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 |
|
#include "../utils/emms.h" |
38 |
|
|
39 |
#include "font.h" /* XXX: remove later */ |
#include "font.h" /* XXX: remove later */ |
40 |
|
|
237 |
} |
} |
238 |
} |
} |
239 |
|
|
|
/* bframe encoding requires image-based u,v interpolation */ |
|
240 |
void |
void |
241 |
image_interpolate(const IMAGE * refn, |
image_interpolate(const uint8_t * refn, |
242 |
IMAGE * refh, |
uint8_t * refh, |
243 |
IMAGE * refv, |
uint8_t * refv, |
244 |
IMAGE * refhv, |
uint8_t * refhv, |
245 |
uint32_t edged_width, |
uint32_t edged_width, |
246 |
uint32_t edged_height, |
uint32_t edged_height, |
247 |
uint32_t quarterpel, |
uint32_t quarterpel, |
249 |
{ |
{ |
250 |
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 */ |
251 |
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; |
|
252 |
|
|
253 |
|
uint8_t *n_ptr; |
254 |
|
uint8_t *h_ptr, *v_ptr, *hv_ptr; |
255 |
|
uint32_t x, y; |
256 |
|
|
257 |
n_ptr = refn->y; |
n_ptr = (uint8_t*)refn; |
258 |
h_ptr = refh->y; |
h_ptr = refh; |
259 |
v_ptr = refv->y; |
v_ptr = refv; |
260 |
|
|
261 |
n_ptr -= offset; |
n_ptr -= offset; |
262 |
h_ptr -= offset; |
h_ptr -= offset; |
285 |
n_ptr += stride_add; |
n_ptr += stride_add; |
286 |
} |
} |
287 |
|
|
288 |
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; |
289 |
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; |
290 |
|
|
291 |
for (y = 0; y < (edged_height - EDGE_SIZE); y = y + 8) { |
for (y = 0; y < (edged_height - EDGE_SIZE); y = y + 8) { |
292 |
hv_ptr -= stride_add; |
hv_ptr -= stride_add; |
302 |
} |
} |
303 |
} else { |
} else { |
304 |
|
|
305 |
hv_ptr = refhv->y; |
hv_ptr = refhv; |
306 |
hv_ptr -= offset; |
hv_ptr -= offset; |
307 |
|
|
308 |
for (y = 0; y < (edged_height - EDGE_SIZE); y += 8) { |
for (y = 0; y < (edged_height - EDGE_SIZE); y += 8) { |
328 |
n_ptr += stride_add; |
n_ptr += stride_add; |
329 |
} |
} |
330 |
} |
} |
|
/* |
|
|
#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); |
|
|
*/ |
|
331 |
} |
} |
332 |
|
|
333 |
|
|
595 |
int height, |
int height, |
596 |
uint32_t edged_width, |
uint32_t edged_width, |
597 |
uint8_t * dst[4], |
uint8_t * dst[4], |
598 |
uint32_t dst_stride[4], |
int dst_stride[4], |
599 |
int csp, |
int csp, |
600 |
int interlacing) |
int interlacing) |
601 |
{ |
{ |
776 |
uint16_t width, |
uint16_t width, |
777 |
uint16_t height) |
uint16_t height) |
778 |
{ |
{ |
779 |
int diff, x, y; |
int y, bwidth, bheight; |
780 |
long sse=0; |
long sse=0; |
781 |
|
|
782 |
for (y = 0; y < height; y++) { |
bwidth = width & (~0x07); |
783 |
|
bheight = height & (~0x07); |
784 |
|
|
785 |
|
/* Compute the 8x8 integer part */ |
786 |
|
for (y = 0; y<bheight; y += 8) { |
787 |
|
int x; |
788 |
|
|
789 |
|
/* Compute sse for the band */ |
790 |
|
for (x = 0; x<bwidth; x += 8) |
791 |
|
sse += sse8_8bit(orig + x, recon + x, stride); |
792 |
|
|
793 |
|
/* remaining pixels of the 8 pixels high band */ |
794 |
|
for (x = bwidth; x < width; x++) { |
795 |
|
int diff; |
796 |
|
diff = *(orig + 0*stride + x) - *(recon + 0*stride + x); |
797 |
|
sse += diff * diff; |
798 |
|
diff = *(orig + 1*stride + x) - *(recon + 1*stride + x); |
799 |
|
sse += diff * diff; |
800 |
|
diff = *(orig + 2*stride + x) - *(recon + 2*stride + x); |
801 |
|
sse += diff * diff; |
802 |
|
diff = *(orig + 3*stride + x) - *(recon + 3*stride + x); |
803 |
|
sse += diff * diff; |
804 |
|
diff = *(orig + 4*stride + x) - *(recon + 4*stride + x); |
805 |
|
sse += diff * diff; |
806 |
|
diff = *(orig + 5*stride + x) - *(recon + 5*stride + x); |
807 |
|
sse += diff * diff; |
808 |
|
diff = *(orig + 6*stride + x) - *(recon + 6*stride + x); |
809 |
|
sse += diff * diff; |
810 |
|
diff = *(orig + 7*stride + x) - *(recon + 7*stride + x); |
811 |
|
sse += diff * diff; |
812 |
|
} |
813 |
|
|
814 |
|
orig += 8*stride; |
815 |
|
recon += 8*stride; |
816 |
|
} |
817 |
|
|
818 |
|
/* Compute the down rectangle sse */ |
819 |
|
for (y = bheight; y < height; y++) { |
820 |
|
int x; |
821 |
for (x = 0; x < width; x++) { |
for (x = 0; x < width; x++) { |
822 |
|
int diff; |
823 |
diff = *(orig + x) - *(recon + x); |
diff = *(orig + x) - *(recon + x); |
824 |
sse += diff * diff; |
sse += diff * diff; |
825 |
} |
} |
826 |
orig += stride; |
orig += stride; |
827 |
recon += stride; |
recon += stride; |
828 |
} |
} |
829 |
return sse; |
|
830 |
|
return (sse); |
831 |
} |
} |
832 |
|
|
833 |
#if 0 |
#if 0 |
956 |
} |
} |
957 |
|
|
958 |
void |
void |
959 |
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) { |
960 |
uint8_t *dY,*dU,*dV,*sY,*sU,*sV; |
uint8_t *dY,*dU,*dV,*sY,*sU,*sV; |
961 |
int std2 = std >> 1; |
int stride2 = stride >> 1; |
962 |
int w = mbl << 4, w2,i; |
int w = mbl << 4, w2,i; |
963 |
|
|
964 |
if(w > width) |
if(w > width) |
968 |
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); |
969 |
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); |
970 |
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); |
971 |
sY = cur->y + (mby << 4) * std + (mbx << 4); |
sY = cur->y + (mby << 4) * stride + (mbx << 4); |
972 |
sU = cur->u + (mby << 3) * std2 + (mbx << 3); |
sU = cur->u + (mby << 3) * stride2 + (mbx << 3); |
973 |
sV = cur->v + (mby << 3) * std2 + (mbx << 3); |
sV = cur->v + (mby << 3) * stride2 + (mbx << 3); |
974 |
|
|
975 |
for(i = 0 ; i < 16 ; i++) { |
for(i = 0 ; i < 16 ; i++) { |
976 |
memcpy(dY,sY,w); |
memcpy(dY,sY,w); |
977 |
dY += out_frm->stride[0]; |
dY += out_frm->stride[0]; |
978 |
sY += std; |
sY += stride; |
979 |
} |
} |
980 |
for(i = 0 ; i < 8 ; i++) { |
for(i = 0 ; i < 8 ; i++) { |
981 |
memcpy(dU,sU,w2); |
memcpy(dU,sU,w2); |
982 |
dU += out_frm->stride[1]; |
dU += out_frm->stride[1]; |
983 |
sU += std2; |
sU += stride2; |
984 |
} |
} |
985 |
for(i = 0 ; i < 8 ; i++) { |
for(i = 0 ; i < 8 ; i++) { |
986 |
memcpy(dV,sV,w2); |
memcpy(dV,sV,w2); |
987 |
dV += out_frm->stride[2]; |
dV += out_frm->stride[2]; |
988 |
sV += std2; |
sV += stride2; |
989 |
} |
} |
990 |
} |
} |
991 |
|
|
1016 |
} |
} |
1017 |
} |
} |
1018 |
|
|
1019 |
|
/****************************************************************************/ |
1020 |
|
|
1021 |
/* reduced resolution deblocking filter |
static void (*deintl_core)(uint8_t *, int width, int height, const int stride) = 0; |
1022 |
block = block size (16=rrv, 8=full resolution) |
extern void xvid_deinterlace_sse(uint8_t *, int width, int height, const int stride); |
|
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; |
|
1023 |
|
|
1024 |
/* luma: j,i in block units */ |
#define CLIP_255(x) ( ((x)&~255) ? ((-(x)) >> (8*sizeof((x))-1))&0xff : (x) ) |
1025 |
|
|
1026 |
for (j = 1; j < mb_height*2; j++) /* horizontal deblocking */ |
static void deinterlace_c(uint8_t *pix, int width, int height, const int bps) |
|
for (i = 0; i < mb_width*2; i++) |
|
1027 |
{ |
{ |
1028 |
if (mbs[(j-1)/2*mb_stride + (i/2)].mode != MODE_NOT_CODED || |
pix += bps; |
1029 |
mbs[(j+0)/2*mb_stride + (i/2)].mode != MODE_NOT_CODED) |
while(width-->0) |
1030 |
{ |
{ |
1031 |
hfilter_31(img->y + (j*block - 1)*edged_width + i*block, |
int p1 = pix[-bps]; |
1032 |
img->y + (j*block + 0)*edged_width + i*block, nblocks); |
int p2 = pix[0]; |
1033 |
} |
int p0 = p2; |
1034 |
} |
int j = (height>>1) - 1; |
1035 |
|
int V; |
1036 |
for (j = 0; j < mb_height*2; j++) /* vertical deblocking */ |
unsigned char *P = pix++; |
1037 |
for (i = 1; i < mb_width*2; i++) |
while(j-->0) |
|
{ |
|
|
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) |
|
1038 |
{ |
{ |
1039 |
vfilter_31(img->y + (j*block)*edged_width + i*block - 1, |
const int p3 = P[ bps]; |
1040 |
img->y + (j*block)*edged_width + i*block + 0, |
const int p4 = P[2*bps]; |
1041 |
edged_width, nblocks); |
V = ((p1+p3+1)>>1) + ((p2 - ((p0+p4+1)>>1)) >> 2); |
1042 |
|
P[0] = CLIP_255( V ); |
1043 |
|
p0 = p2; |
1044 |
|
p1 = p3; |
1045 |
|
p2 = p4; |
1046 |
|
P += 2*bps; |
1047 |
} |
} |
1048 |
} |
V = ((p1+p1+1)>>1) + ((p2 - ((p0+p2+1)>>1)) >> 2); |
1049 |
|
P[0] = CLIP_255( V ); |
|
|
|
|
|
|
|
/* 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); |
|
1050 |
} |
} |
1051 |
} |
} |
1052 |
|
#undef CLIP_255 |
1053 |
|
|
1054 |
for (j = 0; j < mb_height; j++) /* vertical deblocking */ |
int xvid_image_deinterlace(xvid_image_t* img, int width, int height, int bottom_first) |
|
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) |
|
1055 |
{ |
{ |
1056 |
vfilter_31(img->u + (j*block)*edged_width2 + i*block - 1, |
if (height&1) |
1057 |
img->u + (j*block)*edged_width2 + i*block + 0, |
return 0; |
1058 |
edged_width2, nblocks); |
if (img->csp!=XVID_CSP_PLANAR && img->csp!=XVID_CSP_I420 && img->csp!=XVID_CSP_YV12) |
1059 |
vfilter_31(img->v + (j*block)*edged_width2 + i*block - 1, |
return 0; /* not yet supported */ |
1060 |
img->v + (j*block)*edged_width2 + i*block + 0, |
if (deintl_core==0) { |
1061 |
edged_width2, nblocks); |
const int cpu_flags = check_cpu_features(); |
1062 |
|
deintl_core = deinterlace_c; |
1063 |
|
#ifdef ARCH_IS_IA32 |
1064 |
|
if (cpu_flags & XVID_CPU_MMX) |
1065 |
|
deintl_core = xvid_deinterlace_sse; |
1066 |
|
#endif |
1067 |
} |
} |
1068 |
|
if (!bottom_first) { |
1069 |
|
deintl_core(img->plane[0], width, height, img->stride[0]); |
1070 |
|
deintl_core(img->plane[1], width>>1, height>>1, img->stride[1]); |
1071 |
|
deintl_core(img->plane[2], width>>1, height>>1, img->stride[2]); |
1072 |
|
} |
1073 |
|
else { |
1074 |
|
deintl_core((uint8_t *)img->plane[0] + ( height -1)*img->stride[0], width, height, -img->stride[0]); |
1075 |
|
deintl_core((uint8_t *)img->plane[1] + ((height>>1)-1)*img->stride[1], width>>1, height>>1, -img->stride[1]); |
1076 |
|
deintl_core((uint8_t *)img->plane[2] + ((height>>1)-1)*img->stride[2], width>>1, height>>1, -img->stride[2]); |
1077 |
} |
} |
1078 |
|
emms(); |
1079 |
|
|
1080 |
|
return 1; |
1081 |
} |
} |
1082 |
|
|