Skip to content

F
ffmpeg.wasm-core
Commit 2750b827 authored by Michael Niedermayer's avatar Michael Niedermayer
Browse files
Options

motion estimation cleanup 

replace ugly macros by always_inline functions, that way its much more readable and flexible as always_inline can simply be removed while the macros couldnt be
about 0.5 % speedup with default parameters

Originally committed as revision 3037 to svn://svn.ffmpeg.org/ffmpeg/trunk
parent c4a17148
Show whitespace changes
Inline Side-by-side
Showing with 476 additions and 669 deletions
libavcodec/avcodec.h
View file @ 2750b827
...@@ -1575,7 +1575,7 @@ typedef struct AVCodecContext { ...@@ -1575,7 +1575,7 @@ typedef struct AVCodecContext {
* - encoding: set by user * - encoding: set by user
* - decoding: set by user * - decoding: set by user
*/ */
void *me_threshold; int me_threshold;
} AVCodecContext; } AVCodecContext;
......
libavcodec/motion_est.c
View file @ 2750b827
...@@ -33,8 +33,8 @@ ...@@ -33,8 +33,8 @@
#include "dsputil.h" #include "dsputil.h"
#include "mpegvideo.h" #include "mpegvideo.h"
//#undef NDEBUG #undef NDEBUG
//#include <assert.h> #include <assert.h>
#define SQ(a) ((a)*(a)) #define SQ(a) ((a)*(a))
...@@ -46,9 +46,8 @@ ...@@ -46,9 +46,8 @@
static inline int sad_hpel_motion_search(MpegEncContext * s, static inline int sad_hpel_motion_search(MpegEncContext * s,
int *mx_ptr, int *my_ptr, int dmin, int *mx_ptr, int *my_ptr, int dmin,
int pred_x, int pred_y, uint8_t *src_data[3], int src_index, int ref_index,
uint8_t *ref_data[6], int stride, int uvstride, int size, int h);
int size, int h, uint8_t * const mv_penalty);
static inline int update_map_generation(MpegEncContext * s) static inline int update_map_generation(MpegEncContext * s)
{ {
...@@ -74,208 +73,173 @@ static int minima_cmp(const void *a, const void *b){ ...@@ -74,208 +73,173 @@ static int minima_cmp(const void *a, const void *b){
return da->height - db->height; return da->height - db->height;
} }
/* SIMPLE */ #define FLAG_QPEL 1 //must be 1
#define RENAME(a) simple_ ## a #define FLAG_CHROMA 2
#define FLAG_DIRECT 4
#define CMP(d, x, y, size)\ static inline void init_ref(MpegEncContext *s, uint8_t *src[3], uint8_t *ref[3], uint8_t *ref2[3], int x, int y, int ref_index){
d = cmp(s, src_y, (ref_y) + (x) + (y)*(stride), stride, h); MotionEstContext * const c= &s->me;
const int offset[3]= {
#define CMP_HPEL(d, dx, dy, x, y, size)\ y*c-> stride + x,
{\ ((y*c->uvstride + x)>>1),
const int dxy= (dx) + 2*(dy);\ ((y*c->uvstride + x)>>1),
hpel_put[0][dxy](s->me.scratchpad, (ref_y) + (x) + (y)*(stride), stride, h);\ };
d = cmp_sub(s, s->me.scratchpad, src_y, stride, h);\ int i;
for(i=0; i<3; i++){
c->src[0][i]= src [i] + offset[i];
c->ref[0][i]= ref [i] + offset[i];
}
if(ref_index){
for(i=0; i<3; i++){
c->ref[ref_index][i]= ref2[i] + offset[i];
}
}
} }
static int get_flags(MpegEncContext *s, int direct, int chroma){
#define CMP_QPEL(d, dx, dy, x, y, size)\ return ((s->flags&CODEC_FLAG_QPEL) ? FLAG_QPEL : 0)
{\ + (direct ? FLAG_DIRECT : 0)
const int dxy= (dx) + 4*(dy);\ + (chroma ? FLAG_CHROMA : 0);
qpel_put[0][dxy](s->me.scratchpad, (ref_y) + (x) + (y)*(stride), stride);\
d = cmp_sub(s, s->me.scratchpad, src_y, stride, h);\
} }
#include "motion_est_template.c" static inline void init_mc(MpegEncContext *s, int size, int flags){
#undef RENAME MotionEstContext * const c= &s->me;
#undef CMP
#undef CMP_HPEL
#undef CMP_QPEL
#undef INIT
/* SIMPLE CHROMA */
#define RENAME(a) simple_chroma_ ## a
#define CMP(d, x, y, size)\
d = cmp(s, src_y, (ref_y) + (x) + (y)*(stride), stride, h);\
if(chroma_cmp){\
int dxy= ((x)&1) + 2*((y)&1);\
int c= ((x)>>1) + ((y)>>1)*uvstride;\
\
chroma_hpel_put[0][dxy](s->me.scratchpad, ref_u + c, uvstride, h>>1);\
d += chroma_cmp(s, s->me.scratchpad, src_u, uvstride, h>>1);\
chroma_hpel_put[0][dxy](s->me.scratchpad, ref_v + c, uvstride, h>>1);\
d += chroma_cmp(s, s->me.scratchpad, src_v, uvstride, h>>1);\
}
#define CMP_HPEL(d, dx, dy, x, y, size)\ /*FIXME s->no_rounding b_type*/
{\ if(flags & FLAG_CHROMA){
const int dxy= (dx) + 2*(dy);\ if(s->no_rounding) c->chroma_hpel_put= &s->dsp.put_no_rnd_pixels_tab[size+1];
hpel_put[0][dxy](s->me.scratchpad, (ref_y) + (x) + (y)*(stride), stride, h);\ else c->chroma_hpel_put= &s->dsp.put_pixels_tab[size+1];
d = cmp_sub(s, s->me.scratchpad, src_y, stride, h);\ }
if(chroma_cmp_sub){\
int cxy= (dxy) | ((x)&1) | (2*((y)&1));\
int c= ((x)>>1) + ((y)>>1)*uvstride;\
chroma_hpel_put[0][cxy](s->me.scratchpad, ref_u + c, uvstride, h>>1);\
d += chroma_cmp_sub(s, s->me.scratchpad, src_u, uvstride, h>>1);\
chroma_hpel_put[0][cxy](s->me.scratchpad, ref_v + c, uvstride, h>>1);\
d += chroma_cmp_sub(s, s->me.scratchpad, src_v, uvstride, h>>1);\
}\
}
#define CMP_QPEL(d, dx, dy, x, y, size)\ if(flags & FLAG_QPEL){
{\ c->qpel_avg= &s->dsp.avg_qpel_pixels_tab[size];
const int dxy= (dx) + 4*(dy);\ if(s->no_rounding) c->qpel_put= &s->dsp.put_no_rnd_qpel_pixels_tab[size];
qpel_put[0][dxy](s->me.scratchpad, (ref_y) + (x) + (y)*(stride), stride);\ else c->qpel_put= &s->dsp.put_qpel_pixels_tab[size];
d = cmp_sub(s, s->me.scratchpad, src_y, stride, h);\ }else{
if(chroma_cmp_sub){\ c->hpel_avg= &s->dsp.avg_pixels_tab[size];
int cxy, c;\ if(s->no_rounding) c->hpel_put= &s->dsp.put_no_rnd_pixels_tab[size];
int cx= (4*(x) + (dx))/2;\ else c->hpel_put= &s->dsp.put_pixels_tab[size];
int cy= (4*(y) + (dy))/2;\ }
cx= (cx>>1)|(cx&1);\ c->temp= c->scratchpad;
cy= (cy>>1)|(cy&1);\
cxy= (cx&1) + 2*(cy&1);\
c= ((cx)>>1) + ((cy)>>1)*uvstride;\
chroma_hpel_put[0][cxy](s->me.scratchpad, ref_u + c, uvstride, h>>1);\
d += chroma_cmp_sub(s, s->me.scratchpad, src_u, uvstride, h>>1);\
chroma_hpel_put[0][cxy](s->me.scratchpad, ref_v + c, uvstride, h>>1);\
d += chroma_cmp_sub(s, s->me.scratchpad, src_v, uvstride, h>>1);\
}\
} }
#include "motion_est_template.c" static always_inline int cmp(MpegEncContext *s, const int x, const int y, const int subx, const int suby,
#undef RENAME const int size, const int h, int ref_index, int src_index,
#undef CMP me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){
#undef CMP_HPEL MotionEstContext * const c= &s->me;
#undef CMP_QPEL const int stride= c->stride;
#undef INIT const int uvstride= c->uvstride;
const int qpel= flags&FLAG_QPEL;
/* SIMPLE DIRECT HPEL */ const int chroma= flags&FLAG_CHROMA;
#define RENAME(a) simple_direct_hpel_ ## a const int dxy= subx + (suby<<(1+qpel)); //FIXME log2_subpel?
//FIXME precalc divisions stuff const int hx= subx + (x<<(1+qpel));
const int hy= suby + (y<<(1+qpel));
#define CMP_DIRECT(d, dx, dy, x, y, size, cmp_func)\ uint8_t * const * const ref= c->ref[ref_index];
if((x) >= xmin && 2*(x) + (dx) <= 2*xmax && (y) >= ymin && 2*(y) + (dy) <= 2*ymax){\ uint8_t * const * const src= c->src[src_index];
const int hx= 2*(x) + (dx);\ int d;
const int hy= 2*(y) + (dy);\ //FIXME check chroma 4mv, (no crashes ...)
if(s->mv_type==MV_TYPE_8X8){\ if(flags&FLAG_DIRECT){
int i;\ if(x >= c->xmin && hx <= c->xmax<<(qpel+1) && y >= c->ymin && hy <= c->ymax<<(qpel+1)){
for(i=0; i<4; i++){\ const int time_pp= s->pp_time;
int fx = s->me.direct_basis_mv[i][0] + hx;\ const int time_pb= s->pb_time;
int fy = s->me.direct_basis_mv[i][1] + hy;\ const int mask= 2*qpel+1;
int bx = hx ? fx - s->me.co_located_mv[i][0] : s->me.co_located_mv[i][0]*(time_pb - time_pp)/time_pp + (i &1)*16;\ if(s->mv_type==MV_TYPE_8X8){
int by = hy ? fy - s->me.co_located_mv[i][1] : s->me.co_located_mv[i][1]*(time_pb - time_pp)/time_pp + (i>>1)*16;\ int i;
int fxy= (fx&1) + 2*(fy&1);\ for(i=0; i<4; i++){
int bxy= (bx&1) + 2*(by&1);\ int fx = c->direct_basis_mv[i][0] + hx;
\ int fy = c->direct_basis_mv[i][1] + hy;
uint8_t *dst= s->me.scratchpad + 8*(i&1) + 8*stride*(i>>1);\ int bx = hx ? fx - c->co_located_mv[i][0] : c->co_located_mv[i][0]*(time_pb - time_pp)/time_pp + ((i &1)<<(qpel+4));
hpel_put[1][fxy](dst, (ref_y ) + (fx>>1) + (fy>>1)*(stride), stride, 8);\ int by = hy ? fy - c->co_located_mv[i][1] : c->co_located_mv[i][1]*(time_pb - time_pp)/time_pp + ((i>>1)<<(qpel+4));
hpel_avg[1][bxy](dst, (ref_data[3]) + (bx>>1) + (by>>1)*(stride), stride, 8);\ int fxy= (fx&mask) + ((fy&mask)<<(qpel+1));
}\ int bxy= (bx&mask) + ((by&mask)<<(qpel+1));
}else{\
int fx = s->me.direct_basis_mv[0][0] + hx;\ uint8_t *dst= c->temp + 8*(i&1) + 8*stride*(i>>1);
int fy = s->me.direct_basis_mv[0][1] + hy;\ if(qpel){
int bx = hx ? fx - s->me.co_located_mv[0][0] : (s->me.co_located_mv[0][0]*(time_pb - time_pp)/time_pp);\ c->qpel_put[1][fxy](dst, ref[0] + (fx>>2) + (fy>>2)*stride, stride);
int by = hy ? fy - s->me.co_located_mv[0][1] : (s->me.co_located_mv[0][1]*(time_pb - time_pp)/time_pp);\ c->qpel_avg[1][bxy](dst, ref[8] + (bx>>2) + (by>>2)*stride, stride);
int fxy= (fx&1) + 2*(fy&1);\ }else{
int bxy= (bx&1) + 2*(by&1);\ c->hpel_put[1][fxy](dst, ref[0] + (fx>>1) + (fy>>1)*stride, stride, 8);
\ c->hpel_avg[1][bxy](dst, ref[8] + (bx>>1) + (by>>1)*stride, stride, 8);
assert((fx>>1) + 16*s->mb_x >= -16);\ }
assert((fy>>1) + 16*s->mb_y >= -16);\ }
assert((fx>>1) + 16*s->mb_x <= s->width);\ }else{
assert((fy>>1) + 16*s->mb_y <= s->height);\ int fx = c->direct_basis_mv[0][0] + hx;
assert((bx>>1) + 16*s->mb_x >= -16);\ int fy = c->direct_basis_mv[0][1] + hy;
assert((by>>1) + 16*s->mb_y >= -16);\ int bx = hx ? fx - c->co_located_mv[0][0] : (c->co_located_mv[0][0]*(time_pb - time_pp)/time_pp);
assert((bx>>1) + 16*s->mb_x <= s->width);\ int by = hy ? fy - c->co_located_mv[0][1] : (c->co_located_mv[0][1]*(time_pb - time_pp)/time_pp);
assert((by>>1) + 16*s->mb_y <= s->height);\ int fxy= (fx&mask) + ((fy&mask)<<(qpel+1));
\ int bxy= (bx&mask) + ((by&mask)<<(qpel+1));
hpel_put[0][fxy](s->me.scratchpad, (ref_y ) + (fx>>1) + (fy>>1)*(stride), stride, 16);\
hpel_avg[0][bxy](s->me.scratchpad, (ref_data[3]) + (bx>>1) + (by>>1)*(stride), stride, 16);\ if(qpel){
}\ c->qpel_put[1][fxy](c->temp , ref[0] + (fx>>2) + (fy>>2)*stride , stride);
d = cmp_func(s, s->me.scratchpad, src_y, stride, 16);\ c->qpel_put[1][fxy](c->temp + 8 , ref[0] + (fx>>2) + (fy>>2)*stride + 8 , stride);
}else\ c->qpel_put[1][fxy](c->temp + 8*stride, ref[0] + (fx>>2) + (fy>>2)*stride + 8*stride, stride);
d= 256*256*256*32; c->qpel_put[1][fxy](c->temp + 8 + 8*stride, ref[0] + (fx>>2) + (fy>>2)*stride + 8 + 8*stride, stride);
c->qpel_avg[1][bxy](c->temp , ref[8] + (bx>>2) + (by>>2)*stride , stride);
c->qpel_avg[1][bxy](c->temp + 8 , ref[8] + (bx>>2) + (by>>2)*stride + 8 , stride);
#define CMP_HPEL(d, dx, dy, x, y, size)\ c->qpel_avg[1][bxy](c->temp + 8*stride, ref[8] + (bx>>2) + (by>>2)*stride + 8*stride, stride);
CMP_DIRECT(d, dx, dy, x, y, size, cmp_sub) c->qpel_avg[1][bxy](c->temp + 8 + 8*stride, ref[8] + (bx>>2) + (by>>2)*stride + 8 + 8*stride, stride);
}else{
#define CMP(d, x, y, size)\ assert((fx>>1) + 16*s->mb_x >= -16);
CMP_DIRECT(d, 0, 0, x, y, size, cmp) assert((fy>>1) + 16*s->mb_y >= -16);
assert((fx>>1) + 16*s->mb_x <= s->width);
assert((fy>>1) + 16*s->mb_y <= s->height);
assert((bx>>1) + 16*s->mb_x >= -16);
assert((by>>1) + 16*s->mb_y >= -16);
assert((bx>>1) + 16*s->mb_x <= s->width);
assert((by>>1) + 16*s->mb_y <= s->height);
#include "motion_est_template.c" c->hpel_put[0][fxy](c->temp, ref[0] + (fx>>1) + (fy>>1)*stride, stride, 16);
#undef RENAME c->hpel_avg[0][bxy](c->temp, ref[8] + (bx>>1) + (by>>1)*stride, stride, 16);
#undef CMP }
#undef CMP_HPEL }
#undef CMP_QPEL d = cmp_func(s, c->temp, src[0], stride, 16);
#undef INIT }else
#undef CMP_DIRECT
/* SIMPLE DIRECT QPEL */
#define RENAME(a) simple_direct_qpel_ ## a
#define CMP_DIRECT(d, dx, dy, x, y, size, cmp_func)\
if((x) >= xmin && 4*(x) + (dx) <= 4*xmax && (y) >= ymin && 4*(y) + (dy) <= 4*ymax){\
const int qx= 4*(x) + (dx);\
const int qy= 4*(y) + (dy);\
if(s->mv_type==MV_TYPE_8X8){\
int i;\
for(i=0; i<4; i++){\
int fx = s->me.direct_basis_mv[i][0] + qx;\
int fy = s->me.direct_basis_mv[i][1] + qy;\
int bx = qx ? fx - s->me.co_located_mv[i][0] : s->me.co_located_mv[i][0]*(time_pb - time_pp)/time_pp + (i &1)*16;\
int by = qy ? fy - s->me.co_located_mv[i][1] : s->me.co_located_mv[i][1]*(time_pb - time_pp)/time_pp + (i>>1)*16;\
int fxy= (fx&3) + 4*(fy&3);\
int bxy= (bx&3) + 4*(by&3);\
\
uint8_t *dst= s->me.scratchpad + 8*(i&1) + 8*stride*(i>>1);\
qpel_put[1][fxy](dst, (ref_y ) + (fx>>2) + (fy>>2)*(stride), stride);\
qpel_avg[1][bxy](dst, (ref_data[3]) + (bx>>2) + (by>>2)*(stride), stride);\
}\
}else{\
int fx = s->me.direct_basis_mv[0][0] + qx;\
int fy = s->me.direct_basis_mv[0][1] + qy;\
int bx = qx ? fx - s->me.co_located_mv[0][0] : s->me.co_located_mv[0][0]*(time_pb - time_pp)/time_pp;\
int by = qy ? fy - s->me.co_located_mv[0][1] : s->me.co_located_mv[0][1]*(time_pb - time_pp)/time_pp;\
int fxy= (fx&3) + 4*(fy&3);\
int bxy= (bx&3) + 4*(by&3);\
\
qpel_put[1][fxy](s->me.scratchpad , (ref_y ) + (fx>>2) + (fy>>2)*(stride) , stride);\
qpel_put[1][fxy](s->me.scratchpad + 8 , (ref_y ) + (fx>>2) + (fy>>2)*(stride) + 8 , stride);\
qpel_put[1][fxy](s->me.scratchpad + 8*stride, (ref_y ) + (fx>>2) + (fy>>2)*(stride) + 8*stride, stride);\
qpel_put[1][fxy](s->me.scratchpad + 8 + 8*stride, (ref_y ) + (fx>>2) + (fy>>2)*(stride) + 8 + 8*stride, stride);\
qpel_avg[1][bxy](s->me.scratchpad , (ref_data[3]) + (bx>>2) + (by>>2)*(stride) , stride);\
qpel_avg[1][bxy](s->me.scratchpad + 8 , (ref_data[3]) + (bx>>2) + (by>>2)*(stride) + 8 , stride);\
qpel_avg[1][bxy](s->me.scratchpad + 8*stride, (ref_data[3]) + (bx>>2) + (by>>2)*(stride) + 8*stride, stride);\
qpel_avg[1][bxy](s->me.scratchpad + 8 + 8*stride, (ref_data[3]) + (bx>>2) + (by>>2)*(stride) + 8 + 8*stride, stride);\
}\
d = cmp_func(s, s->me.scratchpad, src_y, stride, 16);\
}else\
d= 256*256*256*32; d= 256*256*256*32;
}else{
int uvdxy;
if(dxy){
if(qpel){
c->qpel_put[size][dxy](c->temp, ref[0] + x + y*stride, stride); //FIXME prototype (add h)
if(chroma){
int cx= hx/2;
int cy= hy/2;
cx= (cx>>1)|(cx&1);
cy= (cy>>1)|(cy&1);
uvdxy= (cx&1) + 2*(cy&1);
//FIXME x/y wrong, but mpeg4 qpel is sick anyway, we should drop as much of it as possible in favor for h264
}
}else{
c->hpel_put[size][dxy](c->temp, ref[0] + x + y*stride, stride, h);
if(chroma)
uvdxy= dxy | (x&1) | (2*(y&1));
}
d = cmp_func(s, c->temp, src[0], stride, h);
}else{
d = cmp_func(s, src[0], ref[0] + x + y*stride, stride, h);
if(chroma)
uvdxy= (x&1) + 2*(y&1);
}
if(chroma){
uint8_t * const uvtemp= c->temp + 16*stride;
c->hpel_put[size+1][uvdxy](uvtemp , ref[1] + (x>>1) + (y>>1)*uvstride, uvstride, h>>1);
c->hpel_put[size+1][uvdxy](uvtemp+8, ref[2] + (x>>1) + (y>>1)*uvstride, uvstride, h>>1);
d += chroma_cmp_func(s, uvtemp , src[1], uvstride, h>>1);
d += chroma_cmp_func(s, uvtemp+8, src[2], uvstride, h>>1);
}
}
#if 0
if(full_pel){
const int index= (((y)<<ME_MAP_SHIFT) + (x))&(ME_MAP_SIZE-1);
score_map[index]= d;
}
d += (c->mv_penalty[hx - c->pred_x] + c->mv_penalty[hy - c->pred_y])*c->penalty_factor;
#define CMP_QPEL(d, dx, dy, x, y, size)\ #endif
CMP_DIRECT(d, dx, dy, x, y, size, cmp_sub) return d;
}
#define CMP(d, x, y, size)\
CMP_DIRECT(d, 0, 0, x, y, size, cmp)
#include "motion_est_template.c" #include "motion_est_template.c"
#undef RENAME
#undef CMP
#undef CMP_HPEL
#undef CMP_QPEL
#undef INIT
#undef CMP__DIRECT
static inline int get_penalty_factor(MpegEncContext *s, int type){ static inline int get_penalty_factor(MpegEncContext *s, int type){
switch(type&0xFF){ switch(type&0xFF){
...@@ -302,48 +266,28 @@ void ff_init_me(MpegEncContext *s){ ...@@ -302,48 +266,28 @@ void ff_init_me(MpegEncContext *s){
ff_set_cmp(&s->dsp, s->dsp.me_sub_cmp, s->avctx->me_sub_cmp); ff_set_cmp(&s->dsp, s->dsp.me_sub_cmp, s->avctx->me_sub_cmp);
ff_set_cmp(&s->dsp, s->dsp.mb_cmp, s->avctx->mb_cmp); ff_set_cmp(&s->dsp, s->dsp.mb_cmp, s->avctx->mb_cmp);
s->me.flags = get_flags(s, 0, s->avctx->me_cmp &FF_CMP_CHROMA);
s->me.sub_flags= get_flags(s, 0, s->avctx->me_sub_cmp&FF_CMP_CHROMA);
s->me.mb_flags = get_flags(s, 0, s->avctx->mb_cmp &FF_CMP_CHROMA);
if(s->flags&CODEC_FLAG_QPEL){ if(s->flags&CODEC_FLAG_QPEL){
if(s->avctx->me_sub_cmp&FF_CMP_CHROMA) s->me.sub_motion_search= qpel_motion_search;
s->me.sub_motion_search= simple_chroma_qpel_motion_search;
else
s->me.sub_motion_search= simple_qpel_motion_search;
}else{ }else{
if(s->avctx->me_sub_cmp&FF_CMP_CHROMA) if(s->avctx->me_sub_cmp&FF_CMP_CHROMA)
s->me.sub_motion_search= simple_chroma_hpel_motion_search; s->me.sub_motion_search= hpel_motion_search;
else if( s->avctx->me_sub_cmp == FF_CMP_SAD else if( s->avctx->me_sub_cmp == FF_CMP_SAD
&& s->avctx-> me_cmp == FF_CMP_SAD && s->avctx-> me_cmp == FF_CMP_SAD
&& s->avctx-> mb_cmp == FF_CMP_SAD) && s->avctx-> mb_cmp == FF_CMP_SAD)
s->me.sub_motion_search= sad_hpel_motion_search; // 2050 vs. 2450 cycles s->me.sub_motion_search= sad_hpel_motion_search; // 2050 vs. 2450 cycles
else else
s->me.sub_motion_search= simple_hpel_motion_search; s->me.sub_motion_search= hpel_motion_search;
} }
if(s->linesize){
if(s->avctx->me_cmp&FF_CMP_CHROMA){ s->me.stride = s->linesize;
s->me.motion_search[0]= simple_chroma_epzs_motion_search; s->me.uvstride= s->uvlinesize;
s->me.motion_search[1]= simple_chroma_epzs_motion_search4;
s->me.motion_search[4]= simple_chroma_epzs_motion_search2;
}else{ }else{
s->me.motion_search[0]= simple_epzs_motion_search; s->me.stride = 16*s->mb_width + 32;
s->me.motion_search[1]= simple_epzs_motion_search4; s->me.uvstride= 8*s->mb_width + 16;
s->me.motion_search[4]= simple_epzs_motion_search2;
}
if(s->avctx->me_pre_cmp&FF_CMP_CHROMA){
s->me.pre_motion_search= simple_chroma_epzs_motion_search;
}else{
s->me.pre_motion_search= simple_epzs_motion_search;
}
if(s->flags&CODEC_FLAG_QPEL){
if(s->avctx->mb_cmp&FF_CMP_CHROMA)
s->me.get_mb_score= simple_chroma_qpel_get_mb_score;
else
s->me.get_mb_score= simple_qpel_get_mb_score;
}else{
if(s->avctx->mb_cmp&FF_CMP_CHROMA)
s->me.get_mb_score= simple_chroma_hpel_get_mb_score;
else
s->me.get_mb_score= simple_hpel_get_mb_score;
} }
} }
...@@ -611,18 +555,17 @@ static int phods_motion_search(MpegEncContext * s, ...@@ -611,18 +555,17 @@ static int phods_motion_search(MpegEncContext * s,
static inline int sad_hpel_motion_search(MpegEncContext * s, static inline int sad_hpel_motion_search(MpegEncContext * s,
int *mx_ptr, int *my_ptr, int dmin, int *mx_ptr, int *my_ptr, int dmin,
int pred_x, int pred_y, uint8_t *src_data[3], int src_index, int ref_index,
uint8_t *ref_data[6], int stride, int uvstride, int size, int h)
int size, int h, uint8_t * const mv_penalty)
{ {
uint32_t *score_map= s->me.score_map;
const int penalty_factor= s->me.sub_penalty_factor; const int penalty_factor= s->me.sub_penalty_factor;
int mx, my, dminh; int mx, my, dminh;
uint8_t *pix, *ptr; uint8_t *pix, *ptr;
const int xmin= s->me.xmin; int stride= s->me.stride;
const int ymin= s->me.ymin; const int flags= s->me.sub_flags;
const int xmax= s->me.xmax; LOAD_COMMON
const int ymax= s->me.ymax;
assert(flags == 0);
if(s->me.skip){ if(s->me.skip){
// printf("S"); // printf("S");
...@@ -632,11 +575,11 @@ static inline int sad_hpel_motion_search(MpegEncContext * s, ...@@ -632,11 +575,11 @@ static inline int sad_hpel_motion_search(MpegEncContext * s,
} }
// printf("N"); // printf("N");
pix = src_data[0]; pix = s->me.src[src_index][0];
mx = *mx_ptr; mx = *mx_ptr;
my = *my_ptr; my = *my_ptr;
ptr = ref_data[0] + (my * stride) + mx; ptr = s->me.ref[ref_index][0] + (my * stride) + mx;
dminh = dmin; dminh = dmin;
...@@ -765,15 +708,24 @@ static inline void get_limits(MpegEncContext *s, int x, int y) ...@@ -765,15 +708,24 @@ static inline void get_limits(MpegEncContext *s, int x, int y)
static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift) static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift)
{ {
MotionEstContext * const c= &s->me;
const int size= 1; const int size= 1;
const int h=8; const int h=8;
int block; int block;
int P[10][2]; int P[10][2];
int dmin_sum=0, mx4_sum=0, my4_sum=0; int dmin_sum=0, mx4_sum=0, my4_sum=0;
uint8_t * const mv_penalty= s->me.mv_penalty[s->f_code] + MAX_MV;
int same=1; int same=1;
const int stride= s->linesize; const int stride= s->linesize;
const int uvstride= s->uvlinesize; const int uvstride= s->uvlinesize;
uint8_t *mv_penalty= s->me.current_mv_penalty;
c->ref[1][0] = c->ref[0][0] + 8;
c->ref[2][0] = c->ref[0][0] + 8*stride;
c->ref[3][0] = c->ref[2][0] + 8;
c->src[1][0] = c->src[0][0] + 8;
c->src[2][0] = c->src[0][0] + 8*stride;
c->src[3][0] = c->src[2][0] + 8;
init_mc(s, size, c->flags);
for(block=0; block<4; block++){ for(block=0; block<4; block++){
int mx4, my4; int mx4, my4;
...@@ -782,18 +734,6 @@ static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift) ...@@ -782,18 +734,6 @@ static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift)
static const int off[4]= {2, 1, 1, -1}; static const int off[4]= {2, 1, 1, -1};
const int mot_stride = s->b8_stride; const int mot_stride = s->b8_stride;
const int mot_xy = s->block_index[block]; const int mot_xy = s->block_index[block];
const int block_x= (block&1);
const int block_y= (block>>1);
uint8_t *src_data[3]= {
s->new_picture.data[0] + 8*(2*s->mb_x + block_x) + stride *8*(2*s->mb_y + block_y), //FIXME chroma?
s->new_picture.data[1] + 4*(2*s->mb_x + block_x) + uvstride*4*(2*s->mb_y + block_y),
s->new_picture.data[2] + 4*(2*s->mb_x + block_x) + uvstride*4*(2*s->mb_y + block_y)
};
uint8_t *ref_data[3]= {
s->last_picture.data[0] + 8*(2*s->mb_x + block_x) + stride *8*(2*s->mb_y + block_y), //FIXME chroma?
s->last_picture.data[1] + 4*(2*s->mb_x + block_x) + uvstride*4*(2*s->mb_y + block_y),
s->last_picture.data[2] + 4*(2*s->mb_x + block_x) + uvstride*4*(2*s->mb_y + block_y)
};
P_LEFT[0] = s->current_picture.motion_val[0][mot_xy - 1][0]; P_LEFT[0] = s->current_picture.motion_val[0][mot_xy - 1][0];
P_LEFT[1] = s->current_picture.motion_val[0][mot_xy - 1][1]; P_LEFT[1] = s->current_picture.motion_val[0][mot_xy - 1][1];
...@@ -802,8 +742,8 @@ static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift) ...@@ -802,8 +742,8 @@ static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift)
/* special case for first line */ /* special case for first line */
if (s->first_slice_line && block<2) { if (s->first_slice_line && block<2) {
pred_x4= P_LEFT[0]; s->me.pred_x= pred_x4= P_LEFT[0];
pred_y4= P_LEFT[1]; s->me.pred_y= pred_y4= P_LEFT[1];
} else { } else {
P_TOP[0] = s->current_picture.motion_val[0][mot_xy - mot_stride ][0]; P_TOP[0] = s->current_picture.motion_val[0][mot_xy - mot_stride ][0];
P_TOP[1] = s->current_picture.motion_val[0][mot_xy - mot_stride ][1]; P_TOP[1] = s->current_picture.motion_val[0][mot_xy - mot_stride ][1];
...@@ -817,32 +757,22 @@ static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift) ...@@ -817,32 +757,22 @@ static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift)
P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]); P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]); P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
// if(s->out_format == FMT_H263){ s->me.pred_x= pred_x4 = P_MEDIAN[0];
pred_x4 = P_MEDIAN[0]; s->me.pred_y= pred_y4 = P_MEDIAN[1];
pred_y4 = P_MEDIAN[1];
#if 0
}else { /* mpeg1 at least */
pred_x4= P_LEFT[0];
pred_y4= P_LEFT[1];
}
#endif
} }
P_MV1[0]= mx; P_MV1[0]= mx;
P_MV1[1]= my; P_MV1[1]= my;
dmin4 = s->me.motion_search[1](s, &mx4, &my4, P, pred_x4, pred_y4, dmin4 = epzs_motion_search4(s, &mx4, &my4, P, block, block, s->p_mv_table, (1<<16)>>shift);
src_data, ref_data, stride, uvstride, s->p_mv_table, (1<<16)>>shift, mv_penalty);
dmin4= s->me.sub_motion_search(s, &mx4, &my4, dmin4, dmin4= s->me.sub_motion_search(s, &mx4, &my4, dmin4, block, block, size, h);
pred_x4, pred_y4, src_data, ref_data, stride, uvstride, size, h, mv_penalty);
if(s->dsp.me_sub_cmp[0] != s->dsp.mb_cmp[0] if(s->dsp.me_sub_cmp[0] != s->dsp.mb_cmp[0]){
&& s->avctx->mb_decision == FF_MB_DECISION_SIMPLE){
int dxy; int dxy;
const int offset= ((block&1) + (block>>1)*stride)*8; const int offset= ((block&1) + (block>>1)*stride)*8;
uint8_t *dest_y = s->me.scratchpad + offset; uint8_t *dest_y = s->me.scratchpad + offset;
if(s->quarter_sample){ if(s->quarter_sample){
uint8_t *ref= ref_data[0] + (mx4>>2) + (my4>>2)*stride; uint8_t *ref= c->ref[block][0] + (mx4>>2) + (my4>>2)*stride;
dxy = ((my4 & 3) << 2) | (mx4 & 3); dxy = ((my4 & 3) << 2) | (mx4 & 3);
if(s->no_rounding) if(s->no_rounding)
...@@ -850,7 +780,7 @@ static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift) ...@@ -850,7 +780,7 @@ static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift)
else else
s->dsp.put_qpel_pixels_tab [1][dxy](dest_y , ref , stride); s->dsp.put_qpel_pixels_tab [1][dxy](dest_y , ref , stride);
}else{ }else{
uint8_t *ref= ref_data[0] + (mx4>>1) + (my4>>1)*stride; uint8_t *ref= c->ref[block][0] + (mx4>>1) + (my4>>1)*stride;
dxy = ((my4 & 1) << 1) | (mx4 & 1); dxy = ((my4 & 1) << 1) | (mx4 & 1);
if(s->no_rounding) if(s->no_rounding)
...@@ -906,6 +836,9 @@ static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift) ...@@ -906,6 +836,9 @@ static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift)
dmin_sum += s->dsp.mb_cmp[1](s, s->new_picture.data[2] + s->mb_x*8 + s->mb_y*8*s->uvlinesize, s->me.scratchpad+8, s->uvlinesize, 8); dmin_sum += s->dsp.mb_cmp[1](s, s->new_picture.data[2] + s->mb_x*8 + s->mb_y*8*s->uvlinesize, s->me.scratchpad+8, s->uvlinesize, 8);
} }
s->me.pred_x= mx;
s->me.pred_y= my;
switch(s->avctx->mb_cmp&0xFF){ switch(s->avctx->mb_cmp&0xFF){
/*case FF_CMP_SSE: /*case FF_CMP_SSE:
return dmin_sum+ 32*s->qscale*s->qscale;*/ return dmin_sum+ 32*s->qscale*s->qscale;*/
...@@ -916,14 +849,15 @@ static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift) ...@@ -916,14 +849,15 @@ static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift)
} }
} }
static int interlaced_search(MpegEncContext *s, uint8_t *frame_src_data[3], uint8_t *frame_ref_data[3], static int interlaced_search(MpegEncContext *s, int ref_index,
int16_t (*mv_tables[2][2])[2], uint8_t *field_select_tables[2], int f_code, int mx, int my) int16_t (*mv_tables[2][2])[2], uint8_t *field_select_tables[2], int mx, int my)
{ {
MotionEstContext * const c= &s->me;
const int size=0; const int size=0;
const int h=8; const int h=8;
int block; int block;
int P[10][2]; int P[10][2];
uint8_t * const mv_penalty= s->me.mv_penalty[f_code] + MAX_MV; uint8_t * const mv_penalty= c->current_mv_penalty;
int same=1; int same=1;
const int stride= 2*s->linesize; const int stride= 2*s->linesize;
const int uvstride= 2*s->uvlinesize; const int uvstride= 2*s->uvlinesize;
...@@ -931,45 +865,45 @@ static int interlaced_search(MpegEncContext *s, uint8_t *frame_src_data[3], uint ...@@ -931,45 +865,45 @@ static int interlaced_search(MpegEncContext *s, uint8_t *frame_src_data[3], uint
const int mot_stride= s->mb_stride; const int mot_stride= s->mb_stride;
const int xy= s->mb_x + s->mb_y*mot_stride; const int xy= s->mb_x + s->mb_y*mot_stride;
s->me.ymin>>=1; c->ymin>>=1;
s->me.ymax>>=1; c->ymax>>=1;
c->stride<<=1;
c->uvstride<<=1;
c->ref[1+ref_index][0] = c->ref[0+ref_index][0] + s->linesize;
c->src[1][0] = c->src[0][0] + s->linesize;
if(c->flags & FLAG_CHROMA){
c->ref[1+ref_index][1] = c->ref[0+ref_index][1] + s->uvlinesize;
c->ref[1+ref_index][2] = c->ref[0+ref_index][2] + s->uvlinesize;
c->src[1][1] = c->src[0][1] + s->uvlinesize;
c->src[1][2] = c->src[0][2] + s->uvlinesize;
}
init_mc(s, size, c->flags);
for(block=0; block<2; block++){ for(block=0; block<2; block++){
int field_select; int field_select;
int best_dmin= INT_MAX; int best_dmin= INT_MAX;
int best_field= -1; int best_field= -1;
uint8_t *src_data[3]= {
frame_src_data[0] + s-> linesize*block,
frame_src_data[1] + s->uvlinesize*block,
frame_src_data[2] + s->uvlinesize*block
};
for(field_select=0; field_select<2; field_select++){ for(field_select=0; field_select<2; field_select++){
int dmin, mx_i, my_i, pred_x, pred_y; int dmin, mx_i, my_i;
uint8_t *ref_data[3]= {
frame_ref_data[0] + s-> linesize*field_select,
frame_ref_data[1] + s->uvlinesize*field_select,
frame_ref_data[2] + s->uvlinesize*field_select
};
int16_t (*mv_table)[2]= mv_tables[block][field_select]; int16_t (*mv_table)[2]= mv_tables[block][field_select];
P_LEFT[0] = mv_table[xy - 1][0]; P_LEFT[0] = mv_table[xy - 1][0];
P_LEFT[1] = mv_table[xy - 1][1]; P_LEFT[1] = mv_table[xy - 1][1];
if(P_LEFT[0] > (s->me.xmax<<1)) P_LEFT[0] = (s->me.xmax<<1); if(P_LEFT[0] > (c->xmax<<1)) P_LEFT[0] = (c->xmax<<1);
pred_x= P_LEFT[0]; s->me.pred_x= P_LEFT[0];
pred_y= P_LEFT[1]; s->me.pred_y= P_LEFT[1];
if(!s->first_slice_line){ if(!s->first_slice_line){
P_TOP[0] = mv_table[xy - mot_stride][0]; P_TOP[0] = mv_table[xy - mot_stride][0];
P_TOP[1] = mv_table[xy - mot_stride][1]; P_TOP[1] = mv_table[xy - mot_stride][1];
P_TOPRIGHT[0] = mv_table[xy - mot_stride + 1][0]; P_TOPRIGHT[0] = mv_table[xy - mot_stride + 1][0];
P_TOPRIGHT[1] = mv_table[xy - mot_stride + 1][1]; P_TOPRIGHT[1] = mv_table[xy - mot_stride + 1][1];
if(P_TOP[1] > (s->me.ymax<<1)) P_TOP[1] = (s->me.ymax<<1); if(P_TOP[1] > (c->ymax<<1)) P_TOP[1] = (c->ymax<<1);
if(P_TOPRIGHT[0] < (s->me.xmin<<1)) P_TOPRIGHT[0]= (s->me.xmin<<1); if(P_TOPRIGHT[0] < (c->xmin<<1)) P_TOPRIGHT[0]= (c->xmin<<1);
if(P_TOPRIGHT[0] > (s->me.xmax<<1)) P_TOPRIGHT[0]= (s->me.xmax<<1); if(P_TOPRIGHT[0] > (c->xmax<<1)) P_TOPRIGHT[0]= (c->xmax<<1);
if(P_TOPRIGHT[1] > (s->me.ymax<<1)) P_TOPRIGHT[1]= (s->me.ymax<<1); if(P_TOPRIGHT[1] > (c->ymax<<1)) P_TOPRIGHT[1]= (c->ymax<<1);
P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]); P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]); P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
...@@ -977,32 +911,29 @@ static int interlaced_search(MpegEncContext *s, uint8_t *frame_src_data[3], uint ...@@ -977,32 +911,29 @@ static int interlaced_search(MpegEncContext *s, uint8_t *frame_src_data[3], uint
P_MV1[0]= mx; //FIXME not correct if block != field_select P_MV1[0]= mx; //FIXME not correct if block != field_select
P_MV1[1]= my / 2; P_MV1[1]= my / 2;
dmin = s->me.motion_search[4](s, &mx_i, &my_i, P, pred_x, pred_y, dmin = epzs_motion_search2(s, &mx_i, &my_i, P, block, field_select+ref_index, mv_table, (1<<16)>>1);
src_data, ref_data, stride, uvstride, mv_table, (1<<16)>>1, mv_penalty);
dmin= s->me.sub_motion_search(s, &mx_i, &my_i, dmin, dmin= c->sub_motion_search(s, &mx_i, &my_i, dmin, block, field_select+ref_index, size, h);
pred_x, pred_y, src_data, ref_data, stride, uvstride, size, h, mv_penalty);
mv_table[xy][0]= mx_i; mv_table[xy][0]= mx_i;
mv_table[xy][1]= my_i; mv_table[xy][1]= my_i;
if(s->dsp.me_sub_cmp[0] != s->dsp.mb_cmp[0] if(s->dsp.me_sub_cmp[0] != s->dsp.mb_cmp[0]){
&& s->avctx->mb_decision == FF_MB_DECISION_SIMPLE){
int dxy; int dxy;
//FIXME chroma ME //FIXME chroma ME
uint8_t *ref= ref_data[0] + (mx_i>>1) + (my_i>>1)*stride; uint8_t *ref= c->ref[field_select+ref_index][0] + (mx_i>>1) + (my_i>>1)*stride;
dxy = ((my_i & 1) << 1) | (mx_i & 1); dxy = ((my_i & 1) << 1) | (mx_i & 1);
if(s->no_rounding){ if(s->no_rounding){
s->dsp.put_no_rnd_pixels_tab[size][dxy](s->me.scratchpad, ref , stride, h); s->dsp.put_no_rnd_pixels_tab[size][dxy](c->scratchpad, ref , stride, h);
}else{ }else{
s->dsp.put_pixels_tab [size][dxy](s->me.scratchpad, ref , stride, h); s->dsp.put_pixels_tab [size][dxy](c->scratchpad, ref , stride, h);
} }
dmin= s->dsp.mb_cmp[size](s, src_data[0], s->me.scratchpad, stride, h); dmin= s->dsp.mb_cmp[size](s, c->src[block][0], c->scratchpad, stride, h);
dmin+= (mv_penalty[mx_i-pred_x] + mv_penalty[my_i-pred_y] + 1)*s->me.mb_penalty_factor; dmin+= (mv_penalty[mx_i-s->me.pred_x] + mv_penalty[my_i-s->me.pred_y] + 1)*c->mb_penalty_factor;
}else }else
dmin+= s->me.mb_penalty_factor; //field_select bits dmin+= c->mb_penalty_factor; //field_select bits
dmin += field_select != block; //slightly prefer same field dmin += field_select != block; //slightly prefer same field
...@@ -1024,8 +955,10 @@ static int interlaced_search(MpegEncContext *s, uint8_t *frame_src_data[3], uint ...@@ -1024,8 +955,10 @@ static int interlaced_search(MpegEncContext *s, uint8_t *frame_src_data[3], uint
dmin_sum += best_dmin; dmin_sum += best_dmin;
} }
s->me.ymin<<=1; c->ymin<<=1;
s->me.ymax<<=1; c->ymax<<=1;
c->stride>>=1;
c->uvstride>>=1;
if(same) if(same)
return INT_MAX; return INT_MAX;
...@@ -1036,40 +969,32 @@ static int interlaced_search(MpegEncContext *s, uint8_t *frame_src_data[3], uint ...@@ -1036,40 +969,32 @@ static int interlaced_search(MpegEncContext *s, uint8_t *frame_src_data[3], uint
case FF_CMP_RD: case FF_CMP_RD:
return dmin_sum; return dmin_sum;
default: default:
return dmin_sum+ 11*s->me.mb_penalty_factor; return dmin_sum+ 11*c->mb_penalty_factor;
} }
} }
void ff_estimate_p_frame_motion(MpegEncContext * s, void ff_estimate_p_frame_motion(MpegEncContext * s,
int mb_x, int mb_y) int mb_x, int mb_y)
{ {
MotionEstContext * const c= &s->me;
uint8_t *pix, *ppix; uint8_t *pix, *ppix;
int sum, varc, vard, mx, my, dmin, xx, yy; int sum, varc, vard, mx, my, dmin, xx, yy;
int pred_x=0, pred_y=0;
int P[10][2]; int P[10][2];
const int shift= 1+s->quarter_sample; const int shift= 1+s->quarter_sample;
int mb_type=0; int mb_type=0;
uint8_t *ref_picture= s->last_picture.data[0];
Picture * const pic= &s->current_picture; Picture * const pic= &s->current_picture;
uint8_t * const mv_penalty= s->me.mv_penalty[s->f_code] + MAX_MV;
const int stride= s->linesize; init_ref(s, s->new_picture.data, s->last_picture.data, NULL, 16*mb_x, 16*mb_y, 0);
const int uvstride= s->uvlinesize; init_mc(s, 0, c->flags);
uint8_t *src_data[3]= {
s->new_picture.data[0] + 16*(mb_x + stride*mb_y),
s->new_picture.data[1] + 8*(mb_x + uvstride*mb_y),
s->new_picture.data[2] + 8*(mb_x + uvstride*mb_y)
};
uint8_t *ref_data[3]= {
s->last_picture.data[0] + 16*(mb_x + stride*mb_y),
s->last_picture.data[1] + 8*(mb_x + uvstride*mb_y),
s->last_picture.data[2] + 8*(mb_x + uvstride*mb_y)
};
assert(s->quarter_sample==0 || s->quarter_sample==1); assert(s->quarter_sample==0 || s->quarter_sample==1);
assert(s->linesize == s->me.stride);
assert(s->uvlinesize == s->me.uvstride);
s->me.penalty_factor = get_penalty_factor(s, s->avctx->me_cmp); s->me.penalty_factor = get_penalty_factor(s, s->avctx->me_cmp);
s->me.sub_penalty_factor= get_penalty_factor(s, s->avctx->me_sub_cmp); s->me.sub_penalty_factor= get_penalty_factor(s, s->avctx->me_sub_cmp);
s->me.mb_penalty_factor = get_penalty_factor(s, s->avctx->mb_cmp); s->me.mb_penalty_factor = get_penalty_factor(s, s->avctx->mb_cmp);
s->me.current_mv_penalty= s->me.mv_penalty[s->f_code] + MAX_MV;
get_limits(s, 16*mb_x, 16*mb_y); get_limits(s, 16*mb_x, 16*mb_y);
s->me.skip=0; s->me.skip=0;
...@@ -1123,20 +1048,19 @@ void ff_estimate_p_frame_motion(MpegEncContext * s, ...@@ -1123,20 +1048,19 @@ void ff_estimate_p_frame_motion(MpegEncContext * s,
P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]); P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
if(s->out_format == FMT_H263){ if(s->out_format == FMT_H263){
pred_x = P_MEDIAN[0]; c->pred_x = P_MEDIAN[0];
pred_y = P_MEDIAN[1]; c->pred_y = P_MEDIAN[1];
}else { /* mpeg1 at least */ }else { /* mpeg1 at least */
pred_x= P_LEFT[0]; c->pred_x= P_LEFT[0];
pred_y= P_LEFT[1]; c->pred_y= P_LEFT[1];
} }
}else{ }else{
pred_x= P_LEFT[0]; c->pred_x= P_LEFT[0];
pred_y= P_LEFT[1]; c->pred_y= P_LEFT[1];
} }
} }
dmin = s->me.motion_search[0](s, &mx, &my, P, pred_x, pred_y, dmin = epzs_motion_search(s, &mx, &my, P, 0, 0, s->p_mv_table, (1<<16)>>shift);
src_data, ref_data, stride, uvstride, s->p_mv_table, (1<<16)>>shift, mv_penalty);
break; break;
} }
...@@ -1145,9 +1069,9 @@ void ff_estimate_p_frame_motion(MpegEncContext * s, ...@@ -1145,9 +1069,9 @@ void ff_estimate_p_frame_motion(MpegEncContext * s,
xx = mb_x * 16; xx = mb_x * 16;
yy = mb_y * 16; yy = mb_y * 16;
pix = src_data[0]; pix = c->src[0][0];
/* At this point (mx,my) are full-pell and the relative displacement */ /* At this point (mx,my) are full-pell and the relative displacement */
ppix = ref_data[0] + (my * s->linesize) + mx; ppix = c->ref[0][0] + (my * s->linesize) + mx;
sum = s->dsp.pix_sum(pix, s->linesize); sum = s->dsp.pix_sum(pix, s->linesize);
...@@ -1177,8 +1101,7 @@ void ff_estimate_p_frame_motion(MpegEncContext * s, ...@@ -1177,8 +1101,7 @@ void ff_estimate_p_frame_motion(MpegEncContext * s,
mb_type|= CANDIDATE_MB_TYPE_INTRA; mb_type|= CANDIDATE_MB_TYPE_INTRA;
if (varc*2 + 200 > vard){ if (varc*2 + 200 > vard){
mb_type|= CANDIDATE_MB_TYPE_INTER; mb_type|= CANDIDATE_MB_TYPE_INTER;
s->me.sub_motion_search(s, &mx, &my, dmin, s->me.sub_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
pred_x, pred_y, src_data, ref_data, stride, uvstride, 0, 16, mv_penalty);
if(s->flags&CODEC_FLAG_MV0) if(s->flags&CODEC_FLAG_MV0)
if(mx || my) if(mx || my)
mb_type |= CANDIDATE_MB_TYPE_SKIPED; //FIXME check difference mb_type |= CANDIDATE_MB_TYPE_SKIPED; //FIXME check difference
...@@ -1196,17 +1119,16 @@ void ff_estimate_p_frame_motion(MpegEncContext * s, ...@@ -1196,17 +1119,16 @@ void ff_estimate_p_frame_motion(MpegEncContext * s,
set_p_mv_tables(s, mx, my, 1); set_p_mv_tables(s, mx, my, 1);
if((s->flags&CODEC_FLAG_INTERLACED_ME) if((s->flags&CODEC_FLAG_INTERLACED_ME)
&& !s->me.skip){ //FIXME varc/d checks && !s->me.skip){ //FIXME varc/d checks
if(interlaced_search(s, src_data, ref_data, s->p_field_mv_table, s->p_field_select_table, s->f_code, mx, my) < INT_MAX) if(interlaced_search(s, 0, s->p_field_mv_table, s->p_field_select_table, mx, my) < INT_MAX)
mb_type |= CANDIDATE_MB_TYPE_INTER_I; mb_type |= CANDIDATE_MB_TYPE_INTER_I;
} }
}else{ }else{
int intra_score, i; int intra_score, i;
mb_type= CANDIDATE_MB_TYPE_INTER; mb_type= CANDIDATE_MB_TYPE_INTER;
dmin= s->me.sub_motion_search(s, &mx, &my, dmin, dmin= s->me.sub_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
pred_x, pred_y, src_data, ref_data, stride, uvstride, 0, 16, mv_penalty);
if(s->avctx->me_sub_cmp != s->avctx->mb_cmp && !s->me.skip) if(s->avctx->me_sub_cmp != s->avctx->mb_cmp && !s->me.skip)
dmin= s->me.get_mb_score(s, mx, my, pred_x, pred_y, src_data, ref_data, stride, uvstride, mv_penalty); dmin= get_mb_score(s, mx, my, 0, 0);
if((s->flags&CODEC_FLAG_4MV) if((s->flags&CODEC_FLAG_4MV)
&& !s->me.skip && varc>50 && vard>10){ && !s->me.skip && varc>50 && vard>10){
...@@ -1218,7 +1140,7 @@ void ff_estimate_p_frame_motion(MpegEncContext * s, ...@@ -1218,7 +1140,7 @@ void ff_estimate_p_frame_motion(MpegEncContext * s,
} }
if((s->flags&CODEC_FLAG_INTERLACED_ME) if((s->flags&CODEC_FLAG_INTERLACED_ME)
&& !s->me.skip){ //FIXME varc/d checks && !s->me.skip){ //FIXME varc/d checks
int dmin_i= interlaced_search(s, src_data, ref_data, s->p_field_mv_table, s->p_field_select_table, s->f_code, mx, my); int dmin_i= interlaced_search(s, 0, s->p_field_mv_table, s->p_field_select_table, mx, my);
if(dmin_i < dmin){ if(dmin_i < dmin){
mb_type = CANDIDATE_MB_TYPE_INTER_I; mb_type = CANDIDATE_MB_TYPE_INTER_I;
dmin= dmin_i; dmin= dmin_i;
...@@ -1289,28 +1211,18 @@ void ff_estimate_p_frame_motion(MpegEncContext * s, ...@@ -1289,28 +1211,18 @@ void ff_estimate_p_frame_motion(MpegEncContext * s,
int ff_pre_estimate_p_frame_motion(MpegEncContext * s, int ff_pre_estimate_p_frame_motion(MpegEncContext * s,
int mb_x, int mb_y) int mb_x, int mb_y)
{ {
MotionEstContext * const c= &s->me;
int mx, my, dmin; int mx, my, dmin;
int pred_x=0, pred_y=0;
int P[10][2]; int P[10][2];
const int shift= 1+s->quarter_sample; const int shift= 1+s->quarter_sample;
uint8_t * const mv_penalty= s->me.mv_penalty[s->f_code] + MAX_MV;
const int xy= mb_x + mb_y*s->mb_stride; const int xy= mb_x + mb_y*s->mb_stride;
const int stride= s->linesize; init_ref(s, s->new_picture.data, s->last_picture.data, NULL, 16*mb_x, 16*mb_y, 0);
const int uvstride= s->uvlinesize; init_mc(s, 0, c->flags);
uint8_t *src_data[3]= {
s->new_picture.data[0] + 16*(mb_x + stride*mb_y),
s->new_picture.data[1] + 8*(mb_x + uvstride*mb_y),
s->new_picture.data[2] + 8*(mb_x + uvstride*mb_y)
};
uint8_t *ref_data[3]= {
s->last_picture.data[0] + 16*(mb_x + stride*mb_y),
s->last_picture.data[1] + 8*(mb_x + uvstride*mb_y),
s->last_picture.data[2] + 8*(mb_x + uvstride*mb_y)
};
assert(s->quarter_sample==0 || s->quarter_sample==1); assert(s->quarter_sample==0 || s->quarter_sample==1);
s->me.pre_penalty_factor = get_penalty_factor(s, s->avctx->me_pre_cmp); s->me.pre_penalty_factor = get_penalty_factor(s, s->avctx->me_pre_cmp);
s->me.current_mv_penalty= s->me.mv_penalty[s->f_code] + MAX_MV;
get_limits(s, 16*mb_x, 16*mb_y); get_limits(s, 16*mb_x, 16*mb_y);
s->me.skip=0; s->me.skip=0;
...@@ -1322,8 +1234,8 @@ int ff_pre_estimate_p_frame_motion(MpegEncContext * s, ...@@ -1322,8 +1234,8 @@ int ff_pre_estimate_p_frame_motion(MpegEncContext * s,
/* special case for first line */ /* special case for first line */
if (s->first_slice_line) { if (s->first_slice_line) {
pred_x= P_LEFT[0]; c->pred_x= P_LEFT[0];
pred_y= P_LEFT[1]; c->pred_y= P_LEFT[1];
P_TOP[0]= P_TOPRIGHT[0]= P_MEDIAN[0]= P_TOP[0]= P_TOPRIGHT[0]= P_MEDIAN[0]=
P_TOP[1]= P_TOPRIGHT[1]= P_MEDIAN[1]= 0; //FIXME P_TOP[1]= P_TOPRIGHT[1]= P_MEDIAN[1]= 0; //FIXME
} else { } else {
...@@ -1338,11 +1250,11 @@ int ff_pre_estimate_p_frame_motion(MpegEncContext * s, ...@@ -1338,11 +1250,11 @@ int ff_pre_estimate_p_frame_motion(MpegEncContext * s,
P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]); P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]); P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
pred_x = P_MEDIAN[0]; c->pred_x = P_MEDIAN[0];
pred_y = P_MEDIAN[1]; c->pred_y = P_MEDIAN[1];
} }
dmin = s->me.pre_motion_search(s, &mx, &my, P, pred_x, pred_y,
src_data, ref_data, stride, uvstride, s->p_mv_table, (1<<16)>>shift, mv_penalty); dmin = epzs_motion_search(s, &mx, &my, P, 0, 0, s->p_mv_table, (1<<16)>>shift);
s->p_mv_table[xy][0] = mx<<shift; s->p_mv_table[xy][0] = mx<<shift;
s->p_mv_table[xy][1] = my<<shift; s->p_mv_table[xy][1] = my<<shift;
...@@ -1351,22 +1263,20 @@ int ff_pre_estimate_p_frame_motion(MpegEncContext * s, ...@@ -1351,22 +1263,20 @@ int ff_pre_estimate_p_frame_motion(MpegEncContext * s,
} }
static int ff_estimate_motion_b(MpegEncContext * s, static int ff_estimate_motion_b(MpegEncContext * s,
int mb_x, int mb_y, int16_t (*mv_table)[2], uint8_t *src_data[3], int mb_x, int mb_y, int16_t (*mv_table)[2], int ref_index, int f_code)
uint8_t *ref_data[3], int stride, int uvstride, int f_code)
{ {
int mx, my, dmin; int mx, my, dmin;
int pred_x=0, pred_y=0;
int P[10][2]; int P[10][2];
const int shift= 1+s->quarter_sample; const int shift= 1+s->quarter_sample;
const int mot_stride = s->mb_stride; const int mot_stride = s->mb_stride;
const int mot_xy = mb_y*mot_stride + mb_x; const int mot_xy = mb_y*mot_stride + mb_x;
uint8_t * const ref_picture= ref_data[0] - 16*s->mb_x - 16*s->mb_y*s->linesize; //FIXME ugly
uint8_t * const mv_penalty= s->me.mv_penalty[f_code] + MAX_MV; uint8_t * const mv_penalty= s->me.mv_penalty[f_code] + MAX_MV;
int mv_scale; int mv_scale;
s->me.penalty_factor = get_penalty_factor(s, s->avctx->me_cmp); s->me.penalty_factor = get_penalty_factor(s, s->avctx->me_cmp);
s->me.sub_penalty_factor= get_penalty_factor(s, s->avctx->me_sub_cmp); s->me.sub_penalty_factor= get_penalty_factor(s, s->avctx->me_sub_cmp);
s->me.mb_penalty_factor = get_penalty_factor(s, s->avctx->mb_cmp); s->me.mb_penalty_factor = get_penalty_factor(s, s->avctx->mb_cmp);
s->me.current_mv_penalty= mv_penalty;
get_limits(s, 16*mb_x, 16*mb_y); get_limits(s, 16*mb_x, 16*mb_y);
...@@ -1416,8 +1326,8 @@ static int ff_estimate_motion_b(MpegEncContext * s, ...@@ -1416,8 +1326,8 @@ static int ff_estimate_motion_b(MpegEncContext * s,
P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]); P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]); P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
} }
pred_x= P_LEFT[0]; s->me.pred_x= P_LEFT[0];
pred_y= P_LEFT[1]; s->me.pred_y= P_LEFT[1];
} }
if(mv_table == s->b_forw_mv_table){ if(mv_table == s->b_forw_mv_table){
...@@ -1426,17 +1336,15 @@ static int ff_estimate_motion_b(MpegEncContext * s, ...@@ -1426,17 +1336,15 @@ static int ff_estimate_motion_b(MpegEncContext * s,
mv_scale= ((s->pb_time - s->pp_time)<<16) / (s->pp_time<<shift); mv_scale= ((s->pb_time - s->pp_time)<<16) / (s->pp_time<<shift);
} }
dmin = s->me.motion_search[0](s, &mx, &my, P, pred_x, pred_y, dmin = epzs_motion_search(s, &mx, &my, P, 0, ref_index, s->p_mv_table, mv_scale);
src_data, ref_data, stride, uvstride, s->p_mv_table, mv_scale, mv_penalty);
break; break;
} }
dmin= s->me.sub_motion_search(s, &mx, &my, dmin, dmin= s->me.sub_motion_search(s, &mx, &my, dmin, 0, ref_index, 0, 16);
pred_x, pred_y, src_data, ref_data, stride, uvstride, 0, 16, mv_penalty);
if(s->avctx->me_sub_cmp != s->avctx->mb_cmp && !s->me.skip) if(s->avctx->me_sub_cmp != s->avctx->mb_cmp && !s->me.skip)
dmin= s->me.get_mb_score(s, mx, my, pred_x, pred_y, src_data, ref_data, stride, uvstride, mv_penalty); dmin= get_mb_score(s, mx, my, 0, ref_index);
//printf("%d %d %d %d//", s->mb_x, s->mb_y, mx, my); //printf("%d %d %d %d//", s->mb_x, s->mb_y, mx, my);
// s->mb_type[mb_y*s->mb_width + mb_x]= mb_type; // s->mb_type[mb_y*s->mb_width + mb_x]= mb_type;
...@@ -1446,8 +1354,7 @@ static int ff_estimate_motion_b(MpegEncContext * s, ...@@ -1446,8 +1354,7 @@ static int ff_estimate_motion_b(MpegEncContext * s,
return dmin; return dmin;
} }
static inline int check_bidir_mv(MpegEncContext * s, uint8_t *src_data[3], uint8_t *ref_data[6], static inline int check_bidir_mv(MpegEncContext * s,
int stride, int uvstride,
int motion_fx, int motion_fy, int motion_fx, int motion_fy,
int motion_bx, int motion_by, int motion_bx, int motion_by,
int pred_fx, int pred_fy, int pred_fx, int pred_fy,
...@@ -1455,15 +1362,20 @@ static inline int check_bidir_mv(MpegEncContext * s, uint8_t *src_data[3], uint8 ...@@ -1455,15 +1362,20 @@ static inline int check_bidir_mv(MpegEncContext * s, uint8_t *src_data[3], uint8
int size, int h) int size, int h)
{ {
//FIXME optimize? //FIXME optimize?
//FIXME move into template?
//FIXME better f_code prediction (max mv & distance) //FIXME better f_code prediction (max mv & distance)
//FIXME pointers //FIXME pointers
MotionEstContext * const c= &s->me;
uint8_t * const mv_penalty= s->me.mv_penalty[s->f_code] + MAX_MV; // f_code of the prev frame uint8_t * const mv_penalty= s->me.mv_penalty[s->f_code] + MAX_MV; // f_code of the prev frame
int stride= s->me.stride;
int uvstride= s->me.uvstride;
uint8_t *dest_y = s->me.scratchpad; uint8_t *dest_y = s->me.scratchpad;
uint8_t *ptr; uint8_t *ptr;
int dxy; int dxy;
int src_x, src_y; int src_x, src_y;
int fbmin; int fbmin;
uint8_t **src_data= c->src[0];
uint8_t **ref_data= c->ref[0];
uint8_t **ref2_data= c->ref[2];
if(s->quarter_sample){ if(s->quarter_sample){
dxy = ((motion_fy & 3) << 2) | (motion_fx & 3); dxy = ((motion_fy & 3) << 2) | (motion_fx & 3);
...@@ -1477,7 +1389,7 @@ static inline int check_bidir_mv(MpegEncContext * s, uint8_t *src_data[3], uint8 ...@@ -1477,7 +1389,7 @@ static inline int check_bidir_mv(MpegEncContext * s, uint8_t *src_data[3], uint8
src_x = motion_bx >> 2; src_x = motion_bx >> 2;
src_y = motion_by >> 2; src_y = motion_by >> 2;
ptr = ref_data[3] + (src_y * stride) + src_x; ptr = ref2_data[0] + (src_y * stride) + src_x;
s->dsp.avg_qpel_pixels_tab[size][dxy](dest_y , ptr , stride); s->dsp.avg_qpel_pixels_tab[size][dxy](dest_y , ptr , stride);
}else{ }else{
dxy = ((motion_fy & 1) << 1) | (motion_fx & 1); dxy = ((motion_fy & 1) << 1) | (motion_fx & 1);
...@@ -1491,7 +1403,7 @@ static inline int check_bidir_mv(MpegEncContext * s, uint8_t *src_data[3], uint8 ...@@ -1491,7 +1403,7 @@ static inline int check_bidir_mv(MpegEncContext * s, uint8_t *src_data[3], uint8
src_x = motion_bx >> 1; src_x = motion_bx >> 1;
src_y = motion_by >> 1; src_y = motion_by >> 1;
ptr = ref_data[3] + (src_y * stride) + src_x; ptr = ref2_data[0] + (src_y * stride) + src_x;
s->dsp.avg_pixels_tab[size][dxy](dest_y , ptr , stride, h); s->dsp.avg_pixels_tab[size][dxy](dest_y , ptr , stride, h);
} }
...@@ -1507,9 +1419,7 @@ static inline int check_bidir_mv(MpegEncContext * s, uint8_t *src_data[3], uint8 ...@@ -1507,9 +1419,7 @@ static inline int check_bidir_mv(MpegEncContext * s, uint8_t *src_data[3], uint8
} }
/* refine the bidir vectors in hq mode and return the score in both lq & hq mode*/ /* refine the bidir vectors in hq mode and return the score in both lq & hq mode*/
static inline int bidir_refine(MpegEncContext * s, uint8_t *src_data[3], uint8_t *ref_data[6], static inline int bidir_refine(MpegEncContext * s, int mb_x, int mb_y)
int stride, int uvstride,
int mb_x, int mb_y)
{ {
const int mot_stride = s->mb_stride; const int mot_stride = s->mb_stride;
const int xy = mb_y *mot_stride + mb_x; const int xy = mb_y *mot_stride + mb_x;
...@@ -1525,8 +1435,7 @@ static inline int bidir_refine(MpegEncContext * s, uint8_t *src_data[3], uint8_t ...@@ -1525,8 +1435,7 @@ static inline int bidir_refine(MpegEncContext * s, uint8_t *src_data[3], uint8_t
//FIXME do refinement and add flag //FIXME do refinement and add flag
fbmin= check_bidir_mv(s, src_data, ref_data, stride, uvstride, fbmin= check_bidir_mv(s, motion_fx, motion_fy,
motion_fx, motion_fy,
motion_bx, motion_by, motion_bx, motion_by,
pred_fx, pred_fy, pred_fx, pred_fy,
pred_bx, pred_by, pred_bx, pred_by,
...@@ -1535,9 +1444,7 @@ static inline int bidir_refine(MpegEncContext * s, uint8_t *src_data[3], uint8_t ...@@ -1535,9 +1444,7 @@ static inline int bidir_refine(MpegEncContext * s, uint8_t *src_data[3], uint8_t
return fbmin; return fbmin;
} }
static inline int direct_search(MpegEncContext * s, uint8_t *src_data[3], uint8_t *ref_data[6], static inline int direct_search(MpegEncContext * s, int mb_x, int mb_y)
int stride, int uvstride,
int mb_x, int mb_y)
{ {
int P[10][2]; int P[10][2];
const int mot_stride = s->mb_stride; const int mot_stride = s->mb_stride;
...@@ -1548,8 +1455,8 @@ static inline int direct_search(MpegEncContext * s, uint8_t *src_data[3], uint8_ ...@@ -1548,8 +1455,8 @@ static inline int direct_search(MpegEncContext * s, uint8_t *src_data[3], uint8_
const int time_pb= s->pb_time; const int time_pb= s->pb_time;
int mx, my, xmin, xmax, ymin, ymax; int mx, my, xmin, xmax, ymin, ymax;
int16_t (*mv_table)[2]= s->b_direct_mv_table; int16_t (*mv_table)[2]= s->b_direct_mv_table;
uint8_t * const mv_penalty= s->me.mv_penalty[1] + MAX_MV;
s->me.current_mv_penalty= s->me.mv_penalty[1] + MAX_MV;
ymin= xmin=(-32)>>shift; ymin= xmin=(-32)>>shift;
ymax= xmax= 31>>shift; ymax= xmax= 31>>shift;
...@@ -1600,6 +1507,10 @@ static inline int direct_search(MpegEncContext * s, uint8_t *src_data[3], uint8_ ...@@ -1600,6 +1507,10 @@ static inline int direct_search(MpegEncContext * s, uint8_t *src_data[3], uint8_
s->me.ymin= ymin; s->me.ymin= ymin;
s->me.xmax= xmax; s->me.xmax= xmax;
s->me.ymax= ymax; s->me.ymax= ymax;
s->me.flags |= FLAG_DIRECT;
s->me.sub_flags |= FLAG_DIRECT;
s->me.pred_x=0;
s->me.pred_y=0;
P_LEFT[0] = clip(mv_table[mot_xy - 1][0], xmin<<shift, xmax<<shift); P_LEFT[0] = clip(mv_table[mot_xy - 1][0], xmin<<shift, xmax<<shift);
P_LEFT[1] = clip(mv_table[mot_xy - 1][1], ymin<<shift, ymax<<shift); P_LEFT[1] = clip(mv_table[mot_xy - 1][1], ymin<<shift, ymax<<shift);
...@@ -1615,29 +1526,22 @@ static inline int direct_search(MpegEncContext * s, uint8_t *src_data[3], uint8_ ...@@ -1615,29 +1526,22 @@ static inline int direct_search(MpegEncContext * s, uint8_t *src_data[3], uint8_
P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]); P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
} }
//FIXME direct_search ptr in context!!! (needed for chroma anyway or this will get messy) dmin = epzs_motion_search(s, &mx, &my, P, 0, 0, mv_table, 1<<(16-shift));
if(s->flags&CODEC_FLAG_QPEL){ if(s->me.sub_flags&FLAG_QPEL)
dmin = simple_direct_qpel_epzs_motion_search(s, &mx, &my, P, 0, 0, dmin = qpel_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
src_data, ref_data, stride, uvstride, mv_table, 1<<14, mv_penalty); else
dmin = simple_direct_qpel_qpel_motion_search(s, &mx, &my, dmin, dmin = hpel_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
0, 0, src_data, ref_data, stride, uvstride, 0, 16, mv_penalty);
if(s->avctx->me_sub_cmp != s->avctx->mb_cmp && !s->me.skip)
dmin= simple_direct_qpel_qpel_get_mb_score(s, mx, my, 0, 0, src_data, ref_data, stride, uvstride, mv_penalty);
}else{
dmin = simple_direct_hpel_epzs_motion_search(s, &mx, &my, P, 0, 0,
src_data, ref_data, stride, uvstride, mv_table, 1<<15, mv_penalty);
dmin = simple_direct_hpel_hpel_motion_search(s, &mx, &my, dmin,
0, 0, src_data, ref_data, stride, uvstride, 0, 16, mv_penalty);
if(s->avctx->me_sub_cmp != s->avctx->mb_cmp && !s->me.skip) if(s->avctx->me_sub_cmp != s->avctx->mb_cmp && !s->me.skip)
dmin= simple_direct_hpel_hpel_get_mb_score(s, mx, my, 0, 0, src_data, ref_data, stride, uvstride, mv_penalty); dmin= get_mb_score(s, mx, my, 0, 0);
}
get_limits(s, 16*mb_x, 16*mb_y); //restore s->me.?min/max, maybe not needed get_limits(s, 16*mb_x, 16*mb_y); //restore s->me.?min/max, maybe not needed
s->b_direct_mv_table[mot_xy][0]= mx; s->b_direct_mv_table[mot_xy][0]= mx;
s->b_direct_mv_table[mot_xy][1]= my; s->b_direct_mv_table[mot_xy][1]= my;
s->me.flags &= ~FLAG_DIRECT;
s->me.sub_flags &= ~FLAG_DIRECT;
return dmin; return dmin;
} }
...@@ -1647,39 +1551,24 @@ void ff_estimate_b_frame_motion(MpegEncContext * s, ...@@ -1647,39 +1551,24 @@ void ff_estimate_b_frame_motion(MpegEncContext * s,
const int penalty_factor= s->me.mb_penalty_factor; const int penalty_factor= s->me.mb_penalty_factor;
int fmin, bmin, dmin, fbmin, bimin, fimin; int fmin, bmin, dmin, fbmin, bimin, fimin;
int type=0; int type=0;
const int stride= s->linesize; init_ref(s, s->new_picture.data, s->last_picture.data, s->next_picture.data, 16*mb_x, 16*mb_y, 2);
const int uvstride= s->uvlinesize; init_mc(s, 0, s->me.flags);
uint8_t *src_data[3]= {
s->new_picture.data[0] + 16*(s->mb_x + stride*s->mb_y),
s->new_picture.data[1] + 8*(s->mb_x + uvstride*s->mb_y),
s->new_picture.data[2] + 8*(s->mb_x + uvstride*s->mb_y)
};
uint8_t *ref_data[6]= {
s->last_picture.data[0] + 16*(s->mb_x + stride*s->mb_y),
s->last_picture.data[1] + 8*(s->mb_x + uvstride*s->mb_y),
s->last_picture.data[2] + 8*(s->mb_x + uvstride*s->mb_y),
s->next_picture.data[0] + 16*(s->mb_x + stride*s->mb_y),
s->next_picture.data[1] + 8*(s->mb_x + uvstride*s->mb_y),
s->next_picture.data[2] + 8*(s->mb_x + uvstride*s->mb_y)
};
s->me.skip=0; s->me.skip=0;
if (s->codec_id == CODEC_ID_MPEG4) if (s->codec_id == CODEC_ID_MPEG4)
dmin= direct_search(s, src_data, ref_data, stride, uvstride, mb_x, mb_y); dmin= direct_search(s, mb_x, mb_y);
else else
dmin= INT_MAX; dmin= INT_MAX;
//FIXME penalty stuff for non mpeg4 //FIXME penalty stuff for non mpeg4
s->me.skip=0; s->me.skip=0;
fmin= ff_estimate_motion_b(s, mb_x, mb_y, s->b_forw_mv_table, src_data, fmin= ff_estimate_motion_b(s, mb_x, mb_y, s->b_forw_mv_table, 0, s->f_code) + 3*penalty_factor;
ref_data, stride, uvstride, s->f_code) + 3*penalty_factor;
s->me.skip=0; s->me.skip=0;
bmin= ff_estimate_motion_b(s, mb_x, mb_y, s->b_back_mv_table, src_data, bmin= ff_estimate_motion_b(s, mb_x, mb_y, s->b_back_mv_table, 2, s->b_code) + 2*penalty_factor;
ref_data+3, stride, uvstride, s->b_code) + 2*penalty_factor;
//printf(" %d %d ", s->b_forw_mv_table[xy][0], s->b_forw_mv_table[xy][1]); //printf(" %d %d ", s->b_forw_mv_table[xy][0], s->b_forw_mv_table[xy][1]);
s->me.skip=0; s->me.skip=0;
fbmin= bidir_refine(s, src_data, ref_data, stride, uvstride, mb_x, mb_y) + penalty_factor; fbmin= bidir_refine(s, mb_x, mb_y) + penalty_factor;
//printf("%d %d %d %d\n", dmin, fmin, bmin, fbmin); //printf("%d %d %d %d\n", dmin, fmin, bmin, fbmin);
if(s->flags & CODEC_FLAG_INTERLACED_ME){ if(s->flags & CODEC_FLAG_INTERLACED_ME){
...@@ -1687,11 +1576,13 @@ void ff_estimate_b_frame_motion(MpegEncContext * s, ...@@ -1687,11 +1576,13 @@ void ff_estimate_b_frame_motion(MpegEncContext * s,
//FIXME mb type penalty //FIXME mb type penalty
s->me.skip=0; s->me.skip=0;
fimin= interlaced_search(s, src_data, ref_data , s->me.current_mv_penalty= s->me.mv_penalty[s->f_code] + MAX_MV;
s->b_field_mv_table[0], s->b_field_select_table[0], s->f_code, fimin= interlaced_search(s, 0,
s->b_field_mv_table[0], s->b_field_select_table[0],
s->b_forw_mv_table[xy][0], s->b_forw_mv_table[xy][1]); s->b_forw_mv_table[xy][0], s->b_forw_mv_table[xy][1]);
bimin= interlaced_search(s, src_data, ref_data+3, s->me.current_mv_penalty= s->me.mv_penalty[s->b_code] + MAX_MV;
s->b_field_mv_table[1], s->b_field_select_table[1], s->b_code, bimin= interlaced_search(s, 2,
s->b_field_mv_table[1], s->b_field_select_table[1],
s->b_back_mv_table[xy][0], s->b_back_mv_table[xy][1]); s->b_back_mv_table[xy][0], s->b_back_mv_table[xy][1]);
}else }else
fimin= bimin= INT_MAX; fimin= bimin= INT_MAX;
......
libavcodec/motion_est_template.c
View file @ 2750b827
...@@ -22,58 +22,32 @@ ...@@ -22,58 +22,32 @@
* @file motion_est_template.c * @file motion_est_template.c
* Motion estimation template. * Motion estimation template.
*/ */
//FIXME ref2_y next_pic?
//lets hope gcc will remove the unused vars ...(gcc 3.2.2 seems to do it ...) //lets hope gcc will remove the unused vars ...(gcc 3.2.2 seems to do it ...)
//Note, the last line is there to kill these ugly unused var warnings
#define LOAD_COMMON\ #define LOAD_COMMON\
uint32_t * const score_map= s->me.score_map;\ uint32_t * const score_map= s->me.score_map;\
const int time_pp= s->pp_time;\
const int time_pb= s->pb_time;\
const int xmin= s->me.xmin;\ const int xmin= s->me.xmin;\
const int ymin= s->me.ymin;\ const int ymin= s->me.ymin;\
const int xmax= s->me.xmax;\ const int xmax= s->me.xmax;\
const int ymax= s->me.ymax;\ const int ymax= s->me.ymax;\
uint8_t * const src_y= src_data[0];\ uint8_t *mv_penalty= s->me.current_mv_penalty;\
uint8_t * const src_u= src_data[1];\ const int pred_x= s->me.pred_x;\
uint8_t * const src_v= src_data[2];\ const int pred_y= s->me.pred_y;\
uint8_t * const ref_y= ref_data[0];\
uint8_t * const ref_u= ref_data[1];\
uint8_t * const ref_v= ref_data[2];\
op_pixels_func (*hpel_put)[4];\
op_pixels_func (*hpel_avg)[4]= &s->dsp.avg_pixels_tab[size];\
op_pixels_func (*chroma_hpel_put)[4];\
qpel_mc_func (*qpel_put)[16];\
qpel_mc_func (*qpel_avg)[16]= &s->dsp.avg_qpel_pixels_tab[size];\
const __attribute__((unused)) int unu= time_pp + time_pb + (size_t)src_u + (size_t)src_v + (size_t)ref_u + (size_t)ref_v\
+ (size_t)hpel_avg + (size_t)qpel_avg + (size_t)score_map\
+ xmin + xmax + ymin + ymax;\
if(s->no_rounding /*FIXME b_type*/){\
hpel_put= &s->dsp.put_no_rnd_pixels_tab[size];\
chroma_hpel_put= &s->dsp.put_no_rnd_pixels_tab[size+1];\
qpel_put= &s->dsp.put_no_rnd_qpel_pixels_tab[size];\
}else{\
hpel_put=& s->dsp.put_pixels_tab[size];\
chroma_hpel_put= &s->dsp.put_pixels_tab[size+1];\
qpel_put= &s->dsp.put_qpel_pixels_tab[size];\
}
#ifdef CMP_HPEL
#define CHECK_HALF_MV(dx, dy, x, y)\ #define CHECK_HALF_MV(dx, dy, x, y)\
{\ {\
const int hx= 2*(x)+(dx);\ const int hx= 2*(x)+(dx);\
const int hy= 2*(y)+(dy);\ const int hy= 2*(y)+(dy);\
CMP_HPEL(d, dx, dy, x, y, size);\ d= cmp(s, x, y, dx, dy, size, h, ref_index, src_index, cmp_sub, chroma_cmp_sub, flags);\
d += (mv_penalty[hx - pred_x] + mv_penalty[hy - pred_y])*penalty_factor;\ d += (mv_penalty[hx - pred_x] + mv_penalty[hy - pred_y])*penalty_factor;\
COPY3_IF_LT(dmin, d, bx, hx, by, hy)\ COPY3_IF_LT(dmin, d, bx, hx, by, hy)\
} }
#if 0 #if 0
static int RENAME(hpel_motion_search)(MpegEncContext * s, static int hpel_motion_search)(MpegEncContext * s,
int *mx_ptr, int *my_ptr, int dmin, int *mx_ptr, int *my_ptr, int dmin,
int pred_x, int pred_y, uint8_t *ref_data[3], uint8_t *ref_data[3],
int size, uint8_t * const mv_penalty) int size)
{ {
const int xx = 16 * s->mb_x + 8*(n&1); const int xx = 16 * s->mb_x + 8*(n&1);
const int yy = 16 * s->mb_y + 8*(n>>1); const int yy = 16 * s->mb_y + 8*(n>>1);
...@@ -94,8 +68,8 @@ static int RENAME(hpel_motion_search)(MpegEncContext * s, ...@@ -94,8 +68,8 @@ static int RENAME(hpel_motion_search)(MpegEncContext * s,
hpel_put=& s->dsp.put_pixels_tab[size]; hpel_put=& s->dsp.put_pixels_tab[size];
chroma_hpel_put= &s->dsp.put_pixels_tab[size+1]; chroma_hpel_put= &s->dsp.put_pixels_tab[size+1];
} }
cmp= s->dsp.me_cmp[size]; cmpf= s->dsp.me_cmp[size];
chroma_cmp= s->dsp.me_cmp[size+1]; chroma_cmpf= s->dsp.me_cmp[size+1];
cmp_sub= s->dsp.me_sub_cmp[size]; cmp_sub= s->dsp.me_sub_cmp[size];
chroma_cmp_sub= s->dsp.me_sub_cmp[size+1]; chroma_cmp_sub= s->dsp.me_sub_cmp[size+1];
...@@ -138,11 +112,10 @@ static int RENAME(hpel_motion_search)(MpegEncContext * s, ...@@ -138,11 +112,10 @@ static int RENAME(hpel_motion_search)(MpegEncContext * s,
} }
#else #else
static int RENAME(hpel_motion_search)(MpegEncContext * s, static int hpel_motion_search(MpegEncContext * s,
int *mx_ptr, int *my_ptr, int dmin, int *mx_ptr, int *my_ptr, int dmin,
int pred_x, int pred_y, uint8_t *src_data[3], int src_index, int ref_index,
uint8_t *ref_data[3], int stride, int uvstride, int size, int h)
int size, int h, uint8_t * const mv_penalty)
{ {
const int mx = *mx_ptr; const int mx = *mx_ptr;
const int my = *my_ptr; const int my = *my_ptr;
...@@ -151,6 +124,7 @@ static int RENAME(hpel_motion_search)(MpegEncContext * s, ...@@ -151,6 +124,7 @@ static int RENAME(hpel_motion_search)(MpegEncContext * s,
int bx=2*mx, by=2*my; int bx=2*mx, by=2*my;
LOAD_COMMON LOAD_COMMON
int flags= s->me.sub_flags;
//FIXME factorize //FIXME factorize
...@@ -164,7 +138,7 @@ static int RENAME(hpel_motion_search)(MpegEncContext * s, ...@@ -164,7 +138,7 @@ static int RENAME(hpel_motion_search)(MpegEncContext * s,
} }
if(s->avctx->me_cmp != s->avctx->me_sub_cmp){ if(s->avctx->me_cmp != s->avctx->me_sub_cmp){
CMP_HPEL(dmin, 0, 0, mx, my, size); dmin= cmp(s, mx, my, 0, 0, size, h, ref_index, src_index, cmp_sub, chroma_cmp_sub, flags);
if(mx || my || size>0) if(mx || my || size>0)
dmin += (mv_penalty[2*mx - pred_x] + mv_penalty[2*my - pred_y])*penalty_factor; dmin += (mv_penalty[2*mx - pred_x] + mv_penalty[2*my - pred_y])*penalty_factor;
} }
...@@ -246,14 +220,16 @@ static int RENAME(hpel_motion_search)(MpegEncContext * s, ...@@ -246,14 +220,16 @@ static int RENAME(hpel_motion_search)(MpegEncContext * s,
} }
#endif #endif
static int RENAME(hpel_get_mb_score)(MpegEncContext * s, int mx, int my, int pred_x, int pred_y, uint8_t *src_data[3], static int inline get_mb_score(MpegEncContext * s, int mx, int my, int src_index,
uint8_t *ref_data[3], int stride, int uvstride, int ref_index)
uint8_t * const mv_penalty)
{ {
// const int check_luma= s->dsp.me_sub_cmp != s->dsp.mb_cmp; // const int check_luma= s->dsp.me_sub_cmp != s->dsp.mb_cmp;
const int size= 0; const int size= 0;
const int h= 16; const int h= 16;
const int penalty_factor= s->me.mb_penalty_factor; const int penalty_factor= s->me.mb_penalty_factor;
const int flags= s->me.mb_flags;
const int qpel= flags & FLAG_QPEL;
const int mask= 1+2*qpel;
me_cmp_func cmp_sub, chroma_cmp_sub; me_cmp_func cmp_sub, chroma_cmp_sub;
int d; int d;
...@@ -267,7 +243,7 @@ static int RENAME(hpel_get_mb_score)(MpegEncContext * s, int mx, int my, int pre ...@@ -267,7 +243,7 @@ static int RENAME(hpel_get_mb_score)(MpegEncContext * s, int mx, int my, int pre
assert(!s->me.skip); assert(!s->me.skip);
assert(s->avctx->me_sub_cmp != s->avctx->mb_cmp); assert(s->avctx->me_sub_cmp != s->avctx->mb_cmp);
CMP_HPEL(d, mx&1, my&1, mx>>1, my>>1, size); d= cmp(s, mx>>(qpel+1), my>>(qpel+1), mx&mask, my&mask, size, h, ref_index, src_index, cmp_sub, chroma_cmp_sub, flags);
//FIXME check cbp before adding penalty for (0,0) vector //FIXME check cbp before adding penalty for (0,0) vector
if(mx || my || size>0) if(mx || my || size>0)
d += (mv_penalty[mx - pred_x] + mv_penalty[my - pred_y])*penalty_factor; d += (mv_penalty[mx - pred_x] + mv_penalty[my - pred_y])*penalty_factor;
...@@ -275,26 +251,19 @@ static int RENAME(hpel_get_mb_score)(MpegEncContext * s, int mx, int my, int pre ...@@ -275,26 +251,19 @@ static int RENAME(hpel_get_mb_score)(MpegEncContext * s, int mx, int my, int pre
return d; return d;
} }
#endif /* CMP_HPEL */
#ifdef CMP_QPEL
#define CHECK_QUARTER_MV(dx, dy, x, y)\ #define CHECK_QUARTER_MV(dx, dy, x, y)\
{\ {\
const int hx= 4*(x)+(dx);\ const int hx= 4*(x)+(dx);\
const int hy= 4*(y)+(dy);\ const int hy= 4*(y)+(dy);\
CMP_QPEL(d, dx, dy, x, y, size);\ d= cmp(s, x, y, dx, dy, size, h, ref_index, src_index, cmpf, chroma_cmpf, flags);\
d += (mv_penalty[hx - pred_x] + mv_penalty[hy - pred_y])*penalty_factor;\ d += (mv_penalty[hx - pred_x] + mv_penalty[hy - pred_y])*penalty_factor;\
COPY3_IF_LT(dmin, d, bx, hx, by, hy)\ COPY3_IF_LT(dmin, d, bx, hx, by, hy)\
} }
static int RENAME(qpel_motion_search)(MpegEncContext * s, static int qpel_motion_search(MpegEncContext * s,
int *mx_ptr, int *my_ptr, int dmin, int *mx_ptr, int *my_ptr, int dmin,
int pred_x, int pred_y, uint8_t *src_data[3], int src_index, int ref_index,
uint8_t *ref_data[3], int stride, int uvstride, int size, int h)
int size, int h, uint8_t * const mv_penalty)
{ {
const int mx = *mx_ptr; const int mx = *mx_ptr;
const int my = *my_ptr; const int my = *my_ptr;
...@@ -302,13 +271,14 @@ static int RENAME(qpel_motion_search)(MpegEncContext * s, ...@@ -302,13 +271,14 @@ static int RENAME(qpel_motion_search)(MpegEncContext * s,
const int map_generation= s->me.map_generation; const int map_generation= s->me.map_generation;
const int subpel_quality= s->avctx->me_subpel_quality; const int subpel_quality= s->avctx->me_subpel_quality;
uint32_t *map= s->me.map; uint32_t *map= s->me.map;
me_cmp_func cmp, chroma_cmp; me_cmp_func cmpf, chroma_cmpf;
me_cmp_func cmp_sub, chroma_cmp_sub; me_cmp_func cmp_sub, chroma_cmp_sub;
LOAD_COMMON LOAD_COMMON
int flags= s->me.sub_flags;
cmp= s->dsp.me_cmp[size]; cmpf= s->dsp.me_cmp[size];
chroma_cmp= s->dsp.me_cmp[size+1]; //factorize FIXME chroma_cmpf= s->dsp.me_cmp[size+1]; //factorize FIXME
//FIXME factorize //FIXME factorize
cmp_sub= s->dsp.me_sub_cmp[size]; cmp_sub= s->dsp.me_sub_cmp[size];
...@@ -321,7 +291,7 @@ static int RENAME(qpel_motion_search)(MpegEncContext * s, ...@@ -321,7 +291,7 @@ static int RENAME(qpel_motion_search)(MpegEncContext * s,
} }
if(s->avctx->me_cmp != s->avctx->me_sub_cmp){ if(s->avctx->me_cmp != s->avctx->me_sub_cmp){
CMP_QPEL(dmin, 0, 0, mx, my, size); dmin= cmp(s, mx, my, 0, 0, size, h, ref_index, src_index, cmp_sub, chroma_cmp_sub, flags);
if(mx || my || size>0) if(mx || my || size>0)
dmin += (mv_penalty[4*mx - pred_x] + mv_penalty[4*my - pred_y])*penalty_factor; dmin += (mv_penalty[4*mx - pred_x] + mv_penalty[4*my - pred_y])*penalty_factor;
} }
...@@ -386,7 +356,7 @@ static int RENAME(qpel_motion_search)(MpegEncContext * s, ...@@ -386,7 +356,7 @@ static int RENAME(qpel_motion_search)(MpegEncContext * s,
if(map[(index-(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)] == (my<<ME_MAP_MV_BITS) + mx + map_generation && 0){ //FIXME if(map[(index-(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)] == (my<<ME_MAP_MV_BITS) + mx + map_generation && 0){ //FIXME
tl= score_map[(index-(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)]; tl= score_map[(index-(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)];
}else{ }else{
CMP(tl, mx-1, my-1, size); //FIXME wrong if chroma me is different tl= cmp(s, mx-1, my-1, 0, 0, size, h, ref_index, src_index, cmpf, chroma_cmpf, flags);//FIXME wrong if chroma me is different
} }
cxy= 2*tl + (cx + cy)/4 - (cx2 + cy2) - 2*c; cxy= 2*tl + (cx + cy)/4 - (cx2 + cy2) - 2*c;
...@@ -509,36 +479,6 @@ static int RENAME(qpel_motion_search)(MpegEncContext * s, ...@@ -509,36 +479,6 @@ static int RENAME(qpel_motion_search)(MpegEncContext * s,
return dmin; return dmin;
} }
static int RENAME(qpel_get_mb_score)(MpegEncContext * s, int mx, int my, int pred_x, int pred_y, uint8_t *src_data[3],
uint8_t *ref_data[3], int stride, int uvstride,
uint8_t * const mv_penalty)
{
const int size= 0;
const int h= 16;
const int penalty_factor= s->me.mb_penalty_factor;
me_cmp_func cmp_sub, chroma_cmp_sub;
int d;
LOAD_COMMON
//FIXME factorize
cmp_sub= s->dsp.mb_cmp[size];
chroma_cmp_sub= s->dsp.mb_cmp[size+1];
assert(!s->me.skip);
assert(s->avctx->me_sub_cmp != s->avctx->mb_cmp);
CMP_QPEL(d, mx&3, my&3, mx>>2, my>>2, size);
//FIXME check cbp before adding penalty for (0,0) vector
if(mx || my || size>0)
d += (mv_penalty[mx - pred_x] + mv_penalty[my - pred_y])*penalty_factor;
return d;
}
#endif /* CMP_QPEL */
#define CHECK_MV(x,y)\ #define CHECK_MV(x,y)\
{\ {\
...@@ -546,7 +486,7 @@ static int RENAME(qpel_get_mb_score)(MpegEncContext * s, int mx, int my, int pre ...@@ -546,7 +486,7 @@ static int RENAME(qpel_get_mb_score)(MpegEncContext * s, int mx, int my, int pre
const int index= (((y)<<ME_MAP_SHIFT) + (x))&(ME_MAP_SIZE-1);\ const int index= (((y)<<ME_MAP_SHIFT) + (x))&(ME_MAP_SIZE-1);\
/*printf("check_mv %d %d\n", x, y);*/\ /*printf("check_mv %d %d\n", x, y);*/\
if(map[index]!=key){\ if(map[index]!=key){\
CMP(d, x, y, size);\ d= cmp(s, x, y, 0, 0, size, h, ref_index, src_index, cmpf, chroma_cmpf, flags);\
map[index]= key;\ map[index]= key;\
score_map[index]= d;\ score_map[index]= d;\
d += (mv_penalty[((x)<<shift)-pred_x] + mv_penalty[((y)<<shift)-pred_y])*penalty_factor;\ d += (mv_penalty[((x)<<shift)-pred_x] + mv_penalty[((y)<<shift)-pred_y])*penalty_factor;\
...@@ -570,7 +510,7 @@ static int RENAME(qpel_get_mb_score)(MpegEncContext * s, int mx, int my, int pre ...@@ -570,7 +510,7 @@ static int RENAME(qpel_get_mb_score)(MpegEncContext * s, int mx, int my, int pre
const int index= (((y)<<ME_MAP_SHIFT) + (x))&(ME_MAP_SIZE-1);\ const int index= (((y)<<ME_MAP_SHIFT) + (x))&(ME_MAP_SIZE-1);\
/*printf("check_mv_dir %d %d %d\n", x, y, new_dir);*/\ /*printf("check_mv_dir %d %d %d\n", x, y, new_dir);*/\
if(map[index]!=key){\ if(map[index]!=key){\
CMP(d, x, y, size);\ d= cmp(s, x, y, 0, 0, size, h, ref_index, src_index, cmpf, chroma_cmpf, flags);\
map[index]= key;\ map[index]= key;\
score_map[index]= d;\ score_map[index]= d;\
d += (mv_penalty[((x)<<shift)-pred_x] + mv_penalty[((y)<<shift)-pred_y])*penalty_factor;\ d += (mv_penalty[((x)<<shift)-pred_x] + mv_penalty[((y)<<shift)-pred_y])*penalty_factor;\
...@@ -590,27 +530,29 @@ if( (x)>(xmax<<(S)) ) printf("%d %d %d %d %d xmax" #v, xmax, (x), (y), s->mb_x, ...@@ -590,27 +530,29 @@ if( (x)>(xmax<<(S)) ) printf("%d %d %d %d %d xmax" #v, xmax, (x), (y), s->mb_x,
if( (y)<(ymin<<(S)) ) printf("%d %d %d %d %d ymin" #v, ymin, (x), (y), s->mb_x, s->mb_y);\ if( (y)<(ymin<<(S)) ) printf("%d %d %d %d %d ymin" #v, ymin, (x), (y), s->mb_x, s->mb_y);\
if( (y)>(ymax<<(S)) ) printf("%d %d %d %d %d ymax" #v, ymax, (x), (y), s->mb_x, s->mb_y);\ if( (y)>(ymax<<(S)) ) printf("%d %d %d %d %d ymax" #v, ymax, (x), (y), s->mb_x, s->mb_y);\
#define LOAD_COMMON2\
uint32_t *map= s->me.map;\
const int qpel= flags&FLAG_QPEL;\
const int shift= 1+qpel;\
static inline int RENAME(small_diamond_search)(MpegEncContext * s, int *best, int dmin, static always_inline int small_diamond_search(MpegEncContext * s, int *best, int dmin,
uint8_t *src_data[3], int src_index, int ref_index, int const penalty_factor,
uint8_t *ref_data[3], int stride, int uvstride, int size, int h, int flags)
int const pred_x, int const pred_y, int const penalty_factor,
int const shift,
uint32_t *map, int map_generation, int size, int h, uint8_t * const mv_penalty
)
{ {
me_cmp_func cmp, chroma_cmp; me_cmp_func cmpf, chroma_cmpf;
int next_dir=-1; int next_dir=-1;
LOAD_COMMON LOAD_COMMON
LOAD_COMMON2
int map_generation= s->me.map_generation;
cmp= s->dsp.me_cmp[size]; cmpf= s->dsp.me_cmp[size];
chroma_cmp= s->dsp.me_cmp[size+1]; chroma_cmpf= s->dsp.me_cmp[size+1];
{ /* ensure that the best point is in the MAP as h/qpel refinement needs it */ { /* ensure that the best point is in the MAP as h/qpel refinement needs it */
const int key= (best[1]<<ME_MAP_MV_BITS) + best[0] + map_generation; const int key= (best[1]<<ME_MAP_MV_BITS) + best[0] + map_generation;
const int index= ((best[1]<<ME_MAP_SHIFT) + best[0])&(ME_MAP_SIZE-1); const int index= ((best[1]<<ME_MAP_SHIFT) + best[0])&(ME_MAP_SIZE-1);
if(map[index]!=key){ //this will be executed only very rarey if(map[index]!=key){ //this will be executed only very rarey
CMP(score_map[index], best[0], best[1], size); score_map[index]= cmp(s, best[0], best[1], 0, 0, size, h, ref_index, src_index, cmpf, chroma_cmpf, flags);
map[index]= key; map[index]= key;
} }
} }
...@@ -634,20 +576,18 @@ static inline int RENAME(small_diamond_search)(MpegEncContext * s, int *best, in ...@@ -634,20 +576,18 @@ static inline int RENAME(small_diamond_search)(MpegEncContext * s, int *best, in
} }
} }
static inline int RENAME(funny_diamond_search)(MpegEncContext * s, int *best, int dmin, static int funny_diamond_search(MpegEncContext * s, int *best, int dmin,
uint8_t *src_data[3], int src_index, int ref_index, int const penalty_factor,
uint8_t *ref_data[3], int stride, int uvstride, int size, int h, int flags)
int const pred_x, int const pred_y, int const penalty_factor,
int const shift,
uint32_t *map, int map_generation, int size, int h, uint8_t * const mv_penalty
)
{ {
me_cmp_func cmp, chroma_cmp; me_cmp_func cmpf, chroma_cmpf;
int dia_size; int dia_size;
LOAD_COMMON LOAD_COMMON
LOAD_COMMON2
int map_generation= s->me.map_generation;
cmp= s->dsp.me_cmp[size]; cmpf= s->dsp.me_cmp[size];
chroma_cmp= s->dsp.me_cmp[size+1]; chroma_cmpf= s->dsp.me_cmp[size+1];
for(dia_size=1; dia_size<=4; dia_size++){ for(dia_size=1; dia_size<=4; dia_size++){
int dir; int dir;
...@@ -702,7 +642,7 @@ if(256*256*256*64 % (stats[0]+1)==0){ ...@@ -702,7 +642,7 @@ if(256*256*256*64 % (stats[0]+1)==0){
const int index= (((ay)<<ME_MAP_SHIFT) + (ax))&(ME_MAP_SIZE-1);\ const int index= (((ay)<<ME_MAP_SHIFT) + (ax))&(ME_MAP_SIZE-1);\
/*printf("sab check %d %d\n", ax, ay);*/\ /*printf("sab check %d %d\n", ax, ay);*/\
if(map[index]!=key){\ if(map[index]!=key){\
CMP(d, ax, ay, size);\ d= cmp(s, ax, ay, 0, 0, size, h, ref_index, src_index, cmpf, chroma_cmpf, flags);\
map[index]= key;\ map[index]= key;\
score_map[index]= d;\ score_map[index]= d;\
d += (mv_penalty[((ax)<<shift)-pred_x] + mv_penalty[((ay)<<shift)-pred_y])*penalty_factor;\ d += (mv_penalty[((ax)<<shift)-pred_x] + mv_penalty[((ay)<<shift)-pred_y])*penalty_factor;\
...@@ -726,22 +666,20 @@ if(256*256*256*64 % (stats[0]+1)==0){ ...@@ -726,22 +666,20 @@ if(256*256*256*64 % (stats[0]+1)==0){
} }
#define MAX_SAB_SIZE 16 #define MAX_SAB_SIZE 16
static inline int RENAME(sab_diamond_search)(MpegEncContext * s, int *best, int dmin, static int sab_diamond_search(MpegEncContext * s, int *best, int dmin,
uint8_t *src_data[3], int src_index, int ref_index, int const penalty_factor,
uint8_t *ref_data[3], int stride, int uvstride, int size, int h, int flags)
int const pred_x, int const pred_y, int const penalty_factor,
int const shift,
uint32_t *map, int map_generation, int size, int h, uint8_t * const mv_penalty
)
{ {
me_cmp_func cmp, chroma_cmp; me_cmp_func cmpf, chroma_cmpf;
Minima minima[MAX_SAB_SIZE]; Minima minima[MAX_SAB_SIZE];
const int minima_count= ABS(s->me.dia_size); const int minima_count= ABS(s->me.dia_size);
int i, j; int i, j;
LOAD_COMMON LOAD_COMMON
LOAD_COMMON2
int map_generation= s->me.map_generation;
cmp= s->dsp.me_cmp[size]; cmpf= s->dsp.me_cmp[size];
chroma_cmp= s->dsp.me_cmp[size+1]; chroma_cmpf= s->dsp.me_cmp[size+1];
for(j=i=0; i<ME_MAP_SIZE; i++){ for(j=i=0; i<ME_MAP_SIZE; i++){
uint32_t key= map[i]; uint32_t key= map[i];
...@@ -807,20 +745,18 @@ static inline int RENAME(sab_diamond_search)(MpegEncContext * s, int *best, int ...@@ -807,20 +745,18 @@ static inline int RENAME(sab_diamond_search)(MpegEncContext * s, int *best, int
return dmin; return dmin;
} }
static inline int RENAME(var_diamond_search)(MpegEncContext * s, int *best, int dmin, static int var_diamond_search(MpegEncContext * s, int *best, int dmin,
uint8_t *src_data[3], int src_index, int ref_index, int const penalty_factor,
uint8_t *ref_data[3], int stride, int uvstride, int size, int h, int flags)
int const pred_x, int const pred_y, int const penalty_factor,
int const shift,
uint32_t *map, int map_generation, int size, int h, uint8_t * const mv_penalty
)
{ {
me_cmp_func cmp, chroma_cmp; me_cmp_func cmpf, chroma_cmpf;
int dia_size; int dia_size;
LOAD_COMMON LOAD_COMMON
LOAD_COMMON2
int map_generation= s->me.map_generation;
cmp= s->dsp.me_cmp[size]; cmpf= s->dsp.me_cmp[size];
chroma_cmp= s->dsp.me_cmp[size+1]; chroma_cmpf= s->dsp.me_cmp[size+1];
for(dia_size=1; dia_size<=s->me.dia_size; dia_size++){ for(dia_size=1; dia_size<=s->me.dia_size; dia_size++){
int dir, start, end; int dir, start, end;
...@@ -885,31 +821,42 @@ if(256*256*256*64 % (stats[0]+1)==0){ ...@@ -885,31 +821,42 @@ if(256*256*256*64 % (stats[0]+1)==0){
return dmin; return dmin;
} }
static int RENAME(epzs_motion_search)(MpegEncContext * s, static always_inline int diamond_search(MpegEncContext * s, int *best, int dmin,
int *mx_ptr, int *my_ptr, int src_index, int ref_index, int const penalty_factor,
int P[10][2], int pred_x, int pred_y, uint8_t *src_data[3], int size, int h, int flags){
uint8_t *ref_data[3], int stride, int uvstride, int16_t (*last_mv)[2], if(s->me.dia_size==-1)
int ref_mv_scale, uint8_t * const mv_penalty) return funny_diamond_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags);
else if(s->me.dia_size<-1)
return sab_diamond_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags);
else if(s->me.dia_size<2)
return small_diamond_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags);
else
return var_diamond_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags);
}
static always_inline int epzs_motion_search_internal(MpegEncContext * s, int *mx_ptr, int *my_ptr,
int P[10][2], int src_index, int ref_index, int16_t (*last_mv)[2],
int ref_mv_scale, int flags)
{ {
int best[2]={0, 0}; int best[2]={0, 0};
int d, dmin; int d, dmin;
const int shift= 1+s->quarter_sample;
uint32_t *map= s->me.map;
int map_generation; int map_generation;
const int penalty_factor= s->me.penalty_factor; const int penalty_factor= s->me.penalty_factor;
const int size=0; const int size=0;
const int h=16; const int h=16;
const int ref_mv_stride= s->mb_stride; //pass as arg FIXME const int ref_mv_stride= s->mb_stride; //pass as arg FIXME
const int ref_mv_xy= s->mb_x + s->mb_y*ref_mv_stride; //add to last_mv beforepassing FIXME const int ref_mv_xy= s->mb_x + s->mb_y*ref_mv_stride; //add to last_mv beforepassing FIXME
me_cmp_func cmp, chroma_cmp; me_cmp_func cmpf, chroma_cmpf;
LOAD_COMMON LOAD_COMMON
LOAD_COMMON2
cmp= s->dsp.me_cmp[size]; cmpf= s->dsp.me_cmp[size];
chroma_cmp= s->dsp.me_cmp[size+1]; chroma_cmpf= s->dsp.me_cmp[size+1];
map_generation= update_map_generation(s); map_generation= update_map_generation(s);
CMP(dmin, 0, 0, size); dmin= cmp(s, 0, 0, 0, 0, size, h, ref_index, src_index, cmpf, chroma_cmpf, flags);
map[0]= map_generation; map[0]= map_generation;
score_map[0]= dmin; score_map[0]= dmin;
...@@ -974,22 +921,7 @@ static int RENAME(epzs_motion_search)(MpegEncContext * s, ...@@ -974,22 +921,7 @@ static int RENAME(epzs_motion_search)(MpegEncContext * s,
} }
//check(best[0],best[1],0, b0) //check(best[0],best[1],0, b0)
if(s->me.dia_size==-1) dmin= diamond_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags);
dmin= RENAME(funny_diamond_search)(s, best, dmin, src_data, ref_data, stride, uvstride,
pred_x, pred_y, penalty_factor,
shift, map, map_generation, size, h, mv_penalty);
else if(s->me.dia_size<-1)
dmin= RENAME(sab_diamond_search)(s, best, dmin, src_data, ref_data, stride, uvstride,
pred_x, pred_y, penalty_factor,
shift, map, map_generation, size, h, mv_penalty);
else if(s->me.dia_size<2)
dmin= RENAME(small_diamond_search)(s, best, dmin, src_data, ref_data, stride, uvstride,
pred_x, pred_y, penalty_factor,
shift, map, map_generation, size, h, mv_penalty);
else
dmin= RENAME(var_diamond_search)(s, best, dmin, src_data, ref_data, stride, uvstride,
pred_x, pred_y, penalty_factor,
shift, map, map_generation, size, h, mv_penalty);
//check(best[0],best[1],0, b1) //check(best[0],best[1],0, b1)
*mx_ptr= best[0]; *mx_ptr= best[0];
...@@ -999,29 +931,42 @@ static int RENAME(epzs_motion_search)(MpegEncContext * s, ...@@ -999,29 +931,42 @@ static int RENAME(epzs_motion_search)(MpegEncContext * s,
return dmin; return dmin;
} }
#ifndef CMP_DIRECT /* no 4mv search needed in direct mode */ //this function is dedicated to the braindamaged gcc
static int RENAME(epzs_motion_search4)(MpegEncContext * s, static inline int epzs_motion_search(MpegEncContext * s, int *mx_ptr, int *my_ptr,
int *mx_ptr, int *my_ptr, int P[10][2], int src_index, int ref_index, int16_t (*last_mv)[2],
int P[10][2], int pred_x, int pred_y, int ref_mv_scale)
uint8_t *src_data[3], {
uint8_t *ref_data[3], int stride, int uvstride, int16_t (*last_mv)[2], //FIXME convert other functions in the same way if faster
int ref_mv_scale, uint8_t * const mv_penalty) switch(s->me.flags){
case 0:
return epzs_motion_search_internal(s, mx_ptr, my_ptr, P, src_index, ref_index, last_mv, ref_mv_scale, 0);
// case FLAG_QPEL:
// return epzs_motion_search_internal(s, mx_ptr, my_ptr, P, src_index, ref_index, last_mv, ref_mv_scale, FLAG_QPEL);
default:
return epzs_motion_search_internal(s, mx_ptr, my_ptr, P, src_index, ref_index, last_mv, ref_mv_scale, s->me.flags);
}
}
static int epzs_motion_search4(MpegEncContext * s,
int *mx_ptr, int *my_ptr, int P[10][2],
int src_index, int ref_index, int16_t (*last_mv)[2],
int ref_mv_scale)
{ {
int best[2]={0, 0}; int best[2]={0, 0};
int d, dmin; int d, dmin;
const int shift= 1+s->quarter_sample;
uint32_t *map= s->me.map;
int map_generation; int map_generation;
const int penalty_factor= s->me.penalty_factor; const int penalty_factor= s->me.penalty_factor;
const int size=1; const int size=1;
const int h=8; const int h=8;
const int ref_mv_stride= s->mb_stride; const int ref_mv_stride= s->mb_stride;
const int ref_mv_xy= s->mb_x + s->mb_y *ref_mv_stride; const int ref_mv_xy= s->mb_x + s->mb_y *ref_mv_stride;
me_cmp_func cmp, chroma_cmp; me_cmp_func cmpf, chroma_cmpf;
LOAD_COMMON LOAD_COMMON
int flags= s->me.flags;
LOAD_COMMON2
cmp= s->dsp.me_cmp[size]; cmpf= s->dsp.me_cmp[size];
chroma_cmp= s->dsp.me_cmp[size+1]; chroma_cmpf= s->dsp.me_cmp[size+1];
map_generation= update_map_generation(s); map_generation= update_map_generation(s);
...@@ -1053,23 +998,7 @@ static int RENAME(epzs_motion_search4)(MpegEncContext * s, ...@@ -1053,23 +998,7 @@ static int RENAME(epzs_motion_search4)(MpegEncContext * s,
(last_mv[ref_mv_xy+ref_mv_stride][1]*ref_mv_scale + (1<<15))>>16) (last_mv[ref_mv_xy+ref_mv_stride][1]*ref_mv_scale + (1<<15))>>16)
} }
if(s->me.dia_size==-1) dmin= diamond_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags);
dmin= RENAME(funny_diamond_search)(s, best, dmin, src_data, ref_data, stride, uvstride,
pred_x, pred_y, penalty_factor,
shift, map, map_generation, size, h, mv_penalty);
else if(s->me.dia_size<-1)
dmin= RENAME(sab_diamond_search)(s, best, dmin, src_data, ref_data, stride, uvstride,
pred_x, pred_y, penalty_factor,
shift, map, map_generation, size, h, mv_penalty);
else if(s->me.dia_size<2)
dmin= RENAME(small_diamond_search)(s, best, dmin, src_data, ref_data, stride, uvstride,
pred_x, pred_y, penalty_factor,
shift, map, map_generation, size, h, mv_penalty);
else
dmin= RENAME(var_diamond_search)(s, best, dmin, src_data, ref_data, stride, uvstride,
pred_x, pred_y, penalty_factor,
shift, map, map_generation, size, h, mv_penalty);
*mx_ptr= best[0]; *mx_ptr= best[0];
*my_ptr= best[1]; *my_ptr= best[1];
...@@ -1079,28 +1008,26 @@ static int RENAME(epzs_motion_search4)(MpegEncContext * s, ...@@ -1079,28 +1008,26 @@ static int RENAME(epzs_motion_search4)(MpegEncContext * s,
} }
//try to merge with above FIXME (needs PSNR test) //try to merge with above FIXME (needs PSNR test)
static int RENAME(epzs_motion_search2)(MpegEncContext * s, static int epzs_motion_search2(MpegEncContext * s,
int *mx_ptr, int *my_ptr, int *mx_ptr, int *my_ptr, int P[10][2],
int P[10][2], int pred_x, int pred_y, int src_index, int ref_index, int16_t (*last_mv)[2],
uint8_t *src_data[3], int ref_mv_scale)
uint8_t *ref_data[3], int stride, int uvstride, int16_t (*last_mv)[2],
int ref_mv_scale, uint8_t * const mv_penalty)
{ {
int best[2]={0, 0}; int best[2]={0, 0};
int d, dmin; int d, dmin;
const int shift= 1+s->quarter_sample;
uint32_t *map= s->me.map;
int map_generation; int map_generation;
const int penalty_factor= s->me.penalty_factor; const int penalty_factor= s->me.penalty_factor;
const int size=0; //FIXME pass as arg const int size=0; //FIXME pass as arg
const int h=8; const int h=8;
const int ref_mv_stride= s->mb_stride; const int ref_mv_stride= s->mb_stride;
const int ref_mv_xy= s->mb_x + s->mb_y *ref_mv_stride; const int ref_mv_xy= s->mb_x + s->mb_y *ref_mv_stride;
me_cmp_func cmp, chroma_cmp; me_cmp_func cmpf, chroma_cmpf;
LOAD_COMMON LOAD_COMMON
int flags= s->me.flags;
LOAD_COMMON2
cmp= s->dsp.me_cmp[size]; cmpf= s->dsp.me_cmp[size];
chroma_cmp= s->dsp.me_cmp[size+1]; chroma_cmpf= s->dsp.me_cmp[size+1];
map_generation= update_map_generation(s); map_generation= update_map_generation(s);
...@@ -1132,23 +1059,7 @@ static int RENAME(epzs_motion_search2)(MpegEncContext * s, ...@@ -1132,23 +1059,7 @@ static int RENAME(epzs_motion_search2)(MpegEncContext * s,
(last_mv[ref_mv_xy+ref_mv_stride][1]*ref_mv_scale + (1<<15))>>16) (last_mv[ref_mv_xy+ref_mv_stride][1]*ref_mv_scale + (1<<15))>>16)
} }
if(s->me.dia_size==-1) dmin= diamond_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags);
dmin= RENAME(funny_diamond_search)(s, best, dmin, src_data, ref_data, stride, uvstride,
pred_x, pred_y, penalty_factor,
shift, map, map_generation, size, h, mv_penalty);
else if(s->me.dia_size<-1)
dmin= RENAME(sab_diamond_search)(s, best, dmin, src_data, ref_data, stride, uvstride,
pred_x, pred_y, penalty_factor,
shift, map, map_generation, size, h, mv_penalty);
else if(s->me.dia_size<2)
dmin= RENAME(small_diamond_search)(s, best, dmin, src_data, ref_data, stride, uvstride,
pred_x, pred_y, penalty_factor,
shift, map, map_generation, size, h, mv_penalty);
else
dmin= RENAME(var_diamond_search)(s, best, dmin, src_data, ref_data, stride, uvstride,
pred_x, pred_y, penalty_factor,
shift, map, map_generation, size, h, mv_penalty);
*mx_ptr= best[0]; *mx_ptr= best[0];
*my_ptr= best[1]; *my_ptr= best[1];
...@@ -1156,4 +1067,3 @@ static int RENAME(epzs_motion_search2)(MpegEncContext * s, ...@@ -1156,4 +1067,3 @@ static int RENAME(epzs_motion_search2)(MpegEncContext * s,
// printf("%d %d %d \n", best[0], best[1], dmin); // printf("%d %d %d \n", best[0], best[1], dmin);
return dmin; return dmin;
} }
#endif /* !CMP_DIRECT */
libavcodec/mpegvideo.h
View file @ 2750b827
...@@ -200,6 +200,10 @@ typedef struct MotionEstContext{ ...@@ -200,6 +200,10 @@ typedef struct MotionEstContext{
int co_located_mv[4][2]; ///< mv from last p frame for direct mode ME int co_located_mv[4][2]; ///< mv from last p frame for direct mode ME
int direct_basis_mv[4][2]; int direct_basis_mv[4][2];
uint8_t *scratchpad; ///< data area for the me algo, so that the ME doesnt need to malloc/free uint8_t *scratchpad; ///< data area for the me algo, so that the ME doesnt need to malloc/free
uint8_t *best_mb;
uint8_t *temp_mb[2];
uint8_t *temp;
int best_bits;
uint32_t *map; ///< map to avoid duplicate evaluations uint32_t *map; ///< map to avoid duplicate evaluations
uint32_t *score_map; ///< map to store the scores uint32_t *score_map; ///< map to store the scores
int map_generation; int map_generation;
...@@ -207,31 +211,33 @@ typedef struct MotionEstContext{ ...@@ -207,31 +211,33 @@ typedef struct MotionEstContext{
int penalty_factor; int penalty_factor;
int sub_penalty_factor; int sub_penalty_factor;
int mb_penalty_factor; int mb_penalty_factor;
int flags;
int sub_flags;
int mb_flags;
int pre_pass; ///< = 1 for the pre pass int pre_pass; ///< = 1 for the pre pass
int dia_size; int dia_size;
int xmin; int xmin;
int xmax; int xmax;
int ymin; int ymin;
int ymax; int ymax;
int pred_x;
int pred_y;
uint8_t *src[4][4];
uint8_t *ref[4][4];
int stride;
int uvstride;
/* cmp, chroma_cmp;*/
op_pixels_func (*hpel_put)[4];
op_pixels_func (*hpel_avg)[4];
op_pixels_func (*chroma_hpel_put)[4];
qpel_mc_func (*qpel_put)[16];
qpel_mc_func (*qpel_avg)[16];
uint8_t (*mv_penalty)[MAX_MV*2+1]; ///< amount of bits needed to encode a MV uint8_t (*mv_penalty)[MAX_MV*2+1]; ///< amount of bits needed to encode a MV
uint8_t *current_mv_penalty;
int (*sub_motion_search)(struct MpegEncContext * s, int (*sub_motion_search)(struct MpegEncContext * s,
int *mx_ptr, int *my_ptr, int dmin, int *mx_ptr, int *my_ptr, int dmin,
int pred_x, int pred_y, uint8_t *src_data[3], int src_index, int ref_index,
uint8_t *ref_data[6], int stride, int uvstride, int size, int h);
int size, int h, uint8_t * const mv_penalty);
int (*motion_search[7])(struct MpegEncContext * s,
int *mx_ptr, int *my_ptr,
int P[10][2], int pred_x, int pred_y, uint8_t *src_data[3],
uint8_t *ref_data[6], int stride, int uvstride, int16_t (*last_mv)[2],
int ref_mv_scale, uint8_t * const mv_penalty);
int (*pre_motion_search)(struct MpegEncContext * s,
int *mx_ptr, int *my_ptr,
int P[10][2], int pred_x, int pred_y, uint8_t *src_data[3],
uint8_t *ref_data[6], int stride, int uvstride, int16_t (*last_mv)[2],
int ref_mv_scale, uint8_t * const mv_penalty);
int (*get_mb_score)(struct MpegEncContext * s, int mx, int my, int pred_x, int pred_y, uint8_t *src_data[3],
uint8_t *ref_data[6], int stride, int uvstride,
uint8_t * const mv_penalty);
}MotionEstContext; }MotionEstContext;
/** /**
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment