int src_index, int ref_index,
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);
- if(s->me.map_generation==0){
- s->me.map_generation= 1<<(ME_MAP_MV_BITS*2);
- memset(s->me.map, 0, sizeof(uint32_t)*ME_MAP_SIZE);
+ c->map_generation+= 1<<(ME_MAP_MV_BITS*2);
+ if(c->map_generation==0){
+ c->map_generation= 1<<(ME_MAP_MV_BITS*2);
+ memset(c->map, 0, sizeof(uint32_t)*ME_MAP_SIZE);
}
- return s->me.map_generation;
+ return c->map_generation;
}
/* shape adaptive search stuff */
#define FLAG_CHROMA 2
#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){
- MotionEstContext * const c= &s->me;
+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){
const int offset[3]= {
y*c-> stride + x,
((y*c->uvstride + x)>>1),
}
}
-static int get_flags(MpegEncContext *s, int direct, int chroma){
- return ((s->flags&CODEC_FLAG_QPEL) ? FLAG_QPEL : 0)
+static int get_flags(MotionEstContext *c, int direct, int chroma){
+ return ((c->avctx->flags&CODEC_FLAG_QPEL) ? FLAG_QPEL : 0)
+ (direct ? FLAG_DIRECT : 0)
+ (chroma ? FLAG_CHROMA : 0);
}
-static inline void init_mc(MpegEncContext *s, int size, int flags){
- MotionEstContext * const c= &s->me;
-
-/*FIXME s->no_rounding b_type*/
- if(flags & FLAG_CHROMA){
- if(s->no_rounding) c->chroma_hpel_put= &s->dsp.put_no_rnd_pixels_tab[size+1];
- else c->chroma_hpel_put= &s->dsp.put_pixels_tab[size+1];
- }
-
- if(flags & FLAG_QPEL){
- c->qpel_avg= &s->dsp.avg_qpel_pixels_tab[size];
- if(s->no_rounding) c->qpel_put= &s->dsp.put_no_rnd_qpel_pixels_tab[size];
- else c->qpel_put= &s->dsp.put_qpel_pixels_tab[size];
- }else{
- c->hpel_avg= &s->dsp.avg_pixels_tab[size];
- if(s->no_rounding) c->hpel_put= &s->dsp.put_no_rnd_pixels_tab[size];
- else c->hpel_put= &s->dsp.put_pixels_tab[size];
- }
- c->temp= c->scratchpad;
-}
-
static always_inline int cmp(MpegEncContext *s, const int x, const int y, const int subx, const int suby,
const int size, const int h, int ref_index, int src_index,
me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){
#include "motion_est_template.c"
-static inline int get_penalty_factor(MpegEncContext *s, int type){
- switch(type&0xFF){
- default:
- case FF_CMP_SAD:
- return s->qscale*2;
- case FF_CMP_DCT:
- return s->qscale*3;
- case FF_CMP_SATD:
- return s->qscale*6;
- case FF_CMP_SSE:
- return s->qscale*s->qscale*2;
- case FF_CMP_BIT:
- return 1;
- case FF_CMP_RD:
- case FF_CMP_PSNR:
- return (s->qscale*s->qscale*185 + 64)>>7;
- }
+static int zero_cmp(void *s, uint8_t *a, uint8_t *b, int stride, int h){
+ return 0;
+}
+
+static void zero_hpel(uint8_t *a, const uint8_t *b, int stride, int h){
}
void ff_init_me(MpegEncContext *s){
- ff_set_cmp(&s->dsp, s->dsp.me_pre_cmp, s->avctx->me_pre_cmp);
- ff_set_cmp(&s->dsp, s->dsp.me_cmp, s->avctx->me_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);
+ MotionEstContext * const c= &s->me;
+ c->avctx= s->avctx;
+
+ ff_set_cmp(&s->dsp, s->dsp.me_pre_cmp, c->avctx->me_pre_cmp);
+ ff_set_cmp(&s->dsp, s->dsp.me_cmp, c->avctx->me_cmp);
+ ff_set_cmp(&s->dsp, s->dsp.me_sub_cmp, c->avctx->me_sub_cmp);
+ ff_set_cmp(&s->dsp, s->dsp.mb_cmp, c->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);
+ c->flags = get_flags(c, 0, c->avctx->me_cmp &FF_CMP_CHROMA);
+ c->sub_flags= get_flags(c, 0, c->avctx->me_sub_cmp&FF_CMP_CHROMA);
+ c->mb_flags = get_flags(c, 0, c->avctx->mb_cmp &FF_CMP_CHROMA);
+/*FIXME s->no_rounding b_type*/
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;
+ 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{
- if(s->avctx->me_sub_cmp&FF_CMP_CHROMA)
- s->me.sub_motion_search= hpel_motion_search;
- else if( s->avctx->me_sub_cmp == FF_CMP_SAD
- && s->avctx-> me_cmp == FF_CMP_SAD
- && s->avctx-> mb_cmp == FF_CMP_SAD)
- s->me.sub_motion_search= sad_hpel_motion_search; // 2050 vs. 2450 cycles
+ if(c->avctx->me_sub_cmp&FF_CMP_CHROMA)
+ c->sub_motion_search= hpel_motion_search;
+ else if( c->avctx->me_sub_cmp == FF_CMP_SAD
+ && c->avctx-> me_cmp == FF_CMP_SAD
+ && c->avctx-> mb_cmp == FF_CMP_SAD)
+ c->sub_motion_search= sad_hpel_motion_search; // 2050 vs. 2450 cycles
else
- s->me.sub_motion_search= hpel_motion_search;
+ c->sub_motion_search= hpel_motion_search;
}
+ c->hpel_avg= s->dsp.avg_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;
+
if(s->linesize){
- s->me.stride = s->linesize;
- s->me.uvstride= s->uvlinesize;
+ c->stride = s->linesize;
+ c->uvstride= s->uvlinesize;
}else{
- s->me.stride = 16*s->mb_width + 32;
- s->me.uvstride= 8*s->mb_width + 16;
+ c->stride = 16*s->mb_width + 32;
+ c->uvstride= 8*s->mb_width + 16;
+ }
+
+ // 8x8 fullpel search would need a 4x4 chroma compare, which we dont have yet, and even if we had the motion estimation code doesnt expect it
+ if(s->codec_id != CODEC_ID_SNOW){
+ if((c->avctx->me_cmp&FF_CMP_CHROMA) && !s->dsp.me_cmp[2]){
+ s->dsp.me_cmp[2]= zero_cmp;
+ }
+ if((c->avctx->me_sub_cmp&FF_CMP_CHROMA) && !s->dsp.me_sub_cmp[2]){
+ s->dsp.me_sub_cmp[2]= zero_cmp;
+ }
+ c->hpel_put[2][0]= c->hpel_put[2][1]=
+ c->hpel_put[2][2]= c->hpel_put[2][3]= zero_hpel;
}
+
+ if(s->codec_id == CODEC_ID_H261){
+ c->sub_motion_search= no_sub_motion_search;
+ }
+
+ c->temp= c->scratchpad;
}
#if 0
*my_ptr = 16 * s->mb_y;
}
+#if 0 /* the use of these functions is inside #if 0 */
static int full_motion_search(MpegEncContext * s,
int *mx_ptr, int *my_ptr, int range,
int xmin, int ymin, int xmax, int ymax, uint8_t *ref_picture)
*my_ptr = my;
return dminy;
}
-
+#endif /* 0 */
#define Z_THRESHOLD 256
int src_index, int ref_index,
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;
uint8_t *pix, *ptr;
- int stride= s->me.stride;
- const int flags= s->me.sub_flags;
+ int stride= c->stride;
+ const int flags= c->sub_flags;
LOAD_COMMON
assert(flags == 0);
- if(s->me.skip){
+ if(c->skip){
// printf("S");
*mx_ptr = 0;
*my_ptr = 0;
}
// printf("N");
- pix = s->me.src[src_index][0];
+ pix = c->src[src_index][0];
mx = *mx_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;
*/
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;
- else s->me.range= 16;
+ if(c->avctx->me_range) c->range= c->avctx->me_range >> 1;
+ else c->range= 16;
*/
if (s->unrestricted_mv) {
- s->me.xmin = - x - 16;
- s->me.ymin = - y - 16;
- s->me.xmax = - x + s->mb_width *16;
- s->me.ymax = - y + s->mb_height*16;
+ c->xmin = - x - 16;
+ c->ymin = - y - 16;
+ c->xmax = - x + s->mb_width *16;
+ c->ymax = - y + s->mb_height*16;
+ } else if (s->out_format == FMT_H261){
+ // Search range of H261 is different from other codec standards
+ c->xmin = (x > 15) ? - 15 : 0;
+ c->ymin = (y > 15) ? - 15 : 0;
+ c->xmax = (x < s->mb_width * 16 - 16) ? 15 : 0;
+ c->ymax = (y < s->mb_height * 16 - 16) ? 15 : 0;
} else {
- s->me.xmin = - x;
- s->me.ymin = - y;
- s->me.xmax = - x + s->mb_width *16 - 16;
- s->me.ymax = - y + s->mb_height*16 - 16;
+ c->xmin = - x;
+ c->ymin = - y;
+ c->xmax = - x + s->mb_width *16 - 16;
+ c->ymax = - y + s->mb_height*16 - 16;
}
}
+static inline void init_mv4_ref(MotionEstContext *c){
+ const int stride= c->stride;
+
+ 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;
+}
+
static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift)
{
MotionEstContext * const c= &s->me;
int P[10][2];
int dmin_sum=0, mx4_sum=0, my4_sum=0;
int same=1;
- const int stride= s->linesize;
- const int uvstride= s->uvlinesize;
- uint8_t *mv_penalty= s->me.current_mv_penalty;
+ const int stride= c->stride;
+ uint8_t *mv_penalty= c->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);
+ init_mv4_ref(c);
for(block=0; block<4; block++){
int mx4, my4;
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];
- 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 */
if (s->first_slice_line && block<2) {
- s->me.pred_x= pred_x4= P_LEFT[0];
- s->me.pred_y= pred_y4= P_LEFT[1];
+ c->pred_x= pred_x4= P_LEFT[0];
+ c->pred_y= pred_y4= P_LEFT[1];
} else {
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_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];
- if(P_TOP[1] > (s->me.ymax<<shift)) P_TOP[1] = (s->me.ymax<<shift);
- if(P_TOPRIGHT[0] < (s->me.xmin<<shift)) P_TOPRIGHT[0]= (s->me.xmin<<shift);
- if(P_TOPRIGHT[0] > (s->me.xmax<<shift)) P_TOPRIGHT[0]= (s->me.xmax<<shift);
- if(P_TOPRIGHT[1] > (s->me.ymax<<shift)) P_TOPRIGHT[1]= (s->me.ymax<<shift);
+ if(P_TOP[1] > (c->ymax<<shift)) P_TOP[1] = (c->ymax<<shift);
+ if(P_TOPRIGHT[0] < (c->xmin<<shift)) P_TOPRIGHT[0]= (c->xmin<<shift);
+ if(P_TOPRIGHT[0] > (c->xmax<<shift)) P_TOPRIGHT[0]= (c->xmax<<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[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
- s->me.pred_x= pred_x4 = P_MEDIAN[0];
- s->me.pred_y= pred_y4 = P_MEDIAN[1];
+ c->pred_x= pred_x4 = P_MEDIAN[0];
+ c->pred_y= pred_y4 = P_MEDIAN[1];
}
P_MV1[0]= mx;
P_MV1[1]= my;
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]){
int dxy;
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){
uint8_t *ref= c->ref[block][0] + (mx4>>2) + (my4>>2)*stride;
dxy = ((my4 & 3) << 2) | (mx4 & 3);
else
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
dmin_sum+= dmin4;
return INT_MAX;
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(c->avctx->mb_cmp&FF_CMP_CHROMA){
int dxy;
int mx, my;
int offset;
offset= (s->mb_x*8 + (mx>>1)) + (s->mb_y*8 + (my>>1))*s->uvlinesize;
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](s->me.scratchpad+8 , s->last_picture.data[2] + 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](c->scratchpad+8 , s->last_picture.data[2] + offset, s->uvlinesize, 8);
}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](s->me.scratchpad+8 , s->last_picture.data[2] + 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](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[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[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, c->scratchpad+8, s->uvlinesize, 8);
}
- s->me.pred_x= mx;
- s->me.pred_y= my;
+ c->pred_x= mx;
+ c->pred_y= my;
- switch(s->avctx->mb_cmp&0xFF){
+ switch(c->avctx->mb_cmp&0xFF){
/*case FF_CMP_SSE:
return dmin_sum+ 32*s->qscale*s->qscale;*/
case FF_CMP_RD:
return dmin_sum;
default:
- return dmin_sum+ 11*s->me.mb_penalty_factor;
+ return dmin_sum+ 11*c->mb_penalty_factor;
+ }
+}
+
+static inline void init_interlaced_ref(MpegEncContext *s, int ref_index){
+ MotionEstContext * const c= &s->me;
+
+ 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;
}
}
static int interlaced_search(MpegEncContext *s, int ref_index,
- int16_t (*mv_tables[2][2])[2], uint8_t *field_select_tables[2], int mx, int my)
+ int16_t (*mv_tables[2][2])[2], uint8_t *field_select_tables[2], int mx, int my, int user_field_select)
{
MotionEstContext * const c= &s->me;
const int size=0;
uint8_t * const mv_penalty= c->current_mv_penalty;
int same=1;
const int stride= 2*s->linesize;
- const int uvstride= 2*s->uvlinesize;
int dmin_sum= 0;
const int mot_stride= s->mb_stride;
const int xy= s->mb_x + s->mb_y*mot_stride;
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);
+ init_interlaced_ref(s, ref_index);
for(block=0; block<2; block++){
int field_select;
int dmin, mx_i, my_i;
int16_t (*mv_table)[2]= mv_tables[block][field_select];
+ if(user_field_select){
+ if(field_select_tables[block][xy] != field_select)
+ continue;
+ }
+
P_LEFT[0] = mv_table[xy - 1][0];
P_LEFT[1] = mv_table[xy - 1][1];
if(P_LEFT[0] > (c->xmax<<1)) P_LEFT[0] = (c->xmax<<1);
- s->me.pred_x= P_LEFT[0];
- s->me.pred_y= P_LEFT[1];
+ c->pred_x= P_LEFT[0];
+ c->pred_y= P_LEFT[1];
if(!s->first_slice_line){
P_TOP[0] = mv_table[xy - mot_stride][0];
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+= (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
dmin+= c->mb_penalty_factor; //field_select bits
if(same)
return INT_MAX;
- switch(s->avctx->mb_cmp&0xFF){
+ switch(c->avctx->mb_cmp&0xFF){
/*case FF_CMP_SSE:
return dmin_sum+ 32*s->qscale*s->qscale;*/
case FF_CMP_RD:
}
}
+static void clip_input_mv(MpegEncContext * s, int16_t *mv, int interlaced){
+ int ymax= s->me.ymax>>interlaced;
+ int ymin= s->me.ymin>>interlaced;
+
+ if(mv[0] < s->me.xmin) mv[0] = s->me.xmin;
+ if(mv[0] > s->me.xmax) mv[0] = s->me.xmax;
+ if(mv[1] < ymin) mv[1] = ymin;
+ if(mv[1] > ymax) mv[1] = ymax;
+}
+
static inline int check_input_motion(MpegEncContext * s, int mb_x, int mb_y, int p_type){
MotionEstContext * const c= &s->me;
Picture *p= s->current_picture_ptr;
int flags= c->flags;
int shift= (flags&FLAG_QPEL) + 1;
int mask= (1<<shift)-1;
- int x, y;
+ int x, y, i;
int d=0;
me_cmp_func cmpf= s->dsp.sse[0];
me_cmp_func chroma_cmpf= s->dsp.sse[1];
- assert(p_type==0 || !USES_LIST(mb_type, 1));
+ if(p_type && USES_LIST(mb_type, 1)){
+ av_log(c->avctx, AV_LOG_ERROR, "backward motion vector in P frame\n");
+ return INT_MAX/4;
+ }
assert(IS_INTRA(mb_type) || USES_LIST(mb_type,0) || USES_LIST(mb_type,1));
+ for(i=0; i<4; i++){
+ int xy= s->block_index[i];
+ clip_input_mv(s, p->motion_val[0][xy], !!IS_INTERLACED(mb_type));
+ clip_input_mv(s, p->motion_val[1][xy], !!IS_INTERLACED(mb_type));
+ }
+
if(IS_INTERLACED(mb_type)){
int xy2= xy + s->b8_stride;
s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_INTRA;
c->stride<<=1;
c->uvstride<<=1;
- c->ref[1][0] = c->ref[0][0] + s->linesize;
- c->ref[3][0] = c->ref[2][0] + s->linesize;
- c->src[1][0] = c->src[0][0] + s->linesize;
- if(c->flags & FLAG_CHROMA){
- c->ref[1][1] = c->ref[0][1] + s->uvlinesize;
- c->ref[1][2] = c->ref[0][2] + s->uvlinesize;
- c->ref[3][1] = c->ref[2][1] + s->uvlinesize;
- c->ref[3][2] = c->ref[2][2] + s->uvlinesize;
- c->src[1][1] = c->src[0][1] + s->uvlinesize;
- c->src[1][2] = c->src[0][2] + s->uvlinesize;
+
+ if(!(s->flags & CODEC_FLAG_INTERLACED_ME)){
+ av_log(c->avctx, AV_LOG_ERROR, "Interlaced macroblock selected but interlaced motion estimation disabled\n");
+ return INT_MAX/4;
}
+
if(USES_LIST(mb_type, 0)){
int field_select0= p->ref_index[0][xy ];
int field_select1= p->ref_index[0][xy2];
assert(field_select0==0 ||field_select0==1);
assert(field_select1==0 ||field_select1==1);
+ init_interlaced_ref(s, 0);
+
if(p_type){
s->p_field_select_table[0][mb_xy]= field_select0;
s->p_field_select_table[1][mb_xy]= field_select1;
int field_select1= p->ref_index[1][xy2];
assert(field_select0==0 ||field_select0==1);
assert(field_select1==0 ||field_select1==1);
+ init_interlaced_ref(s, 2);
+
s->b_field_select_table[1][0][mb_xy]= field_select0;
s->b_field_select_table[1][1][mb_xy]= field_select1;
*(uint32_t*)s->b_field_mv_table[1][0][field_select0][mb_xy]= *(uint32_t*)p->motion_val[1][xy ];
}
c->stride>>=1;
c->uvstride>>=1;
+ }else if(IS_8X8(mb_type)){
+ if(!(s->flags & CODEC_FLAG_4MV)){
+ av_log(c->avctx, AV_LOG_ERROR, "4MV macroblock selected but 4MV encoding disabled\n");
+ return INT_MAX/4;
+ }
+ cmpf= s->dsp.sse[1];
+ chroma_cmpf= s->dsp.sse[1];
+ init_mv4_ref(c);
+ for(i=0; i<4; i++){
+ xy= s->block_index[i];
+ x= p->motion_val[0][xy][0];
+ y= p->motion_val[0][xy][1];
+ d+= cmp(s, x>>shift, y>>shift, x&mask, y&mask, 1, 8, i, i, cmpf, chroma_cmpf, flags);
+ }
+ s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_INTER4V;
}else{
if(USES_LIST(mb_type, 0)){
if(p_type){
{
MotionEstContext * const c= &s->me;
uint8_t *pix, *ppix;
- int sum, varc, vard, mx, my, dmin, xx, yy;
+ int sum, varc, vard, mx, my, dmin;
int P[10][2];
const int shift= 1+s->quarter_sample;
int mb_type=0;
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_mc(s, 0, c->flags);
+ 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->linesize == s->me.stride);
- assert(s->uvlinesize == s->me.uvstride);
+ assert(s->linesize == c->stride);
+ assert(s->uvlinesize == c->uvstride);
- 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.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->penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_cmp);
+ c->sub_penalty_factor= get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_sub_cmp);
+ c->mb_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->mb_cmp);
+ c->current_mv_penalty= c->mv_penalty[s->f_code] + MAX_MV;
get_limits(s, 16*mb_x, 16*mb_y);
- s->me.skip=0;
+ c->skip=0;
+
+ /* intra / predictive decision */
+ pix = c->src[0][0];
+ sum = s->dsp.pix_sum(pix, s->linesize);
+ varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500 + 128)>>8;
- if(s->avctx->me_threshold){
+ pic->mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
+ pic->mb_var [s->mb_stride * mb_y + mb_x] = varc;
+ c->mb_var_sum_temp += varc;
+
+ if(c->avctx->me_threshold){
vard= (check_input_motion(s, mb_x, mb_y, 1)+128)>>8;
- if(vard<s->avctx->me_threshold){
- pix = c->src[0][0];
- sum = s->dsp.pix_sum(pix, s->linesize);
- varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500 + 128)>>8;
-
- pic->mb_var [s->mb_stride * mb_y + mb_x] = varc;
+ if(vard<c->avctx->me_threshold){
pic->mc_mb_var[s->mb_stride * mb_y + mb_x] = vard;
- pic->mb_mean [s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
- s->mb_var_sum_temp += varc;
- s->mc_mb_var_sum_temp += vard;
+ c->mc_mb_var_sum_temp += vard;
if (vard <= 64 || vard < varc) { //FIXME
- s->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);
+ c->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);
}else{
- s->scene_change_score+= s->qscale;
+ c->scene_change_score+= s->qscale;
}
return;
}
+ if(vard<c->avctx->mb_threshold)
+ mb_type= s->mb_type[mb_x + mb_y*s->mb_stride];
}
switch(s->me_method) {
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];
- 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) {
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_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];
- if(P_TOP[1] > (s->me.ymax<<shift)) P_TOP[1] = (s->me.ymax<<shift);
- if(P_TOPRIGHT[0] < (s->me.xmin<<shift)) P_TOPRIGHT[0]= (s->me.xmin<<shift);
- if(P_TOPRIGHT[1] > (s->me.ymax<<shift)) P_TOPRIGHT[1]= (s->me.ymax<<shift);
+ if(P_TOP[1] > (c->ymax<<shift)) P_TOP[1] = (c->ymax<<shift);
+ if(P_TOPRIGHT[0] < (c->xmin<<shift)) P_TOPRIGHT[0]= (c->xmin<<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[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
}
}
- dmin = epzs_motion_search(s, &mx, &my, P, 0, 0, s->p_mv_table, (1<<16)>>shift);
+ dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, 0, s->p_mv_table, (1<<16)>>shift, 0, 16);
break;
}
- /* intra / predictive decision */
- xx = mb_x * 16;
- yy = mb_y * 16;
-
- pix = c->src[0][0];
/* At this point (mx,my) are full-pell and the relative displacement */
ppix = c->ref[0][0] + (my * s->linesize) + mx;
-
- sum = s->dsp.pix_sum(pix, s->linesize);
-
- varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500 + 128)>>8;
+
vard = (s->dsp.sse[0](NULL, pix, ppix, s->linesize, 16)+128)>>8;
-//printf("%d %d %d %X %X %X\n", s->mb_width, mb_x, mb_y,(int)s, (int)s->mb_var, (int)s->mc_mb_var); fflush(stdout);
- pic->mb_var [s->mb_stride * mb_y + mb_x] = varc;
pic->mc_mb_var[s->mb_stride * mb_y + mb_x] = vard;
- pic->mb_mean [s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
// pic->mb_cmp_score[s->mb_stride * mb_y + mb_x] = dmin;
- s->mb_var_sum_temp += varc;
- s->mc_mb_var_sum_temp += vard;
-//printf("E%d %d %d %X %X %X\n", s->mb_width, mb_x, mb_y,(int)s, (int)s->mb_var, (int)s->mc_mb_var); fflush(stdout);
+ c->mc_mb_var_sum_temp += vard;
#if 0
printf("varc=%4d avg_var=%4d (sum=%4d) vard=%4d mx=%2d my=%2d\n",
varc, s->avg_mb_var, sum, vard, mx - xx, my - yy);
#endif
- if(s->avctx->mb_decision > FF_MB_DECISION_SIMPLE){
+ if(mb_type){
if (vard <= 64 || vard < varc)
- s->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);
+ c->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);
else
- s->scene_change_score+= s->qscale;
+ c->scene_change_score+= s->qscale;
+
+ if(mb_type == CANDIDATE_MB_TYPE_INTER){
+ c->sub_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
+ set_p_mv_tables(s, mx, my, 1);
+ }else{
+ mx <<=shift;
+ my <<=shift;
+ }
+ if(mb_type == CANDIDATE_MB_TYPE_INTER4V){
+ h263_mv4_search(s, mx, my, shift);
+
+ set_p_mv_tables(s, mx, my, 0);
+ }
+ if(mb_type == CANDIDATE_MB_TYPE_INTER_I){
+ interlaced_search(s, 0, s->p_field_mv_table, s->p_field_select_table, mx, my, 1);
+ }
+ }else if(c->avctx->mb_decision > FF_MB_DECISION_SIMPLE){
+ if (vard <= 64 || vard < varc)
+ c->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);
+ else
+ c->scene_change_score+= s->qscale;
if (vard*2 + 200 > varc)
mb_type|= CANDIDATE_MB_TYPE_INTRA;
if (varc*2 + 200 > vard){
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(mx || my)
mb_type |= CANDIDATE_MB_TYPE_SKIPED; //FIXME check difference
my <<=shift;
}
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)
mb_type|=CANDIDATE_MB_TYPE_INTER4V;
}else
set_p_mv_tables(s, mx, my, 1);
if((s->flags&CODEC_FLAG_INTERLACED_ME)
- && !s->me.skip){ //FIXME varc/d checks
- if(interlaced_search(s, 0, s->p_field_mv_table, s->p_field_select_table, mx, my) < INT_MAX)
+ && !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)
mb_type |= CANDIDATE_MB_TYPE_INTER_I;
}
}else{
int intra_score, i;
mb_type= CANDIDATE_MB_TYPE_INTER;
- dmin= s->me.sub_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
- if(s->avctx->me_sub_cmp != s->avctx->mb_cmp && !s->me.skip)
- dmin= get_mb_score(s, mx, my, 0, 0);
+ dmin= c->sub_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
+ if(c->avctx->me_sub_cmp != c->avctx->mb_cmp && !c->skip)
+ dmin= ff_get_mb_score(s, mx, my, 0, 0, 0, 16, 1);
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);
if(dmin4 < dmin){
mb_type= CANDIDATE_MB_TYPE_INTER4V;
}
}
if((s->flags&CODEC_FLAG_INTERLACED_ME)
- && !s->me.skip){ //FIXME varc/d checks
- int dmin_i= interlaced_search(s, 0, s->p_field_mv_table, s->p_field_select_table, mx, my);
+ && !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);
if(dmin_i < dmin){
mb_type = CANDIDATE_MB_TYPE_INTER_I;
dmin= dmin_i;
set_p_mv_tables(s, mx, my, mb_type!=CANDIDATE_MB_TYPE_INTER4V);
/* get intra luma score */
- if((s->avctx->mb_cmp&0xFF)==FF_CMP_SSE){
+ if((c->avctx->mb_cmp&0xFF)==FF_CMP_SSE){
intra_score= (varc<<8) - 500; //FIXME dont scale it down so we dont have to fix it
}else{
int mean= (sum+128)>>8;
mean*= 0x01010101;
for(i=0; i<16; i++){
- *(uint32_t*)(&s->me.scratchpad[i*s->linesize+ 0]) = mean;
- *(uint32_t*)(&s->me.scratchpad[i*s->linesize+ 4]) = mean;
- *(uint32_t*)(&s->me.scratchpad[i*s->linesize+ 8]) = mean;
- *(uint32_t*)(&s->me.scratchpad[i*s->linesize+12]) = mean;
+ *(uint32_t*)(&c->scratchpad[i*s->linesize+ 0]) = mean;
+ *(uint32_t*)(&c->scratchpad[i*s->linesize+ 4]) = mean;
+ *(uint32_t*)(&c->scratchpad[i*s->linesize+ 8]) = 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
/* get chroma score */
- if(s->avctx->mb_cmp&FF_CMP_CHROMA){
+ if(c->avctx->mb_cmp&FF_CMP_CHROMA){
for(i=1; i<3; i++){
uint8_t *dest_c;
int mean;
mean*= 0x01010101;
for(i=0; i<8; i++){
- *(uint32_t*)(&s->me.scratchpad[i*s->uvlinesize+ 0]) = mean;
- *(uint32_t*)(&s->me.scratchpad[i*s->uvlinesize+ 4]) = mean;
+ *(uint32_t*)(&c->scratchpad[i*s->uvlinesize+ 0]) = 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
- intra_score += s->me.mb_penalty_factor*16;
+ intra_score += c->mb_penalty_factor*16;
if(intra_score < dmin){
mb_type= CANDIDATE_MB_TYPE_INTRA;
s->current_picture.mb_type[mb_y*s->mb_stride + mb_x]= 0;
if (vard <= 64 || vard < varc) { //FIXME
- s->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);
+ c->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);
}else{
- s->scene_change_score+= s->qscale;
+ c->scene_change_score+= s->qscale;
}
}
int P[10][2];
const int shift= 1+s->quarter_sample;
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_mc(s, 0, c->flags);
+ 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);
- 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;
+ c->pre_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_pre_cmp);
+ c->current_mv_penalty= c->mv_penalty[s->f_code] + MAX_MV;
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[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 */
if (s->first_slice_line) {
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[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_TOPRIGHT[0] > (s->me.xmax<<shift)) P_TOPRIGHT[0]= (s->me.xmax<<shift);
- if(P_TOPRIGHT[1] < (s->me.ymin<<shift)) P_TOPRIGHT[1]= (s->me.ymin<<shift);
+ if(P_TOP[1] < (c->ymin<<shift)) P_TOP[1] = (c->ymin<<shift);
+ if(P_TOPRIGHT[0] > (c->xmax<<shift)) P_TOPRIGHT[0]= (c->xmax<<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[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
c->pred_y = P_MEDIAN[1];
}
- dmin = epzs_motion_search(s, &mx, &my, P, 0, 0, s->p_mv_table, (1<<16)>>shift);
+ dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, 0, s->p_mv_table, (1<<16)>>shift, 0, 16);
s->p_mv_table[xy][0] = mx<<shift;
s->p_mv_table[xy][1] = my<<shift;
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)
{
+ MotionEstContext * const c= &s->me;
int mx, my, dmin;
int P[10][2];
const int shift= 1+s->quarter_sample;
const int mot_stride = s->mb_stride;
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;
- 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.mb_penalty_factor = get_penalty_factor(s, s->avctx->mb_cmp);
- s->me.current_mv_penalty= mv_penalty;
+ c->penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_cmp);
+ c->sub_penalty_factor= get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_sub_cmp);
+ c->mb_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->mb_cmp);
+ c->current_mv_penalty= mv_penalty;
get_limits(s, 16*mb_x, 16*mb_y);
P_LEFT[0] = mv_table[mot_xy - 1][0];
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 */
if (!s->first_slice_line) {
P_TOP[1] = mv_table[mot_xy - mot_stride ][1];
P_TOPRIGHT[0] = mv_table[mot_xy - mot_stride + 1 ][0];
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_TOPRIGHT[0] < (s->me.xmin<<shift)) P_TOPRIGHT[0]= (s->me.xmin<<shift);
- if(P_TOPRIGHT[1] > (s->me.ymax<<shift)) P_TOPRIGHT[1]= (s->me.ymax<<shift);
+ if(P_TOP[1] > (c->ymax<<shift)) P_TOP[1]= (c->ymax<<shift);
+ if(P_TOPRIGHT[0] < (c->xmin<<shift)) P_TOPRIGHT[0]= (c->xmin<<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[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
}
- s->me.pred_x= P_LEFT[0];
- s->me.pred_y= P_LEFT[1];
+ c->pred_x= P_LEFT[0];
+ c->pred_y= P_LEFT[1];
}
if(mv_table == s->b_forw_mv_table){
mv_scale= ((s->pb_time - s->pp_time)<<16) / (s->pp_time<<shift);
}
- dmin = epzs_motion_search(s, &mx, &my, P, 0, ref_index, s->p_mv_table, mv_scale);
+ dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, ref_index, s->p_mv_table, mv_scale, 0, 16);
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)
- dmin= get_mb_score(s, mx, my, 0, ref_index);
+ if(c->avctx->me_sub_cmp != c->avctx->mb_cmp && !c->skip)
+ dmin= ff_get_mb_score(s, mx, my, 0, ref_index, 0, 16, 1);
//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;
//FIXME better f_code prediction (max mv & distance)
//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
- int stride= s->me.stride;
- int uvstride= s->me.uvstride;
- uint8_t *dest_y = s->me.scratchpad;
+ uint8_t * const mv_penalty= c->mv_penalty[s->f_code] + MAX_MV; // f_code of the prev frame
+ int stride= c->stride;
+ int uvstride= c->uvstride;
+ uint8_t *dest_y = c->scratchpad;
uint8_t *ptr;
int dxy;
int src_x, src_y;
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
- +(mv_penalty[motion_bx-pred_bx] + mv_penalty[motion_by-pred_by])*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])*c->mb_penalty_factor
+ 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(c->avctx->mb_cmp&FF_CMP_CHROMA){
}
//FIXME CHROMA !!!
static inline int direct_search(MpegEncContext * s, int mb_x, int mb_y)
{
+ MotionEstContext * const c= &s->me;
int P[10][2];
const int mot_stride = s->mb_stride;
const int mot_xy = mb_y*mot_stride + mb_x;
int mx, my, xmin, xmax, ymin, ymax;
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;
ymax= xmax= 31>>shift;
int index= s->block_index[i];
int min, max;
- s->me.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];
- s->me.direct_basis_mv[i][0]= s->me.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));
-// 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);
-// 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);
-
- max= FFMAX(s->me.direct_basis_mv[i][0], s->me.direct_basis_mv[i][0] - s->me.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;
+ c->co_located_mv[i][0]= s->next_picture.motion_val[0][index][0];
+ c->co_located_mv[i][1]= s->next_picture.motion_val[0][index][1];
+ c->direct_basis_mv[i][0]= c->co_located_mv[i][0]*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));
+// c->direct_basis_mv[1][i][0]= c->co_located_mv[i][0]*(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(c->direct_basis_mv[i][0], c->direct_basis_mv[i][0] - c->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
min+= 16*mb_x - 1;
xmax= FFMIN(xmax, s->width - max);
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;
- min= FFMIN(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(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
min+= 16*mb_y - 1;
ymax= FFMIN(ymax, s->height - max);
return 256*256*256*64;
}
- s->me.xmin= xmin;
- s->me.ymin= ymin;
- s->me.xmax= xmax;
- 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;
+ c->xmin= xmin;
+ c->ymin= ymin;
+ c->xmax= xmax;
+ c->ymax= ymax;
+ c->flags |= FLAG_DIRECT;
+ c->sub_flags |= FLAG_DIRECT;
+ c->pred_x=0;
+ c->pred_y=0;
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_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
}
- dmin = epzs_motion_search(s, &mx, &my, P, 0, 0, mv_table, 1<<(16-shift));
- if(s->me.sub_flags&FLAG_QPEL)
+ dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, 0, mv_table, 1<<(16-shift), 0, 16);
+ if(c->sub_flags&FLAG_QPEL)
dmin = qpel_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
else
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)
- dmin= get_mb_score(s, mx, my, 0, 0);
+ if(c->avctx->me_sub_cmp != c->avctx->mb_cmp && !c->skip)
+ dmin= ff_get_mb_score(s, mx, my, 0, 0, 0, 16, 1);
- 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][1]= my;
- s->me.flags &= ~FLAG_DIRECT;
- s->me.sub_flags &= ~FLAG_DIRECT;
+ c->flags &= ~FLAG_DIRECT;
+ c->sub_flags &= ~FLAG_DIRECT;
return dmin;
}
void ff_estimate_b_frame_motion(MpegEncContext * s,
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 type=0;
- init_ref(s, s->new_picture.data, s->last_picture.data, s->next_picture.data, 16*mb_x, 16*mb_y, 2);
- init_mc(s, 0, s->me.flags);
+ const int xy = mb_y*s->mb_stride + mb_x;
+ init_ref(c, s->new_picture.data, s->last_picture.data, s->next_picture.data, 16*mb_x, 16*mb_y, 2);
+
+ get_limits(s, 16*mb_x, 16*mb_y);
- s->me.skip=0;
- if(s->avctx->me_threshold){
+ c->skip=0;
+ if(c->avctx->me_threshold){
int vard= (check_input_motion(s, mb_x, mb_y, 0)+128)>>8;
- if(vard<s->avctx->me_threshold){
+ if(vard<c->avctx->me_threshold){
// pix = c->src[0][0];
// sum = s->dsp.pix_sum(pix, s->linesize);
// varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500 + 128)>>8;
// pic->mb_var [s->mb_stride * mb_y + mb_x] = varc;
s->current_picture.mc_mb_var[s->mb_stride * mb_y + mb_x] = vard;
/* pic->mb_mean [s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
- s->mb_var_sum_temp += varc;*/
- s->mc_mb_var_sum_temp += vard;
+ c->mb_var_sum_temp += varc;*/
+ c->mc_mb_var_sum_temp += vard;
/* if (vard <= 64 || vard < varc) {
- s->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);
+ c->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);
}else{
- s->scene_change_score+= s->qscale;
+ c->scene_change_score+= s->qscale;
}*/
return;
}
+ if(vard<c->avctx->mb_threshold){
+ type= s->mb_type[mb_y*s->mb_stride + mb_x];
+ if(type == CANDIDATE_MB_TYPE_DIRECT){
+ direct_search(s, mb_x, mb_y);
+ }
+ if(type == CANDIDATE_MB_TYPE_FORWARD || type == CANDIDATE_MB_TYPE_BIDIR){
+ c->skip=0;
+ 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){
+ c->skip=0;
+ 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){
+ c->skip=0;
+ c->current_mv_penalty= c->mv_penalty[s->f_code] + MAX_MV;
+ 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], 1);
+ }
+ if(type == CANDIDATE_MB_TYPE_BACKWARD_I || type == CANDIDATE_MB_TYPE_BIDIR_I){
+ c->skip=0;
+ c->current_mv_penalty= c->mv_penalty[s->b_code] + MAX_MV;
+ 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], 1);
+ }
+ return;
+ }
}
if (s->codec_id == CODEC_ID_MPEG4)
else
dmin= INT_MAX;
//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;
- 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;
//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;
//printf("%d %d %d %d\n", dmin, fmin, bmin, fbmin);
if(s->flags & CODEC_FLAG_INTERLACED_ME){
- const int xy = mb_y*s->mb_stride + mb_x;
-
//FIXME mb type penalty
- s->me.skip=0;
- s->me.current_mv_penalty= s->me.mv_penalty[s->f_code] + MAX_MV;
+ c->skip=0;
+ c->current_mv_penalty= c->mv_penalty[s->f_code] + MAX_MV;
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->me.current_mv_penalty= s->me.mv_penalty[s->b_code] + MAX_MV;
+ s->b_forw_mv_table[xy][0], s->b_forw_mv_table[xy][1], 0);
+ c->current_mv_penalty= c->mv_penalty[s->b_code] + MAX_MV;
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], 0);
}else
fimin= bimin= INT_MAX;
}
score= ((unsigned)(score*score + 128*256))>>16;
- s->mc_mb_var_sum_temp += score;
+ c->mc_mb_var_sum_temp += score;
s->current_picture.mc_mb_var[mb_y*s->mb_stride + mb_x] = score; //FIXME use SSE
}
- if(s->avctx->mb_decision > FF_MB_DECISION_SIMPLE){
+ if(c->avctx->mb_decision > FF_MB_DECISION_SIMPLE){
type= CANDIDATE_MB_TYPE_FORWARD | CANDIDATE_MB_TYPE_BACKWARD | CANDIDATE_MB_TYPE_BIDIR | CANDIDATE_MB_TYPE_DIRECT;
if(fimin < INT_MAX)
type |= CANDIDATE_MB_TYPE_FORWARD_I;
{
if(s->me_method>=ME_EPZS){
int score[8];
- int i, y;
+ int i, y, range= s->avctx->me_range;
uint8_t * fcode_tab= s->fcode_tab;
int best_fcode=-1;
int best_score=-10000000;
int xy= y*s->mb_stride;
for(x=0; x<s->mb_width; x++){
if(s->mb_type[xy] & type){
- int fcode= FFMAX(fcode_tab[mv_table[xy][0] + MAX_MV],
- fcode_tab[mv_table[xy][1] + MAX_MV]);
+ int mx= mv_table[xy][0];
+ int my= mv_table[xy][1];
+ int fcode= FFMAX(fcode_tab[mx + MAX_MV],
+ fcode_tab[my + MAX_MV]);
int j;
+ if(range){
+ if(mx >= range || mx < -range ||
+ my >= range || my < -range)
+ continue;
+ }
+
for(j=0; j<fcode && j<8; j++){
if(s->pict_type==B_TYPE || s->current_picture.mc_mb_var[xy] < s->current_picture.mb_var[xy])
score[j]-= 170;
void ff_fix_long_p_mvs(MpegEncContext * s)
{
+ MotionEstContext * const c= &s->me;
const int f_code= s->f_code;
int y, range;
assert(s->pict_type==P_TYPE);
if(s->msmpeg4_version) range= 16;
- if(s->avctx->me_range && range > s->avctx->me_range) range= s->avctx->me_range;
+ if(c->avctx->me_range && range > c->avctx->me_range) range= c->avctx->me_range;
//printf("%d no:%d %d//\n", clip, noclip, f_code);
if(s->flags&CODEC_FLAG_4MV){
void ff_fix_long_mvs(MpegEncContext * s, uint8_t *field_select_table, int field_select,
int16_t (*mv_table)[2], int f_code, int type, int truncate)
{
+ MotionEstContext * const c= &s->me;
int y, h_range, v_range;
// RAL: 8 in MPEG-1, 16 in MPEG-4
int range = (((s->out_format == FMT_MPEG1) ? 8 : 16) << f_code);
if(s->msmpeg4_version) range= 16;
- if(s->avctx->me_range && range > s->avctx->me_range) range= s->avctx->me_range;
+ if(c->avctx->me_range && range > c->avctx->me_range) range= c->avctx->me_range;
h_range= range;
v_range= field_select_table ? range>>1 : range;