summaryrefslogtreecommitdiff
path: root/vp10/encoder/mcomp.c
blob: 5720c63c7c93de3ff87283c557c70d1cb712ebfb (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
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
/*
 *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
 *
 *  Use of this source code is governed by a BSD-style license
 *  that can be found in the LICENSE file in the root of the source
 *  tree. An additional intellectual property rights grant can be found
 *  in the file PATENTS.  All contributing project authors may
 *  be found in the AUTHORS file in the root of the source tree.
 */

#include <limits.h>
#include <math.h>
#include <stdio.h>

#include "./vpx_config.h"
#include "./vpx_dsp_rtcd.h"

#include "vpx_mem/vpx_mem.h"
#include "vpx_ports/mem.h"

#include "vp10/common/common.h"
#include "vp10/common/reconinter.h"

#include "vp10/encoder/encoder.h"
#include "vp10/encoder/mcomp.h"

// #define NEW_DIAMOND_SEARCH

static INLINE const uint8_t *get_buf_from_mv(const struct buf_2d *buf,
                                             const MV *mv) {
  return &buf->buf[mv->row * buf->stride + mv->col];
}

void vp10_set_mv_search_range(MACROBLOCK *x, const MV *mv) {
  int col_min = (mv->col >> 3) - MAX_FULL_PEL_VAL + (mv->col & 7 ? 1 : 0);
  int row_min = (mv->row >> 3) - MAX_FULL_PEL_VAL + (mv->row & 7 ? 1 : 0);
  int col_max = (mv->col >> 3) + MAX_FULL_PEL_VAL;
  int row_max = (mv->row >> 3) + MAX_FULL_PEL_VAL;

  col_min = MAX(col_min, (MV_LOW >> 3) + 1);
  row_min = MAX(row_min, (MV_LOW >> 3) + 1);
  col_max = MIN(col_max, (MV_UPP >> 3) - 1);
  row_max = MIN(row_max, (MV_UPP >> 3) - 1);

  // Get intersection of UMV window and valid MV window to reduce # of checks
  // in diamond search.
  if (x->mv_col_min < col_min)
    x->mv_col_min = col_min;
  if (x->mv_col_max > col_max)
    x->mv_col_max = col_max;
  if (x->mv_row_min < row_min)
    x->mv_row_min = row_min;
  if (x->mv_row_max > row_max)
    x->mv_row_max = row_max;
}

int vp10_init_search_range(int size) {
  int sr = 0;
  // Minimum search size no matter what the passed in value.
  size = MAX(16, size);

  while ((size << sr) < MAX_FULL_PEL_VAL)
    sr++;

  sr = MIN(sr, MAX_MVSEARCH_STEPS - 2);
  return sr;
}

static INLINE int mv_cost(const MV *mv,
                          const int *joint_cost, int *const comp_cost[2]) {
  return joint_cost[vp10_get_mv_joint(mv)] +
             comp_cost[0][mv->row] + comp_cost[1][mv->col];
}

int vp10_mv_bit_cost(const MV *mv, const MV *ref,
                    const int *mvjcost, int *mvcost[2], int weight) {
  const MV diff = { mv->row - ref->row,
                    mv->col - ref->col };
  return ROUND_POWER_OF_TWO(mv_cost(&diff, mvjcost, mvcost) * weight, 7);
}

static int mv_err_cost(const MV *mv, const MV *ref,
                       const int *mvjcost, int *mvcost[2],
                       int error_per_bit) {
  if (mvcost) {
    const MV diff = { mv->row - ref->row,
                      mv->col - ref->col };
    return ROUND_POWER_OF_TWO(mv_cost(&diff, mvjcost, mvcost) *
                                  error_per_bit, 13);
  }
  return 0;
}

static int mvsad_err_cost(const MACROBLOCK *x, const MV *mv, const MV *ref,
                          int error_per_bit) {
  const MV diff = { mv->row - ref->row,
                    mv->col - ref->col };
  return ROUND_POWER_OF_TWO(mv_cost(&diff, x->nmvjointsadcost,
                                    x->nmvsadcost) * error_per_bit, 8);
}

void vp10_init_dsmotion_compensation(search_site_config *cfg, int stride) {
  int len, ss_count = 1;

  cfg->ss[0].mv.col = cfg->ss[0].mv.row = 0;
  cfg->ss[0].offset = 0;

  for (len = MAX_FIRST_STEP; len > 0; len /= 2) {
    // Generate offsets for 4 search sites per step.
    const MV ss_mvs[] = {{-len, 0}, {len, 0}, {0, -len}, {0, len}};
    int i;
    for (i = 0; i < 4; ++i) {
      search_site *const ss = &cfg->ss[ss_count++];
      ss->mv = ss_mvs[i];
      ss->offset = ss->mv.row * stride + ss->mv.col;
    }
  }

  cfg->ss_count = ss_count;
  cfg->searches_per_step = 4;
}

void vp10_init3smotion_compensation(search_site_config *cfg, int stride) {
  int len, ss_count = 1;

  cfg->ss[0].mv.col = cfg->ss[0].mv.row = 0;
  cfg->ss[0].offset = 0;

  for (len = MAX_FIRST_STEP; len > 0; len /= 2) {
    // Generate offsets for 8 search sites per step.
    const MV ss_mvs[8] = {
      {-len,  0  }, {len,  0  }, { 0,   -len}, {0,    len},
      {-len, -len}, {-len, len}, {len,  -len}, {len,  len}
    };
    int i;
    for (i = 0; i < 8; ++i) {
      search_site *const ss = &cfg->ss[ss_count++];
      ss->mv = ss_mvs[i];
      ss->offset = ss->mv.row * stride + ss->mv.col;
    }
  }

  cfg->ss_count = ss_count;
  cfg->searches_per_step = 8;
}

/*
 * To avoid the penalty for crossing cache-line read, preload the reference
 * area in a small buffer, which is aligned to make sure there won't be crossing
 * cache-line read while reading from this buffer. This reduced the cpu
 * cycles spent on reading ref data in sub-pixel filter functions.
 * TODO: Currently, since sub-pixel search range here is -3 ~ 3, copy 22 rows x
 * 32 cols area that is enough for 16x16 macroblock. Later, for SPLITMV, we
 * could reduce the area.
 */

/* estimated cost of a motion vector (r,c) */
#define MVC(r, c)                                       \
    (mvcost ?                                           \
     ((mvjcost[((r) != rr) * 2 + ((c) != rc)] +         \
       mvcost[0][((r) - rr)] + mvcost[1][((c) - rc)]) * \
      error_per_bit + 4096) >> 13 : 0)


// convert motion vector component to offset for sv[a]f calc
static INLINE int sp(int x) {
  return x & 7;
}

static INLINE const uint8_t *pre(const uint8_t *buf, int stride, int r, int c) {
  return &buf[(r >> 3) * stride + (c >> 3)];
}

/* checks if (r, c) has better score than previous best */
#define CHECK_BETTER(v, r, c) \
  if (c >= minc && c <= maxc && r >= minr && r <= maxr) {              \
    if (second_pred == NULL)                                           \
      thismse = vfp->svf(pre(y, y_stride, r, c), y_stride, sp(c), sp(r), z, \
                             src_stride, &sse);                        \
    else                                                               \
      thismse = vfp->svaf(pre(y, y_stride, r, c), y_stride, sp(c), sp(r), \
                              z, src_stride, &sse, second_pred);       \
    if ((v = MVC(r, c) + thismse) < besterr) {                         \
      besterr = v;                                                     \
      br = r;                                                          \
      bc = c;                                                          \
      *distortion = thismse;                                           \
      *sse1 = sse;                                                     \
    }                                                                  \
  } else {                                                             \
    v = INT_MAX;                                                       \
  }

#define FIRST_LEVEL_CHECKS                              \
  {                                                     \
    unsigned int left, right, up, down, diag;           \
    CHECK_BETTER(left, tr, tc - hstep);                 \
    CHECK_BETTER(right, tr, tc + hstep);                \
    CHECK_BETTER(up, tr - hstep, tc);                   \
    CHECK_BETTER(down, tr + hstep, tc);                 \
    whichdir = (left < right ? 0 : 1) +                 \
               (up < down ? 0 : 2);                     \
    switch (whichdir) {                                 \
      case 0:                                           \
        CHECK_BETTER(diag, tr - hstep, tc - hstep);     \
        break;                                          \
      case 1:                                           \
        CHECK_BETTER(diag, tr - hstep, tc + hstep);     \
        break;                                          \
      case 2:                                           \
        CHECK_BETTER(diag, tr + hstep, tc - hstep);     \
        break;                                          \
      case 3:                                           \
        CHECK_BETTER(diag, tr + hstep, tc + hstep);     \
        break;                                          \
    }                                                   \
  }

#define SECOND_LEVEL_CHECKS                             \
  {                                                     \
    int kr, kc;                                         \
    unsigned int second;                                \
    if (tr != br && tc != bc) {                         \
      kr = br - tr;                                     \
      kc = bc - tc;                                     \
      CHECK_BETTER(second, tr + kr, tc + 2 * kc);       \
      CHECK_BETTER(second, tr + 2 * kr, tc + kc);       \
    } else if (tr == br && tc != bc) {                  \
      kc = bc - tc;                                     \
      CHECK_BETTER(second, tr + hstep, tc + 2 * kc);    \
      CHECK_BETTER(second, tr - hstep, tc + 2 * kc);    \
      switch (whichdir) {                               \
        case 0:                                         \
        case 1:                                         \
          CHECK_BETTER(second, tr + hstep, tc + kc);    \
          break;                                        \
        case 2:                                         \
        case 3:                                         \
          CHECK_BETTER(second, tr - hstep, tc + kc);    \
          break;                                        \
      }                                                 \
    } else if (tr != br && tc == bc) {                  \
      kr = br - tr;                                     \
      CHECK_BETTER(second, tr + 2 * kr, tc + hstep);    \
      CHECK_BETTER(second, tr + 2 * kr, tc - hstep);    \
      switch (whichdir) {                               \
        case 0:                                         \
        case 2:                                         \
          CHECK_BETTER(second, tr + kr, tc + hstep);    \
          break;                                        \
        case 1:                                         \
        case 3:                                         \
          CHECK_BETTER(second, tr + kr, tc - hstep);    \
          break;                                        \
      }                                                 \
    }                                                   \
  }

// TODO(yunqingwang): SECOND_LEVEL_CHECKS_BEST was a rewrote of
// SECOND_LEVEL_CHECKS, and SECOND_LEVEL_CHECKS should be rewritten
// later in the same way.
#define SECOND_LEVEL_CHECKS_BEST                        \
  {                                                     \
    unsigned int second;                                \
    int br0 = br;                                       \
    int bc0 = bc;                                       \
    assert(tr == br || tc == bc);                       \
    if (tr == br && tc != bc) {                         \
      kc = bc - tc;                                     \
    } else if (tr != br && tc == bc) {                  \
      kr = br - tr;                                     \
    }                                                   \
    CHECK_BETTER(second, br0 + kr, bc0);                \
    CHECK_BETTER(second, br0, bc0 + kc);                \
    if (br0 != br || bc0 != bc) {                       \
      CHECK_BETTER(second, br0 + kr, bc0 + kc);         \
    }                                                   \
  }

#define SETUP_SUBPEL_SEARCH                                                \
  const uint8_t *const z = x->plane[0].src.buf;                            \
  const int src_stride = x->plane[0].src.stride;                           \
  const MACROBLOCKD *xd = &x->e_mbd;                                       \
  unsigned int besterr = INT_MAX;                                          \
  unsigned int sse;                                                        \
  unsigned int whichdir;                                                   \
  int thismse;                                                             \
  const unsigned int halfiters = iters_per_step;                           \
  const unsigned int quarteriters = iters_per_step;                        \
  const unsigned int eighthiters = iters_per_step;                         \
  const int y_stride = xd->plane[0].pre[0].stride;                         \
  const int offset = bestmv->row * y_stride + bestmv->col;                 \
  const uint8_t *const y = xd->plane[0].pre[0].buf;                        \
                                                                           \
  int rr = ref_mv->row;                                                    \
  int rc = ref_mv->col;                                                    \
  int br = bestmv->row * 8;                                                \
  int bc = bestmv->col * 8;                                                \
  int hstep = 4;                                                           \
  const int minc = MAX(x->mv_col_min * 8, ref_mv->col - MV_MAX);           \
  const int maxc = MIN(x->mv_col_max * 8, ref_mv->col + MV_MAX);           \
  const int minr = MAX(x->mv_row_min * 8, ref_mv->row - MV_MAX);           \
  const int maxr = MIN(x->mv_row_max * 8, ref_mv->row + MV_MAX);           \
  int tr = br;                                                             \
  int tc = bc;                                                             \
                                                                           \
  bestmv->row *= 8;                                                        \
  bestmv->col *= 8;

static unsigned int setup_center_error(const MACROBLOCKD *xd,
                                       const MV *bestmv,
                                       const MV *ref_mv,
                                       int error_per_bit,
                                       const vp9_variance_fn_ptr_t *vfp,
                                       const uint8_t *const src,
                                       const int src_stride,
                                       const uint8_t *const y,
                                       int y_stride,
                                       const uint8_t *second_pred,
                                       int w, int h, int offset,
                                       int *mvjcost, int *mvcost[2],
                                       unsigned int *sse1,
                                       int *distortion) {
  unsigned int besterr;
#if CONFIG_VP9_HIGHBITDEPTH
  if (second_pred != NULL) {
    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
      DECLARE_ALIGNED(16, uint16_t, comp_pred16[64 * 64]);
      vpx_highbd_comp_avg_pred(comp_pred16, second_pred, w, h, y + offset,
                               y_stride);
      besterr = vfp->vf(CONVERT_TO_BYTEPTR(comp_pred16), w, src, src_stride,
                        sse1);
    } else {
      DECLARE_ALIGNED(16, uint8_t, comp_pred[64 * 64]);
      vpx_comp_avg_pred(comp_pred, second_pred, w, h, y + offset, y_stride);
      besterr = vfp->vf(comp_pred, w, src, src_stride, sse1);
    }
  } else {
    besterr = vfp->vf(y + offset, y_stride, src, src_stride, sse1);
  }
  *distortion = besterr;
  besterr += mv_err_cost(bestmv, ref_mv, mvjcost, mvcost, error_per_bit);
#else
  (void) xd;
  if (second_pred != NULL) {
    DECLARE_ALIGNED(16, uint8_t, comp_pred[64 * 64]);
    vpx_comp_avg_pred(comp_pred, second_pred, w, h, y + offset, y_stride);
    besterr = vfp->vf(comp_pred, w, src, src_stride, sse1);
  } else {
    besterr = vfp->vf(y + offset, y_stride, src, src_stride, sse1);
  }
  *distortion = besterr;
  besterr += mv_err_cost(bestmv, ref_mv, mvjcost, mvcost, error_per_bit);
#endif  // CONFIG_VP9_HIGHBITDEPTH
  return besterr;
}

static INLINE int divide_and_round(const int n, const int d) {
  return ((n < 0) ^ (d < 0)) ? ((n - d / 2) / d) : ((n + d / 2) / d);
}

static INLINE int is_cost_list_wellbehaved(int *cost_list) {
  return cost_list[0] < cost_list[1] &&
         cost_list[0] < cost_list[2] &&
         cost_list[0] < cost_list[3] &&
         cost_list[0] < cost_list[4];
}

// Returns surface minima estimate at given precision in 1/2^n bits.
// Assume a model for the cost surface: S = A(x - x0)^2 + B(y - y0)^2 + C
// For a given set of costs S0, S1, S2, S3, S4 at points
// (y, x) = (0, 0), (0, -1), (1, 0), (0, 1) and (-1, 0) respectively,
// the solution for the location of the minima (x0, y0) is given by:
// x0 = 1/2 (S1 - S3)/(S1 + S3 - 2*S0),
// y0 = 1/2 (S4 - S2)/(S4 + S2 - 2*S0).
// The code below is an integerized version of that.
static void get_cost_surf_min(int *cost_list, int *ir, int *ic,
                              int bits) {
  *ic = divide_and_round((cost_list[1] - cost_list[3]) * (1 << (bits - 1)),
                         (cost_list[1] - 2 * cost_list[0] + cost_list[3]));
  *ir = divide_and_round((cost_list[4] - cost_list[2]) * (1 << (bits - 1)),
                         (cost_list[4] - 2 * cost_list[0] + cost_list[2]));
}

int vp10_find_best_sub_pixel_tree_pruned_evenmore(
    const MACROBLOCK *x,
    MV *bestmv, const MV *ref_mv,
    int allow_hp,
    int error_per_bit,
    const vp9_variance_fn_ptr_t *vfp,
    int forced_stop,
    int iters_per_step,
    int *cost_list,
    int *mvjcost, int *mvcost[2],
    int *distortion,
    unsigned int *sse1,
    const uint8_t *second_pred,
    int w, int h) {
  SETUP_SUBPEL_SEARCH;
  besterr = setup_center_error(xd, bestmv, ref_mv, error_per_bit, vfp,
                               z, src_stride, y, y_stride, second_pred,
                               w, h, offset, mvjcost, mvcost,
                               sse1, distortion);
  (void) halfiters;
  (void) quarteriters;
  (void) eighthiters;
  (void) whichdir;
  (void) allow_hp;
  (void) forced_stop;
  (void) hstep;

  if (cost_list &&
      cost_list[0] != INT_MAX && cost_list[1] != INT_MAX &&
      cost_list[2] != INT_MAX && cost_list[3] != INT_MAX &&
      cost_list[4] != INT_MAX &&
      is_cost_list_wellbehaved(cost_list)) {
    int ir, ic;
    unsigned int minpt;
    get_cost_surf_min(cost_list, &ir, &ic, 2);
    if (ir != 0 || ic != 0) {
      CHECK_BETTER(minpt, tr + 2 * ir, tc + 2 * ic);
    }
  } else {
    FIRST_LEVEL_CHECKS;
    if (halfiters > 1) {
      SECOND_LEVEL_CHECKS;
    }

    tr = br;
    tc = bc;

    // Each subsequent iteration checks at least one point in common with
    // the last iteration could be 2 ( if diag selected) 1/4 pel
    // Note forced_stop: 0 - full, 1 - qtr only, 2 - half only
    if (forced_stop != 2) {
      hstep >>= 1;
      FIRST_LEVEL_CHECKS;
      if (quarteriters > 1) {
        SECOND_LEVEL_CHECKS;
      }
    }
  }

  tr = br;
  tc = bc;

  if (allow_hp && vp10_use_mv_hp(ref_mv) && forced_stop == 0) {
    hstep >>= 1;
    FIRST_LEVEL_CHECKS;
    if (eighthiters > 1) {
      SECOND_LEVEL_CHECKS;
    }
  }

  bestmv->row = br;
  bestmv->col = bc;

  if ((abs(bestmv->col - ref_mv->col) > (MAX_FULL_PEL_VAL << 3)) ||
      (abs(bestmv->row - ref_mv->row) > (MAX_FULL_PEL_VAL << 3)))
    return INT_MAX;

  return besterr;
}

int vp10_find_best_sub_pixel_tree_pruned_more(const MACROBLOCK *x,
                                             MV *bestmv, const MV *ref_mv,
                                             int allow_hp,
                                             int error_per_bit,
                                             const vp9_variance_fn_ptr_t *vfp,
                                             int forced_stop,
                                             int iters_per_step,
                                             int *cost_list,
                                             int *mvjcost, int *mvcost[2],
                                             int *distortion,
                                             unsigned int *sse1,
                                             const uint8_t *second_pred,
                                             int w, int h) {
  SETUP_SUBPEL_SEARCH;
  besterr = setup_center_error(xd, bestmv, ref_mv, error_per_bit, vfp,
                               z, src_stride, y, y_stride, second_pred,
                               w, h, offset, mvjcost, mvcost,
                               sse1, distortion);
  if (cost_list &&
      cost_list[0] != INT_MAX && cost_list[1] != INT_MAX &&
      cost_list[2] != INT_MAX && cost_list[3] != INT_MAX &&
      cost_list[4] != INT_MAX &&
      is_cost_list_wellbehaved(cost_list)) {
    unsigned int minpt;
    int ir, ic;
    get_cost_surf_min(cost_list, &ir, &ic, 1);
    if (ir != 0 || ic != 0) {
      CHECK_BETTER(minpt, tr + ir * hstep, tc + ic * hstep);
    }
  } else {
    FIRST_LEVEL_CHECKS;
    if (halfiters > 1) {
      SECOND_LEVEL_CHECKS;
    }
  }

  // Each subsequent iteration checks at least one point in common with
  // the last iteration could be 2 ( if diag selected) 1/4 pel

  // Note forced_stop: 0 - full, 1 - qtr only, 2 - half only
  if (forced_stop != 2) {
    tr = br;
    tc = bc;
    hstep >>= 1;
    FIRST_LEVEL_CHECKS;
    if (quarteriters > 1) {
      SECOND_LEVEL_CHECKS;
    }
  }

  if (allow_hp && vp10_use_mv_hp(ref_mv) && forced_stop == 0) {
    tr = br;
    tc = bc;
    hstep >>= 1;
    FIRST_LEVEL_CHECKS;
    if (eighthiters > 1) {
      SECOND_LEVEL_CHECKS;
    }
  }
  // These lines insure static analysis doesn't warn that
  // tr and tc aren't used after the above point.
  (void) tr;
  (void) tc;

  bestmv->row = br;
  bestmv->col = bc;

  if ((abs(bestmv->col - ref_mv->col) > (MAX_FULL_PEL_VAL << 3)) ||
      (abs(bestmv->row - ref_mv->row) > (MAX_FULL_PEL_VAL << 3)))
    return INT_MAX;

  return besterr;
}

int vp10_find_best_sub_pixel_tree_pruned(const MACROBLOCK *x,
                                        MV *bestmv, const MV *ref_mv,
                                        int allow_hp,
                                        int error_per_bit,
                                        const vp9_variance_fn_ptr_t *vfp,
                                        int forced_stop,
                                        int iters_per_step,
                                        int *cost_list,
                                        int *mvjcost, int *mvcost[2],
                                        int *distortion,
                                        unsigned int *sse1,
                                        const uint8_t *second_pred,
                                        int w, int h) {
  SETUP_SUBPEL_SEARCH;
  besterr = setup_center_error(xd, bestmv, ref_mv, error_per_bit, vfp,
                               z, src_stride, y, y_stride, second_pred,
                               w, h, offset, mvjcost, mvcost,
                               sse1, distortion);
  if (cost_list &&
      cost_list[0] != INT_MAX && cost_list[1] != INT_MAX &&
      cost_list[2] != INT_MAX && cost_list[3] != INT_MAX &&
      cost_list[4] != INT_MAX) {
    unsigned int left, right, up, down, diag;
    whichdir = (cost_list[1] < cost_list[3] ? 0 : 1) +
               (cost_list[2] < cost_list[4] ? 0 : 2);
    switch (whichdir) {
      case 0:
        CHECK_BETTER(left, tr, tc - hstep);
        CHECK_BETTER(down, tr + hstep, tc);
        CHECK_BETTER(diag, tr + hstep, tc - hstep);
        break;
      case 1:
        CHECK_BETTER(right, tr, tc + hstep);
        CHECK_BETTER(down, tr + hstep, tc);
        CHECK_BETTER(diag, tr + hstep, tc + hstep);
        break;
      case 2:
        CHECK_BETTER(left, tr, tc - hstep);
        CHECK_BETTER(up, tr - hstep, tc);
        CHECK_BETTER(diag, tr - hstep, tc - hstep);
        break;
      case 3:
        CHECK_BETTER(right, tr, tc + hstep);
        CHECK_BETTER(up, tr - hstep, tc);
        CHECK_BETTER(diag, tr - hstep, tc + hstep);
        break;
    }
  } else {
    FIRST_LEVEL_CHECKS;
    if (halfiters > 1) {
      SECOND_LEVEL_CHECKS;
    }
  }

  tr = br;
  tc = bc;

  // Each subsequent iteration checks at least one point in common with
  // the last iteration could be 2 ( if diag selected) 1/4 pel

  // Note forced_stop: 0 - full, 1 - qtr only, 2 - half only
  if (forced_stop != 2) {
    hstep >>= 1;
    FIRST_LEVEL_CHECKS;
    if (quarteriters > 1) {
      SECOND_LEVEL_CHECKS;
    }
    tr = br;
    tc = bc;
  }

  if (allow_hp && vp10_use_mv_hp(ref_mv) && forced_stop == 0) {
    hstep >>= 1;
    FIRST_LEVEL_CHECKS;
    if (eighthiters > 1) {
      SECOND_LEVEL_CHECKS;
    }
    tr = br;
    tc = bc;
  }
  // These lines insure static analysis doesn't warn that
  // tr and tc aren't used after the above point.
  (void) tr;
  (void) tc;

  bestmv->row = br;
  bestmv->col = bc;

  if ((abs(bestmv->col - ref_mv->col) > (MAX_FULL_PEL_VAL << 3)) ||
      (abs(bestmv->row - ref_mv->row) > (MAX_FULL_PEL_VAL << 3)))
    return INT_MAX;

  return besterr;
}

static const MV search_step_table[12] = {
    // left, right, up, down
    {0, -4}, {0, 4}, {-4, 0}, {4, 0},
    {0, -2}, {0, 2}, {-2, 0}, {2, 0},
    {0, -1}, {0, 1}, {-1, 0}, {1, 0}
};

int vp10_find_best_sub_pixel_tree(const MACROBLOCK *x,
                                 MV *bestmv, const MV *ref_mv,
                                 int allow_hp,
                                 int error_per_bit,
                                 const vp9_variance_fn_ptr_t *vfp,
                                 int forced_stop,
                                 int iters_per_step,
                                 int *cost_list,
                                 int *mvjcost, int *mvcost[2],
                                 int *distortion,
                                 unsigned int *sse1,
                                 const uint8_t *second_pred,
                                 int w, int h) {
  const uint8_t *const z = x->plane[0].src.buf;
  const uint8_t *const src_address = z;
  const int src_stride = x->plane[0].src.stride;
  const MACROBLOCKD *xd = &x->e_mbd;
  unsigned int besterr = INT_MAX;
  unsigned int sse;
  int thismse;
  const int y_stride = xd->plane[0].pre[0].stride;
  const int offset = bestmv->row * y_stride + bestmv->col;
  const uint8_t *const y = xd->plane[0].pre[0].buf;

  int rr = ref_mv->row;
  int rc = ref_mv->col;
  int br = bestmv->row * 8;
  int bc = bestmv->col * 8;
  int hstep = 4;
  int iter, round = 3 - forced_stop;
  const int minc = MAX(x->mv_col_min * 8, ref_mv->col - MV_MAX);
  const int maxc = MIN(x->mv_col_max * 8, ref_mv->col + MV_MAX);
  const int minr = MAX(x->mv_row_min * 8, ref_mv->row - MV_MAX);
  const int maxr = MIN(x->mv_row_max * 8, ref_mv->row + MV_MAX);
  int tr = br;
  int tc = bc;
  const MV *search_step = search_step_table;
  int idx, best_idx = -1;
  unsigned int cost_array[5];
  int kr, kc;

  if (!(allow_hp && vp10_use_mv_hp(ref_mv)))
    if (round == 3)
      round = 2;

  bestmv->row *= 8;
  bestmv->col *= 8;

  besterr = setup_center_error(xd, bestmv, ref_mv, error_per_bit, vfp,
                               z, src_stride, y, y_stride, second_pred,
                               w, h, offset, mvjcost, mvcost,
                               sse1, distortion);

  (void) cost_list;  // to silence compiler warning

  for (iter = 0; iter < round; ++iter) {
    // Check vertical and horizontal sub-pixel positions.
    for (idx = 0; idx < 4; ++idx) {
      tr = br + search_step[idx].row;
      tc = bc + search_step[idx].col;
      if (tc >= minc && tc <= maxc && tr >= minr && tr <= maxr) {
        const uint8_t *const pre_address = y + (tr >> 3) * y_stride + (tc >> 3);
        MV this_mv;
        this_mv.row = tr;
        this_mv.col = tc;
        if (second_pred == NULL)
          thismse = vfp->svf(pre_address, y_stride, sp(tc), sp(tr),
                             src_address, src_stride, &sse);
        else
          thismse = vfp->svaf(pre_address, y_stride, sp(tc), sp(tr),
                              src_address, src_stride, &sse, second_pred);
        cost_array[idx] = thismse +
            mv_err_cost(&this_mv, ref_mv, mvjcost, mvcost, error_per_bit);

        if (cost_array[idx] < besterr) {
          best_idx = idx;
          besterr = cost_array[idx];
          *distortion = thismse;
          *sse1 = sse;
        }
      } else {
        cost_array[idx] = INT_MAX;
      }
    }

    // Check diagonal sub-pixel position
    kc = (cost_array[0] <= cost_array[1] ? -hstep : hstep);
    kr = (cost_array[2] <= cost_array[3] ? -hstep : hstep);

    tc = bc + kc;
    tr = br + kr;
    if (tc >= minc && tc <= maxc && tr >= minr && tr <= maxr) {
      const uint8_t *const pre_address = y + (tr >> 3) * y_stride + (tc >> 3);
      MV this_mv = {tr, tc};
      if (second_pred == NULL)
        thismse = vfp->svf(pre_address, y_stride, sp(tc), sp(tr),
                           src_address, src_stride, &sse);
      else
        thismse = vfp->svaf(pre_address, y_stride, sp(tc), sp(tr),
                            src_address, src_stride, &sse, second_pred);
      cost_array[4] = thismse +
          mv_err_cost(&this_mv, ref_mv, mvjcost, mvcost, error_per_bit);

      if (cost_array[4] < besterr) {
        best_idx = 4;
        besterr = cost_array[4];
        *distortion = thismse;
        *sse1 = sse;
      }
    } else {
      cost_array[idx] = INT_MAX;
    }

    if (best_idx < 4 && best_idx >= 0) {
      br += search_step[best_idx].row;
      bc += search_step[best_idx].col;
    } else if (best_idx == 4) {
      br = tr;
      bc = tc;
    }

    if (iters_per_step > 1 && best_idx != -1)
      SECOND_LEVEL_CHECKS_BEST;

    tr = br;
    tc = bc;

    search_step += 4;
    hstep >>= 1;
    best_idx = -1;
  }

  // Each subsequent iteration checks at least one point in common with
  // the last iteration could be 2 ( if diag selected) 1/4 pel

  // These lines insure static analysis doesn't warn that
  // tr and tc aren't used after the above point.
  (void) tr;
  (void) tc;

  bestmv->row = br;
  bestmv->col = bc;

  if ((abs(bestmv->col - ref_mv->col) > (MAX_FULL_PEL_VAL << 3)) ||
      (abs(bestmv->row - ref_mv->row) > (MAX_FULL_PEL_VAL << 3)))
    return INT_MAX;

  return besterr;
}

#undef MVC
#undef PRE
#undef CHECK_BETTER

static INLINE int check_bounds(const MACROBLOCK *x, int row, int col,
                               int range) {
  return ((row - range) >= x->mv_row_min) &
         ((row + range) <= x->mv_row_max) &
         ((col - range) >= x->mv_col_min) &
         ((col + range) <= x->mv_col_max);
}

static INLINE int is_mv_in(const MACROBLOCK *x, const MV *mv) {
  return (mv->col >= x->mv_col_min) && (mv->col <= x->mv_col_max) &&
         (mv->row >= x->mv_row_min) && (mv->row <= x->mv_row_max);
}

#define CHECK_BETTER \
  {\
    if (thissad < bestsad) {\
      if (use_mvcost) \
        thissad += mvsad_err_cost(x, &this_mv, &fcenter_mv, sad_per_bit);\
      if (thissad < bestsad) {\
        bestsad = thissad;\
        best_site = i;\
      }\
    }\
  }

#define MAX_PATTERN_SCALES         11
#define MAX_PATTERN_CANDIDATES      8  // max number of canddiates per scale
#define PATTERN_CANDIDATES_REF      3  // number of refinement candidates

// Calculate and return a sad+mvcost list around an integer best pel.
static INLINE void calc_int_cost_list(const MACROBLOCK *x,
                                      const MV *ref_mv,
                                      int sadpb,
                                      const vp9_variance_fn_ptr_t *fn_ptr,
                                      const MV *best_mv,
                                      int *cost_list) {
  static const MV neighbors[4] = {{0, -1}, {1, 0}, {0, 1}, {-1, 0}};
  const struct buf_2d *const what = &x->plane[0].src;
  const struct buf_2d *const in_what = &x->e_mbd.plane[0].pre[0];
  const MV fcenter_mv = {ref_mv->row >> 3, ref_mv->col >> 3};
  int br = best_mv->row;
  int bc = best_mv->col;
  MV this_mv;
  int i;
  unsigned int sse;

  this_mv.row = br;
  this_mv.col = bc;
  cost_list[0] = fn_ptr->vf(what->buf, what->stride,
                            get_buf_from_mv(in_what, &this_mv),
                            in_what->stride, &sse) +
      mvsad_err_cost(x, &this_mv, &fcenter_mv, sadpb);
  if (check_bounds(x, br, bc, 1)) {
    for (i = 0; i < 4; i++) {
      const MV this_mv = {br + neighbors[i].row,
        bc + neighbors[i].col};
      cost_list[i + 1] = fn_ptr->vf(what->buf, what->stride,
                                    get_buf_from_mv(in_what, &this_mv),
                                    in_what->stride, &sse) +
          // mvsad_err_cost(x, &this_mv, &fcenter_mv, sadpb);
          mv_err_cost(&this_mv, &fcenter_mv, x->nmvjointcost, x->mvcost,
                      x->errorperbit);
    }
  } else {
    for (i = 0; i < 4; i++) {
      const MV this_mv = {br + neighbors[i].row,
        bc + neighbors[i].col};
      if (!is_mv_in(x, &this_mv))
        cost_list[i + 1] = INT_MAX;
      else
        cost_list[i + 1] = fn_ptr->vf(what->buf, what->stride,
                                      get_buf_from_mv(in_what, &this_mv),
                                      in_what->stride, &sse) +
            // mvsad_err_cost(x, &this_mv, &fcenter_mv, sadpb);
            mv_err_cost(&this_mv, &fcenter_mv, x->nmvjointcost, x->mvcost,
                        x->errorperbit);
    }
  }
}

// Generic pattern search function that searches over multiple scales.
// Each scale can have a different number of candidates and shape of
// candidates as indicated in the num_candidates and candidates arrays
// passed into this function
//
static int vp10_pattern_search(const MACROBLOCK *x,
                              MV *ref_mv,
                              int search_param,
                              int sad_per_bit,
                              int do_init_search,
                              int *cost_list,
                              const vp9_variance_fn_ptr_t *vfp,
                              int use_mvcost,
                              const MV *center_mv,
                              MV *best_mv,
                              const int num_candidates[MAX_PATTERN_SCALES],
                              const MV candidates[MAX_PATTERN_SCALES]
                                                 [MAX_PATTERN_CANDIDATES]) {
  const MACROBLOCKD *const xd = &x->e_mbd;
  static const int search_param_to_steps[MAX_MVSEARCH_STEPS] = {
    10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0,
  };
  int i, s, t;
  const struct buf_2d *const what = &x->plane[0].src;
  const struct buf_2d *const in_what = &xd->plane[0].pre[0];
  int br, bc;
  int bestsad = INT_MAX;
  int thissad;
  int k = -1;
  const MV fcenter_mv = {center_mv->row >> 3, center_mv->col >> 3};
  int best_init_s = search_param_to_steps[search_param];
  // adjust ref_mv to make sure it is within MV range
  clamp_mv(ref_mv, x->mv_col_min, x->mv_col_max, x->mv_row_min, x->mv_row_max);
  br = ref_mv->row;
  bc = ref_mv->col;

  // Work out the start point for the search
  bestsad = vfp->sdf(what->buf, what->stride,
                     get_buf_from_mv(in_what, ref_mv), in_what->stride) +
      mvsad_err_cost(x, ref_mv, &fcenter_mv, sad_per_bit);

  // Search all possible scales upto the search param around the center point
  // pick the scale of the point that is best as the starting scale of
  // further steps around it.
  if (do_init_search) {
    s = best_init_s;
    best_init_s = -1;
    for (t = 0; t <= s; ++t) {
      int best_site = -1;
      if (check_bounds(x, br, bc, 1 << t)) {
        for (i = 0; i < num_candidates[t]; i++) {
          const MV this_mv = {br + candidates[t][i].row,
                              bc + candidates[t][i].col};
          thissad = vfp->sdf(what->buf, what->stride,
                             get_buf_from_mv(in_what, &this_mv),
                             in_what->stride);
          CHECK_BETTER
        }
      } else {
        for (i = 0; i < num_candidates[t]; i++) {
          const MV this_mv = {br + candidates[t][i].row,
                              bc + candidates[t][i].col};
          if (!is_mv_in(x, &this_mv))
            continue;
          thissad = vfp->sdf(what->buf, what->stride,
                             get_buf_from_mv(in_what, &this_mv),
                             in_what->stride);
          CHECK_BETTER
        }
      }
      if (best_site == -1) {
        continue;
      } else {
        best_init_s = t;
        k = best_site;
      }
    }
    if (best_init_s != -1) {
      br += candidates[best_init_s][k].row;
      bc += candidates[best_init_s][k].col;
    }
  }

  // If the center point is still the best, just skip this and move to
  // the refinement step.
  if (best_init_s != -1) {
    int best_site = -1;
    s = best_init_s;

    do {
      // No need to search all 6 points the 1st time if initial search was used
      if (!do_init_search || s != best_init_s) {
        if (check_bounds(x, br, bc, 1 << s)) {
          for (i = 0; i < num_candidates[s]; i++) {
            const MV this_mv = {br + candidates[s][i].row,
                                bc + candidates[s][i].col};
            thissad = vfp->sdf(what->buf, what->stride,
                               get_buf_from_mv(in_what, &this_mv),
                               in_what->stride);
            CHECK_BETTER
          }
        } else {
          for (i = 0; i < num_candidates[s]; i++) {
            const MV this_mv = {br + candidates[s][i].row,
                                bc + candidates[s][i].col};
            if (!is_mv_in(x, &this_mv))
              continue;
            thissad = vfp->sdf(what->buf, what->stride,
                               get_buf_from_mv(in_what, &this_mv),
                               in_what->stride);
            CHECK_BETTER
          }
        }

        if (best_site == -1) {
          continue;
        } else {
          br += candidates[s][best_site].row;
          bc += candidates[s][best_site].col;
          k = best_site;
        }
      }

      do {
        int next_chkpts_indices[PATTERN_CANDIDATES_REF];
        best_site = -1;
        next_chkpts_indices[0] = (k == 0) ? num_candidates[s] - 1 : k - 1;
        next_chkpts_indices[1] = k;
        next_chkpts_indices[2] = (k == num_candidates[s] - 1) ? 0 : k + 1;

        if (check_bounds(x, br, bc, 1 << s)) {
          for (i = 0; i < PATTERN_CANDIDATES_REF; i++) {
            const MV this_mv = {br + candidates[s][next_chkpts_indices[i]].row,
                                bc + candidates[s][next_chkpts_indices[i]].col};
            thissad = vfp->sdf(what->buf, what->stride,
                               get_buf_from_mv(in_what, &this_mv),
                               in_what->stride);
            CHECK_BETTER
          }
        } else {
          for (i = 0; i < PATTERN_CANDIDATES_REF; i++) {
            const MV this_mv = {br + candidates[s][next_chkpts_indices[i]].row,
                                bc + candidates[s][next_chkpts_indices[i]].col};
            if (!is_mv_in(x, &this_mv))
              continue;
            thissad = vfp->sdf(what->buf, what->stride,
                               get_buf_from_mv(in_what, &this_mv),
                               in_what->stride);
            CHECK_BETTER
          }
        }

        if (best_site != -1) {
          k = next_chkpts_indices[best_site];
          br += candidates[s][k].row;
          bc += candidates[s][k].col;
        }
      } while (best_site != -1);
    } while (s--);
  }

  // Returns the one-away integer pel sad values around the best as follows:
  // cost_list[0]: cost at the best integer pel
  // cost_list[1]: cost at delta {0, -1} (left)   from the best integer pel
  // cost_list[2]: cost at delta { 1, 0} (bottom) from the best integer pel
  // cost_list[3]: cost at delta { 0, 1} (right)  from the best integer pel
  // cost_list[4]: cost at delta {-1, 0} (top)    from the best integer pel
  if (cost_list) {
    const MV best_mv = { br, bc };
    calc_int_cost_list(x, &fcenter_mv, sad_per_bit, vfp, &best_mv, cost_list);
  }
  best_mv->row = br;
  best_mv->col = bc;
  return bestsad;
}

// A specialized function where the smallest scale search candidates
// are 4 1-away neighbors, and cost_list is non-null
// TODO(debargha): Merge this function with the one above. Also remove
// use_mvcost option since it is always 1, to save unnecessary branches.
static int vp10_pattern_search_sad(const MACROBLOCK *x,
                                  MV *ref_mv,
                                  int search_param,
                                  int sad_per_bit,
                                  int do_init_search,
                                  int *cost_list,
                                  const vp9_variance_fn_ptr_t *vfp,
                                  int use_mvcost,
                                  const MV *center_mv,
                                  MV *best_mv,
                                  const int num_candidates[MAX_PATTERN_SCALES],
                                  const MV candidates[MAX_PATTERN_SCALES]
                                                     [MAX_PATTERN_CANDIDATES]) {
  const MACROBLOCKD *const xd = &x->e_mbd;
  static const int search_param_to_steps[MAX_MVSEARCH_STEPS] = {
    10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0,
  };
  int i, s, t;
  const struct buf_2d *const what = &x->plane[0].src;
  const struct buf_2d *const in_what = &xd->plane[0].pre[0];
  int br, bc;
  int bestsad = INT_MAX;
  int thissad;
  int k = -1;
  const MV fcenter_mv = {center_mv->row >> 3, center_mv->col >> 3};
  int best_init_s = search_param_to_steps[search_param];
  // adjust ref_mv to make sure it is within MV range
  clamp_mv(ref_mv, x->mv_col_min, x->mv_col_max, x->mv_row_min, x->mv_row_max);
  br = ref_mv->row;
  bc = ref_mv->col;
  if (cost_list != NULL) {
    cost_list[0] = cost_list[1] = cost_list[2] = cost_list[3] = cost_list[4] =
        INT_MAX;
  }

  // Work out the start point for the search
  bestsad = vfp->sdf(what->buf, what->stride,
                     get_buf_from_mv(in_what, ref_mv), in_what->stride) +
      mvsad_err_cost(x, ref_mv, &fcenter_mv, sad_per_bit);

  // Search all possible scales upto the search param around the center point
  // pick the scale of the point that is best as the starting scale of
  // further steps around it.
  if (do_init_search) {
    s = best_init_s;
    best_init_s = -1;
    for (t = 0; t <= s; ++t) {
      int best_site = -1;
      if (check_bounds(x, br, bc, 1 << t)) {
        for (i = 0; i < num_candidates[t]; i++) {
          const MV this_mv = {br + candidates[t][i].row,
                              bc + candidates[t][i].col};
          thissad = vfp->sdf(what->buf, what->stride,
                             get_buf_from_mv(in_what, &this_mv),
                             in_what->stride);
          CHECK_BETTER
        }
      } else {
        for (i = 0; i < num_candidates[t]; i++) {
          const MV this_mv = {br + candidates[t][i].row,
                              bc + candidates[t][i].col};
          if (!is_mv_in(x, &this_mv))
            continue;
          thissad = vfp->sdf(what->buf, what->stride,
                             get_buf_from_mv(in_what, &this_mv),
                             in_what->stride);
          CHECK_BETTER
        }
      }
      if (best_site == -1) {
        continue;
      } else {
        best_init_s = t;
        k = best_site;
      }
    }
    if (best_init_s != -1) {
      br += candidates[best_init_s][k].row;
      bc += candidates[best_init_s][k].col;
    }
  }

  // If the center point is still the best, just skip this and move to
  // the refinement step.
  if (best_init_s != -1) {
    int do_sad = (num_candidates[0] == 4 && cost_list != NULL);
    int best_site = -1;
    s = best_init_s;

    for (; s >= do_sad; s--) {
      if (!do_init_search || s != best_init_s) {
        if (check_bounds(x, br, bc, 1 << s)) {
          for (i = 0; i < num_candidates[s]; i++) {
            const MV this_mv = {br + candidates[s][i].row,
                                bc + candidates[s][i].col};
            thissad = vfp->sdf(what->buf, what->stride,
                               get_buf_from_mv(in_what, &this_mv),
                               in_what->stride);
            CHECK_BETTER
          }
        } else {
          for (i = 0; i < num_candidates[s]; i++) {
            const MV this_mv = {br + candidates[s][i].row,
                                bc + candidates[s][i].col};
            if (!is_mv_in(x, &this_mv))
              continue;
            thissad = vfp->sdf(what->buf, what->stride,
                               get_buf_from_mv(in_what, &this_mv),
                               in_what->stride);
            CHECK_BETTER
          }
        }

        if (best_site == -1) {
          continue;
        } else {
          br += candidates[s][best_site].row;
          bc += candidates[s][best_site].col;
          k = best_site;
        }
      }

      do {
        int next_chkpts_indices[PATTERN_CANDIDATES_REF];
        best_site = -1;
        next_chkpts_indices[0] = (k == 0) ? num_candidates[s] - 1 : k - 1;
        next_chkpts_indices[1] = k;
        next_chkpts_indices[2] = (k == num_candidates[s] - 1) ? 0 : k + 1;

        if (check_bounds(x, br, bc, 1 << s)) {
          for (i = 0; i < PATTERN_CANDIDATES_REF; i++) {
            const MV this_mv = {br + candidates[s][next_chkpts_indices[i]].row,
                                bc + candidates[s][next_chkpts_indices[i]].col};
            thissad = vfp->sdf(what->buf, what->stride,
                               get_buf_from_mv(in_what, &this_mv),
                               in_what->stride);
            CHECK_BETTER
          }
        } else {
          for (i = 0; i < PATTERN_CANDIDATES_REF; i++) {
            const MV this_mv = {br + candidates[s][next_chkpts_indices[i]].row,
                                bc + candidates[s][next_chkpts_indices[i]].col};
            if (!is_mv_in(x, &this_mv))
              continue;
            thissad = vfp->sdf(what->buf, what->stride,
                               get_buf_from_mv(in_what, &this_mv),
                               in_what->stride);
            CHECK_BETTER
          }
        }

        if (best_site != -1) {
          k = next_chkpts_indices[best_site];
          br += candidates[s][k].row;
          bc += candidates[s][k].col;
        }
      } while (best_site != -1);
    }

    // Note: If we enter the if below, then cost_list must be non-NULL.
    if (s == 0) {
      cost_list[0] = bestsad;
      if (!do_init_search || s != best_init_s) {
        if (check_bounds(x, br, bc, 1 << s)) {
          for (i = 0; i < num_candidates[s]; i++) {
            const MV this_mv = {br + candidates[s][i].row,
                                bc + candidates[s][i].col};
            cost_list[i + 1] =
            thissad = vfp->sdf(what->buf, what->stride,
                               get_buf_from_mv(in_what, &this_mv),
                               in_what->stride);
            CHECK_BETTER
          }
        } else {
          for (i = 0; i < num_candidates[s]; i++) {
            const MV this_mv = {br + candidates[s][i].row,
                                bc + candidates[s][i].col};
            if (!is_mv_in(x, &this_mv))
              continue;
            cost_list[i + 1] =
            thissad = vfp->sdf(what->buf, what->stride,
                               get_buf_from_mv(in_what, &this_mv),
                               in_what->stride);
            CHECK_BETTER
          }
        }

        if (best_site != -1) {
          br += candidates[s][best_site].row;
          bc += candidates[s][best_site].col;
          k = best_site;
        }
      }
      while (best_site != -1) {
        int next_chkpts_indices[PATTERN_CANDIDATES_REF];
        best_site = -1;
        next_chkpts_indices[0] = (k == 0) ? num_candidates[s] - 1 : k - 1;
        next_chkpts_indices[1] = k;
        next_chkpts_indices[2] = (k == num_candidates[s] - 1) ? 0 : k + 1;
        cost_list[1] = cost_list[2] = cost_list[3] = cost_list[4] = INT_MAX;
        cost_list[((k + 2) % 4) + 1] = cost_list[0];
        cost_list[0] = bestsad;

        if (check_bounds(x, br, bc, 1 << s)) {
          for (i = 0; i < PATTERN_CANDIDATES_REF; i++) {
            const MV this_mv = {br + candidates[s][next_chkpts_indices[i]].row,
                                bc + candidates[s][next_chkpts_indices[i]].col};
            cost_list[next_chkpts_indices[i] + 1] =
            thissad = vfp->sdf(what->buf, what->stride,
                               get_buf_from_mv(in_what, &this_mv),
                               in_what->stride);
            CHECK_BETTER
          }
        } else {
          for (i = 0; i < PATTERN_CANDIDATES_REF; i++) {
            const MV this_mv = {br + candidates[s][next_chkpts_indices[i]].row,
                                bc + candidates[s][next_chkpts_indices[i]].col};
            if (!is_mv_in(x, &this_mv)) {
              cost_list[next_chkpts_indices[i] + 1] = INT_MAX;
              continue;
            }
            cost_list[next_chkpts_indices[i] + 1] =
            thissad = vfp->sdf(what->buf, what->stride,
                               get_buf_from_mv(in_what, &this_mv),
                               in_what->stride);
            CHECK_BETTER
          }
        }

        if (best_site != -1) {
          k = next_chkpts_indices[best_site];
          br += candidates[s][k].row;
          bc += candidates[s][k].col;
        }
      }
    }
  }

  // Returns the one-away integer pel sad values around the best as follows:
  // cost_list[0]: sad at the best integer pel
  // cost_list[1]: sad at delta {0, -1} (left)   from the best integer pel
  // cost_list[2]: sad at delta { 1, 0} (bottom) from the best integer pel
  // cost_list[3]: sad at delta { 0, 1} (right)  from the best integer pel
  // cost_list[4]: sad at delta {-1, 0} (top)    from the best integer pel
  if (cost_list) {
    static const MV neighbors[4] = {{0, -1}, {1, 0}, {0, 1}, {-1, 0}};
    if (cost_list[0] == INT_MAX) {
      cost_list[0] = bestsad;
      if (check_bounds(x, br, bc, 1)) {
        for (i = 0; i < 4; i++) {
          const MV this_mv = { br + neighbors[i].row,
                               bc + neighbors[i].col };
          cost_list[i + 1] = vfp->sdf(what->buf, what->stride,
                                     get_buf_from_mv(in_what, &this_mv),
                                     in_what->stride);
        }
      } else {
        for (i = 0; i < 4; i++) {
          const MV this_mv = {br + neighbors[i].row,
            bc + neighbors[i].col};
          if (!is_mv_in(x, &this_mv))
            cost_list[i + 1] = INT_MAX;
          else
            cost_list[i + 1] = vfp->sdf(what->buf, what->stride,
                                       get_buf_from_mv(in_what, &this_mv),
                                       in_what->stride);
        }
      }
    } else {
      if (use_mvcost) {
        for (i = 0; i < 4; i++) {
          const MV this_mv = {br + neighbors[i].row,
            bc + neighbors[i].col};
          if (cost_list[i + 1] != INT_MAX) {
            cost_list[i + 1] +=
                mvsad_err_cost(x, &this_mv, &fcenter_mv, sad_per_bit);
          }
        }
      }
    }
  }
  best_mv->row = br;
  best_mv->col = bc;
  return bestsad;
}

int vp10_get_mvpred_var(const MACROBLOCK *x,
                       const MV *best_mv, const MV *center_mv,
                       const vp9_variance_fn_ptr_t *vfp,
                       int use_mvcost) {
  const MACROBLOCKD *const xd = &x->e_mbd;
  const struct buf_2d *const what = &x->plane[0].src;
  const struct buf_2d *const in_what = &xd->plane[0].pre[0];
  const MV mv = {best_mv->row * 8, best_mv->col * 8};
  unsigned int unused;

  return vfp->vf(what->buf, what->stride,
                 get_buf_from_mv(in_what, best_mv), in_what->stride, &unused) +
      (use_mvcost ?  mv_err_cost(&mv, center_mv, x->nmvjointcost,
                                 x->mvcost, x->errorperbit) : 0);
}

int vp10_get_mvpred_av_var(const MACROBLOCK *x,
                          const MV *best_mv, const MV *center_mv,
                          const uint8_t *second_pred,
                          const vp9_variance_fn_ptr_t *vfp,
                          int use_mvcost) {
  const MACROBLOCKD *const xd = &x->e_mbd;
  const struct buf_2d *const what = &x->plane[0].src;
  const struct buf_2d *const in_what = &xd->plane[0].pre[0];
  const MV mv = {best_mv->row * 8, best_mv->col * 8};
  unsigned int unused;

  return vfp->svaf(get_buf_from_mv(in_what, best_mv), in_what->stride, 0, 0,
                   what->buf, what->stride, &unused, second_pred) +
      (use_mvcost ?  mv_err_cost(&mv, center_mv, x->nmvjointcost,
                                 x->mvcost, x->errorperbit) : 0);
}

int vp10_hex_search(const MACROBLOCK *x,
                   MV *ref_mv,
                   int search_param,
                   int sad_per_bit,
                   int do_init_search,
                   int *cost_list,
                   const vp9_variance_fn_ptr_t *vfp,
                   int use_mvcost,
                   const MV *center_mv, MV *best_mv) {
  // First scale has 8-closest points, the rest have 6 points in hex shape
  // at increasing scales
  static const int hex_num_candidates[MAX_PATTERN_SCALES] = {
    8, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6
  };
  // Note that the largest candidate step at each scale is 2^scale
  static const MV hex_candidates[MAX_PATTERN_SCALES][MAX_PATTERN_CANDIDATES] = {
    {{-1, -1}, {0, -1}, {1, -1}, {1, 0}, {1, 1}, { 0, 1}, { -1, 1}, {-1, 0}},
    {{-1, -2}, {1, -2}, {2, 0}, {1, 2}, { -1, 2}, { -2, 0}},
    {{-2, -4}, {2, -4}, {4, 0}, {2, 4}, { -2, 4}, { -4, 0}},
    {{-4, -8}, {4, -8}, {8, 0}, {4, 8}, { -4, 8}, { -8, 0}},
    {{-8, -16}, {8, -16}, {16, 0}, {8, 16}, { -8, 16}, { -16, 0}},
    {{-16, -32}, {16, -32}, {32, 0}, {16, 32}, { -16, 32}, { -32, 0}},
    {{-32, -64}, {32, -64}, {64, 0}, {32, 64}, { -32, 64}, { -64, 0}},
    {{-64, -128}, {64, -128}, {128, 0}, {64, 128}, { -64, 128}, { -128, 0}},
    {{-128, -256}, {128, -256}, {256, 0}, {128, 256}, { -128, 256}, { -256, 0}},
    {{-256, -512}, {256, -512}, {512, 0}, {256, 512}, { -256, 512}, { -512, 0}},
    {{-512, -1024}, {512, -1024}, {1024, 0}, {512, 1024}, { -512, 1024},
      { -1024, 0}},
  };
  return vp10_pattern_search(x, ref_mv, search_param, sad_per_bit,
                            do_init_search, cost_list, vfp, use_mvcost,
                            center_mv, best_mv,
                            hex_num_candidates, hex_candidates);
}

int vp10_bigdia_search(const MACROBLOCK *x,
                      MV *ref_mv,
                      int search_param,
                      int sad_per_bit,
                      int do_init_search,
                      int *cost_list,
                      const vp9_variance_fn_ptr_t *vfp,
                      int use_mvcost,
                      const MV *center_mv,
                      MV *best_mv) {
  // First scale has 4-closest points, the rest have 8 points in diamond
  // shape at increasing scales
  static const int bigdia_num_candidates[MAX_PATTERN_SCALES] = {
    4, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
  };
  // Note that the largest candidate step at each scale is 2^scale
  static const MV bigdia_candidates[MAX_PATTERN_SCALES]
                                   [MAX_PATTERN_CANDIDATES] = {
    {{0, -1}, {1, 0}, { 0, 1}, {-1, 0}},
    {{-1, -1}, {0, -2}, {1, -1}, {2, 0}, {1, 1}, {0, 2}, {-1, 1}, {-2, 0}},
    {{-2, -2}, {0, -4}, {2, -2}, {4, 0}, {2, 2}, {0, 4}, {-2, 2}, {-4, 0}},
    {{-4, -4}, {0, -8}, {4, -4}, {8, 0}, {4, 4}, {0, 8}, {-4, 4}, {-8, 0}},
    {{-8, -8}, {0, -16}, {8, -8}, {16, 0}, {8, 8}, {0, 16}, {-8, 8}, {-16, 0}},
    {{-16, -16}, {0, -32}, {16, -16}, {32, 0}, {16, 16}, {0, 32},
      {-16, 16}, {-32, 0}},
    {{-32, -32}, {0, -64}, {32, -32}, {64, 0}, {32, 32}, {0, 64},
      {-32, 32}, {-64, 0}},
    {{-64, -64}, {0, -128}, {64, -64}, {128, 0}, {64, 64}, {0, 128},
      {-64, 64}, {-128, 0}},
    {{-128, -128}, {0, -256}, {128, -128}, {256, 0}, {128, 128}, {0, 256},
      {-128, 128}, {-256, 0}},
    {{-256, -256}, {0, -512}, {256, -256}, {512, 0}, {256, 256}, {0, 512},
      {-256, 256}, {-512, 0}},
    {{-512, -512}, {0, -1024}, {512, -512}, {1024, 0}, {512, 512}, {0, 1024},
      {-512, 512}, {-1024, 0}},
  };
  return vp10_pattern_search_sad(x, ref_mv, search_param, sad_per_bit,
                                do_init_search, cost_list, vfp, use_mvcost,
                                center_mv, best_mv,
                                bigdia_num_candidates, bigdia_candidates);
}

int vp10_square_search(const MACROBLOCK *x,
                      MV *ref_mv,
                      int search_param,
                      int sad_per_bit,
                      int do_init_search,
                      int *cost_list,
                      const vp9_variance_fn_ptr_t *vfp,
                      int use_mvcost,
                      const MV *center_mv,
                      MV *best_mv) {
  // All scales have 8 closest points in square shape
  static const int square_num_candidates[MAX_PATTERN_SCALES] = {
    8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
  };
  // Note that the largest candidate step at each scale is 2^scale
  static const MV square_candidates[MAX_PATTERN_SCALES]
                                   [MAX_PATTERN_CANDIDATES] = {
    {{-1, -1}, {0, -1}, {1, -1}, {1, 0}, {1, 1}, {0, 1}, {-1, 1}, {-1, 0}},
    {{-2, -2}, {0, -2}, {2, -2}, {2, 0}, {2, 2}, {0, 2}, {-2, 2}, {-2, 0}},
    {{-4, -4}, {0, -4}, {4, -4}, {4, 0}, {4, 4}, {0, 4}, {-4, 4}, {-4, 0}},
    {{-8, -8}, {0, -8}, {8, -8}, {8, 0}, {8, 8}, {0, 8}, {-8, 8}, {-8, 0}},
    {{-16, -16}, {0, -16}, {16, -16}, {16, 0}, {16, 16}, {0, 16},
      {-16, 16}, {-16, 0}},
    {{-32, -32}, {0, -32}, {32, -32}, {32, 0}, {32, 32}, {0, 32},
      {-32, 32}, {-32, 0}},
    {{-64, -64}, {0, -64}, {64, -64}, {64, 0}, {64, 64}, {0, 64},
      {-64, 64}, {-64, 0}},
    {{-128, -128}, {0, -128}, {128, -128}, {128, 0}, {128, 128}, {0, 128},
      {-128, 128}, {-128, 0}},
    {{-256, -256}, {0, -256}, {256, -256}, {256, 0}, {256, 256}, {0, 256},
      {-256, 256}, {-256, 0}},
    {{-512, -512}, {0, -512}, {512, -512}, {512, 0}, {512, 512}, {0, 512},
      {-512, 512}, {-512, 0}},
    {{-1024, -1024}, {0, -1024}, {1024, -1024}, {1024, 0}, {1024, 1024},
      {0, 1024}, {-1024, 1024}, {-1024, 0}},
  };
  return vp10_pattern_search(x, ref_mv, search_param, sad_per_bit,
                            do_init_search, cost_list, vfp, use_mvcost,
                            center_mv, best_mv,
                            square_num_candidates, square_candidates);
}

int vp10_fast_hex_search(const MACROBLOCK *x,
                        MV *ref_mv,
                        int search_param,
                        int sad_per_bit,
                        int do_init_search,  // must be zero for fast_hex
                        int *cost_list,
                        const vp9_variance_fn_ptr_t *vfp,
                        int use_mvcost,
                        const MV *center_mv,
                        MV *best_mv) {
  return vp10_hex_search(x, ref_mv, MAX(MAX_MVSEARCH_STEPS - 2, search_param),
                        sad_per_bit, do_init_search, cost_list, vfp, use_mvcost,
                        center_mv, best_mv);
}

int vp10_fast_dia_search(const MACROBLOCK *x,
                        MV *ref_mv,
                        int search_param,
                        int sad_per_bit,
                        int do_init_search,
                        int *cost_list,
                        const vp9_variance_fn_ptr_t *vfp,
                        int use_mvcost,
                        const MV *center_mv,
                        MV *best_mv) {
  return vp10_bigdia_search(x, ref_mv, MAX(MAX_MVSEARCH_STEPS - 2, search_param),
                           sad_per_bit, do_init_search, cost_list, vfp,
                           use_mvcost, center_mv, best_mv);
}

#undef CHECK_BETTER

int vp10_full_range_search_c(const MACROBLOCK *x,
                            const search_site_config *cfg,
                            MV *ref_mv, MV *best_mv,
                            int search_param, int sad_per_bit, int *num00,
                            const vp9_variance_fn_ptr_t *fn_ptr,
                            const MV *center_mv) {
  const MACROBLOCKD *const xd = &x->e_mbd;
  const struct buf_2d *const what = &x->plane[0].src;
  const struct buf_2d *const in_what = &xd->plane[0].pre[0];
  const int range = 64;
  const MV fcenter_mv = {center_mv->row >> 3, center_mv->col >> 3};
  unsigned int best_sad = INT_MAX;
  int r, c, i;
  int start_col, end_col, start_row, end_row;

  // The cfg and search_param parameters are not used in this search variant
  (void)cfg;
  (void)search_param;

  clamp_mv(ref_mv, x->mv_col_min, x->mv_col_max, x->mv_row_min, x->mv_row_max);
  *best_mv = *ref_mv;
  *num00 = 11;
  best_sad = fn_ptr->sdf(what->buf, what->stride,
                         get_buf_from_mv(in_what, ref_mv), in_what->stride) +
                 mvsad_err_cost(x, ref_mv, &fcenter_mv, sad_per_bit);
  start_row = MAX(-range, x->mv_row_min - ref_mv->row);
  start_col = MAX(-range, x->mv_col_min - ref_mv->col);
  end_row = MIN(range, x->mv_row_max - ref_mv->row);
  end_col = MIN(range, x->mv_col_max - ref_mv->col);

  for (r = start_row; r <= end_row; ++r) {
    for (c = start_col; c <= end_col; c += 4) {
      if (c + 3 <= end_col) {
        unsigned int sads[4];
        const uint8_t *addrs[4];
        for (i = 0; i < 4; ++i) {
          const MV mv = {ref_mv->row + r, ref_mv->col + c + i};
          addrs[i] = get_buf_from_mv(in_what, &mv);
        }

        fn_ptr->sdx4df(what->buf, what->stride, addrs, in_what->stride, sads);

        for (i = 0; i < 4; ++i) {
          if (sads[i] < best_sad) {
            const MV mv = {ref_mv->row + r, ref_mv->col + c + i};
            const unsigned int sad = sads[i] +
                mvsad_err_cost(x, &mv, &fcenter_mv, sad_per_bit);
            if (sad < best_sad) {
              best_sad = sad;
              *best_mv = mv;
            }
          }
        }
      } else {
        for (i = 0; i < end_col - c; ++i) {
          const MV mv = {ref_mv->row + r, ref_mv->col + c + i};
          unsigned int sad = fn_ptr->sdf(what->buf, what->stride,
              get_buf_from_mv(in_what, &mv), in_what->stride);
          if (sad < best_sad) {
            sad += mvsad_err_cost(x, &mv, &fcenter_mv, sad_per_bit);
            if (sad < best_sad) {
              best_sad = sad;
              *best_mv = mv;
            }
          }
        }
      }
    }
  }

  return best_sad;
}

int vp10_diamond_search_sad_c(const MACROBLOCK *x,
                             const search_site_config *cfg,
                             MV *ref_mv, MV *best_mv, int search_param,
                             int sad_per_bit, int *num00,
                             const vp9_variance_fn_ptr_t *fn_ptr,
                             const MV *center_mv) {
  int i, j, step;

  const MACROBLOCKD *const xd = &x->e_mbd;
  uint8_t *what = x->plane[0].src.buf;
  const int what_stride = x->plane[0].src.stride;
  const uint8_t *in_what;
  const int in_what_stride = xd->plane[0].pre[0].stride;
  const uint8_t *best_address;

  unsigned int bestsad = INT_MAX;
  int best_site = 0;
  int last_site = 0;

  int ref_row;
  int ref_col;

  // search_param determines the length of the initial step and hence the number
  // of iterations.
  // 0 = initial step (MAX_FIRST_STEP) pel
  // 1 = (MAX_FIRST_STEP/2) pel,
  // 2 = (MAX_FIRST_STEP/4) pel...
  const search_site *ss = &cfg->ss[search_param * cfg->searches_per_step];
  const int tot_steps = (cfg->ss_count / cfg->searches_per_step) - search_param;

  const MV fcenter_mv = {center_mv->row >> 3, center_mv->col >> 3};
  clamp_mv(ref_mv, x->mv_col_min, x->mv_col_max, x->mv_row_min, x->mv_row_max);
  ref_row = ref_mv->row;
  ref_col = ref_mv->col;
  *num00 = 0;
  best_mv->row = ref_row;
  best_mv->col = ref_col;

  // Work out the start point for the search
  in_what = xd->plane[0].pre[0].buf + ref_row * in_what_stride + ref_col;
  best_address = in_what;

  // Check the starting position
  bestsad = fn_ptr->sdf(what, what_stride, in_what, in_what_stride)
                + mvsad_err_cost(x, best_mv, &fcenter_mv, sad_per_bit);

  i = 1;

  for (step = 0; step < tot_steps; step++) {
    int all_in = 1, t;

    // All_in is true if every one of the points we are checking are within
    // the bounds of the image.
    all_in &= ((best_mv->row + ss[i].mv.row) > x->mv_row_min);
    all_in &= ((best_mv->row + ss[i + 1].mv.row) < x->mv_row_max);
    all_in &= ((best_mv->col + ss[i + 2].mv.col) > x->mv_col_min);
    all_in &= ((best_mv->col + ss[i + 3].mv.col) < x->mv_col_max);

    // If all the pixels are within the bounds we don't check whether the
    // search point is valid in this loop,  otherwise we check each point
    // for validity..
    if (all_in) {
      unsigned int sad_array[4];

      for (j = 0; j < cfg->searches_per_step; j += 4) {
        unsigned char const *block_offset[4];

        for (t = 0; t < 4; t++)
          block_offset[t] = ss[i + t].offset + best_address;

        fn_ptr->sdx4df(what, what_stride, block_offset, in_what_stride,
                       sad_array);

        for (t = 0; t < 4; t++, i++) {
          if (sad_array[t] < bestsad) {
            const MV this_mv = {best_mv->row + ss[i].mv.row,
                                best_mv->col + ss[i].mv.col};
            sad_array[t] += mvsad_err_cost(x, &this_mv, &fcenter_mv,
                                           sad_per_bit);
            if (sad_array[t] < bestsad) {
              bestsad = sad_array[t];
              best_site = i;
            }
          }
        }
      }
    } else {
      for (j = 0; j < cfg->searches_per_step; j++) {
        // Trap illegal vectors
        const MV this_mv = {best_mv->row + ss[i].mv.row,
                            best_mv->col + ss[i].mv.col};

        if (is_mv_in(x, &this_mv)) {
          const uint8_t *const check_here = ss[i].offset + best_address;
          unsigned int thissad = fn_ptr->sdf(what, what_stride, check_here,
                                             in_what_stride);

          if (thissad < bestsad) {
            thissad += mvsad_err_cost(x, &this_mv, &fcenter_mv, sad_per_bit);
            if (thissad < bestsad) {
              bestsad = thissad;
              best_site = i;
            }
          }
        }
        i++;
      }
    }
    if (best_site != last_site) {
      best_mv->row += ss[best_site].mv.row;
      best_mv->col += ss[best_site].mv.col;
      best_address += ss[best_site].offset;
      last_site = best_site;
#if defined(NEW_DIAMOND_SEARCH)
      while (1) {
        const MV this_mv = {best_mv->row + ss[best_site].mv.row,
                            best_mv->col + ss[best_site].mv.col};
        if (is_mv_in(x, &this_mv)) {
          const uint8_t *const check_here = ss[best_site].offset + best_address;
          unsigned int thissad = fn_ptr->sdf(what, what_stride, check_here,
                                             in_what_stride);
          if (thissad < bestsad) {
            thissad += mvsad_err_cost(x, &this_mv, &fcenter_mv, sad_per_bit);
            if (thissad < bestsad) {
              bestsad = thissad;
              best_mv->row += ss[best_site].mv.row;
              best_mv->col += ss[best_site].mv.col;
              best_address += ss[best_site].offset;
              continue;
            }
          }
        }
        break;
      }
#endif
    } else if (best_address == in_what) {
      (*num00)++;
    }
  }
  return bestsad;
}

static int vector_match(int16_t *ref, int16_t *src, int bwl) {
  int best_sad = INT_MAX;
  int this_sad;
  int d;
  int center, offset = 0;
  int bw = 4 << bwl;  // redundant variable, to be changed in the experiments.
  for (d = 0; d <= bw; d += 16) {
    this_sad = vp10_vector_var(&ref[d], src, bwl);
    if (this_sad < best_sad) {
      best_sad = this_sad;
      offset = d;
    }
  }
  center = offset;

  for (d = -8; d <= 8; d += 16) {
    int this_pos = offset + d;
    // check limit
    if (this_pos < 0 || this_pos > bw)
      continue;
    this_sad = vp10_vector_var(&ref[this_pos], src, bwl);
    if (this_sad < best_sad) {
      best_sad = this_sad;
      center = this_pos;
    }
  }
  offset = center;

  for (d = -4; d <= 4; d += 8) {
    int this_pos = offset + d;
    // check limit
    if (this_pos < 0 || this_pos > bw)
      continue;
    this_sad = vp10_vector_var(&ref[this_pos], src, bwl);
    if (this_sad < best_sad) {
      best_sad = this_sad;
      center = this_pos;
    }
  }
  offset = center;

  for (d = -2; d <= 2; d += 4) {
    int this_pos = offset + d;
    // check limit
    if (this_pos < 0 || this_pos > bw)
      continue;
    this_sad = vp10_vector_var(&ref[this_pos], src, bwl);
    if (this_sad < best_sad) {
      best_sad = this_sad;
      center = this_pos;
    }
  }
  offset = center;

  for (d = -1; d <= 1; d += 2) {
    int this_pos = offset + d;
    // check limit
    if (this_pos < 0 || this_pos > bw)
      continue;
    this_sad = vp10_vector_var(&ref[this_pos], src, bwl);
    if (this_sad < best_sad) {
      best_sad = this_sad;
      center = this_pos;
    }
  }

  return (center - (bw >> 1));
}

static const MV search_pos[4] = {
    {-1, 0}, {0, -1}, {0, 1}, {1, 0},
};

unsigned int vp10_int_pro_motion_estimation(const VP10_COMP *cpi, MACROBLOCK *x,
                                           BLOCK_SIZE bsize,
                                           int mi_row, int mi_col) {
  MACROBLOCKD *xd = &x->e_mbd;
  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
  struct buf_2d backup_yv12[MAX_MB_PLANE] = {{0, 0}};
  DECLARE_ALIGNED(16, int16_t, hbuf[128]);
  DECLARE_ALIGNED(16, int16_t, vbuf[128]);
  DECLARE_ALIGNED(16, int16_t, src_hbuf[64]);
  DECLARE_ALIGNED(16, int16_t, src_vbuf[64]);
  int idx;
  const int bw = 4 << b_width_log2_lookup[bsize];
  const int bh = 4 << b_height_log2_lookup[bsize];
  const int search_width = bw << 1;
  const int search_height = bh << 1;
  const int src_stride = x->plane[0].src.stride;
  const int ref_stride = xd->plane[0].pre[0].stride;
  uint8_t const *ref_buf, *src_buf;
  MV *tmp_mv = &xd->mi[0]->mbmi.mv[0].as_mv;
  unsigned int best_sad, tmp_sad, this_sad[4];
  MV this_mv;
  const int norm_factor = 3 + (bw >> 5);
  const YV12_BUFFER_CONFIG *scaled_ref_frame =
      vp10_get_scaled_ref_frame(cpi, mbmi->ref_frame[0]);

  if (scaled_ref_frame) {
    int i;
    // Swap out the reference frame for a version that's been scaled to
    // match the resolution of the current frame, allowing the existing
    // motion search code to be used without additional modifications.
    for (i = 0; i < MAX_MB_PLANE; i++)
      backup_yv12[i] = xd->plane[i].pre[0];
    vp10_setup_pre_planes(xd, 0, scaled_ref_frame, mi_row, mi_col, NULL);
  }

#if CONFIG_VP9_HIGHBITDEPTH
  {
    unsigned int this_sad;
    tmp_mv->row = 0;
    tmp_mv->col = 0;
    this_sad = cpi->fn_ptr[bsize].sdf(x->plane[0].src.buf, src_stride,
                                      xd->plane[0].pre[0].buf, ref_stride);

    if (scaled_ref_frame) {
      int i;
      for (i = 0; i < MAX_MB_PLANE; i++)
        xd->plane[i].pre[0] = backup_yv12[i];
    }
    return this_sad;
  }
#endif

  // Set up prediction 1-D reference set
  ref_buf = xd->plane[0].pre[0].buf - (bw >> 1);
  for (idx = 0; idx < search_width; idx += 16) {
    vp10_int_pro_row(&hbuf[idx], ref_buf, ref_stride, bh);
    ref_buf += 16;
  }

  ref_buf = xd->plane[0].pre[0].buf - (bh >> 1) * ref_stride;
  for (idx = 0; idx < search_height; ++idx) {
    vbuf[idx] = vp10_int_pro_col(ref_buf, bw) >> norm_factor;
    ref_buf += ref_stride;
  }

  // Set up src 1-D reference set
  for (idx = 0; idx < bw; idx += 16) {
    src_buf = x->plane[0].src.buf + idx;
    vp10_int_pro_row(&src_hbuf[idx], src_buf, src_stride, bh);
  }

  src_buf = x->plane[0].src.buf;
  for (idx = 0; idx < bh; ++idx) {
    src_vbuf[idx] = vp10_int_pro_col(src_buf, bw) >> norm_factor;
    src_buf += src_stride;
  }

  // Find the best match per 1-D search
  tmp_mv->col = vector_match(hbuf, src_hbuf, b_width_log2_lookup[bsize]);
  tmp_mv->row = vector_match(vbuf, src_vbuf, b_height_log2_lookup[bsize]);

  this_mv = *tmp_mv;
  src_buf = x->plane[0].src.buf;
  ref_buf = xd->plane[0].pre[0].buf + this_mv.row * ref_stride + this_mv.col;
  best_sad = cpi->fn_ptr[bsize].sdf(src_buf, src_stride, ref_buf, ref_stride);

  {
    const uint8_t * const pos[4] = {
        ref_buf - ref_stride,
        ref_buf - 1,
        ref_buf + 1,
        ref_buf + ref_stride,
    };

    cpi->fn_ptr[bsize].sdx4df(src_buf, src_stride, pos, ref_stride, this_sad);
  }

  for (idx = 0; idx < 4; ++idx) {
    if (this_sad[idx] < best_sad) {
      best_sad = this_sad[idx];
      tmp_mv->row = search_pos[idx].row + this_mv.row;
      tmp_mv->col = search_pos[idx].col + this_mv.col;
    }
  }

  if (this_sad[0] < this_sad[3])
    this_mv.row -= 1;
  else
    this_mv.row += 1;

  if (this_sad[1] < this_sad[2])
    this_mv.col -= 1;
  else
    this_mv.col += 1;

  ref_buf = xd->plane[0].pre[0].buf + this_mv.row * ref_stride + this_mv.col;

  tmp_sad = cpi->fn_ptr[bsize].sdf(src_buf, src_stride,
                                   ref_buf, ref_stride);
  if (best_sad > tmp_sad) {
    *tmp_mv = this_mv;
    best_sad = tmp_sad;
  }

  tmp_mv->row *= 8;
  tmp_mv->col *= 8;

  if (scaled_ref_frame) {
    int i;
    for (i = 0; i < MAX_MB_PLANE; i++)
      xd->plane[i].pre[0] = backup_yv12[i];
  }

  return best_sad;
}

/* do_refine: If last step (1-away) of n-step search doesn't pick the center
              point as the best match, we will do a final 1-away diamond
              refining search  */
int vp10_full_pixel_diamond(const VP10_COMP *cpi, MACROBLOCK *x,
                           MV *mvp_full, int step_param,
                           int sadpb, int further_steps, int do_refine,
                           int *cost_list,
                           const vp9_variance_fn_ptr_t *fn_ptr,
                           const MV *ref_mv, MV *dst_mv) {
  MV temp_mv;
  int thissme, n, num00 = 0;
  int bestsme = cpi->diamond_search_sad(x, &cpi->ss_cfg, mvp_full, &temp_mv,
                                        step_param, sadpb, &n,
                                        fn_ptr, ref_mv);
  if (bestsme < INT_MAX)
    bestsme = vp10_get_mvpred_var(x, &temp_mv, ref_mv, fn_ptr, 1);
  *dst_mv = temp_mv;

  // If there won't be more n-step search, check to see if refining search is
  // needed.
  if (n > further_steps)
    do_refine = 0;

  while (n < further_steps) {
    ++n;

    if (num00) {
      num00--;
    } else {
      thissme = cpi->diamond_search_sad(x, &cpi->ss_cfg, mvp_full, &temp_mv,
                                        step_param + n, sadpb, &num00,
                                        fn_ptr, ref_mv);
      if (thissme < INT_MAX)
        thissme = vp10_get_mvpred_var(x, &temp_mv, ref_mv, fn_ptr, 1);

      // check to see if refining search is needed.
      if (num00 > further_steps - n)
        do_refine = 0;

      if (thissme < bestsme) {
        bestsme = thissme;
        *dst_mv = temp_mv;
      }
    }
  }

  // final 1-away diamond refining search
  if (do_refine) {
    const int search_range = 8;
    MV best_mv = *dst_mv;
    thissme = vp10_refining_search_sad(x, &best_mv, sadpb, search_range,
                                       fn_ptr, ref_mv);
    if (thissme < INT_MAX)
      thissme = vp10_get_mvpred_var(x, &best_mv, ref_mv, fn_ptr, 1);
    if (thissme < bestsme) {
      bestsme = thissme;
      *dst_mv = best_mv;
    }
  }

  // Return cost list.
  if (cost_list) {
    calc_int_cost_list(x, ref_mv, sadpb, fn_ptr, dst_mv, cost_list);
  }
  return bestsme;
}

int vp10_full_search_sad_c(const MACROBLOCK *x, const MV *ref_mv,
                          int sad_per_bit, int distance,
                          const vp9_variance_fn_ptr_t *fn_ptr,
                          const MV *center_mv, MV *best_mv) {
  int r, c;
  const MACROBLOCKD *const xd = &x->e_mbd;
  const struct buf_2d *const what = &x->plane[0].src;
  const struct buf_2d *const in_what = &xd->plane[0].pre[0];
  const int row_min = MAX(ref_mv->row - distance, x->mv_row_min);
  const int row_max = MIN(ref_mv->row + distance, x->mv_row_max);
  const int col_min = MAX(ref_mv->col - distance, x->mv_col_min);
  const int col_max = MIN(ref_mv->col + distance, x->mv_col_max);
  const MV fcenter_mv = {center_mv->row >> 3, center_mv->col >> 3};
  int best_sad = fn_ptr->sdf(what->buf, what->stride,
      get_buf_from_mv(in_what, ref_mv), in_what->stride) +
      mvsad_err_cost(x, ref_mv, &fcenter_mv, sad_per_bit);
  *best_mv = *ref_mv;

  for (r = row_min; r < row_max; ++r) {
    for (c = col_min; c < col_max; ++c) {
      const MV mv = {r, c};
      const int sad = fn_ptr->sdf(what->buf, what->stride,
          get_buf_from_mv(in_what, &mv), in_what->stride) +
              mvsad_err_cost(x, &mv, &fcenter_mv, sad_per_bit);
      if (sad < best_sad) {
        best_sad = sad;
        *best_mv = mv;
      }
    }
  }
  return best_sad;
}

int vp10_full_search_sadx3(const MACROBLOCK *x, const MV *ref_mv,
                          int sad_per_bit, int distance,
                          const vp9_variance_fn_ptr_t *fn_ptr,
                          const MV *center_mv, MV *best_mv) {
  int r;
  const MACROBLOCKD *const xd = &x->e_mbd;
  const struct buf_2d *const what = &x->plane[0].src;
  const struct buf_2d *const in_what = &xd->plane[0].pre[0];
  const int row_min = MAX(ref_mv->row - distance, x->mv_row_min);
  const int row_max = MIN(ref_mv->row + distance, x->mv_row_max);
  const int col_min = MAX(ref_mv->col - distance, x->mv_col_min);
  const int col_max = MIN(ref_mv->col + distance, x->mv_col_max);
  const MV fcenter_mv = {center_mv->row >> 3, center_mv->col >> 3};
  unsigned int best_sad = fn_ptr->sdf(what->buf, what->stride,
      get_buf_from_mv(in_what, ref_mv), in_what->stride) +
      mvsad_err_cost(x, ref_mv, &fcenter_mv, sad_per_bit);
  *best_mv = *ref_mv;

  for (r = row_min; r < row_max; ++r) {
    int c = col_min;
    const uint8_t *check_here = &in_what->buf[r * in_what->stride + c];

    if (fn_ptr->sdx3f != NULL) {
      while ((c + 2) < col_max) {
        int i;
        DECLARE_ALIGNED(16, uint32_t, sads[3]);

        fn_ptr->sdx3f(what->buf, what->stride, check_here, in_what->stride,
                      sads);

        for (i = 0; i < 3; ++i) {
          unsigned int sad = sads[i];
          if (sad < best_sad) {
            const MV mv = {r, c};
            sad += mvsad_err_cost(x, &mv, &fcenter_mv, sad_per_bit);
            if (sad < best_sad) {
              best_sad = sad;
              *best_mv = mv;
            }
          }
          ++check_here;
          ++c;
        }
      }
    }

    while (c < col_max) {
      unsigned int sad = fn_ptr->sdf(what->buf, what->stride,
                                     check_here, in_what->stride);
      if (sad < best_sad) {
        const MV mv = {r, c};
        sad += mvsad_err_cost(x, &mv, &fcenter_mv, sad_per_bit);
        if (sad < best_sad) {
          best_sad = sad;
          *best_mv = mv;
        }
      }
      ++check_here;
      ++c;
    }
  }

  return best_sad;
}

int vp10_full_search_sadx8(const MACROBLOCK *x, const MV *ref_mv,
                          int sad_per_bit, int distance,
                          const vp9_variance_fn_ptr_t *fn_ptr,
                          const MV *center_mv, MV *best_mv) {
  int r;
  const MACROBLOCKD *const xd = &x->e_mbd;
  const struct buf_2d *const what = &x->plane[0].src;
  const struct buf_2d *const in_what = &xd->plane[0].pre[0];
  const int row_min = MAX(ref_mv->row - distance, x->mv_row_min);
  const int row_max = MIN(ref_mv->row + distance, x->mv_row_max);
  const int col_min = MAX(ref_mv->col - distance, x->mv_col_min);
  const int col_max = MIN(ref_mv->col + distance, x->mv_col_max);
  const MV fcenter_mv = {center_mv->row >> 3, center_mv->col >> 3};
  unsigned int best_sad = fn_ptr->sdf(what->buf, what->stride,
      get_buf_from_mv(in_what, ref_mv), in_what->stride) +
      mvsad_err_cost(x, ref_mv, &fcenter_mv, sad_per_bit);
  *best_mv = *ref_mv;

  for (r = row_min; r < row_max; ++r) {
    int c = col_min;
    const uint8_t *check_here = &in_what->buf[r * in_what->stride + c];

    if (fn_ptr->sdx8f != NULL) {
      while ((c + 7) < col_max) {
        int i;
        DECLARE_ALIGNED(16, uint32_t, sads[8]);

        fn_ptr->sdx8f(what->buf, what->stride, check_here, in_what->stride,
                      sads);

        for (i = 0; i < 8; ++i) {
          unsigned int sad = sads[i];
          if (sad < best_sad) {
            const MV mv = {r, c};
            sad += mvsad_err_cost(x, &mv, &fcenter_mv, sad_per_bit);
            if (sad < best_sad) {
              best_sad = sad;
              *best_mv = mv;
            }
          }
          ++check_here;
          ++c;
        }
      }
    }

    if (fn_ptr->sdx3f != NULL) {
      while ((c + 2) < col_max) {
        int i;
        DECLARE_ALIGNED(16, uint32_t, sads[3]);

        fn_ptr->sdx3f(what->buf, what->stride, check_here, in_what->stride,
                      sads);

        for (i = 0; i < 3; ++i) {
          unsigned int sad = sads[i];
          if (sad < best_sad) {
            const MV mv = {r, c};
            sad += mvsad_err_cost(x, &mv, &fcenter_mv, sad_per_bit);
            if (sad < best_sad) {
              best_sad = sad;
              *best_mv = mv;
            }
          }
          ++check_here;
          ++c;
        }
      }
    }

    while (c < col_max) {
      unsigned int sad = fn_ptr->sdf(what->buf, what->stride,
                                     check_here, in_what->stride);
      if (sad < best_sad) {
        const MV mv = {r, c};
        sad += mvsad_err_cost(x, &mv, &fcenter_mv, sad_per_bit);
        if (sad < best_sad) {
          best_sad = sad;
          *best_mv = mv;
        }
      }
      ++check_here;
      ++c;
    }
  }

  return best_sad;
}

int vp10_refining_search_sad(const MACROBLOCK *x,
                            MV *ref_mv, int error_per_bit,
                            int search_range,
                            const vp9_variance_fn_ptr_t *fn_ptr,
                            const MV *center_mv) {
  const MACROBLOCKD *const xd = &x->e_mbd;
  const MV neighbors[4] = {{ -1, 0}, {0, -1}, {0, 1}, {1, 0}};
  const struct buf_2d *const what = &x->plane[0].src;
  const struct buf_2d *const in_what = &xd->plane[0].pre[0];
  const MV fcenter_mv = {center_mv->row >> 3, center_mv->col >> 3};
  const uint8_t *best_address = get_buf_from_mv(in_what, ref_mv);
  unsigned int best_sad = fn_ptr->sdf(what->buf, what->stride, best_address,
                                    in_what->stride) +
      mvsad_err_cost(x, ref_mv, &fcenter_mv, error_per_bit);
  int i, j;

  for (i = 0; i < search_range; i++) {
    int best_site = -1;
    const int all_in = ((ref_mv->row - 1) > x->mv_row_min) &
                       ((ref_mv->row + 1) < x->mv_row_max) &
                       ((ref_mv->col - 1) > x->mv_col_min) &
                       ((ref_mv->col + 1) < x->mv_col_max);

    if (all_in) {
      unsigned int sads[4];
      const uint8_t *const positions[4] = {
        best_address - in_what->stride,
        best_address - 1,
        best_address + 1,
        best_address + in_what->stride
      };

      fn_ptr->sdx4df(what->buf, what->stride, positions, in_what->stride, sads);

      for (j = 0; j < 4; ++j) {
        if (sads[j] < best_sad) {
          const MV mv = {ref_mv->row + neighbors[j].row,
                         ref_mv->col + neighbors[j].col};
          sads[j] += mvsad_err_cost(x, &mv, &fcenter_mv, error_per_bit);
          if (sads[j] < best_sad) {
            best_sad = sads[j];
            best_site = j;
          }
        }
      }
    } else {
      for (j = 0; j < 4; ++j) {
        const MV mv = {ref_mv->row + neighbors[j].row,
                       ref_mv->col + neighbors[j].col};

        if (is_mv_in(x, &mv)) {
          unsigned int sad = fn_ptr->sdf(what->buf, what->stride,
                                         get_buf_from_mv(in_what, &mv),
                                         in_what->stride);
          if (sad < best_sad) {
            sad += mvsad_err_cost(x, &mv, &fcenter_mv, error_per_bit);
            if (sad < best_sad) {
              best_sad = sad;
              best_site = j;
            }
          }
        }
      }
    }

    if (best_site == -1) {
      break;
    } else {
      ref_mv->row += neighbors[best_site].row;
      ref_mv->col += neighbors[best_site].col;
      best_address = get_buf_from_mv(in_what, ref_mv);
    }
  }

  return best_sad;
}

// This function is called when we do joint motion search in comp_inter_inter
// mode.
int vp10_refining_search_8p_c(const MACROBLOCK *x,
                             MV *ref_mv, int error_per_bit,
                             int search_range,
                             const vp9_variance_fn_ptr_t *fn_ptr,
                             const MV *center_mv,
                             const uint8_t *second_pred) {
  const MV neighbors[8] = {{-1, 0}, {0, -1}, {0, 1}, {1, 0},
                           {-1, -1}, {1, -1}, {-1, 1}, {1, 1}};
  const MACROBLOCKD *const xd = &x->e_mbd;
  const struct buf_2d *const what = &x->plane[0].src;
  const struct buf_2d *const in_what = &xd->plane[0].pre[0];
  const MV fcenter_mv = {center_mv->row >> 3, center_mv->col >> 3};
  unsigned int best_sad = fn_ptr->sdaf(what->buf, what->stride,
      get_buf_from_mv(in_what, ref_mv), in_what->stride, second_pred) +
      mvsad_err_cost(x, ref_mv, &fcenter_mv, error_per_bit);
  int i, j;

  for (i = 0; i < search_range; ++i) {
    int best_site = -1;

    for (j = 0; j < 8; ++j) {
      const MV mv = {ref_mv->row + neighbors[j].row,
                     ref_mv->col + neighbors[j].col};

      if (is_mv_in(x, &mv)) {
        unsigned int sad = fn_ptr->sdaf(what->buf, what->stride,
            get_buf_from_mv(in_what, &mv), in_what->stride, second_pred);
        if (sad < best_sad) {
          sad += mvsad_err_cost(x, &mv, &fcenter_mv, error_per_bit);
          if (sad < best_sad) {
            best_sad = sad;
            best_site = j;
          }
        }
      }
    }

    if (best_site == -1) {
      break;
    } else {
      ref_mv->row += neighbors[best_site].row;
      ref_mv->col += neighbors[best_site].col;
    }
  }
  return best_sad;
}

int vp10_full_pixel_search(VP10_COMP *cpi, MACROBLOCK *x,
                          BLOCK_SIZE bsize, MV *mvp_full,
                          int step_param, int error_per_bit,
                          int *cost_list,
                          const MV *ref_mv, MV *tmp_mv,
                          int var_max, int rd) {
  const SPEED_FEATURES *const sf = &cpi->sf;
  const SEARCH_METHODS method = sf->mv.search_method;
  vp9_variance_fn_ptr_t *fn_ptr = &cpi->fn_ptr[bsize];
  int var = 0;
  if (cost_list) {
    cost_list[0] = INT_MAX;
    cost_list[1] = INT_MAX;
    cost_list[2] = INT_MAX;
    cost_list[3] = INT_MAX;
    cost_list[4] = INT_MAX;
  }

  switch (method) {
    case FAST_DIAMOND:
      var = vp10_fast_dia_search(x, mvp_full, step_param, error_per_bit, 0,
                                cost_list, fn_ptr, 1, ref_mv, tmp_mv);
      break;
    case FAST_HEX:
      var = vp10_fast_hex_search(x, mvp_full, step_param, error_per_bit, 0,
                                cost_list, fn_ptr, 1, ref_mv, tmp_mv);
      break;
    case HEX:
      var = vp10_hex_search(x, mvp_full, step_param, error_per_bit, 1,
                           cost_list, fn_ptr, 1, ref_mv, tmp_mv);
      break;
    case SQUARE:
      var = vp10_square_search(x, mvp_full, step_param, error_per_bit, 1,
                              cost_list, fn_ptr, 1, ref_mv, tmp_mv);
      break;
    case BIGDIA:
      var = vp10_bigdia_search(x, mvp_full, step_param, error_per_bit, 1,
                              cost_list, fn_ptr, 1, ref_mv, tmp_mv);
      break;
    case NSTEP:
      var = vp10_full_pixel_diamond(cpi, x, mvp_full, step_param, error_per_bit,
                                   MAX_MVSEARCH_STEPS - 1 - step_param,
                                   1, cost_list, fn_ptr, ref_mv, tmp_mv);
      break;
    default:
      assert(0 && "Invalid search method.");
  }

  if (method != NSTEP && rd && var < var_max)
    var = vp10_get_mvpred_var(x, tmp_mv, ref_mv, fn_ptr, 1);

  return var;
}