summaryrefslogtreecommitdiffstats
path: root/media/libjpeg/simd/jsimd_arm_neon.S
blob: cd2612724aa995787f713ebfb773d63dcc6d8216 (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
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
2525
2526
2527
2528
2529
2530
2531
2532
2533
2534
2535
2536
2537
2538
2539
2540
2541
2542
2543
2544
2545
2546
2547
2548
2549
2550
2551
2552
2553
2554
2555
2556
2557
2558
2559
2560
2561
2562
2563
2564
2565
2566
2567
2568
2569
2570
2571
2572
2573
2574
2575
2576
2577
2578
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614
2615
2616
2617
2618
2619
2620
2621
2622
2623
2624
2625
2626
2627
2628
2629
2630
2631
2632
2633
2634
2635
2636
2637
2638
2639
2640
2641
2642
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
2669
2670
2671
2672
2673
2674
2675
2676
2677
2678
2679
2680
2681
2682
2683
2684
2685
2686
2687
2688
2689
2690
2691
2692
2693
2694
2695
2696
2697
2698
2699
2700
2701
2702
2703
2704
2705
2706
2707
2708
2709
2710
2711
2712
2713
2714
2715
2716
2717
2718
2719
2720
2721
2722
2723
2724
2725
2726
2727
2728
2729
2730
2731
2732
2733
2734
2735
2736
2737
2738
2739
2740
2741
2742
2743
2744
2745
2746
2747
2748
2749
2750
2751
2752
2753
2754
2755
2756
2757
2758
2759
2760
2761
2762
2763
2764
2765
2766
2767
2768
2769
2770
2771
2772
2773
2774
2775
2776
2777
2778
2779
2780
2781
2782
2783
2784
2785
2786
2787
2788
2789
2790
2791
2792
2793
2794
2795
2796
2797
2798
2799
2800
2801
2802
2803
2804
2805
2806
2807
2808
2809
2810
2811
2812
2813
2814
2815
2816
2817
2818
2819
2820
2821
2822
2823
2824
2825
2826
2827
2828
2829
2830
2831
2832
2833
2834
2835
2836
2837
2838
2839
2840
2841
2842
2843
2844
2845
2846
2847
2848
2849
2850
2851
2852
2853
2854
2855
2856
2857
2858
2859
2860
2861
2862
2863
2864
2865
2866
2867
2868
2869
2870
2871
2872
2873
2874
2875
2876
2877
2878
/*
 * ARMv7 NEON optimizations for libjpeg-turbo
 *
 * Copyright (C) 2009-2011, Nokia Corporation and/or its subsidiary(-ies).
 * All Rights Reserved.
 * Author: Siarhei Siamashka <siarhei.siamashka@nokia.com>
 * Copyright (C) 2014, Siarhei Siamashka.  All Rights Reserved.
 * Copyright (C) 2014, Linaro Limited.  All Rights Reserved.
 * Copyright (C) 2015, D. R. Commander.  All Rights Reserved.
 * Copyright (C) 2015-2016, Matthieu Darbois.  All Rights Reserved.
 *
 * This software is provided 'as-is', without any express or implied
 * warranty.  In no event will the authors be held liable for any damages
 * arising from the use of this software.
 *
 * Permission is granted to anyone to use this software for any purpose,
 * including commercial applications, and to alter it and redistribute it
 * freely, subject to the following restrictions:
 *
 * 1. The origin of this software must not be misrepresented; you must not
 *    claim that you wrote the original software. If you use this software
 *    in a product, an acknowledgment in the product documentation would be
 *    appreciated but is not required.
 * 2. Altered source versions must be plainly marked as such, and must not be
 *    misrepresented as being the original software.
 * 3. This notice may not be removed or altered from any source distribution.
 */

#if defined(__linux__) && defined(__ELF__)
.section .note.GNU-stack, "", %progbits  /* mark stack as non-executable */
#endif

.text
.fpu neon
.arch armv7a
.object_arch armv4
.arm
.syntax unified


#define RESPECT_STRICT_ALIGNMENT 1


/*****************************************************************************/

/* Supplementary macro for setting function attributes */
.macro asm_function fname
#ifdef __APPLE__
    .globl _\fname
_\fname:
#else
    .global \fname
#ifdef __ELF__
    .hidden \fname
    .type \fname, %function
#endif
\fname:
#endif
.endm

/* Transpose a block of 4x4 coefficients in four 64-bit registers */
.macro transpose_4x4 x0, x1, x2, x3
    vtrn.16         \x0, \x1
    vtrn.16         \x2, \x3
    vtrn.32         \x0, \x2
    vtrn.32         \x1, \x3
.endm


#define CENTERJSAMPLE 128

/*****************************************************************************/

/*
 * Perform dequantization and inverse DCT on one block of coefficients.
 *
 * GLOBAL(void)
 * jsimd_idct_islow_neon (void *dct_table, JCOEFPTR coef_block,
 *                        JSAMPARRAY output_buf, JDIMENSION output_col)
 */

#define FIX_0_298631336 (2446)
#define FIX_0_390180644 (3196)
#define FIX_0_541196100 (4433)
#define FIX_0_765366865 (6270)
#define FIX_0_899976223 (7373)
#define FIX_1_175875602 (9633)
#define FIX_1_501321110 (12299)
#define FIX_1_847759065 (15137)
#define FIX_1_961570560 (16069)
#define FIX_2_053119869 (16819)
#define FIX_2_562915447 (20995)
#define FIX_3_072711026 (25172)

#define FIX_1_175875602_MINUS_1_961570560 (FIX_1_175875602 - FIX_1_961570560)
#define FIX_1_175875602_MINUS_0_390180644 (FIX_1_175875602 - FIX_0_390180644)
#define FIX_0_541196100_MINUS_1_847759065 (FIX_0_541196100 - FIX_1_847759065)
#define FIX_3_072711026_MINUS_2_562915447 (FIX_3_072711026 - FIX_2_562915447)
#define FIX_0_298631336_MINUS_0_899976223 (FIX_0_298631336 - FIX_0_899976223)
#define FIX_1_501321110_MINUS_0_899976223 (FIX_1_501321110 - FIX_0_899976223)
#define FIX_2_053119869_MINUS_2_562915447 (FIX_2_053119869 - FIX_2_562915447)
#define FIX_0_541196100_PLUS_0_765366865  (FIX_0_541196100 + FIX_0_765366865)

/*
 * Reference SIMD-friendly 1-D ISLOW iDCT C implementation.
 * Uses some ideas from the comments in 'simd/jiss2int-64.asm'
 */
#define REF_1D_IDCT(xrow0, xrow1, xrow2, xrow3, xrow4, xrow5, xrow6, xrow7)   \
{                                                                             \
    DCTELEM row0, row1, row2, row3, row4, row5, row6, row7;                   \
    JLONG   q1, q2, q3, q4, q5, q6, q7;                                       \
    JLONG   tmp11_plus_tmp2, tmp11_minus_tmp2;                                \
                                                                              \
    /* 1-D iDCT input data */                                                 \
    row0 = xrow0;                                                             \
    row1 = xrow1;                                                             \
    row2 = xrow2;                                                             \
    row3 = xrow3;                                                             \
    row4 = xrow4;                                                             \
    row5 = xrow5;                                                             \
    row6 = xrow6;                                                             \
    row7 = xrow7;                                                             \
                                                                              \
    q5 = row7 + row3;                                                         \
    q4 = row5 + row1;                                                         \
    q6 = MULTIPLY(q5, FIX_1_175875602_MINUS_1_961570560) +                    \
         MULTIPLY(q4, FIX_1_175875602);                                       \
    q7 = MULTIPLY(q5, FIX_1_175875602) +                                      \
         MULTIPLY(q4, FIX_1_175875602_MINUS_0_390180644);                     \
    q2 = MULTIPLY(row2, FIX_0_541196100) +                                    \
         MULTIPLY(row6, FIX_0_541196100_MINUS_1_847759065);                   \
    q4 = q6;                                                                  \
    q3 = ((JLONG) row0 - (JLONG) row4) << 13;                                 \
    q6 += MULTIPLY(row5, -FIX_2_562915447) +                                  \
          MULTIPLY(row3, FIX_3_072711026_MINUS_2_562915447);                  \
    /* now we can use q1 (reloadable constants have been used up) */          \
    q1 = q3 + q2;                                                             \
    q4 += MULTIPLY(row7, FIX_0_298631336_MINUS_0_899976223) +                 \
          MULTIPLY(row1, -FIX_0_899976223);                                   \
    q5 = q7;                                                                  \
    q1 = q1 + q6;                                                             \
    q7 += MULTIPLY(row7, -FIX_0_899976223) +                                  \
          MULTIPLY(row1, FIX_1_501321110_MINUS_0_899976223);                  \
                                                                              \
    /* (tmp11 + tmp2) has been calculated (out_row1 before descale) */        \
    tmp11_plus_tmp2 = q1;                                                     \
    row1 = 0;                                                                 \
                                                                              \
    q1 = q1 - q6;                                                             \
    q5 += MULTIPLY(row5, FIX_2_053119869_MINUS_2_562915447) +                 \
          MULTIPLY(row3, -FIX_2_562915447);                                   \
    q1 = q1 - q6;                                                             \
    q6 = MULTIPLY(row2, FIX_0_541196100_PLUS_0_765366865) +                   \
         MULTIPLY(row6, FIX_0_541196100);                                     \
    q3 = q3 - q2;                                                             \
                                                                              \
    /* (tmp11 - tmp2) has been calculated (out_row6 before descale) */        \
    tmp11_minus_tmp2 = q1;                                                    \
                                                                              \
    q1 = ((JLONG) row0 + (JLONG) row4) << 13;                                 \
    q2 = q1 + q6;                                                             \
    q1 = q1 - q6;                                                             \
                                                                              \
    /* pick up the results */                                                 \
    tmp0  = q4;                                                               \
    tmp1  = q5;                                                               \
    tmp2  = (tmp11_plus_tmp2 - tmp11_minus_tmp2) / 2;                         \
    tmp3  = q7;                                                               \
    tmp10 = q2;                                                               \
    tmp11 = (tmp11_plus_tmp2 + tmp11_minus_tmp2) / 2;                         \
    tmp12 = q3;                                                               \
    tmp13 = q1;                                                               \
}

#define XFIX_0_899976223                   d0[0]
#define XFIX_0_541196100                   d0[1]
#define XFIX_2_562915447                   d0[2]
#define XFIX_0_298631336_MINUS_0_899976223 d0[3]
#define XFIX_1_501321110_MINUS_0_899976223 d1[0]
#define XFIX_2_053119869_MINUS_2_562915447 d1[1]
#define XFIX_0_541196100_PLUS_0_765366865  d1[2]
#define XFIX_1_175875602                   d1[3]
#define XFIX_1_175875602_MINUS_0_390180644 d2[0]
#define XFIX_0_541196100_MINUS_1_847759065 d2[1]
#define XFIX_3_072711026_MINUS_2_562915447 d2[2]
#define XFIX_1_175875602_MINUS_1_961570560 d2[3]

.balign 16
jsimd_idct_islow_neon_consts:
  .short FIX_0_899976223                    /* d0[0] */
  .short FIX_0_541196100                    /* d0[1] */
  .short FIX_2_562915447                    /* d0[2] */
  .short FIX_0_298631336_MINUS_0_899976223  /* d0[3] */
  .short FIX_1_501321110_MINUS_0_899976223  /* d1[0] */
  .short FIX_2_053119869_MINUS_2_562915447  /* d1[1] */
  .short FIX_0_541196100_PLUS_0_765366865   /* d1[2] */
  .short FIX_1_175875602                    /* d1[3] */
  /* reloadable constants */
  .short FIX_1_175875602_MINUS_0_390180644  /* d2[0] */
  .short FIX_0_541196100_MINUS_1_847759065  /* d2[1] */
  .short FIX_3_072711026_MINUS_2_562915447  /* d2[2] */
  .short FIX_1_175875602_MINUS_1_961570560  /* d2[3] */

asm_function jsimd_idct_islow_neon

    DCT_TABLE       .req r0
    COEF_BLOCK      .req r1
    OUTPUT_BUF      .req r2
    OUTPUT_COL      .req r3
    TMP1            .req r0
    TMP2            .req r1
    TMP3            .req r2
    TMP4            .req ip

    ROW0L           .req d16
    ROW0R           .req d17
    ROW1L           .req d18
    ROW1R           .req d19
    ROW2L           .req d20
    ROW2R           .req d21
    ROW3L           .req d22
    ROW3R           .req d23
    ROW4L           .req d24
    ROW4R           .req d25
    ROW5L           .req d26
    ROW5R           .req d27
    ROW6L           .req d28
    ROW6R           .req d29
    ROW7L           .req d30
    ROW7R           .req d31

    /* Load and dequantize coefficients into NEON registers
     * with the following allocation:
     *       0 1 2 3 | 4 5 6 7
     *      ---------+--------
     *   0 | d16     | d17     ( q8  )
     *   1 | d18     | d19     ( q9  )
     *   2 | d20     | d21     ( q10 )
     *   3 | d22     | d23     ( q11 )
     *   4 | d24     | d25     ( q12 )
     *   5 | d26     | d27     ( q13 )
     *   6 | d28     | d29     ( q14 )
     *   7 | d30     | d31     ( q15 )
     */
    adr             ip, jsimd_idct_islow_neon_consts
    vld1.16         {d16, d17, d18, d19}, [COEF_BLOCK, :128]!
    vld1.16         {d0, d1, d2, d3}, [DCT_TABLE, :128]!
    vld1.16         {d20, d21, d22, d23}, [COEF_BLOCK, :128]!
    vmul.s16        q8, q8, q0
    vld1.16         {d4, d5, d6, d7}, [DCT_TABLE, :128]!
    vmul.s16        q9, q9, q1
    vld1.16         {d24, d25, d26, d27}, [COEF_BLOCK, :128]!
    vmul.s16        q10, q10, q2
    vld1.16         {d0, d1, d2, d3}, [DCT_TABLE, :128]!
    vmul.s16        q11, q11, q3
    vld1.16         {d28, d29, d30, d31}, [COEF_BLOCK, :128]
    vmul.s16        q12, q12, q0
    vld1.16         {d4, d5, d6, d7}, [DCT_TABLE, :128]!
    vmul.s16        q14, q14, q2
    vmul.s16        q13, q13, q1
    vld1.16         {d0, d1, d2, d3}, [ip, :128]  /* load constants */
    add             ip, ip, #16
    vmul.s16        q15, q15, q3
    vpush           {d8-d15}                      /* save NEON registers */
    /* 1-D IDCT, pass 1, left 4x8 half */
    vadd.s16        d4, ROW7L, ROW3L
    vadd.s16        d5, ROW5L, ROW1L
    vmull.s16       q6, d4, XFIX_1_175875602_MINUS_1_961570560
    vmlal.s16       q6, d5, XFIX_1_175875602
    vmull.s16       q7, d4, XFIX_1_175875602
      /* Check for the zero coefficients in the right 4x8 half */
      push            {r4, r5}
    vmlal.s16       q7, d5, XFIX_1_175875602_MINUS_0_390180644
    vsubl.s16       q3, ROW0L, ROW4L
      ldrd            r4, [COEF_BLOCK, #(-96 + 2 * (4 + 1 * 8))]
    vmull.s16       q2, ROW2L, XFIX_0_541196100
    vmlal.s16       q2, ROW6L, XFIX_0_541196100_MINUS_1_847759065
      orr             r0, r4, r5
    vmov            q4, q6
    vmlsl.s16       q6, ROW5L, XFIX_2_562915447
      ldrd            r4, [COEF_BLOCK, #(-96 + 2 * (4 + 2 * 8))]
    vmlal.s16       q6, ROW3L, XFIX_3_072711026_MINUS_2_562915447
    vshl.s32        q3, q3, #13
      orr             r0, r0, r4
    vmlsl.s16       q4, ROW1L, XFIX_0_899976223
      orr             r0, r0, r5
    vadd.s32        q1, q3, q2
      ldrd            r4, [COEF_BLOCK, #(-96 + 2 * (4 + 3 * 8))]
    vmov            q5, q7
    vadd.s32        q1, q1, q6
      orr             r0, r0, r4
    vmlsl.s16       q7, ROW7L, XFIX_0_899976223
      orr             r0, r0, r5
    vmlal.s16       q7, ROW1L, XFIX_1_501321110_MINUS_0_899976223
    vrshrn.s32      ROW1L, q1, #11
      ldrd            r4, [COEF_BLOCK, #(-96 + 2 * (4 + 4 * 8))]
    vsub.s32        q1, q1, q6
    vmlal.s16       q5, ROW5L, XFIX_2_053119869_MINUS_2_562915447
      orr             r0, r0, r4
    vmlsl.s16       q5, ROW3L, XFIX_2_562915447
      orr             r0, r0, r5
    vsub.s32        q1, q1, q6
    vmull.s16       q6, ROW2L, XFIX_0_541196100_PLUS_0_765366865
      ldrd            r4, [COEF_BLOCK, #(-96 + 2 * (4 + 5 * 8))]
    vmlal.s16       q6, ROW6L, XFIX_0_541196100
    vsub.s32        q3, q3, q2
      orr             r0, r0, r4
    vrshrn.s32      ROW6L, q1, #11
      orr             r0, r0, r5
    vadd.s32        q1, q3, q5
      ldrd            r4, [COEF_BLOCK, #(-96 + 2 * (4 + 6 * 8))]
    vsub.s32        q3, q3, q5
    vaddl.s16       q5, ROW0L, ROW4L
      orr             r0, r0, r4
    vrshrn.s32      ROW2L, q1, #11
      orr             r0, r0, r5
    vrshrn.s32      ROW5L, q3, #11
      ldrd            r4, [COEF_BLOCK, #(-96 + 2 * (4 + 7 * 8))]
    vshl.s32        q5, q5, #13
    vmlal.s16       q4, ROW7L, XFIX_0_298631336_MINUS_0_899976223
      orr             r0, r0, r4
    vadd.s32        q2, q5, q6
      orrs            r0, r0, r5
    vsub.s32        q1, q5, q6
    vadd.s32        q6, q2, q7
      ldrd            r4, [COEF_BLOCK, #(-96 + 2 * (4 + 0 * 8))]
    vsub.s32        q2, q2, q7
    vadd.s32        q5, q1, q4
      orr             r0, r4, r5
    vsub.s32        q3, q1, q4
      pop             {r4, r5}
    vrshrn.s32      ROW7L, q2, #11
    vrshrn.s32      ROW3L, q5, #11
    vrshrn.s32      ROW0L, q6, #11
    vrshrn.s32      ROW4L, q3, #11

      beq             3f  /* Go to do some special handling for the sparse
                             right 4x8 half */

    /* 1-D IDCT, pass 1, right 4x8 half */
    vld1.s16        {d2}, [ip, :64]  /* reload constants */
    vadd.s16        d10, ROW7R, ROW3R
    vadd.s16        d8, ROW5R, ROW1R
      /* Transpose left 4x8 half */
      vtrn.16         ROW6L, ROW7L
    vmull.s16       q6, d10, XFIX_1_175875602_MINUS_1_961570560
    vmlal.s16       q6, d8, XFIX_1_175875602
      vtrn.16         ROW2L, ROW3L
    vmull.s16       q7, d10, XFIX_1_175875602
    vmlal.s16       q7, d8, XFIX_1_175875602_MINUS_0_390180644
      vtrn.16         ROW0L, ROW1L
    vsubl.s16       q3, ROW0R, ROW4R
    vmull.s16       q2, ROW2R, XFIX_0_541196100
    vmlal.s16       q2, ROW6R, XFIX_0_541196100_MINUS_1_847759065
      vtrn.16         ROW4L, ROW5L
    vmov            q4, q6
    vmlsl.s16       q6, ROW5R, XFIX_2_562915447
    vmlal.s16       q6, ROW3R, XFIX_3_072711026_MINUS_2_562915447
      vtrn.32         ROW1L, ROW3L
    vshl.s32        q3, q3, #13
    vmlsl.s16       q4, ROW1R, XFIX_0_899976223
      vtrn.32         ROW4L, ROW6L
    vadd.s32        q1, q3, q2
    vmov            q5, q7
    vadd.s32        q1, q1, q6
      vtrn.32         ROW0L, ROW2L
    vmlsl.s16       q7, ROW7R, XFIX_0_899976223
    vmlal.s16       q7, ROW1R, XFIX_1_501321110_MINUS_0_899976223
    vrshrn.s32      ROW1R, q1, #11
      vtrn.32         ROW5L, ROW7L
    vsub.s32        q1, q1, q6
    vmlal.s16       q5, ROW5R, XFIX_2_053119869_MINUS_2_562915447
    vmlsl.s16       q5, ROW3R, XFIX_2_562915447
    vsub.s32        q1, q1, q6
    vmull.s16       q6, ROW2R, XFIX_0_541196100_PLUS_0_765366865
    vmlal.s16       q6, ROW6R, XFIX_0_541196100
    vsub.s32        q3, q3, q2
    vrshrn.s32      ROW6R, q1, #11
    vadd.s32        q1, q3, q5
    vsub.s32        q3, q3, q5
    vaddl.s16       q5, ROW0R, ROW4R
    vrshrn.s32      ROW2R, q1, #11
    vrshrn.s32      ROW5R, q3, #11
    vshl.s32        q5, q5, #13
    vmlal.s16       q4, ROW7R, XFIX_0_298631336_MINUS_0_899976223
    vadd.s32        q2, q5, q6
    vsub.s32        q1, q5, q6
    vadd.s32        q6, q2, q7
    vsub.s32        q2, q2, q7
    vadd.s32        q5, q1, q4
    vsub.s32        q3, q1, q4
    vrshrn.s32      ROW7R, q2, #11
    vrshrn.s32      ROW3R, q5, #11
    vrshrn.s32      ROW0R, q6, #11
    vrshrn.s32      ROW4R, q3, #11
    /* Transpose right 4x8 half */
    vtrn.16         ROW6R, ROW7R
    vtrn.16         ROW2R, ROW3R
    vtrn.16         ROW0R, ROW1R
    vtrn.16         ROW4R, ROW5R
    vtrn.32         ROW1R, ROW3R
    vtrn.32         ROW4R, ROW6R
    vtrn.32         ROW0R, ROW2R
    vtrn.32         ROW5R, ROW7R

1:  /* 1-D IDCT, pass 2 (normal variant), left 4x8 half */
    vld1.s16        {d2}, [ip, :64]               /* reload constants */
    vmull.s16       q6, ROW1R, XFIX_1_175875602   /* ROW5L <-> ROW1R */
    vmlal.s16       q6, ROW1L, XFIX_1_175875602
    vmlal.s16       q6, ROW3R, XFIX_1_175875602_MINUS_1_961570560  /* ROW7L <-> ROW3R */
    vmlal.s16       q6, ROW3L, XFIX_1_175875602_MINUS_1_961570560
    vmull.s16       q7, ROW3R, XFIX_1_175875602   /* ROW7L <-> ROW3R */
    vmlal.s16       q7, ROW3L, XFIX_1_175875602
    vmlal.s16       q7, ROW1R, XFIX_1_175875602_MINUS_0_390180644  /* ROW5L <-> ROW1R */
    vmlal.s16       q7, ROW1L, XFIX_1_175875602_MINUS_0_390180644
    vsubl.s16       q3, ROW0L, ROW0R              /* ROW4L <-> ROW0R */
    vmull.s16       q2, ROW2L, XFIX_0_541196100
    vmlal.s16       q2, ROW2R, XFIX_0_541196100_MINUS_1_847759065  /* ROW6L <-> ROW2R */
    vmov            q4, q6
    vmlsl.s16       q6, ROW1R, XFIX_2_562915447   /* ROW5L <-> ROW1R */
    vmlal.s16       q6, ROW3L, XFIX_3_072711026_MINUS_2_562915447
    vshl.s32        q3, q3, #13
    vmlsl.s16       q4, ROW1L, XFIX_0_899976223
    vadd.s32        q1, q3, q2
    vmov            q5, q7
    vadd.s32        q1, q1, q6
    vmlsl.s16       q7, ROW3R, XFIX_0_899976223   /* ROW7L <-> ROW3R */
    vmlal.s16       q7, ROW1L, XFIX_1_501321110_MINUS_0_899976223
    vshrn.s32       ROW1L, q1, #16
    vsub.s32        q1, q1, q6
    vmlal.s16       q5, ROW1R, XFIX_2_053119869_MINUS_2_562915447  /* ROW5L <-> ROW1R */
    vmlsl.s16       q5, ROW3L, XFIX_2_562915447
    vsub.s32        q1, q1, q6
    vmull.s16       q6, ROW2L, XFIX_0_541196100_PLUS_0_765366865
    vmlal.s16       q6, ROW2R, XFIX_0_541196100   /* ROW6L <-> ROW2R */
    vsub.s32        q3, q3, q2
    vshrn.s32       ROW2R, q1, #16                /* ROW6L <-> ROW2R */
    vadd.s32        q1, q3, q5
    vsub.s32        q3, q3, q5
    vaddl.s16       q5, ROW0L, ROW0R              /* ROW4L <-> ROW0R */
    vshrn.s32       ROW2L, q1, #16
    vshrn.s32       ROW1R, q3, #16                /* ROW5L <-> ROW1R */
    vshl.s32        q5, q5, #13
    vmlal.s16       q4, ROW3R, XFIX_0_298631336_MINUS_0_899976223  /* ROW7L <-> ROW3R */
    vadd.s32        q2, q5, q6
    vsub.s32        q1, q5, q6
    vadd.s32        q6, q2, q7
    vsub.s32        q2, q2, q7
    vadd.s32        q5, q1, q4
    vsub.s32        q3, q1, q4
    vshrn.s32       ROW3R, q2, #16                /* ROW7L <-> ROW3R */
    vshrn.s32       ROW3L, q5, #16
    vshrn.s32       ROW0L, q6, #16
    vshrn.s32       ROW0R, q3, #16                /* ROW4L <-> ROW0R */
    /* 1-D IDCT, pass 2, right 4x8 half */
    vld1.s16        {d2}, [ip, :64]               /* reload constants */
    vmull.s16       q6, ROW5R, XFIX_1_175875602
    vmlal.s16       q6, ROW5L, XFIX_1_175875602   /* ROW5L <-> ROW1R */
    vmlal.s16       q6, ROW7R, XFIX_1_175875602_MINUS_1_961570560
    vmlal.s16       q6, ROW7L, XFIX_1_175875602_MINUS_1_961570560  /* ROW7L <-> ROW3R */
    vmull.s16       q7, ROW7R, XFIX_1_175875602
    vmlal.s16       q7, ROW7L, XFIX_1_175875602   /* ROW7L <-> ROW3R */
    vmlal.s16       q7, ROW5R, XFIX_1_175875602_MINUS_0_390180644
    vmlal.s16       q7, ROW5L, XFIX_1_175875602_MINUS_0_390180644  /* ROW5L <-> ROW1R */
    vsubl.s16       q3, ROW4L, ROW4R              /* ROW4L <-> ROW0R */
    vmull.s16       q2, ROW6L, XFIX_0_541196100   /* ROW6L <-> ROW2R */
    vmlal.s16       q2, ROW6R, XFIX_0_541196100_MINUS_1_847759065
    vmov            q4, q6
    vmlsl.s16       q6, ROW5R, XFIX_2_562915447
    vmlal.s16       q6, ROW7L, XFIX_3_072711026_MINUS_2_562915447  /* ROW7L <-> ROW3R */
    vshl.s32        q3, q3, #13
    vmlsl.s16       q4, ROW5L, XFIX_0_899976223   /* ROW5L <-> ROW1R */
    vadd.s32        q1, q3, q2
    vmov            q5, q7
    vadd.s32        q1, q1, q6
    vmlsl.s16       q7, ROW7R, XFIX_0_899976223
    vmlal.s16       q7, ROW5L, XFIX_1_501321110_MINUS_0_899976223  /* ROW5L <-> ROW1R */
    vshrn.s32       ROW5L, q1, #16                /* ROW5L <-> ROW1R */
    vsub.s32        q1, q1, q6
    vmlal.s16       q5, ROW5R, XFIX_2_053119869_MINUS_2_562915447
    vmlsl.s16       q5, ROW7L, XFIX_2_562915447   /* ROW7L <-> ROW3R */
    vsub.s32        q1, q1, q6
    vmull.s16       q6, ROW6L, XFIX_0_541196100_PLUS_0_765366865  /* ROW6L <-> ROW2R */
    vmlal.s16       q6, ROW6R, XFIX_0_541196100
    vsub.s32        q3, q3, q2
    vshrn.s32       ROW6R, q1, #16
    vadd.s32        q1, q3, q5
    vsub.s32        q3, q3, q5
    vaddl.s16       q5, ROW4L, ROW4R              /* ROW4L <-> ROW0R */
    vshrn.s32       ROW6L, q1, #16                /* ROW6L <-> ROW2R */
    vshrn.s32       ROW5R, q3, #16
    vshl.s32        q5, q5, #13
    vmlal.s16       q4, ROW7R, XFIX_0_298631336_MINUS_0_899976223
    vadd.s32        q2, q5, q6
    vsub.s32        q1, q5, q6
    vadd.s32        q6, q2, q7
    vsub.s32        q2, q2, q7
    vadd.s32        q5, q1, q4
    vsub.s32        q3, q1, q4
    vshrn.s32       ROW7R, q2, #16
    vshrn.s32       ROW7L, q5, #16                /* ROW7L <-> ROW3R */
    vshrn.s32       ROW4L, q6, #16                /* ROW4L <-> ROW0R */
    vshrn.s32       ROW4R, q3, #16

2:  /* Descale to 8-bit and range limit */
    vqrshrn.s16     d16, q8, #2
    vqrshrn.s16     d17, q9, #2
    vqrshrn.s16     d18, q10, #2
    vqrshrn.s16     d19, q11, #2
    vpop            {d8-d15}                      /* restore NEON registers */
    vqrshrn.s16     d20, q12, #2
      /* Transpose the final 8-bit samples and do signed->unsigned conversion */
      vtrn.16         q8, q9
    vqrshrn.s16     d21, q13, #2
    vqrshrn.s16     d22, q14, #2
      vmov.u8         q0, #(CENTERJSAMPLE)
    vqrshrn.s16     d23, q15, #2
      vtrn.8          d16, d17
      vtrn.8          d18, d19
      vadd.u8         q8, q8, q0
      vadd.u8         q9, q9, q0
      vtrn.16         q10, q11
        /* Store results to the output buffer */
        ldmia           OUTPUT_BUF!, {TMP1, TMP2}
        add             TMP1, TMP1, OUTPUT_COL
        add             TMP2, TMP2, OUTPUT_COL
        vst1.8          {d16}, [TMP1]
      vtrn.8          d20, d21
        vst1.8          {d17}, [TMP2]
        ldmia           OUTPUT_BUF!, {TMP1, TMP2}
        add             TMP1, TMP1, OUTPUT_COL
        add             TMP2, TMP2, OUTPUT_COL
        vst1.8          {d18}, [TMP1]
      vadd.u8         q10, q10, q0
        vst1.8          {d19}, [TMP2]
        ldmia           OUTPUT_BUF, {TMP1, TMP2, TMP3, TMP4}
        add             TMP1, TMP1, OUTPUT_COL
        add             TMP2, TMP2, OUTPUT_COL
        add             TMP3, TMP3, OUTPUT_COL
        add             TMP4, TMP4, OUTPUT_COL
      vtrn.8          d22, d23
        vst1.8          {d20}, [TMP1]
      vadd.u8         q11, q11, q0
        vst1.8          {d21}, [TMP2]
        vst1.8          {d22}, [TMP3]
        vst1.8          {d23}, [TMP4]
    bx              lr

3:  /* Left 4x8 half is done, right 4x8 half contains mostly zeros */

    /* Transpose left 4x8 half */
    vtrn.16         ROW6L, ROW7L
    vtrn.16         ROW2L, ROW3L
    vtrn.16         ROW0L, ROW1L
    vtrn.16         ROW4L, ROW5L
    vshl.s16        ROW0R, ROW0R, #2  /* PASS1_BITS */
    vtrn.32         ROW1L, ROW3L
    vtrn.32         ROW4L, ROW6L
    vtrn.32         ROW0L, ROW2L
    vtrn.32         ROW5L, ROW7L

    cmp             r0, #0
    beq             4f  /* Right 4x8 half has all zeros, go to 'sparse' second
                           pass */

    /* Only row 0 is non-zero for the right 4x8 half  */
    vdup.s16        ROW1R, ROW0R[1]
    vdup.s16        ROW2R, ROW0R[2]
    vdup.s16        ROW3R, ROW0R[3]
    vdup.s16        ROW4R, ROW0R[0]
    vdup.s16        ROW5R, ROW0R[1]
    vdup.s16        ROW6R, ROW0R[2]
    vdup.s16        ROW7R, ROW0R[3]
    vdup.s16        ROW0R, ROW0R[0]
    b               1b  /* Go to 'normal' second pass */

4:  /* 1-D IDCT, pass 2 (sparse variant with zero rows 4-7), left 4x8 half */
    vld1.s16        {d2}, [ip, :64]               /* reload constants */
    vmull.s16       q6, ROW1L, XFIX_1_175875602
    vmlal.s16       q6, ROW3L, XFIX_1_175875602_MINUS_1_961570560
    vmull.s16       q7, ROW3L, XFIX_1_175875602
    vmlal.s16       q7, ROW1L, XFIX_1_175875602_MINUS_0_390180644
    vmull.s16       q2, ROW2L, XFIX_0_541196100
    vshll.s16       q3, ROW0L, #13
    vmov            q4, q6
    vmlal.s16       q6, ROW3L, XFIX_3_072711026_MINUS_2_562915447
    vmlsl.s16       q4, ROW1L, XFIX_0_899976223
    vadd.s32        q1, q3, q2
    vmov            q5, q7
    vmlal.s16       q7, ROW1L, XFIX_1_501321110_MINUS_0_899976223
    vadd.s32        q1, q1, q6
    vadd.s32        q6, q6, q6
    vmlsl.s16       q5, ROW3L, XFIX_2_562915447
    vshrn.s32       ROW1L, q1, #16
    vsub.s32        q1, q1, q6
    vmull.s16       q6, ROW2L, XFIX_0_541196100_PLUS_0_765366865
    vsub.s32        q3, q3, q2
    vshrn.s32       ROW2R, q1, #16                /* ROW6L <-> ROW2R */
    vadd.s32        q1, q3, q5
    vsub.s32        q3, q3, q5
    vshll.s16       q5, ROW0L, #13
    vshrn.s32       ROW2L, q1, #16
    vshrn.s32       ROW1R, q3, #16                /* ROW5L <-> ROW1R */
    vadd.s32        q2, q5, q6
    vsub.s32        q1, q5, q6
    vadd.s32        q6, q2, q7
    vsub.s32        q2, q2, q7
    vadd.s32        q5, q1, q4
    vsub.s32        q3, q1, q4
    vshrn.s32       ROW3R, q2, #16                /* ROW7L <-> ROW3R */
    vshrn.s32       ROW3L, q5, #16
    vshrn.s32       ROW0L, q6, #16
    vshrn.s32       ROW0R, q3, #16                /* ROW4L <-> ROW0R */
    /* 1-D IDCT, pass 2 (sparse variant with zero rows 4-7), right 4x8 half */
    vld1.s16        {d2}, [ip, :64]               /* reload constants */
    vmull.s16       q6, ROW5L, XFIX_1_175875602
    vmlal.s16       q6, ROW7L, XFIX_1_175875602_MINUS_1_961570560
    vmull.s16       q7, ROW7L, XFIX_1_175875602
    vmlal.s16       q7, ROW5L, XFIX_1_175875602_MINUS_0_390180644
    vmull.s16       q2, ROW6L, XFIX_0_541196100
    vshll.s16       q3, ROW4L, #13
    vmov            q4, q6
    vmlal.s16       q6, ROW7L, XFIX_3_072711026_MINUS_2_562915447
    vmlsl.s16       q4, ROW5L, XFIX_0_899976223
    vadd.s32        q1, q3, q2
    vmov            q5, q7
    vmlal.s16       q7, ROW5L, XFIX_1_501321110_MINUS_0_899976223
    vadd.s32        q1, q1, q6
    vadd.s32        q6, q6, q6
    vmlsl.s16       q5, ROW7L, XFIX_2_562915447
    vshrn.s32       ROW5L, q1, #16                /* ROW5L <-> ROW1R */
    vsub.s32        q1, q1, q6
    vmull.s16       q6, ROW6L, XFIX_0_541196100_PLUS_0_765366865
    vsub.s32        q3, q3, q2
    vshrn.s32       ROW6R, q1, #16
    vadd.s32        q1, q3, q5
    vsub.s32        q3, q3, q5
    vshll.s16       q5, ROW4L, #13
    vshrn.s32       ROW6L, q1, #16                /* ROW6L <-> ROW2R */
    vshrn.s32       ROW5R, q3, #16
    vadd.s32        q2, q5, q6
    vsub.s32        q1, q5, q6
    vadd.s32        q6, q2, q7
    vsub.s32        q2, q2, q7
    vadd.s32        q5, q1, q4
    vsub.s32        q3, q1, q4
    vshrn.s32       ROW7R, q2, #16
    vshrn.s32       ROW7L, q5, #16                /* ROW7L <-> ROW3R */
    vshrn.s32       ROW4L, q6, #16                /* ROW4L <-> ROW0R */
    vshrn.s32       ROW4R, q3, #16
    b               2b                            /* Go to epilogue */

    .unreq          DCT_TABLE
    .unreq          COEF_BLOCK
    .unreq          OUTPUT_BUF
    .unreq          OUTPUT_COL
    .unreq          TMP1
    .unreq          TMP2
    .unreq          TMP3
    .unreq          TMP4

    .unreq          ROW0L
    .unreq          ROW0R
    .unreq          ROW1L
    .unreq          ROW1R
    .unreq          ROW2L
    .unreq          ROW2R
    .unreq          ROW3L
    .unreq          ROW3R
    .unreq          ROW4L
    .unreq          ROW4R
    .unreq          ROW5L
    .unreq          ROW5R
    .unreq          ROW6L
    .unreq          ROW6R
    .unreq          ROW7L
    .unreq          ROW7R


/*****************************************************************************/

/*
 * jsimd_idct_ifast_neon
 *
 * This function contains a fast, not so accurate integer implementation of
 * the inverse DCT (Discrete Cosine Transform). It uses the same calculations
 * and produces exactly the same output as IJG's original 'jpeg_idct_ifast'
 * function from jidctfst.c
 *
 * Normally 1-D AAN DCT needs 5 multiplications and 29 additions.
 * But in ARM NEON case some extra additions are required because VQDMULH
 * instruction can't handle the constants larger than 1. So the expressions
 * like "x * 1.082392200" have to be converted to "x * 0.082392200 + x",
 * which introduces an extra addition. Overall, there are 6 extra additions
 * per 1-D IDCT pass, totalling to 5 VQDMULH and 35 VADD/VSUB instructions.
 */

#define XFIX_1_082392200 d0[0]
#define XFIX_1_414213562 d0[1]
#define XFIX_1_847759065 d0[2]
#define XFIX_2_613125930 d0[3]

.balign 16
jsimd_idct_ifast_neon_consts:
  .short (277 * 128 - 256 * 128)  /* XFIX_1_082392200 */
  .short (362 * 128 - 256 * 128)  /* XFIX_1_414213562 */
  .short (473 * 128 - 256 * 128)  /* XFIX_1_847759065 */
  .short (669 * 128 - 512 * 128)  /* XFIX_2_613125930 */

asm_function jsimd_idct_ifast_neon

    DCT_TABLE       .req r0
    COEF_BLOCK      .req r1
    OUTPUT_BUF      .req r2
    OUTPUT_COL      .req r3
    TMP1            .req r0
    TMP2            .req r1
    TMP3            .req r2
    TMP4            .req ip

    /* Load and dequantize coefficients into NEON registers
     * with the following allocation:
     *       0 1 2 3 | 4 5 6 7
     *      ---------+--------
     *   0 | d16     | d17     ( q8  )
     *   1 | d18     | d19     ( q9  )
     *   2 | d20     | d21     ( q10 )
     *   3 | d22     | d23     ( q11 )
     *   4 | d24     | d25     ( q12 )
     *   5 | d26     | d27     ( q13 )
     *   6 | d28     | d29     ( q14 )
     *   7 | d30     | d31     ( q15 )
     */
    adr             ip, jsimd_idct_ifast_neon_consts
    vld1.16         {d16, d17, d18, d19}, [COEF_BLOCK, :128]!
    vld1.16         {d0, d1, d2, d3}, [DCT_TABLE, :128]!
    vld1.16         {d20, d21, d22, d23}, [COEF_BLOCK, :128]!
    vmul.s16        q8, q8, q0
    vld1.16         {d4, d5, d6, d7}, [DCT_TABLE, :128]!
    vmul.s16        q9, q9, q1
    vld1.16         {d24, d25, d26, d27}, [COEF_BLOCK, :128]!
    vmul.s16        q10, q10, q2
    vld1.16         {d0, d1, d2, d3}, [DCT_TABLE, :128]!
    vmul.s16        q11, q11, q3
    vld1.16         {d28, d29, d30, d31}, [COEF_BLOCK, :128]
    vmul.s16        q12, q12, q0
    vld1.16         {d4, d5, d6, d7}, [DCT_TABLE, :128]!
    vmul.s16        q14, q14, q2
    vmul.s16        q13, q13, q1
    vld1.16         {d0}, [ip, :64]  /* load constants */
    vmul.s16        q15, q15, q3
    vpush           {d8-d13}         /* save NEON registers */
    /* 1-D IDCT, pass 1 */
    vsub.s16        q2, q10, q14
    vadd.s16        q14, q10, q14
    vsub.s16        q1, q11, q13
    vadd.s16        q13, q11, q13
    vsub.s16        q5, q9, q15
    vadd.s16        q15, q9, q15
    vqdmulh.s16     q4, q2, XFIX_1_414213562
    vqdmulh.s16     q6, q1, XFIX_2_613125930
    vadd.s16        q3, q1, q1
    vsub.s16        q1, q5, q1
    vadd.s16        q10, q2, q4
    vqdmulh.s16     q4, q1, XFIX_1_847759065
    vsub.s16        q2, q15, q13
    vadd.s16        q3, q3, q6
    vqdmulh.s16     q6, q2, XFIX_1_414213562
    vadd.s16        q1, q1, q4
    vqdmulh.s16     q4, q5, XFIX_1_082392200
    vsub.s16        q10, q10, q14
    vadd.s16        q2, q2, q6
    vsub.s16        q6, q8, q12
    vadd.s16        q12, q8, q12
    vadd.s16        q9, q5, q4
    vadd.s16        q5, q6, q10
    vsub.s16        q10, q6, q10
    vadd.s16        q6, q15, q13
    vadd.s16        q8, q12, q14
    vsub.s16        q3, q6, q3
    vsub.s16        q12, q12, q14
    vsub.s16        q3, q3, q1
    vsub.s16        q1, q9, q1
    vadd.s16        q2, q3, q2
    vsub.s16        q15, q8, q6
    vadd.s16        q1, q1, q2
    vadd.s16        q8, q8, q6
    vadd.s16        q14, q5, q3
    vsub.s16        q9, q5, q3
    vsub.s16        q13, q10, q2
    vadd.s16        q10, q10, q2
      /* Transpose */
      vtrn.16         q8, q9
    vsub.s16        q11, q12, q1
      vtrn.16         q14, q15
    vadd.s16        q12, q12, q1
      vtrn.16         q10, q11
      vtrn.16         q12, q13
      vtrn.32         q9, q11
      vtrn.32         q12, q14
      vtrn.32         q8, q10
      vtrn.32         q13, q15
      vswp            d28, d21
      vswp            d26, d19
    /* 1-D IDCT, pass 2 */
    vsub.s16        q2, q10, q14
      vswp            d30, d23
    vadd.s16        q14, q10, q14
      vswp            d24, d17
    vsub.s16        q1, q11, q13
    vadd.s16        q13, q11, q13
    vsub.s16        q5, q9, q15
    vadd.s16        q15, q9, q15
    vqdmulh.s16     q4, q2, XFIX_1_414213562
    vqdmulh.s16     q6, q1, XFIX_2_613125930
    vadd.s16        q3, q1, q1
    vsub.s16        q1, q5, q1
    vadd.s16        q10, q2, q4
    vqdmulh.s16     q4, q1, XFIX_1_847759065
    vsub.s16        q2, q15, q13
    vadd.s16        q3, q3, q6
    vqdmulh.s16     q6, q2, XFIX_1_414213562
    vadd.s16        q1, q1, q4
    vqdmulh.s16     q4, q5, XFIX_1_082392200
    vsub.s16        q10, q10, q14
    vadd.s16        q2, q2, q6
    vsub.s16        q6, q8, q12
    vadd.s16        q12, q8, q12
    vadd.s16        q9, q5, q4
    vadd.s16        q5, q6, q10
    vsub.s16        q10, q6, q10
    vadd.s16        q6, q15, q13
    vadd.s16        q8, q12, q14
    vsub.s16        q3, q6, q3
    vsub.s16        q12, q12, q14
    vsub.s16        q3, q3, q1
    vsub.s16        q1, q9, q1
    vadd.s16        q2, q3, q2
    vsub.s16        q15, q8, q6
    vadd.s16        q1, q1, q2
    vadd.s16        q8, q8, q6
    vadd.s16        q14, q5, q3
    vsub.s16        q9, q5, q3
    vsub.s16        q13, q10, q2
    vpop            {d8-d13}      /* restore NEON registers */
    vadd.s16        q10, q10, q2
    vsub.s16        q11, q12, q1
    vadd.s16        q12, q12, q1
    /* Descale to 8-bit and range limit */
    vmov.u8         q0, #0x80
    vqshrn.s16      d16, q8, #5
    vqshrn.s16      d17, q9, #5
    vqshrn.s16      d18, q10, #5
    vqshrn.s16      d19, q11, #5
    vqshrn.s16      d20, q12, #5
    vqshrn.s16      d21, q13, #5
    vqshrn.s16      d22, q14, #5
    vqshrn.s16      d23, q15, #5
    vadd.u8         q8, q8, q0
    vadd.u8         q9, q9, q0
    vadd.u8         q10, q10, q0
    vadd.u8         q11, q11, q0
    /* Transpose the final 8-bit samples */
    vtrn.16         q8, q9
    vtrn.16         q10, q11
    vtrn.32         q8, q10
    vtrn.32         q9, q11
    vtrn.8          d16, d17
    vtrn.8          d18, d19
      /* Store results to the output buffer */
      ldmia           OUTPUT_BUF!, {TMP1, TMP2}
      add             TMP1, TMP1, OUTPUT_COL
      add             TMP2, TMP2, OUTPUT_COL
      vst1.8          {d16}, [TMP1]
      vst1.8          {d17}, [TMP2]
      ldmia           OUTPUT_BUF!, {TMP1, TMP2}
      add             TMP1, TMP1, OUTPUT_COL
      add             TMP2, TMP2, OUTPUT_COL
      vst1.8          {d18}, [TMP1]
    vtrn.8          d20, d21
      vst1.8          {d19}, [TMP2]
      ldmia           OUTPUT_BUF, {TMP1, TMP2, TMP3, TMP4}
      add             TMP1, TMP1, OUTPUT_COL
      add             TMP2, TMP2, OUTPUT_COL
      add             TMP3, TMP3, OUTPUT_COL
      add             TMP4, TMP4, OUTPUT_COL
      vst1.8          {d20}, [TMP1]
    vtrn.8          d22, d23
      vst1.8          {d21}, [TMP2]
      vst1.8          {d22}, [TMP3]
      vst1.8          {d23}, [TMP4]
    bx              lr

    .unreq          DCT_TABLE
    .unreq          COEF_BLOCK
    .unreq          OUTPUT_BUF
    .unreq          OUTPUT_COL
    .unreq          TMP1
    .unreq          TMP2
    .unreq          TMP3
    .unreq          TMP4


/*****************************************************************************/

/*
 * jsimd_idct_4x4_neon
 *
 * This function contains inverse-DCT code for getting reduced-size
 * 4x4 pixels output from an 8x8 DCT block. It uses the same  calculations
 * and produces exactly the same output as IJG's original 'jpeg_idct_4x4'
 * function from jpeg-6b (jidctred.c).
 *
 * NOTE: jpeg-8 has an improved implementation of 4x4 inverse-DCT, which
 *       requires much less arithmetic operations and hence should be faster.
 *       The primary purpose of this particular NEON optimized function is
 *       bit exact compatibility with jpeg-6b.
 *
 * TODO: a bit better instructions scheduling can be achieved by expanding
 *       idct_helper/transpose_4x4 macros and reordering instructions,
 *       but readability will suffer somewhat.
 */

#define CONST_BITS  13

#define FIX_0_211164243 (1730)   /* FIX(0.211164243) */
#define FIX_0_509795579 (4176)   /* FIX(0.509795579) */
#define FIX_0_601344887 (4926)   /* FIX(0.601344887) */
#define FIX_0_720959822 (5906)   /* FIX(0.720959822) */
#define FIX_0_765366865 (6270)   /* FIX(0.765366865) */
#define FIX_0_850430095 (6967)   /* FIX(0.850430095) */
#define FIX_0_899976223 (7373)   /* FIX(0.899976223) */
#define FIX_1_061594337 (8697)   /* FIX(1.061594337) */
#define FIX_1_272758580 (10426)  /* FIX(1.272758580) */
#define FIX_1_451774981 (11893)  /* FIX(1.451774981) */
#define FIX_1_847759065 (15137)  /* FIX(1.847759065) */
#define FIX_2_172734803 (17799)  /* FIX(2.172734803) */
#define FIX_2_562915447 (20995)  /* FIX(2.562915447) */
#define FIX_3_624509785 (29692)  /* FIX(3.624509785) */

.balign 16
jsimd_idct_4x4_neon_consts:
  .short FIX_1_847759065      /* d0[0] */
  .short -FIX_0_765366865     /* d0[1] */
  .short -FIX_0_211164243     /* d0[2] */
  .short FIX_1_451774981      /* d0[3] */
  .short -FIX_2_172734803     /* d1[0] */
  .short FIX_1_061594337      /* d1[1] */
  .short -FIX_0_509795579     /* d1[2] */
  .short -FIX_0_601344887     /* d1[3] */
  .short FIX_0_899976223      /* d2[0] */
  .short FIX_2_562915447      /* d2[1] */
  .short 1 << (CONST_BITS+1)  /* d2[2] */
  .short 0                    /* d2[3] */

.macro idct_helper x4, x6, x8, x10, x12, x14, x16, shift, y26, y27, y28, y29
    vmull.s16       q14, \x4, d2[2]
    vmlal.s16       q14, \x8, d0[0]
    vmlal.s16       q14, \x14, d0[1]

    vmull.s16       q13, \x16, d1[2]
    vmlal.s16       q13, \x12, d1[3]
    vmlal.s16       q13, \x10, d2[0]
    vmlal.s16       q13, \x6, d2[1]

    vmull.s16       q15, \x4, d2[2]
    vmlsl.s16       q15, \x8, d0[0]
    vmlsl.s16       q15, \x14, d0[1]

    vmull.s16       q12, \x16, d0[2]
    vmlal.s16       q12, \x12, d0[3]
    vmlal.s16       q12, \x10, d1[0]
    vmlal.s16       q12, \x6, d1[1]

    vadd.s32        q10, q14, q13
    vsub.s32        q14, q14, q13

  .if \shift > 16
    vrshr.s32       q10, q10, #\shift
    vrshr.s32       q14, q14, #\shift
    vmovn.s32       \y26, q10
    vmovn.s32       \y29, q14
  .else
    vrshrn.s32      \y26, q10, #\shift
    vrshrn.s32      \y29, q14, #\shift
  .endif

    vadd.s32        q10, q15, q12
    vsub.s32        q15, q15, q12

  .if \shift > 16
    vrshr.s32       q10, q10, #\shift
    vrshr.s32       q15, q15, #\shift
    vmovn.s32       \y27, q10
    vmovn.s32       \y28, q15
  .else
    vrshrn.s32      \y27, q10, #\shift
    vrshrn.s32      \y28, q15, #\shift
  .endif
.endm

asm_function jsimd_idct_4x4_neon

    DCT_TABLE       .req r0
    COEF_BLOCK      .req r1
    OUTPUT_BUF      .req r2
    OUTPUT_COL      .req r3
    TMP1            .req r0
    TMP2            .req r1
    TMP3            .req r2
    TMP4            .req ip

    vpush           {d8-d15}

    /* Load constants (d3 is just used for padding) */
    adr             TMP4, jsimd_idct_4x4_neon_consts
    vld1.16         {d0, d1, d2, d3}, [TMP4, :128]

    /* Load all COEF_BLOCK into NEON registers with the following allocation:
     *       0 1 2 3 | 4 5 6 7
     *      ---------+--------
     *   0 | d4      | d5
     *   1 | d6      | d7
     *   2 | d8      | d9
     *   3 | d10     | d11
     *   4 | -       | -
     *   5 | d12     | d13
     *   6 | d14     | d15
     *   7 | d16     | d17
     */
    vld1.16         {d4, d5, d6, d7}, [COEF_BLOCK, :128]!
    vld1.16         {d8, d9, d10, d11}, [COEF_BLOCK, :128]!
    add COEF_BLOCK, COEF_BLOCK, #16
    vld1.16         {d12, d13, d14, d15}, [COEF_BLOCK, :128]!
    vld1.16         {d16, d17}, [COEF_BLOCK, :128]!
    /* dequantize */
    vld1.16         {d18, d19, d20, d21}, [DCT_TABLE, :128]!
    vmul.s16        q2, q2, q9
    vld1.16         {d22, d23, d24, d25}, [DCT_TABLE, :128]!
    vmul.s16        q3, q3, q10
    vmul.s16        q4, q4, q11
    add             DCT_TABLE, DCT_TABLE, #16
    vld1.16         {d26, d27, d28, d29}, [DCT_TABLE, :128]!
    vmul.s16        q5, q5, q12
    vmul.s16        q6, q6, q13
    vld1.16         {d30, d31}, [DCT_TABLE, :128]!
    vmul.s16        q7, q7, q14
    vmul.s16        q8, q8, q15

    /* Pass 1 */
    idct_helper     d4, d6, d8, d10, d12, d14, d16, 12, d4, d6, d8, d10
    transpose_4x4   d4, d6, d8, d10
    idct_helper     d5, d7, d9, d11, d13, d15, d17, 12, d5, d7, d9, d11
    transpose_4x4   d5, d7, d9, d11

    /* Pass 2 */
    idct_helper     d4, d6, d8, d10, d7, d9, d11, 19, d26, d27, d28, d29
    transpose_4x4   d26, d27, d28, d29

    /* Range limit */
    vmov.u16        q15, #0x80
    vadd.s16        q13, q13, q15
    vadd.s16        q14, q14, q15
    vqmovun.s16     d26, q13
    vqmovun.s16     d27, q14

    /* Store results to the output buffer */
    ldmia           OUTPUT_BUF, {TMP1, TMP2, TMP3, TMP4}
    add             TMP1, TMP1, OUTPUT_COL
    add             TMP2, TMP2, OUTPUT_COL
    add             TMP3, TMP3, OUTPUT_COL
    add             TMP4, TMP4, OUTPUT_COL

#if defined(__ARMEL__) && !RESPECT_STRICT_ALIGNMENT
    /* We can use much less instructions on little endian systems if the
     * OS kernel is not configured to trap unaligned memory accesses
     */
    vst1.32         {d26[0]}, [TMP1]!
    vst1.32         {d27[0]}, [TMP3]!
    vst1.32         {d26[1]}, [TMP2]!
    vst1.32         {d27[1]}, [TMP4]!
#else
    vst1.8          {d26[0]}, [TMP1]!
    vst1.8          {d27[0]}, [TMP3]!
    vst1.8          {d26[1]}, [TMP1]!
    vst1.8          {d27[1]}, [TMP3]!
    vst1.8          {d26[2]}, [TMP1]!
    vst1.8          {d27[2]}, [TMP3]!
    vst1.8          {d26[3]}, [TMP1]!
    vst1.8          {d27[3]}, [TMP3]!

    vst1.8          {d26[4]}, [TMP2]!
    vst1.8          {d27[4]}, [TMP4]!
    vst1.8          {d26[5]}, [TMP2]!
    vst1.8          {d27[5]}, [TMP4]!
    vst1.8          {d26[6]}, [TMP2]!
    vst1.8          {d27[6]}, [TMP4]!
    vst1.8          {d26[7]}, [TMP2]!
    vst1.8          {d27[7]}, [TMP4]!
#endif

    vpop            {d8-d15}
    bx              lr

    .unreq          DCT_TABLE
    .unreq          COEF_BLOCK
    .unreq          OUTPUT_BUF
    .unreq          OUTPUT_COL
    .unreq          TMP1
    .unreq          TMP2
    .unreq          TMP3
    .unreq          TMP4

.purgem idct_helper


/*****************************************************************************/

/*
 * jsimd_idct_2x2_neon
 *
 * This function contains inverse-DCT code for getting reduced-size
 * 2x2 pixels output from an 8x8 DCT block. It uses the same  calculations
 * and produces exactly the same output as IJG's original 'jpeg_idct_2x2'
 * function from jpeg-6b (jidctred.c).
 *
 * NOTE: jpeg-8 has an improved implementation of 2x2 inverse-DCT, which
 *       requires much less arithmetic operations and hence should be faster.
 *       The primary purpose of this particular NEON optimized function is
 *       bit exact compatibility with jpeg-6b.
 */

.balign 8
jsimd_idct_2x2_neon_consts:
  .short -FIX_0_720959822  /* d0[0] */
  .short FIX_0_850430095   /* d0[1] */
  .short -FIX_1_272758580  /* d0[2] */
  .short FIX_3_624509785   /* d0[3] */

.macro idct_helper x4, x6, x10, x12, x16, shift, y26, y27
    vshll.s16       q14, \x4, #15
    vmull.s16       q13, \x6, d0[3]
    vmlal.s16       q13, \x10, d0[2]
    vmlal.s16       q13, \x12, d0[1]
    vmlal.s16       q13, \x16, d0[0]

    vadd.s32        q10, q14, q13
    vsub.s32        q14, q14, q13

  .if \shift > 16
    vrshr.s32       q10, q10, #\shift
    vrshr.s32       q14, q14, #\shift
    vmovn.s32       \y26, q10
    vmovn.s32       \y27, q14
  .else
    vrshrn.s32      \y26, q10, #\shift
    vrshrn.s32      \y27, q14, #\shift
  .endif
.endm

asm_function jsimd_idct_2x2_neon

    DCT_TABLE       .req r0
    COEF_BLOCK      .req r1
    OUTPUT_BUF      .req r2
    OUTPUT_COL      .req r3
    TMP1            .req r0
    TMP2            .req ip

    vpush           {d8-d15}

    /* Load constants */
    adr             TMP2, jsimd_idct_2x2_neon_consts
    vld1.16         {d0}, [TMP2, :64]

    /* Load all COEF_BLOCK into NEON registers with the following allocation:
     *       0 1 2 3 | 4 5 6 7
     *      ---------+--------
     *   0 | d4      | d5
     *   1 | d6      | d7
     *   2 | -       | -
     *   3 | d10     | d11
     *   4 | -       | -
     *   5 | d12     | d13
     *   6 | -       | -
     *   7 | d16     | d17
     */
    vld1.16         {d4, d5, d6, d7}, [COEF_BLOCK, :128]!
    add             COEF_BLOCK, COEF_BLOCK, #16
    vld1.16         {d10, d11}, [COEF_BLOCK, :128]!
    add             COEF_BLOCK, COEF_BLOCK, #16
    vld1.16         {d12, d13}, [COEF_BLOCK, :128]!
    add             COEF_BLOCK, COEF_BLOCK, #16
    vld1.16         {d16, d17}, [COEF_BLOCK, :128]!
    /* Dequantize */
    vld1.16         {d18, d19, d20, d21}, [DCT_TABLE, :128]!
    vmul.s16        q2, q2, q9
    vmul.s16        q3, q3, q10
    add             DCT_TABLE, DCT_TABLE, #16
    vld1.16         {d24, d25}, [DCT_TABLE, :128]!
    vmul.s16        q5, q5, q12
    add             DCT_TABLE, DCT_TABLE, #16
    vld1.16         {d26, d27}, [DCT_TABLE, :128]!
    vmul.s16        q6, q6, q13
    add             DCT_TABLE, DCT_TABLE, #16
    vld1.16         {d30, d31}, [DCT_TABLE, :128]!
    vmul.s16        q8, q8, q15

    /* Pass 1 */
#if 0
    idct_helper     d4, d6, d10, d12, d16, 13, d4, d6
    transpose_4x4   d4, d6, d8, d10
    idct_helper     d5, d7, d11, d13, d17, 13, d5, d7
    transpose_4x4   d5, d7, d9, d11
#else
    vmull.s16       q13, d6, d0[3]
    vmlal.s16       q13, d10, d0[2]
    vmlal.s16       q13, d12, d0[1]
    vmlal.s16       q13, d16, d0[0]
    vmull.s16       q12, d7, d0[3]
    vmlal.s16       q12, d11, d0[2]
    vmlal.s16       q12, d13, d0[1]
    vmlal.s16       q12, d17, d0[0]
    vshll.s16       q14, d4, #15
    vshll.s16       q15, d5, #15
    vadd.s32        q10, q14, q13
    vsub.s32        q14, q14, q13
    vrshrn.s32      d4, q10, #13
    vrshrn.s32      d6, q14, #13
    vadd.s32        q10, q15, q12
    vsub.s32        q14, q15, q12
    vrshrn.s32      d5, q10, #13
    vrshrn.s32      d7, q14, #13
    vtrn.16         q2, q3
    vtrn.32         q3, q5
#endif

    /* Pass 2 */
    idct_helper     d4, d6, d10, d7, d11, 20, d26, d27

    /* Range limit */
    vmov.u16        q15, #0x80
    vadd.s16        q13, q13, q15
    vqmovun.s16     d26, q13
    vqmovun.s16     d27, q13

    /* Store results to the output buffer */
    ldmia           OUTPUT_BUF, {TMP1, TMP2}
    add             TMP1, TMP1, OUTPUT_COL
    add             TMP2, TMP2, OUTPUT_COL

    vst1.8          {d26[0]}, [TMP1]!
    vst1.8          {d27[4]}, [TMP1]!
    vst1.8          {d26[1]}, [TMP2]!
    vst1.8          {d27[5]}, [TMP2]!

    vpop            {d8-d15}
    bx              lr

    .unreq          DCT_TABLE
    .unreq          COEF_BLOCK
    .unreq          OUTPUT_BUF
    .unreq          OUTPUT_COL
    .unreq          TMP1
    .unreq          TMP2

.purgem idct_helper


/*****************************************************************************/

/*
 * jsimd_ycc_extrgb_convert_neon
 * jsimd_ycc_extbgr_convert_neon
 * jsimd_ycc_extrgbx_convert_neon
 * jsimd_ycc_extbgrx_convert_neon
 * jsimd_ycc_extxbgr_convert_neon
 * jsimd_ycc_extxrgb_convert_neon
 *
 * Colorspace conversion YCbCr -> RGB
 */


.macro do_load size
  .if \size == 8
    vld1.8          {d4}, [U, :64]!
    vld1.8          {d5}, [V, :64]!
    vld1.8          {d0}, [Y, :64]!
    pld             [U, #64]
    pld             [V, #64]
    pld             [Y, #64]
  .elseif \size == 4
    vld1.8          {d4[0]}, [U]!
    vld1.8          {d4[1]}, [U]!
    vld1.8          {d4[2]}, [U]!
    vld1.8          {d4[3]}, [U]!
    vld1.8          {d5[0]}, [V]!
    vld1.8          {d5[1]}, [V]!
    vld1.8          {d5[2]}, [V]!
    vld1.8          {d5[3]}, [V]!
    vld1.8          {d0[0]}, [Y]!
    vld1.8          {d0[1]}, [Y]!
    vld1.8          {d0[2]}, [Y]!
    vld1.8          {d0[3]}, [Y]!
  .elseif \size == 2
    vld1.8          {d4[4]}, [U]!
    vld1.8          {d4[5]}, [U]!
    vld1.8          {d5[4]}, [V]!
    vld1.8          {d5[5]}, [V]!
    vld1.8          {d0[4]}, [Y]!
    vld1.8          {d0[5]}, [Y]!
  .elseif \size == 1
    vld1.8          {d4[6]}, [U]!
    vld1.8          {d5[6]}, [V]!
    vld1.8          {d0[6]}, [Y]!
  .else
    .error unsupported macroblock size
  .endif
.endm

.macro do_store bpp, size
  .if \bpp == 24
    .if \size == 8
      vst3.8        {d10, d11, d12}, [RGB]!
    .elseif \size == 4
      vst3.8        {d10[0], d11[0], d12[0]}, [RGB]!
      vst3.8        {d10[1], d11[1], d12[1]}, [RGB]!
      vst3.8        {d10[2], d11[2], d12[2]}, [RGB]!
      vst3.8        {d10[3], d11[3], d12[3]}, [RGB]!
    .elseif \size == 2
      vst3.8        {d10[4], d11[4], d12[4]}, [RGB]!
      vst3.8        {d10[5], d11[5], d12[5]}, [RGB]!
    .elseif \size == 1
      vst3.8        {d10[6], d11[6], d12[6]}, [RGB]!
    .else
      .error unsupported macroblock size
    .endif
  .elseif \bpp == 32
    .if \size == 8
      vst4.8        {d10, d11, d12, d13}, [RGB]!
    .elseif \size == 4
      vst4.8        {d10[0], d11[0], d12[0], d13[0]}, [RGB]!
      vst4.8        {d10[1], d11[1], d12[1], d13[1]}, [RGB]!
      vst4.8        {d10[2], d11[2], d12[2], d13[2]}, [RGB]!
      vst4.8        {d10[3], d11[3], d12[3], d13[3]}, [RGB]!
    .elseif \size == 2
      vst4.8        {d10[4], d11[4], d12[4], d13[4]}, [RGB]!
      vst4.8        {d10[5], d11[5], d12[5], d13[5]}, [RGB]!
    .elseif \size == 1
      vst4.8        {d10[6], d11[6], d12[6], d13[6]}, [RGB]!
    .else
      .error unsupported macroblock size
    .endif
  .elseif \bpp == 16
    .if \size == 8
      vst1.16       {q15}, [RGB]!
    .elseif \size == 4
      vst1.16       {d30}, [RGB]!
    .elseif \size == 2
      vst1.16       {d31[0]}, [RGB]!
      vst1.16       {d31[1]}, [RGB]!
    .elseif \size == 1
      vst1.16       {d31[2]}, [RGB]!
    .else
      .error unsupported macroblock size
    .endif
  .else
    .error unsupported bpp
  .endif
.endm

.macro generate_jsimd_ycc_rgb_convert_neon colorid, bpp, r_offs, g_offs, b_offs

/*
 * 2-stage pipelined YCbCr->RGB conversion
 */

.macro do_yuv_to_rgb_stage1
    vaddw.u8        q3, q1, d4      /* q3 = u - 128 */
    vaddw.u8        q4, q1, d5      /* q2 = v - 128 */
    vmull.s16       q10, d6, d1[1]  /* multiply by -11277 */
    vmlal.s16       q10, d8, d1[2]  /* multiply by -23401 */
    vmull.s16       q11, d7, d1[1]  /* multiply by -11277 */
    vmlal.s16       q11, d9, d1[2]  /* multiply by -23401 */
    vmull.s16       q12, d8, d1[0]  /* multiply by 22971 */
    vmull.s16       q13, d9, d1[0]  /* multiply by 22971 */
    vmull.s16       q14, d6, d1[3]  /* multiply by 29033 */
    vmull.s16       q15, d7, d1[3]  /* multiply by 29033 */
.endm

.macro do_yuv_to_rgb_stage2
    vrshrn.s32      d20, q10, #15
    vrshrn.s32      d21, q11, #15
    vrshrn.s32      d24, q12, #14
    vrshrn.s32      d25, q13, #14
    vrshrn.s32      d28, q14, #14
    vrshrn.s32      d29, q15, #14
    vaddw.u8        q11, q10, d0
    vaddw.u8        q12, q12, d0
    vaddw.u8        q14, q14, d0
  .if \bpp != 16
    vqmovun.s16     d1\g_offs, q11
    vqmovun.s16     d1\r_offs, q12
    vqmovun.s16     d1\b_offs, q14
  .else  /* rgb565 */
    vqshlu.s16      q13, q11, #8
    vqshlu.s16      q15, q12, #8
    vqshlu.s16      q14, q14, #8
    vsri.u16        q15, q13, #5
    vsri.u16        q15, q14, #11
  .endif
.endm

.macro do_yuv_to_rgb_stage2_store_load_stage1
                                       /* "do_yuv_to_rgb_stage2" and "store" */
                                       vrshrn.s32      d20, q10, #15
    /* "load" and "do_yuv_to_rgb_stage1" */
    pld             [U, #64]
                                       vrshrn.s32      d21, q11, #15
    pld             [V, #64]
                                       vrshrn.s32      d24, q12, #14
                                       vrshrn.s32      d25, q13, #14
    vld1.8          {d4}, [U, :64]!
                                       vrshrn.s32      d28, q14, #14
    vld1.8          {d5}, [V, :64]!
                                       vrshrn.s32      d29, q15, #14
    vaddw.u8        q3, q1, d4      /* q3 = u - 128 */
    vaddw.u8        q4, q1, d5      /* q2 = v - 128 */
                                       vaddw.u8        q11, q10, d0
    vmull.s16       q10, d6, d1[1]  /* multiply by -11277 */
    vmlal.s16       q10, d8, d1[2]  /* multiply by -23401 */
                                       vaddw.u8        q12, q12, d0
                                       vaddw.u8        q14, q14, d0
  .if \bpp != 16  /**************** rgb24/rgb32 ******************************/
                                       vqmovun.s16     d1\g_offs, q11
    pld             [Y, #64]
                                       vqmovun.s16     d1\r_offs, q12
    vld1.8          {d0}, [Y, :64]!
                                       vqmovun.s16     d1\b_offs, q14
    vmull.s16       q11, d7, d1[1]  /* multiply by -11277 */
    vmlal.s16       q11, d9, d1[2]  /* multiply by -23401 */
                                       do_store        \bpp, 8
    vmull.s16       q12, d8, d1[0]  /* multiply by 22971 */
    vmull.s16       q13, d9, d1[0]  /* multiply by 22971 */
    vmull.s16       q14, d6, d1[3]  /* multiply by 29033 */
    vmull.s16       q15, d7, d1[3]  /* multiply by 29033 */
  .else  /**************************** rgb565 ********************************/
                                       vqshlu.s16      q13, q11, #8
    pld             [Y, #64]
                                       vqshlu.s16      q15, q12, #8
                                       vqshlu.s16      q14, q14, #8
    vld1.8          {d0}, [Y, :64]!
    vmull.s16       q11, d7, d1[1]
    vmlal.s16       q11, d9, d1[2]
                                       vsri.u16        q15, q13, #5
    vmull.s16       q12, d8, d1[0]
                                       vsri.u16        q15, q14, #11
    vmull.s16       q13, d9, d1[0]
    vmull.s16       q14, d6, d1[3]
                                       do_store        \bpp, 8
    vmull.s16       q15, d7, d1[3]
  .endif
.endm

.macro do_yuv_to_rgb
    do_yuv_to_rgb_stage1
    do_yuv_to_rgb_stage2
.endm

/* Apple gas crashes on adrl, work around that by using adr.
 * But this requires a copy of these constants for each function.
 */

.balign 16
jsimd_ycc_\colorid\()_neon_consts:
  .short 0,      0,     0,      0
  .short 22971, -11277, -23401, 29033
  .short -128,  -128,   -128,   -128
  .short -128,  -128,   -128,   -128

asm_function jsimd_ycc_\colorid\()_convert_neon
    OUTPUT_WIDTH    .req r0
    INPUT_BUF       .req r1
    INPUT_ROW       .req r2
    OUTPUT_BUF      .req r3
    NUM_ROWS        .req r4

    INPUT_BUF0      .req r5
    INPUT_BUF1      .req r6
    INPUT_BUF2      .req INPUT_BUF

    RGB             .req r7
    Y               .req r8
    U               .req r9
    V               .req r10
    N               .req ip

    /* Load constants to d1, d2, d3 (d0 is just used for padding) */
    adr             ip, jsimd_ycc_\colorid\()_neon_consts
    vld1.16         {d0, d1, d2, d3}, [ip, :128]

    /* Save ARM registers and handle input arguments */
    push            {r4, r5, r6, r7, r8, r9, r10, lr}
    ldr             NUM_ROWS, [sp, #(4 * 8)]
    ldr             INPUT_BUF0, [INPUT_BUF]
    ldr             INPUT_BUF1, [INPUT_BUF, #4]
    ldr             INPUT_BUF2, [INPUT_BUF, #8]
    .unreq          INPUT_BUF

    /* Save NEON registers */
    vpush           {d8-d15}

    /* Initially set d10, d11, d12, d13 to 0xFF */
    vmov.u8         q5, #255
    vmov.u8         q6, #255

    /* Outer loop over scanlines */
    cmp             NUM_ROWS, #1
    blt             9f
0:
    ldr             Y, [INPUT_BUF0, INPUT_ROW, lsl #2]
    ldr             U, [INPUT_BUF1, INPUT_ROW, lsl #2]
    mov             N, OUTPUT_WIDTH
    ldr             V, [INPUT_BUF2, INPUT_ROW, lsl #2]
    add             INPUT_ROW, INPUT_ROW, #1
    ldr             RGB, [OUTPUT_BUF], #4

    /* Inner loop over pixels */
    subs            N, N, #8
    blt             3f
    do_load         8
    do_yuv_to_rgb_stage1
    subs            N, N, #8
    blt             2f
1:
    do_yuv_to_rgb_stage2_store_load_stage1
    subs            N, N, #8
    bge             1b
2:
    do_yuv_to_rgb_stage2
    do_store        \bpp, 8
    tst             N, #7
    beq             8f
3:
    tst             N, #4
    beq             3f
    do_load         4
3:
    tst             N, #2
    beq             4f
    do_load         2
4:
    tst             N, #1
    beq             5f
    do_load         1
5:
    do_yuv_to_rgb
    tst             N, #4
    beq             6f
    do_store        \bpp, 4
6:
    tst             N, #2
    beq             7f
    do_store        \bpp, 2
7:
    tst             N, #1
    beq             8f
    do_store        \bpp, 1
8:
    subs            NUM_ROWS, NUM_ROWS, #1
    bgt             0b
9:
    /* Restore all registers and return */
    vpop            {d8-d15}
    pop             {r4, r5, r6, r7, r8, r9, r10, pc}

    .unreq          OUTPUT_WIDTH
    .unreq          INPUT_ROW
    .unreq          OUTPUT_BUF
    .unreq          NUM_ROWS
    .unreq          INPUT_BUF0
    .unreq          INPUT_BUF1
    .unreq          INPUT_BUF2
    .unreq          RGB
    .unreq          Y
    .unreq          U
    .unreq          V
    .unreq          N

.purgem do_yuv_to_rgb
.purgem do_yuv_to_rgb_stage1
.purgem do_yuv_to_rgb_stage2
.purgem do_yuv_to_rgb_stage2_store_load_stage1

.endm

/*--------------------------------- id ----- bpp R  G  B */
generate_jsimd_ycc_rgb_convert_neon extrgb,  24, 0, 1, 2
generate_jsimd_ycc_rgb_convert_neon extbgr,  24, 2, 1, 0
generate_jsimd_ycc_rgb_convert_neon extrgbx, 32, 0, 1, 2
generate_jsimd_ycc_rgb_convert_neon extbgrx, 32, 2, 1, 0
generate_jsimd_ycc_rgb_convert_neon extxbgr, 32, 3, 2, 1
generate_jsimd_ycc_rgb_convert_neon extxrgb, 32, 1, 2, 3
generate_jsimd_ycc_rgb_convert_neon rgb565,  16, 0, 0, 0

.purgem do_load
.purgem do_store


/*****************************************************************************/

/*
 * jsimd_extrgb_ycc_convert_neon
 * jsimd_extbgr_ycc_convert_neon
 * jsimd_extrgbx_ycc_convert_neon
 * jsimd_extbgrx_ycc_convert_neon
 * jsimd_extxbgr_ycc_convert_neon
 * jsimd_extxrgb_ycc_convert_neon
 *
 * Colorspace conversion RGB -> YCbCr
 */

.macro do_store size
  .if \size == 8
    vst1.8          {d20}, [Y]!
    vst1.8          {d21}, [U]!
    vst1.8          {d22}, [V]!
  .elseif \size == 4
    vst1.8          {d20[0]}, [Y]!
    vst1.8          {d20[1]}, [Y]!
    vst1.8          {d20[2]}, [Y]!
    vst1.8          {d20[3]}, [Y]!
    vst1.8          {d21[0]}, [U]!
    vst1.8          {d21[1]}, [U]!
    vst1.8          {d21[2]}, [U]!
    vst1.8          {d21[3]}, [U]!
    vst1.8          {d22[0]}, [V]!
    vst1.8          {d22[1]}, [V]!
    vst1.8          {d22[2]}, [V]!
    vst1.8          {d22[3]}, [V]!
  .elseif \size == 2
    vst1.8          {d20[4]}, [Y]!
    vst1.8          {d20[5]}, [Y]!
    vst1.8          {d21[4]}, [U]!
    vst1.8          {d21[5]}, [U]!
    vst1.8          {d22[4]}, [V]!
    vst1.8          {d22[5]}, [V]!
  .elseif \size == 1
    vst1.8          {d20[6]}, [Y]!
    vst1.8          {d21[6]}, [U]!
    vst1.8          {d22[6]}, [V]!
  .else
    .error unsupported macroblock size
  .endif
.endm

.macro do_load bpp, size
  .if \bpp == 24
    .if \size == 8
      vld3.8        {d10, d11, d12}, [RGB]!
      pld           [RGB, #128]
    .elseif \size == 4
      vld3.8        {d10[0], d11[0], d12[0]}, [RGB]!
      vld3.8        {d10[1], d11[1], d12[1]}, [RGB]!
      vld3.8        {d10[2], d11[2], d12[2]}, [RGB]!
      vld3.8        {d10[3], d11[3], d12[3]}, [RGB]!
    .elseif \size == 2
      vld3.8        {d10[4], d11[4], d12[4]}, [RGB]!
      vld3.8        {d10[5], d11[5], d12[5]}, [RGB]!
    .elseif \size == 1
      vld3.8        {d10[6], d11[6], d12[6]}, [RGB]!
    .else
      .error unsupported macroblock size
    .endif
  .elseif \bpp == 32
    .if \size == 8
      vld4.8        {d10, d11, d12, d13}, [RGB]!
      pld           [RGB, #128]
    .elseif \size == 4
      vld4.8        {d10[0], d11[0], d12[0], d13[0]}, [RGB]!
      vld4.8        {d10[1], d11[1], d12[1], d13[1]}, [RGB]!
      vld4.8        {d10[2], d11[2], d12[2], d13[2]}, [RGB]!
      vld4.8        {d10[3], d11[3], d12[3], d13[3]}, [RGB]!
    .elseif \size == 2
      vld4.8        {d10[4], d11[4], d12[4], d13[4]}, [RGB]!
      vld4.8        {d10[5], d11[5], d12[5], d13[5]}, [RGB]!
    .elseif \size == 1
      vld4.8        {d10[6], d11[6], d12[6], d13[6]}, [RGB]!
    .else
      .error unsupported macroblock size
    .endif
  .else
    .error unsupported bpp
  .endif
.endm

.macro generate_jsimd_rgb_ycc_convert_neon colorid, bpp, r_offs, g_offs, b_offs

/*
 * 2-stage pipelined RGB->YCbCr conversion
 */

.macro do_rgb_to_yuv_stage1
    vmovl.u8        q2, d1\r_offs  /* r = { d4, d5 } */
    vmovl.u8        q3, d1\g_offs  /* g = { d6, d7 } */
    vmovl.u8        q4, d1\b_offs  /* b = { d8, d9 } */
    vmull.u16       q7, d4, d0[0]
    vmlal.u16       q7, d6, d0[1]
    vmlal.u16       q7, d8, d0[2]
    vmull.u16       q8, d5, d0[0]
    vmlal.u16       q8, d7, d0[1]
    vmlal.u16       q8, d9, d0[2]
    vrev64.32       q9, q1
    vrev64.32       q13, q1
    vmlsl.u16       q9, d4, d0[3]
    vmlsl.u16       q9, d6, d1[0]
    vmlal.u16       q9, d8, d1[1]
    vmlsl.u16       q13, d5, d0[3]
    vmlsl.u16       q13, d7, d1[0]
    vmlal.u16       q13, d9, d1[1]
    vrev64.32       q14, q1
    vrev64.32       q15, q1
    vmlal.u16       q14, d4, d1[1]
    vmlsl.u16       q14, d6, d1[2]
    vmlsl.u16       q14, d8, d1[3]
    vmlal.u16       q15, d5, d1[1]
    vmlsl.u16       q15, d7, d1[2]
    vmlsl.u16       q15, d9, d1[3]
.endm

.macro do_rgb_to_yuv_stage2
    vrshrn.u32      d20, q7, #16
    vrshrn.u32      d21, q8, #16
    vshrn.u32       d22, q9, #16
    vshrn.u32       d23, q13, #16
    vshrn.u32       d24, q14, #16
    vshrn.u32       d25, q15, #16
    vmovn.u16       d20, q10       /* d20 = y */
    vmovn.u16       d21, q11       /* d21 = u */
    vmovn.u16       d22, q12       /* d22 = v */
.endm

.macro do_rgb_to_yuv
    do_rgb_to_yuv_stage1
    do_rgb_to_yuv_stage2
.endm

.macro do_rgb_to_yuv_stage2_store_load_stage1
      vrshrn.u32      d20, q7, #16
      vrshrn.u32      d21, q8, #16
      vshrn.u32       d22, q9, #16
    vrev64.32       q9, q1
      vshrn.u32       d23, q13, #16
    vrev64.32       q13, q1
      vshrn.u32       d24, q14, #16
      vshrn.u32       d25, q15, #16
    do_load         \bpp, 8
      vmovn.u16       d20, q10     /* d20 = y */
    vmovl.u8        q2, d1\r_offs  /* r = { d4, d5 } */
      vmovn.u16       d21, q11     /* d21 = u */
    vmovl.u8        q3, d1\g_offs  /* g = { d6, d7 } */
      vmovn.u16       d22, q12     /* d22 = v */
    vmovl.u8        q4, d1\b_offs  /* b = { d8, d9 } */
    vmull.u16       q7, d4, d0[0]
    vmlal.u16       q7, d6, d0[1]
    vmlal.u16       q7, d8, d0[2]
      vst1.8          {d20}, [Y]!
    vmull.u16       q8, d5, d0[0]
    vmlal.u16       q8, d7, d0[1]
    vmlal.u16       q8, d9, d0[2]
    vmlsl.u16       q9, d4, d0[3]
    vmlsl.u16       q9, d6, d1[0]
    vmlal.u16       q9, d8, d1[1]
      vst1.8          {d21}, [U]!
    vmlsl.u16       q13, d5, d0[3]
    vmlsl.u16       q13, d7, d1[0]
    vmlal.u16       q13, d9, d1[1]
    vrev64.32       q14, q1
    vrev64.32       q15, q1
    vmlal.u16       q14, d4, d1[1]
    vmlsl.u16       q14, d6, d1[2]
    vmlsl.u16       q14, d8, d1[3]
      vst1.8          {d22}, [V]!
    vmlal.u16       q15, d5, d1[1]
    vmlsl.u16       q15, d7, d1[2]
    vmlsl.u16       q15, d9, d1[3]
.endm

.balign 16
jsimd_\colorid\()_ycc_neon_consts:
  .short 19595, 38470, 7471,  11059
  .short 21709, 32768, 27439, 5329
  .short 32767, 128,   32767, 128
  .short 32767, 128,   32767, 128

asm_function jsimd_\colorid\()_ycc_convert_neon
    OUTPUT_WIDTH    .req r0
    INPUT_BUF       .req r1
    OUTPUT_BUF      .req r2
    OUTPUT_ROW      .req r3
    NUM_ROWS        .req r4

    OUTPUT_BUF0     .req r5
    OUTPUT_BUF1     .req r6
    OUTPUT_BUF2     .req OUTPUT_BUF

    RGB             .req r7
    Y               .req r8
    U               .req r9
    V               .req r10
    N               .req ip

    /* Load constants to d0, d1, d2, d3 */
    adr             ip, jsimd_\colorid\()_ycc_neon_consts
    vld1.16         {d0, d1, d2, d3}, [ip, :128]

    /* Save ARM registers and handle input arguments */
    push            {r4, r5, r6, r7, r8, r9, r10, lr}
    ldr             NUM_ROWS, [sp, #(4 * 8)]
    ldr             OUTPUT_BUF0, [OUTPUT_BUF]
    ldr             OUTPUT_BUF1, [OUTPUT_BUF, #4]
    ldr             OUTPUT_BUF2, [OUTPUT_BUF, #8]
    .unreq          OUTPUT_BUF

    /* Save NEON registers */
    vpush           {d8-d15}

    /* Outer loop over scanlines */
    cmp             NUM_ROWS, #1
    blt             9f
0:
    ldr             Y, [OUTPUT_BUF0, OUTPUT_ROW, lsl #2]
    ldr             U, [OUTPUT_BUF1, OUTPUT_ROW, lsl #2]
    mov             N, OUTPUT_WIDTH
    ldr             V, [OUTPUT_BUF2, OUTPUT_ROW, lsl #2]
    add             OUTPUT_ROW, OUTPUT_ROW, #1
    ldr             RGB, [INPUT_BUF], #4

    /* Inner loop over pixels */
    subs            N, N, #8
    blt             3f
    do_load         \bpp, 8
    do_rgb_to_yuv_stage1
    subs            N, N, #8
    blt             2f
1:
    do_rgb_to_yuv_stage2_store_load_stage1
    subs            N, N, #8
    bge             1b
2:
    do_rgb_to_yuv_stage2
    do_store        8
    tst             N, #7
    beq             8f
3:
    tst             N, #4
    beq             3f
    do_load         \bpp, 4
3:
    tst             N, #2
    beq             4f
    do_load         \bpp, 2
4:
    tst             N, #1
    beq             5f
    do_load         \bpp, 1
5:
    do_rgb_to_yuv
    tst             N, #4
    beq             6f
    do_store        4
6:
    tst             N, #2
    beq             7f
    do_store        2
7:
    tst             N, #1
    beq             8f
    do_store        1
8:
    subs            NUM_ROWS, NUM_ROWS, #1
    bgt             0b
9:
    /* Restore all registers and return */
    vpop            {d8-d15}
    pop             {r4, r5, r6, r7, r8, r9, r10, pc}

    .unreq          OUTPUT_WIDTH
    .unreq          OUTPUT_ROW
    .unreq          INPUT_BUF
    .unreq          NUM_ROWS
    .unreq          OUTPUT_BUF0
    .unreq          OUTPUT_BUF1
    .unreq          OUTPUT_BUF2
    .unreq          RGB
    .unreq          Y
    .unreq          U
    .unreq          V
    .unreq          N

.purgem do_rgb_to_yuv
.purgem do_rgb_to_yuv_stage1
.purgem do_rgb_to_yuv_stage2
.purgem do_rgb_to_yuv_stage2_store_load_stage1

.endm

/*--------------------------------- id ----- bpp R  G  B */
generate_jsimd_rgb_ycc_convert_neon extrgb,  24, 0, 1, 2
generate_jsimd_rgb_ycc_convert_neon extbgr,  24, 2, 1, 0
generate_jsimd_rgb_ycc_convert_neon extrgbx, 32, 0, 1, 2
generate_jsimd_rgb_ycc_convert_neon extbgrx, 32, 2, 1, 0
generate_jsimd_rgb_ycc_convert_neon extxbgr, 32, 3, 2, 1
generate_jsimd_rgb_ycc_convert_neon extxrgb, 32, 1, 2, 3

.purgem do_load
.purgem do_store


/*****************************************************************************/

/*
 * Load data into workspace, applying unsigned->signed conversion
 *
 * TODO: can be combined with 'jsimd_fdct_ifast_neon' to get
 *       rid of VST1.16 instructions
 */

asm_function jsimd_convsamp_neon
    SAMPLE_DATA     .req r0
    START_COL       .req r1
    WORKSPACE       .req r2
    TMP1            .req r3
    TMP2            .req r4
    TMP3            .req r5
    TMP4            .req ip

    push            {r4, r5}
    vmov.u8         d0, #128

    ldmia           SAMPLE_DATA!, {TMP1, TMP2, TMP3, TMP4}
    add             TMP1, TMP1, START_COL
    add             TMP2, TMP2, START_COL
    add             TMP3, TMP3, START_COL
    add             TMP4, TMP4, START_COL
    vld1.8          {d16}, [TMP1]
    vsubl.u8        q8, d16, d0
    vld1.8          {d18}, [TMP2]
    vsubl.u8        q9, d18, d0
    vld1.8          {d20}, [TMP3]
    vsubl.u8        q10, d20, d0
    vld1.8          {d22}, [TMP4]
    ldmia           SAMPLE_DATA!, {TMP1, TMP2, TMP3, TMP4}
    vsubl.u8        q11, d22, d0
    vst1.16         {d16, d17, d18, d19}, [WORKSPACE, :128]!
    add             TMP1, TMP1, START_COL
    add             TMP2, TMP2, START_COL
    vst1.16         {d20, d21, d22, d23}, [WORKSPACE, :128]!
    add             TMP3, TMP3, START_COL
    add             TMP4, TMP4, START_COL
    vld1.8          {d24}, [TMP1]
    vsubl.u8        q12, d24, d0
    vld1.8          {d26}, [TMP2]
    vsubl.u8        q13, d26, d0
    vld1.8          {d28}, [TMP3]
    vsubl.u8        q14, d28, d0
    vld1.8          {d30}, [TMP4]
    vsubl.u8        q15, d30, d0
    vst1.16         {d24, d25, d26, d27}, [WORKSPACE, :128]!
    vst1.16         {d28, d29, d30, d31}, [WORKSPACE, :128]!
    pop             {r4, r5}
    bx              lr

    .unreq          SAMPLE_DATA
    .unreq          START_COL
    .unreq          WORKSPACE
    .unreq          TMP1
    .unreq          TMP2
    .unreq          TMP3
    .unreq          TMP4


/*****************************************************************************/

/*
 * jsimd_fdct_ifast_neon
 *
 * This function contains a fast, not so accurate integer implementation of
 * the forward DCT (Discrete Cosine Transform). It uses the same calculations
 * and produces exactly the same output as IJG's original 'jpeg_fdct_ifast'
 * function from jfdctfst.c
 *
 * TODO: can be combined with 'jsimd_convsamp_neon' to get
 *       rid of a bunch of VLD1.16 instructions
 */

#define XFIX_0_382683433 d0[0]
#define XFIX_0_541196100 d0[1]
#define XFIX_0_707106781 d0[2]
#define XFIX_1_306562965 d0[3]

.balign 16
jsimd_fdct_ifast_neon_consts:
  .short (98 * 128)               /* XFIX_0_382683433 */
  .short (139 * 128)              /* XFIX_0_541196100 */
  .short (181 * 128)              /* XFIX_0_707106781 */
  .short (334 * 128 - 256 * 128)  /* XFIX_1_306562965 */

asm_function jsimd_fdct_ifast_neon

    DATA            .req r0
    TMP             .req ip

    vpush           {d8-d15}

    /* Load constants */
    adr             TMP, jsimd_fdct_ifast_neon_consts
    vld1.16         {d0}, [TMP, :64]

    /* Load all DATA into NEON registers with the following allocation:
     *       0 1 2 3 | 4 5 6 7
     *      ---------+--------
     *   0 | d16     | d17    | q8
     *   1 | d18     | d19    | q9
     *   2 | d20     | d21    | q10
     *   3 | d22     | d23    | q11
     *   4 | d24     | d25    | q12
     *   5 | d26     | d27    | q13
     *   6 | d28     | d29    | q14
     *   7 | d30     | d31    | q15
     */

    vld1.16         {d16, d17, d18, d19}, [DATA, :128]!
    vld1.16         {d20, d21, d22, d23}, [DATA, :128]!
    vld1.16         {d24, d25, d26, d27}, [DATA, :128]!
    vld1.16         {d28, d29, d30, d31}, [DATA, :128]
    sub             DATA, DATA, #(128 - 32)

    mov             TMP, #2
1:
    /* Transpose */
    vtrn.16         q12, q13
    vtrn.16         q10, q11
    vtrn.16         q8, q9
    vtrn.16         q14, q15
    vtrn.32         q9, q11
    vtrn.32         q13, q15
    vtrn.32         q8, q10
    vtrn.32         q12, q14
    vswp            d30, d23
    vswp            d24, d17
    vswp            d26, d19
      /* 1-D FDCT */
      vadd.s16        q2, q11, q12
    vswp            d28, d21
      vsub.s16        q12, q11, q12
      vsub.s16        q6, q10, q13
      vadd.s16        q10, q10, q13
      vsub.s16        q7, q9, q14
      vadd.s16        q9, q9, q14
      vsub.s16        q1, q8, q15
      vadd.s16        q8, q8, q15
      vsub.s16        q4, q9, q10
      vsub.s16        q5, q8, q2
      vadd.s16        q3, q9, q10
      vadd.s16        q4, q4, q5
      vadd.s16        q2, q8, q2
      vqdmulh.s16     q4, q4, XFIX_0_707106781
      vadd.s16        q11, q12, q6
      vadd.s16        q8, q2, q3
      vsub.s16        q12, q2, q3
      vadd.s16        q3, q6, q7
      vadd.s16        q7, q7, q1
      vqdmulh.s16     q3, q3, XFIX_0_707106781
      vsub.s16        q6, q11, q7
      vadd.s16        q10, q5, q4
      vqdmulh.s16     q6, q6, XFIX_0_382683433
      vsub.s16        q14, q5, q4
      vqdmulh.s16     q11, q11, XFIX_0_541196100
      vqdmulh.s16     q5, q7, XFIX_1_306562965
      vadd.s16        q4, q1, q3
      vsub.s16        q3, q1, q3
      vadd.s16        q7, q7, q6
      vadd.s16        q11, q11, q6
      vadd.s16        q7, q7, q5
      vadd.s16        q13, q3, q11
      vsub.s16        q11, q3, q11
      vadd.s16        q9, q4, q7
      vsub.s16        q15, q4, q7
    subs            TMP, TMP, #1
    bne             1b

    /* store results */
    vst1.16         {d16, d17, d18, d19}, [DATA, :128]!
    vst1.16         {d20, d21, d22, d23}, [DATA, :128]!
    vst1.16         {d24, d25, d26, d27}, [DATA, :128]!
    vst1.16         {d28, d29, d30, d31}, [DATA, :128]

    vpop            {d8-d15}
    bx              lr

    .unreq          DATA
    .unreq          TMP


/*****************************************************************************/

/*
 * GLOBAL(void)
 * jsimd_quantize_neon (JCOEFPTR coef_block, DCTELEM *divisors,
 *                      DCTELEM *workspace);
 *
 * Note: the code uses 2 stage pipelining in order to improve instructions
 *       scheduling and eliminate stalls (this provides ~15% better
 *       performance for this function on both ARM Cortex-A8 and
 *       ARM Cortex-A9 when compared to the non-pipelined variant).
 *       The instructions which belong to the second stage use different
 *       indentation for better readiability.
 */
asm_function jsimd_quantize_neon

    COEF_BLOCK      .req r0
    DIVISORS        .req r1
    WORKSPACE       .req r2

    RECIPROCAL      .req DIVISORS
    CORRECTION      .req r3
    SHIFT           .req ip
    LOOP_COUNT      .req r4

    vld1.16         {d0, d1, d2, d3}, [WORKSPACE, :128]!
    vabs.s16        q12, q0
    add             CORRECTION, DIVISORS, #(64 * 2)
    add             SHIFT, DIVISORS, #(64 * 6)
    vld1.16         {d20, d21, d22, d23}, [CORRECTION, :128]!
    vabs.s16        q13, q1
    vld1.16         {d16, d17, d18, d19}, [RECIPROCAL, :128]!
    vadd.u16        q12, q12, q10  /* add correction */
    vadd.u16        q13, q13, q11
    vmull.u16       q10, d24, d16  /* multiply by reciprocal */
    vmull.u16       q11, d25, d17
    vmull.u16       q8, d26, d18
    vmull.u16       q9, d27, d19
    vld1.16         {d24, d25, d26, d27}, [SHIFT, :128]!
    vshrn.u32       d20, q10, #16
    vshrn.u32       d21, q11, #16
    vshrn.u32       d22, q8, #16
    vshrn.u32       d23, q9, #16
    vneg.s16        q12, q12
    vneg.s16        q13, q13
    vshr.s16        q2, q0, #15    /* extract sign */
    vshr.s16        q3, q1, #15
    vshl.u16        q14, q10, q12  /* shift */
    vshl.u16        q15, q11, q13

    push            {r4, r5}
    mov             LOOP_COUNT, #3
1:
    vld1.16         {d0, d1, d2, d3}, [WORKSPACE, :128]!
      veor.u16        q14, q14, q2  /* restore sign */
    vabs.s16        q12, q0
    vld1.16         {d20, d21, d22, d23}, [CORRECTION, :128]!
    vabs.s16        q13, q1
      veor.u16        q15, q15, q3
    vld1.16         {d16, d17, d18, d19}, [RECIPROCAL, :128]!
    vadd.u16        q12, q12, q10  /* add correction */
    vadd.u16        q13, q13, q11
    vmull.u16       q10, d24, d16  /* multiply by reciprocal */
    vmull.u16       q11, d25, d17
    vmull.u16       q8, d26, d18
    vmull.u16       q9, d27, d19
      vsub.u16        q14, q14, q2
    vld1.16         {d24, d25, d26, d27}, [SHIFT, :128]!
      vsub.u16        q15, q15, q3
    vshrn.u32       d20, q10, #16
    vshrn.u32       d21, q11, #16
      vst1.16         {d28, d29, d30, d31}, [COEF_BLOCK, :128]!
    vshrn.u32       d22, q8, #16
    vshrn.u32       d23, q9, #16
    vneg.s16        q12, q12
    vneg.s16        q13, q13
    vshr.s16        q2, q0, #15    /* extract sign */
    vshr.s16        q3, q1, #15
    vshl.u16        q14, q10, q12  /* shift */
    vshl.u16        q15, q11, q13
    subs            LOOP_COUNT, LOOP_COUNT, #1
    bne             1b
    pop             {r4, r5}

      veor.u16        q14, q14, q2  /* restore sign */
      veor.u16        q15, q15, q3
      vsub.u16        q14, q14, q2
      vsub.u16        q15, q15, q3
      vst1.16         {d28, d29, d30, d31}, [COEF_BLOCK, :128]!

    bx              lr  /* return */

    .unreq          COEF_BLOCK
    .unreq          DIVISORS
    .unreq          WORKSPACE
    .unreq          RECIPROCAL
    .unreq          CORRECTION
    .unreq          SHIFT
    .unreq          LOOP_COUNT


/*****************************************************************************/

/*
 * GLOBAL(void)
 * jsimd_h2v1_fancy_upsample_neon (int max_v_samp_factor,
 *                                 JDIMENSION downsampled_width,
 *                                 JSAMPARRAY input_data,
 *                                 JSAMPARRAY *output_data_ptr);
 *
 * Note: the use of unaligned writes is the main remaining bottleneck in
 *       this code, which can be potentially solved to get up to tens
 *       of percents performance improvement on Cortex-A8/Cortex-A9.
 */

/*
 * Upsample 16 source pixels to 32 destination pixels. The new 16 source
 * pixels are loaded to q0. The previous 16 source pixels are in q1. The
 * shifted-by-one source pixels are constructed in q2 by using q0 and q1.
 * Register d28 is used for multiplication by 3. Register q15 is used
 * for adding +1 bias.
 */
.macro upsample16 OUTPTR, INPTR
    vld1.8          {q0}, [\INPTR]!
    vmovl.u8        q8, d0
    vext.8          q2, q1, q0, #15
    vmovl.u8        q9, d1
    vaddw.u8        q10, q15, d4
    vaddw.u8        q11, q15, d5
    vmlal.u8        q8, d4, d28
    vmlal.u8        q9, d5, d28
    vmlal.u8        q10, d0, d28
    vmlal.u8        q11, d1, d28
    vmov            q1, q0        /* backup source pixels to q1 */
    vrshrn.u16      d6, q8, #2
    vrshrn.u16      d7, q9, #2
    vshrn.u16       d8, q10, #2
    vshrn.u16       d9, q11, #2
    vst2.8          {d6, d7, d8, d9}, [\OUTPTR]!
.endm

/*
 * Upsample 32 source pixels to 64 destination pixels. Compared to 'usample16'
 * macro, the roles of q0 and q1 registers are reversed for even and odd
 * groups of 16 pixels, that's why "vmov q1, q0" instructions are not needed.
 * Also this unrolling allows to reorder loads and stores to compensate
 * multiplication latency and reduce stalls.
 */
.macro upsample32 OUTPTR, INPTR
    /* even 16 pixels group */
    vld1.8          {q0}, [\INPTR]!
    vmovl.u8        q8, d0
    vext.8          q2, q1, q0, #15
    vmovl.u8        q9, d1
    vaddw.u8        q10, q15, d4
    vaddw.u8        q11, q15, d5
    vmlal.u8        q8, d4, d28
    vmlal.u8        q9, d5, d28
    vmlal.u8        q10, d0, d28
    vmlal.u8        q11, d1, d28
      /* odd 16 pixels group */
      vld1.8          {q1}, [\INPTR]!
    vrshrn.u16      d6, q8, #2
    vrshrn.u16      d7, q9, #2
    vshrn.u16       d8, q10, #2
    vshrn.u16       d9, q11, #2
      vmovl.u8        q8, d2
      vext.8          q2, q0, q1, #15
      vmovl.u8        q9, d3
      vaddw.u8        q10, q15, d4
      vaddw.u8        q11, q15, d5
      vmlal.u8        q8, d4, d28
      vmlal.u8        q9, d5, d28
      vmlal.u8        q10, d2, d28
      vmlal.u8        q11, d3, d28
    vst2.8          {d6, d7, d8, d9}, [\OUTPTR]!
      vrshrn.u16      d6, q8, #2
      vrshrn.u16      d7, q9, #2
      vshrn.u16       d8, q10, #2
      vshrn.u16       d9, q11, #2
      vst2.8          {d6, d7, d8, d9}, [\OUTPTR]!
.endm

/*
 * Upsample a row of WIDTH pixels from INPTR to OUTPTR.
 */
.macro upsample_row OUTPTR, INPTR, WIDTH, TMP1
    /* special case for the first and last pixels */
    sub             \WIDTH, \WIDTH, #1
    add             \OUTPTR, \OUTPTR, #1
    ldrb            \TMP1, [\INPTR, \WIDTH]
    strb            \TMP1, [\OUTPTR, \WIDTH, asl #1]
    ldrb            \TMP1, [\INPTR], #1
    strb            \TMP1, [\OUTPTR, #-1]
    vmov.8          d3[7], \TMP1

    subs            \WIDTH, \WIDTH, #32
    blt             5f
0:  /* process 32 pixels per iteration */
    upsample32      \OUTPTR, \INPTR
    subs            \WIDTH, \WIDTH, #32
    bge             0b
5:
    adds            \WIDTH, \WIDTH, #16
    blt             1f
0:  /* process 16 pixels if needed */
    upsample16      \OUTPTR, \INPTR
    subs            \WIDTH, \WIDTH, #16
1:
    adds            \WIDTH, \WIDTH, #16
    beq             9f

    /* load the remaining 1-15 pixels */
    add             \INPTR, \INPTR, \WIDTH
    tst             \WIDTH, #1
    beq             2f
    sub             \INPTR, \INPTR, #1
    vld1.8          {d0[0]}, [\INPTR]
2:
    tst             \WIDTH, #2
    beq             2f
    vext.8          d0, d0, d0, #6
    sub             \INPTR, \INPTR, #1
    vld1.8          {d0[1]}, [\INPTR]
    sub             \INPTR, \INPTR, #1
    vld1.8          {d0[0]}, [\INPTR]
2:
    tst             \WIDTH, #4
    beq             2f
    vrev64.32       d0, d0
    sub             \INPTR, \INPTR, #1
    vld1.8          {d0[3]}, [\INPTR]
    sub             \INPTR, \INPTR, #1
    vld1.8          {d0[2]}, [\INPTR]
    sub             \INPTR, \INPTR, #1
    vld1.8          {d0[1]}, [\INPTR]
    sub             \INPTR, \INPTR, #1
    vld1.8          {d0[0]}, [\INPTR]
2:
    tst             \WIDTH, #8
    beq             2f
    vmov            d1, d0
    sub             \INPTR, \INPTR, #8
    vld1.8          {d0}, [\INPTR]
2:  /* upsample the remaining pixels */
    vmovl.u8        q8, d0
    vext.8          q2, q1, q0, #15
    vmovl.u8        q9, d1
    vaddw.u8        q10, q15, d4
    vaddw.u8        q11, q15, d5
    vmlal.u8        q8, d4, d28
    vmlal.u8        q9, d5, d28
    vmlal.u8        q10, d0, d28
    vmlal.u8        q11, d1, d28
    vrshrn.u16      d10, q8, #2
    vrshrn.u16      d12, q9, #2
    vshrn.u16       d11, q10, #2
    vshrn.u16       d13, q11, #2
    vzip.8          d10, d11
    vzip.8          d12, d13
    /* store the remaining pixels */
    tst             \WIDTH, #8
    beq             2f
    vst1.8          {d10, d11}, [\OUTPTR]!
    vmov            q5, q6
2:
    tst             \WIDTH, #4
    beq             2f
    vst1.8          {d10}, [\OUTPTR]!
    vmov            d10, d11
2:
    tst             \WIDTH, #2
    beq             2f
    vst1.8          {d10[0]}, [\OUTPTR]!
    vst1.8          {d10[1]}, [\OUTPTR]!
    vst1.8          {d10[2]}, [\OUTPTR]!
    vst1.8          {d10[3]}, [\OUTPTR]!
    vext.8          d10, d10, d10, #4
2:
    tst             \WIDTH, #1
    beq             2f
    vst1.8          {d10[0]}, [\OUTPTR]!
    vst1.8          {d10[1]}, [\OUTPTR]!
2:
9:
.endm

asm_function jsimd_h2v1_fancy_upsample_neon

    MAX_V_SAMP_FACTOR .req r0
    DOWNSAMPLED_WIDTH .req r1
    INPUT_DATA        .req r2
    OUTPUT_DATA_PTR   .req r3
    OUTPUT_DATA       .req OUTPUT_DATA_PTR

    OUTPTR            .req r4
    INPTR             .req r5
    WIDTH             .req ip
    TMP               .req lr

    push            {r4, r5, r6, lr}
    vpush           {d8-d15}

    ldr             OUTPUT_DATA, [OUTPUT_DATA_PTR]
    cmp             MAX_V_SAMP_FACTOR, #0
    ble             99f

    /* initialize constants */
    vmov.u8         d28, #3
    vmov.u16        q15, #1
11:
    ldr             INPTR, [INPUT_DATA], #4
    ldr             OUTPTR, [OUTPUT_DATA], #4
    mov             WIDTH, DOWNSAMPLED_WIDTH
    upsample_row    OUTPTR, INPTR, WIDTH, TMP
    subs            MAX_V_SAMP_FACTOR, MAX_V_SAMP_FACTOR, #1
    bgt             11b

99:
    vpop            {d8-d15}
    pop             {r4, r5, r6, pc}

    .unreq          MAX_V_SAMP_FACTOR
    .unreq          DOWNSAMPLED_WIDTH
    .unreq          INPUT_DATA
    .unreq          OUTPUT_DATA_PTR
    .unreq          OUTPUT_DATA

    .unreq          OUTPTR
    .unreq          INPTR
    .unreq          WIDTH
    .unreq          TMP

.purgem upsample16
.purgem upsample32
.purgem upsample_row


/*****************************************************************************/

/*
 * GLOBAL(JOCTET*)
 * jsimd_huff_encode_one_block (working_state *state, JOCTET *buffer,
 *                              JCOEFPTR block, int last_dc_val,
 *                              c_derived_tbl *dctbl, c_derived_tbl *actbl)
 *
 */

.macro emit_byte BUFFER, PUT_BUFFER, PUT_BITS, ZERO, TMP
    sub             \PUT_BITS, \PUT_BITS, #0x8
    lsr             \TMP, \PUT_BUFFER, \PUT_BITS
    uxtb            \TMP, \TMP
    strb            \TMP, [\BUFFER, #1]!
    cmp             \TMP, #0xff
    /*it eq*/
    strbeq          \ZERO, [\BUFFER, #1]!
.endm

.macro put_bits PUT_BUFFER, PUT_BITS, CODE, SIZE
    /*lsl             \PUT_BUFFER, \PUT_BUFFER, \SIZE*/
    add             \PUT_BITS, \SIZE
    /*orr             \PUT_BUFFER, \PUT_BUFFER, \CODE*/
    orr             \PUT_BUFFER, \CODE, \PUT_BUFFER, lsl \SIZE
.endm

.macro checkbuf15 BUFFER, PUT_BUFFER, PUT_BITS, ZERO, TMP
  cmp               \PUT_BITS, #0x10
  blt               15f
    eor               \ZERO, \ZERO, \ZERO
    emit_byte         \BUFFER, \PUT_BUFFER, \PUT_BITS, \ZERO, \TMP
    emit_byte         \BUFFER, \PUT_BUFFER, \PUT_BITS, \ZERO, \TMP
15:
.endm

.balign 16
jsimd_huff_encode_one_block_neon_consts:
  .byte 0x01
  .byte 0x02
  .byte 0x04
  .byte 0x08
  .byte 0x10
  .byte 0x20
  .byte 0x40
  .byte 0x80

asm_function jsimd_huff_encode_one_block_neon
    push            {r4, r5, r6, r7, r8, r9, r10, r11, lr}
    add             r7, sp, #0x1c
    sub             r4, sp, #0x40
    bfc             r4, #0, #5
    mov             sp, r4           /* align sp on 32 bytes */
    vst1.64         {d8, d9, d10, d11}, [r4, :128]!
    vst1.64         {d12, d13, d14, d15}, [r4, :128]
    sub             sp, #0x140       /* reserve 320 bytes */
    str             r0, [sp, #0x18]  /* working state > sp + Ox18 */
    add             r4, sp, #0x20    /* r4 = t1 */
    ldr             lr, [r7, #0x8]   /* lr = dctbl */
    sub             r10, r1, #0x1    /* r10=buffer-- */
    ldrsh           r1, [r2]
    mov             r9, #0x10
    mov             r8, #0x1
    adr             r5, jsimd_huff_encode_one_block_neon_consts
    /* prepare data */
    vld1.8          {d26}, [r5, :64]
    veor            q8, q8, q8
    veor            q9, q9, q9
    vdup.16         q14, r9
    vdup.16         q15, r8
    veor            q10, q10, q10
    veor            q11, q11, q11
    sub             r1, r1, r3
    add             r9, r2, #0x22
    add             r8, r2, #0x18
    add             r3, r2, #0x36
    vmov.16         d0[0], r1
    vld1.16         {d2[0]}, [r9, :16]
    vld1.16         {d4[0]}, [r8, :16]
    vld1.16         {d6[0]}, [r3, :16]
    add             r1, r2, #0x2
    add             r9, r2, #0x30
    add             r8, r2, #0x26
    add             r3, r2, #0x28
    vld1.16         {d0[1]}, [r1, :16]
    vld1.16         {d2[1]}, [r9, :16]
    vld1.16         {d4[1]}, [r8, :16]
    vld1.16         {d6[1]}, [r3, :16]
    add             r1, r2, #0x10
    add             r9, r2, #0x40
    add             r8, r2, #0x34
    add             r3, r2, #0x1a
    vld1.16         {d0[2]}, [r1, :16]
    vld1.16         {d2[2]}, [r9, :16]
    vld1.16         {d4[2]}, [r8, :16]
    vld1.16         {d6[2]}, [r3, :16]
    add             r1, r2, #0x20
    add             r9, r2, #0x32
    add             r8, r2, #0x42
    add             r3, r2, #0xc
    vld1.16         {d0[3]}, [r1, :16]
    vld1.16         {d2[3]}, [r9, :16]
    vld1.16         {d4[3]}, [r8, :16]
    vld1.16         {d6[3]}, [r3, :16]
    add             r1, r2, #0x12
    add             r9, r2, #0x24
    add             r8, r2, #0x50
    add             r3, r2, #0xe
    vld1.16         {d1[0]}, [r1, :16]
    vld1.16         {d3[0]}, [r9, :16]
    vld1.16         {d5[0]}, [r8, :16]
    vld1.16         {d7[0]}, [r3, :16]
    add             r1, r2, #0x4
    add             r9, r2, #0x16
    add             r8, r2, #0x60
    add             r3, r2, #0x1c
    vld1.16         {d1[1]}, [r1, :16]
    vld1.16         {d3[1]}, [r9, :16]
    vld1.16         {d5[1]}, [r8, :16]
    vld1.16         {d7[1]}, [r3, :16]
    add             r1, r2, #0x6
    add             r9, r2, #0x8
    add             r8, r2, #0x52
    add             r3, r2, #0x2a
    vld1.16         {d1[2]}, [r1, :16]
    vld1.16         {d3[2]}, [r9, :16]
    vld1.16         {d5[2]}, [r8, :16]
    vld1.16         {d7[2]}, [r3, :16]
    add             r1, r2, #0x14
    add             r9, r2, #0xa
    add             r8, r2, #0x44
    add             r3, r2, #0x38
    vld1.16         {d1[3]}, [r1, :16]
    vld1.16         {d3[3]}, [r9, :16]
    vld1.16         {d5[3]}, [r8, :16]
    vld1.16         {d7[3]}, [r3, :16]
    vcgt.s16        q8, q8, q0
    vcgt.s16        q9, q9, q1
    vcgt.s16        q10, q10, q2
    vcgt.s16        q11, q11, q3
    vabs.s16        q0, q0
    vabs.s16        q1, q1
    vabs.s16        q2, q2
    vabs.s16        q3, q3
    veor            q8, q8, q0
    veor            q9, q9, q1
    veor            q10, q10, q2
    veor            q11, q11, q3
    add             r9, r4, #0x20
    add             r8, r4, #0x80
    add             r3, r4, #0xa0
    vclz.i16        q0, q0
    vclz.i16        q1, q1
    vclz.i16        q2, q2
    vclz.i16        q3, q3
    vsub.i16        q0, q14, q0
    vsub.i16        q1, q14, q1
    vsub.i16        q2, q14, q2
    vsub.i16        q3, q14, q3
    vst1.16         {d0, d1, d2, d3}, [r4, :256]
    vst1.16         {d4, d5, d6, d7}, [r9, :256]
    vshl.s16        q0, q15, q0
    vshl.s16        q1, q15, q1
    vshl.s16        q2, q15, q2
    vshl.s16        q3, q15, q3
    vsub.i16        q0, q0, q15
    vsub.i16        q1, q1, q15
    vsub.i16        q2, q2, q15
    vsub.i16        q3, q3, q15
    vand            q8, q8, q0
    vand            q9, q9, q1
    vand            q10, q10, q2
    vand            q11, q11, q3
    vst1.16         {d16, d17, d18, d19}, [r8, :256]
    vst1.16         {d20, d21, d22, d23}, [r3, :256]
    add             r1, r2, #0x46
    add             r9, r2, #0x3a
    add             r8, r2, #0x74
    add             r3, r2, #0x6a
    vld1.16         {d8[0]}, [r1, :16]
    vld1.16         {d10[0]}, [r9, :16]
    vld1.16         {d12[0]}, [r8, :16]
    vld1.16         {d14[0]}, [r3, :16]
    veor            q8, q8, q8
    veor            q9, q9, q9
    veor            q10, q10, q10
    veor            q11, q11, q11
    add             r1, r2, #0x54
    add             r9, r2, #0x2c
    add             r8, r2, #0x76
    add             r3, r2, #0x78
    vld1.16         {d8[1]}, [r1, :16]
    vld1.16         {d10[1]}, [r9, :16]
    vld1.16         {d12[1]}, [r8, :16]
    vld1.16         {d14[1]}, [r3, :16]
    add             r1, r2, #0x62
    add             r9, r2, #0x1e
    add             r8, r2, #0x68
    add             r3, r2, #0x7a
    vld1.16         {d8[2]}, [r1, :16]
    vld1.16         {d10[2]}, [r9, :16]
    vld1.16         {d12[2]}, [r8, :16]
    vld1.16         {d14[2]}, [r3, :16]
    add             r1, r2, #0x70
    add             r9, r2, #0x2e
    add             r8, r2, #0x5a
    add             r3, r2, #0x6c
    vld1.16         {d8[3]}, [r1, :16]
    vld1.16         {d10[3]}, [r9, :16]
    vld1.16         {d12[3]}, [r8, :16]
    vld1.16         {d14[3]}, [r3, :16]
    add             r1, r2, #0x72
    add             r9, r2, #0x3c
    add             r8, r2, #0x4c
    add             r3, r2, #0x5e
    vld1.16         {d9[0]}, [r1, :16]
    vld1.16         {d11[0]}, [r9, :16]
    vld1.16         {d13[0]}, [r8, :16]
    vld1.16         {d15[0]}, [r3, :16]
    add             r1, r2, #0x64
    add             r9, r2, #0x4a
    add             r8, r2, #0x3e
    add             r3, r2, #0x6e
    vld1.16         {d9[1]}, [r1, :16]
    vld1.16         {d11[1]}, [r9, :16]
    vld1.16         {d13[1]}, [r8, :16]
    vld1.16         {d15[1]}, [r3, :16]
    add             r1, r2, #0x56
    add             r9, r2, #0x58
    add             r8, r2, #0x4e
    add             r3, r2, #0x7c
    vld1.16         {d9[2]}, [r1, :16]
    vld1.16         {d11[2]}, [r9, :16]
    vld1.16         {d13[2]}, [r8, :16]
    vld1.16         {d15[2]}, [r3, :16]
    add             r1, r2, #0x48
    add             r9, r2, #0x66
    add             r8, r2, #0x5c
    add             r3, r2, #0x7e
    vld1.16         {d9[3]}, [r1, :16]
    vld1.16         {d11[3]}, [r9, :16]
    vld1.16         {d13[3]}, [r8, :16]
    vld1.16         {d15[3]}, [r3, :16]
    vcgt.s16        q8, q8, q4
    vcgt.s16        q9, q9, q5
    vcgt.s16        q10, q10, q6
    vcgt.s16        q11, q11, q7
    vabs.s16        q4, q4
    vabs.s16        q5, q5
    vabs.s16        q6, q6
    vabs.s16        q7, q7
    veor            q8, q8, q4
    veor            q9, q9, q5
    veor            q10, q10, q6
    veor            q11, q11, q7
    add             r1, r4, #0x40
    add             r9, r4, #0x60
    add             r8, r4, #0xc0
    add             r3, r4, #0xe0
    vclz.i16        q4, q4
    vclz.i16        q5, q5
    vclz.i16        q6, q6
    vclz.i16        q7, q7
    vsub.i16        q4, q14, q4
    vsub.i16        q5, q14, q5
    vsub.i16        q6, q14, q6
    vsub.i16        q7, q14, q7
    vst1.16         {d8, d9, d10, d11}, [r1, :256]
    vst1.16         {d12, d13, d14, d15}, [r9, :256]
    vshl.s16        q4, q15, q4
    vshl.s16        q5, q15, q5
    vshl.s16        q6, q15, q6
    vshl.s16        q7, q15, q7
    vsub.i16        q4, q4, q15
    vsub.i16        q5, q5, q15
    vsub.i16        q6, q6, q15
    vsub.i16        q7, q7, q15
    vand            q8, q8, q4
    vand            q9, q9, q5
    vand            q10, q10, q6
    vand            q11, q11, q7
    vst1.16         {d16, d17, d18, d19}, [r8, :256]
    vst1.16         {d20, d21, d22, d23}, [r3, :256]
    ldr             r12, [r7, #0xc]       /* r12 = actbl */
    add             r1, lr, #0x400        /* r1 = dctbl->ehufsi */
    mov             r9, r12               /* r9 = actbl */
    add             r6, r4, #0x80         /* r6 = t2 */
    ldr             r11, [r0, #0x8]       /* r11 = put_buffer */
    ldr             r4, [r0, #0xc]        /* r4  = put_bits */
    ldrh            r2, [r6, #-128]       /* r2  = nbits */
    ldrh            r3, [r6]              /* r3  = temp2 & (((JLONG) 1)<<nbits) - 1; */
    ldr             r0, [lr, r2, lsl #2]
    ldrb            r5, [r1, r2]
    put_bits        r11, r4, r0, r5
    checkbuf15      r10, r11, r4, r5, r0
    put_bits        r11, r4, r3, r2
    checkbuf15      r10, r11, r4, r5, r0
    mov             lr, r6                /* lr = t2 */
    add             r5, r9, #0x400        /* r5 = actbl->ehufsi */
    ldrsb           r6, [r5, #0xf0]       /* r6 = actbl->ehufsi[0xf0] */
    veor            q8, q8, q8
    vceq.i16        q0, q0, q8
    vceq.i16        q1, q1, q8
    vceq.i16        q2, q2, q8
    vceq.i16        q3, q3, q8
    vceq.i16        q4, q4, q8
    vceq.i16        q5, q5, q8
    vceq.i16        q6, q6, q8
    vceq.i16        q7, q7, q8
    vmovn.i16       d0, q0
    vmovn.i16       d2, q1
    vmovn.i16       d4, q2
    vmovn.i16       d6, q3
    vmovn.i16       d8, q4
    vmovn.i16       d10, q5
    vmovn.i16       d12, q6
    vmovn.i16       d14, q7
    vand            d0, d0, d26
    vand            d2, d2, d26
    vand            d4, d4, d26
    vand            d6, d6, d26
    vand            d8, d8, d26
    vand            d10, d10, d26
    vand            d12, d12, d26
    vand            d14, d14, d26
    vpadd.i8        d0, d0, d2
    vpadd.i8        d4, d4, d6
    vpadd.i8        d8, d8, d10
    vpadd.i8        d12, d12, d14
    vpadd.i8        d0, d0, d4
    vpadd.i8        d8, d8, d12
    vpadd.i8        d0, d0, d8
    vmov.32         r1, d0[1]
    vmov.32         r8, d0[0]
    mvn             r1, r1
    mvn             r8, r8
    lsrs            r1, r1, #0x1
    rrx             r8, r8            /* shift in last r1 bit while shifting out DC bit */
    rbit            r1, r1            /* r1 = index1 */
    rbit            r8, r8            /* r8 = index0 */
    ldr             r0, [r9, #0x3c0]  /* r0 = actbl->ehufco[0xf0] */
    str             r1, [sp, #0x14]   /* index1 > sp + 0x14 */
    cmp             r8, #0x0
    beq             6f
1:
    clz             r2, r8
    add             lr, lr, r2, lsl #1
    lsl             r8, r8, r2
    ldrh            r1, [lr, #-126]
2:
    cmp             r2, #0x10
    blt             3f
    sub             r2, r2, #0x10
    put_bits        r11, r4, r0, r6
    cmp             r4, #0x10
    blt             2b
    eor             r3, r3, r3
    emit_byte       r10, r11, r4, r3, r12
    emit_byte       r10, r11, r4, r3, r12
    b               2b
3:
    add             r2, r1, r2, lsl #4
    ldrh            r3, [lr, #2]!
    ldr             r12, [r9, r2, lsl #2]
    ldrb            r2, [r5, r2]
    put_bits        r11, r4, r12, r2
    checkbuf15      r10, r11, r4, r2, r12
    put_bits        r11, r4, r3, r1
    checkbuf15      r10, r11, r4, r2, r12
    lsls            r8, r8, #0x1
    bne             1b
6:
    add             r12, sp, #0x20   /* r12 = t1 */
    ldr             r8, [sp, #0x14]  /* r8 = index1 */
    adds            r12, #0xc0       /* r12 = t2 + (DCTSIZE2/2) */
    cmp             r8, #0x0
    beq             6f
    clz             r2, r8
    sub             r12, r12, lr
    lsl             r8, r8, r2
    add             r2, r2, r12, lsr #1
    add             lr, lr, r2, lsl #1
    b               7f
1:
    clz             r2, r8
    add             lr, lr, r2, lsl #1
    lsl             r8, r8, r2
7:
    ldrh            r1, [lr, #-126]
2:
    cmp             r2, #0x10
    blt             3f
    sub             r2, r2, #0x10
    put_bits        r11, r4, r0, r6
    cmp             r4, #0x10
    blt             2b
    eor             r3, r3, r3
    emit_byte       r10, r11, r4, r3, r12
    emit_byte       r10, r11, r4, r3, r12
    b               2b
3:
    add             r2, r1, r2, lsl #4
    ldrh            r3, [lr, #2]!
    ldr             r12, [r9, r2, lsl #2]
    ldrb            r2, [r5, r2]
    put_bits        r11, r4, r12, r2
    checkbuf15      r10, r11, r4, r2, r12
    put_bits        r11, r4, r3, r1
    checkbuf15      r10, r11, r4, r2, r12
    lsls            r8, r8, #0x1
    bne             1b
6:
    add             r0, sp, #0x20
    add             r0, #0xfe
    cmp             lr, r0
    bhs             1f
    ldr             r1, [r9]
    ldrb            r0, [r5]
    put_bits        r11, r4, r1, r0
    checkbuf15      r10, r11, r4, r0, r1
1:
    ldr             r12, [sp, #0x18]
    str             r11, [r12, #0x8]
    str             r4, [r12, #0xc]
    add             r0, r10, #0x1
    add             r4, sp, #0x140
    vld1.64         {d8, d9, d10, d11}, [r4, :128]!
    vld1.64         {d12, d13, d14, d15}, [r4, :128]
    sub             r4, r7, #0x1c
    mov             sp, r4
    pop             {r4, r5, r6, r7, r8, r9, r10, r11, pc}

.purgem emit_byte
.purgem put_bits
.purgem checkbuf15