summaryrefslogtreecommitdiff
path: root/libavcodec/motion_est_template.c
blob: 45e87a0becf29c070fa171d233872ac3001355ca (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
/*
 * Motion estimation
 * Copyright (c) 2002-2004 Michael Niedermayer
 *
 * This file is part of Libav.
 *
 * Libav 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 2.1 of the License, or (at your option) any later version.
 *
 * Libav 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 Libav; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

/**
 * @file
 * Motion estimation template.
 */

//Let us hope gcc will remove the unused vars ...(gcc 3.2.2 seems to do it ...)
#define LOAD_COMMON\
    uint32_t av_unused * const score_map= c->score_map;\
    const int av_unused xmin= c->xmin;\
    const int av_unused ymin= c->ymin;\
    const int av_unused xmax= c->xmax;\
    const int av_unused ymax= c->ymax;\
    uint8_t *mv_penalty= c->current_mv_penalty;\
    const int pred_x= c->pred_x;\
    const int pred_y= c->pred_y;\

#define CHECK_HALF_MV(dx, dy, x, y)\
{\
    const int hx= 2*(x)+(dx);\
    const int hy= 2*(y)+(dy);\
    d= cmp_hpel(s, x, y, dx, dy, size, h, ref_index, src_index, cmp_sub, chroma_cmp_sub, flags);\
    d += (mv_penalty[hx - pred_x] + mv_penalty[hy - pred_y])*penalty_factor;\
    COPY3_IF_LT(dmin, d, bx, hx, by, hy)\
}

#if 0
static int hpel_motion_search)(MpegEncContext * s,
                                  int *mx_ptr, int *my_ptr, int dmin,
                                  uint8_t *ref_data[3],
                                  int size)
{
    const int xx = 16 * s->mb_x + 8*(n&1);
    const int yy = 16 * s->mb_y + 8*(n>>1);
    const int mx = *mx_ptr;
    const int my = *my_ptr;
    const int penalty_factor= c->sub_penalty_factor;

    LOAD_COMMON

 //   INIT;
 //FIXME factorize
    me_cmp_func cmp, chroma_cmp, cmp_sub, chroma_cmp_sub;

    if(s->no_rounding /*FIXME b_type*/){
        hpel_put= &s->dsp.put_no_rnd_pixels_tab[size];
        chroma_hpel_put= &s->dsp.put_no_rnd_pixels_tab[size+1];
    }else{
        hpel_put=& s->dsp.put_pixels_tab[size];
        chroma_hpel_put= &s->dsp.put_pixels_tab[size+1];
    }
    cmpf= s->dsp.me_cmp[size];
    chroma_cmpf= s->dsp.me_cmp[size+1];
    cmp_sub= s->dsp.me_sub_cmp[size];
    chroma_cmp_sub= s->dsp.me_sub_cmp[size+1];

    if(c->skip){ //FIXME somehow move up (benchmark)
        *mx_ptr = 0;
        *my_ptr = 0;
        return dmin;
    }

    if(c->avctx->me_cmp != c->avctx->me_sub_cmp){
        CMP_HPEL(dmin, 0, 0, mx, my, size);
        if(mx || my)
            dmin += (mv_penalty[2*mx - pred_x] + mv_penalty[2*my - pred_y])*penalty_factor;
    }

    if (mx > xmin && mx < xmax &&
        my > ymin && my < ymax) {
        int bx=2*mx, by=2*my;
        int d= dmin;

        CHECK_HALF_MV(1, 1, mx-1, my-1)
        CHECK_HALF_MV(0, 1, mx  , my-1)
        CHECK_HALF_MV(1, 1, mx  , my-1)
        CHECK_HALF_MV(1, 0, mx-1, my  )
        CHECK_HALF_MV(1, 0, mx  , my  )
        CHECK_HALF_MV(1, 1, mx-1, my  )
        CHECK_HALF_MV(0, 1, mx  , my  )
        CHECK_HALF_MV(1, 1, mx  , my  )

        assert(bx >= xmin*2 || bx <= xmax*2 || by >= ymin*2 || by <= ymax*2);

        *mx_ptr = bx;
        *my_ptr = by;
    }else{
        *mx_ptr =2*mx;
        *my_ptr =2*my;
    }

    return dmin;
}

#else
static int hpel_motion_search(MpegEncContext * s,
                                  int *mx_ptr, int *my_ptr, int dmin,
                                  int src_index, int ref_index,
                                  int size, int h)
{
    MotionEstContext * const c= &s->me;
    const int mx = *mx_ptr;
    const int my = *my_ptr;
    const int penalty_factor= c->sub_penalty_factor;
    me_cmp_func cmp_sub, chroma_cmp_sub;
    int bx=2*mx, by=2*my;

    LOAD_COMMON
    int flags= c->sub_flags;

 //FIXME factorize

    cmp_sub= s->dsp.me_sub_cmp[size];
    chroma_cmp_sub= s->dsp.me_sub_cmp[size+1];

    if(c->skip){ //FIXME move out of hpel?
        *mx_ptr = 0;
        *my_ptr = 0;
        return dmin;
    }

    if(c->avctx->me_cmp != c->avctx->me_sub_cmp){
        dmin= cmp(s, mx, my, 0, 0, size, h, ref_index, src_index, cmp_sub, chroma_cmp_sub, flags);
        if(mx || my || size>0)
            dmin += (mv_penalty[2*mx - pred_x] + mv_penalty[2*my - pred_y])*penalty_factor;
    }

    if (mx > xmin && mx < xmax &&
        my > ymin && my < ymax) {
        int d= dmin;
        const int index= (my<<ME_MAP_SHIFT) + mx;
        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])*c->penalty_factor;
        const int l= score_map[(index- 1               )&(ME_MAP_SIZE-1)]
                     + (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)]
                     + (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)]
                     + (mv_penalty[bx   - pred_x] + mv_penalty[by+2 - pred_y])*c->penalty_factor;

        int key;
        int map_generation= c->map_generation;
#ifndef NDEBUG
        uint32_t *map= c->map;
#endif
        key= ((my-1)<<ME_MAP_MV_BITS) + (mx) + map_generation;
        assert(map[(index-(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)] == key);
        key= ((my+1)<<ME_MAP_MV_BITS) + (mx) + map_generation;
        assert(map[(index+(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)] == key);
        key= ((my)<<ME_MAP_MV_BITS) + (mx+1) + map_generation;
        assert(map[(index+1)&(ME_MAP_SIZE-1)] == key);
        key= ((my)<<ME_MAP_MV_BITS) + (mx-1) + map_generation;
        assert(map[(index-1)&(ME_MAP_SIZE-1)] == key);
        if(t<=b){
            CHECK_HALF_MV(0, 1, mx  ,my-1)
            if(l<=r){
                CHECK_HALF_MV(1, 1, mx-1, my-1)
                if(t+r<=b+l){
                    CHECK_HALF_MV(1, 1, mx  , my-1)
                }else{
                    CHECK_HALF_MV(1, 1, mx-1, my  )
                }
                CHECK_HALF_MV(1, 0, mx-1, my  )
            }else{
                CHECK_HALF_MV(1, 1, mx  , my-1)
                if(t+l<=b+r){
                    CHECK_HALF_MV(1, 1, mx-1, my-1)
                }else{
                    CHECK_HALF_MV(1, 1, mx  , my  )
                }
                CHECK_HALF_MV(1, 0, mx  , my  )
            }
        }else{
            if(l<=r){
                if(t+l<=b+r){
                    CHECK_HALF_MV(1, 1, mx-1, my-1)
                }else{
                    CHECK_HALF_MV(1, 1, mx  , my  )
                }
                CHECK_HALF_MV(1, 0, mx-1, my)
                CHECK_HALF_MV(1, 1, mx-1, my)
            }else{
                if(t+r<=b+l){
                    CHECK_HALF_MV(1, 1, mx  , my-1)
                }else{
                    CHECK_HALF_MV(1, 1, mx-1, my)
                }
                CHECK_HALF_MV(1, 0, mx  , my)
                CHECK_HALF_MV(1, 1, mx  , my)
            }
            CHECK_HALF_MV(0, 1, mx  , my)
        }
        assert(bx >= xmin*2 && bx <= xmax*2 && by >= ymin*2 && by <= ymax*2);
    }

    *mx_ptr = bx;
    *my_ptr = by;

    return dmin;
}
#endif

static int no_sub_motion_search(MpegEncContext * s,
          int *mx_ptr, int *my_ptr, int dmin,
                                  int src_index, int ref_index,
                                  int size, int h)
{
    (*mx_ptr)<<=1;
    (*my_ptr)<<=1;
    return dmin;
}

inline int ff_get_mb_score(MpegEncContext * s, int mx, int my, int src_index,
                               int ref_index, int size, int h, int add_rate)
{
//    const int check_luma= s->dsp.me_sub_cmp != s->dsp.mb_cmp;
    MotionEstContext * const c= &s->me;
    const int penalty_factor= c->mb_penalty_factor;
    const int flags= c->mb_flags;
    const int qpel= flags & FLAG_QPEL;
    const int mask= 1+2*qpel;
    me_cmp_func cmp_sub, chroma_cmp_sub;
    int d;

    LOAD_COMMON

 //FIXME factorize

    cmp_sub= s->dsp.mb_cmp[size];
    chroma_cmp_sub= s->dsp.mb_cmp[size+1];

//    assert(!c->skip);
//    assert(c->avctx->me_sub_cmp != c->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);
    //FIXME check cbp before adding penalty for (0,0) vector
    if(add_rate && (mx || my || size>0))
        d += (mv_penalty[mx - pred_x] + mv_penalty[my - pred_y])*penalty_factor;

    return d;
}

#define CHECK_QUARTER_MV(dx, dy, x, y)\
{\
    const int hx= 4*(x)+(dx);\
    const int hy= 4*(y)+(dy);\
    d= cmp_qpel(s, x, y, dx, dy, size, h, ref_index, src_index, cmpf, chroma_cmpf, flags);\
    d += (mv_penalty[hx - pred_x] + mv_penalty[hy - pred_y])*penalty_factor;\
    COPY3_IF_LT(dmin, d, bx, hx, by, hy)\
}

static int qpel_motion_search(MpegEncContext * s,
                                  int *mx_ptr, int *my_ptr, int dmin,
                                  int src_index, int ref_index,
                                  int size, int h)
{
    MotionEstContext * const c= &s->me;
    const int mx = *mx_ptr;
    const int my = *my_ptr;
    const int penalty_factor= c->sub_penalty_factor;
    const int map_generation= c->map_generation;
    const int subpel_quality= c->avctx->me_subpel_quality;
    uint32_t *map= c->map;
    me_cmp_func cmpf, chroma_cmpf;
    me_cmp_func cmp_sub, chroma_cmp_sub;

    LOAD_COMMON
    int flags= c->sub_flags;

    cmpf= s->dsp.me_cmp[size];
    chroma_cmpf= s->dsp.me_cmp[size+1]; //factorize FIXME
 //FIXME factorize

    cmp_sub= s->dsp.me_sub_cmp[size];
    chroma_cmp_sub= s->dsp.me_sub_cmp[size+1];

    if(c->skip){ //FIXME somehow move up (benchmark)
        *mx_ptr = 0;
        *my_ptr = 0;
        return dmin;
    }

    if(c->avctx->me_cmp != c->avctx->me_sub_cmp){
        dmin= cmp(s, mx, my, 0, 0, size, h, ref_index, src_index, cmp_sub, chroma_cmp_sub, flags);
        if(mx || my || size>0)
            dmin += (mv_penalty[4*mx - pred_x] + mv_penalty[4*my - pred_y])*penalty_factor;
    }

    if (mx > xmin && mx < xmax &&
        my > ymin && my < ymax) {
        int bx=4*mx, by=4*my;
        int d= dmin;
        int i, nx, ny;
        const int index= (my<<ME_MAP_SHIFT) + mx;
        const int t= score_map[(index-(1<<ME_MAP_SHIFT)  )&(ME_MAP_SIZE-1)];
        const int l= score_map[(index- 1                 )&(ME_MAP_SIZE-1)];
        const int r= score_map[(index+ 1                 )&(ME_MAP_SIZE-1)];
        const int b= score_map[(index+(1<<ME_MAP_SHIFT)  )&(ME_MAP_SIZE-1)];
        const int c= score_map[(index                    )&(ME_MAP_SIZE-1)];
        int best[8];
        int best_pos[8][2];

        memset(best, 64, sizeof(int)*8);
#if 1
        if(s->me.dia_size>=2){
            const int tl= score_map[(index-(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)];
            const int bl= score_map[(index+(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)];
            const int tr= score_map[(index-(1<<ME_MAP_SHIFT)+1)&(ME_MAP_SIZE-1)];
            const int br= score_map[(index+(1<<ME_MAP_SHIFT)+1)&(ME_MAP_SIZE-1)];

            for(ny= -3; ny <= 3; ny++){
                for(nx= -3; nx <= 3; nx++){
                    //FIXME this could overflow (unlikely though)
                    const int64_t t2= nx*nx*(tr + tl - 2*t) + 4*nx*(tr-tl) + 32*t;
                    const int64_t c2= nx*nx*( r +  l - 2*c) + 4*nx*( r- l) + 32*c;
                    const int64_t b2= nx*nx*(br + bl - 2*b) + 4*nx*(br-bl) + 32*b;
                    int score= (ny*ny*(b2 + t2 - 2*c2) + 4*ny*(b2 - t2) + 32*c2 + 512)>>10;
                    int i;

                    if((nx&3)==0 && (ny&3)==0) continue;

                    score += (mv_penalty[4*mx + nx - pred_x] + mv_penalty[4*my + ny - pred_y])*penalty_factor;

//                    if(nx&1) score-=1024*c->penalty_factor;
//                    if(ny&1) score-=1024*c->penalty_factor;

                    for(i=0; i<8; i++){
                        if(score < best[i]){
                            memmove(&best[i+1], &best[i], sizeof(int)*(7-i));
                            memmove(&best_pos[i+1][0], &best_pos[i][0], sizeof(int)*2*(7-i));
                            best[i]= score;
                            best_pos[i][0]= nx + 4*mx;
                            best_pos[i][1]= ny + 4*my;
                            break;
                        }
                    }
                }
            }
        }else{
            int tl;
            //FIXME this could overflow (unlikely though)
            const int cx = 4*(r - l);
            const int cx2= r + l - 2*c;
            const int cy = 4*(b - t);
            const int cy2= b + t - 2*c;
            int cxy;

            if(map[(index-(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)] == (my<<ME_MAP_MV_BITS) + mx + map_generation && 0){ //FIXME
                tl= score_map[(index-(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)];
            }else{
                tl= cmp(s, mx-1, my-1, 0, 0, size, h, ref_index, src_index, cmpf, chroma_cmpf, flags);//FIXME wrong if chroma me is different
            }

            cxy= 2*tl + (cx + cy)/4 - (cx2 + cy2) - 2*c;

            assert(16*cx2 + 4*cx + 32*c == 32*r);
            assert(16*cx2 - 4*cx + 32*c == 32*l);
            assert(16*cy2 + 4*cy + 32*c == 32*b);
            assert(16*cy2 - 4*cy + 32*c == 32*t);
            assert(16*cxy + 16*cy2 + 16*cx2 - 4*cy - 4*cx + 32*c == 32*tl);

            for(ny= -3; ny <= 3; ny++){
                for(nx= -3; nx <= 3; nx++){
                    //FIXME this could overflow (unlikely though)
                    int score= ny*nx*cxy + nx*nx*cx2 + ny*ny*cy2 + nx*cx + ny*cy + 32*c; //FIXME factor
                    int i;

                    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;
//                    if(nx&1) score-=32*c->penalty_factor;
  //                  if(ny&1) score-=32*c->penalty_factor;

                    for(i=0; i<8; i++){
                        if(score < best[i]){
                            memmove(&best[i+1], &best[i], sizeof(int)*(7-i));
                            memmove(&best_pos[i+1][0], &best_pos[i][0], sizeof(int)*2*(7-i));
                            best[i]= score;
                            best_pos[i][0]= nx + 4*mx;
                            best_pos[i][1]= ny + 4*my;
                            break;
                        }
                    }
                }
            }
        }
        for(i=0; i<subpel_quality; i++){
            nx= best_pos[i][0];
            ny= best_pos[i][1];
            CHECK_QUARTER_MV(nx&3, ny&3, nx>>2, ny>>2)
        }

#if 0
            const int tl= score_map[(index-(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)];
            const int bl= score_map[(index+(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)];
            const int tr= score_map[(index-(1<<ME_MAP_SHIFT)+1)&(ME_MAP_SIZE-1)];
            const int br= score_map[(index+(1<<ME_MAP_SHIFT)+1)&(ME_MAP_SIZE-1)];
//            if(l < r && l < t && l < b && l < tl && l < bl && l < tr && l < br && bl < tl){
            if(tl<br){

//            nx= FFMAX(4*mx - bx, bx - 4*mx);
//            ny= FFMAX(4*my - by, by - 4*my);

            static int stats[7][7], count;
            count++;
            stats[4*mx - bx + 3][4*my - by + 3]++;
            if(256*256*256*64 % count ==0){
                for(i=0; i<49; i++){
                    if((i%7)==0) printf("\n");
                    printf("%6d ", stats[0][i]);
                }
                printf("\n");
            }
            }
#endif
#else

        CHECK_QUARTER_MV(2, 2, mx-1, my-1)
        CHECK_QUARTER_MV(0, 2, mx  , my-1)
        CHECK_QUARTER_MV(2, 2, mx  , my-1)
        CHECK_QUARTER_MV(2, 0, mx  , my  )
        CHECK_QUARTER_MV(2, 2, mx  , my  )
        CHECK_QUARTER_MV(0, 2, mx  , my  )
        CHECK_QUARTER_MV(2, 2, mx-1, my  )
        CHECK_QUARTER_MV(2, 0, mx-1, my  )

        nx= bx;
        ny= by;

        for(i=0; i<8; i++){
            int ox[8]= {0, 1, 1, 1, 0,-1,-1,-1};
            int oy[8]= {1, 1, 0,-1,-1,-1, 0, 1};
            CHECK_QUARTER_MV((nx + ox[i])&3, (ny + oy[i])&3, (nx + ox[i])>>2, (ny + oy[i])>>2)
        }
#endif
#if 0
        //outer ring
        CHECK_QUARTER_MV(1, 3, mx-1, my-1)
        CHECK_QUARTER_MV(1, 2, mx-1, my-1)
        CHECK_QUARTER_MV(1, 1, mx-1, my-1)
        CHECK_QUARTER_MV(2, 1, mx-1, my-1)
        CHECK_QUARTER_MV(3, 1, mx-1, my-1)
        CHECK_QUARTER_MV(0, 1, mx  , my-1)
        CHECK_QUARTER_MV(1, 1, mx  , my-1)
        CHECK_QUARTER_MV(2, 1, mx  , my-1)
        CHECK_QUARTER_MV(3, 1, mx  , my-1)
        CHECK_QUARTER_MV(3, 2, mx  , my-1)
        CHECK_QUARTER_MV(3, 3, mx  , my-1)
        CHECK_QUARTER_MV(3, 0, mx  , my  )
        CHECK_QUARTER_MV(3, 1, mx  , my  )
        CHECK_QUARTER_MV(3, 2, mx  , my  )
        CHECK_QUARTER_MV(3, 3, mx  , my  )
        CHECK_QUARTER_MV(2, 3, mx  , my  )
        CHECK_QUARTER_MV(1, 3, mx  , my  )
        CHECK_QUARTER_MV(0, 3, mx  , my  )
        CHECK_QUARTER_MV(3, 3, mx-1, my  )
        CHECK_QUARTER_MV(2, 3, mx-1, my  )
        CHECK_QUARTER_MV(1, 3, mx-1, my  )
        CHECK_QUARTER_MV(1, 2, mx-1, my  )
        CHECK_QUARTER_MV(1, 1, mx-1, my  )
        CHECK_QUARTER_MV(1, 0, mx-1, my  )
#endif
        assert(bx >= xmin*4 && bx <= xmax*4 && by >= ymin*4 && by <= ymax*4);

        *mx_ptr = bx;
        *my_ptr = by;
    }else{
        *mx_ptr =4*mx;
        *my_ptr =4*my;
    }

    return dmin;
}


#define CHECK_MV(x,y)\
{\
    const int key= ((y)<<ME_MAP_MV_BITS) + (x) + map_generation;\
    const int index= (((y)<<ME_MAP_SHIFT) + (x))&(ME_MAP_SIZE-1);\
    assert((x) >= xmin);\
    assert((x) <= xmax);\
    assert((y) >= ymin);\
    assert((y) <= ymax);\
/*printf("check_mv %d %d\n", x, y);*/\
    if(map[index]!=key){\
        d= cmp(s, x, y, 0, 0, size, h, ref_index, src_index, cmpf, chroma_cmpf, flags);\
        map[index]= key;\
        score_map[index]= d;\
        d += (mv_penalty[((x)<<shift)-pred_x] + mv_penalty[((y)<<shift)-pred_y])*penalty_factor;\
/*printf("score:%d\n", d);*/\
        COPY3_IF_LT(dmin, d, best[0], x, best[1], y)\
    }\
}

#define CHECK_CLIPPED_MV(ax,ay)\
{\
    const int Lx= ax;\
    const int Ly= ay;\
    const int Lx2= FFMAX(xmin, FFMIN(Lx, xmax));\
    const int Ly2= FFMAX(ymin, FFMIN(Ly, ymax));\
    CHECK_MV(Lx2, Ly2)\
}

#define CHECK_MV_DIR(x,y,new_dir)\
{\
    const int key= ((y)<<ME_MAP_MV_BITS) + (x) + map_generation;\
    const int index= (((y)<<ME_MAP_SHIFT) + (x))&(ME_MAP_SIZE-1);\
/*printf("check_mv_dir %d %d %d\n", x, y, new_dir);*/\
    if(map[index]!=key){\
        d= cmp(s, x, y, 0, 0, size, h, ref_index, src_index, cmpf, chroma_cmpf, flags);\
        map[index]= key;\
        score_map[index]= d;\
        d += (mv_penalty[((x)<<shift)-pred_x] + mv_penalty[((y)<<shift)-pred_y])*penalty_factor;\
/*printf("score:%d\n", d);*/\
        if(d<dmin){\
            best[0]=x;\
            best[1]=y;\
            dmin=d;\
            next_dir= new_dir;\
        }\
    }\
}

#define check(x,y,S,v)\
if( (x)<(xmin<<(S)) ) printf("%d %d %d %d %d xmin" #v, xmin, (x), (y), s->mb_x, s->mb_y);\
if( (x)>(xmax<<(S)) ) printf("%d %d %d %d %d xmax" #v, xmax, (x), (y), s->mb_x, s->mb_y);\
if( (y)<(ymin<<(S)) ) printf("%d %d %d %d %d ymin" #v, ymin, (x), (y), s->mb_x, s->mb_y);\
if( (y)>(ymax<<(S)) ) printf("%d %d %d %d %d ymax" #v, ymax, (x), (y), s->mb_x, s->mb_y);\

#define LOAD_COMMON2\
    uint32_t *map= c->map;\
    const int qpel= flags&FLAG_QPEL;\
    const int shift= 1+qpel;\

static av_always_inline int small_diamond_search(MpegEncContext * s, int *best, int dmin,
                                       int src_index, int ref_index, int const penalty_factor,
                                       int size, int h, int flags)
{
    MotionEstContext * const c= &s->me;
    me_cmp_func cmpf, chroma_cmpf;
    int next_dir=-1;
    LOAD_COMMON
    LOAD_COMMON2
    int map_generation= c->map_generation;

    cmpf= s->dsp.me_cmp[size];
    chroma_cmpf= s->dsp.me_cmp[size+1];

    { /* ensure that the best point is in the MAP as h/qpel refinement needs it */
        const int key= (best[1]<<ME_MAP_MV_BITS) + best[0] + map_generation;
        const int index= ((best[1]<<ME_MAP_SHIFT) + best[0])&(ME_MAP_SIZE-1);
        if(map[index]!=key){ //this will be executed only very rarey
            score_map[index]= cmp(s, best[0], best[1], 0, 0, size, h, ref_index, src_index, cmpf, chroma_cmpf, flags);
            map[index]= key;
        }
    }

    for(;;){
        int d;
        const int dir= next_dir;
        const int x= best[0];
        const int y= best[1];
        next_dir=-1;

//printf("%d", dir);
        if(dir!=2 && x>xmin) CHECK_MV_DIR(x-1, y  , 0)
        if(dir!=3 && y>ymin) CHECK_MV_DIR(x  , y-1, 1)
        if(dir!=0 && x<xmax) CHECK_MV_DIR(x+1, y  , 2)
        if(dir!=1 && y<ymax) CHECK_MV_DIR(x  , y+1, 3)

        if(next_dir==-1){
            return dmin;
        }
    }
}

static int funny_diamond_search(MpegEncContext * s, int *best, int dmin,
                                       int src_index, int ref_index, int const penalty_factor,
                                       int size, int h, int flags)
{
    MotionEstContext * const c= &s->me;
    me_cmp_func cmpf, chroma_cmpf;
    int dia_size;
    LOAD_COMMON
    LOAD_COMMON2
    int map_generation= c->map_generation;

    cmpf= s->dsp.me_cmp[size];
    chroma_cmpf= s->dsp.me_cmp[size+1];

    for(dia_size=1; dia_size<=4; dia_size++){
        int dir;
        const int x= best[0];
        const int y= best[1];

        if(dia_size&(dia_size-1)) continue;

        if(   x + dia_size > xmax
           || x - dia_size < xmin
           || y + dia_size > ymax
           || y - dia_size < ymin)
           continue;

        for(dir= 0; dir<dia_size; dir+=2){
            int d;

            CHECK_MV(x + dir           , y + dia_size - dir);
            CHECK_MV(x + dia_size - dir, y - dir           );
            CHECK_MV(x - dir           , y - dia_size + dir);
            CHECK_MV(x - dia_size + dir, y + dir           );
        }

        if(x!=best[0] || y!=best[1])
            dia_size=0;
    }
    return dmin;
}

static int hex_search(MpegEncContext * s, int *best, int dmin,
                                       int src_index, int ref_index, int const penalty_factor,
                                       int size, int h, int flags, int dia_size)
{
    MotionEstContext * const c= &s->me;
    me_cmp_func cmpf, chroma_cmpf;
    LOAD_COMMON
    LOAD_COMMON2
    int map_generation= c->map_generation;
    int x,y,d;
    const int dec= dia_size & (dia_size-1);

    cmpf= s->dsp.me_cmp[size];
    chroma_cmpf= s->dsp.me_cmp[size+1];

    for(;dia_size; dia_size= dec ? dia_size-1 : dia_size>>1){
        do{
            x= best[0];
            y= best[1];

            CHECK_CLIPPED_MV(x  -dia_size    , y);
            CHECK_CLIPPED_MV(x+  dia_size    , y);
            CHECK_CLIPPED_MV(x+( dia_size>>1), y+dia_size);
            CHECK_CLIPPED_MV(x+( dia_size>>1), y-dia_size);
            if(dia_size>1){
                CHECK_CLIPPED_MV(x+(-dia_size>>1), y+dia_size);
                CHECK_CLIPPED_MV(x+(-dia_size>>1), y-dia_size);
            }
        }while(best[0] != x || best[1] != y);
    }

    return dmin;
}

static int l2s_dia_search(MpegEncContext * s, int *best, int dmin,
                                       int src_index, int ref_index, int const penalty_factor,
                                       int size, int h, int flags)
{
    MotionEstContext * const c= &s->me;
    me_cmp_func cmpf, chroma_cmpf;
    LOAD_COMMON
    LOAD_COMMON2
    int map_generation= c->map_generation;
    int x,y,i,d;
    int dia_size= c->dia_size&0xFF;
    const int dec= dia_size & (dia_size-1);
    static const int hex[8][2]={{-2, 0}, {-1,-1}, { 0,-2}, { 1,-1},
                                { 2, 0}, { 1, 1}, { 0, 2}, {-1, 1}};

    cmpf= s->dsp.me_cmp[size];
    chroma_cmpf= s->dsp.me_cmp[size+1];

    for(; dia_size; dia_size= dec ? dia_size-1 : dia_size>>1){
        do{
            x= best[0];
            y= best[1];
            for(i=0; i<8; i++){
                CHECK_CLIPPED_MV(x+hex[i][0]*dia_size, y+hex[i][1]*dia_size);
            }
        }while(best[0] != x || best[1] != y);
    }

    x= best[0];
    y= best[1];
    CHECK_CLIPPED_MV(x+1, y);
    CHECK_CLIPPED_MV(x, y+1);
    CHECK_CLIPPED_MV(x-1, y);
    CHECK_CLIPPED_MV(x, y-1);

    return dmin;
}

static int umh_search(MpegEncContext * s, int *best, int dmin,
                                       int src_index, int ref_index, int const penalty_factor,
                                       int size, int h, int flags)
{
    MotionEstContext * const c= &s->me;
    me_cmp_func cmpf, chroma_cmpf;
    LOAD_COMMON
    LOAD_COMMON2
    int map_generation= c->map_generation;
    int x,y,x2,y2, i, j, d;
    const int dia_size= c->dia_size&0xFE;
    static const int hex[16][2]={{-4,-2}, {-4,-1}, {-4, 0}, {-4, 1}, {-4, 2},
                                 { 4,-2}, { 4,-1}, { 4, 0}, { 4, 1}, { 4, 2},
                                 {-2, 3}, { 0, 4}, { 2, 3},
                                 {-2,-3}, { 0,-4}, { 2,-3},};

    cmpf= s->dsp.me_cmp[size];
    chroma_cmpf= s->dsp.me_cmp[size+1];

    x= best[0];
    y= best[1];
    for(x2=FFMAX(x-dia_size+1, xmin); x2<=FFMIN(x+dia_size-1,xmax); x2+=2){
        CHECK_MV(x2, y);
    }
    for(y2=FFMAX(y-dia_size/2+1, ymin); y2<=FFMIN(y+dia_size/2-1,ymax); y2+=2){
        CHECK_MV(x, y2);
    }

    x= best[0];
    y= best[1];
    for(y2=FFMAX(y-2, ymin); y2<=FFMIN(y+2,ymax); y2++){
        for(x2=FFMAX(x-2, xmin); x2<=FFMIN(x+2,xmax); x2++){
            CHECK_MV(x2, y2);
        }
    }

//FIXME prevent the CLIP stuff

    for(j=1; j<=dia_size/4; j++){
        for(i=0; i<16; i++){
            CHECK_CLIPPED_MV(x+hex[i][0]*j, y+hex[i][1]*j);
        }
    }

    return hex_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags, 2);
}

static int full_search(MpegEncContext * s, int *best, int dmin,
                                       int src_index, int ref_index, int const penalty_factor,
                                       int size, int h, int flags)
{
    MotionEstContext * const c= &s->me;
    me_cmp_func cmpf, chroma_cmpf;
    LOAD_COMMON
    LOAD_COMMON2
    int map_generation= c->map_generation;
    int x,y, d;
    const int dia_size= c->dia_size&0xFF;

    cmpf= s->dsp.me_cmp[size];
    chroma_cmpf= s->dsp.me_cmp[size+1];

    for(y=FFMAX(-dia_size, ymin); y<=FFMIN(dia_size,ymax); y++){
        for(x=FFMAX(-dia_size, xmin); x<=FFMIN(dia_size,xmax); x++){
            CHECK_MV(x, y);
        }
    }

    x= best[0];
    y= best[1];
    d= dmin;
    CHECK_CLIPPED_MV(x  , y);
    CHECK_CLIPPED_MV(x+1, y);
    CHECK_CLIPPED_MV(x, y+1);
    CHECK_CLIPPED_MV(x-1, y);
    CHECK_CLIPPED_MV(x, y-1);
    best[0]= x;
    best[1]= y;

    return d;
}

#define SAB_CHECK_MV(ax,ay)\
{\
    const int key= ((ay)<<ME_MAP_MV_BITS) + (ax) + map_generation;\
    const int index= (((ay)<<ME_MAP_SHIFT) + (ax))&(ME_MAP_SIZE-1);\
/*printf("sab check %d %d\n", ax, ay);*/\
    if(map[index]!=key){\
        d= cmp(s, ax, ay, 0, 0, size, h, ref_index, src_index, cmpf, chroma_cmpf, flags);\
        map[index]= key;\
        score_map[index]= d;\
        d += (mv_penalty[((ax)<<shift)-pred_x] + mv_penalty[((ay)<<shift)-pred_y])*penalty_factor;\
/*printf("score: %d\n", d);*/\
        if(d < minima[minima_count-1].height){\
            int j=0;\
            \
            while(d >= minima[j].height) j++;\
\
            memmove(&minima [j+1], &minima [j], (minima_count - j - 1)*sizeof(Minima));\
\
            minima[j].checked= 0;\
            minima[j].height= d;\
            minima[j].x= ax;\
            minima[j].y= ay;\
            \
            i=-1;\
            continue;\
        }\
    }\
}

#define MAX_SAB_SIZE ME_MAP_SIZE
static int sab_diamond_search(MpegEncContext * s, int *best, int dmin,
                                       int src_index, int ref_index, int const penalty_factor,
                                       int size, int h, int flags)
{
    MotionEstContext * const c= &s->me;
    me_cmp_func cmpf, chroma_cmpf;
    Minima minima[MAX_SAB_SIZE];
    const int minima_count= FFABS(c->dia_size);
    int i, j;
    LOAD_COMMON
    LOAD_COMMON2
    int map_generation= c->map_generation;

    cmpf= s->dsp.me_cmp[size];
    chroma_cmpf= s->dsp.me_cmp[size+1];

    /*Note j<MAX_SAB_SIZE is needed if MAX_SAB_SIZE < ME_MAP_SIZE as j can
      become larger due to MVs overflowing their ME_MAP_MV_BITS bits space in map
     */
    for(j=i=0; i<ME_MAP_SIZE && j<MAX_SAB_SIZE; i++){
        uint32_t key= map[i];

        key += (1<<(ME_MAP_MV_BITS-1)) + (1<<(2*ME_MAP_MV_BITS-1));

        if((key&((-1)<<(2*ME_MAP_MV_BITS))) != map_generation) continue;

        minima[j].height= score_map[i];
        minima[j].x= key & ((1<<ME_MAP_MV_BITS)-1); key>>=ME_MAP_MV_BITS;
        minima[j].y= key & ((1<<ME_MAP_MV_BITS)-1);
        minima[j].x-= (1<<(ME_MAP_MV_BITS-1));
        minima[j].y-= (1<<(ME_MAP_MV_BITS-1));

        // all entries in map should be in range except if the mv overflows their ME_MAP_MV_BITS bits space
        if(   minima[j].x > xmax || minima[j].x < xmin
           || minima[j].y > ymax || minima[j].y < ymin)
            continue;

        minima[j].checked=0;
        if(minima[j].x || minima[j].y)
            minima[j].height+= (mv_penalty[((minima[j].x)<<shift)-pred_x] + mv_penalty[((minima[j].y)<<shift)-pred_y])*penalty_factor;

        j++;
    }

    qsort(minima, j, sizeof(Minima), minima_cmp);

    for(; j<minima_count; j++){
        minima[j].height=256*256*256*64;
        minima[j].checked=0;
        minima[j].x= minima[j].y=0;
    }

    for(i=0; i<minima_count; i++){
        const int x= minima[i].x;
        const int y= minima[i].y;
        int d;

        if(minima[i].checked) continue;

        if(   x >= xmax || x <= xmin
           || y >= ymax || y <= ymin)
           continue;

        SAB_CHECK_MV(x-1, y)
        SAB_CHECK_MV(x+1, y)
        SAB_CHECK_MV(x  , y-1)
        SAB_CHECK_MV(x  , y+1)

        minima[i].checked= 1;
    }

    best[0]= minima[0].x;
    best[1]= minima[0].y;
    dmin= minima[0].height;

    if(   best[0] < xmax && best[0] > xmin
       && best[1] < ymax && best[1] > ymin){
        int d;
        //ensure that the refernece samples for hpel refinement are in the map
        CHECK_MV(best[0]-1, best[1])
        CHECK_MV(best[0]+1, best[1])
        CHECK_MV(best[0], best[1]-1)
        CHECK_MV(best[0], best[1]+1)
    }
    return dmin;
}

static int var_diamond_search(MpegEncContext * s, int *best, int dmin,
                                       int src_index, int ref_index, int const penalty_factor,
                                       int size, int h, int flags)
{
    MotionEstContext * const c= &s->me;
    me_cmp_func cmpf, chroma_cmpf;
    int dia_size;
    LOAD_COMMON
    LOAD_COMMON2
    int map_generation= c->map_generation;

    cmpf= s->dsp.me_cmp[size];
    chroma_cmpf= s->dsp.me_cmp[size+1];

    for(dia_size=1; dia_size<=c->dia_size; dia_size++){
        int dir, start, end;
        const int x= best[0];
        const int y= best[1];

        start= FFMAX(0, y + dia_size - ymax);
        end  = FFMIN(dia_size, xmax - x + 1);
        for(dir= start; dir<end; dir++){
            int d;

//check(x + dir,y + dia_size - dir,0, a0)
            CHECK_MV(x + dir           , y + dia_size - dir);
        }

        start= FFMAX(0, x + dia_size - xmax);
        end  = FFMIN(dia_size, y - ymin + 1);
        for(dir= start; dir<end; dir++){
            int d;

//check(x + dia_size - dir, y - dir,0, a1)
            CHECK_MV(x + dia_size - dir, y - dir           );
        }

        start= FFMAX(0, -y + dia_size + ymin );
        end  = FFMIN(dia_size, x - xmin + 1);
        for(dir= start; dir<end; dir++){
            int d;

//check(x - dir,y - dia_size + dir,0, a2)
            CHECK_MV(x - dir           , y - dia_size + dir);
        }

        start= FFMAX(0, -x + dia_size + xmin );
        end  = FFMIN(dia_size, ymax - y + 1);
        for(dir= start; dir<end; dir++){
            int d;

//check(x - dia_size + dir, y + dir,0, a3)
            CHECK_MV(x - dia_size + dir, y + dir           );
        }

        if(x!=best[0] || y!=best[1])
            dia_size=0;
    }
    return dmin;
}

static av_always_inline int diamond_search(MpegEncContext * s, int *best, int dmin,
                                       int src_index, int ref_index, int const penalty_factor,
                                       int size, int h, int flags){
    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);
    else if(c->dia_size<-1)
        return   sab_diamond_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags);
    else if(c->dia_size<2)
        return small_diamond_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags);
    else if(c->dia_size>1024)
        return          full_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags);
    else if(c->dia_size>768)
        return           umh_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags);
    else if(c->dia_size>512)
        return           hex_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags, c->dia_size&0xFF);
    else if(c->dia_size>256)
        return       l2s_dia_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags);
    else
        return   var_diamond_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags);
}

/*!
   \param P[10][2] a list of candidate mvs to check before starting the
   iterative search. If one of the candidates is close to the optimal mv, then
   it takes fewer iterations. And it increases the chance that we find the
   optimal mv.
 */
static av_always_inline int epzs_motion_search_internal(MpegEncContext * s, int *mx_ptr, int *my_ptr,
                             int P[10][2], int src_index, int ref_index, int16_t (*last_mv)[2],
                             int ref_mv_scale, int flags, int size, int h)
{
    MotionEstContext * const c= &s->me;
    int best[2]={0, 0};      /*!< x and y coordinates of the best motion vector.
                               i.e. the difference between the position of the
                               block currently being encoded and the position of
                               the block chosen to predict it from. */
    int d;                   ///< the score (cmp + penalty) of any given mv
    int dmin;                /*!< the best value of d, i.e. the score
                               corresponding to the mv stored in best[]. */
    int map_generation;
    int penalty_factor;
    const int ref_mv_stride= s->mb_stride; //pass as arg  FIXME
    const int ref_mv_xy= s->mb_x + s->mb_y*ref_mv_stride; //add to last_mv beforepassing FIXME
    me_cmp_func cmpf, chroma_cmpf;

    LOAD_COMMON
    LOAD_COMMON2

    if(c->pre_pass){
        penalty_factor= c->pre_penalty_factor;
        cmpf= s->dsp.me_pre_cmp[size];
        chroma_cmpf= s->dsp.me_pre_cmp[size+1];
    }else{
        penalty_factor= c->penalty_factor;
        cmpf= s->dsp.me_cmp[size];
        chroma_cmpf= s->dsp.me_cmp[size+1];
    }

    map_generation= update_map_generation(c);

    assert(cmpf);
    dmin= cmp(s, 0, 0, 0, 0, size, h, ref_index, src_index, cmpf, chroma_cmpf, flags);
    map[0]= map_generation;
    score_map[0]= dmin;

    //FIXME precalc first term below?
    if((s->pict_type == FF_B_TYPE && !(c->flags & FLAG_DIRECT)) || s->flags&CODEC_FLAG_MV0)
        dmin += (mv_penalty[pred_x] + mv_penalty[pred_y])*penalty_factor;

    /* first line */
    if (s->first_slice_line) {
        CHECK_MV(P_LEFT[0]>>shift, P_LEFT[1]>>shift)
        CHECK_CLIPPED_MV((last_mv[ref_mv_xy][0]*ref_mv_scale + (1<<15))>>16,
                        (last_mv[ref_mv_xy][1]*ref_mv_scale + (1<<15))>>16)
    }else{
        if(dmin<((h*h*s->avctx->mv0_threshold)>>8)
                    && ( P_LEFT[0]    |P_LEFT[1]
                        |P_TOP[0]     |P_TOP[1]
                        |P_TOPRIGHT[0]|P_TOPRIGHT[1])==0){
            *mx_ptr= 0;
            *my_ptr= 0;
            c->skip=1;
            return dmin;
        }
        CHECK_MV(    P_MEDIAN[0] >>shift ,    P_MEDIAN[1] >>shift)
        CHECK_CLIPPED_MV((P_MEDIAN[0]>>shift)  , (P_MEDIAN[1]>>shift)-1)
        CHECK_CLIPPED_MV((P_MEDIAN[0]>>shift)  , (P_MEDIAN[1]>>shift)+1)
        CHECK_CLIPPED_MV((P_MEDIAN[0]>>shift)-1, (P_MEDIAN[1]>>shift)  )
        CHECK_CLIPPED_MV((P_MEDIAN[0]>>shift)+1, (P_MEDIAN[1]>>shift)  )
        CHECK_CLIPPED_MV((last_mv[ref_mv_xy][0]*ref_mv_scale + (1<<15))>>16,
                        (last_mv[ref_mv_xy][1]*ref_mv_scale + (1<<15))>>16)
        CHECK_MV(P_LEFT[0]    >>shift, P_LEFT[1]    >>shift)
        CHECK_MV(P_TOP[0]     >>shift, P_TOP[1]     >>shift)
        CHECK_MV(P_TOPRIGHT[0]>>shift, P_TOPRIGHT[1]>>shift)
    }
    if(dmin>h*h*4){
        if(c->pre_pass){
            CHECK_CLIPPED_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)
            if(!s->first_slice_line)
                CHECK_CLIPPED_MV((last_mv[ref_mv_xy-ref_mv_stride][0]*ref_mv_scale + (1<<15))>>16,
                                (last_mv[ref_mv_xy-ref_mv_stride][1]*ref_mv_scale + (1<<15))>>16)
        }else{
            CHECK_CLIPPED_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)
            if(s->mb_y+1<s->end_mb_y)  //FIXME replace at least with last_slice_line
                CHECK_CLIPPED_MV((last_mv[ref_mv_xy+ref_mv_stride][0]*ref_mv_scale + (1<<15))>>16,
                                (last_mv[ref_mv_xy+ref_mv_stride][1]*ref_mv_scale + (1<<15))>>16)
        }
    }

    if(c->avctx->last_predictor_count){
        const int count= c->avctx->last_predictor_count;
        const int xstart= FFMAX(0, s->mb_x - count);
        const int ystart= FFMAX(0, s->mb_y - count);
        const int xend= FFMIN(s->mb_width , s->mb_x + count + 1);
        const int yend= FFMIN(s->mb_height, s->mb_y + count + 1);
        int mb_y;

        for(mb_y=ystart; mb_y<yend; mb_y++){
            int mb_x;
            for(mb_x=xstart; mb_x<xend; mb_x++){
                const int xy= mb_x + 1 + (mb_y + 1)*ref_mv_stride;
                int mx= (last_mv[xy][0]*ref_mv_scale + (1<<15))>>16;
                int my= (last_mv[xy][1]*ref_mv_scale + (1<<15))>>16;

                if(mx>xmax || mx<xmin || my>ymax || my<ymin) continue;
                CHECK_MV(mx,my)
            }
        }
    }

//check(best[0],best[1],0, b0)
    dmin= diamond_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags);

//check(best[0],best[1],0, b1)
    *mx_ptr= best[0];
    *my_ptr= best[1];

//    printf("%d %d %d \n", best[0], best[1], dmin);
    return dmin;
}

//this function is dedicated to the braindamaged gcc
inline int ff_epzs_motion_search(MpegEncContext * s, int *mx_ptr, int *my_ptr,
                             int P[10][2], int src_index, int ref_index, int16_t (*last_mv)[2],
                             int ref_mv_scale, int size, int h)
{
    MotionEstContext * const c= &s->me;
//FIXME convert other functions in the same way if faster
    if(c->flags==0 && h==16 && size==0){
        return epzs_motion_search_internal(s, mx_ptr, my_ptr, P, src_index, ref_index, last_mv, ref_mv_scale, 0, 0, 16);
//    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);
    }else{
        return epzs_motion_search_internal(s, mx_ptr, my_ptr, P, src_index, ref_index, last_mv, ref_mv_scale, c->flags, size, h);
    }
}

static int epzs_motion_search4(MpegEncContext * s,
                             int *mx_ptr, int *my_ptr, int P[10][2],
                             int src_index, int ref_index, int16_t (*last_mv)[2],
                             int ref_mv_scale)
{
    MotionEstContext * const c= &s->me;
    int best[2]={0, 0};
    int d, dmin;
    int map_generation;
    const int penalty_factor= c->penalty_factor;
    const int size=1;
    const int h=8;
    const int ref_mv_stride= s->mb_stride;
    const int ref_mv_xy= s->mb_x + s->mb_y *ref_mv_stride;
    me_cmp_func cmpf, chroma_cmpf;
    LOAD_COMMON
    int flags= c->flags;
    LOAD_COMMON2

    cmpf= s->dsp.me_cmp[size];
    chroma_cmpf= s->dsp.me_cmp[size+1];

    map_generation= update_map_generation(c);

    dmin = 1000000;
//printf("%d %d %d %d //",xmin, ymin, xmax, ymax);
    /* first line */
    if (s->first_slice_line) {
        CHECK_MV(P_LEFT[0]>>shift, P_LEFT[1]>>shift)
        CHECK_CLIPPED_MV((last_mv[ref_mv_xy][0]*ref_mv_scale + (1<<15))>>16,
                        (last_mv[ref_mv_xy][1]*ref_mv_scale + (1<<15))>>16)
        CHECK_MV(P_MV1[0]>>shift, P_MV1[1]>>shift)
    }else{
        CHECK_MV(P_MV1[0]>>shift, P_MV1[1]>>shift)
        //FIXME try some early stop
        CHECK_MV(P_MEDIAN[0]>>shift, P_MEDIAN[1]>>shift)
        CHECK_MV(P_LEFT[0]>>shift, P_LEFT[1]>>shift)
        CHECK_MV(P_TOP[0]>>shift, P_TOP[1]>>shift)
        CHECK_MV(P_TOPRIGHT[0]>>shift, P_TOPRIGHT[1]>>shift)
        CHECK_CLIPPED_MV((last_mv[ref_mv_xy][0]*ref_mv_scale + (1<<15))>>16,
                        (last_mv[ref_mv_xy][1]*ref_mv_scale + (1<<15))>>16)
    }
    if(dmin>64*4){
        CHECK_CLIPPED_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)
        if(s->mb_y+1<s->end_mb_y)  //FIXME replace at least with last_slice_line
            CHECK_CLIPPED_MV((last_mv[ref_mv_xy+ref_mv_stride][0]*ref_mv_scale + (1<<15))>>16,
                            (last_mv[ref_mv_xy+ref_mv_stride][1]*ref_mv_scale + (1<<15))>>16)
    }

    dmin= diamond_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags);

    *mx_ptr= best[0];
    *my_ptr= best[1];

//    printf("%d %d %d \n", best[0], best[1], dmin);
    return dmin;
}

//try to merge with above FIXME (needs PSNR test)
static int epzs_motion_search2(MpegEncContext * s,
                             int *mx_ptr, int *my_ptr, int P[10][2],
                             int src_index, int ref_index, int16_t (*last_mv)[2],
                             int ref_mv_scale)
{
    MotionEstContext * const c= &s->me;
    int best[2]={0, 0};
    int d, dmin;
    int map_generation;
    const int penalty_factor= c->penalty_factor;
    const int size=0; //FIXME pass as arg
    const int h=8;
    const int ref_mv_stride= s->mb_stride;
    const int ref_mv_xy= s->mb_x + s->mb_y *ref_mv_stride;
    me_cmp_func cmpf, chroma_cmpf;
    LOAD_COMMON
    int flags= c->flags;
    LOAD_COMMON2

    cmpf= s->dsp.me_cmp[size];
    chroma_cmpf= s->dsp.me_cmp[size+1];

    map_generation= update_map_generation(c);

    dmin = 1000000;
//printf("%d %d %d %d //",xmin, ymin, xmax, ymax);
    /* first line */
    if (s->first_slice_line) {
        CHECK_MV(P_LEFT[0]>>shift, P_LEFT[1]>>shift)
        CHECK_CLIPPED_MV((last_mv[ref_mv_xy][0]*ref_mv_scale + (1<<15))>>16,
                        (last_mv[ref_mv_xy][1]*ref_mv_scale + (1<<15))>>16)
        CHECK_MV(P_MV1[0]>>shift, P_MV1[1]>>shift)
    }else{
        CHECK_MV(P_MV1[0]>>shift, P_MV1[1]>>shift)
        //FIXME try some early stop
        CHECK_MV(P_MEDIAN[0]>>shift, P_MEDIAN[1]>>shift)
        CHECK_MV(P_LEFT[0]>>shift, P_LEFT[1]>>shift)
        CHECK_MV(P_TOP[0]>>shift, P_TOP[1]>>shift)
        CHECK_MV(P_TOPRIGHT[0]>>shift, P_TOPRIGHT[1]>>shift)
        CHECK_CLIPPED_MV((last_mv[ref_mv_xy][0]*ref_mv_scale + (1<<15))>>16,
                        (last_mv[ref_mv_xy][1]*ref_mv_scale + (1<<15))>>16)
    }
    if(dmin>64*4){
        CHECK_CLIPPED_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)
        if(s->mb_y+1<s->end_mb_y)  //FIXME replace at least with last_slice_line
            CHECK_CLIPPED_MV((last_mv[ref_mv_xy+ref_mv_stride][0]*ref_mv_scale + (1<<15))>>16,
                            (last_mv[ref_mv_xy+ref_mv_stride][1]*ref_mv_scale + (1<<15))>>16)
    }

    dmin= diamond_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags);

    *mx_ptr= best[0];
    *my_ptr= best[1];

//    printf("%d %d %d \n", best[0], best[1], dmin);
    return dmin;
}