3 |
* XVID MPEG-4 VIDEO CODEC |
* XVID MPEG-4 VIDEO CODEC |
4 |
* - Image management functions - |
* - Image management functions - |
5 |
* |
* |
6 |
* Copyright(C) 2001-2003 Peter Ross <pross@xvid.org> |
* Copyright(C) 2001-2004 Peter Ross <pross@xvid.org> |
7 |
* |
* |
8 |
* This program is free software ; you can redistribute it and/or modify |
* This program is free software ; you can redistribute it and/or modify |
9 |
* it under the terms of the GNU General Public License as published by |
* it under the terms of the GNU General Public License as published by |
26 |
#include <stdlib.h> |
#include <stdlib.h> |
27 |
#include <string.h> /* memcpy, memset */ |
#include <string.h> /* memcpy, memset */ |
28 |
#include <math.h> |
#include <math.h> |
|
|
|
29 |
#include "../portab.h" |
#include "../portab.h" |
30 |
#include "../global.h" /* XVID_CSP_XXX's */ |
#include "../global.h" /* XVID_CSP_XXX's */ |
31 |
#include "../xvid.h" /* XVID_CSP_XXX's */ |
#include "../xvid.h" /* XVID_CSP_XXX's */ |
32 |
#include "image.h" |
#include "image.h" |
33 |
#include "colorspace.h" |
#include "colorspace.h" |
34 |
#include "interpolate8x8.h" |
#include "interpolate8x8.h" |
|
#include "reduced.h" |
|
35 |
#include "../utils/mem_align.h" |
#include "../utils/mem_align.h" |
36 |
|
#include "../motion/sad.h" |
37 |
|
|
38 |
#include "font.h" /* XXX: remove later */ |
#include "font.h" /* XXX: remove later */ |
39 |
|
|
236 |
} |
} |
237 |
} |
} |
238 |
|
|
|
/* bframe encoding requires image-based u,v interpolation */ |
|
239 |
void |
void |
240 |
image_interpolate(const IMAGE * refn, |
image_interpolate(const uint8_t * refn, |
241 |
IMAGE * refh, |
uint8_t * refh, |
242 |
IMAGE * refv, |
uint8_t * refv, |
243 |
IMAGE * refhv, |
uint8_t * refhv, |
244 |
uint32_t edged_width, |
uint32_t edged_width, |
245 |
uint32_t edged_height, |
uint32_t edged_height, |
246 |
uint32_t quarterpel, |
uint32_t quarterpel, |
248 |
{ |
{ |
249 |
const uint32_t offset = EDGE_SIZE2 * (edged_width + 1); /* we only interpolate half of the edge area */ |
const uint32_t offset = EDGE_SIZE2 * (edged_width + 1); /* we only interpolate half of the edge area */ |
250 |
const uint32_t stride_add = 7 * edged_width; |
const uint32_t stride_add = 7 * edged_width; |
|
#if 0 |
|
|
const uint32_t edged_width2 = edged_width / 2; |
|
|
const uint32_t edged_height2 = edged_height / 2; |
|
|
const uint32_t offset2 = EDGE_SIZE2 * (edged_width2 + 1); |
|
|
const uint32_t stride_add2 = 7 * edged_width2; |
|
|
#endif |
|
|
uint8_t *n_ptr, *h_ptr, *v_ptr, *hv_ptr; |
|
|
uint32_t x, y; |
|
251 |
|
|
252 |
|
uint8_t *n_ptr; |
253 |
|
uint8_t *h_ptr, *v_ptr, *hv_ptr; |
254 |
|
uint32_t x, y; |
255 |
|
|
256 |
n_ptr = refn->y; |
n_ptr = (uint8_t*)refn; |
257 |
h_ptr = refh->y; |
h_ptr = refh; |
258 |
v_ptr = refv->y; |
v_ptr = refv; |
259 |
|
|
260 |
n_ptr -= offset; |
n_ptr -= offset; |
261 |
h_ptr -= offset; |
h_ptr -= offset; |
284 |
n_ptr += stride_add; |
n_ptr += stride_add; |
285 |
} |
} |
286 |
|
|
287 |
h_ptr = refh->y + (edged_height - EDGE_SIZE - EDGE_SIZE2)*edged_width - EDGE_SIZE2; |
h_ptr = refh + (edged_height - EDGE_SIZE - EDGE_SIZE2)*edged_width - EDGE_SIZE2; |
288 |
hv_ptr = refhv->y + (edged_height - EDGE_SIZE - EDGE_SIZE2)*edged_width - EDGE_SIZE2; |
hv_ptr = refhv + (edged_height - EDGE_SIZE - EDGE_SIZE2)*edged_width - EDGE_SIZE2; |
289 |
|
|
290 |
for (y = 0; y < (edged_height - EDGE_SIZE); y = y + 8) { |
for (y = 0; y < (edged_height - EDGE_SIZE); y = y + 8) { |
291 |
hv_ptr -= stride_add; |
hv_ptr -= stride_add; |
301 |
} |
} |
302 |
} else { |
} else { |
303 |
|
|
304 |
hv_ptr = refhv->y; |
hv_ptr = refhv; |
305 |
hv_ptr -= offset; |
hv_ptr -= offset; |
306 |
|
|
307 |
for (y = 0; y < (edged_height - EDGE_SIZE); y += 8) { |
for (y = 0; y < (edged_height - EDGE_SIZE); y += 8) { |
327 |
n_ptr += stride_add; |
n_ptr += stride_add; |
328 |
} |
} |
329 |
} |
} |
|
/* |
|
|
#ifdef BFRAMES |
|
|
n_ptr = refn->u; |
|
|
h_ptr = refh->u; |
|
|
v_ptr = refv->u; |
|
|
hv_ptr = refhv->u; |
|
|
|
|
|
n_ptr -= offset2; |
|
|
h_ptr -= offset2; |
|
|
v_ptr -= offset2; |
|
|
hv_ptr -= offset2; |
|
|
|
|
|
for (y = 0; y < edged_height2; y += 8) { |
|
|
for (x = 0; x < edged_width2; x += 8) { |
|
|
interpolate8x8_halfpel_h(h_ptr, n_ptr, edged_width2, rounding); |
|
|
interpolate8x8_halfpel_v(v_ptr, n_ptr, edged_width2, rounding); |
|
|
interpolate8x8_halfpel_hv(hv_ptr, n_ptr, edged_width2, rounding); |
|
|
|
|
|
n_ptr += 8; |
|
|
h_ptr += 8; |
|
|
v_ptr += 8; |
|
|
hv_ptr += 8; |
|
|
} |
|
|
h_ptr += stride_add2; |
|
|
v_ptr += stride_add2; |
|
|
hv_ptr += stride_add2; |
|
|
n_ptr += stride_add2; |
|
|
} |
|
|
|
|
|
n_ptr = refn->v; |
|
|
h_ptr = refh->v; |
|
|
v_ptr = refv->v; |
|
|
hv_ptr = refhv->v; |
|
|
|
|
|
n_ptr -= offset2; |
|
|
h_ptr -= offset2; |
|
|
v_ptr -= offset2; |
|
|
hv_ptr -= offset2; |
|
|
|
|
|
for (y = 0; y < edged_height2; y = y + 8) { |
|
|
for (x = 0; x < edged_width2; x = x + 8) { |
|
|
interpolate8x8_halfpel_h(h_ptr, n_ptr, edged_width2, rounding); |
|
|
interpolate8x8_halfpel_v(v_ptr, n_ptr, edged_width2, rounding); |
|
|
interpolate8x8_halfpel_hv(hv_ptr, n_ptr, edged_width2, rounding); |
|
|
|
|
|
n_ptr += 8; |
|
|
h_ptr += 8; |
|
|
v_ptr += 8; |
|
|
hv_ptr += 8; |
|
|
} |
|
|
h_ptr += stride_add2; |
|
|
v_ptr += stride_add2; |
|
|
hv_ptr += stride_add2; |
|
|
n_ptr += stride_add2; |
|
|
} |
|
|
#endif |
|
|
*/ |
|
|
/* |
|
|
interpolate_halfpel_h( |
|
|
refh->y - offset, |
|
|
refn->y - offset, |
|
|
edged_width, edged_height, |
|
|
rounding); |
|
|
|
|
|
interpolate_halfpel_v( |
|
|
refv->y - offset, |
|
|
refn->y - offset, |
|
|
edged_width, edged_height, |
|
|
rounding); |
|
|
|
|
|
interpolate_halfpel_hv( |
|
|
refhv->y - offset, |
|
|
refn->y - offset, |
|
|
edged_width, edged_height, |
|
|
rounding); |
|
|
*/ |
|
|
|
|
|
/* uv-image-based compensation |
|
|
offset = EDGE_SIZE2 * (edged_width / 2 + 1); |
|
|
|
|
|
interpolate_halfpel_h( |
|
|
refh->u - offset, |
|
|
refn->u - offset, |
|
|
edged_width / 2, edged_height / 2, |
|
|
rounding); |
|
|
|
|
|
interpolate_halfpel_v( |
|
|
refv->u - offset, |
|
|
refn->u - offset, |
|
|
edged_width / 2, edged_height / 2, |
|
|
rounding); |
|
|
|
|
|
interpolate_halfpel_hv( |
|
|
refhv->u - offset, |
|
|
refn->u - offset, |
|
|
edged_width / 2, edged_height / 2, |
|
|
rounding); |
|
|
|
|
|
|
|
|
interpolate_halfpel_h( |
|
|
refh->v - offset, |
|
|
refn->v - offset, |
|
|
edged_width / 2, edged_height / 2, |
|
|
rounding); |
|
|
|
|
|
interpolate_halfpel_v( |
|
|
refv->v - offset, |
|
|
refn->v - offset, |
|
|
edged_width / 2, edged_height / 2, |
|
|
rounding); |
|
|
|
|
|
interpolate_halfpel_hv( |
|
|
refhv->v - offset, |
|
|
refn->v - offset, |
|
|
edged_width / 2, edged_height / 2, |
|
|
rounding); |
|
|
*/ |
|
330 |
} |
} |
331 |
|
|
332 |
|
|
594 |
int height, |
int height, |
595 |
uint32_t edged_width, |
uint32_t edged_width, |
596 |
uint8_t * dst[4], |
uint8_t * dst[4], |
597 |
uint32_t dst_stride[4], |
int dst_stride[4], |
598 |
int csp, |
int csp, |
599 |
int interlacing) |
int interlacing) |
600 |
{ |
{ |
775 |
uint16_t width, |
uint16_t width, |
776 |
uint16_t height) |
uint16_t height) |
777 |
{ |
{ |
778 |
int diff, x, y; |
int y, bwidth, bheight; |
779 |
long sse=0; |
long sse=0; |
780 |
|
|
781 |
for (y = 0; y < height; y++) { |
bwidth = width & (~0x07); |
782 |
|
bheight = height & (~0x07); |
783 |
|
|
784 |
|
/* Compute the 8x8 integer part */ |
785 |
|
for (y = 0; y<bheight; y += 8) { |
786 |
|
int x; |
787 |
|
|
788 |
|
/* Compute sse for the band */ |
789 |
|
for (x = 0; x<bwidth; x += 8) |
790 |
|
sse += sse8_8bit(orig + x, recon + x, stride); |
791 |
|
|
792 |
|
/* remaining pixels of the 8 pixels high band */ |
793 |
|
for (x = bwidth; x < width; x++) { |
794 |
|
int diff; |
795 |
|
diff = *(orig + 0*stride + x) - *(recon + 0*stride + x); |
796 |
|
sse += diff * diff; |
797 |
|
diff = *(orig + 1*stride + x) - *(recon + 1*stride + x); |
798 |
|
sse += diff * diff; |
799 |
|
diff = *(orig + 2*stride + x) - *(recon + 2*stride + x); |
800 |
|
sse += diff * diff; |
801 |
|
diff = *(orig + 3*stride + x) - *(recon + 3*stride + x); |
802 |
|
sse += diff * diff; |
803 |
|
diff = *(orig + 4*stride + x) - *(recon + 4*stride + x); |
804 |
|
sse += diff * diff; |
805 |
|
diff = *(orig + 5*stride + x) - *(recon + 5*stride + x); |
806 |
|
sse += diff * diff; |
807 |
|
diff = *(orig + 6*stride + x) - *(recon + 6*stride + x); |
808 |
|
sse += diff * diff; |
809 |
|
diff = *(orig + 7*stride + x) - *(recon + 7*stride + x); |
810 |
|
sse += diff * diff; |
811 |
|
} |
812 |
|
|
813 |
|
orig += 8*stride; |
814 |
|
recon += 8*stride; |
815 |
|
} |
816 |
|
|
817 |
|
/* Compute the down rectangle sse */ |
818 |
|
for (y = bheight; y < height; y++) { |
819 |
|
int x; |
820 |
for (x = 0; x < width; x++) { |
for (x = 0; x < width; x++) { |
821 |
|
int diff; |
822 |
diff = *(orig + x) - *(recon + x); |
diff = *(orig + x) - *(recon + x); |
823 |
sse += diff * diff; |
sse += diff * diff; |
824 |
} |
} |
825 |
orig += stride; |
orig += stride; |
826 |
recon += stride; |
recon += stride; |
827 |
} |
} |
828 |
return sse; |
|
829 |
|
return (sse); |
830 |
} |
} |
831 |
|
|
832 |
#if 0 |
#if 0 |
955 |
} |
} |
956 |
|
|
957 |
void |
void |
958 |
output_slice(IMAGE * cur, int std, int width, xvid_image_t* out_frm, int mbx, int mby,int mbl) { |
output_slice(IMAGE * cur, int stride, int width, xvid_image_t* out_frm, int mbx, int mby,int mbl) { |
959 |
uint8_t *dY,*dU,*dV,*sY,*sU,*sV; |
uint8_t *dY,*dU,*dV,*sY,*sU,*sV; |
960 |
int std2 = std >> 1; |
int stride2 = stride >> 1; |
961 |
int w = mbl << 4, w2,i; |
int w = mbl << 4, w2,i; |
962 |
|
|
963 |
if(w > width) |
if(w > width) |
967 |
dY = (uint8_t*)out_frm->plane[0] + (mby << 4) * out_frm->stride[0] + (mbx << 4); |
dY = (uint8_t*)out_frm->plane[0] + (mby << 4) * out_frm->stride[0] + (mbx << 4); |
968 |
dU = (uint8_t*)out_frm->plane[1] + (mby << 3) * out_frm->stride[1] + (mbx << 3); |
dU = (uint8_t*)out_frm->plane[1] + (mby << 3) * out_frm->stride[1] + (mbx << 3); |
969 |
dV = (uint8_t*)out_frm->plane[2] + (mby << 3) * out_frm->stride[2] + (mbx << 3); |
dV = (uint8_t*)out_frm->plane[2] + (mby << 3) * out_frm->stride[2] + (mbx << 3); |
970 |
sY = cur->y + (mby << 4) * std + (mbx << 4); |
sY = cur->y + (mby << 4) * stride + (mbx << 4); |
971 |
sU = cur->u + (mby << 3) * std2 + (mbx << 3); |
sU = cur->u + (mby << 3) * stride2 + (mbx << 3); |
972 |
sV = cur->v + (mby << 3) * std2 + (mbx << 3); |
sV = cur->v + (mby << 3) * stride2 + (mbx << 3); |
973 |
|
|
974 |
for(i = 0 ; i < 16 ; i++) { |
for(i = 0 ; i < 16 ; i++) { |
975 |
memcpy(dY,sY,w); |
memcpy(dY,sY,w); |
976 |
dY += out_frm->stride[0]; |
dY += out_frm->stride[0]; |
977 |
sY += std; |
sY += stride; |
978 |
} |
} |
979 |
for(i = 0 ; i < 8 ; i++) { |
for(i = 0 ; i < 8 ; i++) { |
980 |
memcpy(dU,sU,w2); |
memcpy(dU,sU,w2); |
981 |
dU += out_frm->stride[1]; |
dU += out_frm->stride[1]; |
982 |
sU += std2; |
sU += stride2; |
983 |
} |
} |
984 |
for(i = 0 ; i < 8 ; i++) { |
for(i = 0 ; i < 8 ; i++) { |
985 |
memcpy(dV,sV,w2); |
memcpy(dV,sV,w2); |
986 |
dV += out_frm->stride[2]; |
dV += out_frm->stride[2]; |
987 |
sV += std2; |
sV += stride2; |
988 |
} |
} |
989 |
} |
} |
990 |
|
|
1014 |
p += edged_width/2; |
p += edged_width/2; |
1015 |
} |
} |
1016 |
} |
} |
|
|
|
|
|
|
|
/* 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); |
|
|
} |
|
|
} |
|
|
|
|
|
|
|
|
} |
|
|
|
|