image.c

用ADI DSP汇编指令编写的全内插函数（包括水平、垂直、交叉内插）

			dst[0], dst_stride[0], image->y, image->u, image->v,
			edged_width, edged_width2, width, height, (csp & XVID_CSP_VFLIP),
			interlacing?yv12_to_rgbai  :yv12_to_rgba,
			interlacing?yv12_to_rgbai_c:yv12_to_rgba_c, 4);
		return 0;

	case XVID_CSP_ARGB:
		safe_packed_conv(
			dst[0], dst_stride[0], image->y, image->u, image->v,
			edged_width, edged_width2, width, height, (csp & XVID_CSP_VFLIP),
			interlacing?yv12_to_argbi  :yv12_to_argb,
			interlacing?yv12_to_argbi_c:yv12_to_argb_c, 4);
		return 0;

	case XVID_CSP_YUY2:
		safe_packed_conv(
			dst[0], dst_stride[0], image->y, image->u, image->v,
			edged_width, edged_width2, width, height, (csp & XVID_CSP_VFLIP),
			interlacing?yv12_to_yuyvi  :yv12_to_yuyv,
			interlacing?yv12_to_yuyvi_c:yv12_to_yuyv_c, 2);
		return 0;

	case XVID_CSP_YVYU:		/* u,v swapped */
		safe_packed_conv(
			dst[0], dst_stride[0], image->y, image->v, image->u,
			edged_width, edged_width2, width, height, (csp & XVID_CSP_VFLIP),
			interlacing?yv12_to_yuyvi  :yv12_to_yuyv,
			interlacing?yv12_to_yuyvi_c:yv12_to_yuyv_c, 2);
		return 0;

	case XVID_CSP_UYVY:
		safe_packed_conv(
			dst[0], dst_stride[0], image->y, image->u, image->v,
			edged_width, edged_width2, width, height, (csp & XVID_CSP_VFLIP),
			interlacing?yv12_to_uyvyi  :yv12_to_uyvy,
			interlacing?yv12_to_uyvyi_c:yv12_to_uyvy_c, 2);
		return 0;

	case XVID_CSP_I420: /* YCbCr == YUV == internal colorspace for MPEG */
		yv12_to_yv12(dst[0], dst[0] + dst_stride[0]*height, dst[0] + dst_stride[0]*height + (dst_stride[0]/2)*height2,
			dst_stride[0], dst_stride[0]/2,
			image->y, image->u, image->v, edged_width, edged_width2,
			width, height, (csp & XVID_CSP_VFLIP));
		return 0;

	case XVID_CSP_YV12:	/* YCrCb == YVU == U and V plane swapped */
		yv12_to_yv12(dst[0], dst[0] + dst_stride[0]*height, dst[0] + dst_stride[0]*height + (dst_stride[0]/2)*height2,
			dst_stride[0], dst_stride[0]/2,
			image->y, image->v, image->u, edged_width, edged_width2,
			width, height, (csp & XVID_CSP_VFLIP));
		return 0;

	case XVID_CSP_PLANAR:  /* YCbCr with arbitrary pointers and different strides for Y and UV */
		yv12_to_yv12(dst[0], dst[1], dst[2],
			dst_stride[0], dst_stride[1],	/* v: dst_stride[2] not yet supported */
			image->y, image->u, image->v, edged_width, edged_width2,
			width, height, (csp & XVID_CSP_VFLIP));
		return 0;

	case XVID_CSP_INTERNAL :
		dst[0] = image->y;
		dst[1] = image->u;
		dst[2] = image->v;
		dst_stride[0] = edged_width;
		dst_stride[1] = edged_width/2;
		dst_stride[2] = edged_width/2;
		return 0;

	case XVID_CSP_NULL:
	case XVID_CSP_SLICE:
		return 0;

	}

	return -1;
}

float
image_psnr(IMAGE * orig_image,
		   IMAGE * recon_image,
		   uint16_t stride,
		   uint16_t width,
		   uint16_t height)
{
	int32_t diff, x, y, quad = 0;
	uint8_t *orig = orig_image->y;
	uint8_t *recon = recon_image->y;
	float psnr_y;

	for (y = 0; y < height; y++) {
		for (x = 0; x < width; x++) {
			diff = *(orig + x) - *(recon + x);
			quad += diff * diff;
		}
		orig += stride;
		recon += stride;
	}

	psnr_y = (float) quad / (float) (width * height);

	if (psnr_y) 
		psnr_y = (float) (255 * 255) / psnr_y;
		psnr_y = 10 * (float) log10(psnr_y);
	} else
		psnr_y = (float) 99.99;

	return psnr_y;
}


float sse_to_PSNR(long sse, int pixels)
{
        if (sse==0)
                return 99.99F;

        return 48.131F - 10*(float)log10((float)sse/(float)(pixels));   /* log10(255*255)=4.8131 */

}

long plane_sse(uint8_t * orig,
		   uint8_t * recon,
		   uint16_t stride,
		   uint16_t width,
		   uint16_t height)
{
	int diff, x, y;
	long sse=0;

	for (y = 0; y < height; y++) {
		for (x = 0; x < width; x++) {
			diff = *(orig + x) - *(recon + x);
			sse += diff * diff;
		}
		orig += stride;
		recon += stride;
	}
	return sse;
}

#if 0

#include <stdio.h>
#include <string.h>

int image_dump_pgm(uint8_t * bmp, uint32_t width, uint32_t height, char * filename)
{
	FILE * f;
	char hdr[1024];

	f = fopen(filename, "wb");
	if ( f == NULL)
	{
		return -1;
	}
	sprintf(hdr, "P5\n#xvid\n%i %i\n255\n", width, height);
	fwrite(hdr, strlen(hdr), 1, f);
	fwrite(bmp, width, height, f);
	fclose(f);

	return 0;
}


/* dump image+edges to yuv pgm files */

int image_dump(IMAGE * image, uint32_t edged_width, uint32_t edged_height, char * path, int number)
{
	char filename[1024];

	sprintf(filename, "%s_%i_%c.pgm", path, number, 'y');
	image_dump_pgm(
		image->y - (EDGE_SIZE * edged_width + EDGE_SIZE),
		edged_width, edged_height, filename);

	sprintf(filename, "%s_%i_%c.pgm", path, number, 'u');
	image_dump_pgm(
		image->u - (EDGE_SIZE2 * edged_width / 2 + EDGE_SIZE2),
		edged_width / 2, edged_height / 2, filename);

	sprintf(filename, "%s_%i_%c.pgm", path, number, 'v');
	image_dump_pgm(
		image->v - (EDGE_SIZE2 * edged_width / 2 + EDGE_SIZE2),
		edged_width / 2, edged_height / 2, filename);

	return 0;
}
#endif



/* dump image to yuvpgm file */

#include <stdio.h>

int
image_dump_yuvpgm(const IMAGE * image,
				  const uint32_t edged_width,
				  const uint32_t width,
				  const uint32_t height,
				  char *filename)
{
	FILE *f;
	char hdr[1024];
	uint32_t i;
	uint8_t *bmp1;
	uint8_t *bmp2;


	f = fopen(filename, "wb");
	if (f == NULL) {
		return -1;
	}
	sprintf(hdr, "P5\n#xvid\n%i %i\n255\n", width, (3 * height) / 2);
	fwrite(hdr, strlen(hdr), 1, f);

	bmp1 = image->y;
	for (i = 0; i < height; i++) {
		fwrite(bmp1, width, 1, f);
		bmp1 += edged_width;
	}

	bmp1 = image->u;
	bmp2 = image->v;
	for (i = 0; i < height / 2; i++) {
		fwrite(bmp1, width / 2, 1, f);
		fwrite(bmp2, width / 2, 1, f);
		bmp1 += edged_width / 2;
		bmp2 += edged_width / 2;
	}

	fclose(f);
	return 0;
}


float
image_mad(const IMAGE * img1,
		  const IMAGE * img2,
		  uint32_t stride,
		  uint32_t width,
		  uint32_t height)
{
	const uint32_t stride2 = stride / 2;
	const uint32_t width2 = width / 2;
	const uint32_t height2 = height / 2;

	uint32_t x, y;
	uint32_t sum = 0;

	for (y = 0; y < height; y++)
		for (x = 0; x < width; x++)
			sum += abs(img1->y[x + y * stride] - img2->y[x + y * stride]);

	for (y = 0; y < height2; y++)
		for (x = 0; x < width2; x++)
			sum += abs(img1->u[x + y * stride2] - img2->u[x + y * stride2]);

	for (y = 0; y < height2; y++)
		for (x = 0; x < width2; x++)
			sum += abs(img1->v[x + y * stride2] - img2->v[x + y * stride2]);

	return (float) sum / (width * height * 3 / 2);
}

void
output_slice(IMAGE * cur, int std, int width, xvid_image_t* out_frm, int mbx, int mby,int mbl) {
  uint8_t *dY,*dU,*dV,*sY,*sU,*sV;
  int std2 = std >> 1;
  int w = mbl << 4, w2,i;

  if(w > width)
    w = width;
  w2 = w >> 1;

  dY = (uint8_t*)out_frm->plane[0] + (mby << 4) * out_frm->stride[0] + (mbx << 4);
  dU = (uint8_t*)out_frm->plane[1] + (mby << 3) * out_frm->stride[1] + (mbx << 3);
  dV = (uint8_t*)out_frm->plane[2] + (mby << 3) * out_frm->stride[2] + (mbx << 3);
  sY = cur->y + (mby << 4) * std + (mbx << 4);
  sU = cur->u + (mby << 3) * std2 + (mbx << 3);
  sV = cur->v + (mby << 3) * std2 + (mbx << 3);

  for(i = 0 ; i < 16 ; i++) {
    memcpy(dY,sY,w);
    dY += out_frm->stride[0];
    sY += std;
  }
  for(i = 0 ; i < 8 ; i++) {
    memcpy(dU,sU,w2);
    dU += out_frm->stride[1];
    sU += std2;
  }
  for(i = 0 ; i < 8 ; i++) {
    memcpy(dV,sV,w2);
    dV += out_frm->stride[2];
    sV += std2;
  }
}


void
image_clear(IMAGE * img, int width, int height, int edged_width,
					int y, int u, int v)
{
	uint8_t * p;
	int i;

	p = img->y;
	for (i = 0; i < height; i++) {
		memset(p, y, width);
		p += edged_width;
	}

	p = img->u;
	for (i = 0; i < height/2; i++) {
		memset(p, u, width/2);
		p += edged_width/2;
	}

	p = img->v;
	for (i = 0; i < height/2; i++) {
		memset(p, v, width/2);
		p += edged_width/2;
	}
}


/* reduced resolution deblocking filter
	block = block size (16=rrv, 8=full resolution)
	flags = XVID_DEC_YDEBLOCK|XVID_DEC_UVDEBLOCK
*/
void
image_deblock_rrv(IMAGE * img, int edged_width,
				const MACROBLOCK * mbs, int mb_width, int mb_height, int mb_stride,
				int block, int flags)
{
	const int edged_width2 = edged_width /2;
	const int nblocks = block / 8;	/* skals code uses 8pixel block uints */
	int i,j;

	/* luma: j,i in block units */

		for (j = 1; j < mb_height*2; j++)		/* horizontal deblocking */
		for (i = 0; i < mb_width*2; i++)
		{
			if (mbs[(j-1)/2*mb_stride + (i/2)].mode != MODE_NOT_CODED ||
				mbs[(j+0)/2*mb_stride + (i/2)].mode != MODE_NOT_CODED)
			{
				hfilter_31(img->y + (j*block - 1)*edged_width + i*block,
								  img->y + (j*block + 0)*edged_width + i*block, nblocks);
			}
		}

		for (j = 0; j < mb_height*2; j++)		/* vertical deblocking */
		for (i = 1; i < mb_width*2; i++)
		{
			if (mbs[(j/2)*mb_stride + (i-1)/2].mode != MODE_NOT_CODED ||
				mbs[(j/2)*mb_stride + (i+0)/2].mode != MODE_NOT_CODED)
			{
				vfilter_31(img->y + (j*block)*edged_width + i*block - 1,
						   img->y + (j*block)*edged_width + i*block + 0,
						   edged_width, nblocks);
			}
		}



	/* chroma */

		for (j = 1; j < mb_height; j++)		/* horizontal deblocking */
		for (i = 0; i < mb_width; i++)
		{
			if (mbs[(j-1)*mb_stride + i].mode != MODE_NOT_CODED ||
				mbs[(j+0)*mb_stride + i].mode != MODE_NOT_CODED)
			{
				hfilter_31(img->u + (j*block - 1)*edged_width2 + i*block,
						   img->u + (j*block + 0)*edged_width2 + i*block, nblocks);
				hfilter_31(img->v + (j*block - 1)*edged_width2 + i*block,
						   img->v + (j*block + 0)*edged_width2 + i*block, nblocks);
			}
		}

		for (j = 0; j < mb_height; j++)		/* vertical deblocking */
		for (i = 1; i < mb_width; i++)
		{
			if (mbs[j*mb_stride + i - 1].mode != MODE_NOT_CODED ||
				mbs[j*mb_stride + i + 0].mode != MODE_NOT_CODED)
			{
				vfilter_31(img->u + (j*block)*edged_width2 + i*block - 1,
						   img->u + (j*block)*edged_width2 + i*block + 0,
						   edged_width2, nblocks);
				vfilter_31(img->v + (j*block)*edged_width2 + i*block - 1,
						   img->v + (j*block)*edged_width2 + i*block + 0,
						   edged_width2, nblocks);
			}
		}


}