swscale_altivec_template.c

video linux conference

    break;

  case 8:
    {
      for(i=0; i<dstW; i++) {
	register int srcPos = filterPos[i];

	vector unsigned char src_v0 = vec_ld(srcPos, src);
	vector unsigned char src_v1;
	if ((((int)src + srcPos)% 16) > 8) {
	  src_v1 = vec_ld(srcPos + 16, src);
	}
	vector unsigned char src_vF = vec_perm(src_v0, src_v1, vec_lvsl(srcPos, src));

	vector signed short src_v = // vec_unpackh sign-extends...
	  (vector signed short)(vec_mergeh((vector unsigned char)vzero, src_vF));
	vector signed short filter_v = vec_ld(i << 4, filter);
        // the 4 above is 3 (filterSize == 8) + 1 (sizeof(short) == 2)

	vector signed int val_v = vec_msums(src_v, filter_v, (vector signed int)vzero);
	vector signed int val_s = vec_sums(val_v, vzero);
	vec_st(val_s, 0, tempo);
	dst[i] = MIN(MAX(0, tempo[3]>>7), (1<<15)-1);
      }
    }
    break;

  case 16:
    {
      for(i=0; i<dstW; i++) {
	register int srcPos = filterPos[i];

	vector unsigned char src_v0 = vec_ld(srcPos, src);
	vector unsigned char src_v1 = vec_ld(srcPos + 16, src);
	vector unsigned char src_vF = vec_perm(src_v0, src_v1, vec_lvsl(srcPos, src));

	vector signed short src_vA = // vec_unpackh sign-extends...
	  (vector signed short)(vec_mergeh((vector unsigned char)vzero, src_vF));
	vector signed short src_vB = // vec_unpackh sign-extends...
	  (vector signed short)(vec_mergel((vector unsigned char)vzero, src_vF));

	vector signed short filter_v0 = vec_ld(i << 5, filter);
       	vector signed short filter_v1 = vec_ld((i << 5) + 16, filter);
        // the 5 above are 4 (filterSize == 16) + 1 (sizeof(short) == 2)

	vector signed int val_acc = vec_msums(src_vA, filter_v0, (vector signed int)vzero);
	vector signed int val_v = vec_msums(src_vB, filter_v1, val_acc);

	vector signed int val_s = vec_sums(val_v, vzero);

	vec_st(val_s, 0, tempo);
	dst[i] = MIN(MAX(0, tempo[3]>>7), (1<<15)-1);
      }
    }
    break;
    
  default:
    {
      for(i=0; i<dstW; i++) {
	register int j;
	register int srcPos = filterPos[i];

        vector signed int val_v = (vector signed int)vzero;
	vector signed short filter_v0R = vec_ld(i * 2 * filterSize, filter);
        vector unsigned char permF = vec_lvsl((i * 2 * filterSize), filter);

        vector unsigned char src_v0 = vec_ld(srcPos, src);
        vector unsigned char permS = vec_lvsl(srcPos, src);

        for (j = 0 ; j < filterSize - 15; j += 16) {
          vector unsigned char src_v1 = vec_ld(srcPos + j + 16, src);
          vector unsigned char src_vF = vec_perm(src_v0, src_v1, permS);
          
          vector signed short src_vA = // vec_unpackh sign-extends...
            (vector signed short)(vec_mergeh((vector unsigned char)vzero, src_vF));
          vector signed short src_vB = // vec_unpackh sign-extends...
            (vector signed short)(vec_mergel((vector unsigned char)vzero, src_vF));
          
          vector signed short filter_v1R = vec_ld((i * 2 * filterSize) + (j * 2) + 16, filter);
          vector signed short filter_v2R = vec_ld((i * 2 * filterSize) + (j * 2) + 32, filter);
          vector signed short filter_v0 = vec_perm(filter_v0R, filter_v1R, permF);
          vector signed short filter_v1 = vec_perm(filter_v1R, filter_v2R, permF);
          
          vector signed int val_acc = vec_msums(src_vA, filter_v0, val_v);
          val_v = vec_msums(src_vB, filter_v1, val_acc);

          filter_v0R = filter_v2R;
          src_v0 = src_v1;
        }

        if (j < (filterSize-7)) {
          // loading src_v0 is useless, it's already done above
          //vector unsigned char src_v0 = vec_ld(srcPos + j, src);
          vector unsigned char src_v1;
          if ((((int)src + srcPos)% 16) > 8) {
            src_v1 = vec_ld(srcPos + j + 16, src);
          }
          vector unsigned char src_vF = vec_perm(src_v0, src_v1, permS);
          
          vector signed short src_v = // vec_unpackh sign-extends...
            (vector signed short)(vec_mergeh((vector unsigned char)vzero, src_vF));
          // loading filter_v0R is useless, it's already done above
          //vector signed short filter_v0R = vec_ld((i * 2 * filterSize) + j, filter);
          vector signed short filter_v1R = vec_ld((i * 2 * filterSize) + (j * 2) + 16, filter);
          vector signed short filter_v = vec_perm(filter_v0R, filter_v1R, permF);
          
          val_v = vec_msums(src_v, filter_v, val_v);
        }

        vector signed int val_s = vec_sums(val_v, vzero);
          
        vec_st(val_s, 0, tempo);
        dst[i] = MIN(MAX(0, tempo[3]>>7), (1<<15)-1);        
      }
      
    }
  }
}

static inline int yv12toyuy2_unscaled_altivec(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
     int srcSliceH, uint8_t* dstParam[], int dstStride_a[]) {
  uint8_t *dst=dstParam[0] + dstStride_a[0]*srcSliceY;
  // yv12toyuy2( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
  uint8_t *ysrc = src[0];
  uint8_t *usrc = src[1];
  uint8_t *vsrc = src[2];
  const int width = c->srcW;
  const int height = srcSliceH;
  const int lumStride = srcStride[0];
  const int chromStride = srcStride[1];
  const int dstStride = dstStride_a[0];
  const vector unsigned char yperm = vec_lvsl(0, ysrc);
  const int vertLumPerChroma = 2;  
  register unsigned int y;

  /* this code assume:

  1) dst is 16 bytes-aligned
  2) dstStride is a multiple of 16
  3) width is a multiple of 16
  4) lum&chrom stride are multiple of 8
  */
  
  for(y=0; y<height; y++)
    {
      int i;
      for (i = 0; i < width - 31; i+= 32) {
	const unsigned int j = i >> 1;
	vector unsigned char v_yA = vec_ld(i, ysrc);
	vector unsigned char v_yB = vec_ld(i + 16, ysrc);
	vector unsigned char v_yC = vec_ld(i + 32, ysrc);
	vector unsigned char v_y1 = vec_perm(v_yA, v_yB, yperm);
	vector unsigned char v_y2 = vec_perm(v_yB, v_yC, yperm);
	vector unsigned char v_uA = vec_ld(j, usrc);
	vector unsigned char v_uB = vec_ld(j + 16, usrc);
	vector unsigned char v_u = vec_perm(v_uA, v_uB, vec_lvsl(j, usrc));
	vector unsigned char v_vA = vec_ld(j, vsrc);
	vector unsigned char v_vB = vec_ld(j + 16, vsrc);
	vector unsigned char v_v = vec_perm(v_vA, v_vB, vec_lvsl(j, vsrc));
	vector unsigned char v_uv_a = vec_mergeh(v_u, v_v);
	vector unsigned char v_uv_b = vec_mergel(v_u, v_v);
	vector unsigned char v_yuy2_0 = vec_mergeh(v_y1, v_uv_a);
	vector unsigned char v_yuy2_1 = vec_mergel(v_y1, v_uv_a);
	vector unsigned char v_yuy2_2 = vec_mergeh(v_y2, v_uv_b);
	vector unsigned char v_yuy2_3 = vec_mergel(v_y2, v_uv_b);
	vec_st(v_yuy2_0, (i << 1), dst);
	vec_st(v_yuy2_1, (i << 1) + 16, dst);
	vec_st(v_yuy2_2, (i << 1) + 32, dst);
	vec_st(v_yuy2_3, (i << 1) + 48, dst);
      }
      if (i < width) {
	const unsigned int j = i >> 1;
	vector unsigned char v_y1 = vec_ld(i, ysrc);
	vector unsigned char v_u = vec_ld(j, usrc);
	vector unsigned char v_v = vec_ld(j, vsrc);
	vector unsigned char v_uv_a = vec_mergeh(v_u, v_v);
	vector unsigned char v_yuy2_0 = vec_mergeh(v_y1, v_uv_a);
	vector unsigned char v_yuy2_1 = vec_mergel(v_y1, v_uv_a);
	vec_st(v_yuy2_0, (i << 1), dst);
	vec_st(v_yuy2_1, (i << 1) + 16, dst);
      }
      if((y&(vertLumPerChroma-1))==(vertLumPerChroma-1) )
	{
	  usrc += chromStride;
	  vsrc += chromStride;
	}
      ysrc += lumStride;
      dst += dstStride;
    }
  
  return srcSliceH;
}

static inline int yv12touyvy_unscaled_altivec(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
     int srcSliceH, uint8_t* dstParam[], int dstStride_a[]) {
  uint8_t *dst=dstParam[0] + dstStride_a[0]*srcSliceY;
  // yv12toyuy2( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
  uint8_t *ysrc = src[0];
  uint8_t *usrc = src[1];
  uint8_t *vsrc = src[2];
  const int width = c->srcW;
  const int height = srcSliceH;
  const int lumStride = srcStride[0];
  const int chromStride = srcStride[1];
  const int dstStride = dstStride_a[0];
  const int vertLumPerChroma = 2;
  const vector unsigned char yperm = vec_lvsl(0, ysrc);
  register unsigned int y;

  /* this code assume:

  1) dst is 16 bytes-aligned
  2) dstStride is a multiple of 16
  3) width is a multiple of 16
  4) lum&chrom stride are multiple of 8
  */
  
  for(y=0; y<height; y++)
    {
      int i;
      for (i = 0; i < width - 31; i+= 32) {
	const unsigned int j = i >> 1;
	vector unsigned char v_yA = vec_ld(i, ysrc);
	vector unsigned char v_yB = vec_ld(i + 16, ysrc);
	vector unsigned char v_yC = vec_ld(i + 32, ysrc);
	vector unsigned char v_y1 = vec_perm(v_yA, v_yB, yperm);
	vector unsigned char v_y2 = vec_perm(v_yB, v_yC, yperm);
	vector unsigned char v_uA = vec_ld(j, usrc);
	vector unsigned char v_uB = vec_ld(j + 16, usrc);
	vector unsigned char v_u = vec_perm(v_uA, v_uB, vec_lvsl(j, usrc));
	vector unsigned char v_vA = vec_ld(j, vsrc);
	vector unsigned char v_vB = vec_ld(j + 16, vsrc);
	vector unsigned char v_v = vec_perm(v_vA, v_vB, vec_lvsl(j, vsrc));
	vector unsigned char v_uv_a = vec_mergeh(v_u, v_v);
	vector unsigned char v_uv_b = vec_mergel(v_u, v_v);
	vector unsigned char v_uyvy_0 = vec_mergeh(v_uv_a, v_y1);
	vector unsigned char v_uyvy_1 = vec_mergel(v_uv_a, v_y1);
	vector unsigned char v_uyvy_2 = vec_mergeh(v_uv_b, v_y2);
	vector unsigned char v_uyvy_3 = vec_mergel(v_uv_b, v_y2);
	vec_st(v_uyvy_0, (i << 1), dst);
	vec_st(v_uyvy_1, (i << 1) + 16, dst);
	vec_st(v_uyvy_2, (i << 1) + 32, dst);
	vec_st(v_uyvy_3, (i << 1) + 48, dst);
      }
      if (i < width) {
	const unsigned int j = i >> 1;
	vector unsigned char v_y1 = vec_ld(i, ysrc);
	vector unsigned char v_u = vec_ld(j, usrc);
	vector unsigned char v_v = vec_ld(j, vsrc);
	vector unsigned char v_uv_a = vec_mergeh(v_u, v_v);
	vector unsigned char v_uyvy_0 = vec_mergeh(v_uv_a, v_y1);
	vector unsigned char v_uyvy_1 = vec_mergel(v_uv_a, v_y1);
	vec_st(v_uyvy_0, (i << 1), dst);
	vec_st(v_uyvy_1, (i << 1) + 16, dst);
      }
      if((y&(vertLumPerChroma-1))==(vertLumPerChroma-1) )
	{
	  usrc += chromStride;
	  vsrc += chromStride;
	}
      ysrc += lumStride;
      dst += dstStride;
    }
  return srcSliceH;
}