aboutsummaryrefslogtreecommitdiffstats
path: root/libavcodec/aarch64/vp9itxfm_16bpp_neon.S
blob: f53e94ac0c485e6542675e08aff649f5308d7727 (plain) (blame)
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
/*
 * Copyright (c) 2017 Google Inc.
 *
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * FFmpeg is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with FFmpeg; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

#include "libavutil/aarch64/asm.S"
#include "neon.S"

const itxfm4_coeffs, align=4
        .short  11585, 0, 6270, 15137
iadst4_coeffs:
        .short  5283, 15212, 9929, 13377
endconst

const iadst8_coeffs, align=4
        .short  16305, 1606, 14449, 7723, 10394, 12665, 4756, 15679
idct_coeffs:
        .short  11585, 0, 6270, 15137, 3196, 16069, 13623, 9102
        .short  1606, 16305, 12665, 10394, 7723, 14449, 15679, 4756
        .short  804, 16364, 12140, 11003, 7005, 14811, 15426, 5520
        .short  3981, 15893, 14053, 8423, 9760, 13160, 16207, 2404
endconst

const iadst16_coeffs, align=4
        .short  16364, 804, 15893, 3981, 11003, 12140, 8423, 14053
        .short  14811, 7005, 13160, 9760, 5520, 15426, 2404, 16207
endconst

.macro transpose_4x4s r0, r1, r2, r3, r4, r5, r6, r7
        trn1            \r4\().4s,  \r0\().4s,  \r1\().4s
        trn2            \r5\().4s,  \r0\().4s,  \r1\().4s
        trn1            \r6\().4s,  \r2\().4s,  \r3\().4s
        trn2            \r7\().4s,  \r2\().4s,  \r3\().4s
        trn1            \r0\().2d,  \r4\().2d,  \r6\().2d
        trn2            \r2\().2d,  \r4\().2d,  \r6\().2d
        trn1            \r1\().2d,  \r5\().2d,  \r7\().2d
        trn2            \r3\().2d,  \r5\().2d,  \r7\().2d
.endm

// Transpose a 8x8 matrix of 32 bit elements, where each row is spread out
// over two registers.
.macro transpose_8x8s r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11, r12, r13, r14, r15, t0, t1, t2, t3
        transpose_4x4s  \r0,  \r2,  \r4,  \r6,  \t0, \t1, \t2, \t3
        transpose_4x4s  \r9,  \r11, \r13, \r15, \t0, \t1, \t2, \t3

        // Do 4x4 transposes of r1,r3,r5,r7 and r8,r10,r12,r14
        // while swapping the two 4x4 matrices between each other

        // First step of the 4x4 transpose of r1-r7, into t0-t3
        trn1            \t0\().4s,  \r1\().4s,  \r3\().4s
        trn2            \t1\().4s,  \r1\().4s,  \r3\().4s
        trn1            \t2\().4s,  \r5\().4s,  \r7\().4s
        trn2            \t3\().4s,  \r5\().4s,  \r7\().4s

        // First step of the 4x4 transpose of r8-r12, into r1-r7
        trn1            \r1\().4s,  \r8\().4s,  \r10\().4s
        trn2            \r3\().4s,  \r8\().4s,  \r10\().4s
        trn1            \r5\().4s,  \r12\().4s, \r14\().4s
        trn2            \r7\().4s,  \r12\().4s, \r14\().4s

        // Second step of the 4x4 transpose of r1-r7 (now in t0-r3), into r8-r12
        trn1            \r8\().2d,  \t0\().2d,  \t2\().2d
        trn2            \r12\().2d, \t0\().2d,  \t2\().2d
        trn1            \r10\().2d, \t1\().2d,  \t3\().2d
        trn2            \r14\().2d, \t1\().2d,  \t3\().2d

        // Second step of the 4x4 transpose of r8-r12 (now in r1-r7), in place as far as possible
        trn1            \t0\().2d,  \r1\().2d,  \r5\().2d
        trn2            \r5\().2d,  \r1\().2d,  \r5\().2d
        trn1            \t1\().2d,  \r3\().2d,  \r7\().2d
        trn2            \r7\().2d,  \r3\().2d,  \r7\().2d

        // Move the outputs of trn1 back in place
        mov             \r1\().16b,  \t0\().16b
        mov             \r3\().16b,  \t1\().16b
.endm

// out1 = ((in1 + in2) * d0[0] + (1 << 13)) >> 14
// out2 = ((in1 - in2) * d0[0] + (1 << 13)) >> 14
// in/out are .4s registers; this can do with 4 temp registers, but is
// more efficient if 6 temp registers are available.
.macro dmbutterfly0 out1, out2, in1, in2, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, neg=0
.if \neg > 0
        neg             \tmp4\().4s, v0.4s
.endif
        add             \tmp1\().4s, \in1\().4s,  \in2\().4s
        sub             \tmp2\().4s, \in1\().4s,  \in2\().4s
.if \neg > 0
        smull           \tmp3\().2d, \tmp1\().2s, \tmp4\().s[0]
        smull2          \tmp4\().2d, \tmp1\().4s, \tmp4\().s[0]
.else
        smull           \tmp3\().2d, \tmp1\().2s, v0.s[0]
        smull2          \tmp4\().2d, \tmp1\().4s, v0.s[0]
.endif
.ifb \tmp5
        rshrn           \out1\().2s, \tmp3\().2d, #14
        rshrn2          \out1\().4s, \tmp4\().2d, #14
        smull           \tmp3\().2d, \tmp2\().2s, v0.s[0]
        smull2          \tmp4\().2d, \tmp2\().4s, v0.s[0]
        rshrn           \out2\().2s, \tmp3\().2d, #14
        rshrn2          \out2\().4s, \tmp4\().2d, #14
.else
        smull           \tmp5\().2d, \tmp2\().2s, v0.s[0]
        smull2          \tmp6\().2d, \tmp2\().4s, v0.s[0]
        rshrn           \out1\().2s, \tmp3\().2d, #14
        rshrn2          \out1\().4s, \tmp4\().2d, #14
        rshrn           \out2\().2s, \tmp5\().2d, #14
        rshrn2          \out2\().4s, \tmp6\().2d, #14
.endif
.endm

// out1,out2 = in1 * coef1 - in2 * coef2
// out3,out4 = in1 * coef2 + in2 * coef1
// out are 4 x .2d registers, in are 2 x .4s registers
.macro dmbutterfly_l out1, out2, out3, out4, in1, in2, coef1, coef2
        smull           \out1\().2d, \in1\().2s, \coef1
        smull2          \out2\().2d, \in1\().4s, \coef1
        smull           \out3\().2d, \in1\().2s, \coef2
        smull2          \out4\().2d, \in1\().4s, \coef2
        smlsl           \out1\().2d, \in2\().2s, \coef2
        smlsl2          \out2\().2d, \in2\().4s, \coef2
        smlal           \out3\().2d, \in2\().2s, \coef1
        smlal2          \out4\().2d, \in2\().4s, \coef1
.endm

// inout1 = (inout1 * coef1 - inout2 * coef2 + (1 << 13)) >> 14
// inout2 = (inout1 * coef2 + inout2 * coef1 + (1 << 13)) >> 14
// inout are 2 x .4s registers
.macro dmbutterfly inout1, inout2, coef1, coef2, tmp1, tmp2, tmp3, tmp4, neg=0
        dmbutterfly_l   \tmp1, \tmp2, \tmp3, \tmp4, \inout1, \inout2, \coef1, \coef2
.if \neg > 0
        neg             \tmp3\().2d, \tmp3\().2d
        neg             \tmp4\().2d, \tmp4\().2d
.endif
        rshrn           \inout1\().2s, \tmp1\().2d,  #14
        rshrn2          \inout1\().4s, \tmp2\().2d,  #14
        rshrn           \inout2\().2s, \tmp3\().2d,  #14
        rshrn2          \inout2\().4s, \tmp4\().2d,  #14
.endm

// out1 = in1 + in2
// out2 = in1 - in2
.macro butterfly_4s out1, out2, in1, in2
        add             \out1\().4s, \in1\().4s, \in2\().4s
        sub             \out2\().4s, \in1\().4s, \in2\().4s
.endm

// out1 = in1 - in2
// out2 = in1 + in2
.macro butterfly_4s_r out1, out2, in1, in2
        sub             \out1\().4s, \in1\().4s, \in2\().4s
        add             \out2\().4s, \in1\().4s, \in2\().4s
.endm

// out1 = (in1,in2 + in3,in4 + (1 << 13)) >> 14
// out2 = (in1,in2 - in3,in4 + (1 << 13)) >> 14
// out are 2 x .4s registers, in are 4 x .2d registers
.macro dbutterfly_n out1, out2, in1, in2, in3, in4, tmp1, tmp2, tmp3, tmp4
        add             \tmp1\().2d, \in1\().2d, \in3\().2d
        add             \tmp2\().2d, \in2\().2d, \in4\().2d
        sub             \tmp3\().2d, \in1\().2d, \in3\().2d
        sub             \tmp4\().2d, \in2\().2d, \in4\().2d
        rshrn           \out1\().2s, \tmp1\().2d,  #14
        rshrn2          \out1\().4s, \tmp2\().2d,  #14
        rshrn           \out2\().2s, \tmp3\().2d,  #14
        rshrn2          \out2\().4s, \tmp4\().2d,  #14
.endm

.macro iwht4_10 c0, c1, c2, c3
        add             \c0\().4s, \c0\().4s, \c1\().4s
        sub             v17.4s,    \c2\().4s, \c3\().4s
        sub             v16.4s,    \c0\().4s, v17.4s
        sshr            v16.4s,    v16.4s,    #1
        sub             \c2\().4s, v16.4s,    \c1\().4s
        sub             \c1\().4s, v16.4s,    \c3\().4s
        add             \c3\().4s, v17.4s,    \c2\().4s
        sub             \c0\().4s, \c0\().4s, \c1\().4s
.endm

.macro iwht4_12 c0, c1, c2, c3
        iwht4_10        \c0, \c1, \c2, \c3
.endm

.macro idct4_10 c0, c1, c2, c3
        mul             v22.4s,    \c1\().4s, v0.s[3]
        mul             v20.4s,    \c1\().4s, v0.s[2]
        add             v16.4s,    \c0\().4s, \c2\().4s
        sub             v17.4s,    \c0\().4s, \c2\().4s
        mla             v22.4s,    \c3\().4s, v0.s[2]
        mul             v18.4s,    v16.4s,    v0.s[0]
        mul             v24.4s,    v17.4s,    v0.s[0]
        mls             v20.4s,    \c3\().4s, v0.s[3]
        srshr           v22.4s,    v22.4s,    #14
        srshr           v18.4s,    v18.4s,    #14
        srshr           v24.4s,    v24.4s,    #14
        srshr           v20.4s,    v20.4s,    #14
        add             \c0\().4s, v18.4s,    v22.4s
        sub             \c3\().4s, v18.4s,    v22.4s
        add             \c1\().4s, v24.4s,    v20.4s
        sub             \c2\().4s, v24.4s,    v20.4s
.endm

.macro idct4_12 c0, c1, c2, c3
        smull           v22.2d,    \c1\().2s, v0.s[3]
        smull2          v23.2d,    \c1\().4s, v0.s[3]
        smull           v20.2d,    \c1\().2s, v0.s[2]
        smull2          v21.2d,    \c1\().4s, v0.s[2]
        add             v16.4s,    \c0\().4s, \c2\().4s
        sub             v17.4s,    \c0\().4s, \c2\().4s
        smlal           v22.2d,    \c3\().2s, v0.s[2]
        smlal2          v23.2d,    \c3\().4s, v0.s[2]
        smull           v18.2d,    v16.2s,    v0.s[0]
        smull2          v19.2d,    v16.4s,    v0.s[0]
        smull           v24.2d,    v17.2s,    v0.s[0]
        smull2          v25.2d,    v17.4s,    v0.s[0]
        smlsl           v20.2d,    \c3\().2s, v0.s[3]
        smlsl2          v21.2d,    \c3\().4s, v0.s[3]
        rshrn           v22.2s,    v22.2d,    #14
        rshrn2          v22.4s,    v23.2d,    #14
        rshrn           v18.2s,    v18.2d,    #14
        rshrn2          v18.4s,    v19.2d,    #14
        rshrn           v24.2s,    v24.2d,    #14
        rshrn2          v24.4s,    v25.2d,    #14
        rshrn           v20.2s,    v20.2d,    #14
        rshrn2          v20.4s,    v21.2d,    #14
        add             \c0\().4s, v18.4s,    v22.4s
        sub             \c3\().4s, v18.4s,    v22.4s
        add             \c1\().4s, v24.4s,    v20.4s
        sub             \c2\().4s, v24.4s,    v20.4s
.endm

.macro iadst4_10 c0, c1, c2, c3
        mul             v16.4s,    \c0\().4s, v1.s[0]
        mla             v16.4s,    \c2\().4s, v1.s[1]
        mla             v16.4s,    \c3\().4s, v1.s[2]
        mul             v18.4s,    \c0\().4s, v1.s[2]
        mls             v18.4s,    \c2\().4s, v1.s[0]
        sub             \c0\().4s, \c0\().4s, \c2\().4s
        mls             v18.4s,    \c3\().4s, v1.s[1]
        add             \c0\().4s, \c0\().4s, \c3\().4s
        mul             v22.4s,    \c1\().4s, v1.s[3]
        mul             v20.4s,    \c0\().4s, v1.s[3]
        add             v24.4s,    v16.4s,    v22.4s
        add             v26.4s,    v18.4s,    v22.4s
        srshr           \c0\().4s, v24.4s,    #14
        add             v16.4s,    v16.4s,    v18.4s
        srshr           \c1\().4s, v26.4s,    #14
        sub             v16.4s,    v16.4s,    v22.4s
        srshr           \c2\().4s, v20.4s,    #14
        srshr           \c3\().4s, v16.4s,    #14
.endm

.macro iadst4_12 c0, c1, c2, c3
        smull           v16.2d,    \c0\().2s, v1.s[0]
        smull2          v17.2d,    \c0\().4s, v1.s[0]
        smlal           v16.2d,    \c2\().2s, v1.s[1]
        smlal2          v17.2d,    \c2\().4s, v1.s[1]
        smlal           v16.2d,    \c3\().2s, v1.s[2]
        smlal2          v17.2d,    \c3\().4s, v1.s[2]
        smull           v18.2d,    \c0\().2s, v1.s[2]
        smull2          v19.2d,    \c0\().4s, v1.s[2]
        smlsl           v18.2d,    \c2\().2s, v1.s[0]
        smlsl2          v19.2d,    \c2\().4s, v1.s[0]
        sub             \c0\().4s, \c0\().4s, \c2\().4s
        smlsl           v18.2d,    \c3\().2s, v1.s[1]
        smlsl2          v19.2d,    \c3\().4s, v1.s[1]
        add             \c0\().4s, \c0\().4s, \c3\().4s
        smull           v22.2d,    \c1\().2s, v1.s[3]
        smull2          v23.2d,    \c1\().4s, v1.s[3]
        smull           v20.2d,    \c0\().2s, v1.s[3]
        smull2          v21.2d,    \c0\().4s, v1.s[3]
        add             v24.2d,    v16.2d,    v22.2d
        add             v25.2d,    v17.2d,    v23.2d
        add             v26.2d,    v18.2d,    v22.2d
        add             v27.2d,    v19.2d,    v23.2d
        rshrn           \c0\().2s, v24.2d,    #14
        rshrn2          \c0\().4s, v25.2d,    #14
        add             v16.2d,    v16.2d,    v18.2d
        add             v17.2d,    v17.2d,    v19.2d
        rshrn           \c1\().2s, v26.2d,    #14
        rshrn2          \c1\().4s, v27.2d,    #14
        sub             v16.2d,    v16.2d,    v22.2d
        sub             v17.2d,    v17.2d,    v23.2d
        rshrn           \c2\().2s, v20.2d,    #14
        rshrn2          \c2\().4s, v21.2d,    #14
        rshrn           \c3\().2s, v16.2d,    #14
        rshrn2          \c3\().4s, v17.2d,    #14
.endm

// The public functions in this file have got the following signature:
// void itxfm_add(uint8_t *dst, ptrdiff_t stride, int16_t *block, int eob);

.macro itxfm_func4x4 txfm1, txfm2, bpp
function ff_vp9_\txfm1\()_\txfm2\()_4x4_add_\bpp\()_neon, export=1
.ifc \txfm1,\txfm2
.ifc \txfm1,idct
        movrel          x4,  itxfm4_coeffs
        ld1             {v0.4h}, [x4]
        sxtl            v0.4s,  v0.4h
.endif
.ifc \txfm1,iadst
        movrel          x4,  iadst4_coeffs
        ld1             {v0.d}[1], [x4]
        sxtl2           v1.4s,  v0.8h
.endif
.else
        movrel          x4,  itxfm4_coeffs
        ld1             {v0.8h}, [x4]
        sxtl2           v1.4s,  v0.8h
        sxtl            v0.4s,  v0.4h
.endif

        movi            v30.4s, #0
        movi            v31.4s, #0
.ifc \txfm1\()_\txfm2,idct_idct
        cmp             w3,  #1
        b.ne            1f
        // DC-only for idct/idct
        ld1             {v2.s}[0],  [x2]
        smull           v2.2d,  v2.2s, v0.s[0]
        rshrn           v2.2s,  v2.2d, #14
        smull           v2.2d,  v2.2s, v0.s[0]
        rshrn           v2.2s,  v2.2d, #14
        st1             {v31.s}[0], [x2]
        dup             v4.4s,  v2.s[0]
        mov             v5.16b, v4.16b
        mov             v6.16b, v4.16b
        mov             v7.16b, v4.16b
        b               2f
.endif

1:
        ld1             {v4.4s,v5.4s,v6.4s,v7.4s},  [x2]
        st1             {v30.4s,v31.4s}, [x2], #32

.ifc \txfm1,iwht
        sshr            v4.4s,  v4.4s,  #2
        sshr            v5.4s,  v5.4s,  #2
        sshr            v6.4s,  v6.4s,  #2
        sshr            v7.4s,  v7.4s,  #2
.endif

        \txfm1\()4_\bpp v4,  v5,  v6,  v7

        st1             {v30.4s,v31.4s}, [x2], #32
        // Transpose 4x4 with 32 bit elements
        transpose_4x4s  v4,  v5,  v6,  v7,  v16, v17, v18, v19

        \txfm2\()4_\bpp v4,  v5,  v6,  v7
2:
        mvni            v31.8h, #((0xff << (\bpp - 8)) & 0xff), lsl #8
        ld1             {v0.4h},   [x0], x1
        ld1             {v1.4h},   [x0], x1
.ifnc \txfm1,iwht
        srshr           v4.4s,  v4.4s,  #4
        srshr           v5.4s,  v5.4s,  #4
        srshr           v6.4s,  v6.4s,  #4
        srshr           v7.4s,  v7.4s,  #4
.endif
        uaddw           v4.4s,  v4.4s,  v0.4h
        uaddw           v5.4s,  v5.4s,  v1.4h
        ld1             {v2.4h},   [x0], x1
        ld1             {v3.4h},   [x0], x1
        sqxtun          v0.4h,  v4.4s
        sqxtun2         v0.8h,  v5.4s
        sub             x0,  x0,  x1, lsl #2

        uaddw           v6.4s,  v6.4s,  v2.4h
        umin            v0.8h,  v0.8h,  v31.8h
        uaddw           v7.4s,  v7.4s,  v3.4h
        st1             {v0.4h},   [x0], x1
        sqxtun          v2.4h,  v6.4s
        sqxtun2         v2.8h,  v7.4s
        umin            v2.8h,  v2.8h,  v31.8h

        st1             {v0.d}[1], [x0], x1
        st1             {v2.4h},   [x0], x1
        st1             {v2.d}[1], [x0], x1

        ret
endfunc
.endm

.macro itxfm_funcs4x4 bpp
itxfm_func4x4 idct,  idct,  \bpp
itxfm_func4x4 iadst, idct,  \bpp
itxfm_func4x4 idct,  iadst, \bpp
itxfm_func4x4 iadst, iadst, \bpp
itxfm_func4x4 iwht,  iwht,  \bpp
.endm

itxfm_funcs4x4 10
itxfm_funcs4x4 12

function idct8x8_dc_add_neon
        movrel          x4,  idct_coeffs
        ld1             {v0.4h}, [x4]

        movi            v1.4h,  #0
        sxtl            v0.4s,  v0.4h

        ld1             {v2.s}[0],  [x2]
        smull           v2.2d,  v2.2s,  v0.s[0]
        rshrn           v2.2s,  v2.2d,  #14
        smull           v2.2d,  v2.2s,  v0.s[0]
        rshrn           v2.2s,  v2.2d,  #14
        st1             {v1.s}[0],  [x2]
        dup             v2.4s,  v2.s[0]

        srshr           v2.4s,  v2.4s,  #5

        mov             x4,  #8
        mov             x3,  x0
        dup             v31.8h, w5
1:
        // Loop to add the constant from v2 into all 8x8 outputs
        subs            x4,  x4,  #2
        ld1             {v3.8h},  [x0], x1
        ld1             {v4.8h},  [x0], x1
        uaddw           v16.4s, v2.4s,  v3.4h
        uaddw2          v17.4s, v2.4s,  v3.8h
        uaddw           v18.4s, v2.4s,  v4.4h
        uaddw2          v19.4s, v2.4s,  v4.8h
        sqxtun          v3.4h,  v16.4s
        sqxtun2         v3.8h,  v17.4s
        sqxtun          v4.4h,  v18.4s
        sqxtun2         v4.8h,  v19.4s
        umin            v3.8h,  v3.8h,  v31.8h
        umin            v4.8h,  v4.8h,  v31.8h
        st1             {v3.8h},  [x3], x1
        st1             {v4.8h},  [x3], x1
        b.ne            1b

        ret
endfunc

.macro idct8 r0, r1, r2, r3, r4, r5, r6, r7, t0, t1, t2, t3, t4, t5
        dmbutterfly0    \r0, \r4, \r0, \r4, \t0, \t1, \t2, \t3, \t4, \t5 // r0 = t0a, r4 = t1a
        dmbutterfly     \r2, \r6, v0.s[2], v0.s[3], \t0, \t1, \t2, \t3   // r2 = t2a, r6 = t3a
        dmbutterfly     \r1, \r7, v1.s[0], v1.s[1], \t0, \t1, \t2, \t3   // r1 = t4a, r7 = t7a
        dmbutterfly     \r5, \r3, v1.s[2], v1.s[3], \t0, \t1, \t2, \t3   // r5 = t5a, r3 = t6a

        butterfly_4s    \t0, \t1, \r0, \r6 // t0 = t0, t1 = t3
        butterfly_4s    \t2, \r5, \r1, \r5 // t2 = t4, r5 = t5a
        butterfly_4s    \t3, \r6, \r7, \r3 // t3 = t7, r6 = t6a
        butterfly_4s    \r7, \r4, \r4, \r2 // r7 = t1, r4 = t2

        dmbutterfly0    \r6, \r5, \r6, \r5, \r0, \r1, \r2, \r3, \t4, \t5 // r6 = t6, r5 = t5

        butterfly_4s    \r1, \r6, \r7, \r6 // r1 = out[1], r6 = out[6]
        butterfly_4s    \r0, \r7, \t0, \t3 // r0 = out[0], r7 = out[7]
        butterfly_4s    \r2, \r5, \r4, \r5 // r2 = out[2], r5 = out[5]
        butterfly_4s    \r3, \r4, \t1, \t2 // r3 = out[3], r4 = out[4]
.endm

.macro iadst8 r0, r1, r2, r3, r4, r5, r6, r7, t0, t1, t2, t3, t4, t5
        dmbutterfly_l   \t2, \t3, \t0, \t1, \r7, \r0, v2.s[1], v2.s[0]   // t2,t3 = t1a, t0,t1 = t0a
        dmbutterfly_l   \r0, \r7, \t4, \t5, \r3, \r4, v3.s[1], v3.s[0]   // r0,r7 = t5a, t4,t5 = t4a

        dbutterfly_n    \r3, \t0, \t0, \t1, \t4, \t5, \r3, \r4, \t0, \t1 // r3 = t0, t0 = t4
        dbutterfly_n    \r4, \t1, \t2, \t3, \r0, \r7, \r4, \t1, \t4, \t5 // r4 = t1, t1 = t5

        dmbutterfly_l   \t4, \t5, \t2, \t3, \r5, \r2, v2.s[3], v2.s[2]   // t4,t5 = t3a, t2,t3 = t2a
        dmbutterfly_l   \r2, \r5, \r0, \r7, \r1, \r6, v3.s[3], v3.s[2]   // r2,r5 = t7a, r0,r7 = t6a

        dbutterfly_n    \r1, \t2, \t2, \t3, \r0, \r7, \r1, \r6, \t2, \t3 // r1 = t2, t2 = t6
        dbutterfly_n    \r0, \t4, \t4, \t5, \r2, \r5, \r0, \r7, \t4, \t5 // r0 = t3, t4 = t7

        butterfly_4s    \r7, \r4, \r4, \r0   // r7 = -out[7], r4 = t3
        neg             \r7\().4s, \r7\().4s // r7 = out[7]
        butterfly_4s    \r0, \r1, \r3, \r1   // r0 = out[0],  r1 = t2

        dmbutterfly_l   \r2, \r3, \t3, \t5, \t0, \t1, v0.s[2], v0.s[3]   // r2,r3 = t5a, t3,t5 = t4a
        dmbutterfly_l   \t0, \t1, \r5, \r6, \t4, \t2, v0.s[3], v0.s[2]   // t0,t1 = t6a, r5,r6 = t7a

        dbutterfly_n    \r6, \t2, \r2, \r3, \r5, \r6, \t2, \t4, \r2, \r3 // r6 = out[6],  t2 = t7

        dmbutterfly0    \r3, \r4, \r1, \r4, \t4, \r5, \r1, \r2           // r3 = -out[3], r4 = out[4]
        neg             \r3\().4s, \r3\().4s  // r3 = out[3]

        dbutterfly_n    \r1, \t0, \t3, \t5, \t0, \t1, \r1, \r2, \t0, \t1 // r1 = -out[1], t0 = t6
        neg             \r1\().4s, \r1\().4s  // r1 = out[1]

        dmbutterfly0    \r2, \r5, \t0, \t2, \t1, \t3, \t4, \t5           // r2 = out[2],  r5 = -out[5]
        neg             \r5\().4s, \r5\().4s  // r5 = out[5]
.endm


.macro itxfm_func8x8 txfm1, txfm2
function vp9_\txfm1\()_\txfm2\()_8x8_add_16_neon
.ifc \txfm1\()_\txfm2,idct_idct
        cmp             w3,  #1
        b.eq            idct8x8_dc_add_neon
.endif
        // The iadst also uses a few coefficients from
        // idct, so those always need to be loaded.
.ifc \txfm1\()_\txfm2,idct_idct
        movrel          x4,  idct_coeffs
.else
        movrel          x4,  iadst8_coeffs
        ld1             {v1.8h}, [x4], #16
        stp             d8,  d9,  [sp, #-0x10]!
        sxtl2           v3.4s,  v1.8h
        sxtl            v2.4s,  v1.4h
.endif
        ld1             {v0.8h}, [x4]
        sxtl2           v1.4s,  v0.8h
        sxtl            v0.4s,  v0.4h

        movi            v4.4s, #0
        movi            v5.4s, #0
        movi            v6.4s, #0
        movi            v7.4s, #0

1:
        ld1             {v16.4s,v17.4s,v18.4s,v19.4s},  [x2], #64
        ld1             {v20.4s,v21.4s,v22.4s,v23.4s},  [x2], #64
        ld1             {v24.4s,v25.4s,v26.4s,v27.4s},  [x2], #64
        ld1             {v28.4s,v29.4s,v30.4s,v31.4s},  [x2], #64
        sub             x2,  x2,  #256
        st1             {v4.4s,v5.4s,v6.4s,v7.4s},      [x2], #64
        st1             {v4.4s,v5.4s,v6.4s,v7.4s},      [x2], #64
        st1             {v4.4s,v5.4s,v6.4s,v7.4s},      [x2], #64
        st1             {v4.4s,v5.4s,v6.4s,v7.4s},      [x2], #64

.ifc \txfm1\()_\txfm2,idct_idct
        idct8           v16, v18, v20, v22, v24, v26, v28, v30, v2,  v3,  v4,  v5,  v6,  v7
        idct8           v17, v19, v21, v23, v25, v27, v29, v31, v2,  v3,  v4,  v5,  v6,  v7
.else
        \txfm1\()8      v16, v18, v20, v22, v24, v26, v28, v30, v4,  v5,  v6,  v7,  v8,  v9
        \txfm1\()8      v17, v19, v21, v23, v25, v27, v29, v31, v4,  v5,  v6,  v7,  v8,  v9
.endif

        // Transpose 8x8 with 16 bit elements
        transpose_8x8s  v16, v17, v18, v19, v20, v21, v22, v23, v24, v25, v26, v27, v28, v29, v30, v31, v4, v5, v6, v7

.ifc \txfm1\()_\txfm2,idct_idct
        idct8           v16, v18, v20, v22, v24, v26, v28, v30, v2,  v3,  v4,  v5,  v6,  v7
        idct8           v17, v19, v21, v23, v25, v27, v29, v31, v2,  v3,  v4,  v5,  v6,  v7
.else
        \txfm2\()8      v16, v18, v20, v22, v24, v26, v28, v30, v4,  v5,  v6,  v7,  v8,  v9
        \txfm2\()8      v17, v19, v21, v23, v25, v27, v29, v31, v4,  v5,  v6,  v7,  v8,  v9
.endif
2:
        mov             x3,  x0
        // Add into the destination
        ld1             {v0.8h},  [x0], x1
        srshr           v16.4s, v16.4s, #5
        srshr           v17.4s, v17.4s, #5
        ld1             {v1.8h},  [x0], x1
        srshr           v18.4s, v18.4s, #5
        srshr           v19.4s, v19.4s, #5
        ld1             {v2.8h},  [x0], x1
        srshr           v20.4s, v20.4s, #5
        srshr           v21.4s, v21.4s, #5
        uaddw           v16.4s, v16.4s, v0.4h
        uaddw2          v17.4s, v17.4s, v0.8h
        ld1             {v3.8h},  [x0], x1
        srshr           v22.4s, v22.4s, #5
        srshr           v23.4s, v23.4s, #5
        uaddw           v18.4s, v18.4s, v1.4h
        uaddw2          v19.4s, v19.4s, v1.8h
        ld1             {v4.8h},  [x0], x1
        srshr           v24.4s, v24.4s, #5
        srshr           v25.4s, v25.4s, #5
        uaddw           v20.4s, v20.4s, v2.4h
        uaddw2          v21.4s, v21.4s, v2.8h
        sqxtun          v0.4h,  v16.4s
        sqxtun2         v0.8h,  v17.4s
        dup             v16.8h, w5
        ld1             {v5.8h},  [x0], x1
        srshr           v26.4s, v26.4s, #5
        srshr           v27.4s, v27.4s, #5
        uaddw           v22.4s, v22.4s, v3.4h
        uaddw2          v23.4s, v23.4s, v3.8h
        sqxtun          v1.4h,  v18.4s
        sqxtun2         v1.8h,  v19.4s
        umin            v0.8h,  v0.8h,  v16.8h
        ld1             {v6.8h},  [x0], x1
        srshr           v28.4s, v28.4s, #5
        srshr           v29.4s, v29.4s, #5
        uaddw           v24.4s, v24.4s, v4.4h
        uaddw2          v25.4s, v25.4s, v4.8h
        sqxtun          v2.4h,  v20.4s
        sqxtun2         v2.8h,  v21.4s
        umin            v1.8h,  v1.8h,  v16.8h
        ld1             {v7.8h},  [x0], x1
        srshr           v30.4s, v30.4s, #5
        srshr           v31.4s, v31.4s, #5
        uaddw           v26.4s, v26.4s, v5.4h
        uaddw2          v27.4s, v27.4s, v5.8h
        sqxtun          v3.4h,  v22.4s
        sqxtun2         v3.8h,  v23.4s
        umin            v2.8h,  v2.8h,  v16.8h

        st1             {v0.8h},  [x3], x1
        uaddw           v28.4s, v28.4s, v6.4h
        uaddw2          v29.4s, v29.4s, v6.8h
        st1             {v1.8h},  [x3], x1
        sqxtun          v4.4h,  v24.4s
        sqxtun2         v4.8h,  v25.4s
        umin            v3.8h,  v3.8h,  v16.8h
        st1             {v2.8h},  [x3], x1
        uaddw           v30.4s, v30.4s, v7.4h
        uaddw2          v31.4s, v31.4s, v7.8h
        st1             {v3.8h},  [x3], x1
        sqxtun          v5.4h,  v26.4s
        sqxtun2         v5.8h,  v27.4s
        umin            v4.8h,  v4.8h,  v16.8h
        st1             {v4.8h},  [x3], x1
        sqxtun          v6.4h,  v28.4s
        sqxtun2         v6.8h,  v29.4s
        umin            v5.8h,  v5.8h,  v16.8h
        st1             {v5.8h},  [x3], x1
        sqxtun          v7.4h,  v30.4s
        sqxtun2         v7.8h,  v31.4s
        umin            v6.8h,  v6.8h,  v16.8h

        st1             {v6.8h},  [x3], x1
        umin            v7.8h,  v7.8h,  v16.8h
        st1             {v7.8h},  [x3], x1

.ifnc \txfm1\()_\txfm2,idct_idct
        ldp             d8,  d9,  [sp], 0x10
.endif
        ret
endfunc

function ff_vp9_\txfm1\()_\txfm2\()_8x8_add_10_neon, export=1
        mov             x5,  #0x03ff
        b               vp9_\txfm1\()_\txfm2\()_8x8_add_16_neon
endfunc

function ff_vp9_\txfm1\()_\txfm2\()_8x8_add_12_neon, export=1
        mov             x5,  #0x0fff
        b               vp9_\txfm1\()_\txfm2\()_8x8_add_16_neon
endfunc
.endm

itxfm_func8x8 idct,  idct
itxfm_func8x8 iadst, idct
itxfm_func8x8 idct,  iadst
itxfm_func8x8 iadst, iadst


function idct16x16_dc_add_neon
        movrel          x4,  idct_coeffs
        ld1             {v0.4h}, [x4]
        sxtl            v0.4s,  v0.4h

        movi            v1.4h,  #0

        ld1             {v2.s}[0],  [x2]
        smull           v2.2d,  v2.2s,  v0.s[0]
        rshrn           v2.2s,  v2.2d,  #14
        smull           v2.2d,  v2.2s,  v0.s[0]
        rshrn           v2.2s,  v2.2d,  #14
        st1             {v1.s}[0],  [x2]
        dup             v2.4s,  v2.s[0]

        srshr           v0.4s,  v2.4s,  #6

        mov             x3, x0
        mov             x4, #16
        dup             v31.8h, w13
1:
        // Loop to add the constant from v2 into all 16x16 outputs
        subs            x4,  x4,  #2
        ld1             {v1.8h,v2.8h},  [x0], x1
        uaddw           v16.4s, v0.4s,  v1.4h
        uaddw2          v17.4s, v0.4s,  v1.8h
        ld1             {v3.8h,v4.8h},  [x0], x1
        uaddw           v18.4s, v0.4s,  v2.4h
        uaddw2          v19.4s, v0.4s,  v2.8h
        uaddw           v20.4s, v0.4s,  v3.4h
        uaddw2          v21.4s, v0.4s,  v3.8h
        uaddw           v22.4s, v0.4s,  v4.4h
        uaddw2          v23.4s, v0.4s,  v4.8h
        sqxtun          v1.4h,  v16.4s
        sqxtun2         v1.8h,  v17.4s
        sqxtun          v2.4h,  v18.4s
        sqxtun2         v2.8h,  v19.4s
        sqxtun          v3.4h,  v20.4s
        sqxtun2         v3.8h,  v21.4s
        sqxtun          v4.4h,  v22.4s
        sqxtun2         v4.8h,  v23.4s
        umin            v1.8h,  v1.8h,  v31.8h
        umin            v2.8h,  v2.8h,  v31.8h
        st1             {v1.8h,v2.8h},  [x3], x1
        umin            v3.8h,  v3.8h,  v31.8h
        umin            v4.8h,  v4.8h,  v31.8h
        st1             {v3.8h,v4.8h},  [x3], x1
        b.ne            1b

        ret
endfunc

.macro idct16
        dmbutterfly0    v16, v24, v16, v24, v4, v5, v6, v7, v8, v9 // v16 = t0a,  v24 = t1a
        dmbutterfly     v20, v28, v0.s[2], v0.s[3], v4, v5, v6, v7 // v20 = t2a,  v28 = t3a
        dmbutterfly     v18, v30, v1.s[0], v1.s[1], v4, v5, v6, v7 // v18 = t4a,  v30 = t7a
        dmbutterfly     v26, v22, v1.s[2], v1.s[3], v4, v5, v6, v7 // v26 = t5a,  v22 = t6a
        dmbutterfly     v17, v31, v2.s[0], v2.s[1], v4, v5, v6, v7 // v17 = t8a,  v31 = t15a
        dmbutterfly     v25, v23, v2.s[2], v2.s[3], v4, v5, v6, v7 // v25 = t9a,  v23 = t14a
        dmbutterfly     v21, v27, v3.s[0], v3.s[1], v4, v5, v6, v7 // v21 = t10a, v27 = t13a
        dmbutterfly     v29, v19, v3.s[2], v3.s[3], v4, v5, v6, v7 // v29 = t11a, v19 = t12a

        butterfly_4s    v4,  v28, v16, v28               // v4  = t0,   v28 = t3
        butterfly_4s    v5,  v20, v24, v20               // v5  = t1,   v20 = t2
        butterfly_4s    v6,  v26, v18, v26               // v6  = t4,   v26 = t5
        butterfly_4s    v7,  v22, v30, v22               // v7  = t7,   v22 = t6
        butterfly_4s    v16, v25, v17, v25               // v16 = t8,   v25 = t9
        butterfly_4s    v24, v21, v29, v21               // v24 = t11,  v21 = t10
        butterfly_4s    v17, v27, v19, v27               // v17 = t12,  v27 = t13
        butterfly_4s    v29, v23, v31, v23               // v29 = t15,  v23 = t14

        dmbutterfly0    v22, v26, v22, v26, v8, v9, v18, v19, v30, v31        // v22 = t6a,  v26 = t5a
        dmbutterfly     v23, v25, v0.s[2], v0.s[3], v18, v19, v30, v31        // v23 = t9a,  v25 = t14a
        dmbutterfly     v27, v21, v0.s[2], v0.s[3], v18, v19, v30, v31, neg=1 // v27 = t13a, v21 = t10a

        butterfly_4s    v18, v7,  v4,  v7                // v18 = t0a,  v7  = t7a
        butterfly_4s    v19, v22, v5,  v22               // v19 = t1a,  v22 = t6
        butterfly_4s    v4,  v26, v20, v26               // v4  = t2a,  v26 = t5
        butterfly_4s    v5,  v6,  v28, v6                // v5  = t3a,  v6  = t4
        butterfly_4s    v20, v28, v16, v24               // v20 = t8a,  v28 = t11a
        butterfly_4s    v24, v21, v23, v21               // v24 = t9,   v21 = t10
        butterfly_4s    v23, v27, v25, v27               // v23 = t14,  v27 = t13
        butterfly_4s    v25, v29, v29, v17               // v25 = t15a, v29 = t12a

        dmbutterfly0    v8,  v9,  v27, v21, v8,  v9,  v16, v17, v30, v31 // v8  = t13a, v9  = t10a
        dmbutterfly0    v28, v27, v29, v28, v21, v29, v16, v17, v30, v31 // v28 = t12,  v27 = t11

        butterfly_4s    v16, v31, v18, v25               // v16 = out[0], v31 = out[15]
        butterfly_4s    v17, v30, v19, v23               // v17 = out[1], v30 = out[14]
        butterfly_4s_r  v25, v22, v22, v24               // v25 = out[9], v22 = out[6]
        butterfly_4s    v23, v24, v7,  v20               // v23 = out[7], v24 = out[8]
        butterfly_4s    v18, v29, v4,  v8                // v18 = out[2], v29 = out[13]
        butterfly_4s    v19, v28, v5,  v28               // v19 = out[3], v28 = out[12]
        butterfly_4s    v20, v27, v6,  v27               // v20 = out[4], v27 = out[11]
        butterfly_4s    v21, v26, v26, v9                // v21 = out[5], v26 = out[10]
.endm

.macro iadst16
        ld1             {v0.8h,v1.8h}, [x11]
        sxtl            v2.4s,  v1.4h
        sxtl2           v3.4s,  v1.8h
        sxtl2           v1.4s,  v0.8h
        sxtl            v0.4s,  v0.4h

        dmbutterfly_l   v6,  v7,  v4,  v5,  v31, v16, v0.s[1], v0.s[0]   // v6,v7   = t1,   v4,v5   = t0
        dmbutterfly_l   v10, v11, v8,  v9,  v23, v24, v1.s[1], v1.s[0]   // v10,v11 = t9,   v8,v9   = t8
        dbutterfly_n    v31, v24, v6,  v7,  v10, v11, v12, v13, v10, v11 // v31     = t1a,  v24     = t9a
        dmbutterfly_l   v14, v15, v12, v13, v29, v18, v0.s[3], v0.s[2]   // v14,v15 = t3,   v12,v13 = t2
        dbutterfly_n    v16, v23, v4,  v5,  v8,  v9,  v6,  v7,  v8,  v9  // v16     = t0a,  v23     = t8a

        dmbutterfly_l   v6,  v7,  v4,  v5,  v21, v26, v1.s[3], v1.s[2]   // v6,v7   = t11,  v4,v5   = t10
        dbutterfly_n    v29, v26, v14, v15, v6,  v7,  v8,  v9,  v6,  v7  // v29     = t3a,  v26     = t11a
        dmbutterfly_l   v10, v11, v8,  v9,  v27, v20, v2.s[1], v2.s[0]   // v10,v11 = t5,   v8,v9   = t4
        dbutterfly_n    v18, v21, v12, v13, v4,  v5,  v6,  v7,  v4,  v5  // v18     = t2a,  v21     = t10a

        dmbutterfly_l   v14, v15, v12, v13, v19, v28, v3.s[1], v3.s[0]   // v14,v15 = t13,  v12,v13 = t12
        dbutterfly_n    v20, v28, v10, v11, v14, v15, v4,  v5,  v14, v15 // v20     = t5a,  v28     = t13a
        dmbutterfly_l   v6,  v7,  v4,  v5,  v25, v22, v2.s[3], v2.s[2]   // v6,v7   = t7,   v4,v5   = t6
        dbutterfly_n    v27, v19, v8,  v9,  v12, v13, v10, v11, v12, v13 // v27     = t4a,  v19     = t12a

        dmbutterfly_l   v10, v11, v8,  v9,  v17, v30, v3.s[3], v3.s[2]   // v10,v11 = t15,  v8,v9   = t14
        ld1             {v0.8h}, [x10]
        dbutterfly_n    v22, v30, v6,  v7,  v10, v11, v12, v13, v10, v11 // v22     = t7a,  v30     = t15a
        sxtl2           v1.4s,  v0.8h
        sxtl            v0.4s,  v0.4h
        dmbutterfly_l   v14, v15, v12, v13, v23, v24, v1.s[0], v1.s[1]   // v14,v15 = t9,   v12,v13 = t8
        dbutterfly_n    v25, v17, v4,  v5,  v8,  v9,  v6,  v7,  v8,  v9  // v25     = t6a,  v17     = t14a

        dmbutterfly_l   v4,  v5,  v6,  v7,  v28, v19, v1.s[1], v1.s[0]   // v4,v5   = t12,  v6,v7   = t13
        dbutterfly_n    v23, v19, v12, v13, v4,  v5,  v8,  v9,  v4,  v5  // v23     = t8a,  v19     = t12a
        dmbutterfly_l   v10, v11, v8,  v9,  v21, v26, v1.s[2], v1.s[3]   // v10,v11 = t11,  v8,v9   = t10
        butterfly_4s_r  v4,  v27, v16, v27               // v4  = t4,   v27 = t0
        dbutterfly_n    v24, v28, v14, v15, v6,  v7,  v12, v13, v6,  v7  // v24     = t9a,  v28     = t13a

        dmbutterfly_l   v12, v13, v14, v15, v30, v17, v1.s[3], v1.s[2]   // v12,v13 = t14,  v14,v15 = t15
        butterfly_4s_r  v5,  v20, v31, v20               // v5  = t5, v20 = t1
        dbutterfly_n    v21, v17, v8,  v9,  v12, v13, v6,  v7,  v12, v13 // v21     = t10a, v17     = t14a
        dbutterfly_n    v26, v30, v10, v11, v14, v15, v8,  v9,  v14, v15 // v26     = t11a, v30     = t15a

        butterfly_4s_r  v6,  v25, v18, v25               // v6  = t6, v25 = t2
        butterfly_4s_r  v7,  v22, v29, v22               // v7  = t7, v22 = t3

        dmbutterfly_l   v10, v11, v8,  v9,  v19, v28, v0.s[2], v0.s[3]   // v10,v11 = t13,  v8,v9   = t12
        dmbutterfly_l   v12, v13, v14, v15, v30, v17, v0.s[3], v0.s[2]   // v12,v13 = t14,  v14,v15 = t15

        dbutterfly_n    v18, v30, v8,  v9,  v12, v13, v16, v17, v12, v13 // v18   = out[2], v30     = t14a
        dbutterfly_n    v29, v17, v10, v11, v14, v15, v12, v13, v14, v15 // v29 = -out[13], v17     = t15a
        neg             v29.4s, v29.4s                   // v29 = out[13]

        dmbutterfly_l   v10, v11, v8,  v9,  v4,  v5,  v0.s[2], v0.s[3]   // v10,v11 = t5a,  v8,v9   = t4a
        dmbutterfly_l   v12, v13, v14, v15, v7,  v6,  v0.s[3], v0.s[2]   // v12,v13 = t6a,  v14,v15 = t7a

        butterfly_4s    v2,  v6,  v27, v25               // v2 = out[0], v6 = t2a
        butterfly_4s    v3,  v7,  v23, v21               // v3 =-out[1], v7 = t10

        dbutterfly_n    v19, v31, v8,  v9,  v12, v13, v4,  v5,  v8,  v9  // v19 = -out[3],  v31 = t6
        neg             v19.4s, v19.4s                   // v19 = out[3]
        dbutterfly_n    v28, v16, v10, v11, v14, v15, v4,  v5,  v10, v11 // v28 = out[12],  v16 = t7

        butterfly_4s    v5,  v8,  v20, v22               // v5 =-out[15],v8 = t3a
        butterfly_4s    v4,  v9,  v24, v26               // v4 = out[14],v9 = t11

        dmbutterfly0    v23, v24, v6,  v8,  v10, v11, v12, v13, v14, v15, 1 // v23 = out[7], v24 = out[8]
        dmbutterfly0    v21, v26, v30, v17, v10, v11, v12, v13, v14, v15, 1 // v21 = out[5], v26 = out[10]
        dmbutterfly0    v20, v27, v16, v31, v10, v11, v12, v13, v14, v15    // v20 = out[4], v27 = out[11]
        dmbutterfly0    v22, v25, v9,  v7,  v10, v11, v12, v13, v14, v15    // v22 = out[6], v25 = out[9]

        neg             v31.4s,  v5.4s                    // v31 = out[15]
        neg             v17.4s,  v3.4s                    // v17 = out[1]

        mov             v16.16b, v2.16b
        mov             v30.16b, v4.16b
.endm

// Helper macros; we can't use these expressions directly within
// e.g. .irp due to the extra concatenation \(). Therefore wrap
// them in macros to allow using .irp below.
.macro load i, src, inc
        ld1             {v\i\().4s},  [\src], \inc
.endm
.macro store i, dst, inc
        st1             {v\i\().4s},  [\dst], \inc
.endm
.macro movi_v i, size, imm
        movi            v\i\()\size,  \imm
.endm
.macro load_clear i, src, inc
        ld1             {v\i\().4s}, [\src]
        st1             {v4.4s},  [\src], \inc
.endm

// Read a vertical 4x16 slice out of a 16x16 matrix, do a transform on it,
// transpose into a horizontal 16x4 slice and store.
// x0 = dst (temp buffer)
// x1 = slice offset
// x2 = src
// x9 = input stride
.macro itxfm16_1d_funcs txfm
function \txfm\()16_1d_4x16_pass1_neon
        movi            v4.4s, #0
.irp i, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31
        load_clear      \i,  x2,  x9
.endr

        \txfm\()16

        // Do four 4x4 transposes. Originally, v16-v31 contain the
        // 16 rows. Afterwards, v16-v19, v20-v23, v24-v27 and v28-v31
        // contain the four transposed 4x4 blocks.
        transpose_4x4s  v16, v17, v18, v19, v4, v5, v6, v7
        transpose_4x4s  v20, v21, v22, v23, v4, v5, v6, v7
        transpose_4x4s  v24, v25, v26, v27, v4, v5, v6, v7
        transpose_4x4s  v28, v29, v30, v31, v4, v5, v6, v7

        // Store the transposed 8x8 blocks horizontally.
        cmp             x1,  #12
        b.eq            1f
.irp i, 16, 20, 24, 28, 17, 21, 25, 29, 18, 22, 26, 30, 19, 23, 27, 31
        store           \i,  x0,  #16
.endr
        ret
1:
        // Special case: For the last input column (x1 == 12),
        // which would be stored as the last row in the temp buffer,
        // don't store the first 4x4 block, but keep it in registers
        // for the first slice of the second pass (where it is the
        // last 4x4 block).
        add             x0,  x0,  #16
.irp i, 20, 24, 28
        store           \i,  x0,  #16
.endr
        add             x0,  x0,  #16
.irp i, 21, 25, 29
        store           \i,  x0,  #16
.endr
        add             x0,  x0,  #16
.irp i, 22, 26, 30
        store           \i,  x0,  #16
.endr
        add             x0,  x0,  #16
.irp i, 23, 27, 31
        store           \i,  x0,  #16
.endr

        mov             v28.16b, v16.16b
        mov             v29.16b, v17.16b
        mov             v30.16b, v18.16b
        mov             v31.16b, v19.16b
        ret
endfunc

// Read a vertical 4x16 slice out of a 16x16 matrix, do a transform on it,
// load the destination pixels (from a similar 4x16 slice), add and store back.
// x0 = dst
// x1 = dst stride
// x2 = src (temp buffer)
// x3 = slice offset
// x9 = temp buffer stride
function \txfm\()16_1d_4x16_pass2_neon
.irp i, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27
        load            \i,  x2,  x9
.endr
        cbz             x3,  1f
.irp i, 28, 29, 30, 31
        load            \i,  x2,  x9
.endr
1:

        add             x3,  x0,  x1
        lsl             x1,  x1,  #1
        \txfm\()16

        dup             v8.8h, w13
.macro load_add_store coef0, coef1, coef2, coef3, coef4, coef5, coef6, coef7
        srshr           \coef0, \coef0, #6
        ld1             {v4.4h},   [x0], x1
        srshr           \coef1, \coef1, #6
        ld1             {v4.d}[1], [x3], x1
        srshr           \coef2, \coef2, #6
        ld1             {v5.4h},   [x0], x1
        srshr           \coef3, \coef3, #6
        uaddw           \coef0, \coef0, v4.4h
        ld1             {v5.d}[1], [x3], x1
        srshr           \coef4, \coef4, #6
        uaddw2          \coef1, \coef1, v4.8h
        ld1             {v6.4h},   [x0], x1
        srshr           \coef5, \coef5, #6
        uaddw           \coef2, \coef2, v5.4h
        ld1             {v6.d}[1], [x3], x1
        sqxtun          v4.4h,  \coef0
        srshr           \coef6, \coef6, #6
        uaddw2          \coef3, \coef3, v5.8h
        ld1             {v7.4h},   [x0], x1
        sqxtun2         v4.8h,  \coef1
        srshr           \coef7, \coef7, #6
        uaddw           \coef4, \coef4, v6.4h
        ld1             {v7.d}[1], [x3], x1
        umin            v4.8h,  v4.8h,  v8.8h
        sub             x0,  x0,  x1, lsl #2
        sub             x3,  x3,  x1, lsl #2
        sqxtun          v5.4h,  \coef2
        uaddw2          \coef5, \coef5, v6.8h
        st1             {v4.4h},   [x0], x1
        sqxtun2         v5.8h,  \coef3
        uaddw           \coef6, \coef6, v7.4h
        st1             {v4.d}[1], [x3], x1
        umin            v5.8h,  v5.8h,  v8.8h
        sqxtun          v6.4h,  \coef4
        uaddw2          \coef7, \coef7, v7.8h
        st1             {v5.4h},   [x0], x1
        sqxtun2         v6.8h,  \coef5
        st1             {v5.d}[1], [x3], x1
        umin            v6.8h,  v6.8h,  v8.8h
        sqxtun          v7.4h,  \coef6
        st1             {v6.4h},   [x0], x1
        sqxtun2         v7.8h,  \coef7
        st1             {v6.d}[1], [x3], x1
        umin            v7.8h,  v7.8h,  v8.8h
        st1             {v7.4h},   [x0], x1
        st1             {v7.d}[1], [x3], x1
.endm
        load_add_store  v16.4s, v17.4s, v18.4s, v19.4s, v20.4s, v21.4s, v22.4s, v23.4s
        load_add_store  v24.4s, v25.4s, v26.4s, v27.4s, v28.4s, v29.4s, v30.4s, v31.4s
.purgem load_add_store

        ret
endfunc
.endm

itxfm16_1d_funcs idct
itxfm16_1d_funcs iadst

// This is the minimum eob value for each subpartition, in increments of 4
const min_eob_idct_idct_16, align=4
        .short  0, 10, 38, 89
endconst

.macro itxfm_func16x16 txfm1, txfm2
function vp9_\txfm1\()_\txfm2\()_16x16_add_16_neon
.ifc \txfm1\()_\txfm2,idct_idct
        cmp             w3,  #1
        b.eq            idct16x16_dc_add_neon
.endif
        mov             x15, x30
        // iadst16 requires clobbering v8-v15, idct16 only clobbers v8-v9.
.ifnc \txfm1\()_\txfm2,idct_idct
        stp             d14, d15, [sp, #-0x10]!
        stp             d12, d13, [sp, #-0x10]!
        stp             d10, d11, [sp, #-0x10]!
.endif
        stp             d8,  d9,  [sp, #-0x10]!

        sub             sp,  sp,  #1024

        mov             x4,  x0
        mov             x5,  x1
        mov             x6,  x2

        movrel          x10, idct_coeffs
.ifnc \txfm1\()_\txfm2,idct_idct
        movrel          x11, iadst16_coeffs
.endif
        movrel          x12, min_eob_idct_idct_16, 2
.ifc \txfm1,idct
        ld1             {v0.8h,v1.8h}, [x10]
        sxtl            v2.4s,  v1.4h
        sxtl2           v3.4s,  v1.8h
        sxtl2           v1.4s,  v0.8h
        sxtl            v0.4s,  v0.4h
.endif
        mov             x9,  #64

.irp i, 0, 4, 8, 12
        add             x0,  sp,  #(\i*64)
.ifc \txfm1\()_\txfm2,idct_idct
.if \i > 0
        ldrh            w1,  [x12], #2
        cmp             w3,  w1
        mov             x1,  #(16 - \i)/4
        b.le            1f
.endif
.endif
        mov             x1,  #\i
        add             x2,  x6,  #(\i*4)
        bl              \txfm1\()16_1d_4x16_pass1_neon
.endr
.ifc \txfm1\()_\txfm2,iadst_idct
        ld1             {v0.8h,v1.8h}, [x10]
        sxtl            v2.4s,  v1.4h
        sxtl2           v3.4s,  v1.8h
        sxtl2           v1.4s,  v0.8h
        sxtl            v0.4s,  v0.4h
.endif

.ifc \txfm1\()_\txfm2,idct_idct
        b               3f
1:
        // Set v28-v31 to zero, for the in-register passthrough of
        // coefficients to pass 2.
        movi            v28.4s,  #0
        movi            v29.4s,  #0
        movi            v30.4s,  #0
        movi            v31.4s,  #0
2:
        subs            x1,  x1,  #1
.rept 4
        st1             {v28.4s,v29.4s,v30.4s,v31.4s}, [x0], x9
.endr
        b.ne            2b
3:
.endif

.irp i, 0, 4, 8, 12
        add             x0,  x4,  #(\i*2)
        mov             x1,  x5
        add             x2,  sp,  #(\i*4)
        mov             x3,  #\i
        bl              \txfm2\()16_1d_4x16_pass2_neon
.endr

        add             sp,  sp,  #1024
        ldp             d8,  d9,  [sp], 0x10
.ifnc \txfm1\()_\txfm2,idct_idct
        ldp             d10, d11, [sp], 0x10
        ldp             d12, d13, [sp], 0x10
        ldp             d14, d15, [sp], 0x10
.endif
        br              x15
endfunc

function ff_vp9_\txfm1\()_\txfm2\()_16x16_add_10_neon, export=1
        mov             x13, #0x03ff
        b               vp9_\txfm1\()_\txfm2\()_16x16_add_16_neon
endfunc

function ff_vp9_\txfm1\()_\txfm2\()_16x16_add_12_neon, export=1
        mov             x13, #0x0fff
        b               vp9_\txfm1\()_\txfm2\()_16x16_add_16_neon
endfunc
.endm

itxfm_func16x16 idct,  idct
itxfm_func16x16 iadst, idct
itxfm_func16x16 idct,  iadst
itxfm_func16x16 iadst, iadst


function idct32x32_dc_add_neon
        movrel          x4,  idct_coeffs
        ld1             {v0.4h}, [x4]
        sxtl            v0.4s,  v0.4h

        movi            v1.4h,  #0

        ld1             {v2.s}[0],  [x2]
        smull           v2.2d,  v2.2s,  v0.s[0]
        rshrn           v2.2s,  v2.2d,  #14
        smull           v2.2d,  v2.2s,  v0.s[0]
        rshrn           v2.2s,  v2.2d,  #14
        st1             {v1.s}[0],  [x2]
        dup             v2.4s,  v2.s[0]

        srshr           v0.4s,  v2.4s,  #6

        mov             x3,  x0
        mov             x4,  #32
        sub             x1,  x1,  #32
        dup             v31.8h, w13
1:
        // Loop to add the constant v0 into all 32x32 outputs
        subs            x4,  x4,  #1
        ld1             {v1.8h,v2.8h},  [x0], #32
        uaddw           v16.4s, v0.4s,  v1.4h
        uaddw2          v17.4s, v0.4s,  v1.8h
        ld1             {v3.8h,v4.8h},  [x0], x1
        uaddw           v18.4s, v0.4s,  v2.4h
        uaddw2          v19.4s, v0.4s,  v2.8h
        uaddw           v20.4s, v0.4s,  v3.4h
        uaddw2          v21.4s, v0.4s,  v3.8h
        uaddw           v22.4s, v0.4s,  v4.4h
        uaddw2          v23.4s, v0.4s,  v4.8h
        sqxtun          v1.4h,  v16.4s
        sqxtun2         v1.8h,  v17.4s
        sqxtun          v2.4h,  v18.4s
        sqxtun2         v2.8h,  v19.4s
        sqxtun          v3.4h,  v20.4s
        sqxtun2         v3.8h,  v21.4s
        sqxtun          v4.4h,  v22.4s
        sqxtun2         v4.8h,  v23.4s
        umin            v1.8h,  v1.8h,  v31.8h
        umin            v2.8h,  v2.8h,  v31.8h
        st1             {v1.8h,v2.8h},  [x3], #32
        umin            v3.8h,  v3.8h,  v31.8h
        umin            v4.8h,  v4.8h,  v31.8h
        st1             {v3.8h,v4.8h},  [x3], x1
        b.ne            1b

        ret
endfunc

.macro idct32_odd
        dmbutterfly     v16, v31, v10.s[0], v10.s[1], v4, v5, v6, v7 // v16 = t16a, v31 = t31a
        dmbutterfly     v24, v23, v10.s[2], v10.s[3], v4, v5, v6, v7 // v24 = t17a, v23 = t30a
        dmbutterfly     v20, v27, v11.s[0], v11.s[1], v4, v5, v6, v7 // v20 = t18a, v27 = t29a
        dmbutterfly     v28, v19, v11.s[2], v11.s[3], v4, v5, v6, v7 // v28 = t19a, v19 = t28a
        dmbutterfly     v18, v29, v12.s[0], v12.s[1], v4, v5, v6, v7 // v18 = t20a, v29 = t27a
        dmbutterfly     v26, v21, v12.s[2], v12.s[3], v4, v5, v6, v7 // v26 = t21a, v21 = t26a
        dmbutterfly     v22, v25, v13.s[0], v13.s[1], v4, v5, v6, v7 // v22 = t22a, v25 = t25a
        dmbutterfly     v30, v17, v13.s[2], v13.s[3], v4, v5, v6, v7 // v30 = t23a, v17 = t24a

        butterfly_4s    v4,  v24, v16, v24 // v4  = t16, v24 = t17
        butterfly_4s    v5,  v20, v28, v20 // v5  = t19, v20 = t18
        butterfly_4s    v6,  v26, v18, v26 // v6  = t20, v26 = t21
        butterfly_4s    v7,  v22, v30, v22 // v7  = t23, v22 = t22
        butterfly_4s    v28, v25, v17, v25 // v28 = t24, v25 = t25
        butterfly_4s    v30, v21, v29, v21 // v30 = t27, v21 = t26
        butterfly_4s    v29, v23, v31, v23 // v29 = t31, v23 = t30
        butterfly_4s    v31, v27, v19, v27 // v31 = t28, v27 = t29

        dmbutterfly     v23, v24, v1.s[0], v1.s[1], v16, v17, v18, v19        // v23 = t17a, v24 = t30a
        dmbutterfly     v27, v20, v1.s[0], v1.s[1], v16, v17, v18, v19, neg=1 // v27 = t29a, v20 = t18a
        dmbutterfly     v21, v26, v1.s[2], v1.s[3], v16, v17, v18, v19        // v21 = t21a, v26 = t26a
        dmbutterfly     v25, v22, v1.s[2], v1.s[3], v16, v17, v18, v19, neg=1 // v25 = t25a, v22 = t22a

        butterfly_4s    v16, v5,  v4,  v5  // v16 = t16a, v5  = t19a
        butterfly_4s    v17, v20, v23, v20 // v17 = t17,  v20 = t18
        butterfly_4s    v18, v6,  v7,  v6  // v18 = t23a, v6  = t20a
        butterfly_4s    v19, v21, v22, v21 // v19 = t22,  v21 = t21
        butterfly_4s    v4,  v28, v28, v30 // v4  = t24a, v28 = t27a
        butterfly_4s    v23, v26, v25, v26 // v23 = t25,  v26 = t26
        butterfly_4s    v7,  v8,  v29, v31 // v7  = t31a, v3  = t28a
        butterfly_4s    v22, v27, v24, v27 // v22 = t30,  v27 = t29

        dmbutterfly     v27, v20, v0.s[2], v0.s[3], v24, v25, v30, v31        // v27 = t18a, v20 = t29a
        dmbutterfly     v8,  v5,  v0.s[2], v0.s[3], v24, v25, v30, v31        // v3  = t19,  v5  = t28
        dmbutterfly     v28, v6,  v0.s[2], v0.s[3], v24, v25, v30, v31, neg=1 // v28 = t27,  v6  = t20
        dmbutterfly     v26, v21, v0.s[2], v0.s[3], v24, v25, v30, v31, neg=1 // v26 = t26a, v21 = t21a

        butterfly_4s    v31, v24, v7,  v4  // v31 = t31,  v24 = t24
        butterfly_4s    v30, v25, v22, v23 // v30 = t30a, v25 = t25a
        butterfly_4s_r  v23, v16, v16, v18 // v23 = t23,  v16 = t16
        butterfly_4s_r  v22, v17, v17, v19 // v22 = t22a, v17 = t17a
        butterfly_4s    v18, v21, v27, v21 // v18 = t18,  v21 = t21
        butterfly_4s_r  v27, v28, v5,  v28 // v27 = t27a, v28 = t28a
        butterfly_4s    v29, v26, v20, v26 // v29 = t29,  v26 = t26
        butterfly_4s    v19, v20, v8,  v6  // v19 = t19a, v20 = t20

        dmbutterfly0    v27, v20, v27, v20, v4, v5, v6, v7, v8, v9 // v27 = t27,  v20 = t20
        dmbutterfly0    v26, v21, v26, v21, v4, v5, v6, v7, v8, v9 // v26 = t26a, v21 = t21a
        dmbutterfly0    v25, v22, v25, v22, v4, v5, v6, v7, v8, v9 // v25 = t25,  v22 = t22
        dmbutterfly0    v24, v23, v24, v23, v4, v5, v6, v7, v8, v9 // v24 = t24a, v23 = t23a
.endm

// Do an 32-point IDCT of a 4x32 slice out of a 32x32 matrix.
// The 32-point IDCT can be decomposed into two 16-point IDCTs;
// a normal IDCT16 with every other input component (the even ones, with
// each output written twice), followed by a separate 16-point IDCT
// of the odd inputs, added/subtracted onto the outputs of the first idct16.
// x0 = dst (temp buffer)
// x1 = unused
// x2 = src
// x9 = double input stride
function idct32_1d_4x32_pass1_neon
        movi            v4.4s,  #0

        // v16 = IN(0), v17 = IN(2) ... v31 = IN(30)
.irp i, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31
        ld1             {v\i\().4s}, [x2]
        st1             {v4.4s},  [x2], x9
.endr

        idct16

        // Do four 4x4 transposes. Originally, v16-v31 contain the
        // 16 rows. Afterwards, v16-v19, v20-v23, v24-v27 and v28-v31
        // contain the four transposed 4x4 blocks.
        transpose_4x4s  v16, v17, v18, v19, v4, v5, v6, v7
        transpose_4x4s  v20, v21, v22, v23, v4, v5, v6, v7
        transpose_4x4s  v24, v25, v26, v27, v4, v5, v6, v7
        transpose_4x4s  v28, v29, v30, v31, v4, v5, v6, v7

        // Store the registers a, b, c, d horizontally, followed by the
        // same registers d, c, b, a mirrored.
.macro store_rev a, b, c, d
        // There's no rev128 instruction, but we reverse each 64 bit
        // half, and then flip them using an ext with 8 bytes offset.
        rev64           v7.4s, v\d\().4s
        st1             {v\a\().4s},  [x0], #16
        ext             v7.16b, v7.16b, v7.16b, #8
        st1             {v\b\().4s},  [x0], #16
        rev64           v6.4s, v\c\().4s
        st1             {v\c\().4s},  [x0], #16
        ext             v6.16b, v6.16b, v6.16b, #8
        st1             {v\d\().4s},  [x0], #16
        rev64           v5.4s, v\b\().4s
        st1             {v7.4s},  [x0], #16
        ext             v5.16b, v5.16b, v5.16b, #8
        st1             {v6.4s},  [x0], #16
        rev64           v4.4s, v\a\().4s
        st1             {v5.4s},  [x0], #16
        ext             v4.16b, v4.16b, v4.16b, #8
        st1             {v4.4s},  [x0], #16
.endm
        store_rev       16, 20, 24, 28
        store_rev       17, 21, 25, 29
        store_rev       18, 22, 26, 30
        store_rev       19, 23, 27, 31
        sub             x0,  x0,  #512
.purgem store_rev

        // Move x2 back to the start of the input, and move
        // to the first odd row
        sub             x2,  x2,  x9, lsl #4
        add             x2,  x2,  #128

        movi            v4.4s,  #0
        // v16 = IN(1), v17 = IN(3) ... v31 = IN(31)
.irp i, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31
        ld1             {v\i\().4s}, [x2]
        st1             {v4.4s},  [x2], x9
.endr

        idct32_odd

        transpose_4x4s  v31, v30, v29, v28, v4, v5, v6, v7
        transpose_4x4s  v27, v26, v25, v24, v4, v5, v6, v7
        transpose_4x4s  v23, v22, v21, v20, v4, v5, v6, v7
        transpose_4x4s  v19, v18, v17, v16, v4, v5, v6, v7

        // Store the registers a, b, c, d horizontally,
        // adding into the output first, and the mirrored,
        // subtracted from the output.
.macro store_rev a, b, c, d
        ld1             {v4.4s},  [x0]
        rev64           v9.4s, v\d\().4s
        add             v4.4s, v4.4s, v\a\().4s
        st1             {v4.4s},  [x0], #16
        rev64           v8.4s, v\c\().4s
        ld1             {v4.4s},  [x0]
        ext             v9.16b, v9.16b, v9.16b, #8
        add             v4.4s, v4.4s, v\b\().4s
        st1             {v4.4s},  [x0], #16
        ext             v8.16b, v8.16b, v8.16b, #8
        ld1             {v4.4s},  [x0]
        rev64           v\b\().4s, v\b\().4s
        add             v4.4s, v4.4s, v\c\().4s
        st1             {v4.4s},  [x0], #16
        rev64           v\a\().4s, v\a\().4s
        ld1             {v4.4s},  [x0]
        ext             v\b\().16b, v\b\().16b, v\b\().16b, #8
        add             v4.4s, v4.4s, v\d\().4s
        st1             {v4.4s},  [x0], #16
        ext             v\a\().16b, v\a\().16b, v\a\().16b, #8
        ld1             {v4.4s},  [x0]
        sub             v4.4s, v4.4s, v9.4s
        st1             {v4.4s},  [x0], #16
        ld1             {v4.4s},  [x0]
        sub             v4.4s, v4.4s, v8.4s
        st1             {v4.4s},  [x0], #16
        ld1             {v4.4s},  [x0]
        sub             v4.4s, v4.4s, v\b\().4s
        st1             {v4.4s},  [x0], #16
        ld1             {v4.4s},  [x0]
        sub             v4.4s, v4.4s, v\a\().4s
        st1             {v4.4s},  [x0], #16
.endm

        store_rev       31, 27, 23, 19
        store_rev       30, 26, 22, 18
        store_rev       29, 25, 21, 17
        store_rev       28, 24, 20, 16
.purgem store_rev
        ret
endfunc

// This is mostly the same as 4x32_pass1, but without the transpose,
// and use the source as temp buffer between the two idct passes, and
// add into the destination.
// x0 = dst
// x1 = dst stride
// x2 = src (temp buffer)
// x7 = negative double temp buffer stride
// x9 = double temp buffer stride
function idct32_1d_4x32_pass2_neon
        // v16 = IN(0), v17 = IN(2) ... v31 = IN(30)
.irp i, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31
        ld1             {v\i\().4s}, [x2], x9
.endr
        sub             x2,  x2,  x9, lsl #4

        idct16

.irp i, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31
        st1             {v\i\().4s}, [x2], x9
.endr

        sub             x2,  x2,  x9, lsl #4
        add             x2,  x2,  #128

        // v16 = IN(1), v17 = IN(3) ... v31 = IN(31)
.irp i, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31
        ld1             {v\i\().4s}, [x2], x9
.endr
        sub             x2,  x2,  x9, lsl #4
        sub             x2,  x2,  #128

        idct32_odd

.macro load_acc_store a, b, c, d, neg=0
.if \neg == 0
        ld1             {v4.4s},  [x2], x9
        ld1             {v5.4s},  [x2], x9
        add             v4.4s, v4.4s, v\a\().4s
        ld1             {v6.4s},  [x2], x9
        add             v5.4s, v5.4s, v\b\().4s
        ld1             {v7.4s},  [x2], x9
        add             v6.4s, v6.4s, v\c\().4s
        add             v7.4s, v7.4s, v\d\().4s
.else
        ld1             {v4.4s},  [x2], x7
        ld1             {v5.4s},  [x2], x7
        sub             v4.4s, v4.4s, v\a\().4s
        ld1             {v6.4s},  [x2], x7
        sub             v5.4s, v5.4s, v\b\().4s
        ld1             {v7.4s},  [x2], x7
        sub             v6.4s, v6.4s, v\c\().4s
        sub             v7.4s, v7.4s, v\d\().4s
.endif
        ld1             {v8.4h},   [x0], x1
        ld1             {v8.d}[1], [x0], x1
        srshr           v4.4s, v4.4s, #6
        ld1             {v9.4h},   [x0], x1
        srshr           v5.4s, v5.4s, #6
        uaddw           v4.4s, v4.4s, v8.4h
        ld1             {v9.d}[1], [x0], x1
        srshr           v6.4s, v6.4s, #6
        uaddw2          v5.4s, v5.4s, v8.8h
        srshr           v7.4s, v7.4s, #6
        sub             x0,  x0,  x1, lsl #2
        uaddw           v6.4s, v6.4s, v9.4h
        sqxtun          v4.4h, v4.4s
        uaddw2          v7.4s, v7.4s, v9.8h
        sqxtun2         v4.8h, v5.4s
        umin            v4.8h, v4.8h, v15.8h
        st1             {v4.4h},   [x0], x1
        sqxtun          v5.4h, v6.4s
        st1             {v4.d}[1], [x0], x1
        sqxtun2         v5.8h, v7.4s
        umin            v5.8h, v5.8h, v15.8h
        st1             {v5.4h},   [x0], x1
        st1             {v5.d}[1], [x0], x1
.endm
        load_acc_store  31, 30, 29, 28
        load_acc_store  27, 26, 25, 24
        load_acc_store  23, 22, 21, 20
        load_acc_store  19, 18, 17, 16
        sub             x2,  x2,  x9
        load_acc_store  16, 17, 18, 19, 1
        load_acc_store  20, 21, 22, 23, 1
        load_acc_store  24, 25, 26, 27, 1
        load_acc_store  28, 29, 30, 31, 1
.purgem load_acc_store
        ret
endfunc

const min_eob_idct_idct_32, align=4
        .short  0, 9, 34, 70, 135, 240, 336, 448
endconst

function vp9_idct_idct_32x32_add_16_neon
        cmp             w3,  #1
        b.eq            idct32x32_dc_add_neon

        movrel          x10, idct_coeffs
        movrel          x12, min_eob_idct_idct_32, 2

        mov             x15, x30
        stp             d8,  d9,  [sp, #-0x10]!
        stp             d10, d11, [sp, #-0x10]!
        stp             d12, d13, [sp, #-0x10]!
        stp             d14, d15, [sp, #-0x10]!

        sub             sp,  sp,  #4096

        mov             x4,  x0
        mov             x5,  x1
        mov             x6,  x2

        // Double stride of the input, since we only read every other line
        mov             x9,  #256
        neg             x7,  x9

        ld1             {v0.8h,v1.8h},   [x10], #32
        sxtl            v2.4s,  v1.4h
        sxtl2           v3.4s,  v1.8h
        sxtl2           v1.4s,  v0.8h
        sxtl            v0.4s,  v0.4h
        ld1             {v10.8h,v11.8h}, [x10]
        sxtl            v12.4s, v11.4h
        sxtl2           v13.4s, v11.8h
        sxtl2           v11.4s, v10.8h
        sxtl            v10.4s, v10.4h

        dup             v15.8h, w13

.irp i, 0, 4, 8, 12, 16, 20, 24, 28
        add             x0,  sp,  #(\i*128)
.if \i > 0
        ldrh            w1,  [x12], #2
        cmp             w3,  w1
        mov             x1,  #(32 - \i)/4
        b.le            1f
.endif
        add             x2,  x6,  #(\i*4)
        bl              idct32_1d_4x32_pass1_neon
.endr
        b               3f

1:
        // Write zeros to the temp buffer for pass 2
        movi            v16.4s,  #0
        movi            v17.4s,  #0
        movi            v18.4s,  #0
        movi            v19.4s,  #0
2:
        subs            x1,  x1,  #1
.rept 4
        st1             {v16.4s-v19.4s},  [x0], #64
        st1             {v16.4s-v19.4s},  [x0], #64
.endr
        b.ne            2b
3:
.irp i, 0, 4, 8, 12, 16, 20, 24, 28
        add             x0,  x4,  #(\i*2)
        mov             x1,  x5
        add             x2,  sp,  #(\i*4)
        bl              idct32_1d_4x32_pass2_neon
.endr

        add             sp,  sp,  #4096
        ldp             d14, d15, [sp], 0x10
        ldp             d12, d13, [sp], 0x10
        ldp             d10, d11, [sp], 0x10
        ldp             d8,  d9,  [sp], 0x10

        br              x15
endfunc

function ff_vp9_idct_idct_32x32_add_10_neon, export=1
        mov             x13, #0x03ff
        b               vp9_idct_idct_32x32_add_16_neon
endfunc

function ff_vp9_idct_idct_32x32_add_12_neon, export=1
        mov             x13, #0x0fff
        b               vp9_idct_idct_32x32_add_16_neon
endfunc