* Copyright (c) 2002-2004 Michael Niedermayer
*
*
- * This library is free software; you can redistribute it and/or
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
- * version 2 of the License, or (at your option) any later version.
+ * version 2.1 of the License, or (at your option) any later version.
*
- * This library is distributed in the hope that it will be useful,
+ * FFmpeg is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
- * License along with this library; if not, write to the Free Software
- * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*
* new Motion Estimation (X1/EPZS) by Michael Niedermayer <michaelni@gmx.at>
*/
+ (chroma ? FLAG_CHROMA : 0);
}
-static always_inline int cmp(MpegEncContext *s, const int x, const int y, const int subx, const int suby,
+/*! \brief compares a block (either a full macroblock or a partition thereof)
+ against a proposed motion-compensated prediction of that block
+ */
+static av_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){
MotionEstContext * const c= &s->me;
int d;
//FIXME check chroma 4mv, (no crashes ...)
if(flags&FLAG_DIRECT){
+ assert(x >= c->xmin && hx <= c->xmax<<(qpel+1) && y >= c->ymin && hy <= c->ymax<<(qpel+1));
if(x >= c->xmin && hx <= c->xmax<<(qpel+1) && y >= c->ymin && hy <= c->ymax<<(qpel+1)){
const int time_pp= s->pp_time;
const int time_pb= s->pb_time;
void ff_init_me(MpegEncContext *s){
MotionEstContext * const c= &s->me;
+ int cache_size= FFMIN(ME_MAP_SIZE>>ME_MAP_SHIFT, 1<<ME_MAP_SHIFT);
+ int dia_size= FFMAX(FFABS(s->avctx->dia_size)&255, FFABS(s->avctx->pre_dia_size)&255);
c->avctx= s->avctx;
+ if(cache_size < 2*dia_size && !c->stride){
+ av_log(s->avctx, AV_LOG_INFO, "ME_MAP size may be a little small for the selected diamond size\n");
+ }
+
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);
s = 0;
for (i = 0; i < 16; i++) {
for (j = 0; j < 16; j += 8) {
- s += ABS(pix[0]-mean);
- s += ABS(pix[1]-mean);
- s += ABS(pix[2]-mean);
- s += ABS(pix[3]-mean);
- s += ABS(pix[4]-mean);
- s += ABS(pix[5]-mean);
- s += ABS(pix[6]-mean);
- s += ABS(pix[7]-mean);
+ s += FFABS(pix[0]-mean);
+ s += FFABS(pix[1]-mean);
+ s += FFABS(pix[2]-mean);
+ s += FFABS(pix[3]-mean);
+ s += FFABS(pix[4]-mean);
+ s += FFABS(pix[5]-mean);
+ s += FFABS(pix[6]-mean);
+ s += FFABS(pix[7]-mean);
pix += 8;
}
pix += line_size - 16;
#if 0
if (*mx_ptr < -(2 * range) || *mx_ptr >= (2 * range) ||
*my_ptr < -(2 * range) || *my_ptr >= (2 * range)) {
- fprintf(stderr, "error %d %d\n", *mx_ptr, *my_ptr);
+ av_log(NULL, AV_LOG_ERROR, "error %d %d\n", *mx_ptr, *my_ptr);
}
#endif
return dmin;
static inline void get_limits(MpegEncContext *s, int x, int y)
{
MotionEstContext * const c= &s->me;
+ int range= c->avctx->me_range >> (1 + !!(c->flags&FLAG_QPEL));
/*
if(c->avctx->me_range) c->range= c->avctx->me_range >> 1;
else c->range= 16;
c->xmax = - x + s->mb_width *16 - 16;
c->ymax = - y + s->mb_height*16 - 16;
}
+ if(range){
+ c->xmin = FFMAX(c->xmin,-range);
+ c->xmax = FFMIN(c->xmax, range);
+ c->ymin = FFMAX(c->ymin,-range);
+ c->ymax = FFMIN(c->ymax, range);
+ }
}
static inline void init_mv4_ref(MotionEstContext *c){
int16_t (*mv_table)[2]= mv_tables[block][field_select];
if(user_field_select){
+ assert(field_select==0 || field_select==1);
+ assert(field_select_tables[block][xy]==0 || field_select_tables[block][xy]==1);
if(field_select_tables[block][xy] != field_select)
continue;
}
{
MotionEstContext * const c= &s->me;
uint8_t *pix, *ppix;
- int sum, varc, vard, mx, my, dmin;
+ int sum, mx, my, dmin;
+ int varc; ///< the variance of the block (sum of squared (p[y][x]-average))
+ int vard; ///< sum of squared differences with the estimated motion vector
int P[10][2];
const int shift= 1+s->quarter_sample;
int mb_type=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;
+ varc = s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500;
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;
+ pic->mb_var [s->mb_stride * mb_y + mb_x] = (varc+128)>>8;
+ c->mb_var_sum_temp += (varc+128)>>8;
if(c->avctx->me_threshold){
- vard= (check_input_motion(s, mb_x, mb_y, 1)+128)>>8;
-
- if(vard<c->avctx->me_threshold){
- pic->mc_mb_var[s->mb_stride * mb_y + mb_x] = vard;
- c->mc_mb_var_sum_temp += vard;
- if (vard <= 64 || vard < varc) { //FIXME
- c->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);
- }else{
- c->scene_change_score+= s->qscale;
- }
+ vard= check_input_motion(s, mb_x, mb_y, 1);
+
+ if((vard+128)>>8 < c->avctx->me_threshold){
+ int p_score= FFMIN(vard, varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*100);
+ int i_score= varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*20;
+ pic->mc_mb_var[s->mb_stride * mb_y + mb_x] = (vard+128)>>8;
+ c->mc_mb_var_sum_temp += (vard+128)>>8;
+ c->scene_change_score+= ff_sqrt(p_score) - ff_sqrt(i_score);
return;
}
- if(vard<c->avctx->mb_threshold)
+ if((vard+128)>>8 < c->avctx->mb_threshold)
mb_type= s->mb_type[mb_x + mb_y*s->mb_stride];
}
/* At this point (mx,my) are full-pell and the relative displacement */
ppix = c->ref[0][0] + (my * s->linesize) + mx;
- vard = (s->dsp.sse[0](NULL, pix, ppix, s->linesize, 16)+128)>>8;
+ vard = s->dsp.sse[0](NULL, pix, ppix, s->linesize, 16);
- pic->mc_mb_var[s->mb_stride * mb_y + mb_x] = vard;
+ pic->mc_mb_var[s->mb_stride * mb_y + mb_x] = (vard+128)>>8;
// pic->mb_cmp_score[s->mb_stride * mb_y + mb_x] = dmin;
- c->mc_mb_var_sum_temp += vard;
+ c->mc_mb_var_sum_temp += (vard+128)>>8;
#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(mb_type){
- if (vard <= 64 || vard < varc)
- c->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);
- else
- c->scene_change_score+= s->qscale;
+ int p_score= FFMIN(vard, varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*100);
+ int i_score= varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*20;
+ c->scene_change_score+= ff_sqrt(p_score) - ff_sqrt(i_score);
if(mb_type == CANDIDATE_MB_TYPE_INTER){
c->sub_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
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;
+ int p_score= FFMIN(vard, varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*100);
+ int i_score= varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*20;
+ c->scene_change_score+= ff_sqrt(p_score) - ff_sqrt(i_score);
- if (vard*2 + 200 > varc)
+ if (vard*2 + 200*256 > varc)
mb_type|= CANDIDATE_MB_TYPE_INTRA;
- if (varc*2 + 200 > vard){
+ if (varc*2 + 200*256 > vard || s->qscale > 24){
+// if (varc*2 + 200*256 + 50*(s->lambda2>>FF_LAMBDA_SHIFT) > vard){
mb_type|= CANDIDATE_MB_TYPE_INTER;
c->sub_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
if(s->flags&CODEC_FLAG_MV0)
my <<=shift;
}
if((s->flags&CODEC_FLAG_4MV)
- && !c->skip && varc>50 && vard>10){
+ && !c->skip && varc>50<<8 && vard>10<<8){
if(h263_mv4_search(s, mx, my, shift) < INT_MAX)
mb_type|=CANDIDATE_MB_TYPE_INTER4V;
dmin= ff_get_mb_score(s, mx, my, 0, 0, 0, 16, 1);
if((s->flags&CODEC_FLAG_4MV)
- && !c->skip && varc>50 && vard>10){
+ && !c->skip && varc>50<<8 && vard>10<<8){
int dmin4= h263_mv4_search(s, mx, my, shift);
if(dmin4 < dmin){
mb_type= CANDIDATE_MB_TYPE_INTER4V;
/* get intra luma score */
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
+ intra_score= varc - 500;
}else{
int mean= (sum+128)>>8;
mean*= 0x01010101;
}else
s->current_picture.mb_type[mb_y*s->mb_stride + mb_x]= 0;
- if (vard <= 64 || vard < varc) { //FIXME
- c->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);
- }else{
- c->scene_change_score+= s->qscale;
+ {
+ int p_score= FFMIN(vard, varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*100);
+ int i_score= varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*20;
+ c->scene_change_score+= ff_sqrt(p_score) - ff_sqrt(i_score);
}
}
const int ymin= c->ymin<<shift;
const int xmax= c->xmax<<shift;
const int ymax= c->ymax<<shift;
+ uint8_t map[8][8][8][8];
+
+ memset(map,0,sizeof(map));
+#define BIDIR_MAP(fx,fy,bx,by) \
+ map[(motion_fx+fx)&7][(motion_fy+fy)&7][(motion_bx+bx)&7][(motion_by+by)&7]
+ BIDIR_MAP(0,0,0,0) = 1;
fbmin= check_bidir_mv(s, motion_fx, motion_fy,
motion_bx, motion_by,
if(s->avctx->bidir_refine){
int score, end;
#define CHECK_BIDIR(fx,fy,bx,by)\
- if( (fx<=0 || motion_fx+fx<=xmax) && (fy<=0 || motion_fy+fy<=ymax) && (bx<=0 || motion_bx+bx<=xmax) && (by<=0 || motion_by+by<=ymax)\
+ if( !BIDIR_MAP(fx,fy,bx,by)\
+ &&(fx<=0 || motion_fx+fx<=xmax) && (fy<=0 || motion_fy+fy<=ymax) && (bx<=0 || motion_bx+bx<=xmax) && (by<=0 || motion_by+by<=ymax)\
&&(fx>=0 || motion_fx+fx>=xmin) && (fy>=0 || motion_fy+fy>=ymin) && (bx>=0 || motion_bx+bx>=xmin) && (by>=0 || motion_by+by>=ymin)){\
+ BIDIR_MAP(fx,fy,bx,by) = 1;\
score= check_bidir_mv(s, motion_fx+fx, motion_fy+fy, motion_bx+bx, motion_by+by, pred_fx, pred_fy, pred_bx, pred_by, 0, 16);\
if(score < fbmin){\
fbmin= score;\
}
#define CHECK_BIDIR2(a,b,c,d)\
CHECK_BIDIR(a,b,c,d)\
-CHECK_BIDIR(-a,-b,-c,-d)
+CHECK_BIDIR(-(a),-(b),-(c),-(d))
#define CHECK_BIDIRR(a,b,c,d)\
CHECK_BIDIR2(a,b,c,d)\
}while(!end);
}
+ s->b_bidir_forw_mv_table[xy][0]= motion_fx;
+ s->b_bidir_forw_mv_table[xy][1]= motion_fy;
+ s->b_bidir_back_mv_table[xy][0]= motion_bx;
+ s->b_bidir_back_mv_table[xy][1]= motion_by;
+
return fbmin;
}
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_LEFT[0] = av_clip(mv_table[mot_xy - 1][0], xmin<<shift, xmax<<shift);
+ P_LEFT[1] = av_clip(mv_table[mot_xy - 1][1], ymin<<shift, ymax<<shift);
/* special case for first line */
- if (!s->first_slice_line) { //FIXME maybe allow this over thread boundary as its cliped
- P_TOP[0] = clip(mv_table[mot_xy - mot_stride ][0], xmin<<shift, xmax<<shift);
- P_TOP[1] = clip(mv_table[mot_xy - mot_stride ][1], ymin<<shift, ymax<<shift);
- P_TOPRIGHT[0] = clip(mv_table[mot_xy - mot_stride + 1 ][0], xmin<<shift, xmax<<shift);
- P_TOPRIGHT[1] = clip(mv_table[mot_xy - mot_stride + 1 ][1], ymin<<shift, ymax<<shift);
+ if (!s->first_slice_line) { //FIXME maybe allow this over thread boundary as its clipped
+ P_TOP[0] = av_clip(mv_table[mot_xy - mot_stride ][0], xmin<<shift, xmax<<shift);
+ P_TOP[1] = av_clip(mv_table[mot_xy - mot_stride ][1], ymin<<shift, ymax<<shift);
+ P_TOPRIGHT[0] = av_clip(mv_table[mot_xy - mot_stride + 1 ][0], xmin<<shift, xmax<<shift);
+ P_TOPRIGHT[1] = av_clip(mv_table[mot_xy - mot_stride + 1 ][1], ymin<<shift, 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]);
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;
+ mv_table[mot_xy][0]= mx;
+ mv_table[mot_xy][1]= my;
c->flags &= ~FLAG_DIRECT;
c->sub_flags &= ~FLAG_DIRECT;
get_limits(s, 16*mb_x, 16*mb_y);
c->skip=0;
+
+ if(s->codec_id == CODEC_ID_MPEG4 && s->next_picture.mbskip_table[xy]){
+ int score= direct_search(s, mb_x, mb_y); //FIXME just check 0,0
+
+ score= ((unsigned)(score*score + 128*256))>>16;
+ c->mc_mb_var_sum_temp += score;
+ s->current_picture.mc_mb_var[mb_y*s->mb_stride + mb_x] = score; //FIXME use SSE
+ s->mb_type[mb_y*s->mb_stride + mb_x]= CANDIDATE_MB_TYPE_DIRECT0;
+
+ return;
+ }
+
if(c->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);
- if(vard<c->avctx->me_threshold){
+ if((vard+128)>>8 < 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;
+// varc = s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500;
-// 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_var [s->mb_stride * mb_y + mb_x] = (varc+128)>>8;
+ s->current_picture.mc_mb_var[s->mb_stride * mb_y + mb_x] = (vard+128)>>8;
/* pic->mb_mean [s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
- c->mb_var_sum_temp += varc;*/
- c->mc_mb_var_sum_temp += vard;
-/* if (vard <= 64 || vard < varc) {
+ c->mb_var_sum_temp += (varc+128)>>8;*/
+ c->mc_mb_var_sum_temp += (vard+128)>>8;
+/* if (vard <= 64<<8 || vard < varc) {
c->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);
}else{
- c->scene_change_score+= s->qscale;
+ c->scene_change_score+= s->qscale * s->avctx->scenechange_factor;
}*/
return;
}
- if(vard<c->avctx->mb_threshold){
+ if((vard+128)>>8 < 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);
}
//FIXME something smarter
if(dmin>256*256*16) type&= ~CANDIDATE_MB_TYPE_DIRECT; //dont try direct mode if its invalid for this MB
+ if(s->codec_id == CODEC_ID_MPEG4 && type&CANDIDATE_MB_TYPE_DIRECT && s->flags&CODEC_FLAG_MV0 && *(uint32_t*)s->b_direct_mv_table[xy])
+ type |= CANDIDATE_MB_TYPE_DIRECT0;
#if 0
if(s->out_format == FMT_MPEG1)
type |= CANDIDATE_MB_TYPE_INTRA;