7, %%mm0     \n\t"
    "  psrlq       $32, %%mm7       \n\t"
    "  paddw       %%mm0, %%mm7     \n\t"
    "  movq        %%mm7, %%mm0     \n\t"
    "  psrlq       $16, %%mm7       \n\t"
    "  paddw       %%mm0, %%mm7     \n\t"
    "  movd        %%mm7, %0        \n\t"
    "  andl        $0xffff, %0      \n\t"

     : "=m" (DiffVal),
       "+r" (ptr1), 
       "+r" (ptr2)