Skip to content

F
ffmpeg.wasm-core
Commit af4091f1 authored by Michael Niedermayer's avatar Michael Niedermayer
Browse files
Options

reduce dependancy between motion estimation and MpegEncContext 

this is practically just a s/s->me./c->/

Originally committed as revision 3123 to svn://svn.ffmpeg.org/ffmpeg/trunk
parent ca8ad847
Hide whitespace changes
Inline Side-by-side
Showing with 226 additions and 210 deletions
libavcodec/motion_est.c
View file @ af4091f1
...@@ -49,14 +49,14 @@ static inline int sad_hpel_motion_search(MpegEncContext * s, ...@@ -49,14 +49,14 @@ static inline int sad_hpel_motion_search(MpegEncContext * s,
int src_index, int ref_index, int src_index, int ref_index,
int size, int h); int size, int h);
static inline int update_map_generation(MpegEncContext * s) static inline int update_map_generation(MotionEstContext *c)
{ {
s->me.map_generation+= 1<<(ME_MAP_MV_BITS*2); c->map_generation+= 1<<(ME_MAP_MV_BITS*2);
if(s->me.map_generation==0){ if(c->map_generation==0){
s->me.map_generation= 1<<(ME_MAP_MV_BITS*2); c->map_generation= 1<<(ME_MAP_MV_BITS*2);
memset(s->me.map, 0, sizeof(uint32_t)*ME_MAP_SIZE); memset(c->map, 0, sizeof(uint32_t)*ME_MAP_SIZE);
} }
return s->me.map_generation; return c->map_generation;
} }
/* shape adaptive search stuff */ /* shape adaptive search stuff */
...@@ -77,8 +77,7 @@ static int minima_cmp(const void *a, const void *b){ ...@@ -77,8 +77,7 @@ static int minima_cmp(const void *a, const void *b){
#define FLAG_CHROMA 2 #define FLAG_CHROMA 2
#define FLAG_DIRECT 4 #define FLAG_DIRECT 4
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){ static inline void init_ref(MotionEstContext *c, uint8_t *src[3], uint8_t *ref[3], uint8_t *ref2[3], int x, int y, int ref_index){
MotionEstContext * const c= &s->me;
const int offset[3]= { const int offset[3]= {
y*c-> stride + x, y*c-> stride + x,
((y*c->uvstride + x)>>1), ((y*c->uvstride + x)>>1),
...@@ -247,35 +246,35 @@ void ff_init_me(MpegEncContext *s){ ...@@ -247,35 +246,35 @@ 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); c->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); c->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); c->mb_flags = get_flags(s, 0, s->avctx->mb_cmp &FF_CMP_CHROMA);
/*FIXME s->no_rounding b_type*/ /*FIXME s->no_rounding b_type*/
if(s->flags&CODEC_FLAG_QPEL){ if(s->flags&CODEC_FLAG_QPEL){
s->me.sub_motion_search= qpel_motion_search; c->sub_motion_search= qpel_motion_search;
c->qpel_avg= s->dsp.avg_qpel_pixels_tab; c->qpel_avg= s->dsp.avg_qpel_pixels_tab;
if(s->no_rounding) c->qpel_put= s->dsp.put_no_rnd_qpel_pixels_tab; if(s->no_rounding) c->qpel_put= s->dsp.put_no_rnd_qpel_pixels_tab;
else c->qpel_put= s->dsp.put_qpel_pixels_tab; else c->qpel_put= s->dsp.put_qpel_pixels_tab;
}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= hpel_motion_search; c->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 c->sub_motion_search= sad_hpel_motion_search; // 2050 vs. 2450 cycles
else else
s->me.sub_motion_search= hpel_motion_search; c->sub_motion_search= hpel_motion_search;
c->hpel_avg= s->dsp.avg_pixels_tab; c->hpel_avg= s->dsp.avg_pixels_tab;
if(s->no_rounding) c->hpel_put= s->dsp.put_no_rnd_pixels_tab; if(s->no_rounding) c->hpel_put= s->dsp.put_no_rnd_pixels_tab;
else c->hpel_put= s->dsp.put_pixels_tab; else c->hpel_put= s->dsp.put_pixels_tab;
} }
if(s->linesize){ if(s->linesize){
s->me.stride = s->linesize; c->stride = s->linesize;
s->me.uvstride= s->uvlinesize; c->uvstride= s->uvlinesize;
}else{ }else{
s->me.stride = 16*s->mb_width + 32; c->stride = 16*s->mb_width + 32;
s->me.uvstride= 8*s->mb_width + 16; c->uvstride= 8*s->mb_width + 16;
} }
c->temp= c->scratchpad; c->temp= c->scratchpad;
...@@ -548,16 +547,17 @@ static inline int sad_hpel_motion_search(MpegEncContext * s, ...@@ -548,16 +547,17 @@ static inline int sad_hpel_motion_search(MpegEncContext * s,
int src_index, int ref_index, int src_index, int ref_index,
int size, int h) int size, int h)
{ {
const int penalty_factor= s->me.sub_penalty_factor; MotionEstContext * const c= &s->me;
const int penalty_factor= c->sub_penalty_factor;
int mx, my, dminh; int mx, my, dminh;
uint8_t *pix, *ptr; uint8_t *pix, *ptr;
int stride= s->me.stride; int stride= c->stride;
const int flags= s->me.sub_flags; const int flags= c->sub_flags;
LOAD_COMMON LOAD_COMMON
assert(flags == 0); assert(flags == 0);
if(s->me.skip){ if(c->skip){
// printf("S"); // printf("S");
*mx_ptr = 0; *mx_ptr = 0;
*my_ptr = 0; *my_ptr = 0;
...@@ -565,11 +565,11 @@ static inline int sad_hpel_motion_search(MpegEncContext * s, ...@@ -565,11 +565,11 @@ static inline int sad_hpel_motion_search(MpegEncContext * s,
} }
// printf("N"); // printf("N");
pix = s->me.src[src_index][0]; pix = c->src[src_index][0];
mx = *mx_ptr; mx = *mx_ptr;
my = *my_ptr; my = *my_ptr;
ptr = s->me.ref[ref_index][0] + (my * stride) + mx; ptr = c->ref[ref_index][0] + (my * stride) + mx;
dminh = dmin; dminh = dmin;
...@@ -679,20 +679,21 @@ static inline void set_p_mv_tables(MpegEncContext * s, int mx, int my, int mv4) ...@@ -679,20 +679,21 @@ static inline void set_p_mv_tables(MpegEncContext * s, int mx, int my, int mv4)
*/ */
static inline void get_limits(MpegEncContext *s, int x, int y) static inline void get_limits(MpegEncContext *s, int x, int y)
{ {
MotionEstContext * const c= &s->me;
/* /*
if(s->avctx->me_range) s->me.range= s->avctx->me_range >> 1; if(s->avctx->me_range) c->range= s->avctx->me_range >> 1;
else s->me.range= 16; else c->range= 16;
*/ */
if (s->unrestricted_mv) { if (s->unrestricted_mv) {
s->me.xmin = - x - 16; c->xmin = - x - 16;
s->me.ymin = - y - 16; c->ymin = - y - 16;
s->me.xmax = - x + s->mb_width *16; c->xmax = - x + s->mb_width *16;
s->me.ymax = - y + s->mb_height*16; c->ymax = - y + s->mb_height*16;
} else { } else {
s->me.xmin = - x; c->xmin = - x;
s->me.ymin = - y; c->ymin = - y;
s->me.xmax = - x + s->mb_width *16 - 16; c->xmax = - x + s->mb_width *16 - 16;
s->me.ymax = - y + s->mb_height*16 - 16; c->ymax = - y + s->mb_height*16 - 16;
} }
} }
...@@ -719,7 +720,7 @@ static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift) ...@@ -719,7 +720,7 @@ static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift)
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; uint8_t *mv_penalty= c->current_mv_penalty;
init_mv4_ref(s); init_mv4_ref(s);
...@@ -734,39 +735,39 @@ static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift) ...@@ -734,39 +735,39 @@ static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift)
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];
if(P_LEFT[0] > (s->me.xmax<<shift)) P_LEFT[0] = (s->me.xmax<<shift); if(P_LEFT[0] > (c->xmax<<shift)) P_LEFT[0] = (c->xmax<<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) {
s->me.pred_x= pred_x4= P_LEFT[0]; c->pred_x= pred_x4= P_LEFT[0];
s->me.pred_y= pred_y4= P_LEFT[1]; c->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];
P_TOPRIGHT[0] = s->current_picture.motion_val[0][mot_xy - mot_stride + off[block]][0]; P_TOPRIGHT[0] = s->current_picture.motion_val[0][mot_xy - mot_stride + off[block]][0];
P_TOPRIGHT[1] = s->current_picture.motion_val[0][mot_xy - mot_stride + off[block]][1]; P_TOPRIGHT[1] = s->current_picture.motion_val[0][mot_xy - mot_stride + off[block]][1];
if(P_TOP[1] > (s->me.ymax<<shift)) P_TOP[1] = (s->me.ymax<<shift); if(P_TOP[1] > (c->ymax<<shift)) P_TOP[1] = (c->ymax<<shift);
if(P_TOPRIGHT[0] < (s->me.xmin<<shift)) P_TOPRIGHT[0]= (s->me.xmin<<shift); if(P_TOPRIGHT[0] < (c->xmin<<shift)) P_TOPRIGHT[0]= (c->xmin<<shift);
if(P_TOPRIGHT[0] > (s->me.xmax<<shift)) P_TOPRIGHT[0]= (s->me.xmax<<shift); if(P_TOPRIGHT[0] > (c->xmax<<shift)) P_TOPRIGHT[0]= (c->xmax<<shift);
if(P_TOPRIGHT[1] > (s->me.ymax<<shift)) P_TOPRIGHT[1]= (s->me.ymax<<shift); if(P_TOPRIGHT[1] > (c->ymax<<shift)) P_TOPRIGHT[1]= (c->ymax<<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]);
s->me.pred_x= pred_x4 = P_MEDIAN[0]; c->pred_x= pred_x4 = P_MEDIAN[0];
s->me.pred_y= pred_y4 = P_MEDIAN[1]; c->pred_y= pred_y4 = P_MEDIAN[1];
} }
P_MV1[0]= mx; P_MV1[0]= mx;
P_MV1[1]= my; P_MV1[1]= my;
dmin4 = epzs_motion_search4(s, &mx4, &my4, P, block, block, s->p_mv_table, (1<<16)>>shift); dmin4 = epzs_motion_search4(s, &mx4, &my4, P, block, block, s->p_mv_table, (1<<16)>>shift);
dmin4= s->me.sub_motion_search(s, &mx4, &my4, dmin4, block, block, size, h); dmin4= c->sub_motion_search(s, &mx4, &my4, dmin4, block, block, size, h);
if(s->dsp.me_sub_cmp[0] != s->dsp.mb_cmp[0]){ if(s->dsp.me_sub_cmp[0] != s->dsp.mb_cmp[0]){
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 = c->scratchpad + offset;
if(s->quarter_sample){ if(s->quarter_sample){
uint8_t *ref= c->ref[block][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);
...@@ -784,7 +785,7 @@ static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift) ...@@ -784,7 +785,7 @@ static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift)
else else
s->dsp.put_pixels_tab [1][dxy](dest_y , ref , stride, h); s->dsp.put_pixels_tab [1][dxy](dest_y , ref , stride, h);
} }
dmin_sum+= (mv_penalty[mx4-pred_x4] + mv_penalty[my4-pred_y4])*s->me.mb_penalty_factor; dmin_sum+= (mv_penalty[mx4-pred_x4] + mv_penalty[my4-pred_y4])*c->mb_penalty_factor;
}else }else
dmin_sum+= dmin4; dmin_sum+= dmin4;
...@@ -806,7 +807,7 @@ static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift) ...@@ -806,7 +807,7 @@ static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift)
return INT_MAX; return INT_MAX;
if(s->dsp.me_sub_cmp[0] != s->dsp.mb_cmp[0]){ if(s->dsp.me_sub_cmp[0] != s->dsp.mb_cmp[0]){
dmin_sum += s->dsp.mb_cmp[0](s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*16*stride, s->me.scratchpad, stride, 16); dmin_sum += s->dsp.mb_cmp[0](s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*16*stride, c->scratchpad, stride, 16);
} }
if(s->avctx->mb_cmp&FF_CMP_CHROMA){ if(s->avctx->mb_cmp&FF_CMP_CHROMA){
...@@ -821,19 +822,19 @@ static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift) ...@@ -821,19 +822,19 @@ static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift)
offset= (s->mb_x*8 + (mx>>1)) + (s->mb_y*8 + (my>>1))*s->uvlinesize; offset= (s->mb_x*8 + (mx>>1)) + (s->mb_y*8 + (my>>1))*s->uvlinesize;
if(s->no_rounding){ if(s->no_rounding){
s->dsp.put_no_rnd_pixels_tab[1][dxy](s->me.scratchpad , s->last_picture.data[1] + offset, s->uvlinesize, 8); s->dsp.put_no_rnd_pixels_tab[1][dxy](c->scratchpad , s->last_picture.data[1] + offset, s->uvlinesize, 8);
s->dsp.put_no_rnd_pixels_tab[1][dxy](s->me.scratchpad+8 , s->last_picture.data[2] + offset, s->uvlinesize, 8); s->dsp.put_no_rnd_pixels_tab[1][dxy](c->scratchpad+8 , s->last_picture.data[2] + offset, s->uvlinesize, 8);
}else{ }else{
s->dsp.put_pixels_tab [1][dxy](s->me.scratchpad , s->last_picture.data[1] + offset, s->uvlinesize, 8); s->dsp.put_pixels_tab [1][dxy](c->scratchpad , s->last_picture.data[1] + offset, s->uvlinesize, 8);
s->dsp.put_pixels_tab [1][dxy](s->me.scratchpad+8 , s->last_picture.data[2] + offset, s->uvlinesize, 8); s->dsp.put_pixels_tab [1][dxy](c->scratchpad+8 , s->last_picture.data[2] + offset, s->uvlinesize, 8);
} }
dmin_sum += s->dsp.mb_cmp[1](s, s->new_picture.data[1] + s->mb_x*8 + s->mb_y*8*s->uvlinesize, s->me.scratchpad , s->uvlinesize, 8); dmin_sum += s->dsp.mb_cmp[1](s, s->new_picture.data[1] + s->mb_x*8 + s->mb_y*8*s->uvlinesize, c->scratchpad , 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); dmin_sum += s->dsp.mb_cmp[1](s, s->new_picture.data[2] + s->mb_x*8 + s->mb_y*8*s->uvlinesize, c->scratchpad+8, s->uvlinesize, 8);
} }
s->me.pred_x= mx; c->pred_x= mx;
s->me.pred_y= my; c->pred_y= my;
switch(s->avctx->mb_cmp&0xFF){ switch(s->avctx->mb_cmp&0xFF){
/*case FF_CMP_SSE: /*case FF_CMP_SSE:
...@@ -841,7 +842,7 @@ static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift) ...@@ -841,7 +842,7 @@ static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift)
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;
} }
} }
...@@ -898,8 +899,8 @@ static int interlaced_search(MpegEncContext *s, int ref_index, ...@@ -898,8 +899,8 @@ static int interlaced_search(MpegEncContext *s, int ref_index,
P_LEFT[1] = mv_table[xy - 1][1]; P_LEFT[1] = mv_table[xy - 1][1];
if(P_LEFT[0] > (c->xmax<<1)) P_LEFT[0] = (c->xmax<<1); if(P_LEFT[0] > (c->xmax<<1)) P_LEFT[0] = (c->xmax<<1);
s->me.pred_x= P_LEFT[0]; c->pred_x= P_LEFT[0];
s->me.pred_y= P_LEFT[1]; c->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];
...@@ -937,7 +938,7 @@ static int interlaced_search(MpegEncContext *s, int ref_index, ...@@ -937,7 +938,7 @@ static int interlaced_search(MpegEncContext *s, int ref_index,
s->dsp.put_pixels_tab [size][dxy](c->scratchpad, ref , stride, h); s->dsp.put_pixels_tab [size][dxy](c->scratchpad, ref , stride, h);
} }
dmin= s->dsp.mb_cmp[size](s, c->src[block][0], c->scratchpad, stride, h); dmin= s->dsp.mb_cmp[size](s, c->src[block][0], c->scratchpad, stride, h);
dmin+= (mv_penalty[mx_i-s->me.pred_x] + mv_penalty[my_i-s->me.pred_y] + 1)*c->mb_penalty_factor; dmin+= (mv_penalty[mx_i-c->pred_x] + mv_penalty[my_i-c->pred_y] + 1)*c->mb_penalty_factor;
}else }else
dmin+= c->mb_penalty_factor; //field_select bits dmin+= c->mb_penalty_factor; //field_select bits
...@@ -1117,19 +1118,19 @@ void ff_estimate_p_frame_motion(MpegEncContext * s, ...@@ -1117,19 +1118,19 @@ void ff_estimate_p_frame_motion(MpegEncContext * s,
int mb_type=0; int mb_type=0;
Picture * const pic= &s->current_picture; Picture * const pic= &s->current_picture;
init_ref(s, s->new_picture.data, s->last_picture.data, NULL, 16*mb_x, 16*mb_y, 0); init_ref(c, s->new_picture.data, s->last_picture.data, NULL, 16*mb_x, 16*mb_y, 0);
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->linesize == c->stride);
assert(s->uvlinesize == s->me.uvstride); assert(s->uvlinesize == c->uvstride);
s->me.penalty_factor = get_penalty_factor(s, s->avctx->me_cmp); c->penalty_factor = get_penalty_factor(s, s->avctx->me_cmp);
s->me.sub_penalty_factor= get_penalty_factor(s, s->avctx->me_sub_cmp); c->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); c->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; c->current_mv_penalty= c->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; c->skip=0;
/* intra / predictive decision */ /* intra / predictive decision */
pix = c->src[0][0]; pix = c->src[0][0];
...@@ -1191,16 +1192,16 @@ void ff_estimate_p_frame_motion(MpegEncContext * s, ...@@ -1191,16 +1192,16 @@ void ff_estimate_p_frame_motion(MpegEncContext * s,
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];
if(P_LEFT[0] > (s->me.xmax<<shift)) P_LEFT[0] = (s->me.xmax<<shift); if(P_LEFT[0] > (c->xmax<<shift)) P_LEFT[0] = (c->xmax<<shift);
if(!s->first_slice_line) { if(!s->first_slice_line) {
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];
P_TOPRIGHT[0] = s->current_picture.motion_val[0][mot_xy - mot_stride + 2][0]; P_TOPRIGHT[0] = s->current_picture.motion_val[0][mot_xy - mot_stride + 2][0];
P_TOPRIGHT[1] = s->current_picture.motion_val[0][mot_xy - mot_stride + 2][1]; P_TOPRIGHT[1] = s->current_picture.motion_val[0][mot_xy - mot_stride + 2][1];
if(P_TOP[1] > (s->me.ymax<<shift)) P_TOP[1] = (s->me.ymax<<shift); if(P_TOP[1] > (c->ymax<<shift)) P_TOP[1] = (c->ymax<<shift);
if(P_TOPRIGHT[0] < (s->me.xmin<<shift)) P_TOPRIGHT[0]= (s->me.xmin<<shift); if(P_TOPRIGHT[0] < (c->xmin<<shift)) P_TOPRIGHT[0]= (c->xmin<<shift);
if(P_TOPRIGHT[1] > (s->me.ymax<<shift)) P_TOPRIGHT[1]= (s->me.ymax<<shift); if(P_TOPRIGHT[1] > (c->ymax<<shift)) P_TOPRIGHT[1]= (c->ymax<<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]);
...@@ -1243,7 +1244,7 @@ void ff_estimate_p_frame_motion(MpegEncContext * s, ...@@ -1243,7 +1244,7 @@ void ff_estimate_p_frame_motion(MpegEncContext * s,
s->scene_change_score+= s->qscale; s->scene_change_score+= s->qscale;
if(mb_type == CANDIDATE_MB_TYPE_INTER){ if(mb_type == CANDIDATE_MB_TYPE_INTER){
s->me.sub_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16); c->sub_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
set_p_mv_tables(s, mx, my, 1); set_p_mv_tables(s, mx, my, 1);
}else{ }else{
mx <<=shift; mx <<=shift;
...@@ -1267,7 +1268,7 @@ void ff_estimate_p_frame_motion(MpegEncContext * s, ...@@ -1267,7 +1268,7 @@ 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, 0, 0, 0, 16); c->sub_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
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
...@@ -1276,7 +1277,7 @@ void ff_estimate_p_frame_motion(MpegEncContext * s, ...@@ -1276,7 +1277,7 @@ void ff_estimate_p_frame_motion(MpegEncContext * s,
my <<=shift; my <<=shift;
} }
if((s->flags&CODEC_FLAG_4MV) if((s->flags&CODEC_FLAG_4MV)
&& !s->me.skip && varc>50 && vard>10){ && !c->skip && varc>50 && vard>10){
if(h263_mv4_search(s, mx, my, shift) < INT_MAX) if(h263_mv4_search(s, mx, my, shift) < INT_MAX)
mb_type|=CANDIDATE_MB_TYPE_INTER4V; mb_type|=CANDIDATE_MB_TYPE_INTER4V;
...@@ -1284,7 +1285,7 @@ void ff_estimate_p_frame_motion(MpegEncContext * s, ...@@ -1284,7 +1285,7 @@ void ff_estimate_p_frame_motion(MpegEncContext * s,
}else }else
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 && !c->skip){ //FIXME varc/d checks
if(interlaced_search(s, 0, s->p_field_mv_table, s->p_field_select_table, mx, my, 0) < INT_MAX) if(interlaced_search(s, 0, s->p_field_mv_table, s->p_field_select_table, mx, my, 0) < INT_MAX)
mb_type |= CANDIDATE_MB_TYPE_INTER_I; mb_type |= CANDIDATE_MB_TYPE_INTER_I;
} }
...@@ -1292,12 +1293,12 @@ void ff_estimate_p_frame_motion(MpegEncContext * s, ...@@ -1292,12 +1293,12 @@ void ff_estimate_p_frame_motion(MpegEncContext * s,
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, 0, 0, 0, 16); dmin= c->sub_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
if(s->avctx->me_sub_cmp != s->avctx->mb_cmp && !s->me.skip) if(s->avctx->me_sub_cmp != s->avctx->mb_cmp && !c->skip)
dmin= get_mb_score(s, mx, my, 0, 0); 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){ && !c->skip && varc>50 && vard>10){
int dmin4= h263_mv4_search(s, mx, my, shift); int dmin4= h263_mv4_search(s, mx, my, shift);
if(dmin4 < dmin){ if(dmin4 < dmin){
mb_type= CANDIDATE_MB_TYPE_INTER4V; mb_type= CANDIDATE_MB_TYPE_INTER4V;
...@@ -1305,7 +1306,7 @@ void ff_estimate_p_frame_motion(MpegEncContext * s, ...@@ -1305,7 +1306,7 @@ 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 && !c->skip){ //FIXME varc/d checks
int dmin_i= interlaced_search(s, 0, s->p_field_mv_table, s->p_field_select_table, mx, my, 0); int dmin_i= interlaced_search(s, 0, s->p_field_mv_table, s->p_field_select_table, mx, my, 0);
if(dmin_i < dmin){ if(dmin_i < dmin){
mb_type = CANDIDATE_MB_TYPE_INTER_I; mb_type = CANDIDATE_MB_TYPE_INTER_I;
...@@ -1324,13 +1325,13 @@ void ff_estimate_p_frame_motion(MpegEncContext * s, ...@@ -1324,13 +1325,13 @@ void ff_estimate_p_frame_motion(MpegEncContext * s,
mean*= 0x01010101; mean*= 0x01010101;
for(i=0; i<16; i++){ for(i=0; i<16; i++){
*(uint32_t*)(&s->me.scratchpad[i*s->linesize+ 0]) = mean; *(uint32_t*)(&c->scratchpad[i*s->linesize+ 0]) = mean;
*(uint32_t*)(&s->me.scratchpad[i*s->linesize+ 4]) = mean; *(uint32_t*)(&c->scratchpad[i*s->linesize+ 4]) = mean;
*(uint32_t*)(&s->me.scratchpad[i*s->linesize+ 8]) = mean; *(uint32_t*)(&c->scratchpad[i*s->linesize+ 8]) = mean;
*(uint32_t*)(&s->me.scratchpad[i*s->linesize+12]) = mean; *(uint32_t*)(&c->scratchpad[i*s->linesize+12]) = mean;
} }
intra_score= s->dsp.mb_cmp[0](s, s->me.scratchpad, pix, s->linesize, 16); intra_score= s->dsp.mb_cmp[0](s, c->scratchpad, pix, s->linesize, 16);
} }
#if 0 //FIXME #if 0 //FIXME
/* get chroma score */ /* get chroma score */
...@@ -1348,15 +1349,15 @@ void ff_estimate_p_frame_motion(MpegEncContext * s, ...@@ -1348,15 +1349,15 @@ void ff_estimate_p_frame_motion(MpegEncContext * s,
mean*= 0x01010101; mean*= 0x01010101;
for(i=0; i<8; i++){ for(i=0; i<8; i++){
*(uint32_t*)(&s->me.scratchpad[i*s->uvlinesize+ 0]) = mean; *(uint32_t*)(&c->scratchpad[i*s->uvlinesize+ 0]) = mean;
*(uint32_t*)(&s->me.scratchpad[i*s->uvlinesize+ 4]) = mean; *(uint32_t*)(&c->scratchpad[i*s->uvlinesize+ 4]) = mean;
} }
intra_score+= s->dsp.mb_cmp[1](s, s->me.scratchpad, dest_c, s->uvlinesize); intra_score+= s->dsp.mb_cmp[1](s, c->scratchpad, dest_c, s->uvlinesize);
} }
} }
#endif #endif
intra_score += s->me.mb_penalty_factor*16; intra_score += c->mb_penalty_factor*16;
if(intra_score < dmin){ if(intra_score < dmin){
mb_type= CANDIDATE_MB_TYPE_INTRA; mb_type= CANDIDATE_MB_TYPE_INTRA;
...@@ -1382,20 +1383,20 @@ int ff_pre_estimate_p_frame_motion(MpegEncContext * s, ...@@ -1382,20 +1383,20 @@ int ff_pre_estimate_p_frame_motion(MpegEncContext * s,
int P[10][2]; int P[10][2];
const int shift= 1+s->quarter_sample; const int shift= 1+s->quarter_sample;
const int xy= mb_x + mb_y*s->mb_stride; const int xy= mb_x + mb_y*s->mb_stride;
init_ref(s, s->new_picture.data, s->last_picture.data, NULL, 16*mb_x, 16*mb_y, 0); init_ref(c, s->new_picture.data, s->last_picture.data, NULL, 16*mb_x, 16*mb_y, 0);
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); c->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; c->current_mv_penalty= c->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; c->skip=0;
P_LEFT[0] = s->p_mv_table[xy + 1][0]; P_LEFT[0] = s->p_mv_table[xy + 1][0];
P_LEFT[1] = s->p_mv_table[xy + 1][1]; P_LEFT[1] = s->p_mv_table[xy + 1][1];
if(P_LEFT[0] < (s->me.xmin<<shift)) P_LEFT[0] = (s->me.xmin<<shift); if(P_LEFT[0] < (c->xmin<<shift)) P_LEFT[0] = (c->xmin<<shift);
/* special case for first line */ /* special case for first line */
if (s->first_slice_line) { if (s->first_slice_line) {
...@@ -1408,9 +1409,9 @@ int ff_pre_estimate_p_frame_motion(MpegEncContext * s, ...@@ -1408,9 +1409,9 @@ int ff_pre_estimate_p_frame_motion(MpegEncContext * s,
P_TOP[1] = s->p_mv_table[xy + s->mb_stride ][1]; P_TOP[1] = s->p_mv_table[xy + s->mb_stride ][1];
P_TOPRIGHT[0] = s->p_mv_table[xy + s->mb_stride - 1][0]; P_TOPRIGHT[0] = s->p_mv_table[xy + s->mb_stride - 1][0];
P_TOPRIGHT[1] = s->p_mv_table[xy + s->mb_stride - 1][1]; P_TOPRIGHT[1] = s->p_mv_table[xy + s->mb_stride - 1][1];
if(P_TOP[1] < (s->me.ymin<<shift)) P_TOP[1] = (s->me.ymin<<shift); if(P_TOP[1] < (c->ymin<<shift)) P_TOP[1] = (c->ymin<<shift);
if(P_TOPRIGHT[0] > (s->me.xmax<<shift)) P_TOPRIGHT[0]= (s->me.xmax<<shift); if(P_TOPRIGHT[0] > (c->xmax<<shift)) P_TOPRIGHT[0]= (c->xmax<<shift);
if(P_TOPRIGHT[1] < (s->me.ymin<<shift)) P_TOPRIGHT[1]= (s->me.ymin<<shift); if(P_TOPRIGHT[1] < (c->ymin<<shift)) P_TOPRIGHT[1]= (c->ymin<<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]);
...@@ -1430,18 +1431,19 @@ int ff_pre_estimate_p_frame_motion(MpegEncContext * s, ...@@ -1430,18 +1431,19 @@ 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], int ref_index, int f_code) int mb_x, int mb_y, int16_t (*mv_table)[2], int ref_index, int f_code)
{ {
MotionEstContext * const c= &s->me;
int mx, my, dmin; int mx, my, dmin;
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 mv_penalty= s->me.mv_penalty[f_code] + MAX_MV; uint8_t * const mv_penalty= c->mv_penalty[f_code] + MAX_MV;
int mv_scale; int mv_scale;
s->me.penalty_factor = get_penalty_factor(s, s->avctx->me_cmp); c->penalty_factor = get_penalty_factor(s, s->avctx->me_cmp);
s->me.sub_penalty_factor= get_penalty_factor(s, s->avctx->me_sub_cmp); c->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); c->mb_penalty_factor = get_penalty_factor(s, s->avctx->mb_cmp);
s->me.current_mv_penalty= mv_penalty; c->current_mv_penalty= mv_penalty;
get_limits(s, 16*mb_x, 16*mb_y); get_limits(s, 16*mb_x, 16*mb_y);
...@@ -1476,7 +1478,7 @@ static int ff_estimate_motion_b(MpegEncContext * s, ...@@ -1476,7 +1478,7 @@ static int ff_estimate_motion_b(MpegEncContext * s,
P_LEFT[0] = mv_table[mot_xy - 1][0]; P_LEFT[0] = mv_table[mot_xy - 1][0];
P_LEFT[1] = mv_table[mot_xy - 1][1]; P_LEFT[1] = mv_table[mot_xy - 1][1];
if(P_LEFT[0] > (s->me.xmax<<shift)) P_LEFT[0] = (s->me.xmax<<shift); if(P_LEFT[0] > (c->xmax<<shift)) P_LEFT[0] = (c->xmax<<shift);
/* special case for first line */ /* special case for first line */
if (!s->first_slice_line) { if (!s->first_slice_line) {
...@@ -1484,15 +1486,15 @@ static int ff_estimate_motion_b(MpegEncContext * s, ...@@ -1484,15 +1486,15 @@ static int ff_estimate_motion_b(MpegEncContext * s,
P_TOP[1] = mv_table[mot_xy - mot_stride ][1]; P_TOP[1] = mv_table[mot_xy - mot_stride ][1];
P_TOPRIGHT[0] = mv_table[mot_xy - mot_stride + 1 ][0]; P_TOPRIGHT[0] = mv_table[mot_xy - mot_stride + 1 ][0];
P_TOPRIGHT[1] = mv_table[mot_xy - mot_stride + 1 ][1]; P_TOPRIGHT[1] = mv_table[mot_xy - mot_stride + 1 ][1];
if(P_TOP[1] > (s->me.ymax<<shift)) P_TOP[1]= (s->me.ymax<<shift); if(P_TOP[1] > (c->ymax<<shift)) P_TOP[1]= (c->ymax<<shift);
if(P_TOPRIGHT[0] < (s->me.xmin<<shift)) P_TOPRIGHT[0]= (s->me.xmin<<shift); if(P_TOPRIGHT[0] < (c->xmin<<shift)) P_TOPRIGHT[0]= (c->xmin<<shift);
if(P_TOPRIGHT[1] > (s->me.ymax<<shift)) P_TOPRIGHT[1]= (s->me.ymax<<shift); if(P_TOPRIGHT[1] > (c->ymax<<shift)) P_TOPRIGHT[1]= (c->ymax<<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]);
} }
s->me.pred_x= P_LEFT[0]; c->pred_x= P_LEFT[0];
s->me.pred_y= P_LEFT[1]; c->pred_y= P_LEFT[1];
} }
if(mv_table == s->b_forw_mv_table){ if(mv_table == s->b_forw_mv_table){
...@@ -1506,9 +1508,9 @@ static int ff_estimate_motion_b(MpegEncContext * s, ...@@ -1506,9 +1508,9 @@ static int ff_estimate_motion_b(MpegEncContext * s,
break; break;
} }
dmin= s->me.sub_motion_search(s, &mx, &my, dmin, 0, ref_index, 0, 16); dmin= c->sub_motion_search(s, &mx, &my, dmin, 0, ref_index, 0, 16);
if(s->avctx->me_sub_cmp != s->avctx->mb_cmp && !s->me.skip) if(s->avctx->me_sub_cmp != s->avctx->mb_cmp && !c->skip)
dmin= get_mb_score(s, mx, my, 0, ref_index); 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);
...@@ -1530,10 +1532,10 @@ static inline int check_bidir_mv(MpegEncContext * s, ...@@ -1530,10 +1532,10 @@ static inline int check_bidir_mv(MpegEncContext * s,
//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; 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= c->mv_penalty[s->f_code] + MAX_MV; // f_code of the prev frame
int stride= s->me.stride; int stride= c->stride;
int uvstride= s->me.uvstride; int uvstride= c->uvstride;
uint8_t *dest_y = s->me.scratchpad; uint8_t *dest_y = c->scratchpad;
uint8_t *ptr; uint8_t *ptr;
int dxy; int dxy;
int src_x, src_y; int src_x, src_y;
...@@ -1572,8 +1574,8 @@ static inline int check_bidir_mv(MpegEncContext * s, ...@@ -1572,8 +1574,8 @@ static inline int check_bidir_mv(MpegEncContext * s,
s->dsp.avg_pixels_tab[size][dxy](dest_y , ptr , stride, h); s->dsp.avg_pixels_tab[size][dxy](dest_y , ptr , stride, h);
} }
fbmin = (mv_penalty[motion_fx-pred_fx] + mv_penalty[motion_fy-pred_fy])*s->me.mb_penalty_factor fbmin = (mv_penalty[motion_fx-pred_fx] + mv_penalty[motion_fy-pred_fy])*c->mb_penalty_factor
+(mv_penalty[motion_bx-pred_bx] + mv_penalty[motion_by-pred_by])*s->me.mb_penalty_factor +(mv_penalty[motion_bx-pred_bx] + mv_penalty[motion_by-pred_by])*c->mb_penalty_factor
+ s->dsp.mb_cmp[size](s, src_data[0], dest_y, stride, h); //FIXME new_pic + s->dsp.mb_cmp[size](s, src_data[0], dest_y, stride, h); //FIXME new_pic
if(s->avctx->mb_cmp&FF_CMP_CHROMA){ if(s->avctx->mb_cmp&FF_CMP_CHROMA){
...@@ -1611,6 +1613,7 @@ static inline int bidir_refine(MpegEncContext * s, int mb_x, int mb_y) ...@@ -1611,6 +1613,7 @@ static inline int bidir_refine(MpegEncContext * s, int mb_x, int mb_y)
static inline int direct_search(MpegEncContext * s, int mb_x, int mb_y) static inline int direct_search(MpegEncContext * s, int mb_x, int mb_y)
{ {
MotionEstContext * const c= &s->me;
int P[10][2]; int P[10][2];
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;
...@@ -1621,7 +1624,7 @@ static inline int direct_search(MpegEncContext * s, int mb_x, int mb_y) ...@@ -1621,7 +1624,7 @@ static inline int direct_search(MpegEncContext * s, int mb_x, int mb_y)
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;
s->me.current_mv_penalty= s->me.mv_penalty[1] + MAX_MV; c->current_mv_penalty= c->mv_penalty[1] + MAX_MV;
ymin= xmin=(-32)>>shift; ymin= xmin=(-32)>>shift;
ymax= xmax= 31>>shift; ymax= xmax= 31>>shift;
...@@ -1635,22 +1638,22 @@ static inline int direct_search(MpegEncContext * s, int mb_x, int mb_y) ...@@ -1635,22 +1638,22 @@ static inline int direct_search(MpegEncContext * s, int mb_x, int mb_y)
int index= s->block_index[i]; int index= s->block_index[i];
int min, max; int min, max;
s->me.co_located_mv[i][0]= s->next_picture.motion_val[0][index][0]; c->co_located_mv[i][0]= s->next_picture.motion_val[0][index][0];
s->me.co_located_mv[i][1]= s->next_picture.motion_val[0][index][1]; c->co_located_mv[i][1]= s->next_picture.motion_val[0][index][1];
s->me.direct_basis_mv[i][0]= s->me.co_located_mv[i][0]*time_pb/time_pp + ((i& 1)<<(shift+3)); c->direct_basis_mv[i][0]= c->co_located_mv[i][0]*time_pb/time_pp + ((i& 1)<<(shift+3));
s->me.direct_basis_mv[i][1]= s->me.co_located_mv[i][1]*time_pb/time_pp + ((i>>1)<<(shift+3)); c->direct_basis_mv[i][1]= c->co_located_mv[i][1]*time_pb/time_pp + ((i>>1)<<(shift+3));
// s->me.direct_basis_mv[1][i][0]= s->me.co_located_mv[i][0]*(time_pb - time_pp)/time_pp + ((i &1)<<(shift+3); // c->direct_basis_mv[1][i][0]= c->co_located_mv[i][0]*(time_pb - time_pp)/time_pp + ((i &1)<<(shift+3);
// s->me.direct_basis_mv[1][i][1]= s->me.co_located_mv[i][1]*(time_pb - time_pp)/time_pp + ((i>>1)<<(shift+3); // c->direct_basis_mv[1][i][1]= c->co_located_mv[i][1]*(time_pb - time_pp)/time_pp + ((i>>1)<<(shift+3);
max= FFMAX(s->me.direct_basis_mv[i][0], s->me.direct_basis_mv[i][0] - s->me.co_located_mv[i][0])>>shift; max= FFMAX(c->direct_basis_mv[i][0], c->direct_basis_mv[i][0] - c->co_located_mv[i][0])>>shift;
min= FFMIN(s->me.direct_basis_mv[i][0], s->me.direct_basis_mv[i][0] - s->me.co_located_mv[i][0])>>shift; min= FFMIN(c->direct_basis_mv[i][0], c->direct_basis_mv[i][0] - c->co_located_mv[i][0])>>shift;
max+= 16*mb_x + 1; // +-1 is for the simpler rounding max+= 16*mb_x + 1; // +-1 is for the simpler rounding
min+= 16*mb_x - 1; min+= 16*mb_x - 1;
xmax= FFMIN(xmax, s->width - max); xmax= FFMIN(xmax, s->width - max);
xmin= FFMAX(xmin, - 16 - min); xmin= FFMAX(xmin, - 16 - min);
max= FFMAX(s->me.direct_basis_mv[i][1], s->me.direct_basis_mv[i][1] - s->me.co_located_mv[i][1])>>shift; max= FFMAX(c->direct_basis_mv[i][1], c->direct_basis_mv[i][1] - c->co_located_mv[i][1])>>shift;
min= FFMIN(s->me.direct_basis_mv[i][1], s->me.direct_basis_mv[i][1] - s->me.co_located_mv[i][1])>>shift; min= FFMIN(c->direct_basis_mv[i][1], c->direct_basis_mv[i][1] - c->co_located_mv[i][1])>>shift;
max+= 16*mb_y + 1; // +-1 is for the simpler rounding max+= 16*mb_y + 1; // +-1 is for the simpler rounding
min+= 16*mb_y - 1; min+= 16*mb_y - 1;
ymax= FFMIN(ymax, s->height - max); ymax= FFMIN(ymax, s->height - max);
...@@ -1668,14 +1671,14 @@ static inline int direct_search(MpegEncContext * s, int mb_x, int mb_y) ...@@ -1668,14 +1671,14 @@ static inline int direct_search(MpegEncContext * s, int mb_x, int mb_y)
return 256*256*256*64; return 256*256*256*64;
} }
s->me.xmin= xmin; c->xmin= xmin;
s->me.ymin= ymin; c->ymin= ymin;
s->me.xmax= xmax; c->xmax= xmax;
s->me.ymax= ymax; c->ymax= ymax;
s->me.flags |= FLAG_DIRECT; c->flags |= FLAG_DIRECT;
s->me.sub_flags |= FLAG_DIRECT; c->sub_flags |= FLAG_DIRECT;
s->me.pred_x=0; c->pred_x=0;
s->me.pred_y=0; c->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);
...@@ -1692,20 +1695,20 @@ static inline int direct_search(MpegEncContext * s, int mb_x, int mb_y) ...@@ -1692,20 +1695,20 @@ static inline int direct_search(MpegEncContext * s, int mb_x, int mb_y)
} }
dmin = epzs_motion_search(s, &mx, &my, P, 0, 0, mv_table, 1<<(16-shift)); dmin = epzs_motion_search(s, &mx, &my, P, 0, 0, mv_table, 1<<(16-shift));
if(s->me.sub_flags&FLAG_QPEL) if(c->sub_flags&FLAG_QPEL)
dmin = qpel_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16); dmin = qpel_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
else else
dmin = hpel_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16); dmin = hpel_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
if(s->avctx->me_sub_cmp != s->avctx->mb_cmp && !s->me.skip) if(s->avctx->me_sub_cmp != s->avctx->mb_cmp && !c->skip)
dmin= get_mb_score(s, mx, my, 0, 0); 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 c->?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; c->flags &= ~FLAG_DIRECT;
s->me.sub_flags &= ~FLAG_DIRECT; c->sub_flags &= ~FLAG_DIRECT;
return dmin; return dmin;
} }
...@@ -1713,14 +1716,15 @@ static inline int direct_search(MpegEncContext * s, int mb_x, int mb_y) ...@@ -1713,14 +1716,15 @@ static inline int direct_search(MpegEncContext * s, int mb_x, int mb_y)
void ff_estimate_b_frame_motion(MpegEncContext * s, void ff_estimate_b_frame_motion(MpegEncContext * s,
int mb_x, int mb_y) int mb_x, int mb_y)
{ {
const int penalty_factor= s->me.mb_penalty_factor; MotionEstContext * const c= &s->me;
const int penalty_factor= c->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 xy = mb_y*s->mb_stride + mb_x; const int xy = mb_y*s->mb_stride + mb_x;
init_ref(s, s->new_picture.data, s->last_picture.data, s->next_picture.data, 16*mb_x, 16*mb_y, 2); init_ref(c, s->new_picture.data, s->last_picture.data, s->next_picture.data, 16*mb_x, 16*mb_y, 2);
s->me.skip=0; c->skip=0;
if(s->avctx->me_threshold){ if(s->avctx->me_threshold){
int vard= (check_input_motion(s, mb_x, mb_y, 0)+128)>>8; int vard= (check_input_motion(s, mb_x, mb_y, 0)+128)>>8;
...@@ -1747,23 +1751,23 @@ void ff_estimate_b_frame_motion(MpegEncContext * s, ...@@ -1747,23 +1751,23 @@ void ff_estimate_b_frame_motion(MpegEncContext * s,
direct_search(s, mb_x, mb_y); direct_search(s, mb_x, mb_y);
} }
if(type == CANDIDATE_MB_TYPE_FORWARD || type == CANDIDATE_MB_TYPE_BIDIR){ if(type == CANDIDATE_MB_TYPE_FORWARD || type == CANDIDATE_MB_TYPE_BIDIR){
s->me.skip=0; c->skip=0;
ff_estimate_motion_b(s, mb_x, mb_y, s->b_forw_mv_table, 0, s->f_code); ff_estimate_motion_b(s, mb_x, mb_y, s->b_forw_mv_table, 0, s->f_code);
} }
if(type == CANDIDATE_MB_TYPE_BACKWARD || type == CANDIDATE_MB_TYPE_BIDIR){ if(type == CANDIDATE_MB_TYPE_BACKWARD || type == CANDIDATE_MB_TYPE_BIDIR){
s->me.skip=0; c->skip=0;
ff_estimate_motion_b(s, mb_x, mb_y, s->b_back_mv_table, 2, s->b_code); ff_estimate_motion_b(s, mb_x, mb_y, s->b_back_mv_table, 2, s->b_code);
} }
if(type == CANDIDATE_MB_TYPE_FORWARD_I || type == CANDIDATE_MB_TYPE_BIDIR_I){ if(type == CANDIDATE_MB_TYPE_FORWARD_I || type == CANDIDATE_MB_TYPE_BIDIR_I){
s->me.skip=0; c->skip=0;
s->me.current_mv_penalty= s->me.mv_penalty[s->f_code] + MAX_MV; c->current_mv_penalty= c->mv_penalty[s->f_code] + MAX_MV;
interlaced_search(s, 0, interlaced_search(s, 0,
s->b_field_mv_table[0], s->b_field_select_table[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], 1); s->b_forw_mv_table[xy][0], s->b_forw_mv_table[xy][1], 1);
} }
if(type == CANDIDATE_MB_TYPE_BACKWARD_I || type == CANDIDATE_MB_TYPE_BIDIR_I){ if(type == CANDIDATE_MB_TYPE_BACKWARD_I || type == CANDIDATE_MB_TYPE_BIDIR_I){
s->me.skip=0; c->skip=0;
s->me.current_mv_penalty= s->me.mv_penalty[s->b_code] + MAX_MV; c->current_mv_penalty= c->mv_penalty[s->b_code] + MAX_MV;
interlaced_search(s, 2, interlaced_search(s, 2,
s->b_field_mv_table[1], s->b_field_select_table[1], 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], 1); s->b_back_mv_table[xy][0], s->b_back_mv_table[xy][1], 1);
...@@ -1777,25 +1781,25 @@ void ff_estimate_b_frame_motion(MpegEncContext * s, ...@@ -1777,25 +1781,25 @@ void ff_estimate_b_frame_motion(MpegEncContext * s,
else else
dmin= INT_MAX; dmin= INT_MAX;
//FIXME penalty stuff for non mpeg4 //FIXME penalty stuff for non mpeg4
s->me.skip=0; c->skip=0;
fmin= ff_estimate_motion_b(s, mb_x, mb_y, s->b_forw_mv_table, 0, s->f_code) + 3*penalty_factor; fmin= ff_estimate_motion_b(s, mb_x, mb_y, s->b_forw_mv_table, 0, s->f_code) + 3*penalty_factor;
s->me.skip=0; c->skip=0;
bmin= ff_estimate_motion_b(s, mb_x, mb_y, s->b_back_mv_table, 2, s->b_code) + 2*penalty_factor; bmin= ff_estimate_motion_b(s, mb_x, mb_y, s->b_back_mv_table, 2, 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; c->skip=0;
fbmin= bidir_refine(s, 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){
//FIXME mb type penalty //FIXME mb type penalty
s->me.skip=0; c->skip=0;
s->me.current_mv_penalty= s->me.mv_penalty[s->f_code] + MAX_MV; c->current_mv_penalty= c->mv_penalty[s->f_code] + MAX_MV;
fimin= interlaced_search(s, 0, fimin= interlaced_search(s, 0,
s->b_field_mv_table[0], s->b_field_select_table[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], 0); s->b_forw_mv_table[xy][0], s->b_forw_mv_table[xy][1], 0);
s->me.current_mv_penalty= s->me.mv_penalty[s->b_code] + MAX_MV; c->current_mv_penalty= c->mv_penalty[s->b_code] + MAX_MV;
bimin= interlaced_search(s, 2, bimin= interlaced_search(s, 2,
s->b_field_mv_table[1], s->b_field_select_table[1], 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], 0); s->b_back_mv_table[xy][0], s->b_back_mv_table[xy][1], 0);
......
libavcodec/motion_est_template.c
View file @ af4091f1
...@@ -25,14 +25,14 @@ ...@@ -25,14 +25,14 @@
//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 ...)
#define LOAD_COMMON\ #define LOAD_COMMON\
uint32_t * const score_map= s->me.score_map;\ uint32_t * const score_map= c->score_map;\
const int xmin= s->me.xmin;\ const int xmin= c->xmin;\
const int ymin= s->me.ymin;\ const int ymin= c->ymin;\
const int xmax= s->me.xmax;\ const int xmax= c->xmax;\
const int ymax= s->me.ymax;\ const int ymax= c->ymax;\
uint8_t *mv_penalty= s->me.current_mv_penalty;\ uint8_t *mv_penalty= c->current_mv_penalty;\
const int pred_x= s->me.pred_x;\ const int pred_x= c->pred_x;\
const int pred_y= s->me.pred_y;\ const int pred_y= c->pred_y;\
#define CHECK_HALF_MV(dx, dy, x, y)\ #define CHECK_HALF_MV(dx, dy, x, y)\
{\ {\
...@@ -53,7 +53,7 @@ static int hpel_motion_search)(MpegEncContext * s, ...@@ -53,7 +53,7 @@ static int hpel_motion_search)(MpegEncContext * s,
const int yy = 16 * s->mb_y + 8*(n>>1); const int yy = 16 * s->mb_y + 8*(n>>1);
const int mx = *mx_ptr; const int mx = *mx_ptr;
const int my = *my_ptr; const int my = *my_ptr;
const int penalty_factor= s->me.sub_penalty_factor; const int penalty_factor= c->sub_penalty_factor;
LOAD_COMMON LOAD_COMMON
...@@ -73,7 +73,7 @@ static int hpel_motion_search)(MpegEncContext * s, ...@@ -73,7 +73,7 @@ static int hpel_motion_search)(MpegEncContext * s,
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];
if(s->me.skip){ //FIXME somehow move up (benchmark) if(c->skip){ //FIXME somehow move up (benchmark)
*mx_ptr = 0; *mx_ptr = 0;
*my_ptr = 0; *my_ptr = 0;
return dmin; return dmin;
...@@ -117,21 +117,22 @@ static int hpel_motion_search(MpegEncContext * s, ...@@ -117,21 +117,22 @@ static int hpel_motion_search(MpegEncContext * s,
int src_index, int ref_index, int src_index, int ref_index,
int size, int h) int size, int h)
{ {
MotionEstContext * const c= &s->me;
const int mx = *mx_ptr; const int mx = *mx_ptr;
const int my = *my_ptr; const int my = *my_ptr;
const int penalty_factor= s->me.sub_penalty_factor; const int penalty_factor= c->sub_penalty_factor;
me_cmp_func cmp_sub, chroma_cmp_sub; me_cmp_func cmp_sub, chroma_cmp_sub;
int bx=2*mx, by=2*my; int bx=2*mx, by=2*my;
LOAD_COMMON LOAD_COMMON
int flags= s->me.sub_flags; int flags= c->sub_flags;
//FIXME factorize //FIXME factorize
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];
if(s->me.skip){ //FIXME move out of hpel? if(c->skip){ //FIXME move out of hpel?
*mx_ptr = 0; *mx_ptr = 0;
*my_ptr = 0; *my_ptr = 0;
return dmin; return dmin;
...@@ -148,19 +149,19 @@ static int hpel_motion_search(MpegEncContext * s, ...@@ -148,19 +149,19 @@ static int hpel_motion_search(MpegEncContext * s,
int d= dmin; int d= dmin;
const int index= (my<<ME_MAP_SHIFT) + mx; const int index= (my<<ME_MAP_SHIFT) + mx;
const int t= score_map[(index-(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)] const int t= score_map[(index-(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)]
+ (mv_penalty[bx - pred_x] + mv_penalty[by-2 - pred_y])*s->me.penalty_factor; + (mv_penalty[bx - pred_x] + mv_penalty[by-2 - pred_y])*c->penalty_factor;
const int l= score_map[(index- 1 )&(ME_MAP_SIZE-1)] const int l= score_map[(index- 1 )&(ME_MAP_SIZE-1)]
+ (mv_penalty[bx-2 - pred_x] + mv_penalty[by - pred_y])*s->me.penalty_factor; + (mv_penalty[bx-2 - pred_x] + mv_penalty[by - pred_y])*c->penalty_factor;
const int r= score_map[(index+ 1 )&(ME_MAP_SIZE-1)] const int r= score_map[(index+ 1 )&(ME_MAP_SIZE-1)]
+ (mv_penalty[bx+2 - pred_x] + mv_penalty[by - pred_y])*s->me.penalty_factor; + (mv_penalty[bx+2 - pred_x] + mv_penalty[by - pred_y])*c->penalty_factor;
const int b= score_map[(index+(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)] const int b= score_map[(index+(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)]
+ (mv_penalty[bx - pred_x] + mv_penalty[by+2 - pred_y])*s->me.penalty_factor; + (mv_penalty[bx - pred_x] + mv_penalty[by+2 - pred_y])*c->penalty_factor;
#if 1 #if 1
int key; int key;
int map_generation= s->me.map_generation; int map_generation= c->map_generation;
#ifndef NDEBUG #ifndef NDEBUG
uint32_t *map= s->me.map; uint32_t *map= c->map;
#endif #endif
key= ((my-1)<<ME_MAP_MV_BITS) + (mx) + map_generation; key= ((my-1)<<ME_MAP_MV_BITS) + (mx) + map_generation;
assert(map[(index-(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)] == key); assert(map[(index-(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)] == key);
...@@ -224,10 +225,11 @@ static int inline get_mb_score(MpegEncContext * s, int mx, int my, int src_index ...@@ -224,10 +225,11 @@ static int inline get_mb_score(MpegEncContext * s, int mx, int my, int src_index
int ref_index) int ref_index)
{ {
// 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;
MotionEstContext * const c= &s->me;
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= c->mb_penalty_factor;
const int flags= s->me.mb_flags; const int flags= c->mb_flags;
const int qpel= flags & FLAG_QPEL; const int qpel= flags & FLAG_QPEL;
const int mask= 1+2*qpel; const int mask= 1+2*qpel;
me_cmp_func cmp_sub, chroma_cmp_sub; me_cmp_func cmp_sub, chroma_cmp_sub;
...@@ -240,7 +242,7 @@ static int inline get_mb_score(MpegEncContext * s, int mx, int my, int src_index ...@@ -240,7 +242,7 @@ static int inline get_mb_score(MpegEncContext * s, int mx, int my, int src_index
cmp_sub= s->dsp.mb_cmp[size]; cmp_sub= s->dsp.mb_cmp[size];
chroma_cmp_sub= s->dsp.mb_cmp[size+1]; chroma_cmp_sub= s->dsp.mb_cmp[size+1];
assert(!s->me.skip); assert(!c->skip);
assert(s->avctx->me_sub_cmp != s->avctx->mb_cmp); assert(s->avctx->me_sub_cmp != s->avctx->mb_cmp);
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); 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);
...@@ -265,17 +267,18 @@ static int qpel_motion_search(MpegEncContext * s, ...@@ -265,17 +267,18 @@ static int qpel_motion_search(MpegEncContext * s,
int src_index, int ref_index, int src_index, int ref_index,
int size, int h) int size, int h)
{ {
MotionEstContext * const c= &s->me;
const int mx = *mx_ptr; const int mx = *mx_ptr;
const int my = *my_ptr; const int my = *my_ptr;
const int penalty_factor= s->me.sub_penalty_factor; const int penalty_factor= c->sub_penalty_factor;
const int map_generation= s->me.map_generation; const int map_generation= c->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= c->map;
me_cmp_func cmpf, chroma_cmpf; 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; int flags= c->sub_flags;
cmpf= s->dsp.me_cmp[size]; cmpf= s->dsp.me_cmp[size];
chroma_cmpf= s->dsp.me_cmp[size+1]; //factorize FIXME chroma_cmpf= s->dsp.me_cmp[size+1]; //factorize FIXME
...@@ -284,7 +287,7 @@ static int qpel_motion_search(MpegEncContext * s, ...@@ -284,7 +287,7 @@ static int qpel_motion_search(MpegEncContext * s,
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];
if(s->me.skip){ //FIXME somehow move up (benchmark) if(c->skip){ //FIXME somehow move up (benchmark)
*mx_ptr = 0; *mx_ptr = 0;
*my_ptr = 0; *my_ptr = 0;
return dmin; return dmin;
...@@ -330,8 +333,8 @@ static int qpel_motion_search(MpegEncContext * s, ...@@ -330,8 +333,8 @@ static int qpel_motion_search(MpegEncContext * s,
score += 1024*(mv_penalty[4*mx + nx - pred_x] + mv_penalty[4*my + ny - pred_y])*penalty_factor; score += 1024*(mv_penalty[4*mx + nx - pred_x] + mv_penalty[4*my + ny - pred_y])*penalty_factor;
// if(nx&1) score-=1024*s->me.penalty_factor; // if(nx&1) score-=1024*c->penalty_factor;
// if(ny&1) score-=1024*s->me.penalty_factor; // if(ny&1) score-=1024*c->penalty_factor;
for(i=0; i<8; i++){ for(i=0; i<8; i++){
if(score < best[i]){ if(score < best[i]){
...@@ -375,8 +378,8 @@ static int qpel_motion_search(MpegEncContext * s, ...@@ -375,8 +378,8 @@ static int qpel_motion_search(MpegEncContext * s,
if((nx&3)==0 && (ny&3)==0) continue; if((nx&3)==0 && (ny&3)==0) continue;
score += 32*(mv_penalty[4*mx + nx - pred_x] + mv_penalty[4*my + ny - pred_y])*penalty_factor; score += 32*(mv_penalty[4*mx + nx - pred_x] + mv_penalty[4*my + ny - pred_y])*penalty_factor;
// if(nx&1) score-=32*s->me.penalty_factor; // if(nx&1) score-=32*c->penalty_factor;
// if(ny&1) score-=32*s->me.penalty_factor; // if(ny&1) score-=32*c->penalty_factor;
for(i=0; i<8; i++){ for(i=0; i<8; i++){
if(score < best[i]){ if(score < best[i]){
...@@ -531,7 +534,7 @@ if( (y)<(ymin<<(S)) ) printf("%d %d %d %d %d ymin" #v, ymin, (x), (y), s->mb_x, ...@@ -531,7 +534,7 @@ if( (y)<(ymin<<(S)) ) printf("%d %d %d %d %d ymin" #v, ymin, (x), (y), s->mb_x,
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\ #define LOAD_COMMON2\
uint32_t *map= s->me.map;\ uint32_t *map= c->map;\
const int qpel= flags&FLAG_QPEL;\ const int qpel= flags&FLAG_QPEL;\
const int shift= 1+qpel;\ const int shift= 1+qpel;\
...@@ -539,11 +542,12 @@ static always_inline int small_diamond_search(MpegEncContext * s, int *best, int ...@@ -539,11 +542,12 @@ static always_inline int small_diamond_search(MpegEncContext * s, int *best, int
int src_index, int ref_index, int const penalty_factor, int src_index, int ref_index, int const penalty_factor,
int size, int h, int flags) int size, int h, int flags)
{ {
MotionEstContext * const c= &s->me;
me_cmp_func cmpf, chroma_cmpf; me_cmp_func cmpf, chroma_cmpf;
int next_dir=-1; int next_dir=-1;
LOAD_COMMON LOAD_COMMON
LOAD_COMMON2 LOAD_COMMON2
int map_generation= s->me.map_generation; int map_generation= c->map_generation;
cmpf= s->dsp.me_cmp[size]; cmpf= s->dsp.me_cmp[size];
chroma_cmpf= s->dsp.me_cmp[size+1]; chroma_cmpf= s->dsp.me_cmp[size+1];
...@@ -580,11 +584,12 @@ static int funny_diamond_search(MpegEncContext * s, int *best, int dmin, ...@@ -580,11 +584,12 @@ static int funny_diamond_search(MpegEncContext * s, int *best, int dmin,
int src_index, int ref_index, int const penalty_factor, int src_index, int ref_index, int const penalty_factor,
int size, int h, int flags) int size, int h, int flags)
{ {
MotionEstContext * const c= &s->me;
me_cmp_func cmpf, chroma_cmpf; me_cmp_func cmpf, chroma_cmpf;
int dia_size; int dia_size;
LOAD_COMMON LOAD_COMMON
LOAD_COMMON2 LOAD_COMMON2
int map_generation= s->me.map_generation; int map_generation= c->map_generation;
cmpf= s->dsp.me_cmp[size]; cmpf= s->dsp.me_cmp[size];
chroma_cmpf= s->dsp.me_cmp[size+1]; chroma_cmpf= s->dsp.me_cmp[size+1];
...@@ -670,13 +675,14 @@ static int sab_diamond_search(MpegEncContext * s, int *best, int dmin, ...@@ -670,13 +675,14 @@ static int sab_diamond_search(MpegEncContext * s, int *best, int dmin,
int src_index, int ref_index, int const penalty_factor, int src_index, int ref_index, int const penalty_factor,
int size, int h, int flags) int size, int h, int flags)
{ {
MotionEstContext * const c= &s->me;
me_cmp_func cmpf, chroma_cmpf; 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(c->dia_size);
int i, j; int i, j;
LOAD_COMMON LOAD_COMMON
LOAD_COMMON2 LOAD_COMMON2
int map_generation= s->me.map_generation; int map_generation= c->map_generation;
cmpf= s->dsp.me_cmp[size]; cmpf= s->dsp.me_cmp[size];
chroma_cmpf= s->dsp.me_cmp[size+1]; chroma_cmpf= s->dsp.me_cmp[size+1];
...@@ -749,16 +755,17 @@ static int var_diamond_search(MpegEncContext * s, int *best, int dmin, ...@@ -749,16 +755,17 @@ static int var_diamond_search(MpegEncContext * s, int *best, int dmin,
int src_index, int ref_index, int const penalty_factor, int src_index, int ref_index, int const penalty_factor,
int size, int h, int flags) int size, int h, int flags)
{ {
MotionEstContext * const c= &s->me;
me_cmp_func cmpf, chroma_cmpf; me_cmp_func cmpf, chroma_cmpf;
int dia_size; int dia_size;
LOAD_COMMON LOAD_COMMON
LOAD_COMMON2 LOAD_COMMON2
int map_generation= s->me.map_generation; int map_generation= c->map_generation;
cmpf= s->dsp.me_cmp[size]; cmpf= s->dsp.me_cmp[size];
chroma_cmpf= 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<=c->dia_size; dia_size++){
int dir, start, end; int dir, start, end;
const int x= best[0]; const int x= best[0];
const int y= best[1]; const int y= best[1];
...@@ -824,11 +831,12 @@ if(256*256*256*64 % (stats[0]+1)==0){ ...@@ -824,11 +831,12 @@ if(256*256*256*64 % (stats[0]+1)==0){
static always_inline int diamond_search(MpegEncContext * s, int *best, int dmin, static always_inline int diamond_search(MpegEncContext * s, int *best, int dmin,
int src_index, int ref_index, int const penalty_factor, int src_index, int ref_index, int const penalty_factor,
int size, int h, int flags){ int size, int h, int flags){
if(s->me.dia_size==-1) MotionEstContext * const c= &s->me;
if(c->dia_size==-1)
return funny_diamond_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags); return funny_diamond_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags);
else if(s->me.dia_size<-1) else if(c->dia_size<-1)
return sab_diamond_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags); return sab_diamond_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags);
else if(s->me.dia_size<2) else if(c->dia_size<2)
return small_diamond_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags); return small_diamond_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags);
else else
return var_diamond_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags); return var_diamond_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags);
...@@ -838,10 +846,11 @@ static always_inline int epzs_motion_search_internal(MpegEncContext * s, int *mx ...@@ -838,10 +846,11 @@ static always_inline int epzs_motion_search_internal(MpegEncContext * s, int *mx
int P[10][2], int src_index, int ref_index, int16_t (*last_mv)[2], int P[10][2], int src_index, int ref_index, int16_t (*last_mv)[2],
int ref_mv_scale, int flags) int ref_mv_scale, int flags)
{ {
MotionEstContext * const c= &s->me;
int best[2]={0, 0}; int best[2]={0, 0};
int d, dmin; int d, dmin;
int map_generation; int map_generation;
const int penalty_factor= s->me.penalty_factor; const int penalty_factor= c->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
...@@ -854,7 +863,7 @@ static always_inline int epzs_motion_search_internal(MpegEncContext * s, int *mx ...@@ -854,7 +863,7 @@ static always_inline int epzs_motion_search_internal(MpegEncContext * s, int *mx
cmpf= s->dsp.me_cmp[size]; cmpf= s->dsp.me_cmp[size];
chroma_cmpf= 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(c);
dmin= cmp(s, 0, 0, 0, 0, size, h, ref_index, src_index, cmpf, chroma_cmpf, flags); 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;
...@@ -871,7 +880,7 @@ static always_inline int epzs_motion_search_internal(MpegEncContext * s, int *mx ...@@ -871,7 +880,7 @@ static always_inline int epzs_motion_search_internal(MpegEncContext * s, int *mx
|P_TOPRIGHT[0]|P_TOPRIGHT[1])==0){ |P_TOPRIGHT[0]|P_TOPRIGHT[1])==0){
*mx_ptr= 0; *mx_ptr= 0;
*my_ptr= 0; *my_ptr= 0;
s->me.skip=1; c->skip=1;
return dmin; return dmin;
} }
CHECK_MV(P_MEDIAN[0]>>shift, P_MEDIAN[1]>>shift) CHECK_MV(P_MEDIAN[0]>>shift, P_MEDIAN[1]>>shift)
...@@ -884,7 +893,7 @@ static always_inline int epzs_motion_search_internal(MpegEncContext * s, int *mx ...@@ -884,7 +893,7 @@ static always_inline int epzs_motion_search_internal(MpegEncContext * s, int *mx
} }
} }
if(dmin>256*4){ if(dmin>256*4){
if(s->me.pre_pass){ if(c->pre_pass){
CHECK_CLIPED_MV((last_mv[ref_mv_xy-1][0]*ref_mv_scale + (1<<15))>>16, CHECK_CLIPED_MV((last_mv[ref_mv_xy-1][0]*ref_mv_scale + (1<<15))>>16,
(last_mv[ref_mv_xy-1][1]*ref_mv_scale + (1<<15))>>16) (last_mv[ref_mv_xy-1][1]*ref_mv_scale + (1<<15))>>16)
if(!s->first_slice_line) if(!s->first_slice_line)
...@@ -936,14 +945,15 @@ static inline int epzs_motion_search(MpegEncContext * s, int *mx_ptr, int *my_pt ...@@ -936,14 +945,15 @@ static inline int epzs_motion_search(MpegEncContext * s, int *mx_ptr, int *my_pt
int P[10][2], int src_index, int ref_index, int16_t (*last_mv)[2], int P[10][2], int src_index, int ref_index, int16_t (*last_mv)[2],
int ref_mv_scale) int ref_mv_scale)
{ {
MotionEstContext * const c= &s->me;
//FIXME convert other functions in the same way if faster //FIXME convert other functions in the same way if faster
switch(s->me.flags){ switch(c->flags){
case 0: case 0:
return epzs_motion_search_internal(s, mx_ptr, my_ptr, P, src_index, ref_index, last_mv, ref_mv_scale, 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: // 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); // return epzs_motion_search_internal(s, mx_ptr, my_ptr, P, src_index, ref_index, last_mv, ref_mv_scale, FLAG_QPEL);
default: default:
return epzs_motion_search_internal(s, mx_ptr, my_ptr, P, src_index, ref_index, last_mv, ref_mv_scale, s->me.flags); return epzs_motion_search_internal(s, mx_ptr, my_ptr, P, src_index, ref_index, last_mv, ref_mv_scale, c->flags);
} }
} }
...@@ -952,23 +962,24 @@ static int epzs_motion_search4(MpegEncContext * s, ...@@ -952,23 +962,24 @@ static int epzs_motion_search4(MpegEncContext * s,
int src_index, int ref_index, int16_t (*last_mv)[2], int src_index, int ref_index, int16_t (*last_mv)[2],
int ref_mv_scale) int ref_mv_scale)
{ {
MotionEstContext * const c= &s->me;
int best[2]={0, 0}; int best[2]={0, 0};
int d, dmin; int d, dmin;
int map_generation; int map_generation;
const int penalty_factor= s->me.penalty_factor; const int penalty_factor= c->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 cmpf, chroma_cmpf; me_cmp_func cmpf, chroma_cmpf;
LOAD_COMMON LOAD_COMMON
int flags= s->me.flags; int flags= c->flags;
LOAD_COMMON2 LOAD_COMMON2
cmpf= s->dsp.me_cmp[size]; cmpf= s->dsp.me_cmp[size];
chroma_cmpf= 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(c);
dmin = 1000000; dmin = 1000000;
//printf("%d %d %d %d //",xmin, ymin, xmax, ymax); //printf("%d %d %d %d //",xmin, ymin, xmax, ymax);
...@@ -1013,23 +1024,24 @@ static int epzs_motion_search2(MpegEncContext * s, ...@@ -1013,23 +1024,24 @@ static int epzs_motion_search2(MpegEncContext * s,
int src_index, int ref_index, int16_t (*last_mv)[2], int src_index, int ref_index, int16_t (*last_mv)[2],
int ref_mv_scale) int ref_mv_scale)
{ {
MotionEstContext * const c= &s->me;
int best[2]={0, 0}; int best[2]={0, 0};
int d, dmin; int d, dmin;
int map_generation; int map_generation;
const int penalty_factor= s->me.penalty_factor; const int penalty_factor= c->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 cmpf, chroma_cmpf; me_cmp_func cmpf, chroma_cmpf;
LOAD_COMMON LOAD_COMMON
int flags= s->me.flags; int flags= c->flags;
LOAD_COMMON2 LOAD_COMMON2
cmpf= s->dsp.me_cmp[size]; cmpf= s->dsp.me_cmp[size];
chroma_cmpf= 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(c);
dmin = 1000000; dmin = 1000000;
//printf("%d %d %d %d //",xmin, ymin, xmax, ymax); //printf("%d %d %d %d //",xmin, ymin, xmax, ymax);
......
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