summaryrefslogtreecommitdiffstats
path: root/include/import/xxhash.h
blob: 7c3c3fcbabbb05341b525558a8773090a2813d36 (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
2879
2880
2881
2882
2883
2884
2885
2886
2887
2888
2889
2890
2891
2892
2893
2894
2895
2896
2897
2898
2899
2900
2901
2902
2903
2904
2905
2906
2907
2908
2909
2910
2911
2912
2913
2914
2915
2916
2917
2918
2919
2920
2921
2922
2923
2924
2925
2926
2927
2928
2929
2930
2931
2932
2933
2934
2935
2936
2937
2938
2939
2940
2941
2942
2943
2944
2945
2946
2947
2948
2949
2950
2951
2952
2953
2954
2955
2956
2957
2958
2959
2960
2961
2962
2963
2964
2965
2966
2967
2968
2969
2970
2971
2972
2973
2974
2975
2976
2977
2978
2979
2980
2981
2982
2983
2984
2985
2986
2987
2988
2989
2990
2991
2992
2993
2994
2995
2996
2997
2998
2999
3000
3001
3002
3003
3004
3005
3006
3007
3008
3009
3010
3011
3012
3013
3014
3015
3016
3017
3018
3019
3020
3021
3022
3023
3024
3025
3026
3027
3028
3029
3030
3031
3032
3033
3034
3035
3036
3037
3038
3039
3040
3041
3042
3043
3044
3045
3046
3047
3048
3049
3050
3051
3052
3053
3054
3055
3056
3057
3058
3059
3060
3061
3062
3063
3064
3065
3066
3067
3068
3069
3070
3071
3072
3073
3074
3075
3076
3077
3078
3079
3080
3081
3082
3083
3084
3085
3086
3087
3088
3089
3090
3091
3092
3093
3094
3095
3096
3097
3098
3099
3100
3101
3102
3103
3104
3105
3106
3107
3108
3109
3110
3111
3112
3113
3114
3115
3116
3117
3118
3119
3120
3121
3122
3123
3124
3125
3126
3127
3128
3129
3130
3131
3132
3133
3134
3135
3136
3137
3138
3139
3140
3141
3142
3143
3144
3145
3146
3147
3148
3149
3150
3151
3152
3153
3154
3155
3156
3157
3158
3159
3160
3161
3162
3163
3164
3165
3166
3167
3168
3169
3170
3171
3172
3173
3174
3175
3176
3177
3178
3179
3180
3181
3182
3183
3184
3185
3186
3187
3188
3189
3190
3191
3192
3193
3194
3195
3196
3197
3198
3199
3200
3201
3202
3203
3204
3205
3206
3207
3208
3209
3210
3211
3212
3213
3214
3215
3216
3217
3218
3219
3220
3221
3222
3223
3224
3225
3226
3227
3228
3229
3230
3231
3232
3233
3234
3235
3236
3237
3238
3239
3240
3241
3242
3243
3244
3245
3246
3247
3248
3249
3250
3251
3252
3253
3254
3255
3256
3257
3258
3259
3260
3261
3262
3263
3264
3265
3266
3267
3268
3269
3270
3271
3272
3273
3274
3275
3276
3277
3278
3279
3280
3281
3282
3283
3284
3285
3286
3287
3288
3289
3290
3291
3292
3293
3294
3295
3296
3297
3298
3299
3300
3301
3302
3303
3304
3305
3306
3307
3308
3309
3310
3311
3312
3313
3314
3315
3316
3317
3318
3319
3320
3321
3322
3323
3324
3325
3326
3327
3328
3329
3330
3331
3332
3333
3334
3335
3336
3337
3338
3339
3340
3341
3342
3343
3344
3345
3346
3347
3348
3349
3350
3351
3352
3353
3354
3355
3356
3357
3358
3359
3360
3361
3362
3363
3364
3365
3366
3367
3368
3369
3370
3371
3372
3373
3374
3375
3376
3377
3378
3379
3380
3381
3382
3383
3384
3385
3386
3387
3388
3389
3390
3391
3392
3393
3394
3395
3396
3397
3398
3399
3400
3401
3402
3403
3404
3405
3406
3407
3408
3409
3410
3411
3412
3413
3414
3415
3416
3417
3418
3419
3420
3421
3422
3423
3424
3425
3426
3427
3428
3429
3430
3431
3432
3433
3434
3435
3436
3437
3438
3439
3440
3441
3442
3443
3444
3445
3446
3447
3448
3449
3450
3451
3452
3453
3454
3455
3456
3457
3458
3459
3460
3461
3462
3463
3464
3465
3466
3467
3468
3469
3470
3471
3472
3473
3474
3475
3476
3477
3478
3479
3480
3481
3482
3483
3484
3485
3486
3487
3488
3489
3490
3491
3492
3493
3494
3495
3496
3497
3498
3499
3500
3501
3502
3503
3504
3505
3506
3507
3508
3509
3510
3511
3512
3513
3514
3515
3516
3517
3518
3519
3520
3521
3522
3523
3524
3525
3526
3527
3528
3529
3530
3531
3532
3533
3534
3535
3536
3537
3538
3539
3540
3541
3542
3543
3544
3545
3546
3547
3548
3549
3550
3551
3552
3553
3554
3555
3556
3557
3558
3559
3560
3561
3562
3563
3564
3565
3566
3567
3568
3569
3570
3571
3572
3573
3574
3575
3576
3577
3578
3579
3580
3581
3582
3583
3584
3585
3586
3587
3588
3589
3590
3591
3592
3593
3594
3595
3596
3597
3598
3599
3600
3601
3602
3603
3604
3605
3606
3607
3608
3609
3610
3611
3612
3613
3614
3615
3616
3617
3618
3619
3620
3621
3622
3623
3624
3625
3626
3627
3628
3629
3630
3631
3632
3633
3634
3635
3636
3637
3638
3639
3640
3641
3642
3643
3644
3645
3646
3647
3648
3649
3650
3651
3652
3653
3654
3655
3656
3657
3658
3659
3660
3661
3662
3663
3664
3665
3666
3667
3668
3669
3670
3671
3672
3673
3674
3675
3676
3677
3678
3679
3680
3681
3682
3683
3684
3685
3686
3687
3688
3689
3690
3691
3692
3693
3694
3695
3696
3697
3698
3699
3700
3701
3702
3703
3704
3705
3706
3707
3708
3709
3710
3711
3712
3713
3714
3715
3716
3717
3718
3719
3720
3721
3722
3723
3724
3725
3726
3727
3728
3729
3730
3731
3732
3733
3734
3735
3736
3737
3738
3739
3740
3741
3742
3743
3744
3745
3746
3747
3748
3749
3750
3751
3752
3753
3754
3755
3756
3757
3758
3759
3760
3761
3762
3763
3764
3765
3766
3767
3768
3769
3770
3771
3772
3773
3774
3775
3776
3777
3778
3779
3780
3781
3782
3783
3784
3785
3786
3787
3788
3789
3790
3791
3792
3793
3794
3795
3796
3797
3798
3799
3800
3801
3802
3803
3804
3805
3806
3807
3808
3809
3810
3811
3812
3813
3814
3815
3816
3817
3818
3819
3820
3821
3822
3823
3824
3825
3826
3827
3828
3829
3830
3831
3832
3833
3834
3835
3836
3837
3838
3839
3840
3841
3842
3843
3844
3845
3846
3847
3848
3849
3850
3851
3852
3853
3854
3855
3856
3857
3858
3859
3860
3861
3862
3863
3864
3865
3866
3867
3868
3869
3870
3871
3872
3873
3874
3875
3876
3877
3878
3879
3880
3881
3882
3883
3884
3885
3886
3887
3888
3889
3890
3891
3892
3893
3894
3895
3896
3897
3898
3899
3900
3901
3902
3903
3904
3905
3906
3907
3908
3909
3910
3911
3912
3913
3914
3915
3916
3917
3918
3919
3920
3921
3922
3923
3924
3925
3926
3927
3928
3929
3930
3931
3932
3933
3934
3935
3936
3937
3938
3939
3940
3941
3942
3943
3944
3945
3946
3947
3948
3949
3950
3951
3952
3953
3954
3955
3956
3957
3958
3959
3960
3961
3962
3963
3964
3965
3966
3967
3968
3969
3970
3971
3972
3973
3974
3975
3976
3977
3978
3979
3980
3981
3982
3983
3984
3985
3986
3987
3988
3989
3990
3991
3992
3993
3994
3995
3996
3997
3998
3999
4000
4001
4002
4003
4004
4005
4006
4007
4008
4009
4010
4011
4012
4013
4014
4015
4016
4017
4018
4019
4020
4021
4022
4023
4024
4025
4026
4027
4028
4029
4030
4031
4032
4033
4034
4035
4036
4037
4038
4039
4040
4041
4042
4043
4044
4045
4046
4047
4048
4049
4050
4051
4052
4053
4054
4055
4056
4057
4058
4059
4060
4061
4062
4063
4064
4065
4066
4067
4068
4069
4070
4071
4072
4073
4074
4075
4076
4077
4078
4079
4080
4081
4082
4083
4084
4085
4086
4087
4088
4089
4090
4091
4092
4093
4094
4095
4096
4097
4098
4099
4100
4101
4102
4103
4104
4105
4106
4107
4108
4109
4110
4111
4112
4113
4114
4115
4116
4117
4118
4119
4120
4121
4122
4123
4124
4125
4126
4127
4128
4129
4130
4131
4132
4133
4134
4135
4136
4137
4138
4139
4140
4141
4142
4143
4144
4145
4146
4147
4148
4149
4150
4151
4152
4153
4154
4155
4156
4157
4158
4159
4160
4161
4162
4163
4164
4165
4166
4167
4168
4169
4170
4171
4172
4173
4174
4175
4176
4177
4178
4179
4180
4181
4182
4183
4184
4185
4186
4187
4188
4189
4190
4191
4192
4193
4194
4195
4196
4197
4198
4199
4200
4201
4202
4203
4204
4205
4206
4207
4208
4209
4210
4211
4212
4213
4214
4215
4216
4217
4218
4219
4220
4221
4222
4223
4224
4225
4226
4227
4228
4229
4230
4231
4232
4233
4234
4235
4236
4237
4238
4239
4240
4241
4242
4243
4244
4245
4246
4247
4248
4249
4250
4251
4252
4253
4254
4255
4256
4257
4258
4259
4260
4261
4262
4263
4264
4265
4266
4267
4268
4269
4270
4271
4272
4273
4274
4275
4276
4277
4278
4279
4280
4281
4282
4283
4284
4285
4286
4287
4288
4289
4290
4291
4292
4293
4294
4295
4296
4297
4298
4299
4300
4301
4302
4303
4304
4305
4306
4307
4308
4309
4310
4311
4312
4313
4314
4315
4316
4317
4318
4319
4320
4321
4322
4323
4324
4325
4326
4327
4328
4329
4330
4331
4332
4333
4334
4335
4336
4337
4338
4339
4340
4341
4342
4343
4344
4345
4346
4347
4348
4349
4350
4351
4352
4353
4354
4355
4356
4357
4358
4359
4360
4361
4362
4363
4364
4365
4366
4367
4368
4369
4370
4371
4372
4373
4374
4375
4376
4377
4378
4379
4380
4381
4382
4383
4384
4385
4386
4387
4388
4389
4390
4391
4392
4393
4394
4395
4396
4397
4398
4399
4400
4401
4402
4403
4404
4405
4406
4407
4408
4409
4410
4411
4412
4413
4414
4415
4416
4417
4418
4419
4420
4421
4422
4423
4424
4425
4426
4427
4428
4429
4430
4431
4432
4433
4434
4435
4436
4437
4438
4439
4440
4441
4442
4443
4444
4445
4446
4447
4448
4449
4450
4451
4452
4453
4454
4455
4456
4457
4458
4459
4460
4461
4462
4463
4464
4465
4466
4467
4468
4469
4470
4471
4472
4473
4474
4475
4476
4477
4478
4479
4480
4481
4482
4483
4484
4485
4486
4487
4488
4489
4490
4491
4492
4493
4494
4495
4496
4497
4498
4499
4500
4501
4502
4503
4504
4505
4506
4507
4508
4509
4510
4511
4512
4513
4514
4515
4516
4517
4518
4519
4520
4521
4522
4523
4524
4525
4526
4527
4528
4529
4530
4531
4532
4533
4534
4535
4536
4537
4538
4539
4540
4541
4542
4543
4544
4545
4546
4547
4548
4549
4550
4551
4552
4553
4554
4555
4556
4557
4558
4559
4560
4561
4562
4563
4564
4565
4566
4567
4568
4569
4570
4571
4572
4573
4574
4575
4576
4577
4578
4579
4580
4581
4582
4583
4584
4585
4586
4587
4588
4589
4590
4591
4592
4593
4594
4595
4596
4597
4598
4599
4600
4601
4602
4603
4604
4605
4606
4607
4608
4609
4610
4611
4612
4613
4614
4615
4616
4617
4618
4619
4620
4621
4622
4623
4624
4625
4626
4627
4628
4629
4630
4631
4632
4633
4634
4635
4636
4637
4638
4639
4640
4641
4642
4643
4644
4645
4646
4647
4648
4649
4650
4651
4652
4653
4654
4655
4656
4657
4658
4659
4660
4661
4662
4663
4664
4665
4666
4667
4668
4669
4670
4671
4672
4673
4674
4675
4676
4677
4678
4679
4680
4681
4682
4683
4684
4685
4686
4687
4688
4689
4690
4691
4692
4693
4694
4695
4696
4697
4698
4699
4700
4701
4702
4703
4704
4705
4706
4707
4708
4709
4710
4711
4712
4713
4714
4715
4716
4717
4718
4719
4720
4721
4722
4723
4724
4725
4726
4727
4728
4729
4730
4731
4732
4733
4734
4735
4736
4737
4738
4739
4740
4741
4742
4743
4744
4745
4746
4747
4748
4749
4750
4751
4752
4753
4754
4755
4756
4757
4758
4759
4760
4761
4762
4763
4764
4765
4766
4767
4768
4769
4770
4771
4772
4773
4774
4775
4776
4777
4778
4779
4780
4781
4782
4783
4784
4785
4786
4787
4788
4789
4790
4791
4792
4793
4794
4795
4796
4797
4798
4799
4800
4801
4802
4803
4804
4805
4806
4807
4808
4809
4810
4811
4812
4813
4814
4815
4816
4817
4818
4819
4820
4821
4822
4823
4824
4825
4826
4827
4828
4829
4830
4831
4832
4833
4834
4835
4836
4837
4838
4839
4840
4841
4842
4843
4844
4845
4846
4847
4848
4849
4850
4851
4852
4853
4854
4855
4856
4857
4858
4859
4860
4861
4862
4863
4864
4865
4866
4867
4868
4869
4870
4871
4872
4873
4874
4875
4876
4877
4878
4879
4880
4881
4882
4883
4884
4885
4886
4887
4888
4889
4890
4891
4892
4893
4894
4895
4896
4897
4898
4899
4900
4901
4902
4903
4904
4905
4906
4907
4908
4909
4910
4911
4912
4913
4914
4915
4916
4917
4918
4919
4920
4921
4922
4923
4924
4925
4926
4927
4928
4929
4930
4931
4932
4933
4934
4935
4936
4937
4938
4939
4940
4941
4942
4943
4944
4945
4946
4947
4948
4949
4950
4951
4952
4953
4954
4955
4956
4957
4958
4959
4960
4961
4962
4963
4964
4965
4966
4967
4968
4969
4970
4971
4972
4973
4974
4975
4976
4977
4978
4979
4980
4981
4982
4983
4984
4985
4986
4987
4988
4989
4990
4991
4992
4993
4994
4995
4996
4997
4998
4999
5000
5001
5002
5003
5004
5005
5006
5007
5008
5009
5010
5011
5012
5013
5014
5015
5016
5017
5018
5019
5020
5021
5022
5023
5024
5025
5026
5027
5028
5029
5030
5031
5032
5033
5034
5035
5036
5037
5038
5039
5040
5041
5042
5043
5044
5045
5046
5047
5048
5049
5050
5051
5052
5053
5054
5055
5056
5057
5058
5059
5060
5061
5062
5063
5064
5065
5066
5067
5068
5069
5070
5071
5072
5073
5074
5075
5076
5077
5078
5079
5080
5081
5082
5083
5084
5085
5086
5087
5088
5089
5090
5091
5092
5093
5094
5095
5096
5097
5098
5099
5100
5101
5102
5103
5104
5105
5106
5107
5108
5109
5110
5111
5112
5113
5114
5115
5116
5117
5118
5119
5120
5121
5122
5123
5124
5125
5126
5127
5128
5129
5130
5131
5132
5133
5134
5135
5136
5137
5138
5139
5140
5141
5142
5143
5144
5145
5146
5147
5148
5149
5150
5151
5152
5153
5154
5155
5156
5157
5158
5159
5160
5161
5162
5163
5164
5165
5166
5167
5168
5169
5170
5171
5172
5173
5174
5175
5176
5177
5178
5179
5180
5181
5182
5183
5184
5185
5186
5187
5188
5189
5190
5191
5192
5193
5194
5195
5196
5197
5198
5199
5200
5201
5202
5203
5204
5205
5206
5207
5208
5209
5210
5211
5212
5213
5214
5215
5216
5217
5218
5219
5220
5221
5222
5223
5224
5225
5226
5227
5228
5229
5230
5231
5232
5233
5234
5235
5236
5237
5238
5239
5240
5241
5242
5243
5244
5245
5246
5247
5248
5249
5250
5251
5252
5253
5254
5255
5256
5257
5258
5259
5260
5261
5262
5263
5264
5265
5266
5267
5268
5269
5270
5271
5272
5273
5274
5275
5276
5277
5278
5279
5280
5281
5282
5283
5284
5285
5286
5287
5288
5289
5290
5291
5292
5293
5294
5295
5296
5297
5298
5299
5300
5301
5302
5303
5304
5305
5306
5307
5308
5309
5310
5311
5312
5313
5314
5315
5316
5317
5318
5319
5320
5321
5322
5323
5324
5325
5326
5327
5328
5329
5330
5331
5332
5333
5334
5335
5336
5337
5338
5339
5340
5341
5342
5343
5344
5345
5346
5347
5348
5349
5350
5351
5352
5353
5354
5355
5356
5357
5358
5359
5360
5361
5362
5363
5364
5365
5366
5367
5368
5369
5370
5371
5372
5373
5374
5375
5376
5377
5378
5379
5380
5381
5382
5383
5384
5385
5386
5387
5388
5389
5390
5391
5392
5393
5394
5395
5396
5397
5398
5399
5400
5401
5402
5403
5404
5405
5406
5407
5408
5409
5410
5411
5412
5413
5414
5415
5416
5417
5418
5419
5420
5421
5422
5423
5424
5425
5426
5427
5428
5429
5430
5431
5432
5433
5434
5435
5436
5437
5438
5439
5440
5441
5442
5443
5444
5445
5446
5447
5448
5449
5450
5451
5452
5453
5454
5455
5456
5457
5458
5459
5460
5461
5462
5463
5464
5465
5466
5467
5468
5469
5470
5471
5472
5473
5474
5475
5476
5477
5478
5479
5480
5481
5482
5483
5484
5485
5486
5487
5488
5489
5490
5491
5492
5493
5494
5495
5496
5497
5498
5499
5500
5501
5502
5503
5504
5505
5506
5507
5508
5509
5510
5511
5512
5513
5514
5515
5516
5517
5518
5519
5520
5521
5522
5523
5524
5525
5526
5527
5528
5529
5530
5531
5532
5533
5534
5535
5536
5537
5538
5539
5540
5541
5542
5543
5544
5545
5546
5547
5548
5549
5550
5551
5552
5553
5554
5555
5556
5557
5558
5559
5560
5561
5562
5563
5564
5565
5566
5567
5568
5569
5570
5571
5572
5573
5574
5575
5576
5577
5578
5579
5580
5581
5582
5583
5584
5585
5586
5587
5588
5589
5590
5591
5592
5593
5594
5595
5596
5597
5598
5599
5600
5601
5602
5603
5604
5605
5606
5607
5608
5609
5610
5611
5612
5613
5614
5615
5616
5617
5618
5619
5620
5621
5622
5623
5624
5625
5626
5627
5628
5629
5630
5631
5632
5633
5634
5635
5636
5637
5638
5639
5640
5641
5642
5643
5644
5645
5646
5647
5648
5649
5650
5651
5652
5653
5654
5655
5656
5657
5658
5659
5660
5661
5662
5663
5664
5665
5666
5667
5668
5669
5670
5671
5672
5673
5674
5675
5676
5677
5678
5679
5680
5681
5682
5683
5684
5685
5686
5687
5688
5689
5690
5691
5692
5693
5694
5695
5696
5697
5698
5699
5700
5701
5702
5703
5704
5705
5706
5707
5708
5709
5710
5711
5712
5713
5714
5715
5716
5717
5718
5719
5720
5721
5722
5723
5724
5725
5726
5727
5728
5729
5730
5731
5732
5733
5734
5735
5736
5737
5738
5739
5740
5741
5742
5743
5744
5745
5746
5747
5748
5749
5750
5751
5752
5753
5754
5755
5756
5757
5758
5759
5760
5761
5762
5763
5764
5765
5766
5767
5768
5769
5770
5771
5772
5773
5774
5775
5776
5777
5778
5779
5780
5781
5782
5783
5784
5785
5786
5787
5788
5789
5790
5791
5792
5793
5794
5795
5796
5797
5798
5799
5800
5801
5802
5803
5804
5805
5806
5807
5808
5809
5810
5811
5812
5813
5814
5815
5816
5817
5818
5819
5820
5821
5822
5823
5824
5825
5826
5827
5828
5829
5830
5831
5832
5833
5834
5835
5836
5837
5838
5839
5840
5841
5842
5843
5844
5845
5846
5847
5848
5849
5850
5851
5852
5853
5854
5855
5856
5857
5858
5859
5860
5861
5862
5863
5864
5865
5866
5867
5868
5869
5870
5871
5872
5873
5874
5875
5876
5877
5878
5879
5880
5881
5882
5883
5884
5885
5886
5887
5888
5889
5890
5891
5892
5893
5894
5895
5896
5897
5898
5899
5900
5901
5902
5903
5904
5905
5906
5907
5908
5909
5910
5911
5912
5913
5914
5915
5916
5917
5918
5919
5920
5921
5922
5923
5924
5925
5926
5927
5928
5929
5930
5931
5932
5933
5934
5935
5936
5937
5938
5939
5940
5941
5942
5943
5944
5945
5946
5947
5948
5949
5950
5951
5952
5953
5954
5955
5956
5957
5958
5959
5960
5961
5962
5963
5964
5965
5966
5967
5968
5969
5970
5971
5972
5973
5974
5975
5976
5977
5978
5979
5980
5981
5982
5983
5984
5985
5986
5987
5988
5989
5990
5991
5992
5993
5994
5995
5996
5997
5998
5999
6000
6001
6002
6003
6004
6005
6006
6007
6008
6009
6010
6011
6012
6013
6014
6015
6016
6017
6018
6019
6020
6021
6022
6023
6024
6025
6026
6027
6028
6029
6030
6031
6032
6033
6034
6035
6036
6037
6038
6039
6040
6041
6042
6043
6044
6045
6046
6047
6048
6049
6050
6051
6052
6053
6054
6055
6056
6057
6058
6059
6060
6061
6062
6063
6064
6065
6066
6067
6068
6069
6070
6071
6072
6073
6074
6075
6076
6077
6078
6079
6080
6081
6082
6083
6084
6085
6086
6087
6088
6089
6090
6091
6092
6093
6094
6095
6096
6097
6098
6099
6100
6101
6102
6103
6104
6105
6106
6107
6108
6109
6110
6111
6112
6113
6114
6115
6116
6117
6118
6119
6120
6121
6122
6123
6124
6125
6126
6127
6128
6129
6130
6131
6132
6133
6134
6135
6136
6137
6138
6139
6140
6141
6142
6143
6144
6145
6146
6147
6148
6149
6150
6151
6152
6153
6154
6155
6156
6157
6158
6159
6160
6161
6162
6163
6164
6165
6166
6167
6168
6169
6170
6171
6172
6173
6174
6175
6176
6177
6178
6179
6180
6181
6182
6183
6184
6185
6186
6187
6188
6189
6190
6191
6192
6193
6194
6195
6196
6197
6198
6199
6200
6201
6202
6203
6204
6205
6206
6207
6208
6209
6210
6211
6212
6213
6214
6215
6216
6217
6218
6219
6220
6221
6222
6223
6224
6225
6226
6227
6228
6229
6230
6231
6232
6233
6234
6235
6236
6237
6238
6239
6240
6241
6242
6243
6244
6245
6246
6247
6248
6249
6250
6251
6252
6253
6254
6255
6256
6257
6258
6259
6260
6261
6262
6263
6264
6265
6266
6267
6268
6269
6270
6271
6272
6273
6274
6275
6276
6277
6278
6279
6280
6281
6282
6283
6284
6285
6286
6287
6288
6289
6290
6291
6292
6293
6294
6295
6296
6297
6298
6299
6300
6301
6302
6303
6304
6305
6306
6307
6308
6309
6310
6311
6312
6313
6314
6315
6316
6317
6318
6319
6320
6321
6322
6323
6324
6325
6326
6327
6328
6329
6330
6331
6332
6333
6334
6335
6336
6337
6338
6339
6340
6341
6342
6343
6344
6345
6346
6347
6348
6349
6350
6351
6352
6353
6354
6355
6356
6357
6358
6359
6360
6361
6362
6363
6364
6365
6366
6367
6368
6369
6370
6371
6372
6373
6374
6375
6376
6377
6378
6379
6380
6381
6382
6383
6384
6385
6386
6387
6388
6389
6390
6391
6392
6393
6394
6395
6396
6397
6398
6399
6400
6401
6402
6403
6404
6405
6406
6407
6408
6409
6410
6411
6412
6413
6414
6415
6416
6417
6418
6419
6420
6421
6422
6423
6424
6425
6426
6427
6428
6429
6430
6431
6432
6433
6434
6435
6436
6437
6438
6439
6440
6441
6442
6443
6444
6445
6446
6447
6448
6449
6450
6451
6452
6453
6454
6455
6456
6457
6458
6459
6460
6461
6462
6463
6464
6465
6466
6467
6468
6469
6470
6471
6472
6473
6474
6475
6476
6477
6478
6479
6480
6481
6482
6483
6484
6485
6486
6487
6488
6489
6490
6491
6492
6493
6494
6495
6496
6497
6498
6499
6500
6501
6502
6503
6504
6505
6506
6507
6508
6509
6510
6511
6512
6513
6514
6515
6516
6517
6518
6519
6520
6521
6522
6523
6524
6525
6526
6527
6528
6529
6530
6531
6532
6533
6534
6535
6536
6537
6538
6539
6540
6541
6542
6543
6544
6545
6546
6547
6548
6549
6550
6551
6552
6553
6554
6555
6556
6557
6558
6559
6560
6561
6562
6563
6564
6565
6566
6567
6568
6569
6570
6571
6572
6573
6574
6575
6576
6577
6578
6579
6580
6581
6582
6583
6584
6585
6586
6587
6588
6589
6590
6591
6592
6593
6594
6595
6596
6597
6598
6599
6600
6601
6602
6603
6604
6605
6606
6607
6608
6609
6610
6611
6612
6613
6614
6615
6616
6617
6618
6619
6620
6621
6622
6623
6624
6625
6626
6627
6628
6629
6630
6631
6632
6633
6634
6635
6636
6637
6638
6639
6640
6641
6642
6643
6644
6645
6646
6647
6648
6649
6650
6651
6652
6653
6654
6655
6656
6657
6658
6659
6660
6661
6662
6663
6664
6665
6666
6667
6668
6669
6670
6671
6672
6673
6674
6675
6676
6677
6678
6679
6680
6681
6682
6683
6684
6685
6686
6687
6688
6689
6690
6691
6692
6693
6694
6695
6696
6697
6698
6699
6700
6701
6702
6703
6704
6705
6706
6707
6708
6709
6710
6711
6712
6713
6714
6715
6716
6717
6718
6719
6720
6721
6722
6723
6724
6725
6726
6727
6728
6729
6730
6731
6732
6733
6734
6735
6736
6737
6738
6739
6740
6741
6742
6743
6744
6745
6746
6747
6748
6749
6750
6751
6752
6753
6754
6755
6756
6757
6758
6759
6760
6761
6762
6763
6764
6765
6766
6767
6768
6769
6770
6771
6772
6773
/*
 * xxHash - Extremely Fast Hash algorithm
 * Header File
 * Copyright (C) 2012-2021 Yann Collet
 *
 * BSD 2-Clause License (https://www.opensource.org/licenses/bsd-license.php)
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions are
 * met:
 *
 *    * Redistributions of source code must retain the above copyright
 *      notice, this list of conditions and the following disclaimer.
 *    * Redistributions in binary form must reproduce the above
 *      copyright notice, this list of conditions and the following disclaimer
 *      in the documentation and/or other materials provided with the
 *      distribution.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *
 * You can contact the author at:
 *   - xxHash homepage: https://www.xxhash.com
 *   - xxHash source repository: https://github.com/Cyan4973/xxHash
 */

/*!
 * @mainpage xxHash
 *
 * xxHash is an extremely fast non-cryptographic hash algorithm, working at RAM speed
 * limits.
 *
 * It is proposed in four flavors, in three families:
 * 1. @ref XXH32_family
 *   - Classic 32-bit hash function. Simple, compact, and runs on almost all
 *     32-bit and 64-bit systems.
 * 2. @ref XXH64_family
 *   - Classic 64-bit adaptation of XXH32. Just as simple, and runs well on most
 *     64-bit systems (but _not_ 32-bit systems).
 * 3. @ref XXH3_family
 *   - Modern 64-bit and 128-bit hash function family which features improved
 *     strength and performance across the board, especially on smaller data.
 *     It benefits greatly from SIMD and 64-bit without requiring it.
 *
 * Benchmarks
 * ---
 * The reference system uses an Intel i7-9700K CPU, and runs Ubuntu x64 20.04.
 * The open source benchmark program is compiled with clang v10.0 using -O3 flag.
 *
 * | Hash Name            | ISA ext | Width | Large Data Speed | Small Data Velocity |
 * | -------------------- | ------- | ----: | ---------------: | ------------------: |
 * | XXH3_64bits()        | @b AVX2 |    64 |        59.4 GB/s |               133.1 |
 * | MeowHash             | AES-NI  |   128 |        58.2 GB/s |                52.5 |
 * | XXH3_128bits()       | @b AVX2 |   128 |        57.9 GB/s |               118.1 |
 * | CLHash               | PCLMUL  |    64 |        37.1 GB/s |                58.1 |
 * | XXH3_64bits()        | @b SSE2 |    64 |        31.5 GB/s |               133.1 |
 * | XXH3_128bits()       | @b SSE2 |   128 |        29.6 GB/s |               118.1 |
 * | RAM sequential read  |         |   N/A |        28.0 GB/s |                 N/A |
 * | ahash                | AES-NI  |    64 |        22.5 GB/s |               107.2 |
 * | City64               |         |    64 |        22.0 GB/s |                76.6 |
 * | T1ha2                |         |    64 |        22.0 GB/s |                99.0 |
 * | City128              |         |   128 |        21.7 GB/s |                57.7 |
 * | FarmHash             | AES-NI  |    64 |        21.3 GB/s |                71.9 |
 * | XXH64()              |         |    64 |        19.4 GB/s |                71.0 |
 * | SpookyHash           |         |    64 |        19.3 GB/s |                53.2 |
 * | Mum                  |         |    64 |        18.0 GB/s |                67.0 |
 * | CRC32C               | SSE4.2  |    32 |        13.0 GB/s |                57.9 |
 * | XXH32()              |         |    32 |         9.7 GB/s |                71.9 |
 * | City32               |         |    32 |         9.1 GB/s |                66.0 |
 * | Blake3*              | @b AVX2 |   256 |         4.4 GB/s |                 8.1 |
 * | Murmur3              |         |    32 |         3.9 GB/s |                56.1 |
 * | SipHash*             |         |    64 |         3.0 GB/s |                43.2 |
 * | Blake3*              | @b SSE2 |   256 |         2.4 GB/s |                 8.1 |
 * | HighwayHash          |         |    64 |         1.4 GB/s |                 6.0 |
 * | FNV64                |         |    64 |         1.2 GB/s |                62.7 |
 * | Blake2*              |         |   256 |         1.1 GB/s |                 5.1 |
 * | SHA1*                |         |   160 |         0.8 GB/s |                 5.6 |
 * | MD5*                 |         |   128 |         0.6 GB/s |                 7.8 |
 * @note
 *   - Hashes which require a specific ISA extension are noted. SSE2 is also noted,
 *     even though it is mandatory on x64.
 *   - Hashes with an asterisk are cryptographic. Note that MD5 is non-cryptographic
 *     by modern standards.
 *   - Small data velocity is a rough average of algorithm's efficiency for small
 *     data. For more accurate information, see the wiki.
 *   - More benchmarks and strength tests are found on the wiki:
 *         https://github.com/Cyan4973/xxHash/wiki
 *
 * Usage
 * ------
 * All xxHash variants use a similar API. Changing the algorithm is a trivial
 * substitution.
 *
 * @pre
 *    For functions which take an input and length parameter, the following
 *    requirements are assumed:
 *    - The range from [`input`, `input + length`) is valid, readable memory.
 *      - The only exception is if the `length` is `0`, `input` may be `NULL`.
 *    - For C++, the objects must have the *TriviallyCopyable* property, as the
 *      functions access bytes directly as if it was an array of `unsigned char`.
 *
 * @anchor single_shot_example
 * **Single Shot**
 *
 * These functions are stateless functions which hash a contiguous block of memory,
 * immediately returning the result. They are the easiest and usually the fastest
 * option.
 *
 * XXH32(), XXH64(), XXH3_64bits(), XXH3_128bits()
 *
 * @code{.c}
 *   #include <string.h>
 *   #include "xxhash.h"
 *
 *   // Example for a function which hashes a null terminated string with XXH32().
 *   XXH32_hash_t hash_string(const char* string, XXH32_hash_t seed)
 *   {
 *       // NULL pointers are only valid if the length is zero
 *       size_t length = (string == NULL) ? 0 : strlen(string);
 *       return XXH32(string, length, seed);
 *   }
 * @endcode
 *
 * @anchor streaming_example
 * **Streaming**
 *
 * These groups of functions allow incremental hashing of unknown size, even
 * more than what would fit in a size_t.
 *
 * XXH32_reset(), XXH64_reset(), XXH3_64bits_reset(), XXH3_128bits_reset()
 *
 * @code{.c}
 *   #include <stdio.h>
 *   #include <assert.h>
 *   #include "xxhash.h"
 *   // Example for a function which hashes a FILE incrementally with XXH3_64bits().
 *   XXH64_hash_t hashFile(FILE* f)
 *   {
 *       // Allocate a state struct. Do not just use malloc() or new.
 *       XXH3_state_t* state = XXH3_createState();
 *       assert(state != NULL && "Out of memory!");
 *       // Reset the state to start a new hashing session.
 *       XXH3_64bits_reset(state);
 *       char buffer[4096];
 *       size_t count;
 *       // Read the file in chunks
 *       while ((count = fread(buffer, 1, sizeof(buffer), f)) != 0) {
 *           // Run update() as many times as necessary to process the data
 *           XXH3_64bits_update(state, buffer, count);
 *       }
 *       // Retrieve the finalized hash. This will not change the state.
 *       XXH64_hash_t result = XXH3_64bits_digest(state);
 *       // Free the state. Do not use free().
 *       XXH3_freeState(state);
 *       return result;
 *   }
 * @endcode
 *
 * @file xxhash.h
 * xxHash prototypes and implementation
 */

#if defined (__cplusplus)
extern "C" {
#endif

/* ****************************
 *  INLINE mode
 ******************************/
/*!
 * @defgroup public Public API
 * Contains details on the public xxHash functions.
 * @{
 */
#ifdef XXH_DOXYGEN
/*!
 * @brief Gives access to internal state declaration, required for static allocation.
 *
 * Incompatible with dynamic linking, due to risks of ABI changes.
 *
 * Usage:
 * @code{.c}
 *     #define XXH_STATIC_LINKING_ONLY
 *     #include "xxhash.h"
 * @endcode
 */
#  define XXH_STATIC_LINKING_ONLY
/* Do not undef XXH_STATIC_LINKING_ONLY for Doxygen */

/*!
 * @brief Gives access to internal definitions.
 *
 * Usage:
 * @code{.c}
 *     #define XXH_STATIC_LINKING_ONLY
 *     #define XXH_IMPLEMENTATION
 *     #include "xxhash.h"
 * @endcode
 */
#  define XXH_IMPLEMENTATION
/* Do not undef XXH_IMPLEMENTATION for Doxygen */

/*!
 * @brief Exposes the implementation and marks all functions as `inline`.
 *
 * Use these build macros to inline xxhash into the target unit.
 * Inlining improves performance on small inputs, especially when the length is
 * expressed as a compile-time constant:
 *
 *  https://fastcompression.blogspot.com/2018/03/xxhash-for-small-keys-impressive-power.html
 *
 * It also keeps xxHash symbols private to the unit, so they are not exported.
 *
 * Usage:
 * @code{.c}
 *     #define XXH_INLINE_ALL
 *     #include "xxhash.h"
 * @endcode
 * Do not compile and link xxhash.o as a separate object, as it is not useful.
 */
#  define XXH_INLINE_ALL
#  undef XXH_INLINE_ALL
/*!
 * @brief Exposes the implementation without marking functions as inline.
 */
#  define XXH_PRIVATE_API
#  undef XXH_PRIVATE_API
/*!
 * @brief Emulate a namespace by transparently prefixing all symbols.
 *
 * If you want to include _and expose_ xxHash functions from within your own
 * library, but also want to avoid symbol collisions with other libraries which
 * may also include xxHash, you can use @ref XXH_NAMESPACE to automatically prefix
 * any public symbol from xxhash library with the value of @ref XXH_NAMESPACE
 * (therefore, avoid empty or numeric values).
 *
 * Note that no change is required within the calling program as long as it
 * includes `xxhash.h`: Regular symbol names will be automatically translated
 * by this header.
 */
#  define XXH_NAMESPACE /* YOUR NAME HERE */
#  undef XXH_NAMESPACE
#endif

#if (defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API)) \
    && !defined(XXH_INLINE_ALL_31684351384)
   /* this section should be traversed only once */
#  define XXH_INLINE_ALL_31684351384
   /* give access to the advanced API, required to compile implementations */
#  undef XXH_STATIC_LINKING_ONLY   /* avoid macro redef */
#  define XXH_STATIC_LINKING_ONLY
   /* make all functions private */
#  undef XXH_PUBLIC_API
#  if defined(__GNUC__)
#    define XXH_PUBLIC_API static __inline __attribute__((unused))
#  elif defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
#    define XXH_PUBLIC_API static inline
#  elif defined(_MSC_VER)
#    define XXH_PUBLIC_API static __inline
#  else
     /* note: this version may generate warnings for unused static functions */
#    define XXH_PUBLIC_API static
#  endif

   /*
    * This part deals with the special case where a unit wants to inline xxHash,
    * but "xxhash.h" has previously been included without XXH_INLINE_ALL,
    * such as part of some previously included *.h header file.
    * Without further action, the new include would just be ignored,
    * and functions would effectively _not_ be inlined (silent failure).
    * The following macros solve this situation by prefixing all inlined names,
    * avoiding naming collision with previous inclusions.
    */
   /* Before that, we unconditionally #undef all symbols,
    * in case they were already defined with XXH_NAMESPACE.
    * They will then be redefined for XXH_INLINE_ALL
    */
#  undef XXH_versionNumber
    /* XXH32 */
#  undef XXH32
#  undef XXH32_createState
#  undef XXH32_freeState
#  undef XXH32_reset
#  undef XXH32_update
#  undef XXH32_digest
#  undef XXH32_copyState
#  undef XXH32_canonicalFromHash
#  undef XXH32_hashFromCanonical
    /* XXH64 */
#  undef XXH64
#  undef XXH64_createState
#  undef XXH64_freeState
#  undef XXH64_reset
#  undef XXH64_update
#  undef XXH64_digest
#  undef XXH64_copyState
#  undef XXH64_canonicalFromHash
#  undef XXH64_hashFromCanonical
    /* XXH3_64bits */
#  undef XXH3_64bits
#  undef XXH3_64bits_withSecret
#  undef XXH3_64bits_withSeed
#  undef XXH3_64bits_withSecretandSeed
#  undef XXH3_createState
#  undef XXH3_freeState
#  undef XXH3_copyState
#  undef XXH3_64bits_reset
#  undef XXH3_64bits_reset_withSeed
#  undef XXH3_64bits_reset_withSecret
#  undef XXH3_64bits_update
#  undef XXH3_64bits_digest
#  undef XXH3_generateSecret
    /* XXH3_128bits */
#  undef XXH128
#  undef XXH3_128bits
#  undef XXH3_128bits_withSeed
#  undef XXH3_128bits_withSecret
#  undef XXH3_128bits_reset
#  undef XXH3_128bits_reset_withSeed
#  undef XXH3_128bits_reset_withSecret
#  undef XXH3_128bits_reset_withSecretandSeed
#  undef XXH3_128bits_update
#  undef XXH3_128bits_digest
#  undef XXH128_isEqual
#  undef XXH128_cmp
#  undef XXH128_canonicalFromHash
#  undef XXH128_hashFromCanonical
    /* Finally, free the namespace itself */
#  undef XXH_NAMESPACE

    /* employ the namespace for XXH_INLINE_ALL */
#  define XXH_NAMESPACE XXH_INLINE_
   /*
    * Some identifiers (enums, type names) are not symbols,
    * but they must nonetheless be renamed to avoid redeclaration.
    * Alternative solution: do not redeclare them.
    * However, this requires some #ifdefs, and has a more dispersed impact.
    * Meanwhile, renaming can be achieved in a single place.
    */
#  define XXH_IPREF(Id)   XXH_NAMESPACE ## Id
#  define XXH_OK XXH_IPREF(XXH_OK)
#  define XXH_ERROR XXH_IPREF(XXH_ERROR)
#  define XXH_errorcode XXH_IPREF(XXH_errorcode)
#  define XXH32_canonical_t  XXH_IPREF(XXH32_canonical_t)
#  define XXH64_canonical_t  XXH_IPREF(XXH64_canonical_t)
#  define XXH128_canonical_t XXH_IPREF(XXH128_canonical_t)
#  define XXH32_state_s XXH_IPREF(XXH32_state_s)
#  define XXH32_state_t XXH_IPREF(XXH32_state_t)
#  define XXH64_state_s XXH_IPREF(XXH64_state_s)
#  define XXH64_state_t XXH_IPREF(XXH64_state_t)
#  define XXH3_state_s  XXH_IPREF(XXH3_state_s)
#  define XXH3_state_t  XXH_IPREF(XXH3_state_t)
#  define XXH128_hash_t XXH_IPREF(XXH128_hash_t)
   /* Ensure the header is parsed again, even if it was previously included */
#  undef XXHASH_H_5627135585666179
#  undef XXHASH_H_STATIC_13879238742
#endif /* XXH_INLINE_ALL || XXH_PRIVATE_API */

/* ****************************************************************
 *  Stable API
 *****************************************************************/
#ifndef XXHASH_H_5627135585666179
#define XXHASH_H_5627135585666179 1

/*! @brief Marks a global symbol. */
#if !defined(XXH_INLINE_ALL) && !defined(XXH_PRIVATE_API)
#  if defined(WIN32) && defined(_MSC_VER) && (defined(XXH_IMPORT) || defined(XXH_EXPORT))
#    ifdef XXH_EXPORT
#      define XXH_PUBLIC_API __declspec(dllexport)
#    elif XXH_IMPORT
#      define XXH_PUBLIC_API __declspec(dllimport)
#    endif
#  else
#    define XXH_PUBLIC_API   /* do nothing */
#  endif
#endif

#ifdef XXH_NAMESPACE
#  define XXH_CAT(A,B) A##B
#  define XXH_NAME2(A,B) XXH_CAT(A,B)
#  define XXH_versionNumber XXH_NAME2(XXH_NAMESPACE, XXH_versionNumber)
/* XXH32 */
#  define XXH32 XXH_NAME2(XXH_NAMESPACE, XXH32)
#  define XXH32_createState XXH_NAME2(XXH_NAMESPACE, XXH32_createState)
#  define XXH32_freeState XXH_NAME2(XXH_NAMESPACE, XXH32_freeState)
#  define XXH32_reset XXH_NAME2(XXH_NAMESPACE, XXH32_reset)
#  define XXH32_update XXH_NAME2(XXH_NAMESPACE, XXH32_update)
#  define XXH32_digest XXH_NAME2(XXH_NAMESPACE, XXH32_digest)
#  define XXH32_copyState XXH_NAME2(XXH_NAMESPACE, XXH32_copyState)
#  define XXH32_canonicalFromHash XXH_NAME2(XXH_NAMESPACE, XXH32_canonicalFromHash)
#  define XXH32_hashFromCanonical XXH_NAME2(XXH_NAMESPACE, XXH32_hashFromCanonical)
/* XXH64 */
#  define XXH64 XXH_NAME2(XXH_NAMESPACE, XXH64)
#  define XXH64_createState XXH_NAME2(XXH_NAMESPACE, XXH64_createState)
#  define XXH64_freeState XXH_NAME2(XXH_NAMESPACE, XXH64_freeState)
#  define XXH64_reset XXH_NAME2(XXH_NAMESPACE, XXH64_reset)
#  define XXH64_update XXH_NAME2(XXH_NAMESPACE, XXH64_update)
#  define XXH64_digest XXH_NAME2(XXH_NAMESPACE, XXH64_digest)
#  define XXH64_copyState XXH_NAME2(XXH_NAMESPACE, XXH64_copyState)
#  define XXH64_canonicalFromHash XXH_NAME2(XXH_NAMESPACE, XXH64_canonicalFromHash)
#  define XXH64_hashFromCanonical XXH_NAME2(XXH_NAMESPACE, XXH64_hashFromCanonical)
/* XXH3_64bits */
#  define XXH3_64bits XXH_NAME2(XXH_NAMESPACE, XXH3_64bits)
#  define XXH3_64bits_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_withSecret)
#  define XXH3_64bits_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_withSeed)
#  define XXH3_64bits_withSecretandSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_withSecretandSeed)
#  define XXH3_createState XXH_NAME2(XXH_NAMESPACE, XXH3_createState)
#  define XXH3_freeState XXH_NAME2(XXH_NAMESPACE, XXH3_freeState)
#  define XXH3_copyState XXH_NAME2(XXH_NAMESPACE, XXH3_copyState)
#  define XXH3_64bits_reset XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset)
#  define XXH3_64bits_reset_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset_withSeed)
#  define XXH3_64bits_reset_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset_withSecret)
#  define XXH3_64bits_reset_withSecretandSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset_withSecretandSeed)
#  define XXH3_64bits_update XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_update)
#  define XXH3_64bits_digest XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_digest)
#  define XXH3_generateSecret XXH_NAME2(XXH_NAMESPACE, XXH3_generateSecret)
#  define XXH3_generateSecret_fromSeed XXH_NAME2(XXH_NAMESPACE, XXH3_generateSecret_fromSeed)
/* XXH3_128bits */
#  define XXH128 XXH_NAME2(XXH_NAMESPACE, XXH128)
#  define XXH3_128bits XXH_NAME2(XXH_NAMESPACE, XXH3_128bits)
#  define XXH3_128bits_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_withSeed)
#  define XXH3_128bits_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_withSecret)
#  define XXH3_128bits_withSecretandSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_withSecretandSeed)
#  define XXH3_128bits_reset XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset)
#  define XXH3_128bits_reset_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset_withSeed)
#  define XXH3_128bits_reset_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset_withSecret)
#  define XXH3_128bits_reset_withSecretandSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset_withSecretandSeed)
#  define XXH3_128bits_update XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_update)
#  define XXH3_128bits_digest XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_digest)
#  define XXH128_isEqual XXH_NAME2(XXH_NAMESPACE, XXH128_isEqual)
#  define XXH128_cmp     XXH_NAME2(XXH_NAMESPACE, XXH128_cmp)
#  define XXH128_canonicalFromHash XXH_NAME2(XXH_NAMESPACE, XXH128_canonicalFromHash)
#  define XXH128_hashFromCanonical XXH_NAME2(XXH_NAMESPACE, XXH128_hashFromCanonical)
#endif


/* *************************************
*  Compiler specifics
***************************************/

/* specific declaration modes for Windows */
#if !defined(XXH_INLINE_ALL) && !defined(XXH_PRIVATE_API)
#  if defined(WIN32) && defined(_MSC_VER) && (defined(XXH_IMPORT) || defined(XXH_EXPORT))
#    ifdef XXH_EXPORT
#      define XXH_PUBLIC_API __declspec(dllexport)
#    elif XXH_IMPORT
#      define XXH_PUBLIC_API __declspec(dllimport)
#    endif
#  else
#    define XXH_PUBLIC_API   /* do nothing */
#  endif
#endif

#if defined (__GNUC__)
# define XXH_CONSTF  __attribute__((const))
# define XXH_PUREF   __attribute__((pure))
# define XXH_MALLOCF __attribute__((malloc))
#else
# define XXH_CONSTF  /* disable */
# define XXH_PUREF
# define XXH_MALLOCF
#endif

/* *************************************
*  Version
***************************************/
#define XXH_VERSION_MAJOR    0
#define XXH_VERSION_MINOR    8
#define XXH_VERSION_RELEASE  2
/*! @brief Version number, encoded as two digits each */
#define XXH_VERSION_NUMBER  (XXH_VERSION_MAJOR *100*100 + XXH_VERSION_MINOR *100 + XXH_VERSION_RELEASE)

/*!
 * @brief Obtains the xxHash version.
 *
 * This is mostly useful when xxHash is compiled as a shared library,
 * since the returned value comes from the library, as opposed to header file.
 *
 * @return @ref XXH_VERSION_NUMBER of the invoked library.
 */
XXH_PUBLIC_API XXH_CONSTF unsigned XXH_versionNumber (void);


/* ****************************
*  Common basic types
******************************/
#include <stddef.h>   /* size_t */
/*!
 * @brief Exit code for the streaming API.
 */
typedef enum {
    XXH_OK = 0, /*!< OK */
    XXH_ERROR   /*!< Error */
} XXH_errorcode;


/*-**********************************************************************
*  32-bit hash
************************************************************************/
#if defined(XXH_DOXYGEN) /* Don't show <stdint.h> include */
/*!
 * @brief An unsigned 32-bit integer.
 *
 * Not necessarily defined to `uint32_t` but functionally equivalent.
 */
typedef uint32_t XXH32_hash_t;

#elif !defined (__VMS) \
  && (defined (__cplusplus) \
  || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
#   include <stdint.h>
    typedef uint32_t XXH32_hash_t;

#else
#   include <limits.h>
#   if UINT_MAX == 0xFFFFFFFFUL
      typedef unsigned int XXH32_hash_t;
#   elif ULONG_MAX == 0xFFFFFFFFUL
      typedef unsigned long XXH32_hash_t;
#   else
#     error "unsupported platform: need a 32-bit type"
#   endif
#endif

/*!
 * @}
 *
 * @defgroup XXH32_family XXH32 family
 * @ingroup public
 * Contains functions used in the classic 32-bit xxHash algorithm.
 *
 * @note
 *   XXH32 is useful for older platforms, with no or poor 64-bit performance.
 *   Note that the @ref XXH3_family provides competitive speed for both 32-bit
 *   and 64-bit systems, and offers true 64/128 bit hash results.
 *
 * @see @ref XXH64_family, @ref XXH3_family : Other xxHash families
 * @see @ref XXH32_impl for implementation details
 * @{
 */

/*!
 * @brief Calculates the 32-bit hash of @p input using xxHash32.
 *
 * Speed on Core 2 Duo @ 3 GHz (single thread, SMHasher benchmark): 5.4 GB/s
 *
 * See @ref single_shot_example "Single Shot Example" for an example.
 *
 * @param input The block of data to be hashed, at least @p length bytes in size.
 * @param length The length of @p input, in bytes.
 * @param seed The 32-bit seed to alter the hash's output predictably.
 *
 * @pre
 *   The memory between @p input and @p input + @p length must be valid,
 *   readable, contiguous memory. However, if @p length is `0`, @p input may be
 *   `NULL`. In C++, this also must be *TriviallyCopyable*.
 *
 * @return The calculated 32-bit hash value.
 *
 * @see
 *    XXH64(), XXH3_64bits_withSeed(), XXH3_128bits_withSeed(), XXH128():
 *    Direct equivalents for the other variants of xxHash.
 * @see
 *    XXH32_createState(), XXH32_update(), XXH32_digest(): Streaming version.
 */
XXH_PUBLIC_API XXH_PUREF XXH32_hash_t XXH32 (const void* input, size_t length, XXH32_hash_t seed);

#ifndef XXH_NO_STREAM
/*!
 * Streaming functions generate the xxHash value from an incremental input.
 * This method is slower than single-call functions, due to state management.
 * For small inputs, prefer `XXH32()` and `XXH64()`, which are better optimized.
 *
 * An XXH state must first be allocated using `XXH*_createState()`.
 *
 * Start a new hash by initializing the state with a seed using `XXH*_reset()`.
 *
 * Then, feed the hash state by calling `XXH*_update()` as many times as necessary.
 *
 * The function returns an error code, with 0 meaning OK, and any other value
 * meaning there is an error.
 *
 * Finally, a hash value can be produced anytime, by using `XXH*_digest()`.
 * This function returns the nn-bits hash as an int or long long.
 *
 * It's still possible to continue inserting input into the hash state after a
 * digest, and generate new hash values later on by invoking `XXH*_digest()`.
 *
 * When done, release the state using `XXH*_freeState()`.
 *
 * @see streaming_example at the top of @ref xxhash.h for an example.
 */

/*!
 * @typedef struct XXH32_state_s XXH32_state_t
 * @brief The opaque state struct for the XXH32 streaming API.
 *
 * @see XXH32_state_s for details.
 */
typedef struct XXH32_state_s XXH32_state_t;

/*!
 * @brief Allocates an @ref XXH32_state_t.
 *
 * Must be freed with XXH32_freeState().
 * @return An allocated XXH32_state_t on success, `NULL` on failure.
 */
XXH_PUBLIC_API XXH_MALLOCF XXH32_state_t* XXH32_createState(void);
/*!
 * @brief Frees an @ref XXH32_state_t.
 *
 * Must be allocated with XXH32_createState().
 * @param statePtr A pointer to an @ref XXH32_state_t allocated with @ref XXH32_createState().
 * @return XXH_OK.
 */
XXH_PUBLIC_API XXH_errorcode  XXH32_freeState(XXH32_state_t* statePtr);
/*!
 * @brief Copies one @ref XXH32_state_t to another.
 *
 * @param dst_state The state to copy to.
 * @param src_state The state to copy from.
 * @pre
 *   @p dst_state and @p src_state must not be `NULL` and must not overlap.
 */
XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* dst_state, const XXH32_state_t* src_state);

/*!
 * @brief Resets an @ref XXH32_state_t to begin a new hash.
 *
 * This function resets and seeds a state. Call it before @ref XXH32_update().
 *
 * @param statePtr The state struct to reset.
 * @param seed The 32-bit seed to alter the hash result predictably.
 *
 * @pre
 *   @p statePtr must not be `NULL`.
 *
 * @return @ref XXH_OK on success, @ref XXH_ERROR on failure.
 */
XXH_PUBLIC_API XXH_errorcode XXH32_reset  (XXH32_state_t* statePtr, XXH32_hash_t seed);

/*!
 * @brief Consumes a block of @p input to an @ref XXH32_state_t.
 *
 * Call this to incrementally consume blocks of data.
 *
 * @param statePtr The state struct to update.
 * @param input The block of data to be hashed, at least @p length bytes in size.
 * @param length The length of @p input, in bytes.
 *
 * @pre
 *   @p statePtr must not be `NULL`.
 * @pre
 *   The memory between @p input and @p input + @p length must be valid,
 *   readable, contiguous memory. However, if @p length is `0`, @p input may be
 *   `NULL`. In C++, this also must be *TriviallyCopyable*.
 *
 * @return @ref XXH_OK on success, @ref XXH_ERROR on failure.
 */
XXH_PUBLIC_API XXH_errorcode XXH32_update (XXH32_state_t* statePtr, const void* input, size_t length);

/*!
 * @brief Returns the calculated hash value from an @ref XXH32_state_t.
 *
 * @note
 *   Calling XXH32_digest() will not affect @p statePtr, so you can update,
 *   digest, and update again.
 *
 * @param statePtr The state struct to calculate the hash from.
 *
 * @pre
 *  @p statePtr must not be `NULL`.
 *
 * @return The calculated xxHash32 value from that state.
 */
XXH_PUBLIC_API XXH_PUREF XXH32_hash_t XXH32_digest (const XXH32_state_t* statePtr);
#endif /* !XXH_NO_STREAM */

/*******   Canonical representation   *******/

/*
 * The default return values from XXH functions are unsigned 32 and 64 bit
 * integers.
 * This the simplest and fastest format for further post-processing.
 *
 * However, this leaves open the question of what is the order on the byte level,
 * since little and big endian conventions will store the same number differently.
 *
 * The canonical representation settles this issue by mandating big-endian
 * convention, the same convention as human-readable numbers (large digits first).
 *
 * When writing hash values to storage, sending them over a network, or printing
 * them, it's highly recommended to use the canonical representation to ensure
 * portability across a wider range of systems, present and future.
 *
 * The following functions allow transformation of hash values to and from
 * canonical format.
 */

/*!
 * @brief Canonical (big endian) representation of @ref XXH32_hash_t.
 */
typedef struct {
    unsigned char digest[4]; /*!< Hash bytes, big endian */
} XXH32_canonical_t;

/*!
 * @brief Converts an @ref XXH32_hash_t to a big endian @ref XXH32_canonical_t.
 *
 * @param dst The @ref XXH32_canonical_t pointer to be stored to.
 * @param hash The @ref XXH32_hash_t to be converted.
 *
 * @pre
 *   @p dst must not be `NULL`.
 */
XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t hash);

/*!
 * @brief Converts an @ref XXH32_canonical_t to a native @ref XXH32_hash_t.
 *
 * @param src The @ref XXH32_canonical_t to convert.
 *
 * @pre
 *   @p src must not be `NULL`.
 *
 * @return The converted hash.
 */
XXH_PUBLIC_API XXH_PUREF XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src);


/*! @cond Doxygen ignores this part */
#ifdef __has_attribute
# define XXH_HAS_ATTRIBUTE(x) __has_attribute(x)
#else
# define XXH_HAS_ATTRIBUTE(x) 0
#endif
/*! @endcond */

/*! @cond Doxygen ignores this part */
/*
 * C23 __STDC_VERSION__ number hasn't been specified yet. For now
 * leave as `201711L` (C17 + 1).
 * TODO: Update to correct value when its been specified.
 */
#define XXH_C23_VN 201711L
/*! @endcond */

/*! @cond Doxygen ignores this part */
/* C-language Attributes are added in C23. */
#if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= XXH_C23_VN) && defined(__has_c_attribute)
# define XXH_HAS_C_ATTRIBUTE(x) __has_c_attribute(x)
#else
# define XXH_HAS_C_ATTRIBUTE(x) 0
#endif
/*! @endcond */

/*! @cond Doxygen ignores this part */
#if defined(__cplusplus) && defined(__has_cpp_attribute)
# define XXH_HAS_CPP_ATTRIBUTE(x) __has_cpp_attribute(x)
#else
# define XXH_HAS_CPP_ATTRIBUTE(x) 0
#endif
/*! @endcond */

/*! @cond Doxygen ignores this part */
/*
 * Define XXH_FALLTHROUGH macro for annotating switch case with the 'fallthrough' attribute
 * introduced in CPP17 and C23.
 * CPP17 : https://en.cppreference.com/w/cpp/language/attributes/fallthrough
 * C23   : https://en.cppreference.com/w/c/language/attributes/fallthrough
 */
#if XXH_HAS_C_ATTRIBUTE(fallthrough) || XXH_HAS_CPP_ATTRIBUTE(fallthrough)
# define XXH_FALLTHROUGH [[fallthrough]]
#elif XXH_HAS_ATTRIBUTE(__fallthrough__)
# define XXH_FALLTHROUGH __attribute__ ((__fallthrough__))
#else
# define XXH_FALLTHROUGH /* fallthrough */
#endif
/*! @endcond */

/*! @cond Doxygen ignores this part */
/*
 * Define XXH_NOESCAPE for annotated pointers in public API.
 * https://clang.llvm.org/docs/AttributeReference.html#noescape
 * As of writing this, only supported by clang.
 */
#if XXH_HAS_ATTRIBUTE(noescape)
# define XXH_NOESCAPE __attribute__((noescape))
#else
# define XXH_NOESCAPE
#endif
/*! @endcond */


/*!
 * @}
 * @ingroup public
 * @{
 */

#ifndef XXH_NO_LONG_LONG
/*-**********************************************************************
*  64-bit hash
************************************************************************/
#if defined(XXH_DOXYGEN) /* don't include <stdint.h> */
/*!
 * @brief An unsigned 64-bit integer.
 *
 * Not necessarily defined to `uint64_t` but functionally equivalent.
 */
typedef uint64_t XXH64_hash_t;
#elif !defined (__VMS) \
  && (defined (__cplusplus) \
  || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
#  include <stdint.h>
   typedef uint64_t XXH64_hash_t;
#else
#  include <limits.h>
#  if defined(__LP64__) && ULONG_MAX == 0xFFFFFFFFFFFFFFFFULL
     /* LP64 ABI says uint64_t is unsigned long */
     typedef unsigned long XXH64_hash_t;
#  else
     /* the following type must have a width of 64-bit */
     typedef unsigned long long XXH64_hash_t;
#  endif
#endif

/*!
 * @}
 *
 * @defgroup XXH64_family XXH64 family
 * @ingroup public
 * @{
 * Contains functions used in the classic 64-bit xxHash algorithm.
 *
 * @note
 *   XXH3 provides competitive speed for both 32-bit and 64-bit systems,
 *   and offers true 64/128 bit hash results.
 *   It provides better speed for systems with vector processing capabilities.
 */

/*!
 * @brief Calculates the 64-bit hash of @p input using xxHash64.
 *
 * This function usually runs faster on 64-bit systems, but slower on 32-bit
 * systems (see benchmark).
 *
 * @param input The block of data to be hashed, at least @p length bytes in size.
 * @param length The length of @p input, in bytes.
 * @param seed The 64-bit seed to alter the hash's output predictably.
 *
 * @pre
 *   The memory between @p input and @p input + @p length must be valid,
 *   readable, contiguous memory. However, if @p length is `0`, @p input may be
 *   `NULL`. In C++, this also must be *TriviallyCopyable*.
 *
 * @return The calculated 64-bit hash.
 *
 * @see
 *    XXH32(), XXH3_64bits_withSeed(), XXH3_128bits_withSeed(), XXH128():
 *    Direct equivalents for the other variants of xxHash.
 * @see
 *    XXH64_createState(), XXH64_update(), XXH64_digest(): Streaming version.
 */
XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH64(XXH_NOESCAPE const void* input, size_t length, XXH64_hash_t seed);

/*******   Streaming   *******/
#ifndef XXH_NO_STREAM
/*!
 * @brief The opaque state struct for the XXH64 streaming API.
 *
 * @see XXH64_state_s for details.
 */
typedef struct XXH64_state_s XXH64_state_t;   /* incomplete type */

/*!
 * @brief Allocates an @ref XXH64_state_t.
 *
 * Must be freed with XXH64_freeState().
 * @return An allocated XXH64_state_t on success, `NULL` on failure.
 */
XXH_PUBLIC_API XXH_MALLOCF XXH64_state_t* XXH64_createState(void);

/*!
 * @brief Frees an @ref XXH64_state_t.
 *
 * Must be allocated with XXH64_createState().
 * @param statePtr A pointer to an @ref XXH64_state_t allocated with @ref XXH64_createState().
 * @return XXH_OK.
 */
XXH_PUBLIC_API XXH_errorcode  XXH64_freeState(XXH64_state_t* statePtr);

/*!
 * @brief Copies one @ref XXH64_state_t to another.
 *
 * @param dst_state The state to copy to.
 * @param src_state The state to copy from.
 * @pre
 *   @p dst_state and @p src_state must not be `NULL` and must not overlap.
 */
XXH_PUBLIC_API void XXH64_copyState(XXH_NOESCAPE XXH64_state_t* dst_state, const XXH64_state_t* src_state);

/*!
 * @brief Resets an @ref XXH64_state_t to begin a new hash.
 *
 * This function resets and seeds a state. Call it before @ref XXH64_update().
 *
 * @param statePtr The state struct to reset.
 * @param seed The 64-bit seed to alter the hash result predictably.
 *
 * @pre
 *   @p statePtr must not be `NULL`.
 *
 * @return @ref XXH_OK on success, @ref XXH_ERROR on failure.
 */
XXH_PUBLIC_API XXH_errorcode XXH64_reset  (XXH_NOESCAPE XXH64_state_t* statePtr, XXH64_hash_t seed);

/*!
 * @brief Consumes a block of @p input to an @ref XXH64_state_t.
 *
 * Call this to incrementally consume blocks of data.
 *
 * @param statePtr The state struct to update.
 * @param input The block of data to be hashed, at least @p length bytes in size.
 * @param length The length of @p input, in bytes.
 *
 * @pre
 *   @p statePtr must not be `NULL`.
 * @pre
 *   The memory between @p input and @p input + @p length must be valid,
 *   readable, contiguous memory. However, if @p length is `0`, @p input may be
 *   `NULL`. In C++, this also must be *TriviallyCopyable*.
 *
 * @return @ref XXH_OK on success, @ref XXH_ERROR on failure.
 */
XXH_PUBLIC_API XXH_errorcode XXH64_update (XXH_NOESCAPE XXH64_state_t* statePtr, XXH_NOESCAPE const void* input, size_t length);

/*!
 * @brief Returns the calculated hash value from an @ref XXH64_state_t.
 *
 * @note
 *   Calling XXH64_digest() will not affect @p statePtr, so you can update,
 *   digest, and update again.
 *
 * @param statePtr The state struct to calculate the hash from.
 *
 * @pre
 *  @p statePtr must not be `NULL`.
 *
 * @return The calculated xxHash64 value from that state.
 */
XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH64_digest (XXH_NOESCAPE const XXH64_state_t* statePtr);
#endif /* !XXH_NO_STREAM */
/*******   Canonical representation   *******/

/*!
 * @brief Canonical (big endian) representation of @ref XXH64_hash_t.
 */
typedef struct { unsigned char digest[sizeof(XXH64_hash_t)]; } XXH64_canonical_t;

/*!
 * @brief Converts an @ref XXH64_hash_t to a big endian @ref XXH64_canonical_t.
 *
 * @param dst The @ref XXH64_canonical_t pointer to be stored to.
 * @param hash The @ref XXH64_hash_t to be converted.
 *
 * @pre
 *   @p dst must not be `NULL`.
 */
XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH_NOESCAPE XXH64_canonical_t* dst, XXH64_hash_t hash);

/*!
 * @brief Converts an @ref XXH64_canonical_t to a native @ref XXH64_hash_t.
 *
 * @param src The @ref XXH64_canonical_t to convert.
 *
 * @pre
 *   @p src must not be `NULL`.
 *
 * @return The converted hash.
 */
XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH64_hashFromCanonical(XXH_NOESCAPE const XXH64_canonical_t* src);

#ifndef XXH_NO_XXH3

/*!
 * @}
 * ************************************************************************
 * @defgroup XXH3_family XXH3 family
 * @ingroup public
 * @{
 *
 * XXH3 is a more recent hash algorithm featuring:
 *  - Improved speed for both small and large inputs
 *  - True 64-bit and 128-bit outputs
 *  - SIMD acceleration
 *  - Improved 32-bit viability
 *
 * Speed analysis methodology is explained here:
 *
 *    https://fastcompression.blogspot.com/2019/03/presenting-xxh3.html
 *
 * Compared to XXH64, expect XXH3 to run approximately
 * ~2x faster on large inputs and >3x faster on small ones,
 * exact differences vary depending on platform.
 *
 * XXH3's speed benefits greatly from SIMD and 64-bit arithmetic,
 * but does not require it.
 * Most 32-bit and 64-bit targets that can run XXH32 smoothly can run XXH3
 * at competitive speeds, even without vector support. Further details are
 * explained in the implementation.
 *
 * XXH3 has a fast scalar implementation, but it also includes accelerated SIMD
 * implementations for many common platforms:
 *   - AVX512
 *   - AVX2
 *   - SSE2
 *   - ARM NEON
 *   - WebAssembly SIMD128
 *   - POWER8 VSX
 *   - s390x ZVector
 * This can be controlled via the @ref XXH_VECTOR macro, but it automatically
 * selects the best version according to predefined macros. For the x86 family, an
 * automatic runtime dispatcher is included separately in @ref xxh_x86dispatch.c.
 *
 * XXH3 implementation is portable:
 * it has a generic C90 formulation that can be compiled on any platform,
 * all implementations generate exactly the same hash value on all platforms.
 * Starting from v0.8.0, it's also labelled "stable", meaning that
 * any future version will also generate the same hash value.
 *
 * XXH3 offers 2 variants, _64bits and _128bits.
 *
 * When only 64 bits are needed, prefer invoking the _64bits variant, as it
 * reduces the amount of mixing, resulting in faster speed on small inputs.
 * It's also generally simpler to manipulate a scalar return type than a struct.
 *
 * The API supports one-shot hashing, streaming mode, and custom secrets.
 */
/*-**********************************************************************
*  XXH3 64-bit variant
************************************************************************/

/*!
 * @brief 64-bit unseeded variant of XXH3.
 *
 * This is equivalent to @ref XXH3_64bits_withSeed() with a seed of 0, however
 * it may have slightly better performance due to constant propagation of the
 * defaults.
 *
 * @see
 *    XXH32(), XXH64(), XXH3_128bits(): equivalent for the other xxHash algorithms
 * @see
 *    XXH3_64bits_withSeed(), XXH3_64bits_withSecret(): other seeding variants
 * @see
 *    XXH3_64bits_reset(), XXH3_64bits_update(), XXH3_64bits_digest(): Streaming version.
 */
XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH3_64bits(XXH_NOESCAPE const void* input, size_t length);

/*!
 * @brief 64-bit seeded variant of XXH3
 *
 * This variant generates a custom secret on the fly based on default secret
 * altered using the `seed` value.
 *
 * While this operation is decently fast, note that it's not completely free.
 *
 * @note
 *    seed == 0 produces the same results as @ref XXH3_64bits().
 *
 * @param input The data to hash
 * @param length The length
 * @param seed The 64-bit seed to alter the state.
 */
XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH3_64bits_withSeed(XXH_NOESCAPE const void* input, size_t length, XXH64_hash_t seed);

/*!
 * The bare minimum size for a custom secret.
 *
 * @see
 *  XXH3_64bits_withSecret(), XXH3_64bits_reset_withSecret(),
 *  XXH3_128bits_withSecret(), XXH3_128bits_reset_withSecret().
 */
#define XXH3_SECRET_SIZE_MIN 136

/*!
 * @brief 64-bit variant of XXH3 with a custom "secret".
 *
 * It's possible to provide any blob of bytes as a "secret" to generate the hash.
 * This makes it more difficult for an external actor to prepare an intentional collision.
 * The main condition is that secretSize *must* be large enough (>= XXH3_SECRET_SIZE_MIN).
 * However, the quality of the secret impacts the dispersion of the hash algorithm.
 * Therefore, the secret _must_ look like a bunch of random bytes.
 * Avoid "trivial" or structured data such as repeated sequences or a text document.
 * Whenever in doubt about the "randomness" of the blob of bytes,
 * consider employing "XXH3_generateSecret()" instead (see below).
 * It will generate a proper high entropy secret derived from the blob of bytes.
 * Another advantage of using XXH3_generateSecret() is that
 * it guarantees that all bits within the initial blob of bytes
 * will impact every bit of the output.
 * This is not necessarily the case when using the blob of bytes directly
 * because, when hashing _small_ inputs, only a portion of the secret is employed.
 */
XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH3_64bits_withSecret(XXH_NOESCAPE const void* data, size_t len, XXH_NOESCAPE const void* secret, size_t secretSize);


/*******   Streaming   *******/
#ifndef XXH_NO_STREAM
/*
 * Streaming requires state maintenance.
 * This operation costs memory and CPU.
 * As a consequence, streaming is slower than one-shot hashing.
 * For better performance, prefer one-shot functions whenever applicable.
 */

/*!
 * @brief The state struct for the XXH3 streaming API.
 *
 * @see XXH3_state_s for details.
 */
typedef struct XXH3_state_s XXH3_state_t;
XXH_PUBLIC_API XXH_MALLOCF XXH3_state_t* XXH3_createState(void);
XXH_PUBLIC_API XXH_errorcode XXH3_freeState(XXH3_state_t* statePtr);

/*!
 * @brief Copies one @ref XXH3_state_t to another.
 *
 * @param dst_state The state to copy to.
 * @param src_state The state to copy from.
 * @pre
 *   @p dst_state and @p src_state must not be `NULL` and must not overlap.
 */
XXH_PUBLIC_API void XXH3_copyState(XXH_NOESCAPE XXH3_state_t* dst_state, XXH_NOESCAPE const XXH3_state_t* src_state);

/*!
 * @brief Resets an @ref XXH3_state_t to begin a new hash.
 *
 * This function resets `statePtr` and generate a secret with default parameters. Call it before @ref XXH3_64bits_update().
 * Digest will be equivalent to `XXH3_64bits()`.
 *
 * @param statePtr The state struct to reset.
 *
 * @pre
 *   @p statePtr must not be `NULL`.
 *
 * @return @ref XXH_OK on success, @ref XXH_ERROR on failure.
 *
 */
XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset(XXH_NOESCAPE XXH3_state_t* statePtr);

/*!
 * @brief Resets an @ref XXH3_state_t with 64-bit seed to begin a new hash.
 *
 * This function resets `statePtr` and generate a secret from `seed`. Call it before @ref XXH3_64bits_update().
 * Digest will be equivalent to `XXH3_64bits_withSeed()`.
 *
 * @param statePtr The state struct to reset.
 * @param seed     The 64-bit seed to alter the state.
 *
 * @pre
 *   @p statePtr must not be `NULL`.
 *
 * @return @ref XXH_OK on success, @ref XXH_ERROR on failure.
 *
 */
XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH64_hash_t seed);

/*!
 * XXH3_64bits_reset_withSecret():
 * `secret` is referenced, it _must outlive_ the hash streaming session.
 * Similar to one-shot API, `secretSize` must be >= `XXH3_SECRET_SIZE_MIN`,
 * and the quality of produced hash values depends on secret's entropy
 * (secret's content should look like a bunch of random bytes).
 * When in doubt about the randomness of a candidate `secret`,
 * consider employing `XXH3_generateSecret()` instead (see below).
 */
XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSecret(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* secret, size_t secretSize);

/*!
 * @brief Consumes a block of @p input to an @ref XXH3_state_t.
 *
 * Call this to incrementally consume blocks of data.
 *
 * @param statePtr The state struct to update.
 * @param input The block of data to be hashed, at least @p length bytes in size.
 * @param length The length of @p input, in bytes.
 *
 * @pre
 *   @p statePtr must not be `NULL`.
 * @pre
 *   The memory between @p input and @p input + @p length must be valid,
 *   readable, contiguous memory. However, if @p length is `0`, @p input may be
 *   `NULL`. In C++, this also must be *TriviallyCopyable*.
 *
 * @return @ref XXH_OK on success, @ref XXH_ERROR on failure.
 */
XXH_PUBLIC_API XXH_errorcode XXH3_64bits_update (XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* input, size_t length);

/*!
 * @brief Returns the calculated XXH3 64-bit hash value from an @ref XXH3_state_t.
 *
 * @note
 *   Calling XXH3_64bits_digest() will not affect @p statePtr, so you can update,
 *   digest, and update again.
 *
 * @param statePtr The state struct to calculate the hash from.
 *
 * @pre
 *  @p statePtr must not be `NULL`.
 *
 * @return The calculated XXH3 64-bit hash value from that state.
 */
XXH_PUBLIC_API XXH_PUREF XXH64_hash_t  XXH3_64bits_digest (XXH_NOESCAPE const XXH3_state_t* statePtr);
#endif /* !XXH_NO_STREAM */

/* note : canonical representation of XXH3 is the same as XXH64
 * since they both produce XXH64_hash_t values */


/*-**********************************************************************
*  XXH3 128-bit variant
************************************************************************/

/*!
 * @brief The return value from 128-bit hashes.
 *
 * Stored in little endian order, although the fields themselves are in native
 * endianness.
 */
typedef struct {
    XXH64_hash_t low64;   /*!< `value & 0xFFFFFFFFFFFFFFFF` */
    XXH64_hash_t high64;  /*!< `value >> 64` */
} XXH128_hash_t;

/*!
 * @brief Unseeded 128-bit variant of XXH3
 *
 * The 128-bit variant of XXH3 has more strength, but it has a bit of overhead
 * for shorter inputs.
 *
 * This is equivalent to @ref XXH3_128bits_withSeed() with a seed of 0, however
 * it may have slightly better performance due to constant propagation of the
 * defaults.
 *
 * @see
 *    XXH32(), XXH64(), XXH3_64bits(): equivalent for the other xxHash algorithms
 * @see
 *    XXH3_128bits_withSeed(), XXH3_128bits_withSecret(): other seeding variants
 * @see
 *    XXH3_128bits_reset(), XXH3_128bits_update(), XXH3_128bits_digest(): Streaming version.
 */
XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH3_128bits(XXH_NOESCAPE const void* data, size_t len);
/*! @brief Seeded 128-bit variant of XXH3. @see XXH3_64bits_withSeed(). */
XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH3_128bits_withSeed(XXH_NOESCAPE const void* data, size_t len, XXH64_hash_t seed);
/*! @brief Custom secret 128-bit variant of XXH3. @see XXH3_64bits_withSecret(). */
XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH3_128bits_withSecret(XXH_NOESCAPE const void* data, size_t len, XXH_NOESCAPE const void* secret, size_t secretSize);

/*******   Streaming   *******/
#ifndef XXH_NO_STREAM
/*
 * Streaming requires state maintenance.
 * This operation costs memory and CPU.
 * As a consequence, streaming is slower than one-shot hashing.
 * For better performance, prefer one-shot functions whenever applicable.
 *
 * XXH3_128bits uses the same XXH3_state_t as XXH3_64bits().
 * Use already declared XXH3_createState() and XXH3_freeState().
 *
 * All reset and streaming functions have same meaning as their 64-bit counterpart.
 */

/*!
 * @brief Resets an @ref XXH3_state_t to begin a new hash.
 *
 * This function resets `statePtr` and generate a secret with default parameters. Call it before @ref XXH3_128bits_update().
 * Digest will be equivalent to `XXH3_128bits()`.
 *
 * @param statePtr The state struct to reset.
 *
 * @pre
 *   @p statePtr must not be `NULL`.
 *
 * @return @ref XXH_OK on success, @ref XXH_ERROR on failure.
 *
 */
XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset(XXH_NOESCAPE XXH3_state_t* statePtr);

/*!
 * @brief Resets an @ref XXH3_state_t with 64-bit seed to begin a new hash.
 *
 * This function resets `statePtr` and generate a secret from `seed`. Call it before @ref XXH3_128bits_update().
 * Digest will be equivalent to `XXH3_128bits_withSeed()`.
 *
 * @param statePtr The state struct to reset.
 * @param seed     The 64-bit seed to alter the state.
 *
 * @pre
 *   @p statePtr must not be `NULL`.
 *
 * @return @ref XXH_OK on success, @ref XXH_ERROR on failure.
 *
 */
XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH64_hash_t seed);
/*! @brief Custom secret 128-bit variant of XXH3. @see XXH_64bits_reset_withSecret(). */
XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSecret(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* secret, size_t secretSize);

/*!
 * @brief Consumes a block of @p input to an @ref XXH3_state_t.
 *
 * Call this to incrementally consume blocks of data.
 *
 * @param statePtr The state struct to update.
 * @param input The block of data to be hashed, at least @p length bytes in size.
 * @param length The length of @p input, in bytes.
 *
 * @pre
 *   @p statePtr must not be `NULL`.
 * @pre
 *   The memory between @p input and @p input + @p length must be valid,
 *   readable, contiguous memory. However, if @p length is `0`, @p input may be
 *   `NULL`. In C++, this also must be *TriviallyCopyable*.
 *
 * @return @ref XXH_OK on success, @ref XXH_ERROR on failure.
 */
XXH_PUBLIC_API XXH_errorcode XXH3_128bits_update (XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* input, size_t length);

/*!
 * @brief Returns the calculated XXH3 128-bit hash value from an @ref XXH3_state_t.
 *
 * @note
 *   Calling XXH3_128bits_digest() will not affect @p statePtr, so you can update,
 *   digest, and update again.
 *
 * @param statePtr The state struct to calculate the hash from.
 *
 * @pre
 *  @p statePtr must not be `NULL`.
 *
 * @return The calculated XXH3 128-bit hash value from that state.
 */
XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH3_128bits_digest (XXH_NOESCAPE const XXH3_state_t* statePtr);
#endif /* !XXH_NO_STREAM */

/* Following helper functions make it possible to compare XXH128_hast_t values.
 * Since XXH128_hash_t is a structure, this capability is not offered by the language.
 * Note: For better performance, these functions can be inlined using XXH_INLINE_ALL */

/*!
 * XXH128_isEqual():
 * Return: 1 if `h1` and `h2` are equal, 0 if they are not.
 */
XXH_PUBLIC_API XXH_PUREF int XXH128_isEqual(XXH128_hash_t h1, XXH128_hash_t h2);

/*!
 * @brief Compares two @ref XXH128_hash_t
 * This comparator is compatible with stdlib's `qsort()`/`bsearch()`.
 *
 * @return: >0 if *h128_1  > *h128_2
 *          =0 if *h128_1 == *h128_2
 *          <0 if *h128_1  < *h128_2
 */
XXH_PUBLIC_API XXH_PUREF int XXH128_cmp(XXH_NOESCAPE const void* h128_1, XXH_NOESCAPE const void* h128_2);


/*******   Canonical representation   *******/
typedef struct { unsigned char digest[sizeof(XXH128_hash_t)]; } XXH128_canonical_t;


/*!
 * @brief Converts an @ref XXH128_hash_t to a big endian @ref XXH128_canonical_t.
 *
 * @param dst The @ref XXH128_canonical_t pointer to be stored to.
 * @param hash The @ref XXH128_hash_t to be converted.
 *
 * @pre
 *   @p dst must not be `NULL`.
 */
XXH_PUBLIC_API void XXH128_canonicalFromHash(XXH_NOESCAPE XXH128_canonical_t* dst, XXH128_hash_t hash);

/*!
 * @brief Converts an @ref XXH128_canonical_t to a native @ref XXH128_hash_t.
 *
 * @param src The @ref XXH128_canonical_t to convert.
 *
 * @pre
 *   @p src must not be `NULL`.
 *
 * @return The converted hash.
 */
XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH128_hashFromCanonical(XXH_NOESCAPE const XXH128_canonical_t* src);


#endif  /* !XXH_NO_XXH3 */
#endif  /* XXH_NO_LONG_LONG */

/*!
 * @}
 */
#endif /* XXHASH_H_5627135585666179 */



#if defined(XXH_STATIC_LINKING_ONLY) && !defined(XXHASH_H_STATIC_13879238742)
#define XXHASH_H_STATIC_13879238742
/* ****************************************************************************
 * This section contains declarations which are not guaranteed to remain stable.
 * They may change in future versions, becoming incompatible with a different
 * version of the library.
 * These declarations should only be used with static linking.
 * Never use them in association with dynamic linking!
 ***************************************************************************** */

/*
 * These definitions are only present to allow static allocation
 * of XXH states, on stack or in a struct, for example.
 * Never **ever** access their members directly.
 */

/*!
 * @internal
 * @brief Structure for XXH32 streaming API.
 *
 * @note This is only defined when @ref XXH_STATIC_LINKING_ONLY,
 * @ref XXH_INLINE_ALL, or @ref XXH_IMPLEMENTATION is defined. Otherwise it is
 * an opaque type. This allows fields to safely be changed.
 *
 * Typedef'd to @ref XXH32_state_t.
 * Do not access the members of this struct directly.
 * @see XXH64_state_s, XXH3_state_s
 */
struct XXH32_state_s {
   XXH32_hash_t total_len_32; /*!< Total length hashed, modulo 2^32 */
   XXH32_hash_t large_len;    /*!< Whether the hash is >= 16 (handles @ref total_len_32 overflow) */
   XXH32_hash_t v[4];         /*!< Accumulator lanes */
   XXH32_hash_t mem32[4];     /*!< Internal buffer for partial reads. Treated as unsigned char[16]. */
   XXH32_hash_t memsize;      /*!< Amount of data in @ref mem32 */
   XXH32_hash_t reserved;     /*!< Reserved field. Do not read nor write to it. */
};   /* typedef'd to XXH32_state_t */


#ifndef XXH_NO_LONG_LONG  /* defined when there is no 64-bit support */

/*!
 * @internal
 * @brief Structure for XXH64 streaming API.
 *
 * @note This is only defined when @ref XXH_STATIC_LINKING_ONLY,
 * @ref XXH_INLINE_ALL, or @ref XXH_IMPLEMENTATION is defined. Otherwise it is
 * an opaque type. This allows fields to safely be changed.
 *
 * Typedef'd to @ref XXH64_state_t.
 * Do not access the members of this struct directly.
 * @see XXH32_state_s, XXH3_state_s
 */
struct XXH64_state_s {
   XXH64_hash_t total_len;    /*!< Total length hashed. This is always 64-bit. */
   XXH64_hash_t v[4];         /*!< Accumulator lanes */
   XXH64_hash_t mem64[4];     /*!< Internal buffer for partial reads. Treated as unsigned char[32]. */
   XXH32_hash_t memsize;      /*!< Amount of data in @ref mem64 */
   XXH32_hash_t reserved32;   /*!< Reserved field, needed for padding anyways*/
   XXH64_hash_t reserved64;   /*!< Reserved field. Do not read or write to it. */
};   /* typedef'd to XXH64_state_t */

#ifndef XXH_NO_XXH3

#if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L) /* >= C11 */
#  include <stdalign.h>
#  define XXH_ALIGN(n)      alignas(n)
#elif defined(__cplusplus) && (__cplusplus >= 201103L) /* >= C++11 */
/* In C++ alignas() is a keyword */
#  define XXH_ALIGN(n)      alignas(n)
#elif defined(__GNUC__)
#  define XXH_ALIGN(n)      __attribute__ ((aligned(n)))
#elif defined(_MSC_VER)
#  define XXH_ALIGN(n)      __declspec(align(n))
#else
#  define XXH_ALIGN(n)   /* disabled */
#endif

/* Old GCC versions only accept the attribute after the type in structures. */
#if !(defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L))   /* C11+ */ \
    && ! (defined(__cplusplus) && (__cplusplus >= 201103L)) /* >= C++11 */ \
    && defined(__GNUC__)
#   define XXH_ALIGN_MEMBER(align, type) type XXH_ALIGN(align)
#else
#   define XXH_ALIGN_MEMBER(align, type) XXH_ALIGN(align) type
#endif

/*!
 * @brief The size of the internal XXH3 buffer.
 *
 * This is the optimal update size for incremental hashing.
 *
 * @see XXH3_64b_update(), XXH3_128b_update().
 */
#define XXH3_INTERNALBUFFER_SIZE 256

/*!
 * @internal
 * @brief Default size of the secret buffer (and @ref XXH3_kSecret).
 *
 * This is the size used in @ref XXH3_kSecret and the seeded functions.
 *
 * Not to be confused with @ref XXH3_SECRET_SIZE_MIN.
 */
#define XXH3_SECRET_DEFAULT_SIZE 192

/*!
 * @internal
 * @brief Structure for XXH3 streaming API.
 *
 * @note This is only defined when @ref XXH_STATIC_LINKING_ONLY,
 * @ref XXH_INLINE_ALL, or @ref XXH_IMPLEMENTATION is defined.
 * Otherwise it is an opaque type.
 * Never use this definition in combination with dynamic library.
 * This allows fields to safely be changed in the future.
 *
 * @note ** This structure has a strict alignment requirement of 64 bytes!! **
 * Do not allocate this with `malloc()` or `new`,
 * it will not be sufficiently aligned.
 * Use @ref XXH3_createState() and @ref XXH3_freeState(), or stack allocation.
 *
 * Typedef'd to @ref XXH3_state_t.
 * Do never access the members of this struct directly.
 *
 * @see XXH3_INITSTATE() for stack initialization.
 * @see XXH3_createState(), XXH3_freeState().
 * @see XXH32_state_s, XXH64_state_s
 */
struct XXH3_state_s {
   XXH_ALIGN_MEMBER(64, XXH64_hash_t acc[8]);
       /*!< The 8 accumulators. See @ref XXH32_state_s::v and @ref XXH64_state_s::v */
   XXH_ALIGN_MEMBER(64, unsigned char customSecret[XXH3_SECRET_DEFAULT_SIZE]);
       /*!< Used to store a custom secret generated from a seed. */
   XXH_ALIGN_MEMBER(64, unsigned char buffer[XXH3_INTERNALBUFFER_SIZE]);
       /*!< The internal buffer. @see XXH32_state_s::mem32 */
   XXH32_hash_t bufferedSize;
       /*!< The amount of memory in @ref buffer, @see XXH32_state_s::memsize */
   XXH32_hash_t useSeed;
       /*!< Reserved field. Needed for padding on 64-bit. */
   size_t nbStripesSoFar;
       /*!< Number or stripes processed. */
   XXH64_hash_t totalLen;
       /*!< Total length hashed. 64-bit even on 32-bit targets. */
   size_t nbStripesPerBlock;
       /*!< Number of stripes per block. */
   size_t secretLimit;
       /*!< Size of @ref customSecret or @ref extSecret */
   XXH64_hash_t seed;
       /*!< Seed for _withSeed variants. Must be zero otherwise, @see XXH3_INITSTATE() */
   XXH64_hash_t reserved64;
       /*!< Reserved field. */
   const unsigned char* extSecret;
       /*!< Reference to an external secret for the _withSecret variants, NULL
        *   for other variants. */
   /* note: there may be some padding at the end due to alignment on 64 bytes */
}; /* typedef'd to XXH3_state_t */

#undef XXH_ALIGN_MEMBER

/*!
 * @brief Initializes a stack-allocated `XXH3_state_s`.
 *
 * When the @ref XXH3_state_t structure is merely emplaced on stack,
 * it should be initialized with XXH3_INITSTATE() or a memset()
 * in case its first reset uses XXH3_NNbits_reset_withSeed().
 * This init can be omitted if the first reset uses default or _withSecret mode.
 * This operation isn't necessary when the state is created with XXH3_createState().
 * Note that this doesn't prepare the state for a streaming operation,
 * it's still necessary to use XXH3_NNbits_reset*() afterwards.
 */
#define XXH3_INITSTATE(XXH3_state_ptr)                       \
    do {                                                     \
        XXH3_state_t* tmp_xxh3_state_ptr = (XXH3_state_ptr); \
        tmp_xxh3_state_ptr->seed = 0;                        \
        tmp_xxh3_state_ptr->extSecret = NULL;                \
    } while(0)


/*!
 * simple alias to pre-selected XXH3_128bits variant
 */
XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH128(XXH_NOESCAPE const void* data, size_t len, XXH64_hash_t seed);


/* ===   Experimental API   === */
/* Symbols defined below must be considered tied to a specific library version. */

/*!
 * XXH3_generateSecret():
 *
 * Derive a high-entropy secret from any user-defined content, named customSeed.
 * The generated secret can be used in combination with `*_withSecret()` functions.
 * The `_withSecret()` variants are useful to provide a higher level of protection
 * than 64-bit seed, as it becomes much more difficult for an external actor to
 * guess how to impact the calculation logic.
 *
 * The function accepts as input a custom seed of any length and any content,
 * and derives from it a high-entropy secret of length @p secretSize into an
 * already allocated buffer @p secretBuffer.
 *
 * The generated secret can then be used with any `*_withSecret()` variant.
 * The functions @ref XXH3_128bits_withSecret(), @ref XXH3_64bits_withSecret(),
 * @ref XXH3_128bits_reset_withSecret() and @ref XXH3_64bits_reset_withSecret()
 * are part of this list. They all accept a `secret` parameter
 * which must be large enough for implementation reasons (>= @ref XXH3_SECRET_SIZE_MIN)
 * _and_ feature very high entropy (consist of random-looking bytes).
 * These conditions can be a high bar to meet, so @ref XXH3_generateSecret() can
 * be employed to ensure proper quality.
 *
 * @p customSeed can be anything. It can have any size, even small ones,
 * and its content can be anything, even "poor entropy" sources such as a bunch
 * of zeroes. The resulting `secret` will nonetheless provide all required qualities.
 *
 * @pre
 *   - @p secretSize must be >= @ref XXH3_SECRET_SIZE_MIN
 *   - When @p customSeedSize > 0, supplying NULL as customSeed is undefined behavior.
 *
 * Example code:
 * @code{.c}
 *    #include <stdio.h>
 *    #include <stdlib.h>
 *    #include <string.h>
 *    #define XXH_STATIC_LINKING_ONLY // expose unstable API
 *    #include "xxhash.h"
 *    // Hashes argv[2] using the entropy from argv[1].
 *    int main(int argc, char* argv[])
 *    {
 *        char secret[XXH3_SECRET_SIZE_MIN];
 *        if (argv != 3) { return 1; }
 *        XXH3_generateSecret(secret, sizeof(secret), argv[1], strlen(argv[1]));
 *        XXH64_hash_t h = XXH3_64bits_withSecret(
 *             argv[2], strlen(argv[2]),
 *             secret, sizeof(secret)
 *        );
 *        printf("%016llx\n", (unsigned long long) h);
 *    }
 * @endcode
 */
XXH_PUBLIC_API XXH_errorcode XXH3_generateSecret(XXH_NOESCAPE void* secretBuffer, size_t secretSize, XXH_NOESCAPE const void* customSeed, size_t customSeedSize);

/*!
 * @brief Generate the same secret as the _withSeed() variants.
 *
 * The generated secret can be used in combination with
 *`*_withSecret()` and `_withSecretandSeed()` variants.
 *
 * Example C++ `std::string` hash class:
 * @code{.cpp}
 *    #include <string>
 *    #define XXH_STATIC_LINKING_ONLY // expose unstable API
 *    #include "xxhash.h"
 *    // Slow, seeds each time
 *    class HashSlow {
 *        XXH64_hash_t seed;
 *    public:
 *        HashSlow(XXH64_hash_t s) : seed{s} {}
 *        size_t operator()(const std::string& x) const {
 *            return size_t{XXH3_64bits_withSeed(x.c_str(), x.length(), seed)};
 *        }
 *    };
 *    // Fast, caches the seeded secret for future uses.
 *    class HashFast {
 *        unsigned char secret[XXH3_SECRET_SIZE_MIN];
 *    public:
 *        HashFast(XXH64_hash_t s) {
 *            XXH3_generateSecret_fromSeed(secret, seed);
 *        }
 *        size_t operator()(const std::string& x) const {
 *            return size_t{
 *                XXH3_64bits_withSecret(x.c_str(), x.length(), secret, sizeof(secret))
 *            };
 *        }
 *    };
 * @endcode
 * @param secretBuffer A writable buffer of @ref XXH3_SECRET_SIZE_MIN bytes
 * @param seed The seed to seed the state.
 */
XXH_PUBLIC_API void XXH3_generateSecret_fromSeed(XXH_NOESCAPE void* secretBuffer, XXH64_hash_t seed);

/*!
 * These variants generate hash values using either
 * @p seed for "short" keys (< XXH3_MIDSIZE_MAX = 240 bytes)
 * or @p secret for "large" keys (>= XXH3_MIDSIZE_MAX).
 *
 * This generally benefits speed, compared to `_withSeed()` or `_withSecret()`.
 * `_withSeed()` has to generate the secret on the fly for "large" keys.
 * It's fast, but can be perceptible for "not so large" keys (< 1 KB).
 * `_withSecret()` has to generate the masks on the fly for "small" keys,
 * which requires more instructions than _withSeed() variants.
 * Therefore, _withSecretandSeed variant combines the best of both worlds.
 *
 * When @p secret has been generated by XXH3_generateSecret_fromSeed(),
 * this variant produces *exactly* the same results as `_withSeed()` variant,
 * hence offering only a pure speed benefit on "large" input,
 * by skipping the need to regenerate the secret for every large input.
 *
 * Another usage scenario is to hash the secret to a 64-bit hash value,
 * for example with XXH3_64bits(), which then becomes the seed,
 * and then employ both the seed and the secret in _withSecretandSeed().
 * On top of speed, an added benefit is that each bit in the secret
 * has a 50% chance to swap each bit in the output, via its impact to the seed.
 *
 * This is not guaranteed when using the secret directly in "small data" scenarios,
 * because only portions of the secret are employed for small data.
 */
XXH_PUBLIC_API XXH_PUREF XXH64_hash_t
XXH3_64bits_withSecretandSeed(XXH_NOESCAPE const void* data, size_t len,
                              XXH_NOESCAPE const void* secret, size_t secretSize,
                              XXH64_hash_t seed);
/*! @copydoc XXH3_64bits_withSecretandSeed() */
XXH_PUBLIC_API XXH_PUREF XXH128_hash_t
XXH3_128bits_withSecretandSeed(XXH_NOESCAPE const void* input, size_t length,
                               XXH_NOESCAPE const void* secret, size_t secretSize,
                               XXH64_hash_t seed64);
#ifndef XXH_NO_STREAM
/*! @copydoc XXH3_64bits_withSecretandSeed() */
XXH_PUBLIC_API XXH_errorcode
XXH3_64bits_reset_withSecretandSeed(XXH_NOESCAPE XXH3_state_t* statePtr,
                                    XXH_NOESCAPE const void* secret, size_t secretSize,
                                    XXH64_hash_t seed64);
/*! @copydoc XXH3_64bits_withSecretandSeed() */
XXH_PUBLIC_API XXH_errorcode
XXH3_128bits_reset_withSecretandSeed(XXH_NOESCAPE XXH3_state_t* statePtr,
                                     XXH_NOESCAPE const void* secret, size_t secretSize,
                                     XXH64_hash_t seed64);
#endif /* !XXH_NO_STREAM */

#endif  /* !XXH_NO_XXH3 */
#endif  /* XXH_NO_LONG_LONG */
#if defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API)
#  define XXH_IMPLEMENTATION
#endif

#endif  /* defined(XXH_STATIC_LINKING_ONLY) && !defined(XXHASH_H_STATIC_13879238742) */


/* ======================================================================== */
/* ======================================================================== */
/* ======================================================================== */


/*-**********************************************************************
 * xxHash implementation
 *-**********************************************************************
 * xxHash's implementation used to be hosted inside xxhash.c.
 *
 * However, inlining requires implementation to be visible to the compiler,
 * hence be included alongside the header.
 * Previously, implementation was hosted inside xxhash.c,
 * which was then #included when inlining was activated.
 * This construction created issues with a few build and install systems,
 * as it required xxhash.c to be stored in /include directory.
 *
 * xxHash implementation is now directly integrated within xxhash.h.
 * As a consequence, xxhash.c is no longer needed in /include.
 *
 * xxhash.c is still available and is still useful.
 * In a "normal" setup, when xxhash is not inlined,
 * xxhash.h only exposes the prototypes and public symbols,
 * while xxhash.c can be built into an object file xxhash.o
 * which can then be linked into the final binary.
 ************************************************************************/

#if ( defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API) \
   || defined(XXH_IMPLEMENTATION) ) && !defined(XXH_IMPLEM_13a8737387)
#  define XXH_IMPLEM_13a8737387

/* *************************************
*  Tuning parameters
***************************************/

/*!
 * @defgroup tuning Tuning parameters
 * @{
 *
 * Various macros to control xxHash's behavior.
 */
#ifdef XXH_DOXYGEN
/*!
 * @brief Define this to disable 64-bit code.
 *
 * Useful if only using the @ref XXH32_family and you have a strict C90 compiler.
 */
#  define XXH_NO_LONG_LONG
#  undef XXH_NO_LONG_LONG /* don't actually */
/*!
 * @brief Controls how unaligned memory is accessed.
 *
 * By default, access to unaligned memory is controlled by `memcpy()`, which is
 * safe and portable.
 *
 * Unfortunately, on some target/compiler combinations, the generated assembly
 * is sub-optimal.
 *
 * The below switch allow selection of a different access method
 * in the search for improved performance.
 *
 * @par Possible options:
 *
 *  - `XXH_FORCE_MEMORY_ACCESS=0` (default): `memcpy`
 *   @par
 *     Use `memcpy()`. Safe and portable. Note that most modern compilers will
 *     eliminate the function call and treat it as an unaligned access.
 *
 *  - `XXH_FORCE_MEMORY_ACCESS=1`: `__attribute__((aligned(1)))`
 *   @par
 *     Depends on compiler extensions and is therefore not portable.
 *     This method is safe _if_ your compiler supports it,
 *     and *generally* as fast or faster than `memcpy`.
 *
 *  - `XXH_FORCE_MEMORY_ACCESS=2`: Direct cast
 *  @par
 *     Casts directly and dereferences. This method doesn't depend on the
 *     compiler, but it violates the C standard as it directly dereferences an
 *     unaligned pointer. It can generate buggy code on targets which do not
 *     support unaligned memory accesses, but in some circumstances, it's the
 *     only known way to get the most performance.
 *
 *  - `XXH_FORCE_MEMORY_ACCESS=3`: Byteshift
 *  @par
 *     Also portable. This can generate the best code on old compilers which don't
 *     inline small `memcpy()` calls, and it might also be faster on big-endian
 *     systems which lack a native byteswap instruction. However, some compilers
 *     will emit literal byteshifts even if the target supports unaligned access.
 *
 *
 * @warning
 *   Methods 1 and 2 rely on implementation-defined behavior. Use these with
 *   care, as what works on one compiler/platform/optimization level may cause
 *   another to read garbage data or even crash.
 *
 * See https://fastcompression.blogspot.com/2015/08/accessing-unaligned-memory.html for details.
 *
 * Prefer these methods in priority order (0 > 3 > 1 > 2)
 */
#  define XXH_FORCE_MEMORY_ACCESS 0

/*!
 * @def XXH_SIZE_OPT
 * @brief Controls how much xxHash optimizes for size.
 *
 * xxHash, when compiled, tends to result in a rather large binary size. This
 * is mostly due to heavy usage to forced inlining and constant folding of the
 * @ref XXH3_family to increase performance.
 *
 * However, some developers prefer size over speed. This option can
 * significantly reduce the size of the generated code. When using the `-Os`
 * or `-Oz` options on GCC or Clang, this is defined to 1 by default,
 * otherwise it is defined to 0.
 *
 * Most of these size optimizations can be controlled manually.
 *
 * This is a number from 0-2.
 *  - `XXH_SIZE_OPT` == 0: Default. xxHash makes no size optimizations. Speed
 *    comes first.
 *  - `XXH_SIZE_OPT` == 1: Default for `-Os` and `-Oz`. xxHash is more
 *    conservative and disables hacks that increase code size. It implies the
 *    options @ref XXH_NO_INLINE_HINTS == 1, @ref XXH_FORCE_ALIGN_CHECK == 0,
 *    and @ref XXH3_NEON_LANES == 8 if they are not already defined.
 *  - `XXH_SIZE_OPT` == 2: xxHash tries to make itself as small as possible.
 *    Performance may cry. For example, the single shot functions just use the
 *    streaming API.
 */
#  define XXH_SIZE_OPT 0

/*!
 * @def XXH_FORCE_ALIGN_CHECK
 * @brief If defined to non-zero, adds a special path for aligned inputs (XXH32()
 * and XXH64() only).
 *
 * This is an important performance trick for architectures without decent
 * unaligned memory access performance.
 *
 * It checks for input alignment, and when conditions are met, uses a "fast
 * path" employing direct 32-bit/64-bit reads, resulting in _dramatically
 * faster_ read speed.
 *
 * The check costs one initial branch per hash, which is generally negligible,
 * but not zero.
 *
 * Moreover, it's not useful to generate an additional code path if memory
 * access uses the same instruction for both aligned and unaligned
 * addresses (e.g. x86 and aarch64).
 *
 * In these cases, the alignment check can be removed by setting this macro to 0.
 * Then the code will always use unaligned memory access.
 * Align check is automatically disabled on x86, x64, ARM64, and some ARM chips
 * which are platforms known to offer good unaligned memory accesses performance.
 *
 * It is also disabled by default when @ref XXH_SIZE_OPT >= 1.
 *
 * This option does not affect XXH3 (only XXH32 and XXH64).
 */
#  define XXH_FORCE_ALIGN_CHECK 0

/*!
 * @def XXH_NO_INLINE_HINTS
 * @brief When non-zero, sets all functions to `static`.
 *
 * By default, xxHash tries to force the compiler to inline almost all internal
 * functions.
 *
 * This can usually improve performance due to reduced jumping and improved
 * constant folding, but significantly increases the size of the binary which
 * might not be favorable.
 *
 * Additionally, sometimes the forced inlining can be detrimental to performance,
 * depending on the architecture.
 *
 * XXH_NO_INLINE_HINTS marks all internal functions as static, giving the
 * compiler full control on whether to inline or not.
 *
 * When not optimizing (-O0), using `-fno-inline` with GCC or Clang, or if
 * @ref XXH_SIZE_OPT >= 1, this will automatically be defined.
 */
#  define XXH_NO_INLINE_HINTS 0

/*!
 * @def XXH3_INLINE_SECRET
 * @brief Determines whether to inline the XXH3 withSecret code.
 *
 * When the secret size is known, the compiler can improve the performance
 * of XXH3_64bits_withSecret() and XXH3_128bits_withSecret().
 *
 * However, if the secret size is not known, it doesn't have any benefit. This
 * happens when xxHash is compiled into a global symbol. Therefore, if
 * @ref XXH_INLINE_ALL is *not* defined, this will be defined to 0.
 *
 * Additionally, this defaults to 0 on GCC 12+, which has an issue with function pointers
 * that are *sometimes* force inline on -Og, and it is impossible to automatically
 * detect this optimization level.
 */
#  define XXH3_INLINE_SECRET 0

/*!
 * @def XXH32_ENDJMP
 * @brief Whether to use a jump for `XXH32_finalize`.
 *
 * For performance, `XXH32_finalize` uses multiple branches in the finalizer.
 * This is generally preferable for performance,
 * but depending on exact architecture, a jmp may be preferable.
 *
 * This setting is only possibly making a difference for very small inputs.
 */
#  define XXH32_ENDJMP 0

/*!
 * @internal
 * @brief Redefines old internal names.
 *
 * For compatibility with code that uses xxHash's internals before the names
 * were changed to improve namespacing. There is no other reason to use this.
 */
#  define XXH_OLD_NAMES
#  undef XXH_OLD_NAMES /* don't actually use, it is ugly. */

/*!
 * @def XXH_NO_STREAM
 * @brief Disables the streaming API.
 *
 * When xxHash is not inlined and the streaming functions are not used, disabling
 * the streaming functions can improve code size significantly, especially with
 * the @ref XXH3_family which tends to make constant folded copies of itself.
 */
#  define XXH_NO_STREAM
#  undef XXH_NO_STREAM /* don't actually */
#endif /* XXH_DOXYGEN */
/*!
 * @}
 */

#ifndef XXH_FORCE_MEMORY_ACCESS   /* can be defined externally, on command line for example */
   /* prefer __packed__ structures (method 1) for GCC
    * < ARMv7 with unaligned access (e.g. Raspbian armhf) still uses byte shifting, so we use memcpy
    * which for some reason does unaligned loads. */
#  if defined(__GNUC__) && !(defined(__ARM_ARCH) && __ARM_ARCH < 7 && defined(__ARM_FEATURE_UNALIGNED))
#    define XXH_FORCE_MEMORY_ACCESS 1
#  endif
#endif

#ifndef XXH_SIZE_OPT
   /* default to 1 for -Os or -Oz */
#  if (defined(__GNUC__) || defined(__clang__)) && defined(__OPTIMIZE_SIZE__)
#    define XXH_SIZE_OPT 1
#  else
#    define XXH_SIZE_OPT 0
#  endif
#endif

#ifndef XXH_FORCE_ALIGN_CHECK  /* can be defined externally */
   /* don't check on sizeopt, x86, aarch64, or arm when unaligned access is available */
#  if XXH_SIZE_OPT >= 1 || \
      defined(__i386)  || defined(__x86_64__) || defined(__aarch64__) || defined(__ARM_FEATURE_UNALIGNED) \
   || defined(_M_IX86) || defined(_M_X64)     || defined(_M_ARM64)    || defined(_M_ARM) /* visual */
#    define XXH_FORCE_ALIGN_CHECK 0
#  else
#    define XXH_FORCE_ALIGN_CHECK 1
#  endif
#endif

#ifndef XXH_NO_INLINE_HINTS
#  if XXH_SIZE_OPT >= 1 || defined(__NO_INLINE__)  /* -O0, -fno-inline */
#    define XXH_NO_INLINE_HINTS 1
#  else
#    define XXH_NO_INLINE_HINTS 0
#  endif
#endif

#ifndef XXH3_INLINE_SECRET
#  if (defined(__GNUC__) && !defined(__clang__) && __GNUC__ >= 12) \
     || !defined(XXH_INLINE_ALL)
#    define XXH3_INLINE_SECRET 0
#  else
#    define XXH3_INLINE_SECRET 1
#  endif
#endif

#ifndef XXH32_ENDJMP
/* generally preferable for performance */
#  define XXH32_ENDJMP 0
#endif

/*!
 * @defgroup impl Implementation
 * @{
 */


/* *************************************
*  Includes & Memory related functions
***************************************/
#if defined(XXH_NO_STREAM)
/* nothing */
#elif defined(XXH_NO_STDLIB)

/* When requesting to disable any mention of stdlib,
 * the library loses the ability to invoked malloc / free.
 * In practice, it means that functions like `XXH*_createState()`
 * will always fail, and return NULL.
 * This flag is useful in situations where
 * xxhash.h is integrated into some kernel, embedded or limited environment
 * without access to dynamic allocation.
 */

static XXH_CONSTF void* XXH_malloc(size_t s) { (void)s; return NULL; }
static void XXH_free(void* p) { (void)p; }

#else

/*
 * Modify the local functions below should you wish to use
 * different memory routines for malloc() and free()
 */
#include <stdlib.h>

/*!
 * @internal
 * @brief Modify this function to use a different routine than malloc().
 */
static XXH_MALLOCF void* XXH_malloc(size_t s) { return malloc(s); }

/*!
 * @internal
 * @brief Modify this function to use a different routine than free().
 */
static void XXH_free(void* p) { free(p); }

#endif  /* XXH_NO_STDLIB */

#include <string.h>

/*!
 * @internal
 * @brief Modify this function to use a different routine than memcpy().
 */
static void* XXH_memcpy(void* dest, const void* src, size_t size)
{
    return memcpy(dest,src,size);
}

#include <limits.h>   /* ULLONG_MAX */


/* *************************************
*  Compiler Specific Options
***************************************/
#ifdef _MSC_VER /* Visual Studio warning fix */
#  pragma warning(disable : 4127) /* disable: C4127: conditional expression is constant */
#endif

#if XXH_NO_INLINE_HINTS  /* disable inlining hints */
#  if defined(__GNUC__) || defined(__clang__)
#    define XXH_FORCE_INLINE static __attribute__((unused))
#  else
#    define XXH_FORCE_INLINE static
#  endif
#  define XXH_NO_INLINE static
/* enable inlining hints */
#elif defined(__GNUC__) || defined(__clang__)
#  define XXH_FORCE_INLINE static __inline__ __attribute__((always_inline, unused))
#  define XXH_NO_INLINE static __attribute__((noinline))
#elif defined(_MSC_VER)  /* Visual Studio */
#  define XXH_FORCE_INLINE static __forceinline
#  define XXH_NO_INLINE static __declspec(noinline)
#elif defined (__cplusplus) \
  || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L))   /* C99 */
#  define XXH_FORCE_INLINE static inline
#  define XXH_NO_INLINE static
#else
#  define XXH_FORCE_INLINE static
#  define XXH_NO_INLINE static
#endif

#if XXH3_INLINE_SECRET
#  define XXH3_WITH_SECRET_INLINE XXH_FORCE_INLINE
#else
#  define XXH3_WITH_SECRET_INLINE XXH_NO_INLINE
#endif


/* *************************************
*  Debug
***************************************/
/*!
 * @ingroup tuning
 * @def XXH_DEBUGLEVEL
 * @brief Sets the debugging level.
 *
 * XXH_DEBUGLEVEL is expected to be defined externally, typically via the
 * compiler's command line options. The value must be a number.
 */
#ifndef XXH_DEBUGLEVEL
#  ifdef DEBUGLEVEL /* backwards compat */
#    define XXH_DEBUGLEVEL DEBUGLEVEL
#  else
#    define XXH_DEBUGLEVEL 0
#  endif
#endif

#if (XXH_DEBUGLEVEL>=1)
#  include <assert.h>   /* note: can still be disabled with NDEBUG */
#  define XXH_ASSERT(c)   assert(c)
#else
#  if defined(__INTEL_COMPILER)
#    define XXH_ASSERT(c)   XXH_ASSUME((unsigned char) (c))
#  else
#    define XXH_ASSERT(c)   XXH_ASSUME(c)
#  endif
#endif

/* note: use after variable declarations */
#ifndef XXH_STATIC_ASSERT
#  if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L)    /* C11 */
#    define XXH_STATIC_ASSERT_WITH_MESSAGE(c,m) do { _Static_assert((c),m); } while(0)
#  elif defined(__cplusplus) && (__cplusplus >= 201103L)            /* C++11 */
#    define XXH_STATIC_ASSERT_WITH_MESSAGE(c,m) do { static_assert((c),m); } while(0)
#  else
#    define XXH_STATIC_ASSERT_WITH_MESSAGE(c,m) do { struct xxh_sa { char x[(c) ? 1 : -1]; }; } while(0)
#  endif
#  define XXH_STATIC_ASSERT(c) XXH_STATIC_ASSERT_WITH_MESSAGE((c),#c)
#endif

/*!
 * @internal
 * @def XXH_COMPILER_GUARD(var)
 * @brief Used to prevent unwanted optimizations for @p var.
 *
 * It uses an empty GCC inline assembly statement with a register constraint
 * which forces @p var into a general purpose register (eg eax, ebx, ecx
 * on x86) and marks it as modified.
 *
 * This is used in a few places to avoid unwanted autovectorization (e.g.
 * XXH32_round()). All vectorization we want is explicit via intrinsics,
 * and _usually_ isn't wanted elsewhere.
 *
 * We also use it to prevent unwanted constant folding for AArch64 in
 * XXH3_initCustomSecret_scalar().
 */
#if defined(__GNUC__) || defined(__clang__)
#  define XXH_COMPILER_GUARD(var) __asm__("" : "+r" (var))
#else
#  define XXH_COMPILER_GUARD(var) ((void)0)
#endif

/* Specifically for NEON vectors which use the "w" constraint, on
 * Clang. */
#if defined(__clang__) && defined(__ARM_ARCH) && !defined(__wasm__)
#  define XXH_COMPILER_GUARD_CLANG_NEON(var) __asm__("" : "+w" (var))
#else
#  define XXH_COMPILER_GUARD_CLANG_NEON(var) ((void)0)
#endif

/* *************************************
*  Basic Types
***************************************/
#if !defined (__VMS) \
 && (defined (__cplusplus) \
 || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
# include <stdint.h>
  typedef uint8_t xxh_u8;
#else
  typedef unsigned char xxh_u8;
#endif
typedef XXH32_hash_t xxh_u32;

#ifdef XXH_OLD_NAMES
#  warning "XXH_OLD_NAMES is planned to be removed starting v0.9. If the program depends on it, consider moving away from it by employing newer type names directly"
#  define BYTE xxh_u8
#  define U8   xxh_u8
#  define U32  xxh_u32
#endif

/* ***   Memory access   *** */

/*!
 * @internal
 * @fn xxh_u32 XXH_read32(const void* ptr)
 * @brief Reads an unaligned 32-bit integer from @p ptr in native endianness.
 *
 * Affected by @ref XXH_FORCE_MEMORY_ACCESS.
 *
 * @param ptr The pointer to read from.
 * @return The 32-bit native endian integer from the bytes at @p ptr.
 */

/*!
 * @internal
 * @fn xxh_u32 XXH_readLE32(const void* ptr)
 * @brief Reads an unaligned 32-bit little endian integer from @p ptr.
 *
 * Affected by @ref XXH_FORCE_MEMORY_ACCESS.
 *
 * @param ptr The pointer to read from.
 * @return The 32-bit little endian integer from the bytes at @p ptr.
 */

/*!
 * @internal
 * @fn xxh_u32 XXH_readBE32(const void* ptr)
 * @brief Reads an unaligned 32-bit big endian integer from @p ptr.
 *
 * Affected by @ref XXH_FORCE_MEMORY_ACCESS.
 *
 * @param ptr The pointer to read from.
 * @return The 32-bit big endian integer from the bytes at @p ptr.
 */

/*!
 * @internal
 * @fn xxh_u32 XXH_readLE32_align(const void* ptr, XXH_alignment align)
 * @brief Like @ref XXH_readLE32(), but has an option for aligned reads.
 *
 * Affected by @ref XXH_FORCE_MEMORY_ACCESS.
 * Note that when @ref XXH_FORCE_ALIGN_CHECK == 0, the @p align parameter is
 * always @ref XXH_alignment::XXH_unaligned.
 *
 * @param ptr The pointer to read from.
 * @param align Whether @p ptr is aligned.
 * @pre
 *   If @p align == @ref XXH_alignment::XXH_aligned, @p ptr must be 4 byte
 *   aligned.
 * @return The 32-bit little endian integer from the bytes at @p ptr.
 */

#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==3))
/*
 * Manual byteshift. Best for old compilers which don't inline memcpy.
 * We actually directly use XXH_readLE32 and XXH_readBE32.
 */
#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==2))

/*
 * Force direct memory access. Only works on CPU which support unaligned memory
 * access in hardware.
 */
static xxh_u32 XXH_read32(const void* memPtr) { return *(const xxh_u32*) memPtr; }

#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==1))

/*
 * __attribute__((aligned(1))) is supported by gcc and clang. Originally the
 * documentation claimed that it only increased the alignment, but actually it
 * can decrease it on gcc, clang, and icc:
 * https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69502,
 * https://gcc.godbolt.org/z/xYez1j67Y.
 */
#ifdef XXH_OLD_NAMES
typedef union { xxh_u32 u32; } __attribute__((packed)) unalign;
#endif
static xxh_u32 XXH_read32(const void* ptr)
{
    typedef __attribute__((aligned(1))) xxh_u32 xxh_unalign32;
    return *((const xxh_unalign32*)ptr);
}

#else

/*
 * Portable and safe solution. Generally efficient.
 * see: https://fastcompression.blogspot.com/2015/08/accessing-unaligned-memory.html
 */
static xxh_u32 XXH_read32(const void* memPtr)
{
    xxh_u32 val;
    XXH_memcpy(&val, memPtr, sizeof(val));
    return val;
}

#endif   /* XXH_FORCE_DIRECT_MEMORY_ACCESS */


/* ***   Endianness   *** */

/*!
 * @ingroup tuning
 * @def XXH_CPU_LITTLE_ENDIAN
 * @brief Whether the target is little endian.
 *
 * Defined to 1 if the target is little endian, or 0 if it is big endian.
 * It can be defined externally, for example on the compiler command line.
 *
 * If it is not defined,
 * a runtime check (which is usually constant folded) is used instead.
 *
 * @note
 *   This is not necessarily defined to an integer constant.
 *
 * @see XXH_isLittleEndian() for the runtime check.
 */
#ifndef XXH_CPU_LITTLE_ENDIAN
/*
 * Try to detect endianness automatically, to avoid the nonstandard behavior
 * in `XXH_isLittleEndian()`
 */
#  if defined(_WIN32) /* Windows is always little endian */ \
     || defined(__LITTLE_ENDIAN__) \
     || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)
#    define XXH_CPU_LITTLE_ENDIAN 1
#  elif defined(__BIG_ENDIAN__) \
     || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)
#    define XXH_CPU_LITTLE_ENDIAN 0
#  else
/*!
 * @internal
 * @brief Runtime check for @ref XXH_CPU_LITTLE_ENDIAN.
 *
 * Most compilers will constant fold this.
 */
static int XXH_isLittleEndian(void)
{
    /*
     * Portable and well-defined behavior.
     * Don't use static: it is detrimental to performance.
     */
    const union { xxh_u32 u; xxh_u8 c[4]; } one = { 1 };
    return one.c[0];
}
#   define XXH_CPU_LITTLE_ENDIAN   XXH_isLittleEndian()
#  endif
#endif




/* ****************************************
*  Compiler-specific Functions and Macros
******************************************/
#define XXH_GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)

#ifdef __has_builtin
#  define XXH_HAS_BUILTIN(x) __has_builtin(x)
#else
#  define XXH_HAS_BUILTIN(x) 0
#endif



/*
 * C23 and future versions have standard "unreachable()".
 * Once it has been implemented reliably we can add it as an
 * additional case:
 *
 * ```
 * #if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= XXH_C23_VN)
 * #  include <stddef.h>
 * #  ifdef unreachable
 * #    define XXH_UNREACHABLE() unreachable()
 * #  endif
 * #endif
 * ```
 *
 * Note C++23 also has std::unreachable() which can be detected
 * as follows:
 * ```
 * #if defined(__cpp_lib_unreachable) && (__cpp_lib_unreachable >= 202202L)
 * #  include <utility>
 * #  define XXH_UNREACHABLE() std::unreachable()
 * #endif
 * ```
 * NB: `__cpp_lib_unreachable` is defined in the `<version>` header.
 * We don't use that as including `<utility>` in `extern "C"` blocks
 * doesn't work on GCC12
 */

#if XXH_HAS_BUILTIN(__builtin_unreachable)
#  define XXH_UNREACHABLE() __builtin_unreachable()

#elif defined(_MSC_VER)
#  define XXH_UNREACHABLE() __assume(0)

#else
#  define XXH_UNREACHABLE()
#endif

#if XXH_HAS_BUILTIN(__builtin_assume)
#  define XXH_ASSUME(c) __builtin_assume(c)
#else
#  define XXH_ASSUME(c) if (!(c)) { XXH_UNREACHABLE(); }
#endif

/*!
 * @internal
 * @def XXH_rotl32(x,r)
 * @brief 32-bit rotate left.
 *
 * @param x The 32-bit integer to be rotated.
 * @param r The number of bits to rotate.
 * @pre
 *   @p r > 0 && @p r < 32
 * @note
 *   @p x and @p r may be evaluated multiple times.
 * @return The rotated result.
 */
#if !defined(NO_CLANG_BUILTIN) && XXH_HAS_BUILTIN(__builtin_rotateleft32) \
                               && XXH_HAS_BUILTIN(__builtin_rotateleft64)
#  define XXH_rotl32 __builtin_rotateleft32
#  define XXH_rotl64 __builtin_rotateleft64
/* Note: although _rotl exists for minGW (GCC under windows), performance seems poor */
#elif defined(_MSC_VER)
#  define XXH_rotl32(x,r) _rotl(x,r)
#  define XXH_rotl64(x,r) _rotl64(x,r)
#else
#  define XXH_rotl32(x,r) (((x) << (r)) | ((x) >> (32 - (r))))
#  define XXH_rotl64(x,r) (((x) << (r)) | ((x) >> (64 - (r))))
#endif

/*!
 * @internal
 * @fn xxh_u32 XXH_swap32(xxh_u32 x)
 * @brief A 32-bit byteswap.
 *
 * @param x The 32-bit integer to byteswap.
 * @return @p x, byteswapped.
 */
#if defined(_MSC_VER)     /* Visual Studio */
#  define XXH_swap32 _byteswap_ulong
#elif XXH_GCC_VERSION >= 403
#  define XXH_swap32 __builtin_bswap32
#else
static xxh_u32 XXH_swap32 (xxh_u32 x)
{
    return  ((x << 24) & 0xff000000 ) |
            ((x <<  8) & 0x00ff0000 ) |
            ((x >>  8) & 0x0000ff00 ) |
            ((x >> 24) & 0x000000ff );
}
#endif


/* ***************************
*  Memory reads
*****************************/

/*!
 * @internal
 * @brief Enum to indicate whether a pointer is aligned.
 */
typedef enum {
    XXH_aligned,  /*!< Aligned */
    XXH_unaligned /*!< Possibly unaligned */
} XXH_alignment;

/*
 * XXH_FORCE_MEMORY_ACCESS==3 is an endian-independent byteshift load.
 *
 * This is ideal for older compilers which don't inline memcpy.
 */
#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==3))

XXH_FORCE_INLINE xxh_u32 XXH_readLE32(const void* memPtr)
{
    const xxh_u8* bytePtr = (const xxh_u8 *)memPtr;
    return bytePtr[0]
         | ((xxh_u32)bytePtr[1] << 8)
         | ((xxh_u32)bytePtr[2] << 16)
         | ((xxh_u32)bytePtr[3] << 24);
}

XXH_FORCE_INLINE xxh_u32 XXH_readBE32(const void* memPtr)
{
    const xxh_u8* bytePtr = (const xxh_u8 *)memPtr;
    return bytePtr[3]
         | ((xxh_u32)bytePtr[2] << 8)
         | ((xxh_u32)bytePtr[1] << 16)
         | ((xxh_u32)bytePtr[0] << 24);
}

#else
XXH_FORCE_INLINE xxh_u32 XXH_readLE32(const void* ptr)
{
    return XXH_CPU_LITTLE_ENDIAN ? XXH_read32(ptr) : XXH_swap32(XXH_read32(ptr));
}

static xxh_u32 XXH_readBE32(const void* ptr)
{
    return XXH_CPU_LITTLE_ENDIAN ? XXH_swap32(XXH_read32(ptr)) : XXH_read32(ptr);
}
#endif

XXH_FORCE_INLINE xxh_u32
XXH_readLE32_align(const void* ptr, XXH_alignment align)
{
    if (align==XXH_unaligned) {
        return XXH_readLE32(ptr);
    } else {
        return XXH_CPU_LITTLE_ENDIAN ? *(const xxh_u32*)ptr : XXH_swap32(*(const xxh_u32*)ptr);
    }
}


/* *************************************
*  Misc
***************************************/
/*! @ingroup public */
XXH_PUBLIC_API unsigned XXH_versionNumber (void) { return XXH_VERSION_NUMBER; }


/* *******************************************************************
*  32-bit hash functions
*********************************************************************/
/*!
 * @}
 * @defgroup XXH32_impl XXH32 implementation
 * @ingroup impl
 *
 * Details on the XXH32 implementation.
 * @{
 */
 /* #define instead of static const, to be used as initializers */
#define XXH_PRIME32_1  0x9E3779B1U  /*!< 0b10011110001101110111100110110001 */
#define XXH_PRIME32_2  0x85EBCA77U  /*!< 0b10000101111010111100101001110111 */
#define XXH_PRIME32_3  0xC2B2AE3DU  /*!< 0b11000010101100101010111000111101 */
#define XXH_PRIME32_4  0x27D4EB2FU  /*!< 0b00100111110101001110101100101111 */
#define XXH_PRIME32_5  0x165667B1U  /*!< 0b00010110010101100110011110110001 */

#ifdef XXH_OLD_NAMES
#  define PRIME32_1 XXH_PRIME32_1
#  define PRIME32_2 XXH_PRIME32_2
#  define PRIME32_3 XXH_PRIME32_3
#  define PRIME32_4 XXH_PRIME32_4
#  define PRIME32_5 XXH_PRIME32_5
#endif

/*!
 * @internal
 * @brief Normal stripe processing routine.
 *
 * This shuffles the bits so that any bit from @p input impacts several bits in
 * @p acc.
 *
 * @param acc The accumulator lane.
 * @param input The stripe of input to mix.
 * @return The mixed accumulator lane.
 */
static xxh_u32 XXH32_round(xxh_u32 acc, xxh_u32 input)
{
    acc += input * XXH_PRIME32_2;
    acc  = XXH_rotl32(acc, 13);
    acc *= XXH_PRIME32_1;
#if (defined(__SSE4_1__) || defined(__aarch64__) || defined(__wasm_simd128__)) && !defined(XXH_ENABLE_AUTOVECTORIZE)
    /*
     * UGLY HACK:
     * A compiler fence is the only thing that prevents GCC and Clang from
     * autovectorizing the XXH32 loop (pragmas and attributes don't work for some
     * reason) without globally disabling SSE4.1.
     *
     * The reason we want to avoid vectorization is because despite working on
     * 4 integers at a time, there are multiple factors slowing XXH32 down on
     * SSE4:
     * - There's a ridiculous amount of lag from pmulld (10 cycles of latency on
     *   newer chips!) making it slightly slower to multiply four integers at
     *   once compared to four integers independently. Even when pmulld was
     *   fastest, Sandy/Ivy Bridge, it is still not worth it to go into SSE
     *   just to multiply unless doing a long operation.
     *
     * - Four instructions are required to rotate,
     *      movqda tmp,  v // not required with VEX encoding
     *      pslld  tmp, 13 // tmp <<= 13
     *      psrld  v,   19 // x >>= 19
     *      por    v,  tmp // x |= tmp
     *   compared to one for scalar:
     *      roll   v, 13    // reliably fast across the board
     *      shldl  v, v, 13 // Sandy Bridge and later prefer this for some reason
     *
     * - Instruction level parallelism is actually more beneficial here because
     *   the SIMD actually serializes this operation: While v1 is rotating, v2
     *   can load data, while v3 can multiply. SSE forces them to operate
     *   together.
     *
     * This is also enabled on AArch64, as Clang is *very aggressive* in vectorizing
     * the loop. NEON is only faster on the A53, and with the newer cores, it is less
     * than half the speed.
     *
     * Additionally, this is used on WASM SIMD128 because it JITs to the same
     * SIMD instructions and has the same issue.
     */
    XXH_COMPILER_GUARD(acc);
#endif
    return acc;
}

/*!
 * @internal
 * @brief Mixes all bits to finalize the hash.
 *
 * The final mix ensures that all input bits have a chance to impact any bit in
 * the output digest, resulting in an unbiased distribution.
 *
 * @param hash The hash to avalanche.
 * @return The avalanched hash.
 */
static xxh_u32 XXH32_avalanche(xxh_u32 hash)
{
    hash ^= hash >> 15;
    hash *= XXH_PRIME32_2;
    hash ^= hash >> 13;
    hash *= XXH_PRIME32_3;
    hash ^= hash >> 16;
    return hash;
}

#define XXH_get32bits(p) XXH_readLE32_align(p, align)

/*!
 * @internal
 * @brief Processes the last 0-15 bytes of @p ptr.
 *
 * There may be up to 15 bytes remaining to consume from the input.
 * This final stage will digest them to ensure that all input bytes are present
 * in the final mix.
 *
 * @param hash The hash to finalize.
 * @param ptr The pointer to the remaining input.
 * @param len The remaining length, modulo 16.
 * @param align Whether @p ptr is aligned.
 * @return The finalized hash.
 * @see XXH64_finalize().
 */
static XXH_PUREF xxh_u32
XXH32_finalize(xxh_u32 hash, const xxh_u8* ptr, size_t len, XXH_alignment align)
{
#define XXH_PROCESS1 do {                             \
    hash += (*ptr++) * XXH_PRIME32_5;                 \
    hash = XXH_rotl32(hash, 11) * XXH_PRIME32_1;      \
} while (0)

#define XXH_PROCESS4 do {                             \
    hash += XXH_get32bits(ptr) * XXH_PRIME32_3;       \
    ptr += 4;                                         \
    hash  = XXH_rotl32(hash, 17) * XXH_PRIME32_4;     \
} while (0)

    if (ptr==NULL) XXH_ASSERT(len == 0);

    /* Compact rerolled version; generally faster */
    if (!XXH32_ENDJMP) {
        len &= 15;
        while (len >= 4) {
            XXH_PROCESS4;
            len -= 4;
        }
        while (len > 0) {
            XXH_PROCESS1;
            --len;
        }
        return XXH32_avalanche(hash);
    } else {
         switch(len&15) /* or switch(bEnd - p) */ {
           case 12:      XXH_PROCESS4;
                         XXH_FALLTHROUGH;  /* fallthrough */
           case 8:       XXH_PROCESS4;
                         XXH_FALLTHROUGH;  /* fallthrough */
           case 4:       XXH_PROCESS4;
                         return XXH32_avalanche(hash);

           case 13:      XXH_PROCESS4;
                         XXH_FALLTHROUGH;  /* fallthrough */
           case 9:       XXH_PROCESS4;
                         XXH_FALLTHROUGH;  /* fallthrough */
           case 5:       XXH_PROCESS4;
                         XXH_PROCESS1;
                         return XXH32_avalanche(hash);

           case 14:      XXH_PROCESS4;
                         XXH_FALLTHROUGH;  /* fallthrough */
           case 10:      XXH_PROCESS4;
                         XXH_FALLTHROUGH;  /* fallthrough */
           case 6:       XXH_PROCESS4;
                         XXH_PROCESS1;
                         XXH_PROCESS1;
                         return XXH32_avalanche(hash);

           case 15:      XXH_PROCESS4;
                         XXH_FALLTHROUGH;  /* fallthrough */
           case 11:      XXH_PROCESS4;
                         XXH_FALLTHROUGH;  /* fallthrough */
           case 7:       XXH_PROCESS4;
                         XXH_FALLTHROUGH;  /* fallthrough */
           case 3:       XXH_PROCESS1;
                         XXH_FALLTHROUGH;  /* fallthrough */
           case 2:       XXH_PROCESS1;
                         XXH_FALLTHROUGH;  /* fallthrough */
           case 1:       XXH_PROCESS1;
                         XXH_FALLTHROUGH;  /* fallthrough */
           case 0:       return XXH32_avalanche(hash);
        }
        XXH_ASSERT(0);
        return hash;   /* reaching this point is deemed impossible */
    }
}

#ifdef XXH_OLD_NAMES
#  define PROCESS1 XXH_PROCESS1
#  define PROCESS4 XXH_PROCESS4
#else
#  undef XXH_PROCESS1
#  undef XXH_PROCESS4
#endif

/*!
 * @internal
 * @brief The implementation for @ref XXH32().
 *
 * @param input , len , seed Directly passed from @ref XXH32().
 * @param align Whether @p input is aligned.
 * @return The calculated hash.
 */
XXH_FORCE_INLINE XXH_PUREF xxh_u32
XXH32_endian_align(const xxh_u8* input, size_t len, xxh_u32 seed, XXH_alignment align)
{
    xxh_u32 h32;

    if (input==NULL) XXH_ASSERT(len == 0);

    if (len>=16) {
        const xxh_u8* const bEnd = input + len;
        const xxh_u8* const limit = bEnd - 15;
        xxh_u32 v1 = seed + XXH_PRIME32_1 + XXH_PRIME32_2;
        xxh_u32 v2 = seed + XXH_PRIME32_2;
        xxh_u32 v3 = seed + 0;
        xxh_u32 v4 = seed - XXH_PRIME32_1;

        do {
            v1 = XXH32_round(v1, XXH_get32bits(input)); input += 4;
            v2 = XXH32_round(v2, XXH_get32bits(input)); input += 4;
            v3 = XXH32_round(v3, XXH_get32bits(input)); input += 4;
            v4 = XXH32_round(v4, XXH_get32bits(input)); input += 4;
        } while (input < limit);

        h32 = XXH_rotl32(v1, 1)  + XXH_rotl32(v2, 7)
            + XXH_rotl32(v3, 12) + XXH_rotl32(v4, 18);
    } else {
        h32  = seed + XXH_PRIME32_5;
    }

    h32 += (xxh_u32)len;

    return XXH32_finalize(h32, input, len&15, align);
}

/*! @ingroup XXH32_family */
XXH_PUBLIC_API XXH32_hash_t XXH32 (const void* input, size_t len, XXH32_hash_t seed)
{
#if !defined(XXH_NO_STREAM) && XXH_SIZE_OPT >= 2
    /* Simple version, good for code maintenance, but unfortunately slow for small inputs */
    XXH32_state_t state;
    XXH32_reset(&state, seed);
    XXH32_update(&state, (const xxh_u8*)input, len);
    return XXH32_digest(&state);
#else
    if (XXH_FORCE_ALIGN_CHECK) {
        if ((((size_t)input) & 3) == 0) {   /* Input is 4-bytes aligned, leverage the speed benefit */
            return XXH32_endian_align((const xxh_u8*)input, len, seed, XXH_aligned);
    }   }

    return XXH32_endian_align((const xxh_u8*)input, len, seed, XXH_unaligned);
#endif
}



/*******   Hash streaming   *******/
#ifndef XXH_NO_STREAM
/*! @ingroup XXH32_family */
XXH_PUBLIC_API XXH32_state_t* XXH32_createState(void)
{
    return (XXH32_state_t*)XXH_malloc(sizeof(XXH32_state_t));
}
/*! @ingroup XXH32_family */
XXH_PUBLIC_API XXH_errorcode XXH32_freeState(XXH32_state_t* statePtr)
{
    XXH_free(statePtr);
    return XXH_OK;
}

/*! @ingroup XXH32_family */
XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* dstState, const XXH32_state_t* srcState)
{
    XXH_memcpy(dstState, srcState, sizeof(*dstState));
}

/*! @ingroup XXH32_family */
XXH_PUBLIC_API XXH_errorcode XXH32_reset(XXH32_state_t* statePtr, XXH32_hash_t seed)
{
    XXH_ASSERT(statePtr != NULL);
    memset(statePtr, 0, sizeof(*statePtr));
    statePtr->v[0] = seed + XXH_PRIME32_1 + XXH_PRIME32_2;
    statePtr->v[1] = seed + XXH_PRIME32_2;
    statePtr->v[2] = seed + 0;
    statePtr->v[3] = seed - XXH_PRIME32_1;
    return XXH_OK;
}


/*! @ingroup XXH32_family */
XXH_PUBLIC_API XXH_errorcode
XXH32_update(XXH32_state_t* state, const void* input, size_t len)
{
    if (input==NULL) {
        XXH_ASSERT(len == 0);
        return XXH_OK;
    }

    {   const xxh_u8* p = (const xxh_u8*)input;
        const xxh_u8* const bEnd = p + len;

        state->total_len_32 += (XXH32_hash_t)len;
        state->large_len |= (XXH32_hash_t)((len>=16) | (state->total_len_32>=16));

        if (state->memsize + len < 16)  {   /* fill in tmp buffer */
            XXH_memcpy((xxh_u8*)(state->mem32) + state->memsize, input, len);
            state->memsize += (XXH32_hash_t)len;
            return XXH_OK;
        }

        if (state->memsize) {   /* some data left from previous update */
            XXH_memcpy((xxh_u8*)(state->mem32) + state->memsize, input, 16-state->memsize);
            {   const xxh_u32* p32 = state->mem32;
                state->v[0] = XXH32_round(state->v[0], XXH_readLE32(p32)); p32++;
                state->v[1] = XXH32_round(state->v[1], XXH_readLE32(p32)); p32++;
                state->v[2] = XXH32_round(state->v[2], XXH_readLE32(p32)); p32++;
                state->v[3] = XXH32_round(state->v[3], XXH_readLE32(p32));
            }
            p += 16-state->memsize;
            state->memsize = 0;
        }

        if (p <= bEnd-16) {
            const xxh_u8* const limit = bEnd - 16;

            do {
                state->v[0] = XXH32_round(state->v[0], XXH_readLE32(p)); p+=4;
                state->v[1] = XXH32_round(state->v[1], XXH_readLE32(p)); p+=4;
                state->v[2] = XXH32_round(state->v[2], XXH_readLE32(p)); p+=4;
                state->v[3] = XXH32_round(state->v[3], XXH_readLE32(p)); p+=4;
            } while (p<=limit);

        }

        if (p < bEnd) {
            XXH_memcpy(state->mem32, p, (size_t)(bEnd-p));
            state->memsize = (unsigned)(bEnd-p);
        }
    }

    return XXH_OK;
}


/*! @ingroup XXH32_family */
XXH_PUBLIC_API XXH32_hash_t XXH32_digest(const XXH32_state_t* state)
{
    xxh_u32 h32;

    if (state->large_len) {
        h32 = XXH_rotl32(state->v[0], 1)
            + XXH_rotl32(state->v[1], 7)
            + XXH_rotl32(state->v[2], 12)
            + XXH_rotl32(state->v[3], 18);
    } else {
        h32 = state->v[2] /* == seed */ + XXH_PRIME32_5;
    }

    h32 += state->total_len_32;

    return XXH32_finalize(h32, (const xxh_u8*)state->mem32, state->memsize, XXH_aligned);
}
#endif /* !XXH_NO_STREAM */

/*******   Canonical representation   *******/

/*!
 * @ingroup XXH32_family
 * The default return values from XXH functions are unsigned 32 and 64 bit
 * integers.
 *
 * The canonical representation uses big endian convention, the same convention
 * as human-readable numbers (large digits first).
 *
 * This way, hash values can be written into a file or buffer, remaining
 * comparable across different systems.
 *
 * The following functions allow transformation of hash values to and from their
 * canonical format.
 */
XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t hash)
{
    XXH_STATIC_ASSERT(sizeof(XXH32_canonical_t) == sizeof(XXH32_hash_t));
    if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap32(hash);
    XXH_memcpy(dst, &hash, sizeof(*dst));
}
/*! @ingroup XXH32_family */
XXH_PUBLIC_API XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src)
{
    return XXH_readBE32(src);
}


#ifndef XXH_NO_LONG_LONG

/* *******************************************************************
*  64-bit hash functions
*********************************************************************/
/*!
 * @}
 * @ingroup impl
 * @{
 */
/*******   Memory access   *******/

typedef XXH64_hash_t xxh_u64;

#ifdef XXH_OLD_NAMES
#  define U64 xxh_u64
#endif

#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==3))
/*
 * Manual byteshift. Best for old compilers which don't inline memcpy.
 * We actually directly use XXH_readLE64 and XXH_readBE64.
 */
#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==2))

/* Force direct memory access. Only works on CPU which support unaligned memory access in hardware */
static xxh_u64 XXH_read64(const void* memPtr)
{
    return *(const xxh_u64*) memPtr;
}

#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==1))

/*
 * __attribute__((aligned(1))) is supported by gcc and clang. Originally the
 * documentation claimed that it only increased the alignment, but actually it
 * can decrease it on gcc, clang, and icc:
 * https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69502,
 * https://gcc.godbolt.org/z/xYez1j67Y.
 */
#ifdef XXH_OLD_NAMES
typedef union { xxh_u32 u32; xxh_u64 u64; } __attribute__((packed)) unalign64;
#endif
static xxh_u64 XXH_read64(const void* ptr)
{
    typedef __attribute__((aligned(1))) xxh_u64 xxh_unalign64;
    return *((const xxh_unalign64*)ptr);
}

#else

/*
 * Portable and safe solution. Generally efficient.
 * see: https://fastcompression.blogspot.com/2015/08/accessing-unaligned-memory.html
 */
static xxh_u64 XXH_read64(const void* memPtr)
{
    xxh_u64 val;
    XXH_memcpy(&val, memPtr, sizeof(val));
    return val;
}

#endif   /* XXH_FORCE_DIRECT_MEMORY_ACCESS */

#if defined(_MSC_VER)     /* Visual Studio */
#  define XXH_swap64 _byteswap_uint64
#elif XXH_GCC_VERSION >= 403
#  define XXH_swap64 __builtin_bswap64
#else
static xxh_u64 XXH_swap64(xxh_u64 x)
{
    return  ((x << 56) & 0xff00000000000000ULL) |
            ((x << 40) & 0x00ff000000000000ULL) |
            ((x << 24) & 0x0000ff0000000000ULL) |
            ((x << 8)  & 0x000000ff00000000ULL) |
            ((x >> 8)  & 0x00000000ff000000ULL) |
            ((x >> 24) & 0x0000000000ff0000ULL) |
            ((x >> 40) & 0x000000000000ff00ULL) |
            ((x >> 56) & 0x00000000000000ffULL);
}
#endif


/* XXH_FORCE_MEMORY_ACCESS==3 is an endian-independent byteshift load. */
#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==3))

XXH_FORCE_INLINE xxh_u64 XXH_readLE64(const void* memPtr)
{
    const xxh_u8* bytePtr = (const xxh_u8 *)memPtr;
    return bytePtr[0]
         | ((xxh_u64)bytePtr[1] << 8)
         | ((xxh_u64)bytePtr[2] << 16)
         | ((xxh_u64)bytePtr[3] << 24)
         | ((xxh_u64)bytePtr[4] << 32)
         | ((xxh_u64)bytePtr[5] << 40)
         | ((xxh_u64)bytePtr[6] << 48)
         | ((xxh_u64)bytePtr[7] << 56);
}

XXH_FORCE_INLINE xxh_u64 XXH_readBE64(const void* memPtr)
{
    const xxh_u8* bytePtr = (const xxh_u8 *)memPtr;
    return bytePtr[7]
         | ((xxh_u64)bytePtr[6] << 8)
         | ((xxh_u64)bytePtr[5] << 16)
         | ((xxh_u64)bytePtr[4] << 24)
         | ((xxh_u64)bytePtr[3] << 32)
         | ((xxh_u64)bytePtr[2] << 40)
         | ((xxh_u64)bytePtr[1] << 48)
         | ((xxh_u64)bytePtr[0] << 56);
}

#else
XXH_FORCE_INLINE xxh_u64 XXH_readLE64(const void* ptr)
{
    return XXH_CPU_LITTLE_ENDIAN ? XXH_read64(ptr) : XXH_swap64(XXH_read64(ptr));
}

static xxh_u64 XXH_readBE64(const void* ptr)
{
    return XXH_CPU_LITTLE_ENDIAN ? XXH_swap64(XXH_read64(ptr)) : XXH_read64(ptr);
}
#endif

XXH_FORCE_INLINE xxh_u64
XXH_readLE64_align(const void* ptr, XXH_alignment align)
{
    if (align==XXH_unaligned)
        return XXH_readLE64(ptr);
    else
        return XXH_CPU_LITTLE_ENDIAN ? *(const xxh_u64*)ptr : XXH_swap64(*(const xxh_u64*)ptr);
}


/*******   xxh64   *******/
/*!
 * @}
 * @defgroup XXH64_impl XXH64 implementation
 * @ingroup impl
 *
 * Details on the XXH64 implementation.
 * @{
 */
/* #define rather that static const, to be used as initializers */
#define XXH_PRIME64_1  0x9E3779B185EBCA87ULL  /*!< 0b1001111000110111011110011011000110000101111010111100101010000111 */
#define XXH_PRIME64_2  0xC2B2AE3D27D4EB4FULL  /*!< 0b1100001010110010101011100011110100100111110101001110101101001111 */
#define XXH_PRIME64_3  0x165667B19E3779F9ULL  /*!< 0b0001011001010110011001111011000110011110001101110111100111111001 */
#define XXH_PRIME64_4  0x85EBCA77C2B2AE63ULL  /*!< 0b1000010111101011110010100111011111000010101100101010111001100011 */
#define XXH_PRIME64_5  0x27D4EB2F165667C5ULL  /*!< 0b0010011111010100111010110010111100010110010101100110011111000101 */

#ifdef XXH_OLD_NAMES
#  define PRIME64_1 XXH_PRIME64_1
#  define PRIME64_2 XXH_PRIME64_2
#  define PRIME64_3 XXH_PRIME64_3
#  define PRIME64_4 XXH_PRIME64_4
#  define PRIME64_5 XXH_PRIME64_5
#endif

/*! @copydoc XXH32_round */
static xxh_u64 XXH64_round(xxh_u64 acc, xxh_u64 input)
{
    acc += input * XXH_PRIME64_2;
    acc  = XXH_rotl64(acc, 31);
    acc *= XXH_PRIME64_1;
    return acc;
}

static xxh_u64 XXH64_mergeRound(xxh_u64 acc, xxh_u64 val)
{
    val  = XXH64_round(0, val);
    acc ^= val;
    acc  = acc * XXH_PRIME64_1 + XXH_PRIME64_4;
    return acc;
}

/*! @copydoc XXH32_avalanche */
static xxh_u64 XXH64_avalanche(xxh_u64 hash)
{
    hash ^= hash >> 33;
    hash *= XXH_PRIME64_2;
    hash ^= hash >> 29;
    hash *= XXH_PRIME64_3;
    hash ^= hash >> 32;
    return hash;
}


#define XXH_get64bits(p) XXH_readLE64_align(p, align)

/*!
 * @internal
 * @brief Processes the last 0-31 bytes of @p ptr.
 *
 * There may be up to 31 bytes remaining to consume from the input.
 * This final stage will digest them to ensure that all input bytes are present
 * in the final mix.
 *
 * @param hash The hash to finalize.
 * @param ptr The pointer to the remaining input.
 * @param len The remaining length, modulo 32.
 * @param align Whether @p ptr is aligned.
 * @return The finalized hash
 * @see XXH32_finalize().
 */
static XXH_PUREF xxh_u64
XXH64_finalize(xxh_u64 hash, const xxh_u8* ptr, size_t len, XXH_alignment align)
{
    if (ptr==NULL) XXH_ASSERT(len == 0);
    len &= 31;
    while (len >= 8) {
        xxh_u64 const k1 = XXH64_round(0, XXH_get64bits(ptr));
        ptr += 8;
        hash ^= k1;
        hash  = XXH_rotl64(hash,27) * XXH_PRIME64_1 + XXH_PRIME64_4;
        len -= 8;
    }
    if (len >= 4) {
        hash ^= (xxh_u64)(XXH_get32bits(ptr)) * XXH_PRIME64_1;
        ptr += 4;
        hash = XXH_rotl64(hash, 23) * XXH_PRIME64_2 + XXH_PRIME64_3;
        len -= 4;
    }
    while (len > 0) {
        hash ^= (*ptr++) * XXH_PRIME64_5;
        hash = XXH_rotl64(hash, 11) * XXH_PRIME64_1;
        --len;
    }
    return  XXH64_avalanche(hash);
}

#ifdef XXH_OLD_NAMES
#  define PROCESS1_64 XXH_PROCESS1_64
#  define PROCESS4_64 XXH_PROCESS4_64
#  define PROCESS8_64 XXH_PROCESS8_64
#else
#  undef XXH_PROCESS1_64
#  undef XXH_PROCESS4_64
#  undef XXH_PROCESS8_64
#endif

/*!
 * @internal
 * @brief The implementation for @ref XXH64().
 *
 * @param input , len , seed Directly passed from @ref XXH64().
 * @param align Whether @p input is aligned.
 * @return The calculated hash.
 */
XXH_FORCE_INLINE XXH_PUREF xxh_u64
XXH64_endian_align(const xxh_u8* input, size_t len, xxh_u64 seed, XXH_alignment align)
{
    xxh_u64 h64;
    if (input==NULL) XXH_ASSERT(len == 0);

    if (len>=32) {
        const xxh_u8* const bEnd = input + len;
        const xxh_u8* const limit = bEnd - 31;
        xxh_u64 v1 = seed + XXH_PRIME64_1 + XXH_PRIME64_2;
        xxh_u64 v2 = seed + XXH_PRIME64_2;
        xxh_u64 v3 = seed + 0;
        xxh_u64 v4 = seed - XXH_PRIME64_1;

        do {
            v1 = XXH64_round(v1, XXH_get64bits(input)); input+=8;
            v2 = XXH64_round(v2, XXH_get64bits(input)); input+=8;
            v3 = XXH64_round(v3, XXH_get64bits(input)); input+=8;
            v4 = XXH64_round(v4, XXH_get64bits(input)); input+=8;
        } while (input<limit);

        h64 = XXH_rotl64(v1, 1) + XXH_rotl64(v2, 7) + XXH_rotl64(v3, 12) + XXH_rotl64(v4, 18);
        h64 = XXH64_mergeRound(h64, v1);
        h64 = XXH64_mergeRound(h64, v2);
        h64 = XXH64_mergeRound(h64, v3);
        h64 = XXH64_mergeRound(h64, v4);

    } else {
        h64  = seed + XXH_PRIME64_5;
    }

    h64 += (xxh_u64) len;

    return XXH64_finalize(h64, input, len, align);
}


/*! @ingroup XXH64_family */
XXH_PUBLIC_API XXH64_hash_t XXH64 (XXH_NOESCAPE const void* input, size_t len, XXH64_hash_t seed)
{
#if !defined(XXH_NO_STREAM) && XXH_SIZE_OPT >= 2
    /* Simple version, good for code maintenance, but unfortunately slow for small inputs */
    XXH64_state_t state;
    XXH64_reset(&state, seed);
    XXH64_update(&state, (const xxh_u8*)input, len);
    return XXH64_digest(&state);
#else
    if (XXH_FORCE_ALIGN_CHECK) {
        if ((((size_t)input) & 7)==0) {  /* Input is aligned, let's leverage the speed advantage */
            return XXH64_endian_align((const xxh_u8*)input, len, seed, XXH_aligned);
    }   }

    return XXH64_endian_align((const xxh_u8*)input, len, seed, XXH_unaligned);

#endif
}

/*******   Hash Streaming   *******/
#ifndef XXH_NO_STREAM
/*! @ingroup XXH64_family*/
XXH_PUBLIC_API XXH64_state_t* XXH64_createState(void)
{
    return (XXH64_state_t*)XXH_malloc(sizeof(XXH64_state_t));
}
/*! @ingroup XXH64_family */
XXH_PUBLIC_API XXH_errorcode XXH64_freeState(XXH64_state_t* statePtr)
{
    XXH_free(statePtr);
    return XXH_OK;
}

/*! @ingroup XXH64_family */
XXH_PUBLIC_API void XXH64_copyState(XXH_NOESCAPE XXH64_state_t* dstState, const XXH64_state_t* srcState)
{
    XXH_memcpy(dstState, srcState, sizeof(*dstState));
}

/*! @ingroup XXH64_family */
XXH_PUBLIC_API XXH_errorcode XXH64_reset(XXH_NOESCAPE XXH64_state_t* statePtr, XXH64_hash_t seed)
{
    XXH_ASSERT(statePtr != NULL);
    memset(statePtr, 0, sizeof(*statePtr));
    statePtr->v[0] = seed + XXH_PRIME64_1 + XXH_PRIME64_2;
    statePtr->v[1] = seed + XXH_PRIME64_2;
    statePtr->v[2] = seed + 0;
    statePtr->v[3] = seed - XXH_PRIME64_1;
    return XXH_OK;
}

/*! @ingroup XXH64_family */
XXH_PUBLIC_API XXH_errorcode
XXH64_update (XXH_NOESCAPE XXH64_state_t* state, XXH_NOESCAPE const void* input, size_t len)
{
    if (input==NULL) {
        XXH_ASSERT(len == 0);
        return XXH_OK;
    }

    {   const xxh_u8* p = (const xxh_u8*)input;
        const xxh_u8* const bEnd = p + len;

        state->total_len += len;

        if (state->memsize + len < 32) {  /* fill in tmp buffer */
            XXH_memcpy(((xxh_u8*)state->mem64) + state->memsize, input, len);
            state->memsize += (xxh_u32)len;
            return XXH_OK;
        }

        if (state->memsize) {   /* tmp buffer is full */
            XXH_memcpy(((xxh_u8*)state->mem64) + state->memsize, input, 32-state->memsize);
            state->v[0] = XXH64_round(state->v[0], XXH_readLE64(state->mem64+0));
            state->v[1] = XXH64_round(state->v[1], XXH_readLE64(state->mem64+1));
            state->v[2] = XXH64_round(state->v[2], XXH_readLE64(state->mem64+2));
            state->v[3] = XXH64_round(state->v[3], XXH_readLE64(state->mem64+3));
            p += 32 - state->memsize;
            state->memsize = 0;
        }

        if (p+32 <= bEnd) {
            const xxh_u8* const limit = bEnd - 32;

            do {
                state->v[0] = XXH64_round(state->v[0], XXH_readLE64(p)); p+=8;
                state->v[1] = XXH64_round(state->v[1], XXH_readLE64(p)); p+=8;
                state->v[2] = XXH64_round(state->v[2], XXH_readLE64(p)); p+=8;
                state->v[3] = XXH64_round(state->v[3], XXH_readLE64(p)); p+=8;
            } while (p<=limit);

        }

        if (p < bEnd) {
            XXH_memcpy(state->mem64, p, (size_t)(bEnd-p));
            state->memsize = (unsigned)(bEnd-p);
        }
    }

    return XXH_OK;
}


/*! @ingroup XXH64_family */
XXH_PUBLIC_API XXH64_hash_t XXH64_digest(XXH_NOESCAPE const XXH64_state_t* state)
{
    xxh_u64 h64;

    if (state->total_len >= 32) {
        h64 = XXH_rotl64(state->v[0], 1) + XXH_rotl64(state->v[1], 7) + XXH_rotl64(state->v[2], 12) + XXH_rotl64(state->v[3], 18);
        h64 = XXH64_mergeRound(h64, state->v[0]);
        h64 = XXH64_mergeRound(h64, state->v[1]);
        h64 = XXH64_mergeRound(h64, state->v[2]);
        h64 = XXH64_mergeRound(h64, state->v[3]);
    } else {
        h64  = state->v[2] /*seed*/ + XXH_PRIME64_5;
    }

    h64 += (xxh_u64) state->total_len;

    return XXH64_finalize(h64, (const xxh_u8*)state->mem64, (size_t)state->total_len, XXH_aligned);
}
#endif /* !XXH_NO_STREAM */

/******* Canonical representation   *******/

/*! @ingroup XXH64_family */
XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH_NOESCAPE XXH64_canonical_t* dst, XXH64_hash_t hash)
{
    XXH_STATIC_ASSERT(sizeof(XXH64_canonical_t) == sizeof(XXH64_hash_t));
    if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap64(hash);
    XXH_memcpy(dst, &hash, sizeof(*dst));
}

/*! @ingroup XXH64_family */
XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(XXH_NOESCAPE const XXH64_canonical_t* src)
{
    return XXH_readBE64(src);
}

#ifndef XXH_NO_XXH3

/* *********************************************************************
*  XXH3
*  New generation hash designed for speed on small keys and vectorization
************************************************************************ */
/*!
 * @}
 * @defgroup XXH3_impl XXH3 implementation
 * @ingroup impl
 * @{
 */

/* ===   Compiler specifics   === */

#if ((defined(sun) || defined(__sun)) && defined(__cplusplus) && __cplusplus) /* Solaris includes __STDC_VERSION__ with C++. Tested with GCC 5.5 */
#  define XXH_RESTRICT   /* disable */
#elif defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* >= C99 */
#  define XXH_RESTRICT   restrict
#elif (defined (__GNUC__) && ((__GNUC__ > 3) || (__GNUC__ == 3 && __GNUC_MINOR__ >= 1))) \
   || (defined (__clang__)) \
   || (defined (_MSC_VER) && (_MSC_VER >= 1400)) \
   || (defined (__INTEL_COMPILER) && (__INTEL_COMPILER >= 1300))
/*
 * There are a LOT more compilers that recognize __restrict but this
 * covers the major ones.
 */
#  define XXH_RESTRICT   __restrict
#else
#  define XXH_RESTRICT   /* disable */
#endif

#if (defined(__GNUC__) && (__GNUC__ >= 3))  \
  || (defined(__INTEL_COMPILER) && (__INTEL_COMPILER >= 800)) \
  || defined(__clang__)
#    define XXH_likely(x) __builtin_expect(x, 1)
#    define XXH_unlikely(x) __builtin_expect(x, 0)
#else
#    define XXH_likely(x) (x)
#    define XXH_unlikely(x) (x)
#endif

#ifndef XXH_HAS_INCLUDE
#  ifdef __has_include
#    define XXH_HAS_INCLUDE(x) __has_include(x)
#  else
#    define XXH_HAS_INCLUDE(x) 0
#  endif
#endif

#if defined(__GNUC__) || defined(__clang__)
#  if defined(__ARM_FEATURE_SVE)
#    include <arm_sve.h>
#  endif
#  if defined(__ARM_NEON__) || defined(__ARM_NEON) \
   || (defined(_M_ARM) && _M_ARM >= 7) \
   || defined(_M_ARM64) || defined(_M_ARM64EC) \
   || (defined(__wasm_simd128__) && XXH_HAS_INCLUDE(<arm_neon.h>)) /* WASM SIMD128 via SIMDe */
#    define inline __inline__  /* circumvent a clang bug */
#    include <arm_neon.h>
#    undef inline
#  elif defined(__AVX2__)
#    include <immintrin.h>
#  elif defined(__SSE2__)
#    include <emmintrin.h>
#  endif
#endif

#if defined(_MSC_VER)
#  include <intrin.h>
#endif

/*
 * One goal of XXH3 is to make it fast on both 32-bit and 64-bit, while
 * remaining a true 64-bit/128-bit hash function.
 *
 * This is done by prioritizing a subset of 64-bit operations that can be
 * emulated without too many steps on the average 32-bit machine.
 *
 * For example, these two lines seem similar, and run equally fast on 64-bit:
 *
 *   xxh_u64 x;
 *   x ^= (x >> 47); // good
 *   x ^= (x >> 13); // bad
 *
 * However, to a 32-bit machine, there is a major difference.
 *
 * x ^= (x >> 47) looks like this:
 *
 *   x.lo ^= (x.hi >> (47 - 32));
 *
 * while x ^= (x >> 13) looks like this:
 *
 *   // note: funnel shifts are not usually cheap.
 *   x.lo ^= (x.lo >> 13) | (x.hi << (32 - 13));
 *   x.hi ^= (x.hi >> 13);
 *
 * The first one is significantly faster than the second, simply because the
 * shift is larger than 32. This means:
 *  - All the bits we need are in the upper 32 bits, so we can ignore the lower
 *    32 bits in the shift.
 *  - The shift result will always fit in the lower 32 bits, and therefore,
 *    we can ignore the upper 32 bits in the xor.
 *
 * Thanks to this optimization, XXH3 only requires these features to be efficient:
 *
 *  - Usable unaligned access
 *  - A 32-bit or 64-bit ALU
 *      - If 32-bit, a decent ADC instruction
 *  - A 32 or 64-bit multiply with a 64-bit result
 *  - For the 128-bit variant, a decent byteswap helps short inputs.
 *
 * The first two are already required by XXH32, and almost all 32-bit and 64-bit
 * platforms which can run XXH32 can run XXH3 efficiently.
 *
 * Thumb-1, the classic 16-bit only subset of ARM's instruction set, is one
 * notable exception.
 *
 * First of all, Thumb-1 lacks support for the UMULL instruction which
 * performs the important long multiply. This means numerous __aeabi_lmul
 * calls.
 *
 * Second of all, the 8 functional registers are just not enough.
 * Setup for __aeabi_lmul, byteshift loads, pointers, and all arithmetic need
 * Lo registers, and this shuffling results in thousands more MOVs than A32.
 *
 * A32 and T32 don't have this limitation. They can access all 14 registers,
 * do a 32->64 multiply with UMULL, and the flexible operand allowing free
 * shifts is helpful, too.
 *
 * Therefore, we do a quick sanity check.
 *
 * If compiling Thumb-1 for a target which supports ARM instructions, we will
 * emit a warning, as it is not a "sane" platform to compile for.
 *
 * Usually, if this happens, it is because of an accident and you probably need
 * to specify -march, as you likely meant to compile for a newer architecture.
 *
 * Credit: large sections of the vectorial and asm source code paths
 *         have been contributed by @easyaspi314
 */
#if defined(__thumb__) && !defined(__thumb2__) && defined(__ARM_ARCH_ISA_ARM)
#   warning "XXH3 is highly inefficient without ARM or Thumb-2."
#endif

/* ==========================================
 * Vectorization detection
 * ========================================== */

#ifdef XXH_DOXYGEN
/*!
 * @ingroup tuning
 * @brief Overrides the vectorization implementation chosen for XXH3.
 *
 * Can be defined to 0 to disable SIMD or any of the values mentioned in
 * @ref XXH_VECTOR_TYPE.
 *
 * If this is not defined, it uses predefined macros to determine the best
 * implementation.
 */
#  define XXH_VECTOR XXH_SCALAR
/*!
 * @ingroup tuning
 * @brief Possible values for @ref XXH_VECTOR.
 *
 * Note that these are actually implemented as macros.
 *
 * If this is not defined, it is detected automatically.
 * internal macro XXH_X86DISPATCH overrides this.
 */
enum XXH_VECTOR_TYPE /* fake enum */ {
    XXH_SCALAR = 0,  /*!< Portable scalar version */
    XXH_SSE2   = 1,  /*!<
                      * SSE2 for Pentium 4, Opteron, all x86_64.
                      *
                      * @note SSE2 is also guaranteed on Windows 10, macOS, and
                      * Android x86.
                      */
    XXH_AVX2   = 2,  /*!< AVX2 for Haswell and Bulldozer */
    XXH_AVX512 = 3,  /*!< AVX512 for Skylake and Icelake */
    XXH_NEON   = 4,  /*!<
                       * NEON for most ARMv7-A, all AArch64, and WASM SIMD128
                       * via the SIMDeverywhere polyfill provided with the
                       * Emscripten SDK.
                       */
    XXH_VSX    = 5,  /*!< VSX and ZVector for POWER8/z13 (64-bit) */
    XXH_SVE    = 6,  /*!< SVE for some ARMv8-A and ARMv9-A */
};
/*!
 * @ingroup tuning
 * @brief Selects the minimum alignment for XXH3's accumulators.
 *
 * When using SIMD, this should match the alignment required for said vector
 * type, so, for example, 32 for AVX2.
 *
 * Default: Auto detected.
 */
#  define XXH_ACC_ALIGN 8
#endif

/* Actual definition */
#ifndef XXH_DOXYGEN
#  define XXH_SCALAR 0
#  define XXH_SSE2   1
#  define XXH_AVX2   2
#  define XXH_AVX512 3
#  define XXH_NEON   4
#  define XXH_VSX    5
#  define XXH_SVE    6
#endif

#ifndef XXH_VECTOR    /* can be defined on command line */
#  if defined(__ARM_FEATURE_SVE)
#    define XXH_VECTOR XXH_SVE
#  elif ( \
        defined(__ARM_NEON__) || defined(__ARM_NEON) /* gcc */ \
     || defined(_M_ARM) || defined(_M_ARM64) || defined(_M_ARM64EC) /* msvc */ \
     || (defined(__wasm_simd128__) && XXH_HAS_INCLUDE(<arm_neon.h>)) /* wasm simd128 via SIMDe */ \
   ) && ( \
        defined(_WIN32) || defined(__LITTLE_ENDIAN__) /* little endian only */ \
    || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \
   )
#    define XXH_VECTOR XXH_NEON
#  elif defined(__AVX512F__)
#    define XXH_VECTOR XXH_AVX512
#  elif defined(__AVX2__)
#    define XXH_VECTOR XXH_AVX2
#  elif defined(__SSE2__) || defined(_M_AMD64) || defined(_M_X64) || (defined(_M_IX86_FP) && (_M_IX86_FP == 2))
#    define XXH_VECTOR XXH_SSE2
#  elif (defined(__PPC64__) && defined(__POWER8_VECTOR__)) \
     || (defined(__s390x__) && defined(__VEC__)) \
     && defined(__GNUC__) /* TODO: IBM XL */
#    define XXH_VECTOR XXH_VSX
#  else
#    define XXH_VECTOR XXH_SCALAR
#  endif
#endif

/* __ARM_FEATURE_SVE is only supported by GCC & Clang. */
#if (XXH_VECTOR == XXH_SVE) && !defined(__ARM_FEATURE_SVE)
#  ifdef _MSC_VER
#    pragma warning(once : 4606)
#  else
#    warning "__ARM_FEATURE_SVE isn't supported. Use SCALAR instead."
#  endif
#  undef XXH_VECTOR
#  define XXH_VECTOR XXH_SCALAR
#endif

/*
 * Controls the alignment of the accumulator,
 * for compatibility with aligned vector loads, which are usually faster.
 */
#ifndef XXH_ACC_ALIGN
#  if defined(XXH_X86DISPATCH)
#     define XXH_ACC_ALIGN 64  /* for compatibility with avx512 */
#  elif XXH_VECTOR == XXH_SCALAR  /* scalar */
#     define XXH_ACC_ALIGN 8
#  elif XXH_VECTOR == XXH_SSE2  /* sse2 */
#     define XXH_ACC_ALIGN 16
#  elif XXH_VECTOR == XXH_AVX2  /* avx2 */
#     define XXH_ACC_ALIGN 32
#  elif XXH_VECTOR == XXH_NEON  /* neon */
#     define XXH_ACC_ALIGN 16
#  elif XXH_VECTOR == XXH_VSX   /* vsx */
#     define XXH_ACC_ALIGN 16
#  elif XXH_VECTOR == XXH_AVX512  /* avx512 */
#     define XXH_ACC_ALIGN 64
#  elif XXH_VECTOR == XXH_SVE   /* sve */
#     define XXH_ACC_ALIGN 64
#  endif
#endif

#if defined(XXH_X86DISPATCH) || XXH_VECTOR == XXH_SSE2 \
    || XXH_VECTOR == XXH_AVX2 || XXH_VECTOR == XXH_AVX512
#  define XXH_SEC_ALIGN XXH_ACC_ALIGN
#elif XXH_VECTOR == XXH_SVE
#  define XXH_SEC_ALIGN XXH_ACC_ALIGN
#else
#  define XXH_SEC_ALIGN 8
#endif

#if defined(__GNUC__) || defined(__clang__)
#  define XXH_ALIASING __attribute__((may_alias))
#else
#  define XXH_ALIASING /* nothing */
#endif

/*
 * UGLY HACK:
 * GCC usually generates the best code with -O3 for xxHash.
 *
 * However, when targeting AVX2, it is overzealous in its unrolling resulting
 * in code roughly 3/4 the speed of Clang.
 *
 * There are other issues, such as GCC splitting _mm256_loadu_si256 into
 * _mm_loadu_si128 + _mm256_inserti128_si256. This is an optimization which
 * only applies to Sandy and Ivy Bridge... which don't even support AVX2.
 *
 * That is why when compiling the AVX2 version, it is recommended to use either
 *   -O2 -mavx2 -march=haswell
 * or
 *   -O2 -mavx2 -mno-avx256-split-unaligned-load
 * for decent performance, or to use Clang instead.
 *
 * Fortunately, we can control the first one with a pragma that forces GCC into
 * -O2, but the other one we can't control without "failed to inline always
 * inline function due to target mismatch" warnings.
 */
#if XXH_VECTOR == XXH_AVX2 /* AVX2 */ \
  && defined(__GNUC__) && !defined(__clang__) /* GCC, not Clang */ \
  && defined(__OPTIMIZE__) && XXH_SIZE_OPT <= 0 /* respect -O0 and -Os */
#  pragma GCC push_options
#  pragma GCC optimize("-O2")
#endif

#if XXH_VECTOR == XXH_NEON

/*
 * UGLY HACK: While AArch64 GCC on Linux does not seem to care, on macOS, GCC -O3
 * optimizes out the entire hashLong loop because of the aliasing violation.
 *
 * However, GCC is also inefficient at load-store optimization with vld1q/vst1q,
 * so the only option is to mark it as aliasing.
 */
typedef uint64x2_t xxh_aliasing_uint64x2_t XXH_ALIASING;

/*!
 * @internal
 * @brief `vld1q_u64` but faster and alignment-safe.
 *
 * On AArch64, unaligned access is always safe, but on ARMv7-a, it is only
 * *conditionally* safe (`vld1` has an alignment bit like `movdq[ua]` in x86).
 *
 * GCC for AArch64 sees `vld1q_u8` as an intrinsic instead of a load, so it
 * prohibits load-store optimizations. Therefore, a direct dereference is used.
 *
 * Otherwise, `vld1q_u8` is used with `vreinterpretq_u8_u64` to do a safe
 * unaligned load.
 */
#if defined(__aarch64__) && defined(__GNUC__) && !defined(__clang__)
XXH_FORCE_INLINE uint64x2_t XXH_vld1q_u64(void const* ptr) /* silence -Wcast-align */
{
    return *(xxh_aliasing_uint64x2_t const *)ptr;
}
#else
XXH_FORCE_INLINE uint64x2_t XXH_vld1q_u64(void const* ptr)
{
    return vreinterpretq_u64_u8(vld1q_u8((uint8_t const*)ptr));
}
#endif

/*!
 * @internal
 * @brief `vmlal_u32` on low and high halves of a vector.
 *
 * This is a workaround for AArch64 GCC < 11 which implemented arm_neon.h with
 * inline assembly and were therefore incapable of merging the `vget_{low, high}_u32`
 * with `vmlal_u32`.
 */
#if defined(__aarch64__) && defined(__GNUC__) && !defined(__clang__) && __GNUC__ < 11
XXH_FORCE_INLINE uint64x2_t
XXH_vmlal_low_u32(uint64x2_t acc, uint32x4_t lhs, uint32x4_t rhs)
{
    /* Inline assembly is the only way */
    __asm__("umlal   %0.2d, %1.2s, %2.2s" : "+w" (acc) : "w" (lhs), "w" (rhs));
    return acc;
}
XXH_FORCE_INLINE uint64x2_t
XXH_vmlal_high_u32(uint64x2_t acc, uint32x4_t lhs, uint32x4_t rhs)
{
    /* This intrinsic works as expected */
    return vmlal_high_u32(acc, lhs, rhs);
}
#else
/* Portable intrinsic versions */
XXH_FORCE_INLINE uint64x2_t
XXH_vmlal_low_u32(uint64x2_t acc, uint32x4_t lhs, uint32x4_t rhs)
{
    return vmlal_u32(acc, vget_low_u32(lhs), vget_low_u32(rhs));
}
/*! @copydoc XXH_vmlal_low_u32
 * Assume the compiler converts this to vmlal_high_u32 on aarch64 */
XXH_FORCE_INLINE uint64x2_t
XXH_vmlal_high_u32(uint64x2_t acc, uint32x4_t lhs, uint32x4_t rhs)
{
    return vmlal_u32(acc, vget_high_u32(lhs), vget_high_u32(rhs));
}
#endif

/*!
 * @ingroup tuning
 * @brief Controls the NEON to scalar ratio for XXH3
 *
 * This can be set to 2, 4, 6, or 8.
 *
 * ARM Cortex CPUs are _very_ sensitive to how their pipelines are used.
 *
 * For example, the Cortex-A73 can dispatch 3 micro-ops per cycle, but only 2 of those
 * can be NEON. If you are only using NEON instructions, you are only using 2/3 of the CPU
 * bandwidth.
 *
 * This is even more noticeable on the more advanced cores like the Cortex-A76 which
 * can dispatch 8 micro-ops per cycle, but still only 2 NEON micro-ops at once.
 *
 * Therefore, to make the most out of the pipeline, it is beneficial to run 6 NEON lanes
 * and 2 scalar lanes, which is chosen by default.
 *
 * This does not apply to Apple processors or 32-bit processors, which run better with
 * full NEON. These will default to 8. Additionally, size-optimized builds run 8 lanes.
 *
 * This change benefits CPUs with large micro-op buffers without negatively affecting
 * most other CPUs:
 *
 *  | Chipset               | Dispatch type       | NEON only | 6:2 hybrid | Diff. |
 *  |:----------------------|:--------------------|----------:|-----------:|------:|
 *  | Snapdragon 730 (A76)  | 2 NEON/8 micro-ops  |  8.8 GB/s |  10.1 GB/s |  ~16% |
 *  | Snapdragon 835 (A73)  | 2 NEON/3 micro-ops  |  5.1 GB/s |   5.3 GB/s |   ~5% |
 *  | Marvell PXA1928 (A53) | In-order dual-issue |  1.9 GB/s |   1.9 GB/s |    0% |
 *  | Apple M1              | 4 NEON/8 micro-ops  | 37.3 GB/s |  36.1 GB/s |  ~-3% |
 *
 * It also seems to fix some bad codegen on GCC, making it almost as fast as clang.
 *
 * When using WASM SIMD128, if this is 2 or 6, SIMDe will scalarize 2 of the lanes meaning
 * it effectively becomes worse 4.
 *
 * @see XXH3_accumulate_512_neon()
 */
# ifndef XXH3_NEON_LANES
#  if (defined(__aarch64__) || defined(__arm64__) || defined(_M_ARM64) || defined(_M_ARM64EC)) \
   && !defined(__APPLE__) && XXH_SIZE_OPT <= 0
#   define XXH3_NEON_LANES 6
#  else
#   define XXH3_NEON_LANES XXH_ACC_NB
#  endif
# endif
#endif  /* XXH_VECTOR == XXH_NEON */

/*
 * VSX and Z Vector helpers.
 *
 * This is very messy, and any pull requests to clean this up are welcome.
 *
 * There are a lot of problems with supporting VSX and s390x, due to
 * inconsistent intrinsics, spotty coverage, and multiple endiannesses.
 */
#if XXH_VECTOR == XXH_VSX
/* Annoyingly, these headers _may_ define three macros: `bool`, `vector`,
 * and `pixel`. This is a problem for obvious reasons.
 *
 * These keywords are unnecessary; the spec literally says they are
 * equivalent to `__bool`, `__vector`, and `__pixel` and may be undef'd
 * after including the header.
 *
 * We use pragma push_macro/pop_macro to keep the namespace clean. */
#  pragma push_macro("bool")
#  pragma push_macro("vector")
#  pragma push_macro("pixel")
/* silence potential macro redefined warnings */
#  undef bool
#  undef vector
#  undef pixel

#  if defined(__s390x__)
#    include <s390intrin.h>
#  else
#    include <altivec.h>
#  endif

/* Restore the original macro values, if applicable. */
#  pragma pop_macro("pixel")
#  pragma pop_macro("vector")
#  pragma pop_macro("bool")

typedef __vector unsigned long long xxh_u64x2;
typedef __vector unsigned char xxh_u8x16;
typedef __vector unsigned xxh_u32x4;

/*
 * UGLY HACK: Similar to aarch64 macOS GCC, s390x GCC has the same aliasing issue.
 */
typedef xxh_u64x2 xxh_aliasing_u64x2 XXH_ALIASING;

# ifndef XXH_VSX_BE
#  if defined(__BIG_ENDIAN__) \
  || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)
#    define XXH_VSX_BE 1
#  elif defined(__VEC_ELEMENT_REG_ORDER__) && __VEC_ELEMENT_REG_ORDER__ == __ORDER_BIG_ENDIAN__
#    warning "-maltivec=be is not recommended. Please use native endianness."
#    define XXH_VSX_BE 1
#  else
#    define XXH_VSX_BE 0
#  endif
# endif /* !defined(XXH_VSX_BE) */

# if XXH_VSX_BE
#  if defined(__POWER9_VECTOR__) || (defined(__clang__) && defined(__s390x__))
#    define XXH_vec_revb vec_revb
#  else
/*!
 * A polyfill for POWER9's vec_revb().
 */
XXH_FORCE_INLINE xxh_u64x2 XXH_vec_revb(xxh_u64x2 val)
{
    xxh_u8x16 const vByteSwap = { 0x07, 0x06, 0x05, 0x04, 0x03, 0x02, 0x01, 0x00,
                                  0x0F, 0x0E, 0x0D, 0x0C, 0x0B, 0x0A, 0x09, 0x08 };
    return vec_perm(val, val, vByteSwap);
}
#  endif
# endif /* XXH_VSX_BE */

/*!
 * Performs an unaligned vector load and byte swaps it on big endian.
 */
XXH_FORCE_INLINE xxh_u64x2 XXH_vec_loadu(const void *ptr)
{
    xxh_u64x2 ret;
    XXH_memcpy(&ret, ptr, sizeof(xxh_u64x2));
# if XXH_VSX_BE
    ret = XXH_vec_revb(ret);
# endif
    return ret;
}

/*
 * vec_mulo and vec_mule are very problematic intrinsics on PowerPC
 *
 * These intrinsics weren't added until GCC 8, despite existing for a while,
 * and they are endian dependent. Also, their meaning swap depending on version.
 * */
# if defined(__s390x__)
 /* s390x is always big endian, no issue on this platform */
#  define XXH_vec_mulo vec_mulo
#  define XXH_vec_mule vec_mule
# elif defined(__clang__) && XXH_HAS_BUILTIN(__builtin_altivec_vmuleuw) && !defined(__ibmxl__)
/* Clang has a better way to control this, we can just use the builtin which doesn't swap. */
 /* The IBM XL Compiler (which defined __clang__) only implements the vec_* operations */
#  define XXH_vec_mulo __builtin_altivec_vmulouw
#  define XXH_vec_mule __builtin_altivec_vmuleuw
# else
/* gcc needs inline assembly */
/* Adapted from https://github.com/google/highwayhash/blob/master/highwayhash/hh_vsx.h. */
XXH_FORCE_INLINE xxh_u64x2 XXH_vec_mulo(xxh_u32x4 a, xxh_u32x4 b)
{
    xxh_u64x2 result;
    __asm__("vmulouw %0, %1, %2" : "=v" (result) : "v" (a), "v" (b));
    return result;
}
XXH_FORCE_INLINE xxh_u64x2 XXH_vec_mule(xxh_u32x4 a, xxh_u32x4 b)
{
    xxh_u64x2 result;
    __asm__("vmuleuw %0, %1, %2" : "=v" (result) : "v" (a), "v" (b));
    return result;
}
# endif /* XXH_vec_mulo, XXH_vec_mule */
#endif /* XXH_VECTOR == XXH_VSX */

#if XXH_VECTOR == XXH_SVE
#define ACCRND(acc, offset) \
do { \
    svuint64_t input_vec = svld1_u64(mask, xinput + offset);         \
    svuint64_t secret_vec = svld1_u64(mask, xsecret + offset);       \
    svuint64_t mixed = sveor_u64_x(mask, secret_vec, input_vec);     \
    svuint64_t swapped = svtbl_u64(input_vec, kSwap);                \
    svuint64_t mixed_lo = svextw_u64_x(mask, mixed);                 \
    svuint64_t mixed_hi = svlsr_n_u64_x(mask, mixed, 32);            \
    svuint64_t mul = svmad_u64_x(mask, mixed_lo, mixed_hi, swapped); \
    acc = svadd_u64_x(mask, acc, mul);                               \
} while (0)
#endif /* XXH_VECTOR == XXH_SVE */

/* prefetch
 * can be disabled, by declaring XXH_NO_PREFETCH build macro */
#if defined(XXH_NO_PREFETCH)
#  define XXH_PREFETCH(ptr)  (void)(ptr)  /* disabled */
#else
#  if XXH_SIZE_OPT >= 1
#    define XXH_PREFETCH(ptr) (void)(ptr)
#  elif defined(_MSC_VER) && (defined(_M_X64) || defined(_M_IX86))  /* _mm_prefetch() not defined outside of x86/x64 */
#    include <mmintrin.h>   /* https://msdn.microsoft.com/fr-fr/library/84szxsww(v=vs.90).aspx */
#    define XXH_PREFETCH(ptr)  _mm_prefetch((const char*)(ptr), _MM_HINT_T0)
#  elif defined(__GNUC__) && ( (__GNUC__ >= 4) || ( (__GNUC__ == 3) && (__GNUC_MINOR__ >= 1) ) )
#    define XXH_PREFETCH(ptr)  __builtin_prefetch((ptr), 0 /* rw==read */, 3 /* locality */)
#  else
#    define XXH_PREFETCH(ptr) (void)(ptr)  /* disabled */
#  endif
#endif  /* XXH_NO_PREFETCH */


/* ==========================================
 * XXH3 default settings
 * ========================================== */

#define XXH_SECRET_DEFAULT_SIZE 192   /* minimum XXH3_SECRET_SIZE_MIN */

#if (XXH_SECRET_DEFAULT_SIZE < XXH3_SECRET_SIZE_MIN)
#  error "default keyset is not large enough"
#endif

/*! Pseudorandom secret taken directly from FARSH. */
XXH_ALIGN(64) static const xxh_u8 XXH3_kSecret[XXH_SECRET_DEFAULT_SIZE] = {
    0xb8, 0xfe, 0x6c, 0x39, 0x23, 0xa4, 0x4b, 0xbe, 0x7c, 0x01, 0x81, 0x2c, 0xf7, 0x21, 0xad, 0x1c,
    0xde, 0xd4, 0x6d, 0xe9, 0x83, 0x90, 0x97, 0xdb, 0x72, 0x40, 0xa4, 0xa4, 0xb7, 0xb3, 0x67, 0x1f,
    0xcb, 0x79, 0xe6, 0x4e, 0xcc, 0xc0, 0xe5, 0x78, 0x82, 0x5a, 0xd0, 0x7d, 0xcc, 0xff, 0x72, 0x21,
    0xb8, 0x08, 0x46, 0x74, 0xf7, 0x43, 0x24, 0x8e, 0xe0, 0x35, 0x90, 0xe6, 0x81, 0x3a, 0x26, 0x4c,
    0x3c, 0x28, 0x52, 0xbb, 0x91, 0xc3, 0x00, 0xcb, 0x88, 0xd0, 0x65, 0x8b, 0x1b, 0x53, 0x2e, 0xa3,
    0x71, 0x64, 0x48, 0x97, 0xa2, 0x0d, 0xf9, 0x4e, 0x38, 0x19, 0xef, 0x46, 0xa9, 0xde, 0xac, 0xd8,
    0xa8, 0xfa, 0x76, 0x3f, 0xe3, 0x9c, 0x34, 0x3f, 0xf9, 0xdc, 0xbb, 0xc7, 0xc7, 0x0b, 0x4f, 0x1d,
    0x8a, 0x51, 0xe0, 0x4b, 0xcd, 0xb4, 0x59, 0x31, 0xc8, 0x9f, 0x7e, 0xc9, 0xd9, 0x78, 0x73, 0x64,
    0xea, 0xc5, 0xac, 0x83, 0x34, 0xd3, 0xeb, 0xc3, 0xc5, 0x81, 0xa0, 0xff, 0xfa, 0x13, 0x63, 0xeb,
    0x17, 0x0d, 0xdd, 0x51, 0xb7, 0xf0, 0xda, 0x49, 0xd3, 0x16, 0x55, 0x26, 0x29, 0xd4, 0x68, 0x9e,
    0x2b, 0x16, 0xbe, 0x58, 0x7d, 0x47, 0xa1, 0xfc, 0x8f, 0xf8, 0xb8, 0xd1, 0x7a, 0xd0, 0x31, 0xce,
    0x45, 0xcb, 0x3a, 0x8f, 0x95, 0x16, 0x04, 0x28, 0xaf, 0xd7, 0xfb, 0xca, 0xbb, 0x4b, 0x40, 0x7e,
};

static const xxh_u64 PRIME_MX1 = 0x165667919E3779F9ULL;  /*!< 0b0001011001010110011001111001000110011110001101110111100111111001 */
static const xxh_u64 PRIME_MX2 = 0x9FB21C651E98DF25ULL;  /*!< 0b1001111110110010000111000110010100011110100110001101111100100101 */

#ifdef XXH_OLD_NAMES
#  define kSecret XXH3_kSecret
#endif

#ifdef XXH_DOXYGEN
/*!
 * @brief Calculates a 32-bit to 64-bit long multiply.
 *
 * Implemented as a macro.
 *
 * Wraps `__emulu` on MSVC x86 because it tends to call `__allmul` when it doesn't
 * need to (but it shouldn't need to anyways, it is about 7 instructions to do
 * a 64x64 multiply...). Since we know that this will _always_ emit `MULL`, we
 * use that instead of the normal method.
 *
 * If you are compiling for platforms like Thumb-1 and don't have a better option,
 * you may also want to write your own long multiply routine here.
 *
 * @param x, y Numbers to be multiplied
 * @return 64-bit product of the low 32 bits of @p x and @p y.
 */
XXH_FORCE_INLINE xxh_u64
XXH_mult32to64(xxh_u64 x, xxh_u64 y)
{
   return (x & 0xFFFFFFFF) * (y & 0xFFFFFFFF);
}
#elif defined(_MSC_VER) && defined(_M_IX86)
#    define XXH_mult32to64(x, y) __emulu((unsigned)(x), (unsigned)(y))
#else
/*
 * Downcast + upcast is usually better than masking on older compilers like
 * GCC 4.2 (especially 32-bit ones), all without affecting newer compilers.
 *
 * The other method, (x & 0xFFFFFFFF) * (y & 0xFFFFFFFF), will AND both operands
 * and perform a full 64x64 multiply -- entirely redundant on 32-bit.
 */
#    define XXH_mult32to64(x, y) ((xxh_u64)(xxh_u32)(x) * (xxh_u64)(xxh_u32)(y))
#endif

/*!
 * @brief Calculates a 64->128-bit long multiply.
 *
 * Uses `__uint128_t` and `_umul128` if available, otherwise uses a scalar
 * version.
 *
 * @param lhs , rhs The 64-bit integers to be multiplied
 * @return The 128-bit result represented in an @ref XXH128_hash_t.
 */
static XXH128_hash_t
XXH_mult64to128(xxh_u64 lhs, xxh_u64 rhs)
{
    /*
     * GCC/Clang __uint128_t method.
     *
     * On most 64-bit targets, GCC and Clang define a __uint128_t type.
     * This is usually the best way as it usually uses a native long 64-bit
     * multiply, such as MULQ on x86_64 or MUL + UMULH on aarch64.
     *
     * Usually.
     *
     * Despite being a 32-bit platform, Clang (and emscripten) define this type
     * despite not having the arithmetic for it. This results in a laggy
     * compiler builtin call which calculates a full 128-bit multiply.
     * In that case it is best to use the portable one.
     * https://github.com/Cyan4973/xxHash/issues/211#issuecomment-515575677
     */
#if (defined(__GNUC__) || defined(__clang__)) && !defined(__wasm__) \
    && defined(__SIZEOF_INT128__) \
    || (defined(_INTEGRAL_MAX_BITS) && _INTEGRAL_MAX_BITS >= 128)

    __uint128_t const product = (__uint128_t)lhs * (__uint128_t)rhs;
    XXH128_hash_t r128;
    r128.low64  = (xxh_u64)(product);
    r128.high64 = (xxh_u64)(product >> 64);
    return r128;

    /*
     * MSVC for x64's _umul128 method.
     *
     * xxh_u64 _umul128(xxh_u64 Multiplier, xxh_u64 Multiplicand, xxh_u64 *HighProduct);
     *
     * This compiles to single operand MUL on x64.
     */
#elif (defined(_M_X64) || defined(_M_IA64)) && !defined(_M_ARM64EC)

#ifndef _MSC_VER
#   pragma intrinsic(_umul128)
#endif
    xxh_u64 product_high;
    xxh_u64 const product_low = _umul128(lhs, rhs, &product_high);
    XXH128_hash_t r128;
    r128.low64  = product_low;
    r128.high64 = product_high;
    return r128;

    /*
     * MSVC for ARM64's __umulh method.
     *
     * This compiles to the same MUL + UMULH as GCC/Clang's __uint128_t method.
     */
#elif defined(_M_ARM64) || defined(_M_ARM64EC)

#ifndef _MSC_VER
#   pragma intrinsic(__umulh)
#endif
    XXH128_hash_t r128;
    r128.low64  = lhs * rhs;
    r128.high64 = __umulh(lhs, rhs);
    return r128;

#else
    /*
     * Portable scalar method. Optimized for 32-bit and 64-bit ALUs.
     *
     * This is a fast and simple grade school multiply, which is shown below
     * with base 10 arithmetic instead of base 0x100000000.
     *
     *           9 3 // D2 lhs = 93
     *         x 7 5 // D2 rhs = 75
     *     ----------
     *           1 5 // D2 lo_lo = (93 % 10) * (75 % 10) = 15
     *         4 5 | // D2 hi_lo = (93 / 10) * (75 % 10) = 45
     *         2 1 | // D2 lo_hi = (93 % 10) * (75 / 10) = 21
     *     + 6 3 | | // D2 hi_hi = (93 / 10) * (75 / 10) = 63
     *     ---------
     *         2 7 | // D2 cross = (15 / 10) + (45 % 10) + 21 = 27
     *     + 6 7 | | // D2 upper = (27 / 10) + (45 / 10) + 63 = 67
     *     ---------
     *       6 9 7 5 // D4 res = (27 * 10) + (15 % 10) + (67 * 100) = 6975
     *
     * The reasons for adding the products like this are:
     *  1. It avoids manual carry tracking. Just like how
     *     (9 * 9) + 9 + 9 = 99, the same applies with this for UINT64_MAX.
     *     This avoids a lot of complexity.
     *
     *  2. It hints for, and on Clang, compiles to, the powerful UMAAL
     *     instruction available in ARM's Digital Signal Processing extension
     *     in 32-bit ARMv6 and later, which is shown below:
     *
     *         void UMAAL(xxh_u32 *RdLo, xxh_u32 *RdHi, xxh_u32 Rn, xxh_u32 Rm)
     *         {
     *             xxh_u64 product = (xxh_u64)*RdLo * (xxh_u64)*RdHi + Rn + Rm;
     *             *RdLo = (xxh_u32)(product & 0xFFFFFFFF);
     *             *RdHi = (xxh_u32)(product >> 32);
     *         }
     *
     *     This instruction was designed for efficient long multiplication, and
     *     allows this to be calculated in only 4 instructions at speeds
     *     comparable to some 64-bit ALUs.
     *
     *  3. It isn't terrible on other platforms. Usually this will be a couple
     *     of 32-bit ADD/ADCs.
     */

    /* First calculate all of the cross products. */
    xxh_u64 const lo_lo = XXH_mult32to64(lhs & 0xFFFFFFFF, rhs & 0xFFFFFFFF);
    xxh_u64 const hi_lo = XXH_mult32to64(lhs >> 32,        rhs & 0xFFFFFFFF);
    xxh_u64 const lo_hi = XXH_mult32to64(lhs & 0xFFFFFFFF, rhs >> 32);
    xxh_u64 const hi_hi = XXH_mult32to64(lhs >> 32,        rhs >> 32);

    /* Now add the products together. These will never overflow. */
    xxh_u64 const cross = (lo_lo >> 32) + (hi_lo & 0xFFFFFFFF) + lo_hi;
    xxh_u64 const upper = (hi_lo >> 32) + (cross >> 32)        + hi_hi;
    xxh_u64 const lower = (cross << 32) | (lo_lo & 0xFFFFFFFF);

    XXH128_hash_t r128;
    r128.low64  = lower;
    r128.high64 = upper;
    return r128;
#endif
}

/*!
 * @brief Calculates a 64-bit to 128-bit multiply, then XOR folds it.
 *
 * The reason for the separate function is to prevent passing too many structs
 * around by value. This will hopefully inline the multiply, but we don't force it.
 *
 * @param lhs , rhs The 64-bit integers to multiply
 * @return The low 64 bits of the product XOR'd by the high 64 bits.
 * @see XXH_mult64to128()
 */
static xxh_u64
XXH3_mul128_fold64(xxh_u64 lhs, xxh_u64 rhs)
{
    XXH128_hash_t product = XXH_mult64to128(lhs, rhs);
    return product.low64 ^ product.high64;
}

/*! Seems to produce slightly better code on GCC for some reason. */
XXH_FORCE_INLINE XXH_CONSTF xxh_u64 XXH_xorshift64(xxh_u64 v64, int shift)
{
    XXH_ASSERT(0 <= shift && shift < 64);
    return v64 ^ (v64 >> shift);
}

/*
 * This is a fast avalanche stage,
 * suitable when input bits are already partially mixed
 */
static XXH64_hash_t XXH3_avalanche(xxh_u64 h64)
{
    h64 = XXH_xorshift64(h64, 37);
    h64 *= PRIME_MX1;
    h64 = XXH_xorshift64(h64, 32);
    return h64;
}

/*
 * This is a stronger avalanche,
 * inspired by Pelle Evensen's rrmxmx
 * preferable when input has not been previously mixed
 */
static XXH64_hash_t XXH3_rrmxmx(xxh_u64 h64, xxh_u64 len)
{
    /* this mix is inspired by Pelle Evensen's rrmxmx */
    h64 ^= XXH_rotl64(h64, 49) ^ XXH_rotl64(h64, 24);
    h64 *= PRIME_MX2;
    h64 ^= (h64 >> 35) + len ;
    h64 *= PRIME_MX2;
    return XXH_xorshift64(h64, 28);
}


/* ==========================================
 * Short keys
 * ==========================================
 * One of the shortcomings of XXH32 and XXH64 was that their performance was
 * sub-optimal on short lengths. It used an iterative algorithm which strongly
 * favored lengths that were a multiple of 4 or 8.
 *
 * Instead of iterating over individual inputs, we use a set of single shot
 * functions which piece together a range of lengths and operate in constant time.
 *
 * Additionally, the number of multiplies has been significantly reduced. This
 * reduces latency, especially when emulating 64-bit multiplies on 32-bit.
 *
 * Depending on the platform, this may or may not be faster than XXH32, but it
 * is almost guaranteed to be faster than XXH64.
 */

/*
 * At very short lengths, there isn't enough input to fully hide secrets, or use
 * the entire secret.
 *
 * There is also only a limited amount of mixing we can do before significantly
 * impacting performance.
 *
 * Therefore, we use different sections of the secret and always mix two secret
 * samples with an XOR. This should have no effect on performance on the
 * seedless or withSeed variants because everything _should_ be constant folded
 * by modern compilers.
 *
 * The XOR mixing hides individual parts of the secret and increases entropy.
 *
 * This adds an extra layer of strength for custom secrets.
 */
XXH_FORCE_INLINE XXH_PUREF XXH64_hash_t
XXH3_len_1to3_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
{
    XXH_ASSERT(input != NULL);
    XXH_ASSERT(1 <= len && len <= 3);
    XXH_ASSERT(secret != NULL);
    /*
     * len = 1: combined = { input[0], 0x01, input[0], input[0] }
     * len = 2: combined = { input[1], 0x02, input[0], input[1] }
     * len = 3: combined = { input[2], 0x03, input[0], input[1] }
     */
    {   xxh_u8  const c1 = input[0];
        xxh_u8  const c2 = input[len >> 1];
        xxh_u8  const c3 = input[len - 1];
        xxh_u32 const combined = ((xxh_u32)c1 << 16) | ((xxh_u32)c2  << 24)
                               | ((xxh_u32)c3 <<  0) | ((xxh_u32)len << 8);
        xxh_u64 const bitflip = (XXH_readLE32(secret) ^ XXH_readLE32(secret+4)) + seed;
        xxh_u64 const keyed = (xxh_u64)combined ^ bitflip;
        return XXH64_avalanche(keyed);
    }
}

XXH_FORCE_INLINE XXH_PUREF XXH64_hash_t
XXH3_len_4to8_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
{
    XXH_ASSERT(input != NULL);
    XXH_ASSERT(secret != NULL);
    XXH_ASSERT(4 <= len && len <= 8);
    seed ^= (xxh_u64)XXH_swap32((xxh_u32)seed) << 32;
    {   xxh_u32 const input1 = XXH_readLE32(input);
        xxh_u32 const input2 = XXH_readLE32(input + len - 4);
        xxh_u64 const bitflip = (XXH_readLE64(secret+8) ^ XXH_readLE64(secret+16)) - seed;
        xxh_u64 const input64 = input2 + (((xxh_u64)input1) << 32);
        xxh_u64 const keyed = input64 ^ bitflip;
        return XXH3_rrmxmx(keyed, len);
    }
}

XXH_FORCE_INLINE XXH_PUREF XXH64_hash_t
XXH3_len_9to16_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
{
    XXH_ASSERT(input != NULL);
    XXH_ASSERT(secret != NULL);
    XXH_ASSERT(9 <= len && len <= 16);
    {   xxh_u64 const bitflip1 = (XXH_readLE64(secret+24) ^ XXH_readLE64(secret+32)) + seed;
        xxh_u64 const bitflip2 = (XXH_readLE64(secret+40) ^ XXH_readLE64(secret+48)) - seed;
        xxh_u64 const input_lo = XXH_readLE64(input)           ^ bitflip1;
        xxh_u64 const input_hi = XXH_readLE64(input + len - 8) ^ bitflip2;
        xxh_u64 const acc = len
                          + XXH_swap64(input_lo) + input_hi
                          + XXH3_mul128_fold64(input_lo, input_hi);
        return XXH3_avalanche(acc);
    }
}

XXH_FORCE_INLINE XXH_PUREF XXH64_hash_t
XXH3_len_0to16_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
{
    XXH_ASSERT(len <= 16);
    {   if (XXH_likely(len >  8)) return XXH3_len_9to16_64b(input, len, secret, seed);
        if (XXH_likely(len >= 4)) return XXH3_len_4to8_64b(input, len, secret, seed);
        if (len) return XXH3_len_1to3_64b(input, len, secret, seed);
        return XXH64_avalanche(seed ^ (XXH_readLE64(secret+56) ^ XXH_readLE64(secret+64)));
    }
}

/*
 * DISCLAIMER: There are known *seed-dependent* multicollisions here due to
 * multiplication by zero, affecting hashes of lengths 17 to 240.
 *
 * However, they are very unlikely.
 *
 * Keep this in mind when using the unseeded XXH3_64bits() variant: As with all
 * unseeded non-cryptographic hashes, it does not attempt to defend itself
 * against specially crafted inputs, only random inputs.
 *
 * Compared to classic UMAC where a 1 in 2^31 chance of 4 consecutive bytes
 * cancelling out the secret is taken an arbitrary number of times (addressed
 * in XXH3_accumulate_512), this collision is very unlikely with random inputs
 * and/or proper seeding:
 *
 * This only has a 1 in 2^63 chance of 8 consecutive bytes cancelling out, in a
 * function that is only called up to 16 times per hash with up to 240 bytes of
 * input.
 *
 * This is not too bad for a non-cryptographic hash function, especially with
 * only 64 bit outputs.
 *
 * The 128-bit variant (which trades some speed for strength) is NOT affected
 * by this, although it is always a good idea to use a proper seed if you care
 * about strength.
 */
XXH_FORCE_INLINE xxh_u64 XXH3_mix16B(const xxh_u8* XXH_RESTRICT input,
                                     const xxh_u8* XXH_RESTRICT secret, xxh_u64 seed64)
{
#if defined(__GNUC__) && !defined(__clang__) /* GCC, not Clang */ \
  && defined(__i386__) && defined(__SSE2__)  /* x86 + SSE2 */ \
  && !defined(XXH_ENABLE_AUTOVECTORIZE)      /* Define to disable like XXH32 hack */
    /*
     * UGLY HACK:
     * GCC for x86 tends to autovectorize the 128-bit multiply, resulting in
     * slower code.
     *
     * By forcing seed64 into a register, we disrupt the cost model and
     * cause it to scalarize. See `XXH32_round()`
     *
     * FIXME: Clang's output is still _much_ faster -- On an AMD Ryzen 3600,
     * XXH3_64bits @ len=240 runs at 4.6 GB/s with Clang 9, but 3.3 GB/s on
     * GCC 9.2, despite both emitting scalar code.
     *
     * GCC generates much better scalar code than Clang for the rest of XXH3,
     * which is why finding a more optimal codepath is an interest.
     */
    XXH_COMPILER_GUARD(seed64);
#endif
    {   xxh_u64 const input_lo = XXH_readLE64(input);
        xxh_u64 const input_hi = XXH_readLE64(input+8);
        return XXH3_mul128_fold64(
            input_lo ^ (XXH_readLE64(secret)   + seed64),
            input_hi ^ (XXH_readLE64(secret+8) - seed64)
        );
    }
}

/* For mid range keys, XXH3 uses a Mum-hash variant. */
XXH_FORCE_INLINE XXH_PUREF XXH64_hash_t
XXH3_len_17to128_64b(const xxh_u8* XXH_RESTRICT input, size_t len,
                     const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
                     XXH64_hash_t seed)
{
    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); (void)secretSize;
    XXH_ASSERT(16 < len && len <= 128);

    {   xxh_u64 acc = len * XXH_PRIME64_1;
#if XXH_SIZE_OPT >= 1
        /* Smaller and cleaner, but slightly slower. */
        unsigned int i = (unsigned int)(len - 1) / 32;
        do {
            acc += XXH3_mix16B(input+16 * i, secret+32*i, seed);
            acc += XXH3_mix16B(input+len-16*(i+1), secret+32*i+16, seed);
        } while (i-- != 0);
#else
        if (len > 32) {
            if (len > 64) {
                if (len > 96) {
                    acc += XXH3_mix16B(input+48, secret+96, seed);
                    acc += XXH3_mix16B(input+len-64, secret+112, seed);
                }
                acc += XXH3_mix16B(input+32, secret+64, seed);
                acc += XXH3_mix16B(input+len-48, secret+80, seed);
            }
            acc += XXH3_mix16B(input+16, secret+32, seed);
            acc += XXH3_mix16B(input+len-32, secret+48, seed);
        }
        acc += XXH3_mix16B(input+0, secret+0, seed);
        acc += XXH3_mix16B(input+len-16, secret+16, seed);
#endif
        return XXH3_avalanche(acc);
    }
}

#define XXH3_MIDSIZE_MAX 240

XXH_NO_INLINE XXH_PUREF XXH64_hash_t
XXH3_len_129to240_64b(const xxh_u8* XXH_RESTRICT input, size_t len,
                      const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
                      XXH64_hash_t seed)
{
    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); (void)secretSize;
    XXH_ASSERT(128 < len && len <= XXH3_MIDSIZE_MAX);

    #define XXH3_MIDSIZE_STARTOFFSET 3
    #define XXH3_MIDSIZE_LASTOFFSET  17

    {   xxh_u64 acc = len * XXH_PRIME64_1;
        xxh_u64 acc_end;
        unsigned int const nbRounds = (unsigned int)len / 16;
        unsigned int i;
        XXH_ASSERT(128 < len && len <= XXH3_MIDSIZE_MAX);
        for (i=0; i<8; i++) {
            acc += XXH3_mix16B(input+(16*i), secret+(16*i), seed);
        }
        /* last bytes */
        acc_end = XXH3_mix16B(input + len - 16, secret + XXH3_SECRET_SIZE_MIN - XXH3_MIDSIZE_LASTOFFSET, seed);
        XXH_ASSERT(nbRounds >= 8);
        acc = XXH3_avalanche(acc);
#if defined(__clang__)                                /* Clang */ \
    && (defined(__ARM_NEON) || defined(__ARM_NEON__)) /* NEON */ \
    && !defined(XXH_ENABLE_AUTOVECTORIZE)             /* Define to disable */
        /*
         * UGLY HACK:
         * Clang for ARMv7-A tries to vectorize this loop, similar to GCC x86.
         * In everywhere else, it uses scalar code.
         *
         * For 64->128-bit multiplies, even if the NEON was 100% optimal, it
         * would still be slower than UMAAL (see XXH_mult64to128).
         *
         * Unfortunately, Clang doesn't handle the long multiplies properly and
         * converts them to the nonexistent "vmulq_u64" intrinsic, which is then
         * scalarized into an ugly mess of VMOV.32 instructions.
         *
         * This mess is difficult to avoid without turning autovectorization
         * off completely, but they are usually relatively minor and/or not
         * worth it to fix.
         *
         * This loop is the easiest to fix, as unlike XXH32, this pragma
         * _actually works_ because it is a loop vectorization instead of an
         * SLP vectorization.
         */
        #pragma clang loop vectorize(disable)
#endif
        for (i=8 ; i < nbRounds; i++) {
            /*
             * Prevents clang for unrolling the acc loop and interleaving with this one.
             */
            XXH_COMPILER_GUARD(acc);
            acc_end += XXH3_mix16B(input+(16*i), secret+(16*(i-8)) + XXH3_MIDSIZE_STARTOFFSET, seed);
        }
        return XXH3_avalanche(acc + acc_end);
    }
}


/* =======     Long Keys     ======= */

#define XXH_STRIPE_LEN 64
#define XXH_SECRET_CONSUME_RATE 8   /* nb of secret bytes consumed at each accumulation */
#define XXH_ACC_NB (XXH_STRIPE_LEN / sizeof(xxh_u64))

#ifdef XXH_OLD_NAMES
#  define STRIPE_LEN XXH_STRIPE_LEN
#  define ACC_NB XXH_ACC_NB
#endif

#ifndef XXH_PREFETCH_DIST
#  ifdef __clang__
#    define XXH_PREFETCH_DIST 320
#  else
#    if (XXH_VECTOR == XXH_AVX512)
#      define XXH_PREFETCH_DIST 512
#    else
#      define XXH_PREFETCH_DIST 384
#    endif
#  endif  /* __clang__ */
#endif  /* XXH_PREFETCH_DIST */

/*
 * These macros are to generate an XXH3_accumulate() function.
 * The two arguments select the name suffix and target attribute.
 *
 * The name of this symbol is XXH3_accumulate_<name>() and it calls
 * XXH3_accumulate_512_<name>().
 *
 * It may be useful to hand implement this function if the compiler fails to
 * optimize the inline function.
 */
#define XXH3_ACCUMULATE_TEMPLATE(name)                      \
void                                                        \
XXH3_accumulate_##name(xxh_u64* XXH_RESTRICT acc,           \
                       const xxh_u8* XXH_RESTRICT input,    \
                       const xxh_u8* XXH_RESTRICT secret,   \
                       size_t nbStripes)                    \
{                                                           \
    size_t n;                                               \
    for (n = 0; n < nbStripes; n++ ) {                      \
        const xxh_u8* const in = input + n*XXH_STRIPE_LEN;  \
        XXH_PREFETCH(in + XXH_PREFETCH_DIST);               \
        XXH3_accumulate_512_##name(                         \
                 acc,                                       \
                 in,                                        \
                 secret + n*XXH_SECRET_CONSUME_RATE);       \
    }                                                       \
}


XXH_FORCE_INLINE void XXH_writeLE64(void* dst, xxh_u64 v64)
{
    if (!XXH_CPU_LITTLE_ENDIAN) v64 = XXH_swap64(v64);
    XXH_memcpy(dst, &v64, sizeof(v64));
}

/* Several intrinsic functions below are supposed to accept __int64 as argument,
 * as documented in https://software.intel.com/sites/landingpage/IntrinsicsGuide/ .
 * However, several environments do not define __int64 type,
 * requiring a workaround.
 */
#if !defined (__VMS) \
  && (defined (__cplusplus) \
  || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
    typedef int64_t xxh_i64;
#else
    /* the following type must have a width of 64-bit */
    typedef long long xxh_i64;
#endif


/*
 * XXH3_accumulate_512 is the tightest loop for long inputs, and it is the most optimized.
 *
 * It is a hardened version of UMAC, based off of FARSH's implementation.
 *
 * This was chosen because it adapts quite well to 32-bit, 64-bit, and SIMD
 * implementations, and it is ridiculously fast.
 *
 * We harden it by mixing the original input to the accumulators as well as the product.
 *
 * This means that in the (relatively likely) case of a multiply by zero, the
 * original input is preserved.
 *
 * On 128-bit inputs, we swap 64-bit pairs when we add the input to improve
 * cross-pollination, as otherwise the upper and lower halves would be
 * essentially independent.
 *
 * This doesn't matter on 64-bit hashes since they all get merged together in
 * the end, so we skip the extra step.
 *
 * Both XXH3_64bits and XXH3_128bits use this subroutine.
 */

#if (XXH_VECTOR == XXH_AVX512) \
     || (defined(XXH_DISPATCH_AVX512) && XXH_DISPATCH_AVX512 != 0)

#ifndef XXH_TARGET_AVX512
# define XXH_TARGET_AVX512  /* disable attribute target */
#endif

XXH_FORCE_INLINE XXH_TARGET_AVX512 void
XXH3_accumulate_512_avx512(void* XXH_RESTRICT acc,
                     const void* XXH_RESTRICT input,
                     const void* XXH_RESTRICT secret)
{
    __m512i* const xacc = (__m512i *) acc;
    XXH_ASSERT((((size_t)acc) & 63) == 0);
    XXH_STATIC_ASSERT(XXH_STRIPE_LEN == sizeof(__m512i));

    {
        /* data_vec    = input[0]; */
        __m512i const data_vec    = _mm512_loadu_si512   (input);
        /* key_vec     = secret[0]; */
        __m512i const key_vec     = _mm512_loadu_si512   (secret);
        /* data_key    = data_vec ^ key_vec; */
        __m512i const data_key    = _mm512_xor_si512     (data_vec, key_vec);
        /* data_key_lo = data_key >> 32; */
        __m512i const data_key_lo = _mm512_srli_epi64 (data_key, 32);
        /* product     = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */
        __m512i const product     = _mm512_mul_epu32     (data_key, data_key_lo);
        /* xacc[0] += swap(data_vec); */
        __m512i const data_swap = _mm512_shuffle_epi32(data_vec, (_MM_PERM_ENUM)_MM_SHUFFLE(1, 0, 3, 2));
        __m512i const sum       = _mm512_add_epi64(*xacc, data_swap);
        /* xacc[0] += product; */
        *xacc = _mm512_add_epi64(product, sum);
    }
}
XXH_FORCE_INLINE XXH_TARGET_AVX512 XXH3_ACCUMULATE_TEMPLATE(avx512)

/*
 * XXH3_scrambleAcc: Scrambles the accumulators to improve mixing.
 *
 * Multiplication isn't perfect, as explained by Google in HighwayHash:
 *
 *  // Multiplication mixes/scrambles bytes 0-7 of the 64-bit result to
 *  // varying degrees. In descending order of goodness, bytes
 *  // 3 4 2 5 1 6 0 7 have quality 228 224 164 160 100 96 36 32.
 *  // As expected, the upper and lower bytes are much worse.
 *
 * Source: https://github.com/google/highwayhash/blob/0aaf66b/highwayhash/hh_avx2.h#L291
 *
 * Since our algorithm uses a pseudorandom secret to add some variance into the
 * mix, we don't need to (or want to) mix as often or as much as HighwayHash does.
 *
 * This isn't as tight as XXH3_accumulate, but still written in SIMD to avoid
 * extraction.
 *
 * Both XXH3_64bits and XXH3_128bits use this subroutine.
 */

XXH_FORCE_INLINE XXH_TARGET_AVX512 void
XXH3_scrambleAcc_avx512(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
{
    XXH_ASSERT((((size_t)acc) & 63) == 0);
    XXH_STATIC_ASSERT(XXH_STRIPE_LEN == sizeof(__m512i));
    {   __m512i* const xacc = (__m512i*) acc;
        const __m512i prime32 = _mm512_set1_epi32((int)XXH_PRIME32_1);

        /* xacc[0] ^= (xacc[0] >> 47) */
        __m512i const acc_vec     = *xacc;
        __m512i const shifted     = _mm512_srli_epi64    (acc_vec, 47);
        /* xacc[0] ^= secret; */
        __m512i const key_vec     = _mm512_loadu_si512   (secret);
        __m512i const data_key    = _mm512_ternarylogic_epi32(key_vec, acc_vec, shifted, 0x96 /* key_vec ^ acc_vec ^ shifted */);

        /* xacc[0] *= XXH_PRIME32_1; */
        __m512i const data_key_hi = _mm512_srli_epi64 (data_key, 32);
        __m512i const prod_lo     = _mm512_mul_epu32     (data_key, prime32);
        __m512i const prod_hi     = _mm512_mul_epu32     (data_key_hi, prime32);
        *xacc = _mm512_add_epi64(prod_lo, _mm512_slli_epi64(prod_hi, 32));
    }
}

XXH_FORCE_INLINE XXH_TARGET_AVX512 void
XXH3_initCustomSecret_avx512(void* XXH_RESTRICT customSecret, xxh_u64 seed64)
{
    XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 63) == 0);
    XXH_STATIC_ASSERT(XXH_SEC_ALIGN == 64);
    XXH_ASSERT(((size_t)customSecret & 63) == 0);
    (void)(&XXH_writeLE64);
    {   int const nbRounds = XXH_SECRET_DEFAULT_SIZE / sizeof(__m512i);
        __m512i const seed_pos = _mm512_set1_epi64((xxh_i64)seed64);
        __m512i const seed     = _mm512_mask_sub_epi64(seed_pos, 0xAA, _mm512_set1_epi8(0), seed_pos);

        const __m512i* const src  = (const __m512i*) ((const void*) XXH3_kSecret);
              __m512i* const dest = (      __m512i*) customSecret;
        int i;
        XXH_ASSERT(((size_t)src & 63) == 0); /* control alignment */
        XXH_ASSERT(((size_t)dest & 63) == 0);
        for (i=0; i < nbRounds; ++i) {
            dest[i] = _mm512_add_epi64(_mm512_load_si512(src + i), seed);
    }   }
}

#endif

#if (XXH_VECTOR == XXH_AVX2) \
    || (defined(XXH_DISPATCH_AVX2) && XXH_DISPATCH_AVX2 != 0)

#ifndef XXH_TARGET_AVX2
# define XXH_TARGET_AVX2  /* disable attribute target */
#endif

XXH_FORCE_INLINE XXH_TARGET_AVX2 void
XXH3_accumulate_512_avx2( void* XXH_RESTRICT acc,
                    const void* XXH_RESTRICT input,
                    const void* XXH_RESTRICT secret)
{
    XXH_ASSERT((((size_t)acc) & 31) == 0);
    {   __m256i* const xacc    =       (__m256i *) acc;
        /* Unaligned. This is mainly for pointer arithmetic, and because
         * _mm256_loadu_si256 requires  a const __m256i * pointer for some reason. */
        const         __m256i* const xinput  = (const __m256i *) input;
        /* Unaligned. This is mainly for pointer arithmetic, and because
         * _mm256_loadu_si256 requires a const __m256i * pointer for some reason. */
        const         __m256i* const xsecret = (const __m256i *) secret;

        size_t i;
        for (i=0; i < XXH_STRIPE_LEN/sizeof(__m256i); i++) {
            /* data_vec    = xinput[i]; */
            __m256i const data_vec    = _mm256_loadu_si256    (xinput+i);
            /* key_vec     = xsecret[i]; */
            __m256i const key_vec     = _mm256_loadu_si256   (xsecret+i);
            /* data_key    = data_vec ^ key_vec; */
            __m256i const data_key    = _mm256_xor_si256     (data_vec, key_vec);
            /* data_key_lo = data_key >> 32; */
            __m256i const data_key_lo = _mm256_srli_epi64 (data_key, 32);
            /* product     = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */
            __m256i const product     = _mm256_mul_epu32     (data_key, data_key_lo);
            /* xacc[i] += swap(data_vec); */
            __m256i const data_swap = _mm256_shuffle_epi32(data_vec, _MM_SHUFFLE(1, 0, 3, 2));
            __m256i const sum       = _mm256_add_epi64(xacc[i], data_swap);
            /* xacc[i] += product; */
            xacc[i] = _mm256_add_epi64(product, sum);
    }   }
}
XXH_FORCE_INLINE XXH_TARGET_AVX2 XXH3_ACCUMULATE_TEMPLATE(avx2)

XXH_FORCE_INLINE XXH_TARGET_AVX2 void
XXH3_scrambleAcc_avx2(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
{
    XXH_ASSERT((((size_t)acc) & 31) == 0);
    {   __m256i* const xacc = (__m256i*) acc;
        /* Unaligned. This is mainly for pointer arithmetic, and because
         * _mm256_loadu_si256 requires a const __m256i * pointer for some reason. */
        const         __m256i* const xsecret = (const __m256i *) secret;
        const __m256i prime32 = _mm256_set1_epi32((int)XXH_PRIME32_1);

        size_t i;
        for (i=0; i < XXH_STRIPE_LEN/sizeof(__m256i); i++) {
            /* xacc[i] ^= (xacc[i] >> 47) */
            __m256i const acc_vec     = xacc[i];
            __m256i const shifted     = _mm256_srli_epi64    (acc_vec, 47);
            __m256i const data_vec    = _mm256_xor_si256     (acc_vec, shifted);
            /* xacc[i] ^= xsecret; */
            __m256i const key_vec     = _mm256_loadu_si256   (xsecret+i);
            __m256i const data_key    = _mm256_xor_si256     (data_vec, key_vec);

            /* xacc[i] *= XXH_PRIME32_1; */
            __m256i const data_key_hi = _mm256_srli_epi64 (data_key, 32);
            __m256i const prod_lo     = _mm256_mul_epu32     (data_key, prime32);
            __m256i const prod_hi     = _mm256_mul_epu32     (data_key_hi, prime32);
            xacc[i] = _mm256_add_epi64(prod_lo, _mm256_slli_epi64(prod_hi, 32));
        }
    }
}

XXH_FORCE_INLINE XXH_TARGET_AVX2 void XXH3_initCustomSecret_avx2(void* XXH_RESTRICT customSecret, xxh_u64 seed64)
{
    XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 31) == 0);
    XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE / sizeof(__m256i)) == 6);
    XXH_STATIC_ASSERT(XXH_SEC_ALIGN <= 64);
    (void)(&XXH_writeLE64);
    XXH_PREFETCH(customSecret);
    {   __m256i const seed = _mm256_set_epi64x((xxh_i64)(0U - seed64), (xxh_i64)seed64, (xxh_i64)(0U - seed64), (xxh_i64)seed64);

        const __m256i* const src  = (const __m256i*) ((const void*) XXH3_kSecret);
              __m256i*       dest = (      __m256i*) customSecret;

#       if defined(__GNUC__) || defined(__clang__)
        /*
         * On GCC & Clang, marking 'dest' as modified will cause the compiler:
         *   - do not extract the secret from sse registers in the internal loop
         *   - use less common registers, and avoid pushing these reg into stack
         */
        XXH_COMPILER_GUARD(dest);
#       endif
        XXH_ASSERT(((size_t)src & 31) == 0); /* control alignment */
        XXH_ASSERT(((size_t)dest & 31) == 0);

        /* GCC -O2 need unroll loop manually */
        dest[0] = _mm256_add_epi64(_mm256_load_si256(src+0), seed);
        dest[1] = _mm256_add_epi64(_mm256_load_si256(src+1), seed);
        dest[2] = _mm256_add_epi64(_mm256_load_si256(src+2), seed);
        dest[3] = _mm256_add_epi64(_mm256_load_si256(src+3), seed);
        dest[4] = _mm256_add_epi64(_mm256_load_si256(src+4), seed);
        dest[5] = _mm256_add_epi64(_mm256_load_si256(src+5), seed);
    }
}

#endif

/* x86dispatch always generates SSE2 */
#if (XXH_VECTOR == XXH_SSE2) || defined(XXH_X86DISPATCH)

#ifndef XXH_TARGET_SSE2
# define XXH_TARGET_SSE2  /* disable attribute target */
#endif

XXH_FORCE_INLINE XXH_TARGET_SSE2 void
XXH3_accumulate_512_sse2( void* XXH_RESTRICT acc,
                    const void* XXH_RESTRICT input,
                    const void* XXH_RESTRICT secret)
{
    /* SSE2 is just a half-scale version of the AVX2 version. */
    XXH_ASSERT((((size_t)acc) & 15) == 0);
    {   __m128i* const xacc    =       (__m128i *) acc;
        /* Unaligned. This is mainly for pointer arithmetic, and because
         * _mm_loadu_si128 requires a const __m128i * pointer for some reason. */
        const         __m128i* const xinput  = (const __m128i *) input;
        /* Unaligned. This is mainly for pointer arithmetic, and because
         * _mm_loadu_si128 requires a const __m128i * pointer for some reason. */
        const         __m128i* const xsecret = (const __m128i *) secret;

        size_t i;
        for (i=0; i < XXH_STRIPE_LEN/sizeof(__m128i); i++) {
            /* data_vec    = xinput[i]; */
            __m128i const data_vec    = _mm_loadu_si128   (xinput+i);
            /* key_vec     = xsecret[i]; */
            __m128i const key_vec     = _mm_loadu_si128   (xsecret+i);
            /* data_key    = data_vec ^ key_vec; */
            __m128i const data_key    = _mm_xor_si128     (data_vec, key_vec);
            /* data_key_lo = data_key >> 32; */
            __m128i const data_key_lo = _mm_shuffle_epi32 (data_key, _MM_SHUFFLE(0, 3, 0, 1));
            /* product     = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */
            __m128i const product     = _mm_mul_epu32     (data_key, data_key_lo);
            /* xacc[i] += swap(data_vec); */
            __m128i const data_swap = _mm_shuffle_epi32(data_vec, _MM_SHUFFLE(1,0,3,2));
            __m128i const sum       = _mm_add_epi64(xacc[i], data_swap);
            /* xacc[i] += product; */
            xacc[i] = _mm_add_epi64(product, sum);
    }   }
}
XXH_FORCE_INLINE XXH_TARGET_SSE2 XXH3_ACCUMULATE_TEMPLATE(sse2)

XXH_FORCE_INLINE XXH_TARGET_SSE2 void
XXH3_scrambleAcc_sse2(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
{
    XXH_ASSERT((((size_t)acc) & 15) == 0);
    {   __m128i* const xacc = (__m128i*) acc;
        /* Unaligned. This is mainly for pointer arithmetic, and because
         * _mm_loadu_si128 requires a const __m128i * pointer for some reason. */
        const         __m128i* const xsecret = (const __m128i *) secret;
        const __m128i prime32 = _mm_set1_epi32((int)XXH_PRIME32_1);

        size_t i;
        for (i=0; i < XXH_STRIPE_LEN/sizeof(__m128i); i++) {
            /* xacc[i] ^= (xacc[i] >> 47) */
            __m128i const acc_vec     = xacc[i];
            __m128i const shifted     = _mm_srli_epi64    (acc_vec, 47);
            __m128i const data_vec    = _mm_xor_si128     (acc_vec, shifted);
            /* xacc[i] ^= xsecret[i]; */
            __m128i const key_vec     = _mm_loadu_si128   (xsecret+i);
            __m128i const data_key    = _mm_xor_si128     (data_vec, key_vec);

            /* xacc[i] *= XXH_PRIME32_1; */
            __m128i const data_key_hi = _mm_shuffle_epi32 (data_key, _MM_SHUFFLE(0, 3, 0, 1));
            __m128i const prod_lo     = _mm_mul_epu32     (data_key, prime32);
            __m128i const prod_hi     = _mm_mul_epu32     (data_key_hi, prime32);
            xacc[i] = _mm_add_epi64(prod_lo, _mm_slli_epi64(prod_hi, 32));
        }
    }
}

XXH_FORCE_INLINE XXH_TARGET_SSE2 void XXH3_initCustomSecret_sse2(void* XXH_RESTRICT customSecret, xxh_u64 seed64)
{
    XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 15) == 0);
    (void)(&XXH_writeLE64);
    {   int const nbRounds = XXH_SECRET_DEFAULT_SIZE / sizeof(__m128i);

#       if defined(_MSC_VER) && defined(_M_IX86) && _MSC_VER < 1900
        /* MSVC 32bit mode does not support _mm_set_epi64x before 2015 */
        XXH_ALIGN(16) const xxh_i64 seed64x2[2] = { (xxh_i64)seed64, (xxh_i64)(0U - seed64) };
        __m128i const seed = _mm_load_si128((__m128i const*)seed64x2);
#       else
        __m128i const seed = _mm_set_epi64x((xxh_i64)(0U - seed64), (xxh_i64)seed64);
#       endif
        int i;

        const void* const src16 = XXH3_kSecret;
        __m128i* dst16 = (__m128i*) customSecret;
#       if defined(__GNUC__) || defined(__clang__)
        /*
         * On GCC & Clang, marking 'dest' as modified will cause the compiler:
         *   - do not extract the secret from sse registers in the internal loop
         *   - use less common registers, and avoid pushing these reg into stack
         */
        XXH_COMPILER_GUARD(dst16);
#       endif
        XXH_ASSERT(((size_t)src16 & 15) == 0); /* control alignment */
        XXH_ASSERT(((size_t)dst16 & 15) == 0);

        for (i=0; i < nbRounds; ++i) {
            dst16[i] = _mm_add_epi64(_mm_load_si128((const __m128i *)src16+i), seed);
    }   }
}

#endif

#if (XXH_VECTOR == XXH_NEON)

/* forward declarations for the scalar routines */
XXH_FORCE_INLINE void
XXH3_scalarRound(void* XXH_RESTRICT acc, void const* XXH_RESTRICT input,
                 void const* XXH_RESTRICT secret, size_t lane);

XXH_FORCE_INLINE void
XXH3_scalarScrambleRound(void* XXH_RESTRICT acc,
                         void const* XXH_RESTRICT secret, size_t lane);

/*!
 * @internal
 * @brief The bulk processing loop for NEON and WASM SIMD128.
 *
 * The NEON code path is actually partially scalar when running on AArch64. This
 * is to optimize the pipelining and can have up to 15% speedup depending on the
 * CPU, and it also mitigates some GCC codegen issues.
 *
 * @see XXH3_NEON_LANES for configuring this and details about this optimization.
 *
 * NEON's 32-bit to 64-bit long multiply takes a half vector of 32-bit
 * integers instead of the other platforms which mask full 64-bit vectors,
 * so the setup is more complicated than just shifting right.
 *
 * Additionally, there is an optimization for 4 lanes at once noted below.
 *
 * Since, as stated, the most optimal amount of lanes for Cortexes is 6,
 * there needs to be *three* versions of the accumulate operation used
 * for the remaining 2 lanes.
 *
 * WASM's SIMD128 uses SIMDe's arm_neon.h polyfill because the intrinsics overlap
 * nearly perfectly.
 */

XXH_FORCE_INLINE void
XXH3_accumulate_512_neon( void* XXH_RESTRICT acc,
                    const void* XXH_RESTRICT input,
                    const void* XXH_RESTRICT secret)
{
    XXH_ASSERT((((size_t)acc) & 15) == 0);
    XXH_STATIC_ASSERT(XXH3_NEON_LANES > 0 && XXH3_NEON_LANES <= XXH_ACC_NB && XXH3_NEON_LANES % 2 == 0);
    {   /* GCC for darwin arm64 does not like aliasing here */
        xxh_aliasing_uint64x2_t* const xacc = (xxh_aliasing_uint64x2_t*) acc;
        /* We don't use a uint32x4_t pointer because it causes bus errors on ARMv7. */
        uint8_t const* xinput = (const uint8_t *) input;
        uint8_t const* xsecret  = (const uint8_t *) secret;

        size_t i;
#ifdef __wasm_simd128__
        /*
         * On WASM SIMD128, Clang emits direct address loads when XXH3_kSecret
         * is constant propagated, which results in it converting it to this
         * inside the loop:
         *
         *    a = v128.load(XXH3_kSecret +  0 + $secret_offset, offset = 0)
         *    b = v128.load(XXH3_kSecret + 16 + $secret_offset, offset = 0)
         *    ...
         *
         * This requires a full 32-bit address immediate (and therefore a 6 byte
         * instruction) as well as an add for each offset.
         *
         * Putting an asm guard prevents it from folding (at the cost of losing
         * the alignment hint), and uses the free offset in `v128.load` instead
         * of adding secret_offset each time which overall reduces code size by
         * about a kilobyte and improves performance.
         */
        XXH_COMPILER_GUARD(xsecret);
#endif
        /* Scalar lanes use the normal scalarRound routine */
        for (i = XXH3_NEON_LANES; i < XXH_ACC_NB; i++) {
            XXH3_scalarRound(acc, input, secret, i);
        }
        i = 0;
        /* 4 NEON lanes at a time. */
        for (; i+1 < XXH3_NEON_LANES / 2; i+=2) {
            /* data_vec = xinput[i]; */
            uint64x2_t data_vec_1 = XXH_vld1q_u64(xinput  + (i * 16));
            uint64x2_t data_vec_2 = XXH_vld1q_u64(xinput  + ((i+1) * 16));
            /* key_vec  = xsecret[i];  */
            uint64x2_t key_vec_1  = XXH_vld1q_u64(xsecret + (i * 16));
            uint64x2_t key_vec_2  = XXH_vld1q_u64(xsecret + ((i+1) * 16));
            /* data_swap = swap(data_vec) */
            uint64x2_t data_swap_1 = vextq_u64(data_vec_1, data_vec_1, 1);
            uint64x2_t data_swap_2 = vextq_u64(data_vec_2, data_vec_2, 1);
            /* data_key = data_vec ^ key_vec; */
            uint64x2_t data_key_1 = veorq_u64(data_vec_1, key_vec_1);
            uint64x2_t data_key_2 = veorq_u64(data_vec_2, key_vec_2);

            /*
             * If we reinterpret the 64x2 vectors as 32x4 vectors, we can use a
             * de-interleave operation for 4 lanes in 1 step with `vuzpq_u32` to
             * get one vector with the low 32 bits of each lane, and one vector
             * with the high 32 bits of each lane.
             *
             * The intrinsic returns a double vector because the original ARMv7-a
             * instruction modified both arguments in place. AArch64 and SIMD128 emit
             * two instructions from this intrinsic.
             *
             *  [ dk11L | dk11H | dk12L | dk12H ] -> [ dk11L | dk12L | dk21L | dk22L ]
             *  [ dk21L | dk21H | dk22L | dk22H ] -> [ dk11H | dk12H | dk21H | dk22H ]
             */
            uint32x4x2_t unzipped = vuzpq_u32(
                vreinterpretq_u32_u64(data_key_1),
                vreinterpretq_u32_u64(data_key_2)
            );
            /* data_key_lo = data_key & 0xFFFFFFFF */
            uint32x4_t data_key_lo = unzipped.val[0];
            /* data_key_hi = data_key >> 32 */
            uint32x4_t data_key_hi = unzipped.val[1];
            /*
             * Then, we can split the vectors horizontally and multiply which, as for most
             * widening intrinsics, have a variant that works on both high half vectors
             * for free on AArch64. A similar instruction is available on SIMD128.
             *
             * sum = data_swap + (u64x2) data_key_lo * (u64x2) data_key_hi
             */
            uint64x2_t sum_1 = XXH_vmlal_low_u32(data_swap_1, data_key_lo, data_key_hi);
            uint64x2_t sum_2 = XXH_vmlal_high_u32(data_swap_2, data_key_lo, data_key_hi);
            /*
             * Clang reorders
             *    a += b * c;     // umlal   swap.2d, dkl.2s, dkh.2s
             *    c += a;         // add     acc.2d, acc.2d, swap.2d
             * to
             *    c += a;         // add     acc.2d, acc.2d, swap.2d
             *    c += b * c;     // umlal   acc.2d, dkl.2s, dkh.2s
             *
             * While it would make sense in theory since the addition is faster,
             * for reasons likely related to umlal being limited to certain NEON
             * pipelines, this is worse. A compiler guard fixes this.
             */
            XXH_COMPILER_GUARD_CLANG_NEON(sum_1);
            XXH_COMPILER_GUARD_CLANG_NEON(sum_2);
            /* xacc[i] = acc_vec + sum; */
            xacc[i]   = vaddq_u64(xacc[i], sum_1);
            xacc[i+1] = vaddq_u64(xacc[i+1], sum_2);
        }
        /* Operate on the remaining NEON lanes 2 at a time. */
        for (; i < XXH3_NEON_LANES / 2; i++) {
            /* data_vec = xinput[i]; */
            uint64x2_t data_vec = XXH_vld1q_u64(xinput  + (i * 16));
            /* key_vec  = xsecret[i];  */
            uint64x2_t key_vec  = XXH_vld1q_u64(xsecret + (i * 16));
            /* acc_vec_2 = swap(data_vec) */
            uint64x2_t data_swap = vextq_u64(data_vec, data_vec, 1);
            /* data_key = data_vec ^ key_vec; */
            uint64x2_t data_key = veorq_u64(data_vec, key_vec);
            /* For two lanes, just use VMOVN and VSHRN. */
            /* data_key_lo = data_key & 0xFFFFFFFF; */
            uint32x2_t data_key_lo = vmovn_u64(data_key);
            /* data_key_hi = data_key >> 32; */
            uint32x2_t data_key_hi = vshrn_n_u64(data_key, 32);
            /* sum = data_swap + (u64x2) data_key_lo * (u64x2) data_key_hi; */
            uint64x2_t sum = vmlal_u32(data_swap, data_key_lo, data_key_hi);
            /* Same Clang workaround as before */
            XXH_COMPILER_GUARD_CLANG_NEON(sum);
            /* xacc[i] = acc_vec + sum; */
            xacc[i] = vaddq_u64 (xacc[i], sum);
        }
    }
}
XXH_FORCE_INLINE XXH3_ACCUMULATE_TEMPLATE(neon)

XXH_FORCE_INLINE void
XXH3_scrambleAcc_neon(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
{
    XXH_ASSERT((((size_t)acc) & 15) == 0);

    {   xxh_aliasing_uint64x2_t* xacc       = (xxh_aliasing_uint64x2_t*) acc;
        uint8_t const* xsecret = (uint8_t const*) secret;

        size_t i;
        /* WASM uses operator overloads and doesn't need these. */
#ifndef __wasm_simd128__
        /* { prime32_1, prime32_1 } */
        uint32x2_t const kPrimeLo = vdup_n_u32(XXH_PRIME32_1);
        /* { 0, prime32_1, 0, prime32_1 } */
        uint32x4_t const kPrimeHi = vreinterpretq_u32_u64(vdupq_n_u64((xxh_u64)XXH_PRIME32_1 << 32));
#endif

        /* AArch64 uses both scalar and neon at the same time */
        for (i = XXH3_NEON_LANES; i < XXH_ACC_NB; i++) {
            XXH3_scalarScrambleRound(acc, secret, i);
        }
        for (i=0; i < XXH3_NEON_LANES / 2; i++) {
            /* xacc[i] ^= (xacc[i] >> 47); */
            uint64x2_t acc_vec  = xacc[i];
            uint64x2_t shifted  = vshrq_n_u64(acc_vec, 47);
            uint64x2_t data_vec = veorq_u64(acc_vec, shifted);

            /* xacc[i] ^= xsecret[i]; */
            uint64x2_t key_vec  = XXH_vld1q_u64(xsecret + (i * 16));
            uint64x2_t data_key = veorq_u64(data_vec, key_vec);
            /* xacc[i] *= XXH_PRIME32_1 */
#ifdef __wasm_simd128__
            /* SIMD128 has multiply by u64x2, use it instead of expanding and scalarizing */
            xacc[i] = data_key * XXH_PRIME32_1;
#else
            /*
             * Expanded version with portable NEON intrinsics
             *
             *    lo(x) * lo(y) + (hi(x) * lo(y) << 32)
             *
             * prod_hi = hi(data_key) * lo(prime) << 32
             *
             * Since we only need 32 bits of this multiply a trick can be used, reinterpreting the vector
             * as a uint32x4_t and multiplying by { 0, prime, 0, prime } to cancel out the unwanted bits
             * and avoid the shift.
             */
            uint32x4_t prod_hi = vmulq_u32 (vreinterpretq_u32_u64(data_key), kPrimeHi);
            /* Extract low bits for vmlal_u32  */
            uint32x2_t data_key_lo = vmovn_u64(data_key);
            /* xacc[i] = prod_hi + lo(data_key) * XXH_PRIME32_1; */
            xacc[i] = vmlal_u32(vreinterpretq_u64_u32(prod_hi), data_key_lo, kPrimeLo);
#endif
        }
    }
}
#endif

#if (XXH_VECTOR == XXH_VSX)

XXH_FORCE_INLINE void
XXH3_accumulate_512_vsx(  void* XXH_RESTRICT acc,
                    const void* XXH_RESTRICT input,
                    const void* XXH_RESTRICT secret)
{
    /* presumed aligned */
    xxh_aliasing_u64x2* const xacc = (xxh_aliasing_u64x2*) acc;
    xxh_u8 const* const xinput   = (xxh_u8 const*) input;   /* no alignment restriction */
    xxh_u8 const* const xsecret  = (xxh_u8 const*) secret;    /* no alignment restriction */
    xxh_u64x2 const v32 = { 32, 32 };
    size_t i;
    for (i = 0; i < XXH_STRIPE_LEN / sizeof(xxh_u64x2); i++) {
        /* data_vec = xinput[i]; */
        xxh_u64x2 const data_vec = XXH_vec_loadu(xinput + 16*i);
        /* key_vec = xsecret[i]; */
        xxh_u64x2 const key_vec  = XXH_vec_loadu(xsecret + 16*i);
        xxh_u64x2 const data_key = data_vec ^ key_vec;
        /* shuffled = (data_key << 32) | (data_key >> 32); */
        xxh_u32x4 const shuffled = (xxh_u32x4)vec_rl(data_key, v32);
        /* product = ((xxh_u64x2)data_key & 0xFFFFFFFF) * ((xxh_u64x2)shuffled & 0xFFFFFFFF); */
        xxh_u64x2 const product  = XXH_vec_mulo((xxh_u32x4)data_key, shuffled);
        /* acc_vec = xacc[i]; */
        xxh_u64x2 acc_vec        = xacc[i];
        acc_vec += product;

        /* swap high and low halves */
#ifdef __s390x__
        acc_vec += vec_permi(data_vec, data_vec, 2);
#else
        acc_vec += vec_xxpermdi(data_vec, data_vec, 2);
#endif
        xacc[i] = acc_vec;
    }
}
XXH_FORCE_INLINE XXH3_ACCUMULATE_TEMPLATE(vsx)

XXH_FORCE_INLINE void
XXH3_scrambleAcc_vsx(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
{
    XXH_ASSERT((((size_t)acc) & 15) == 0);

    {   xxh_aliasing_u64x2* const xacc = (xxh_aliasing_u64x2*) acc;
        const xxh_u8* const xsecret = (const xxh_u8*) secret;
        /* constants */
        xxh_u64x2 const v32  = { 32, 32 };
        xxh_u64x2 const v47 = { 47, 47 };
        xxh_u32x4 const prime = { XXH_PRIME32_1, XXH_PRIME32_1, XXH_PRIME32_1, XXH_PRIME32_1 };
        size_t i;
        for (i = 0; i < XXH_STRIPE_LEN / sizeof(xxh_u64x2); i++) {
            /* xacc[i] ^= (xacc[i] >> 47); */
            xxh_u64x2 const acc_vec  = xacc[i];
            xxh_u64x2 const data_vec = acc_vec ^ (acc_vec >> v47);

            /* xacc[i] ^= xsecret[i]; */
            xxh_u64x2 const key_vec  = XXH_vec_loadu(xsecret + 16*i);
            xxh_u64x2 const data_key = data_vec ^ key_vec;

            /* xacc[i] *= XXH_PRIME32_1 */
            /* prod_lo = ((xxh_u64x2)data_key & 0xFFFFFFFF) * ((xxh_u64x2)prime & 0xFFFFFFFF);  */
            xxh_u64x2 const prod_even  = XXH_vec_mule((xxh_u32x4)data_key, prime);
            /* prod_hi = ((xxh_u64x2)data_key >> 32) * ((xxh_u64x2)prime >> 32);  */
            xxh_u64x2 const prod_odd  = XXH_vec_mulo((xxh_u32x4)data_key, prime);
            xacc[i] = prod_odd + (prod_even << v32);
    }   }
}

#endif

#if (XXH_VECTOR == XXH_SVE)

XXH_FORCE_INLINE void
XXH3_accumulate_512_sve( void* XXH_RESTRICT acc,
                   const void* XXH_RESTRICT input,
                   const void* XXH_RESTRICT secret)
{
    uint64_t *xacc = (uint64_t *)acc;
    const uint64_t *xinput = (const uint64_t *)(const void *)input;
    const uint64_t *xsecret = (const uint64_t *)(const void *)secret;
    svuint64_t kSwap = sveor_n_u64_z(svptrue_b64(), svindex_u64(0, 1), 1);
    uint64_t element_count = svcntd();
    if (element_count >= 8) {
        svbool_t mask = svptrue_pat_b64(SV_VL8);
        svuint64_t vacc = svld1_u64(mask, xacc);
        ACCRND(vacc, 0);
        svst1_u64(mask, xacc, vacc);
    } else if (element_count == 2) {   /* sve128 */
        svbool_t mask = svptrue_pat_b64(SV_VL2);
        svuint64_t acc0 = svld1_u64(mask, xacc + 0);
        svuint64_t acc1 = svld1_u64(mask, xacc + 2);
        svuint64_t acc2 = svld1_u64(mask, xacc + 4);
        svuint64_t acc3 = svld1_u64(mask, xacc + 6);
        ACCRND(acc0, 0);
        ACCRND(acc1, 2);
        ACCRND(acc2, 4);
        ACCRND(acc3, 6);
        svst1_u64(mask, xacc + 0, acc0);
        svst1_u64(mask, xacc + 2, acc1);
        svst1_u64(mask, xacc + 4, acc2);
        svst1_u64(mask, xacc + 6, acc3);
    } else {
        svbool_t mask = svptrue_pat_b64(SV_VL4);
        svuint64_t acc0 = svld1_u64(mask, xacc + 0);
        svuint64_t acc1 = svld1_u64(mask, xacc + 4);
        ACCRND(acc0, 0);
        ACCRND(acc1, 4);
        svst1_u64(mask, xacc + 0, acc0);
        svst1_u64(mask, xacc + 4, acc1);
    }
}

XXH_FORCE_INLINE void
XXH3_accumulate_sve(xxh_u64* XXH_RESTRICT acc,
               const xxh_u8* XXH_RESTRICT input,
               const xxh_u8* XXH_RESTRICT secret,
               size_t nbStripes)
{
    if (nbStripes != 0) {
        uint64_t *xacc = (uint64_t *)acc;
        const uint64_t *xinput = (const uint64_t *)(const void *)input;
        const uint64_t *xsecret = (const uint64_t *)(const void *)secret;
        svuint64_t kSwap = sveor_n_u64_z(svptrue_b64(), svindex_u64(0, 1), 1);
        uint64_t element_count = svcntd();
        if (element_count >= 8) {
            svbool_t mask = svptrue_pat_b64(SV_VL8);
            svuint64_t vacc = svld1_u64(mask, xacc + 0);
            do {
                /* svprfd(svbool_t, void *, enum svfprop); */
                svprfd(mask, xinput + 128, SV_PLDL1STRM);
                ACCRND(vacc, 0);
                xinput += 8;
                xsecret += 1;
                nbStripes--;
           } while (nbStripes != 0);

           svst1_u64(mask, xacc + 0, vacc);
        } else if (element_count == 2) { /* sve128 */
            svbool_t mask = svptrue_pat_b64(SV_VL2);
            svuint64_t acc0 = svld1_u64(mask, xacc + 0);
            svuint64_t acc1 = svld1_u64(mask, xacc + 2);
            svuint64_t acc2 = svld1_u64(mask, xacc + 4);
            svuint64_t acc3 = svld1_u64(mask, xacc + 6);
            do {
                svprfd(mask, xinput + 128, SV_PLDL1STRM);
                ACCRND(acc0, 0);
                ACCRND(acc1, 2);
                ACCRND(acc2, 4);
                ACCRND(acc3, 6);
                xinput += 8;
                xsecret += 1;
                nbStripes--;
           } while (nbStripes != 0);

           svst1_u64(mask, xacc + 0, acc0);
           svst1_u64(mask, xacc + 2, acc1);
           svst1_u64(mask, xacc + 4, acc2);
           svst1_u64(mask, xacc + 6, acc3);
        } else {
            svbool_t mask = svptrue_pat_b64(SV_VL4);
            svuint64_t acc0 = svld1_u64(mask, xacc + 0);
            svuint64_t acc1 = svld1_u64(mask, xacc + 4);
            do {
                svprfd(mask, xinput + 128, SV_PLDL1STRM);
                ACCRND(acc0, 0);
                ACCRND(acc1, 4);
                xinput += 8;
                xsecret += 1;
                nbStripes--;
           } while (nbStripes != 0);

           svst1_u64(mask, xacc + 0, acc0);
           svst1_u64(mask, xacc + 4, acc1);
       }
    }
}

#endif

/* scalar variants - universal */

#if defined(__aarch64__) && (defined(__GNUC__) || defined(__clang__))
/*
 * In XXH3_scalarRound(), GCC and Clang have a similar codegen issue, where they
 * emit an excess mask and a full 64-bit multiply-add (MADD X-form).
 *
 * While this might not seem like much, as AArch64 is a 64-bit architecture, only
 * big Cortex designs have a full 64-bit multiplier.
 *
 * On the little cores, the smaller 32-bit multiplier is used, and full 64-bit
 * multiplies expand to 2-3 multiplies in microcode. This has a major penalty
 * of up to 4 latency cycles and 2 stall cycles in the multiply pipeline.
 *
 * Thankfully, AArch64 still provides the 32-bit long multiply-add (UMADDL) which does
 * not have this penalty and does the mask automatically.
 */
XXH_FORCE_INLINE xxh_u64
XXH_mult32to64_add64(xxh_u64 lhs, xxh_u64 rhs, xxh_u64 acc)
{
    xxh_u64 ret;
    /* note: %x = 64-bit register, %w = 32-bit register */
    __asm__("umaddl %x0, %w1, %w2, %x3" : "=r" (ret) : "r" (lhs), "r" (rhs), "r" (acc));
    return ret;
}
#else
XXH_FORCE_INLINE xxh_u64
XXH_mult32to64_add64(xxh_u64 lhs, xxh_u64 rhs, xxh_u64 acc)
{
    return XXH_mult32to64((xxh_u32)lhs, (xxh_u32)rhs) + acc;
}
#endif

/*!
 * @internal
 * @brief Scalar round for @ref XXH3_accumulate_512_scalar().
 *
 * This is extracted to its own function because the NEON path uses a combination
 * of NEON and scalar.
 */
XXH_FORCE_INLINE void
XXH3_scalarRound(void* XXH_RESTRICT acc,
                 void const* XXH_RESTRICT input,
                 void const* XXH_RESTRICT secret,
                 size_t lane)
{
    xxh_u64* xacc = (xxh_u64*) acc;
    xxh_u8 const* xinput  = (xxh_u8 const*) input;
    xxh_u8 const* xsecret = (xxh_u8 const*) secret;
    XXH_ASSERT(lane < XXH_ACC_NB);
    XXH_ASSERT(((size_t)acc & (XXH_ACC_ALIGN-1)) == 0);
    {
        xxh_u64 const data_val = XXH_readLE64(xinput + lane * 8);
        xxh_u64 const data_key = data_val ^ XXH_readLE64(xsecret + lane * 8);
        xacc[lane ^ 1] += data_val; /* swap adjacent lanes */
        xacc[lane] = XXH_mult32to64_add64(data_key /* & 0xFFFFFFFF */, data_key >> 32, xacc[lane]);
    }
}

/*!
 * @internal
 * @brief Processes a 64 byte block of data using the scalar path.
 */
XXH_FORCE_INLINE void
XXH3_accumulate_512_scalar(void* XXH_RESTRICT acc,
                     const void* XXH_RESTRICT input,
                     const void* XXH_RESTRICT secret)
{
    size_t i;
    /* ARM GCC refuses to unroll this loop, resulting in a 24% slowdown on ARMv6. */
#if defined(__GNUC__) && !defined(__clang__) \
  && (defined(__arm__) || defined(__thumb2__)) \
  && defined(__ARM_FEATURE_UNALIGNED) /* no unaligned access just wastes bytes */ \
  && XXH_SIZE_OPT <= 0
#  pragma GCC unroll 8
#endif
    for (i=0; i < XXH_ACC_NB; i++) {
        XXH3_scalarRound(acc, input, secret, i);
    }
}
XXH_FORCE_INLINE XXH3_ACCUMULATE_TEMPLATE(scalar)

/*!
 * @internal
 * @brief Scalar scramble step for @ref XXH3_scrambleAcc_scalar().
 *
 * This is extracted to its own function because the NEON path uses a combination
 * of NEON and scalar.
 */
XXH_FORCE_INLINE void
XXH3_scalarScrambleRound(void* XXH_RESTRICT acc,
                         void const* XXH_RESTRICT secret,
                         size_t lane)
{
    xxh_u64* const xacc = (xxh_u64*) acc;   /* presumed aligned */
    const xxh_u8* const xsecret = (const xxh_u8*) secret;   /* no alignment restriction */
    XXH_ASSERT((((size_t)acc) & (XXH_ACC_ALIGN-1)) == 0);
    XXH_ASSERT(lane < XXH_ACC_NB);
    {
        xxh_u64 const key64 = XXH_readLE64(xsecret + lane * 8);
        xxh_u64 acc64 = xacc[lane];
        acc64 = XXH_xorshift64(acc64, 47);
        acc64 ^= key64;
        acc64 *= XXH_PRIME32_1;
        xacc[lane] = acc64;
    }
}

/*!
 * @internal
 * @brief Scrambles the accumulators after a large chunk has been read
 */
XXH_FORCE_INLINE void
XXH3_scrambleAcc_scalar(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
{
    size_t i;
    for (i=0; i < XXH_ACC_NB; i++) {
        XXH3_scalarScrambleRound(acc, secret, i);
    }
}

XXH_FORCE_INLINE void
XXH3_initCustomSecret_scalar(void* XXH_RESTRICT customSecret, xxh_u64 seed64)
{
    /*
     * We need a separate pointer for the hack below,
     * which requires a non-const pointer.
     * Any decent compiler will optimize this out otherwise.
     */
    const xxh_u8* kSecretPtr = XXH3_kSecret;
    XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 15) == 0);

#if defined(__GNUC__) && defined(__aarch64__)
    /*
     * UGLY HACK:
     * GCC and Clang generate a bunch of MOV/MOVK pairs for aarch64, and they are
     * placed sequentially, in order, at the top of the unrolled loop.
     *
     * While MOVK is great for generating constants (2 cycles for a 64-bit
     * constant compared to 4 cycles for LDR), it fights for bandwidth with
     * the arithmetic instructions.
     *
     *   I   L   S
     * MOVK
     * MOVK
     * MOVK
     * MOVK
     * ADD
     * SUB      STR
     *          STR
     * By forcing loads from memory (as the asm line causes the compiler to assume
     * that XXH3_kSecretPtr has been changed), the pipelines are used more
     * efficiently:
     *   I   L   S
     *      LDR
     *  ADD LDR
     *  SUB     STR
     *          STR
     *
     * See XXH3_NEON_LANES for details on the pipsline.
     *
     * XXH3_64bits_withSeed, len == 256, Snapdragon 835
     *   without hack: 2654.4 MB/s
     *   with hack:    3202.9 MB/s
     */
    XXH_COMPILER_GUARD(kSecretPtr);
#endif
    {   int const nbRounds = XXH_SECRET_DEFAULT_SIZE / 16;
        int i;
        for (i=0; i < nbRounds; i++) {
            /*
             * The asm hack causes the compiler to assume that kSecretPtr aliases with
             * customSecret, and on aarch64, this prevented LDP from merging two
             * loads together for free. Putting the loads together before the stores
             * properly generates LDP.
             */
            xxh_u64 lo = XXH_readLE64(kSecretPtr + 16*i)     + seed64;
            xxh_u64 hi = XXH_readLE64(kSecretPtr + 16*i + 8) - seed64;
            XXH_writeLE64((xxh_u8*)customSecret + 16*i,     lo);
            XXH_writeLE64((xxh_u8*)customSecret + 16*i + 8, hi);
    }   }
}


typedef void (*XXH3_f_accumulate)(xxh_u64* XXH_RESTRICT, const xxh_u8* XXH_RESTRICT, const xxh_u8* XXH_RESTRICT, size_t);
typedef void (*XXH3_f_scrambleAcc)(void* XXH_RESTRICT, const void*);
typedef void (*XXH3_f_initCustomSecret)(void* XXH_RESTRICT, xxh_u64);


#if (XXH_VECTOR == XXH_AVX512)

#define XXH3_accumulate_512 XXH3_accumulate_512_avx512
#define XXH3_accumulate     XXH3_accumulate_avx512
#define XXH3_scrambleAcc    XXH3_scrambleAcc_avx512
#define XXH3_initCustomSecret XXH3_initCustomSecret_avx512

#elif (XXH_VECTOR == XXH_AVX2)

#define XXH3_accumulate_512 XXH3_accumulate_512_avx2
#define XXH3_accumulate     XXH3_accumulate_avx2
#define XXH3_scrambleAcc    XXH3_scrambleAcc_avx2
#define XXH3_initCustomSecret XXH3_initCustomSecret_avx2

#elif (XXH_VECTOR == XXH_SSE2)

#define XXH3_accumulate_512 XXH3_accumulate_512_sse2
#define XXH3_accumulate     XXH3_accumulate_sse2
#define XXH3_scrambleAcc    XXH3_scrambleAcc_sse2
#define XXH3_initCustomSecret XXH3_initCustomSecret_sse2

#elif (XXH_VECTOR == XXH_NEON)

#define XXH3_accumulate_512 XXH3_accumulate_512_neon
#define XXH3_accumulate     XXH3_accumulate_neon
#define XXH3_scrambleAcc    XXH3_scrambleAcc_neon
#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar

#elif (XXH_VECTOR == XXH_VSX)

#define XXH3_accumulate_512 XXH3_accumulate_512_vsx
#define XXH3_accumulate     XXH3_accumulate_vsx
#define XXH3_scrambleAcc    XXH3_scrambleAcc_vsx
#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar

#elif (XXH_VECTOR == XXH_SVE)
#define XXH3_accumulate_512 XXH3_accumulate_512_sve
#define XXH3_accumulate     XXH3_accumulate_sve
#define XXH3_scrambleAcc    XXH3_scrambleAcc_scalar
#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar

#else /* scalar */

#define XXH3_accumulate_512 XXH3_accumulate_512_scalar
#define XXH3_accumulate     XXH3_accumulate_scalar
#define XXH3_scrambleAcc    XXH3_scrambleAcc_scalar
#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar

#endif

#if XXH_SIZE_OPT >= 1 /* don't do SIMD for initialization */
#  undef XXH3_initCustomSecret
#  define XXH3_initCustomSecret XXH3_initCustomSecret_scalar
#endif

XXH_FORCE_INLINE void
XXH3_hashLong_internal_loop(xxh_u64* XXH_RESTRICT acc,
                      const xxh_u8* XXH_RESTRICT input, size_t len,
                      const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
                            XXH3_f_accumulate f_acc,
                            XXH3_f_scrambleAcc f_scramble)
{
    size_t const nbStripesPerBlock = (secretSize - XXH_STRIPE_LEN) / XXH_SECRET_CONSUME_RATE;
    size_t const block_len = XXH_STRIPE_LEN * nbStripesPerBlock;
    size_t const nb_blocks = (len - 1) / block_len;

    size_t n;

    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN);

    for (n = 0; n < nb_blocks; n++) {
        f_acc(acc, input + n*block_len, secret, nbStripesPerBlock);
        f_scramble(acc, secret + secretSize - XXH_STRIPE_LEN);
    }

    /* last partial block */
    XXH_ASSERT(len > XXH_STRIPE_LEN);
    {   size_t const nbStripes = ((len - 1) - (block_len * nb_blocks)) / XXH_STRIPE_LEN;
        XXH_ASSERT(nbStripes <= (secretSize / XXH_SECRET_CONSUME_RATE));
        f_acc(acc, input + nb_blocks*block_len, secret, nbStripes);

        /* last stripe */
        {   const xxh_u8* const p = input + len - XXH_STRIPE_LEN;
#define XXH_SECRET_LASTACC_START 7  /* not aligned on 8, last secret is different from acc & scrambler */
            XXH3_accumulate_512(acc, p, secret + secretSize - XXH_STRIPE_LEN - XXH_SECRET_LASTACC_START);
    }   }
}

XXH_FORCE_INLINE xxh_u64
XXH3_mix2Accs(const xxh_u64* XXH_RESTRICT acc, const xxh_u8* XXH_RESTRICT secret)
{
    return XXH3_mul128_fold64(
               acc[0] ^ XXH_readLE64(secret),
               acc[1] ^ XXH_readLE64(secret+8) );
}

static XXH64_hash_t
XXH3_mergeAccs(const xxh_u64* XXH_RESTRICT acc, const xxh_u8* XXH_RESTRICT secret, xxh_u64 start)
{
    xxh_u64 result64 = start;
    size_t i = 0;

    for (i = 0; i < 4; i++) {
        result64 += XXH3_mix2Accs(acc+2*i, secret + 16*i);
#if defined(__clang__)                                /* Clang */ \
    && (defined(__arm__) || defined(__thumb__))       /* ARMv7 */ \
    && (defined(__ARM_NEON) || defined(__ARM_NEON__)) /* NEON */  \
    && !defined(XXH_ENABLE_AUTOVECTORIZE)             /* Define to disable */
        /*
         * UGLY HACK:
         * Prevent autovectorization on Clang ARMv7-a. Exact same problem as
         * the one in XXH3_len_129to240_64b. Speeds up shorter keys > 240b.
         * XXH3_64bits, len == 256, Snapdragon 835:
         *   without hack: 2063.7 MB/s
         *   with hack:    2560.7 MB/s
         */
        XXH_COMPILER_GUARD(result64);
#endif
    }

    return XXH3_avalanche(result64);
}

#define XXH3_INIT_ACC { XXH_PRIME32_3, XXH_PRIME64_1, XXH_PRIME64_2, XXH_PRIME64_3, \
                        XXH_PRIME64_4, XXH_PRIME32_2, XXH_PRIME64_5, XXH_PRIME32_1 }

XXH_FORCE_INLINE XXH64_hash_t
XXH3_hashLong_64b_internal(const void* XXH_RESTRICT input, size_t len,
                           const void* XXH_RESTRICT secret, size_t secretSize,
                           XXH3_f_accumulate f_acc,
                           XXH3_f_scrambleAcc f_scramble)
{
    XXH_ALIGN(XXH_ACC_ALIGN) xxh_u64 acc[XXH_ACC_NB] = XXH3_INIT_ACC;

    XXH3_hashLong_internal_loop(acc, (const xxh_u8*)input, len, (const xxh_u8*)secret, secretSize, f_acc, f_scramble);

    /* converge into final hash */
    XXH_STATIC_ASSERT(sizeof(acc) == 64);
    /* do not align on 8, so that the secret is different from the accumulator */
#define XXH_SECRET_MERGEACCS_START 11
    XXH_ASSERT(secretSize >= sizeof(acc) + XXH_SECRET_MERGEACCS_START);
    return XXH3_mergeAccs(acc, (const xxh_u8*)secret + XXH_SECRET_MERGEACCS_START, (xxh_u64)len * XXH_PRIME64_1);
}

/*
 * It's important for performance to transmit secret's size (when it's static)
 * so that the compiler can properly optimize the vectorized loop.
 * This makes a big performance difference for "medium" keys (<1 KB) when using AVX instruction set.
 * When the secret size is unknown, or on GCC 12 where the mix of NO_INLINE and FORCE_INLINE
 * breaks -Og, this is XXH_NO_INLINE.
 */
XXH3_WITH_SECRET_INLINE XXH64_hash_t
XXH3_hashLong_64b_withSecret(const void* XXH_RESTRICT input, size_t len,
                             XXH64_hash_t seed64, const xxh_u8* XXH_RESTRICT secret, size_t secretLen)
{
    (void)seed64;
    return XXH3_hashLong_64b_internal(input, len, secret, secretLen, XXH3_accumulate, XXH3_scrambleAcc);
}

/*
 * It's preferable for performance that XXH3_hashLong is not inlined,
 * as it results in a smaller function for small data, easier to the instruction cache.
 * Note that inside this no_inline function, we do inline the internal loop,
 * and provide a statically defined secret size to allow optimization of vector loop.
 */
XXH_NO_INLINE XXH_PUREF XXH64_hash_t
XXH3_hashLong_64b_default(const void* XXH_RESTRICT input, size_t len,
                          XXH64_hash_t seed64, const xxh_u8* XXH_RESTRICT secret, size_t secretLen)
{
    (void)seed64; (void)secret; (void)secretLen;
    return XXH3_hashLong_64b_internal(input, len, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_accumulate, XXH3_scrambleAcc);
}

/*
 * XXH3_hashLong_64b_withSeed():
 * Generate a custom key based on alteration of default XXH3_kSecret with the seed,
 * and then use this key for long mode hashing.
 *
 * This operation is decently fast but nonetheless costs a little bit of time.
 * Try to avoid it whenever possible (typically when seed==0).
 *
 * It's important for performance that XXH3_hashLong is not inlined. Not sure
 * why (uop cache maybe?), but the difference is large and easily measurable.
 */
XXH_FORCE_INLINE XXH64_hash_t
XXH3_hashLong_64b_withSeed_internal(const void* input, size_t len,
                                    XXH64_hash_t seed,
                                    XXH3_f_accumulate f_acc,
                                    XXH3_f_scrambleAcc f_scramble,
                                    XXH3_f_initCustomSecret f_initSec)
{
#if XXH_SIZE_OPT <= 0
    if (seed == 0)
        return XXH3_hashLong_64b_internal(input, len,
                                          XXH3_kSecret, sizeof(XXH3_kSecret),
                                          f_acc, f_scramble);
#endif
    {   XXH_ALIGN(XXH_SEC_ALIGN) xxh_u8 secret[XXH_SECRET_DEFAULT_SIZE];
        f_initSec(secret, seed);
        return XXH3_hashLong_64b_internal(input, len, secret, sizeof(secret),
                                          f_acc, f_scramble);
    }
}

/*
 * It's important for performance that XXH3_hashLong is not inlined.
 */
XXH_NO_INLINE XXH64_hash_t
XXH3_hashLong_64b_withSeed(const void* XXH_RESTRICT input, size_t len,
                           XXH64_hash_t seed, const xxh_u8* XXH_RESTRICT secret, size_t secretLen)
{
    (void)secret; (void)secretLen;
    return XXH3_hashLong_64b_withSeed_internal(input, len, seed,
                XXH3_accumulate, XXH3_scrambleAcc, XXH3_initCustomSecret);
}


typedef XXH64_hash_t (*XXH3_hashLong64_f)(const void* XXH_RESTRICT, size_t,
                                          XXH64_hash_t, const xxh_u8* XXH_RESTRICT, size_t);

XXH_FORCE_INLINE XXH64_hash_t
XXH3_64bits_internal(const void* XXH_RESTRICT input, size_t len,
                     XXH64_hash_t seed64, const void* XXH_RESTRICT secret, size_t secretLen,
                     XXH3_hashLong64_f f_hashLong)
{
    XXH_ASSERT(secretLen >= XXH3_SECRET_SIZE_MIN);
    /*
     * If an action is to be taken if `secretLen` condition is not respected,
     * it should be done here.
     * For now, it's a contract pre-condition.
     * Adding a check and a branch here would cost performance at every hash.
     * Also, note that function signature doesn't offer room to return an error.
     */
    if (len <= 16)
        return XXH3_len_0to16_64b((const xxh_u8*)input, len, (const xxh_u8*)secret, seed64);
    if (len <= 128)
        return XXH3_len_17to128_64b((const xxh_u8*)input, len, (const xxh_u8*)secret, secretLen, seed64);
    if (len <= XXH3_MIDSIZE_MAX)
        return XXH3_len_129to240_64b((const xxh_u8*)input, len, (const xxh_u8*)secret, secretLen, seed64);
    return f_hashLong(input, len, seed64, (const xxh_u8*)secret, secretLen);
}


/* ===   Public entry point   === */

/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH64_hash_t XXH3_64bits(XXH_NOESCAPE const void* input, size_t length)
{
    return XXH3_64bits_internal(input, length, 0, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_hashLong_64b_default);
}

/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH64_hash_t
XXH3_64bits_withSecret(XXH_NOESCAPE const void* input, size_t length, XXH_NOESCAPE const void* secret, size_t secretSize)
{
    return XXH3_64bits_internal(input, length, 0, secret, secretSize, XXH3_hashLong_64b_withSecret);
}

/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH64_hash_t
XXH3_64bits_withSeed(XXH_NOESCAPE const void* input, size_t length, XXH64_hash_t seed)
{
    return XXH3_64bits_internal(input, length, seed, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_hashLong_64b_withSeed);
}

XXH_PUBLIC_API XXH64_hash_t
XXH3_64bits_withSecretandSeed(XXH_NOESCAPE const void* input, size_t length, XXH_NOESCAPE const void* secret, size_t secretSize, XXH64_hash_t seed)
{
    if (length <= XXH3_MIDSIZE_MAX)
        return XXH3_64bits_internal(input, length, seed, XXH3_kSecret, sizeof(XXH3_kSecret), NULL);
    return XXH3_hashLong_64b_withSecret(input, length, seed, (const xxh_u8*)secret, secretSize);
}


/* ===   XXH3 streaming   === */
#ifndef XXH_NO_STREAM
/*
 * Malloc's a pointer that is always aligned to align.
 *
 * This must be freed with `XXH_alignedFree()`.
 *
 * malloc typically guarantees 16 byte alignment on 64-bit systems and 8 byte
 * alignment on 32-bit. This isn't enough for the 32 byte aligned loads in AVX2
 * or on 32-bit, the 16 byte aligned loads in SSE2 and NEON.
 *
 * This underalignment previously caused a rather obvious crash which went
 * completely unnoticed due to XXH3_createState() not actually being tested.
 * Credit to RedSpah for noticing this bug.
 *
 * The alignment is done manually: Functions like posix_memalign or _mm_malloc
 * are avoided: To maintain portability, we would have to write a fallback
 * like this anyways, and besides, testing for the existence of library
 * functions without relying on external build tools is impossible.
 *
 * The method is simple: Overallocate, manually align, and store the offset
 * to the original behind the returned pointer.
 *
 * Align must be a power of 2 and 8 <= align <= 128.
 */
static XXH_MALLOCF void* XXH_alignedMalloc(size_t s, size_t align)
{
    XXH_ASSERT(align <= 128 && align >= 8); /* range check */
    XXH_ASSERT((align & (align-1)) == 0);   /* power of 2 */
    XXH_ASSERT(s != 0 && s < (s + align));  /* empty/overflow */
    {   /* Overallocate to make room for manual realignment and an offset byte */
        xxh_u8* base = (xxh_u8*)XXH_malloc(s + align);
        if (base != NULL) {
            /*
             * Get the offset needed to align this pointer.
             *
             * Even if the returned pointer is aligned, there will always be
             * at least one byte to store the offset to the original pointer.
             */
            size_t offset = align - ((size_t)base & (align - 1)); /* base % align */
            /* Add the offset for the now-aligned pointer */
            xxh_u8* ptr = base + offset;

            XXH_ASSERT((size_t)ptr % align == 0);

            /* Store the offset immediately before the returned pointer. */
            ptr[-1] = (xxh_u8)offset;
            return ptr;
        }
        return NULL;
    }
}
/*
 * Frees an aligned pointer allocated by XXH_alignedMalloc(). Don't pass
 * normal malloc'd pointers, XXH_alignedMalloc has a specific data layout.
 */
static void XXH_alignedFree(void* p)
{
    if (p != NULL) {
        xxh_u8* ptr = (xxh_u8*)p;
        /* Get the offset byte we added in XXH_malloc. */
        xxh_u8 offset = ptr[-1];
        /* Free the original malloc'd pointer */
        xxh_u8* base = ptr - offset;
        XXH_free(base);
    }
}
/*! @ingroup XXH3_family */
/*!
 * @brief Allocate an @ref XXH3_state_t.
 *
 * Must be freed with XXH3_freeState().
 * @return An allocated XXH3_state_t on success, `NULL` on failure.
 */
XXH_PUBLIC_API XXH3_state_t* XXH3_createState(void)
{
    XXH3_state_t* const state = (XXH3_state_t*)XXH_alignedMalloc(sizeof(XXH3_state_t), 64);
    if (state==NULL) return NULL;
    XXH3_INITSTATE(state);
    return state;
}

/*! @ingroup XXH3_family */
/*!
 * @brief Frees an @ref XXH3_state_t.
 *
 * Must be allocated with XXH3_createState().
 * @param statePtr A pointer to an @ref XXH3_state_t allocated with @ref XXH3_createState().
 * @return XXH_OK.
 */
XXH_PUBLIC_API XXH_errorcode XXH3_freeState(XXH3_state_t* statePtr)
{
    XXH_alignedFree(statePtr);
    return XXH_OK;
}

/*! @ingroup XXH3_family */
XXH_PUBLIC_API void
XXH3_copyState(XXH_NOESCAPE XXH3_state_t* dst_state, XXH_NOESCAPE const XXH3_state_t* src_state)
{
    XXH_memcpy(dst_state, src_state, sizeof(*dst_state));
}

static void
XXH3_reset_internal(XXH3_state_t* statePtr,
                    XXH64_hash_t seed,
                    const void* secret, size_t secretSize)
{
    size_t const initStart = offsetof(XXH3_state_t, bufferedSize);
    size_t const initLength = offsetof(XXH3_state_t, nbStripesPerBlock) - initStart;
    XXH_ASSERT(offsetof(XXH3_state_t, nbStripesPerBlock) > initStart);
    XXH_ASSERT(statePtr != NULL);
    /* set members from bufferedSize to nbStripesPerBlock (excluded) to 0 */
    memset((char*)statePtr + initStart, 0, initLength);
    statePtr->acc[0] = XXH_PRIME32_3;
    statePtr->acc[1] = XXH_PRIME64_1;
    statePtr->acc[2] = XXH_PRIME64_2;
    statePtr->acc[3] = XXH_PRIME64_3;
    statePtr->acc[4] = XXH_PRIME64_4;
    statePtr->acc[5] = XXH_PRIME32_2;
    statePtr->acc[6] = XXH_PRIME64_5;
    statePtr->acc[7] = XXH_PRIME32_1;
    statePtr->seed = seed;
    statePtr->useSeed = (seed != 0);
    statePtr->extSecret = (const unsigned char*)secret;
    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN);
    statePtr->secretLimit = secretSize - XXH_STRIPE_LEN;
    statePtr->nbStripesPerBlock = statePtr->secretLimit / XXH_SECRET_CONSUME_RATE;
}

/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH_errorcode
XXH3_64bits_reset(XXH_NOESCAPE XXH3_state_t* statePtr)
{
    if (statePtr == NULL) return XXH_ERROR;
    XXH3_reset_internal(statePtr, 0, XXH3_kSecret, XXH_SECRET_DEFAULT_SIZE);
    return XXH_OK;
}

/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH_errorcode
XXH3_64bits_reset_withSecret(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* secret, size_t secretSize)
{
    if (statePtr == NULL) return XXH_ERROR;
    XXH3_reset_internal(statePtr, 0, secret, secretSize);
    if (secret == NULL) return XXH_ERROR;
    if (secretSize < XXH3_SECRET_SIZE_MIN) return XXH_ERROR;
    return XXH_OK;
}

/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH_errorcode
XXH3_64bits_reset_withSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH64_hash_t seed)
{
    if (statePtr == NULL) return XXH_ERROR;
    if (seed==0) return XXH3_64bits_reset(statePtr);
    if ((seed != statePtr->seed) || (statePtr->extSecret != NULL))
        XXH3_initCustomSecret(statePtr->customSecret, seed);
    XXH3_reset_internal(statePtr, seed, NULL, XXH_SECRET_DEFAULT_SIZE);
    return XXH_OK;
}

/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH_errorcode
XXH3_64bits_reset_withSecretandSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* secret, size_t secretSize, XXH64_hash_t seed64)
{
    if (statePtr == NULL) return XXH_ERROR;
    if (secret == NULL) return XXH_ERROR;
    if (secretSize < XXH3_SECRET_SIZE_MIN) return XXH_ERROR;
    XXH3_reset_internal(statePtr, seed64, secret, secretSize);
    statePtr->useSeed = 1; /* always, even if seed64==0 */
    return XXH_OK;
}

/*!
 * @internal
 * @brief Processes a large input for XXH3_update() and XXH3_digest_long().
 *
 * Unlike XXH3_hashLong_internal_loop(), this can process data that overlaps a block.
 *
 * @param acc                Pointer to the 8 accumulator lanes
 * @param nbStripesSoFarPtr  In/out pointer to the number of leftover stripes in the block*
 * @param nbStripesPerBlock  Number of stripes in a block
 * @param input              Input pointer
 * @param nbStripes          Number of stripes to process
 * @param secret             Secret pointer
 * @param secretLimit        Offset of the last block in @p secret
 * @param f_acc              Pointer to an XXH3_accumulate implementation
 * @param f_scramble         Pointer to an XXH3_scrambleAcc implementation
 * @return                   Pointer past the end of @p input after processing
 */
XXH_FORCE_INLINE const xxh_u8 *
XXH3_consumeStripes(xxh_u64* XXH_RESTRICT acc,
                    size_t* XXH_RESTRICT nbStripesSoFarPtr, size_t nbStripesPerBlock,
                    const xxh_u8* XXH_RESTRICT input, size_t nbStripes,
                    const xxh_u8* XXH_RESTRICT secret, size_t secretLimit,
                    XXH3_f_accumulate f_acc,
                    XXH3_f_scrambleAcc f_scramble)
{
    const xxh_u8* initialSecret = secret + *nbStripesSoFarPtr * XXH_SECRET_CONSUME_RATE;
    /* Process full blocks */
    if (nbStripes >= (nbStripesPerBlock - *nbStripesSoFarPtr)) {
        /* Process the initial partial block... */
        size_t nbStripesThisIter = nbStripesPerBlock - *nbStripesSoFarPtr;

        do {
            /* Accumulate and scramble */
            f_acc(acc, input, initialSecret, nbStripesThisIter);
            f_scramble(acc, secret + secretLimit);
            input += nbStripesThisIter * XXH_STRIPE_LEN;
            nbStripes -= nbStripesThisIter;
            /* Then continue the loop with the full block size */
            nbStripesThisIter = nbStripesPerBlock;
            initialSecret = secret;
        } while (nbStripes >= nbStripesPerBlock);
        *nbStripesSoFarPtr = 0;
    }
    /* Process a partial block */
    if (nbStripes > 0) {
        f_acc(acc, input, initialSecret, nbStripes);
        input += nbStripes * XXH_STRIPE_LEN;
        *nbStripesSoFarPtr += nbStripes;
    }
    /* Return end pointer */
    return input;
}

#ifndef XXH3_STREAM_USE_STACK
# if XXH_SIZE_OPT <= 0 && !defined(__clang__) /* clang doesn't need additional stack space */
#   define XXH3_STREAM_USE_STACK 1
# endif
#endif
/*
 * Both XXH3_64bits_update and XXH3_128bits_update use this routine.
 */
XXH_FORCE_INLINE XXH_errorcode
XXH3_update(XXH3_state_t* XXH_RESTRICT const state,
            const xxh_u8* XXH_RESTRICT input, size_t len,
            XXH3_f_accumulate f_acc,
            XXH3_f_scrambleAcc f_scramble)
{
    if (input==NULL) {
        XXH_ASSERT(len == 0);
        return XXH_OK;
    }

    XXH_ASSERT(state != NULL);
    {   const xxh_u8* const bEnd = input + len;
        const unsigned char* const secret = (state->extSecret == NULL) ? state->customSecret : state->extSecret;
#if defined(XXH3_STREAM_USE_STACK) && XXH3_STREAM_USE_STACK >= 1
        /* For some reason, gcc and MSVC seem to suffer greatly
         * when operating accumulators directly into state.
         * Operating into stack space seems to enable proper optimization.
         * clang, on the other hand, doesn't seem to need this trick */
        XXH_ALIGN(XXH_ACC_ALIGN) xxh_u64 acc[8];
        XXH_memcpy(acc, state->acc, sizeof(acc));
#else
        xxh_u64* XXH_RESTRICT const acc = state->acc;
#endif
        state->totalLen += len;
        XXH_ASSERT(state->bufferedSize <= XXH3_INTERNALBUFFER_SIZE);

        /* small input : just fill in tmp buffer */
        if (len <= XXH3_INTERNALBUFFER_SIZE - state->bufferedSize) {
            XXH_memcpy(state->buffer + state->bufferedSize, input, len);
            state->bufferedSize += (XXH32_hash_t)len;
            return XXH_OK;
        }

        /* total input is now > XXH3_INTERNALBUFFER_SIZE */
        #define XXH3_INTERNALBUFFER_STRIPES (XXH3_INTERNALBUFFER_SIZE / XXH_STRIPE_LEN)
        XXH_STATIC_ASSERT(XXH3_INTERNALBUFFER_SIZE % XXH_STRIPE_LEN == 0);   /* clean multiple */

        /*
         * Internal buffer is partially filled (always, except at beginning)
         * Complete it, then consume it.
         */
        if (state->bufferedSize) {
            size_t const loadSize = XXH3_INTERNALBUFFER_SIZE - state->bufferedSize;
            XXH_memcpy(state->buffer + state->bufferedSize, input, loadSize);
            input += loadSize;
            XXH3_consumeStripes(acc,
                               &state->nbStripesSoFar, state->nbStripesPerBlock,
                                state->buffer, XXH3_INTERNALBUFFER_STRIPES,
                                secret, state->secretLimit,
                                f_acc, f_scramble);
            state->bufferedSize = 0;
        }
        XXH_ASSERT(input < bEnd);
        if (bEnd - input > XXH3_INTERNALBUFFER_SIZE) {
            size_t nbStripes = (size_t)(bEnd - 1 - input) / XXH_STRIPE_LEN;
            input = XXH3_consumeStripes(acc,
                                       &state->nbStripesSoFar, state->nbStripesPerBlock,
                                       input, nbStripes,
                                       secret, state->secretLimit,
                                       f_acc, f_scramble);
            XXH_memcpy(state->buffer + sizeof(state->buffer) - XXH_STRIPE_LEN, input - XXH_STRIPE_LEN, XXH_STRIPE_LEN);

        }
        /* Some remaining input (always) : buffer it */
        XXH_ASSERT(input < bEnd);
        XXH_ASSERT(bEnd - input <= XXH3_INTERNALBUFFER_SIZE);
        XXH_ASSERT(state->bufferedSize == 0);
        XXH_memcpy(state->buffer, input, (size_t)(bEnd-input));
        state->bufferedSize = (XXH32_hash_t)(bEnd-input);
#if defined(XXH3_STREAM_USE_STACK) && XXH3_STREAM_USE_STACK >= 1
        /* save stack accumulators into state */
        XXH_memcpy(state->acc, acc, sizeof(acc));
#endif
    }

    return XXH_OK;
}

/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH_errorcode
XXH3_64bits_update(XXH_NOESCAPE XXH3_state_t* state, XXH_NOESCAPE const void* input, size_t len)
{
    return XXH3_update(state, (const xxh_u8*)input, len,
                       XXH3_accumulate, XXH3_scrambleAcc);
}


XXH_FORCE_INLINE void
XXH3_digest_long (XXH64_hash_t* acc,
                  const XXH3_state_t* state,
                  const unsigned char* secret)
{
    xxh_u8 lastStripe[XXH_STRIPE_LEN];
    const xxh_u8* lastStripePtr;

    /*
     * Digest on a local copy. This way, the state remains unaltered, and it can
     * continue ingesting more input afterwards.
     */
    XXH_memcpy(acc, state->acc, sizeof(state->acc));
    if (state->bufferedSize >= XXH_STRIPE_LEN) {
        /* Consume remaining stripes then point to remaining data in buffer */
        size_t const nbStripes = (state->bufferedSize - 1) / XXH_STRIPE_LEN;
        size_t nbStripesSoFar = state->nbStripesSoFar;
        XXH3_consumeStripes(acc,
                           &nbStripesSoFar, state->nbStripesPerBlock,
                            state->buffer, nbStripes,
                            secret, state->secretLimit,
                            XXH3_accumulate, XXH3_scrambleAcc);
        lastStripePtr = state->buffer + state->bufferedSize - XXH_STRIPE_LEN;
    } else {  /* bufferedSize < XXH_STRIPE_LEN */
        /* Copy to temp buffer */
        size_t const catchupSize = XXH_STRIPE_LEN - state->bufferedSize;
        XXH_ASSERT(state->bufferedSize > 0);  /* there is always some input buffered */
        XXH_memcpy(lastStripe, state->buffer + sizeof(state->buffer) - catchupSize, catchupSize);
        XXH_memcpy(lastStripe + catchupSize, state->buffer, state->bufferedSize);
        lastStripePtr = lastStripe;
    }
    /* Last stripe */
    XXH3_accumulate_512(acc,
                        lastStripePtr,
                        secret + state->secretLimit - XXH_SECRET_LASTACC_START);
}

/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_digest (XXH_NOESCAPE const XXH3_state_t* state)
{
    const unsigned char* const secret = (state->extSecret == NULL) ? state->customSecret : state->extSecret;
    if (state->totalLen > XXH3_MIDSIZE_MAX) {
        XXH_ALIGN(XXH_ACC_ALIGN) XXH64_hash_t acc[XXH_ACC_NB];
        XXH3_digest_long(acc, state, secret);
        return XXH3_mergeAccs(acc,
                              secret + XXH_SECRET_MERGEACCS_START,
                              (xxh_u64)state->totalLen * XXH_PRIME64_1);
    }
    /* totalLen <= XXH3_MIDSIZE_MAX: digesting a short input */
    if (state->useSeed)
        return XXH3_64bits_withSeed(state->buffer, (size_t)state->totalLen, state->seed);
    return XXH3_64bits_withSecret(state->buffer, (size_t)(state->totalLen),
                                  secret, state->secretLimit + XXH_STRIPE_LEN);
}
#endif /* !XXH_NO_STREAM */


/* ==========================================
 * XXH3 128 bits (a.k.a XXH128)
 * ==========================================
 * XXH3's 128-bit variant has better mixing and strength than the 64-bit variant,
 * even without counting the significantly larger output size.
 *
 * For example, extra steps are taken to avoid the seed-dependent collisions
 * in 17-240 byte inputs (See XXH3_mix16B and XXH128_mix32B).
 *
 * This strength naturally comes at the cost of some speed, especially on short
 * lengths. Note that longer hashes are about as fast as the 64-bit version
 * due to it using only a slight modification of the 64-bit loop.
 *
 * XXH128 is also more oriented towards 64-bit machines. It is still extremely
 * fast for a _128-bit_ hash on 32-bit (it usually clears XXH64).
 */

XXH_FORCE_INLINE XXH_PUREF XXH128_hash_t
XXH3_len_1to3_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
{
    /* A doubled version of 1to3_64b with different constants. */
    XXH_ASSERT(input != NULL);
    XXH_ASSERT(1 <= len && len <= 3);
    XXH_ASSERT(secret != NULL);
    /*
     * len = 1: combinedl = { input[0], 0x01, input[0], input[0] }
     * len = 2: combinedl = { input[1], 0x02, input[0], input[1] }
     * len = 3: combinedl = { input[2], 0x03, input[0], input[1] }
     */
    {   xxh_u8 const c1 = input[0];
        xxh_u8 const c2 = input[len >> 1];
        xxh_u8 const c3 = input[len - 1];
        xxh_u32 const combinedl = ((xxh_u32)c1 <<16) | ((xxh_u32)c2 << 24)
                                | ((xxh_u32)c3 << 0) | ((xxh_u32)len << 8);
        xxh_u32 const combinedh = XXH_rotl32(XXH_swap32(combinedl), 13);
        xxh_u64 const bitflipl = (XXH_readLE32(secret) ^ XXH_readLE32(secret+4)) + seed;
        xxh_u64 const bitfliph = (XXH_readLE32(secret+8) ^ XXH_readLE32(secret+12)) - seed;
        xxh_u64 const keyed_lo = (xxh_u64)combinedl ^ bitflipl;
        xxh_u64 const keyed_hi = (xxh_u64)combinedh ^ bitfliph;
        XXH128_hash_t h128;
        h128.low64  = XXH64_avalanche(keyed_lo);
        h128.high64 = XXH64_avalanche(keyed_hi);
        return h128;
    }
}

XXH_FORCE_INLINE XXH_PUREF XXH128_hash_t
XXH3_len_4to8_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
{
    XXH_ASSERT(input != NULL);
    XXH_ASSERT(secret != NULL);
    XXH_ASSERT(4 <= len && len <= 8);
    seed ^= (xxh_u64)XXH_swap32((xxh_u32)seed) << 32;
    {   xxh_u32 const input_lo = XXH_readLE32(input);
        xxh_u32 const input_hi = XXH_readLE32(input + len - 4);
        xxh_u64 const input_64 = input_lo + ((xxh_u64)input_hi << 32);
        xxh_u64 const bitflip = (XXH_readLE64(secret+16) ^ XXH_readLE64(secret+24)) + seed;
        xxh_u64 const keyed = input_64 ^ bitflip;

        /* Shift len to the left to ensure it is even, this avoids even multiplies. */
        XXH128_hash_t m128 = XXH_mult64to128(keyed, XXH_PRIME64_1 + (len << 2));

        m128.high64 += (m128.low64 << 1);
        m128.low64  ^= (m128.high64 >> 3);

        m128.low64   = XXH_xorshift64(m128.low64, 35);
        m128.low64  *= PRIME_MX2;
        m128.low64   = XXH_xorshift64(m128.low64, 28);
        m128.high64  = XXH3_avalanche(m128.high64);
        return m128;
    }
}

XXH_FORCE_INLINE XXH_PUREF XXH128_hash_t
XXH3_len_9to16_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
{
    XXH_ASSERT(input != NULL);
    XXH_ASSERT(secret != NULL);
    XXH_ASSERT(9 <= len && len <= 16);
    {   xxh_u64 const bitflipl = (XXH_readLE64(secret+32) ^ XXH_readLE64(secret+40)) - seed;
        xxh_u64 const bitfliph = (XXH_readLE64(secret+48) ^ XXH_readLE64(secret+56)) + seed;
        xxh_u64 const input_lo = XXH_readLE64(input);
        xxh_u64       input_hi = XXH_readLE64(input + len - 8);
        XXH128_hash_t m128 = XXH_mult64to128(input_lo ^ input_hi ^ bitflipl, XXH_PRIME64_1);
        /*
         * Put len in the middle of m128 to ensure that the length gets mixed to
         * both the low and high bits in the 128x64 multiply below.
         */
        m128.low64 += (xxh_u64)(len - 1) << 54;
        input_hi   ^= bitfliph;
        /*
         * Add the high 32 bits of input_hi to the high 32 bits of m128, then
         * add the long product of the low 32 bits of input_hi and XXH_PRIME32_2 to
         * the high 64 bits of m128.
         *
         * The best approach to this operation is different on 32-bit and 64-bit.
         */
        if (sizeof(void *) < sizeof(xxh_u64)) { /* 32-bit */
            /*
             * 32-bit optimized version, which is more readable.
             *
             * On 32-bit, it removes an ADC and delays a dependency between the two
             * halves of m128.high64, but it generates an extra mask on 64-bit.
             */
            m128.high64 += (input_hi & 0xFFFFFFFF00000000ULL) + XXH_mult32to64((xxh_u32)input_hi, XXH_PRIME32_2);
        } else {
            /*
             * 64-bit optimized (albeit more confusing) version.
             *
             * Uses some properties of addition and multiplication to remove the mask:
             *
             * Let:
             *    a = input_hi.lo = (input_hi & 0x00000000FFFFFFFF)
             *    b = input_hi.hi = (input_hi & 0xFFFFFFFF00000000)
             *    c = XXH_PRIME32_2
             *
             *    a + (b * c)
             * Inverse Property: x + y - x == y
             *    a + (b * (1 + c - 1))
             * Distributive Property: x * (y + z) == (x * y) + (x * z)
             *    a + (b * 1) + (b * (c - 1))
             * Identity Property: x * 1 == x
             *    a + b + (b * (c - 1))
             *
             * Substitute a, b, and c:
             *    input_hi.hi + input_hi.lo + ((xxh_u64)input_hi.lo * (XXH_PRIME32_2 - 1))
             *
             * Since input_hi.hi + input_hi.lo == input_hi, we get this:
             *    input_hi + ((xxh_u64)input_hi.lo * (XXH_PRIME32_2 - 1))
             */
            m128.high64 += input_hi + XXH_mult32to64((xxh_u32)input_hi, XXH_PRIME32_2 - 1);
        }
        /* m128 ^= XXH_swap64(m128 >> 64); */
        m128.low64  ^= XXH_swap64(m128.high64);

        {   /* 128x64 multiply: h128 = m128 * XXH_PRIME64_2; */
            XXH128_hash_t h128 = XXH_mult64to128(m128.low64, XXH_PRIME64_2);
            h128.high64 += m128.high64 * XXH_PRIME64_2;

            h128.low64   = XXH3_avalanche(h128.low64);
            h128.high64  = XXH3_avalanche(h128.high64);
            return h128;
    }   }
}

/*
 * Assumption: `secret` size is >= XXH3_SECRET_SIZE_MIN
 */
XXH_FORCE_INLINE XXH_PUREF XXH128_hash_t
XXH3_len_0to16_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
{
    XXH_ASSERT(len <= 16);
    {   if (len > 8) return XXH3_len_9to16_128b(input, len, secret, seed);
        if (len >= 4) return XXH3_len_4to8_128b(input, len, secret, seed);
        if (len) return XXH3_len_1to3_128b(input, len, secret, seed);
        {   XXH128_hash_t h128;
            xxh_u64 const bitflipl = XXH_readLE64(secret+64) ^ XXH_readLE64(secret+72);
            xxh_u64 const bitfliph = XXH_readLE64(secret+80) ^ XXH_readLE64(secret+88);
            h128.low64 = XXH64_avalanche(seed ^ bitflipl);
            h128.high64 = XXH64_avalanche( seed ^ bitfliph);
            return h128;
    }   }
}

/*
 * A bit slower than XXH3_mix16B, but handles multiply by zero better.
 */
XXH_FORCE_INLINE XXH128_hash_t
XXH128_mix32B(XXH128_hash_t acc, const xxh_u8* input_1, const xxh_u8* input_2,
              const xxh_u8* secret, XXH64_hash_t seed)
{
    acc.low64  += XXH3_mix16B (input_1, secret+0, seed);
    acc.low64  ^= XXH_readLE64(input_2) + XXH_readLE64(input_2 + 8);
    acc.high64 += XXH3_mix16B (input_2, secret+16, seed);
    acc.high64 ^= XXH_readLE64(input_1) + XXH_readLE64(input_1 + 8);
    return acc;
}


XXH_FORCE_INLINE XXH_PUREF XXH128_hash_t
XXH3_len_17to128_128b(const xxh_u8* XXH_RESTRICT input, size_t len,
                      const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
                      XXH64_hash_t seed)
{
    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); (void)secretSize;
    XXH_ASSERT(16 < len && len <= 128);

    {   XXH128_hash_t acc;
        acc.low64 = len * XXH_PRIME64_1;
        acc.high64 = 0;

#if XXH_SIZE_OPT >= 1
        {
            /* Smaller, but slightly slower. */
            unsigned int i = (unsigned int)(len - 1) / 32;
            do {
                acc = XXH128_mix32B(acc, input+16*i, input+len-16*(i+1), secret+32*i, seed);
            } while (i-- != 0);
        }
#else
        if (len > 32) {
            if (len > 64) {
                if (len > 96) {
                    acc = XXH128_mix32B(acc, input+48, input+len-64, secret+96, seed);
                }
                acc = XXH128_mix32B(acc, input+32, input+len-48, secret+64, seed);
            }
            acc = XXH128_mix32B(acc, input+16, input+len-32, secret+32, seed);
        }
        acc = XXH128_mix32B(acc, input, input+len-16, secret, seed);
#endif
        {   XXH128_hash_t h128;
            h128.low64  = acc.low64 + acc.high64;
            h128.high64 = (acc.low64    * XXH_PRIME64_1)
                        + (acc.high64   * XXH_PRIME64_4)
                        + ((len - seed) * XXH_PRIME64_2);
            h128.low64  = XXH3_avalanche(h128.low64);
            h128.high64 = (XXH64_hash_t)0 - XXH3_avalanche(h128.high64);
            return h128;
        }
    }
}

XXH_NO_INLINE XXH_PUREF XXH128_hash_t
XXH3_len_129to240_128b(const xxh_u8* XXH_RESTRICT input, size_t len,
                       const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
                       XXH64_hash_t seed)
{
    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); (void)secretSize;
    XXH_ASSERT(128 < len && len <= XXH3_MIDSIZE_MAX);

    {   XXH128_hash_t acc;
        unsigned i;
        acc.low64 = len * XXH_PRIME64_1;
        acc.high64 = 0;
        /*
         *  We set as `i` as offset + 32. We do this so that unchanged
         * `len` can be used as upper bound. This reaches a sweet spot
         * where both x86 and aarch64 get simple agen and good codegen
         * for the loop.
         */
        for (i = 32; i < 160; i += 32) {
            acc = XXH128_mix32B(acc,
                                input  + i - 32,
                                input  + i - 16,
                                secret + i - 32,
                                seed);
        }
        acc.low64 = XXH3_avalanche(acc.low64);
        acc.high64 = XXH3_avalanche(acc.high64);
        /*
         * NB: `i <= len` will duplicate the last 32-bytes if
         * len % 32 was zero. This is an unfortunate necessity to keep
         * the hash result stable.
         */
        for (i=160; i <= len; i += 32) {
            acc = XXH128_mix32B(acc,
                                input + i - 32,
                                input + i - 16,
                                secret + XXH3_MIDSIZE_STARTOFFSET + i - 160,
                                seed);
        }
        /* last bytes */
        acc = XXH128_mix32B(acc,
                            input + len - 16,
                            input + len - 32,
                            secret + XXH3_SECRET_SIZE_MIN - XXH3_MIDSIZE_LASTOFFSET - 16,
                            (XXH64_hash_t)0 - seed);

        {   XXH128_hash_t h128;
            h128.low64  = acc.low64 + acc.high64;
            h128.high64 = (acc.low64    * XXH_PRIME64_1)
                        + (acc.high64   * XXH_PRIME64_4)
                        + ((len - seed) * XXH_PRIME64_2);
            h128.low64  = XXH3_avalanche(h128.low64);
            h128.high64 = (XXH64_hash_t)0 - XXH3_avalanche(h128.high64);
            return h128;
        }
    }
}

XXH_FORCE_INLINE XXH128_hash_t
XXH3_hashLong_128b_internal(const void* XXH_RESTRICT input, size_t len,
                            const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
                            XXH3_f_accumulate f_acc,
                            XXH3_f_scrambleAcc f_scramble)
{
    XXH_ALIGN(XXH_ACC_ALIGN) xxh_u64 acc[XXH_ACC_NB] = XXH3_INIT_ACC;

    XXH3_hashLong_internal_loop(acc, (const xxh_u8*)input, len, secret, secretSize, f_acc, f_scramble);

    /* converge into final hash */
    XXH_STATIC_ASSERT(sizeof(acc) == 64);
    XXH_ASSERT(secretSize >= sizeof(acc) + XXH_SECRET_MERGEACCS_START);
    {   XXH128_hash_t h128;
        h128.low64  = XXH3_mergeAccs(acc,
                                     secret + XXH_SECRET_MERGEACCS_START,
                                     (xxh_u64)len * XXH_PRIME64_1);
        h128.high64 = XXH3_mergeAccs(acc,
                                     secret + secretSize
                                            - sizeof(acc) - XXH_SECRET_MERGEACCS_START,
                                     ~((xxh_u64)len * XXH_PRIME64_2));
        return h128;
    }
}

/*
 * It's important for performance that XXH3_hashLong() is not inlined.
 */
XXH_NO_INLINE XXH_PUREF XXH128_hash_t
XXH3_hashLong_128b_default(const void* XXH_RESTRICT input, size_t len,
                           XXH64_hash_t seed64,
                           const void* XXH_RESTRICT secret, size_t secretLen)
{
    (void)seed64; (void)secret; (void)secretLen;
    return XXH3_hashLong_128b_internal(input, len, XXH3_kSecret, sizeof(XXH3_kSecret),
                                       XXH3_accumulate, XXH3_scrambleAcc);
}

/*
 * It's important for performance to pass @p secretLen (when it's static)
 * to the compiler, so that it can properly optimize the vectorized loop.
 *
 * When the secret size is unknown, or on GCC 12 where the mix of NO_INLINE and FORCE_INLINE
 * breaks -Og, this is XXH_NO_INLINE.
 */
XXH3_WITH_SECRET_INLINE XXH128_hash_t
XXH3_hashLong_128b_withSecret(const void* XXH_RESTRICT input, size_t len,
                              XXH64_hash_t seed64,
                              const void* XXH_RESTRICT secret, size_t secretLen)
{
    (void)seed64;
    return XXH3_hashLong_128b_internal(input, len, (const xxh_u8*)secret, secretLen,
                                       XXH3_accumulate, XXH3_scrambleAcc);
}

XXH_FORCE_INLINE XXH128_hash_t
XXH3_hashLong_128b_withSeed_internal(const void* XXH_RESTRICT input, size_t len,
                                XXH64_hash_t seed64,
                                XXH3_f_accumulate f_acc,
                                XXH3_f_scrambleAcc f_scramble,
                                XXH3_f_initCustomSecret f_initSec)
{
    if (seed64 == 0)
        return XXH3_hashLong_128b_internal(input, len,
                                           XXH3_kSecret, sizeof(XXH3_kSecret),
                                           f_acc, f_scramble);
    {   XXH_ALIGN(XXH_SEC_ALIGN) xxh_u8 secret[XXH_SECRET_DEFAULT_SIZE];
        f_initSec(secret, seed64);
        return XXH3_hashLong_128b_internal(input, len, (const xxh_u8*)secret, sizeof(secret),
                                           f_acc, f_scramble);
    }
}

/*
 * It's important for performance that XXH3_hashLong is not inlined.
 */
XXH_NO_INLINE XXH128_hash_t
XXH3_hashLong_128b_withSeed(const void* input, size_t len,
                            XXH64_hash_t seed64, const void* XXH_RESTRICT secret, size_t secretLen)
{
    (void)secret; (void)secretLen;
    return XXH3_hashLong_128b_withSeed_internal(input, len, seed64,
                XXH3_accumulate, XXH3_scrambleAcc, XXH3_initCustomSecret);
}

typedef XXH128_hash_t (*XXH3_hashLong128_f)(const void* XXH_RESTRICT, size_t,
                                            XXH64_hash_t, const void* XXH_RESTRICT, size_t);

XXH_FORCE_INLINE XXH128_hash_t
XXH3_128bits_internal(const void* input, size_t len,
                      XXH64_hash_t seed64, const void* XXH_RESTRICT secret, size_t secretLen,
                      XXH3_hashLong128_f f_hl128)
{
    XXH_ASSERT(secretLen >= XXH3_SECRET_SIZE_MIN);
    /*
     * If an action is to be taken if `secret` conditions are not respected,
     * it should be done here.
     * For now, it's a contract pre-condition.
     * Adding a check and a branch here would cost performance at every hash.
     */
    if (len <= 16)
        return XXH3_len_0to16_128b((const xxh_u8*)input, len, (const xxh_u8*)secret, seed64);
    if (len <= 128)
        return XXH3_len_17to128_128b((const xxh_u8*)input, len, (const xxh_u8*)secret, secretLen, seed64);
    if (len <= XXH3_MIDSIZE_MAX)
        return XXH3_len_129to240_128b((const xxh_u8*)input, len, (const xxh_u8*)secret, secretLen, seed64);
    return f_hl128(input, len, seed64, secret, secretLen);
}


/* ===   Public XXH128 API   === */

/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH128_hash_t XXH3_128bits(XXH_NOESCAPE const void* input, size_t len)
{
    return XXH3_128bits_internal(input, len, 0,
                                 XXH3_kSecret, sizeof(XXH3_kSecret),
                                 XXH3_hashLong_128b_default);
}

/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH128_hash_t
XXH3_128bits_withSecret(XXH_NOESCAPE const void* input, size_t len, XXH_NOESCAPE const void* secret, size_t secretSize)
{
    return XXH3_128bits_internal(input, len, 0,
                                 (const xxh_u8*)secret, secretSize,
                                 XXH3_hashLong_128b_withSecret);
}

/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH128_hash_t
XXH3_128bits_withSeed(XXH_NOESCAPE const void* input, size_t len, XXH64_hash_t seed)
{
    return XXH3_128bits_internal(input, len, seed,
                                 XXH3_kSecret, sizeof(XXH3_kSecret),
                                 XXH3_hashLong_128b_withSeed);
}

/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH128_hash_t
XXH3_128bits_withSecretandSeed(XXH_NOESCAPE const void* input, size_t len, XXH_NOESCAPE const void* secret, size_t secretSize, XXH64_hash_t seed)
{
    if (len <= XXH3_MIDSIZE_MAX)
        return XXH3_128bits_internal(input, len, seed, XXH3_kSecret, sizeof(XXH3_kSecret), NULL);
    return XXH3_hashLong_128b_withSecret(input, len, seed, secret, secretSize);
}

/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH128_hash_t
XXH128(XXH_NOESCAPE const void* input, size_t len, XXH64_hash_t seed)
{
    return XXH3_128bits_withSeed(input, len, seed);
}


/* ===   XXH3 128-bit streaming   === */
#ifndef XXH_NO_STREAM
/*
 * All initialization and update functions are identical to 64-bit streaming variant.
 * The only difference is the finalization routine.
 */

/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH_errorcode
XXH3_128bits_reset(XXH_NOESCAPE XXH3_state_t* statePtr)
{
    return XXH3_64bits_reset(statePtr);
}

/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH_errorcode
XXH3_128bits_reset_withSecret(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* secret, size_t secretSize)
{
    return XXH3_64bits_reset_withSecret(statePtr, secret, secretSize);
}

/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH_errorcode
XXH3_128bits_reset_withSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH64_hash_t seed)
{
    return XXH3_64bits_reset_withSeed(statePtr, seed);
}

/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH_errorcode
XXH3_128bits_reset_withSecretandSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* secret, size_t secretSize, XXH64_hash_t seed)
{
    return XXH3_64bits_reset_withSecretandSeed(statePtr, secret, secretSize, seed);
}

/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH_errorcode
XXH3_128bits_update(XXH_NOESCAPE XXH3_state_t* state, XXH_NOESCAPE const void* input, size_t len)
{
    return XXH3_64bits_update(state, input, len);
}

/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_digest (XXH_NOESCAPE const XXH3_state_t* state)
{
    const unsigned char* const secret = (state->extSecret == NULL) ? state->customSecret : state->extSecret;
    if (state->totalLen > XXH3_MIDSIZE_MAX) {
        XXH_ALIGN(XXH_ACC_ALIGN) XXH64_hash_t acc[XXH_ACC_NB];
        XXH3_digest_long(acc, state, secret);
        XXH_ASSERT(state->secretLimit + XXH_STRIPE_LEN >= sizeof(acc) + XXH_SECRET_MERGEACCS_START);
        {   XXH128_hash_t h128;
            h128.low64  = XXH3_mergeAccs(acc,
                                         secret + XXH_SECRET_MERGEACCS_START,
                                         (xxh_u64)state->totalLen * XXH_PRIME64_1);
            h128.high64 = XXH3_mergeAccs(acc,
                                         secret + state->secretLimit + XXH_STRIPE_LEN
                                                - sizeof(acc) - XXH_SECRET_MERGEACCS_START,
                                         ~((xxh_u64)state->totalLen * XXH_PRIME64_2));
            return h128;
        }
    }
    /* len <= XXH3_MIDSIZE_MAX : short code */
    if (state->seed)
        return XXH3_128bits_withSeed(state->buffer, (size_t)state->totalLen, state->seed);
    return XXH3_128bits_withSecret(state->buffer, (size_t)(state->totalLen),
                                   secret, state->secretLimit + XXH_STRIPE_LEN);
}
#endif /* !XXH_NO_STREAM */
/* 128-bit utility functions */

#include <string.h>   /* memcmp, memcpy */

/* return : 1 is equal, 0 if different */
/*! @ingroup XXH3_family */
XXH_PUBLIC_API int XXH128_isEqual(XXH128_hash_t h1, XXH128_hash_t h2)
{
    /* note : XXH128_hash_t is compact, it has no padding byte */
    return !(memcmp(&h1, &h2, sizeof(h1)));
}

/* This prototype is compatible with stdlib's qsort().
 * @return : >0 if *h128_1  > *h128_2
 *           <0 if *h128_1  < *h128_2
 *           =0 if *h128_1 == *h128_2  */
/*! @ingroup XXH3_family */
XXH_PUBLIC_API int XXH128_cmp(XXH_NOESCAPE const void* h128_1, XXH_NOESCAPE const void* h128_2)
{
    XXH128_hash_t const h1 = *(const XXH128_hash_t*)h128_1;
    XXH128_hash_t const h2 = *(const XXH128_hash_t*)h128_2;
    int const hcmp = (h1.high64 > h2.high64) - (h2.high64 > h1.high64);
    /* note : bets that, in most cases, hash values are different */
    if (hcmp) return hcmp;
    return (h1.low64 > h2.low64) - (h2.low64 > h1.low64);
}


/*======   Canonical representation   ======*/
/*! @ingroup XXH3_family */
XXH_PUBLIC_API void
XXH128_canonicalFromHash(XXH_NOESCAPE XXH128_canonical_t* dst, XXH128_hash_t hash)
{
    XXH_STATIC_ASSERT(sizeof(XXH128_canonical_t) == sizeof(XXH128_hash_t));
    if (XXH_CPU_LITTLE_ENDIAN) {
        hash.high64 = XXH_swap64(hash.high64);
        hash.low64  = XXH_swap64(hash.low64);
    }
    XXH_memcpy(dst, &hash.high64, sizeof(hash.high64));
    XXH_memcpy((char*)dst + sizeof(hash.high64), &hash.low64, sizeof(hash.low64));
}

/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH128_hash_t
XXH128_hashFromCanonical(XXH_NOESCAPE const XXH128_canonical_t* src)
{
    XXH128_hash_t h;
    h.high64 = XXH_readBE64(src);
    h.low64  = XXH_readBE64(src->digest + 8);
    return h;
}



/* ==========================================
 * Secret generators
 * ==========================================
 */
#define XXH_MIN(x, y) (((x) > (y)) ? (y) : (x))

XXH_FORCE_INLINE void XXH3_combine16(void* dst, XXH128_hash_t h128)
{
    XXH_writeLE64( dst, XXH_readLE64(dst) ^ h128.low64 );
    XXH_writeLE64( (char*)dst+8, XXH_readLE64((char*)dst+8) ^ h128.high64 );
}

/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH_errorcode
XXH3_generateSecret(XXH_NOESCAPE void* secretBuffer, size_t secretSize, XXH_NOESCAPE const void* customSeed, size_t customSeedSize)
{
#if (XXH_DEBUGLEVEL >= 1)
    XXH_ASSERT(secretBuffer != NULL);
    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN);
#else
    /* production mode, assert() are disabled */
    if (secretBuffer == NULL) return XXH_ERROR;
    if (secretSize < XXH3_SECRET_SIZE_MIN) return XXH_ERROR;
#endif

    if (customSeedSize == 0) {
        customSeed = XXH3_kSecret;
        customSeedSize = XXH_SECRET_DEFAULT_SIZE;
    }
#if (XXH_DEBUGLEVEL >= 1)
    XXH_ASSERT(customSeed != NULL);
#else
    if (customSeed == NULL) return XXH_ERROR;
#endif

    /* Fill secretBuffer with a copy of customSeed - repeat as needed */
    {   size_t pos = 0;
        while (pos < secretSize) {
            size_t const toCopy = XXH_MIN((secretSize - pos), customSeedSize);
            memcpy((char*)secretBuffer + pos, customSeed, toCopy);
            pos += toCopy;
    }   }

    {   size_t const nbSeg16 = secretSize / 16;
        size_t n;
        XXH128_canonical_t scrambler;
        XXH128_canonicalFromHash(&scrambler, XXH128(customSeed, customSeedSize, 0));
        for (n=0; n<nbSeg16; n++) {
            XXH128_hash_t const h128 = XXH128(&scrambler, sizeof(scrambler), n);
            XXH3_combine16((char*)secretBuffer + n*16, h128);
        }
        /* last segment */
        XXH3_combine16((char*)secretBuffer + secretSize - 16, XXH128_hashFromCanonical(&scrambler));
    }
    return XXH_OK;
}

/*! @ingroup XXH3_family */
XXH_PUBLIC_API void
XXH3_generateSecret_fromSeed(XXH_NOESCAPE void* secretBuffer, XXH64_hash_t seed)
{
    XXH_ALIGN(XXH_SEC_ALIGN) xxh_u8 secret[XXH_SECRET_DEFAULT_SIZE];
    XXH3_initCustomSecret(secret, seed);
    XXH_ASSERT(secretBuffer != NULL);
    memcpy(secretBuffer, secret, XXH_SECRET_DEFAULT_SIZE);
}



/* Pop our optimization override from above */
#if XXH_VECTOR == XXH_AVX2 /* AVX2 */ \
  && defined(__GNUC__) && !defined(__clang__) /* GCC, not Clang */ \
  && defined(__OPTIMIZE__) && XXH_SIZE_OPT <= 0 /* respect -O0 and -Os */
#  pragma GCC pop_options
#endif

#endif  /* XXH_NO_LONG_LONG */

#endif  /* XXH_NO_XXH3 */

/*!
 * @}
 */
#endif  /* XXH_IMPLEMENTATION */


#if defined (__cplusplus)
} /* extern "C" */
#endif