yuv2rgb_altivec.c

uclinux 下的vlc播放器源代码

  }									   \
  return srcSliceH;							   \
}


#define out_abgr(a,b,c,ptr)  vec_mstrgb32(typeof(a),((typeof (a))(0)),c,b,a,ptr)
#define out_bgra(a,b,c,ptr)  vec_mstrgb32(typeof(a),c,b,a,((typeof (a))(0)),ptr)
#define out_rgba(a,b,c,ptr)  vec_mstrgb32(typeof(a),a,b,c,((typeof (a))(0)),ptr)
#define out_argb(a,b,c,ptr)  vec_mstrgb32(typeof(a),((typeof (a))(0)),a,b,c,ptr)
#define out_rgb24(a,b,c,ptr) vec_mstrgb24(a,b,c,ptr)
#define out_bgr24(a,b,c,ptr) vec_mstrgb24(c,b,a,ptr)

DEFCSP420_CVT (yuv2_abgr32, out_abgr)
DEFCSP420_CVT (yuv2_bgra32, out_argb)
DEFCSP420_CVT (yuv2_rgba32, out_rgba)
DEFCSP420_CVT (yuv2_argb32, out_argb)
DEFCSP420_CVT (yuv2_rgb24,  out_rgb24)
DEFCSP420_CVT (yuv2_bgr24,  out_bgr24)


// uyvy|uyvy|uyvy|uyvy
// 0123 4567 89ab cdef
static
const vector unsigned char
  demux_u = (vector unsigned char)(0x10,0x00,0x10,0x00,
				   0x10,0x04,0x10,0x04,
				   0x10,0x08,0x10,0x08,
				   0x10,0x0c,0x10,0x0c),
  demux_v = (vector unsigned char)(0x10,0x02,0x10,0x02,
				   0x10,0x06,0x10,0x06,
				   0x10,0x0A,0x10,0x0A,
				   0x10,0x0E,0x10,0x0E),
  demux_y = (vector unsigned char)(0x10,0x01,0x10,0x03,
				   0x10,0x05,0x10,0x07,
				   0x10,0x09,0x10,0x0B,
				   0x10,0x0D,0x10,0x0F);

/*
  this is so I can play live CCIR raw video
*/
static int altivec_uyvy_rgb32 (SwsContext *c,
			       unsigned char **in, int *instrides,
			       int srcSliceY,	int srcSliceH,
			       unsigned char **oplanes, int *outstrides)
{
  int w = c->srcW;
  int h = srcSliceH;
  int i,j;
  vector unsigned char uyvy;
  vector signed   short Y,U,V;
  vector signed   short vx,ux,uvx;
  vector signed   short R0,G0,B0,R1,G1,B1;
  vector unsigned char  R,G,B;
  vector unsigned char *out;
  ubyte *img;

  img = in[0];
  out = (vector unsigned char *)(oplanes[0]+srcSliceY*outstrides[0]);

  for (i=0;i<h;i++) {
    for (j=0;j<w/16;j++) {
      uyvy = vec_ld (0, img);
      U = (vector signed short)
	vec_perm (uyvy, (vector unsigned char)(0), demux_u);

      V = (vector signed short)
	vec_perm (uyvy, (vector unsigned char)(0), demux_v);

      Y = (vector signed short)
	vec_perm (uyvy, (vector unsigned char)(0), demux_y);

      cvtyuvtoRGB (c, Y,U,V,&R0,&G0,&B0);

      uyvy = vec_ld (16, img);
      U = (vector signed short)
	vec_perm (uyvy, (vector unsigned char)(0), demux_u);

      V = (vector signed short)
	vec_perm (uyvy, (vector unsigned char)(0), demux_v);

      Y = (vector signed short)
	vec_perm (uyvy, (vector unsigned char)(0), demux_y);

      cvtyuvtoRGB (c, Y,U,V,&R1,&G1,&B1);

      R  = vec_packclp (R0,R1);
      G  = vec_packclp (G0,G1);
      B  = vec_packclp (B0,B1);

      //      vec_mstbgr24 (R,G,B, out);
      out_rgba (R,G,B,out);

      img += 32;
    }
  }
  return srcSliceH;
}



/* Ok currently the acceleration routine only supports
   inputs of widths a multiple of 16
   and heights a multiple 2

   So we just fall back to the C codes for this.
*/
SwsFunc yuv2rgb_init_altivec (SwsContext *c)
{
  if (!(c->flags & SWS_CPU_CAPS_ALTIVEC))    
    return NULL;

  /*
    and this seems not to matter too much I tried a bunch of 
    videos with abnormal widths and mplayer crashes else where.
    mplayer -vo x11 -rawvideo on:w=350:h=240 raw-350x240.eyuv 
    boom with X11 bad match.
    
  */
  if ((c->srcW & 0xf) != 0)    return NULL;

  switch (c->srcFormat) {
  case IMGFMT_YVU9:
  case IMGFMT_IF09:
  case IMGFMT_YV12:
  case IMGFMT_I420:
  case IMGFMT_IYUV:
  case IMGFMT_CLPL:
  case IMGFMT_Y800:
  case IMGFMT_Y8:
  case IMGFMT_NV12:
  case IMGFMT_NV21:
    if ((c->srcH & 0x1) != 0)
      return NULL;

    switch(c->dstFormat){
    case IMGFMT_RGB24:
      MSG_WARN("ALTIVEC: Color Space RGB24\n");
      return altivec_yuv2_rgb24;
    case IMGFMT_BGR24:
      MSG_WARN("ALTIVEC: Color Space BGR24\n");
      return altivec_yuv2_bgr24;
    case IMGFMT_RGB32:
      MSG_WARN("ALTIVEC: Color Space ARGB32\n");
      return altivec_yuv2_argb32;
    case IMGFMT_BGR32:
      MSG_WARN("ALTIVEC: Color Space BGRA32\n");
      //      return profile_altivec_bgra32;

      return altivec_yuv2_bgra32;
    default: return NULL;
    }
    break;

  case IMGFMT_UYVY:
    switch(c->dstFormat){
    case IMGFMT_RGB32:
      MSG_WARN("ALTIVEC: Color Space UYVY -> RGB32\n");
      return altivec_uyvy_rgb32;
    default: return NULL;
    }
    break;

  }
  return NULL;
}

void yuv2rgb_altivec_init_tables (SwsContext *c, const int inv_table[4])
{
  vector signed short CY, CRV, CBU, CGU, CGV, OY, Y0;
  int64_t crv __attribute__ ((aligned(16))) = inv_table[0];
  int64_t cbu __attribute__ ((aligned(16))) = inv_table[1];
  int64_t cgu __attribute__ ((aligned(16))) = inv_table[2];
  int64_t cgv __attribute__ ((aligned(16))) = inv_table[3];
  int64_t cy = (1<<16)-1;
  int64_t oy = 0;
  short tmp __attribute__ ((aligned(16)));

  if ((c->flags & SWS_CPU_CAPS_ALTIVEC) == 0)
    return;

  cy = (cy *c->contrast             )>>17;
  crv= (crv*c->contrast * c->saturation)>>32;
  cbu= (cbu*c->contrast * c->saturation)>>32;
  cgu= (cgu*c->contrast * c->saturation)>>32;
  cgv= (cgv*c->contrast * c->saturation)>>32;

  oy -= 256*c->brightness;

  tmp = cy;
  CY = vec_lde (0, &tmp);
  CY  = vec_splat (CY, 0);

  tmp = oy;
  OY = vec_lde (0, &tmp);
  OY  = vec_splat (OY, 0);

  tmp = crv>>3;
  CRV = vec_lde (0, &tmp);
  CRV  = vec_splat (CRV, 0);
  tmp = cbu>>3;
  CBU = vec_lde (0, &tmp);
  CBU  = vec_splat (CBU, 0);

  tmp = -(cgu>>1);
  CGU = vec_lde (0, &tmp);
  CGU  = vec_splat (CGU, 0);
  tmp = -(cgv>>1);
  CGV = vec_lde (0, &tmp);
  CGV  = vec_splat (CGV, 0);

  c->CSHIFT = (vector unsigned short)(2);
  c->CY = CY;
  c->OY = OY;
  c->CRV = CRV;
  c->CBU = CBU;
  c->CGU = CGU;
  c->CGV = CGV;

#if 0
  printf ("cy:  %hvx\n", CY);
  printf ("oy:  %hvx\n", OY);
  printf ("crv: %hvx\n", CRV);
  printf ("cbu: %hvx\n", CBU);
  printf ("cgv: %hvx\n", CGV);
  printf ("cgu: %hvx\n", CGU);
#endif

 return;
}


void
altivec_yuv2packedX (SwsContext *c,
		       int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
		       int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
		       uint8_t *dest, int dstW, int dstY)
{
  int i,j;
  short tmp __attribute__((aligned (16)));
  short *p;
  short *f;
  vector signed short X,X0,X1,Y0,U0,V0,Y1,U1,V1,U,V;
  vector signed short R0,G0,B0,R1,G1,B1;

  vector unsigned char R,G,B,pels[3];
  vector unsigned char *out,*nout;
  vector signed short   RND = (vector signed short)(1<<3);
  vector unsigned short SCL = (vector unsigned short)(4);
  unsigned long scratch[16] __attribute__ ((aligned (16)));

  vector signed short *vYCoeffsBank, *vCCoeffsBank;

  vector signed short *YCoeffs, *CCoeffs;

  vYCoeffsBank = malloc (sizeof (vector signed short)*lumFilterSize*dstW);
  vCCoeffsBank = malloc (sizeof (vector signed short)*chrFilterSize*dstW);

  for (i=0;i<lumFilterSize*dstW;i++) {
    tmp = c->vLumFilter[i];
    p = &vYCoeffsBank[i];
    for (j=0;j<8;j++)
      p[j] = tmp;
  }

  for (i=0;i<chrFilterSize*dstW;i++) {
    tmp = c->vChrFilter[i];
    p = &vCCoeffsBank[i];
    for (j=0;j<8;j++)
      p[j] = tmp;
  }

  YCoeffs = vYCoeffsBank+dstY*lumFilterSize;
  CCoeffs = vCCoeffsBank+dstY*chrFilterSize;

  out = (vector unsigned char *)dest;

  for(i=0; i<dstW; i+=16){
    Y0 = RND;
    Y1 = RND;
    /* extract 16 coeffs from lumSrc */
    for(j=0; j<lumFilterSize; j++) {
      X0 = vec_ld (0,  &lumSrc[j][i]);
      X1 = vec_ld (16, &lumSrc[j][i]);
      Y0 = vec_mradds (X0, YCoeffs[j], Y0);
      Y1 = vec_mradds (X1, YCoeffs[j], Y1);
    }

    U = RND;
    V = RND;
    /* extract 8 coeffs from U,V */
    for(j=0; j<chrFilterSize; j++) {
      X  = vec_ld (0, &chrSrc[j][i/2]);
      U  = vec_mradds (X, CCoeffs[j], U);
      X  = vec_ld (0, &chrSrc[j][i/2+2048]);
      V  = vec_mradds (X, CCoeffs[j], V);
    }

    /* scale and clip signals */
    Y0 = vec_sra (Y0, SCL);
    Y1 = vec_sra (Y1, SCL);
    U  = vec_sra (U,  SCL);
    V  = vec_sra (V,  SCL);

    Y0 = vec_clip (Y0);
    Y1 = vec_clip (Y1);
    U  = vec_clip (U);
    V  = vec_clip (V);

    /* now we have
      Y0= y0 y1 y2 y3 y4 y5 y6 y7     Y1= y8 y9 y10 y11 y12 y13 y14 y15
      U= u0 u1 u2 u3 u4 u5 u6 u7      V= v0 v1 v2 v3 v4 v5 v6 v7

      Y0= y0 y1 y2 y3 y4 y5 y6 y7    Y1= y8 y9 y10 y11 y12 y13 y14 y15
      U0= u0 u0 u1 u1 u2 u2 u3 u3    U1= u4 u4 u5 u5 u6 u6 u7 u7
      V0= v0 v0 v1 v1 v2 v2 v3 v3    V1= v4 v4 v5 v5 v6 v6 v7 v7
    */

    U0 = vec_mergeh (U,U);
    V0 = vec_mergeh (V,V);

    U1 = vec_mergel (U,U);
    V1 = vec_mergel (V,V);

    cvtyuvtoRGB (c, Y0,U0,V0,&R0,&G0,&B0);
    cvtyuvtoRGB (c, Y1,U1,V1,&R1,&G1,&B1);

    R  = vec_packclp (R0,R1);
    G  = vec_packclp (G0,G1);
    B  = vec_packclp (B0,B1);

    out_rgba (R,G,B,out);
  }

  if (i < dstW) {
    i -= 16;

    Y0 = RND;
    Y1 = RND;
    /* extract 16 coeffs from lumSrc */
    for(j=0; j<lumFilterSize; j++) {
      X0 = vec_ld (0,  &lumSrc[j][i]);
      X1 = vec_ld (16, &lumSrc[j][i]);
      Y0 = vec_mradds (X0, YCoeffs[j], Y0);
      Y1 = vec_mradds (X1, YCoeffs[j], Y1);
    }

    U = RND;
    V = RND;
    /* extract 8 coeffs from U,V */
    for(j=0; j<chrFilterSize; j++) {
      X  = vec_ld (0, &chrSrc[j][i/2]);
      U  = vec_mradds (X, CCoeffs[j], U);
      X  = vec_ld (0, &chrSrc[j][i/2+2048]);
      V  = vec_mradds (X, CCoeffs[j], V);
    }

    /* scale and clip signals */
    Y0 = vec_sra (Y0, SCL);
    Y1 = vec_sra (Y1, SCL);
    U  = vec_sra (U,  SCL);
    V  = vec_sra (V,  SCL);

    Y0 = vec_clip (Y0);
    Y1 = vec_clip (Y1);
    U  = vec_clip (U);
    V  = vec_clip (V);

    /* now we have
       Y0= y0 y1 y2 y3 y4 y5 y6 y7     Y1= y8 y9 y10 y11 y12 y13 y14 y15
       U= u0 u1 u2 u3 u4 u5 u6 u7      V= v0 v1 v2 v3 v4 v5 v6 v7

       Y0= y0 y1 y2 y3 y4 y5 y6 y7    Y1= y8 y9 y10 y11 y12 y13 y14 y15
       U0= u0 u0 u1 u1 u2 u2 u3 u3    U1= u4 u4 u5 u5 u6 u6 u7 u7
       V0= v0 v0 v1 v1 v2 v2 v3 v3    V1= v4 v4 v5 v5 v6 v6 v7 v7
    */

    U0 = vec_mergeh (U,U);
    V0 = vec_mergeh (V,V);

    U1 = vec_mergel (U,U);
    V1 = vec_mergel (V,V);

    cvtyuvtoRGB (c, Y0,U0,V0,&R0,&G0,&B0);
    cvtyuvtoRGB (c, Y1,U1,V1,&R1,&G1,&B1);

    R  = vec_packclp (R0,R1);
    G  = vec_packclp (G0,G1);
    B  = vec_packclp (B0,B1);

    nout = (vector unsigned char *)scratch;
    out_rgba (R,G,B,nout);

    memcpy (&((uint32_t*)dest)[i], scratch, (dstW-i)/4);
  }

  if (vYCoeffsBank) free (vYCoeffsBank);
  if (vCCoeffsBank) free (vCCoeffsBank);

}