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
// Copyright Mozilla Foundation. See the COPYRIGHT
// file at the top-level directory of this distribution.
//
// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
// https://www.apache.org/licenses/LICENSE-2.0> or the MIT license
// <LICENSE-MIT or https://opensource.org/licenses/MIT>, at your
// option. This file may not be copied, modified, or distributed
// except according to those terms.

// It's assumed that in due course Rust will have explicit SIMD but will not
// be good at run-time selection of SIMD vs. no-SIMD. In such a future,
// x86_64 will always use SSE2 and 32-bit x86 will use SSE2 when compiled with
// a Mozilla-shipped rustc. SIMD support and especially detection on ARM is a
// mess. Under the circumstances, it seems to make sense to optimize the ALU
// case for ARMv7 rather than x86. Annoyingly, I was unable to get useful
// numbers of the actual ARMv7 CPU I have access to, because (thermal?)
// throttling kept interfering. Since Raspberry Pi 3 (ARMv8 core but running
// ARMv7 code) produced reproducible performance numbers, that's the ARM
// computer that this code ended up being optimized for in the ALU case.
// Less popular CPU architectures simply get the approach that was chosen based
// on Raspberry Pi 3 measurements. The UTF-16 and UTF-8 ALU cases take
// different approaches based on benchmarking on Raspberry Pi 3.

#[cfg(all(
    feature = "simd-accel",
    any(
        target_feature = "sse2",
        all(target_endian = "little", target_arch = "aarch64"),
        all(target_endian = "little", target_feature = "neon")
    )
))]
use crate::simd_funcs::*;

cfg_if! {
    if #[cfg(feature = "simd-accel")] {
        #[allow(unused_imports)]
        use ::core::intrinsics::unlikely;
        #[allow(unused_imports)]
        use ::core::intrinsics::likely;
    } else {
        #[allow(dead_code)]
        #[inline(always)]
        fn unlikely(b: bool) -> bool {
            b
        }
        #[allow(dead_code)]
        #[inline(always)]
        fn likely(b: bool) -> bool {
            b
        }
    }
}

// Safety invariants for masks: data & mask = 0 for valid ASCII or basic latin utf-16

// `as` truncates, so works on 32-bit, too.
#[allow(dead_code)]
pub const ASCII_MASK: usize = 0x8080_8080_8080_8080u64 as usize;

// `as` truncates, so works on 32-bit, too.
#[allow(dead_code)]
pub const BASIC_LATIN_MASK: usize = 0xFF80_FF80_FF80_FF80u64 as usize;

#[allow(unused_macros)]
macro_rules! ascii_naive {
    ($name:ident, $src_unit:ty, $dst_unit:ty) => {
        /// Safety: src and dst must have len_unit elements and be aligned
        /// Safety-usable invariant: will return Some() when it fails
        /// to convert. The first value will be a u8 that is > 127.
        #[inline(always)]
        pub unsafe fn $name(
            src: *const $src_unit,
            dst: *mut $dst_unit,
            len: usize,
        ) -> Option<($src_unit, usize)> {
            // Yes, manually omitting the bound check here matters
            // a lot for perf.
            for i in 0..len {
                // Safety: len invariant used here
                let code_unit = *(src.add(i));
                // Safety: Upholds safety-usable invariant here
                if code_unit > 127 {
                    return Some((code_unit, i));
                }
                // Safety: len invariant used here
                *(dst.add(i)) = code_unit as $dst_unit;
            }
            return None;
        }
    };
}

#[allow(unused_macros)]
macro_rules! ascii_alu {
    ($name:ident,
     // safety invariant: src/dst MUST be u8
     $src_unit:ty,
     $dst_unit:ty,
     // Safety invariant: stride_fn must consume and produce two usizes, and return the index of the first non-ascii when it fails
     $stride_fn:ident) => {
        /// Safety: src and dst must have len elements, src is valid for read, dst is valid for
        /// write
        /// Safety-usable invariant: will return Some() when it fails
        /// to convert. The first value will be a u8 that is > 127.
        #[cfg_attr(feature = "cargo-clippy", allow(never_loop, cast_ptr_alignment))]
        #[inline(always)]
        pub unsafe fn $name(
            src: *const $src_unit,
            dst: *mut $dst_unit,
            len: usize,
        ) -> Option<($src_unit, usize)> {
            let mut offset = 0usize;
            // This loop is only broken out of as a `goto` forward
            loop {
                // Safety: until_alignment becomes the number of bytes we need to munch until we are aligned to usize
                let mut until_alignment = {
                    // Check if the other unit aligns if we move the narrower unit
                    // to alignment.
                    //               if ::core::mem::size_of::<$src_unit>() == ::core::mem::size_of::<$dst_unit>() {
                    // ascii_to_ascii
                    let src_alignment = (src as usize) & ALU_ALIGNMENT_MASK;
                    let dst_alignment = (dst as usize) & ALU_ALIGNMENT_MASK;
                    if src_alignment != dst_alignment {
                        // Safety: bails early and ends up in the naïve branch where usize-alignment doesn't matter
                        break;
                    }
                    (ALU_ALIGNMENT - src_alignment) & ALU_ALIGNMENT_MASK
                    //               } else if ::core::mem::size_of::<$src_unit>() < ::core::mem::size_of::<$dst_unit>() {
                    // ascii_to_basic_latin
                    //                   let src_until_alignment = (ALIGNMENT - ((src as usize) & ALIGNMENT_MASK)) & ALIGNMENT_MASK;
                    //                   if (dst.add(src_until_alignment) as usize) & ALIGNMENT_MASK != 0 {
                    //                       break;
                    //                   }
                    //                   src_until_alignment
                    //               } else {
                    // basic_latin_to_ascii
                    //                   let dst_until_alignment = (ALIGNMENT - ((dst as usize) & ALIGNMENT_MASK)) & ALIGNMENT_MASK;
                    //                   if (src.add(dst_until_alignment) as usize) & ALIGNMENT_MASK != 0 {
                    //                       break;
                    //                   }
                    //                   dst_until_alignment
                    //               }
                };
                if until_alignment + ALU_STRIDE_SIZE <= len {
                    // Moving pointers to alignment seems to be a pessimization on
                    // x86_64 for operations that have UTF-16 as the internal
                    // Unicode representation. However, since it seems to be a win
                    // on ARM (tested ARMv7 code running on ARMv8 [rpi3]), except
                    // mixed results when encoding from UTF-16 and since x86 and
                    // x86_64 should be using SSE2 in due course, keeping the move
                    // to alignment here. It would be good to test on more ARM CPUs
                    // and on real MIPS and POWER hardware.
                    //
                    // Safety: This is the naïve code once again, for `until_alignment` bytes
                    while until_alignment != 0 {
                        let code_unit = *(src.add(offset));
                        if code_unit > 127 {
                            // Safety: Upholds safety-usable invariant here
                            return Some((code_unit, offset));
                        }
                        *(dst.add(offset)) = code_unit as $dst_unit;
                        // Safety: offset is the number of bytes copied so far
                        offset += 1;
                        until_alignment -= 1;
                    }
                    let len_minus_stride = len - ALU_STRIDE_SIZE;
                    loop {
                        // Safety: num_ascii is known to be a byte index of a non-ascii byte due to stride_fn's invariant
                        if let Some(num_ascii) = $stride_fn(
                            // Safety: These are known to be valid and aligned since we have at
                            // least ALU_STRIDE_SIZE data in these buffers, and offset is the
                            // number of elements copied so far, which according to the
                            // until_alignment calculation above will cause both src and dst to be
                            // aligned to usize after this add
                            src.add(offset) as *const usize,
                            dst.add(offset) as *mut usize,
                        ) {
                            offset += num_ascii;
                            // Safety: Upholds safety-usable invariant here by indexing into non-ascii byte
                            return Some((*(src.add(offset)), offset));
                        }
                        // Safety: offset continues to be the number of bytes copied so far, and
                        // maintains usize alignment for the next loop iteration
                        offset += ALU_STRIDE_SIZE;
                        // Safety: This is `offset > len - stride. This loop will continue as long as
                        // `offset <= len - stride`, which means there are `stride` bytes to still be read.
                        if offset > len_minus_stride {
                            break;
                        }
                    }
                }
                break;
            }

            // Safety: This is the naïve code, same as ascii_naive, and has no requirements
            // other than src/dst being valid for the the right lens
            while offset < len {
                // Safety: len invariant used here
                let code_unit = *(src.add(offset));
                if code_unit > 127 {
                    // Safety: Upholds safety-usable invariant here
                    return Some((code_unit, offset));
                }
                // Safety: len invariant used here
                *(dst.add(offset)) = code_unit as $dst_unit;
                offset += 1;
            }
            None
        }
    };
}

#[allow(unused_macros)]
macro_rules! basic_latin_alu {
    ($name:ident,
    // safety invariant: use u8 for src/dest for ascii, and u16 for basic_latin
     $src_unit:ty,
     $dst_unit:ty,
    // safety invariant: stride function must munch ALU_STRIDE_SIZE*size(src_unit) bytes off of src and
    // write ALU_STRIDE_SIZE*size(dst_unit) bytes to dst
     $stride_fn:ident) => {
        /// Safety: src and dst must have len elements, src is valid for read, dst is valid for
        /// write
        /// Safety-usable invariant: will return Some() when it fails
        /// to convert. The first value will be a u8 that is > 127.
        #[cfg_attr(
            feature = "cargo-clippy",
            allow(never_loop, cast_ptr_alignment, cast_lossless)
        )]
        #[inline(always)]
        pub unsafe fn $name(
            src: *const $src_unit,
            dst: *mut $dst_unit,
            len: usize,
        ) -> Option<($src_unit, usize)> {
            let mut offset = 0usize;
            // This loop is only broken out of as a `goto` forward
            loop {
                // Safety: until_alignment becomes the number of bytes we need to munch from src/dest until we are aligned to usize
                // We ensure basic-latin has the same alignment as ascii, starting with ascii since it is smaller.
                let mut until_alignment = {
                    // Check if the other unit aligns if we move the narrower unit
                    // to alignment.
                    //               if ::core::mem::size_of::<$src_unit>() == ::core::mem::size_of::<$dst_unit>() {
                    // ascii_to_ascii
                    //                   let src_alignment = (src as usize) & ALIGNMENT_MASK;
                    //                   let dst_alignment = (dst as usize) & ALIGNMENT_MASK;
                    //                   if src_alignment != dst_alignment {
                    //                       break;
                    //                   }
                    //                   (ALIGNMENT - src_alignment) & ALIGNMENT_MASK
                    //               } else
                    if ::core::mem::size_of::<$src_unit>() < ::core::mem::size_of::<$dst_unit>() {
                        // ascii_to_basic_latin
                        let src_until_alignment = (ALU_ALIGNMENT
                            - ((src as usize) & ALU_ALIGNMENT_MASK))
                            & ALU_ALIGNMENT_MASK;
                        if (dst.wrapping_add(src_until_alignment) as usize) & ALU_ALIGNMENT_MASK
                            != 0
                        {
                            break;
                        }
                        src_until_alignment
                    } else {
                        // basic_latin_to_ascii
                        let dst_until_alignment = (ALU_ALIGNMENT
                            - ((dst as usize) & ALU_ALIGNMENT_MASK))
                            & ALU_ALIGNMENT_MASK;
                        if (src.wrapping_add(dst_until_alignment) as usize) & ALU_ALIGNMENT_MASK
                            != 0
                        {
                            break;
                        }
                        dst_until_alignment
                    }
                };
                if until_alignment + ALU_STRIDE_SIZE <= len {
                    // Moving pointers to alignment seems to be a pessimization on
                    // x86_64 for operations that have UTF-16 as the internal
                    // Unicode representation. However, since it seems to be a win
                    // on ARM (tested ARMv7 code running on ARMv8 [rpi3]), except
                    // mixed results when encoding from UTF-16 and since x86 and
                    // x86_64 should be using SSE2 in due course, keeping the move
                    // to alignment here. It would be good to test on more ARM CPUs
                    // and on real MIPS and POWER hardware.
                    //
                    // Safety: This is the naïve code once again, for `until_alignment` bytes
                    while until_alignment != 0 {
                        let code_unit = *(src.add(offset));
                        if code_unit > 127 {
                            // Safety: Upholds safety-usable invariant here
                            return Some((code_unit, offset));
                        }
                        *(dst.add(offset)) = code_unit as $dst_unit;
                        // Safety: offset is the number of bytes copied so far
                        offset += 1;
                        until_alignment -= 1;
                    }
                    let len_minus_stride = len - ALU_STRIDE_SIZE;
                    loop {
                        if !$stride_fn(
                            // Safety: These are known to be valid and aligned since we have at
                            // least ALU_STRIDE_SIZE data in these buffers, and offset is the
                            // number of elements copied so far, which according to the
                            // until_alignment calculation above will cause both src and dst to be
                            // aligned to usize after this add
                            src.add(offset) as *const usize,
                            dst.add(offset) as *mut usize,
                        ) {
                            break;
                        }
                        // Safety: offset continues to be the number of bytes copied so far, and
                        // maintains usize alignment for the next loop iteration
                        offset += ALU_STRIDE_SIZE;
                        // Safety: This is `offset > len - stride. This loop will continue as long as
                        // `offset <= len - stride`, which means there are `stride` bytes to still be read.
                        if offset > len_minus_stride {
                            break;
                        }
                    }
                }
                break;
            }
            // Safety: This is the naïve code once again, for leftover bytes
            while offset < len {
                // Safety: len invariant used here
                let code_unit = *(src.add(offset));
                if code_unit > 127 {
                    // Safety: Upholds safety-usable invariant here
                    return Some((code_unit, offset));
                }
                // Safety: len invariant used here
                *(dst.add(offset)) = code_unit as $dst_unit;
                offset += 1;
            }
            None
        }
    };
}

#[allow(unused_macros)]
macro_rules! latin1_alu {
    // safety invariant: stride function must munch ALU_STRIDE_SIZE*size(src_unit) bytes off of src and
    // write ALU_STRIDE_SIZE*size(dst_unit) bytes to dst
    ($name:ident, $src_unit:ty, $dst_unit:ty, $stride_fn:ident) => {
        /// Safety: src and dst must have len elements, src is valid for read, dst is valid for
        /// write
        #[cfg_attr(
            feature = "cargo-clippy",
            allow(never_loop, cast_ptr_alignment, cast_lossless)
        )]
        #[inline(always)]
        pub unsafe fn $name(src: *const $src_unit, dst: *mut $dst_unit, len: usize) {
            let mut offset = 0usize;
            // This loop is only broken out of as a `goto` forward
            loop {
                // Safety: until_alignment becomes the number of bytes we need to munch from src/dest until we are aligned to usize
                // We ensure the UTF-16 side has the same alignment as the Latin-1 side, starting with Latin-1 since it is smaller.
                let mut until_alignment = {
                    if ::core::mem::size_of::<$src_unit>() < ::core::mem::size_of::<$dst_unit>() {
                        // unpack
                        let src_until_alignment = (ALU_ALIGNMENT
                            - ((src as usize) & ALU_ALIGNMENT_MASK))
                            & ALU_ALIGNMENT_MASK;
                        if (dst.wrapping_add(src_until_alignment) as usize) & ALU_ALIGNMENT_MASK
                            != 0
                        {
                            break;
                        }
                        src_until_alignment
                    } else {
                        // pack
                        let dst_until_alignment = (ALU_ALIGNMENT
                            - ((dst as usize) & ALU_ALIGNMENT_MASK))
                            & ALU_ALIGNMENT_MASK;
                        if (src.wrapping_add(dst_until_alignment) as usize) & ALU_ALIGNMENT_MASK
                            != 0
                        {
                            break;
                        }
                        dst_until_alignment
                    }
                };
                if until_alignment + ALU_STRIDE_SIZE <= len {
                    // Safety: This is the naïve code once again, for `until_alignment` bytes
                    while until_alignment != 0 {
                        let code_unit = *(src.add(offset));
                        *(dst.add(offset)) = code_unit as $dst_unit;
                        // Safety: offset is the number of bytes copied so far
                        offset += 1;
                        until_alignment -= 1;
                    }
                    let len_minus_stride = len - ALU_STRIDE_SIZE;
                    loop {
                        $stride_fn(
                            // Safety: These are known to be valid and aligned since we have at
                            // least ALU_STRIDE_SIZE data in these buffers, and offset is the
                            // number of elements copied so far, which according to the
                            // until_alignment calculation above will cause both src and dst to be
                            // aligned to usize after this add
                            src.add(offset) as *const usize,
                            dst.add(offset) as *mut usize,
                        );
                        // Safety: offset continues to be the number of bytes copied so far, and
                        // maintains usize alignment for the next loop iteration
                        offset += ALU_STRIDE_SIZE;
                        // Safety: This is `offset > len - stride. This loop will continue as long as
                        // `offset <= len - stride`, which means there are `stride` bytes to still be read.
                        if offset > len_minus_stride {
                            break;
                        }
                    }
                }
                break;
            }
            // Safety: This is the naïve code once again, for leftover bytes
            while offset < len {
                // Safety: len invariant used here
                let code_unit = *(src.add(offset));
                *(dst.add(offset)) = code_unit as $dst_unit;
                offset += 1;
            }
        }
    };
}

#[allow(unused_macros)]
macro_rules! ascii_simd_check_align {
    (
        $name:ident,
        $src_unit:ty,
        $dst_unit:ty,
        // Safety: This function must require aligned src/dest that are valid for reading/writing SIMD_STRIDE_SIZE src_unit/dst_unit
        $stride_both_aligned:ident,
        // Safety: This function must require aligned/unaligned src/dest that are valid for reading/writing SIMD_STRIDE_SIZE src_unit/dst_unit
        $stride_src_aligned:ident,
        // Safety: This function must require unaligned/aligned src/dest that are valid for reading/writing SIMD_STRIDE_SIZE src_unit/dst_unit
        $stride_dst_aligned:ident,
        // Safety: This function must require unaligned src/dest that are valid for reading/writing SIMD_STRIDE_SIZE src_unit/dst_unit
        $stride_neither_aligned:ident
    ) => {
        /// Safety: src/dst must be valid for reads/writes of `len` elements of their units.
        ///
        /// Safety-usable invariant: will return Some() when it encounters non-ASCII, with the first element in the Some being
        /// guaranteed to be non-ASCII (> 127), and the second being the offset where it is found
        #[inline(always)]
        pub unsafe fn $name(
            src: *const $src_unit,
            dst: *mut $dst_unit,
            len: usize,
        ) -> Option<($src_unit, usize)> {
            let mut offset = 0usize;
            // Safety: if this check succeeds we're valid for reading/writing at least `SIMD_STRIDE_SIZE` elements.
            if SIMD_STRIDE_SIZE <= len {
                let len_minus_stride = len - SIMD_STRIDE_SIZE;
                // XXX Should we first process one stride unconditionally as unaligned to
                // avoid the cost of the branchiness below if the first stride fails anyway?
                // XXX Should we just use unaligned SSE2 access unconditionally? It seems that
                // on Haswell, it would make sense to just use unaligned and not bother
                // checking. Need to benchmark older architectures before deciding.
                let dst_masked = (dst as usize) & SIMD_ALIGNMENT_MASK;
                // Safety: checking whether src is aligned
                if ((src as usize) & SIMD_ALIGNMENT_MASK) == 0 {
                    // Safety: Checking whether dst is aligned
                    if dst_masked == 0 {
                        loop {
                            // Safety: We're valid to read/write SIMD_STRIDE_SIZE elements and have the appropriate alignments
                            if !$stride_both_aligned(src.add(offset), dst.add(offset)) {
                                break;
                            }
                            offset += SIMD_STRIDE_SIZE;
                            // Safety: This is `offset > len - SIMD_STRIDE_SIZE` which means we always have at least `SIMD_STRIDE_SIZE` elements to munch next time.
                            if offset > len_minus_stride {
                                break;
                            }
                        }
                    } else {
                        loop {
                            // Safety: We're valid to read/write SIMD_STRIDE_SIZE elements and have the appropriate alignments
                            if !$stride_src_aligned(src.add(offset), dst.add(offset)) {
                                break;
                            }
                            offset += SIMD_STRIDE_SIZE;
                            // Safety: This is `offset > len - SIMD_STRIDE_SIZE` which means we always have at least `SIMD_STRIDE_SIZE` elements to munch next time.
                            if offset > len_minus_stride {
                                break;
                            }
                        }
                    }
                } else {
                    if dst_masked == 0 {
                        loop {
                            // Safety: We're valid to read/write SIMD_STRIDE_SIZE elements and have the appropriate alignments
                            if !$stride_dst_aligned(src.add(offset), dst.add(offset)) {
                                break;
                            }
                            offset += SIMD_STRIDE_SIZE;
                            // Safety: This is `offset > len - SIMD_STRIDE_SIZE` which means we always have at least `SIMD_STRIDE_SIZE` elements to munch next time.
                            if offset > len_minus_stride {
                                break;
                            }
                        }
                    } else {
                        loop {
                            // Safety: We're valid to read/write SIMD_STRIDE_SIZE elements and have the appropriate alignments
                            if !$stride_neither_aligned(src.add(offset), dst.add(offset)) {
                                break;
                            }
                            offset += SIMD_STRIDE_SIZE;
                            // Safety: This is `offset > len - SIMD_STRIDE_SIZE` which means we always have at least `SIMD_STRIDE_SIZE` elements to munch next time.
                            if offset > len_minus_stride {
                                break;
                            }
                        }
                    }
                }
            }
            while offset < len {
                // Safety: uses len invariant here and below
                let code_unit = *(src.add(offset));
                if code_unit > 127 {
                    // Safety: upholds safety-usable invariant
                    return Some((code_unit, offset));
                }
                *(dst.add(offset)) = code_unit as $dst_unit;
                offset += 1;
            }
            None
        }
    };
}

#[allow(unused_macros)]
macro_rules! ascii_simd_check_align_unrolled {
    (
        $name:ident,
        $src_unit:ty,
        $dst_unit:ty,
        // Safety: This function must require aligned src/dest that are valid for reading/writing SIMD_STRIDE_SIZE src_unit/dst_unit
        $stride_both_aligned:ident,
        // Safety: This function must require aligned/unaligned src/dest that are valid for reading/writing SIMD_STRIDE_SIZE src_unit/dst_unit
        $stride_src_aligned:ident,
        // Safety: This function must require unaligned src/dest that are valid for reading/writing SIMD_STRIDE_SIZE src_unit/dst_unit
        $stride_neither_aligned:ident,
        // Safety: This function must require aligned src/dest that are valid for reading/writing 2*SIMD_STRIDE_SIZE src_unit/dst_unit
        $double_stride_both_aligned:ident,
        // Safety: This function must require aligned/unaligned src/dest that are valid for reading/writing 2*SIMD_STRIDE_SIZE src_unit/dst_unit
        $double_stride_src_aligned:ident
    ) => {
        /// Safety: src/dst must be valid for reads/writes of `len` elements of their units.
        ///
        /// Safety-usable invariant: will return Some() when it encounters non-ASCII, with the first element in the Some being
        /// guaranteed to be non-ASCII (> 127), and the second being the offset where it is found        #[inline(always)]
        pub unsafe fn $name(
            src: *const $src_unit,
            dst: *mut $dst_unit,
            len: usize,
        ) -> Option<($src_unit, usize)> {
            let unit_size = ::core::mem::size_of::<$src_unit>();
            let mut offset = 0usize;
            // This loop is only broken out of as a goto forward without
            // actually looping
            'outer: loop {
                // Safety: if this check succeeds we're valid for reading/writing at least `SIMD_STRIDE_SIZE` elements.
                if SIMD_STRIDE_SIZE <= len {
                    // First, process one unaligned
                    // Safety: this is safe to call since we're valid for this read/write
                    if !$stride_neither_aligned(src, dst) {
                        break 'outer;
                    }
                    offset = SIMD_STRIDE_SIZE;

                    // We have now seen 16 ASCII bytes. Let's guess that
                    // there will be enough more to justify more expense
                    // in the case of non-ASCII.
                    // Use aligned reads for the sake of old microachitectures.
                    //
                    // Safety: this correctly calculates the number of src_units that need to be read before the remaining list is aligned.
                    // This is less that SIMD_ALIGNMENT, which is also SIMD_STRIDE_SIZE (as documented)
                    let until_alignment = ((SIMD_ALIGNMENT
                        - ((src.add(offset) as usize) & SIMD_ALIGNMENT_MASK))
                        & SIMD_ALIGNMENT_MASK)
                        / unit_size;
                    // Safety: This addition won't overflow, because even in the 32-bit PAE case the
                    // address space holds enough code that the slice length can't be that
                    // close to address space size.
                    // offset now equals SIMD_STRIDE_SIZE, hence times 3 below.
                    //
                    // Safety: if this check succeeds we're valid for reading/writing at least `2 * SIMD_STRIDE_SIZE` elements plus `until_alignment`.
                    // The extra SIMD_STRIDE_SIZE in the condition is because `offset` is already `SIMD_STRIDE_SIZE`.
                    if until_alignment + (SIMD_STRIDE_SIZE * 3) <= len {
                        if until_alignment != 0 {
                            // Safety: this is safe to call since we're valid for this read/write (and more), and don't care about alignment
                            // This will copy over bytes that get decoded twice since it's not incrementing `offset` by SIMD_STRIDE_SIZE. This is fine.
                            if !$stride_neither_aligned(src.add(offset), dst.add(offset)) {
                                break;
                            }
                            offset += until_alignment;
                        }
                        // Safety: At this point we're valid for reading/writing 2*SIMD_STRIDE_SIZE elements
                        // Safety: Now `offset` is aligned for `src`
                        let len_minus_stride_times_two = len - (SIMD_STRIDE_SIZE * 2);
                        // Safety: This is whether dst is aligned
                        let dst_masked = (dst.add(offset) as usize) & SIMD_ALIGNMENT_MASK;
                        if dst_masked == 0 {
                            loop {
                                // Safety: both are aligned, we can call the aligned function. We're valid for reading/writing double stride from the initial condition
                                // and the loop break condition below
                                if let Some(advance) =
                                    $double_stride_both_aligned(src.add(offset), dst.add(offset))
                                {
                                    offset += advance;
                                    let code_unit = *(src.add(offset));
                                    // Safety: uses safety-usable invariant on ascii_to_ascii_simd_double_stride to return
                                    // guaranteed non-ascii
                                    return Some((code_unit, offset));
                                }
                                offset += SIMD_STRIDE_SIZE * 2;
                                // Safety: This is `offset > len - 2 * SIMD_STRIDE_SIZE` which means we always have at least `2 * SIMD_STRIDE_SIZE` elements to munch next time.
                                if offset > len_minus_stride_times_two {
                                    break;
                                }
                            }
                            // Safety: We're valid for reading/writing one more, and can still assume alignment
                            if offset + SIMD_STRIDE_SIZE <= len {
                                if !$stride_both_aligned(src.add(offset), dst.add(offset)) {
                                    break 'outer;
                                }
                                offset += SIMD_STRIDE_SIZE;
                            }
                        } else {
                            loop {
                                // Safety: only src is aligned here. We're valid for reading/writing double stride from the initial condition
                                // and the loop break condition below
                                if let Some(advance) =
                                    $double_stride_src_aligned(src.add(offset), dst.add(offset))
                                {
                                    offset += advance;
                                    let code_unit = *(src.add(offset));
                                    // Safety: uses safety-usable invariant on ascii_to_ascii_simd_double_stride to return
                                    // guaranteed non-ascii
                                    return Some((code_unit, offset));
                                }
                                offset += SIMD_STRIDE_SIZE * 2;
                                // Safety: This is `offset > len - 2 * SIMD_STRIDE_SIZE` which means we always have at least `2 * SIMD_STRIDE_SIZE` elements to munch next time.

                                if offset > len_minus_stride_times_two {
                                    break;
                                }
                            }
                            // Safety: We're valid for reading/writing one more, and can still assume alignment
                            if offset + SIMD_STRIDE_SIZE <= len {
                                if !$stride_src_aligned(src.add(offset), dst.add(offset)) {
                                    break 'outer;
                                }
                                offset += SIMD_STRIDE_SIZE;
                            }
                        }
                    } else {
                        // At most two iterations, so unroll
                        if offset + SIMD_STRIDE_SIZE <= len {
                            // Safety: The check above ensures we're allowed to read/write this, and we don't use alignment
                            if !$stride_neither_aligned(src.add(offset), dst.add(offset)) {
                                break;
                            }
                            offset += SIMD_STRIDE_SIZE;
                            if offset + SIMD_STRIDE_SIZE <= len {
                                // Safety: The check above ensures we're allowed to read/write this, and we don't use alignment
                                if !$stride_neither_aligned(src.add(offset), dst.add(offset)) {
                                    break;
                                }
                                offset += SIMD_STRIDE_SIZE;
                            }
                        }
                    }
                }
                break 'outer;
            }
            while offset < len {
                // Safety: relies straightforwardly on the `len` invariant
                let code_unit = *(src.add(offset));
                if code_unit > 127 {
                    // Safety-usable invariant upheld here
                    return Some((code_unit, offset));
                }
                *(dst.add(offset)) = code_unit as $dst_unit;
                offset += 1;
            }
            None
        }
    };
}

#[allow(unused_macros)]
macro_rules! latin1_simd_check_align {
    (
        $name:ident,
        $src_unit:ty,
        $dst_unit:ty,
        // Safety: This function must require aligned src/dest that are valid for reading/writing SIMD_STRIDE_SIZE src_unit/dst_unit
        $stride_both_aligned:ident,
        // Safety: This function must require aligned/unaligned src/dest that are valid for reading/writing SIMD_STRIDE_SIZE src_unit/dst_unit
        $stride_src_aligned:ident,
        // Safety: This function must require unaligned/aligned src/dest that are valid for reading/writing SIMD_STRIDE_SIZE src_unit/dst_unit
        $stride_dst_aligned:ident,
        // Safety: This function must require unaligned src/dest that are valid for reading/writing SIMD_STRIDE_SIZE src_unit/dst_unit
        $stride_neither_aligned:ident

    ) => {
        /// Safety: src/dst must be valid for reads/writes of `len` elements of their units.
        #[inline(always)]
        pub unsafe fn $name(src: *const $src_unit, dst: *mut $dst_unit, len: usize) {
            let mut offset = 0usize;
            // Safety: if this check succeeds we're valid for reading/writing at least `SIMD_STRIDE_SIZE` elements.
            if SIMD_STRIDE_SIZE <= len {
                let len_minus_stride = len - SIMD_STRIDE_SIZE;
                // Whether dst is aligned
                let dst_masked = (dst as usize) & SIMD_ALIGNMENT_MASK;
                // Whether src is aligned
                if ((src as usize) & SIMD_ALIGNMENT_MASK) == 0 {
                    if dst_masked == 0 {
                        loop {
                            // Safety: Both were aligned, we can use the aligned function
                            $stride_both_aligned(src.add(offset), dst.add(offset));
                            offset += SIMD_STRIDE_SIZE;
                            // Safety: This is `offset > len - SIMD_STRIDE_SIZE`, which means in the next iteration we're valid for
                            // reading/writing at least SIMD_STRIDE_SIZE elements.
                            if offset > len_minus_stride {
                                break;
                            }
                        }
                    } else {
                        loop {
                            // Safety: src was aligned, dst was not
                            $stride_src_aligned(src.add(offset), dst.add(offset));
                            offset += SIMD_STRIDE_SIZE;
                            // Safety: This is `offset > len - SIMD_STRIDE_SIZE`, which means in the next iteration we're valid for
                            // reading/writing at least SIMD_STRIDE_SIZE elements.
                            if offset > len_minus_stride {
                                break;
                            }
                        }
                    }
                } else {
                    if dst_masked == 0 {
                        loop {
                            // Safety: src was aligned, dst was not
                            $stride_dst_aligned(src.add(offset), dst.add(offset));
                            offset += SIMD_STRIDE_SIZE;
                            // Safety: This is `offset > len - SIMD_STRIDE_SIZE`, which means in the next iteration we're valid for
                            // reading/writing at least SIMD_STRIDE_SIZE elements.
                            if offset > len_minus_stride {
                                break;
                            }
                        }
                    } else {
                        loop {
                            // Safety: Neither were aligned
                            $stride_neither_aligned(src.add(offset), dst.add(offset));
                            offset += SIMD_STRIDE_SIZE;
                            // Safety: This is `offset > len - SIMD_STRIDE_SIZE`, which means in the next iteration we're valid for
                            // reading/writing at least SIMD_STRIDE_SIZE elements.
                            if offset > len_minus_stride {
                                break;
                            }
                        }
                    }
                }
            }
            while offset < len {
                // Safety: relies straightforwardly on the `len` invariant
                let code_unit = *(src.add(offset));
                *(dst.add(offset)) = code_unit as $dst_unit;
                offset += 1;
            }
        }
    };
}

#[allow(unused_macros)]
macro_rules! latin1_simd_check_align_unrolled {
    (
        $name:ident,
        $src_unit:ty,
        $dst_unit:ty,
        // Safety: This function must require aligned src/dest that are valid for reading/writing SIMD_STRIDE_SIZE src_unit/dst_unit
        $stride_both_aligned:ident,
        // Safety: This function must require aligned/unaligned src/dest that are valid for reading/writing SIMD_STRIDE_SIZE src_unit/dst_unit
        $stride_src_aligned:ident,
        // Safety: This function must require unaligned/aligned src/dest that are valid for reading/writing SIMD_STRIDE_SIZE src_unit/dst_unit
        $stride_dst_aligned:ident,
        // Safety: This function must require unaligned src/dest that are valid for reading/writing SIMD_STRIDE_SIZE src_unit/dst_unit
        $stride_neither_aligned:ident
    ) => {
        /// Safety: src/dst must be valid for reads/writes of `len` elements of their units.
        #[inline(always)]
        pub unsafe fn $name(src: *const $src_unit, dst: *mut $dst_unit, len: usize) {
            let unit_size = ::core::mem::size_of::<$src_unit>();
            let mut offset = 0usize;
            // Safety: if this check succeeds we're valid for reading/writing at least `SIMD_STRIDE_SIZE` elements.
            if SIMD_STRIDE_SIZE <= len {
                // Safety: this correctly calculates the number of src_units that need to be read before the remaining list is aligned.
                // This is by definition less than SIMD_STRIDE_SIZE.
                let mut until_alignment = ((SIMD_STRIDE_SIZE
                    - ((src as usize) & SIMD_ALIGNMENT_MASK))
                    & SIMD_ALIGNMENT_MASK)
                    / unit_size;
                while until_alignment != 0 {
                    // Safety: This is a straightforward copy, since until_alignment is < SIMD_STRIDE_SIZE < len, this is in-bounds
                    *(dst.add(offset)) = *(src.add(offset)) as $dst_unit;
                    offset += 1;
                    until_alignment -= 1;
                }
                // Safety: here offset will be `until_alignment`, i.e. enough to align `src`.
                let len_minus_stride = len - SIMD_STRIDE_SIZE;
                // Safety: if this check succeeds we're valid for reading/writing at least `2 * SIMD_STRIDE_SIZE` elements.
                if offset + SIMD_STRIDE_SIZE * 2 <= len {
                    let len_minus_stride_times_two = len_minus_stride - SIMD_STRIDE_SIZE;
                    // Safety: at this point src is known to be aligned at offset, dst is not.
                    if (dst.add(offset) as usize) & SIMD_ALIGNMENT_MASK == 0 {
                        loop {
                            // Safety: We checked alignment of dst above, we can use the alignment functions. We're allowed to read/write 2*SIMD_STRIDE_SIZE elements, which we do.
                            $stride_both_aligned(src.add(offset), dst.add(offset));
                            offset += SIMD_STRIDE_SIZE;
                            $stride_both_aligned(src.add(offset), dst.add(offset));
                            offset += SIMD_STRIDE_SIZE;
                            // Safety: This is `offset > len - 2 * SIMD_STRIDE_SIZE` which means we always have at least `2 * SIMD_STRIDE_SIZE` elements to munch next time.
                            if offset > len_minus_stride_times_two {
                                break;
                            }
                        }
                    } else {
                        loop {
                            // Safety: we ensured alignment of src already.
                            $stride_src_aligned(src.add(offset), dst.add(offset));
                            offset += SIMD_STRIDE_SIZE;
                            $stride_src_aligned(src.add(offset), dst.add(offset));
                            offset += SIMD_STRIDE_SIZE;
                            // Safety: This is `offset > len - 2 * SIMD_STRIDE_SIZE` which means we always have at least `2 * SIMD_STRIDE_SIZE` elements to munch next time.
                            if offset > len_minus_stride_times_two {
                                break;
                            }
                        }
                    }
                }
                // Safety: This is `offset > len - SIMD_STRIDE_SIZE` which means we are valid to munch SIMD_STRIDE_SIZE more elements, which we do
                if offset < len_minus_stride {
                    $stride_src_aligned(src.add(offset), dst.add(offset));
                    offset += SIMD_STRIDE_SIZE;
                }
            }
            while offset < len {
                // Safety: uses len invariant here and below
                let code_unit = *(src.add(offset));
                // On x86_64, this loop autovectorizes but in the pack
                // case there are instructions whose purpose is to make sure
                // each u16 in the vector is truncated before packing. However,
                // since we don't care about saturating behavior of SSE2 packing
                // when the input isn't Latin1, those instructions are useless.
                // Unfortunately, using the `assume` intrinsic to lie to the
                // optimizer doesn't make LLVM omit the trunctation that we
                // don't need. Possibly this loop could be manually optimized
                // to do the sort of thing that LLVM does but without the
                // ANDing the read vectors of u16 with a constant that discards
                // the high half of each u16. As far as I can tell, the
                // optimization assumes that doing a SIMD read past the end of
                // the array is OK.
                *(dst.add(offset)) = code_unit as $dst_unit;
                offset += 1;
            }
        }
    };
}

#[allow(unused_macros)]
macro_rules! ascii_simd_unalign {
    // Safety: stride_neither_aligned must be a function that requires src/dest be valid for unaligned reads/writes for SIMD_STRIDE_SIZE elements of type src_unit/dest_unit
    ($name:ident, $src_unit:ty, $dst_unit:ty, $stride_neither_aligned:ident) => {
        /// Safety: src and dst must be valid for reads/writes of len elements of type src_unit/dst_unit
        ///
        /// Safety-usable invariant: will return Some() when it encounters non-ASCII, with the first element in the Some being
        /// guaranteed to be non-ASCII (> 127), and the second being the offset where it is found
        #[inline(always)]
        pub unsafe fn $name(
            src: *const $src_unit,
            dst: *mut $dst_unit,
            len: usize,
        ) -> Option<($src_unit, usize)> {
            let mut offset = 0usize;
            // Safety: if this check succeeds we're valid for reading/writing at least `stride` elements.
            if SIMD_STRIDE_SIZE <= len {
                let len_minus_stride = len - SIMD_STRIDE_SIZE;
                loop {
                    // Safety: We know we're valid for `stride` reads/writes, so we can call this function. We don't need alignment.
                    if !$stride_neither_aligned(src.add(offset), dst.add(offset)) {
                        break;
                    }
                    offset += SIMD_STRIDE_SIZE;
                    // This is `offset > len - stride` which means we always have at least `stride` elements to munch next time.
                    if offset > len_minus_stride {
                        break;
                    }
                }
            }
            while offset < len {
                // Safety: Uses len invariant here and below
                let code_unit = *(src.add(offset));
                if code_unit > 127 {
                    // Safety-usable invariant upheld here
                    return Some((code_unit, offset));
                }
                *(dst.add(offset)) = code_unit as $dst_unit;
                offset += 1;
            }
            None
        }
    };
}

#[allow(unused_macros)]
macro_rules! latin1_simd_unalign {
    // Safety: stride_neither_aligned must be a function that requires src/dest be valid for unaligned reads/writes for SIMD_STRIDE_SIZE elements of type src_unit/dest_unit
    ($name:ident, $src_unit:ty, $dst_unit:ty, $stride_neither_aligned:ident) => {
        /// Safety: src and dst must be valid for unaligned reads/writes of len elements of type src_unit/dst_unit
        #[inline(always)]
        pub unsafe fn $name(src: *const $src_unit, dst: *mut $dst_unit, len: usize) {
            let mut offset = 0usize;
            // Safety: if this check succeeds we're valid for reading/writing at least `stride` elements.
            if SIMD_STRIDE_SIZE <= len {
                let len_minus_stride = len - SIMD_STRIDE_SIZE;
                loop {
                    // Safety: We know we're valid for `stride` reads/writes, so we can call this function. We don't need alignment.
                    $stride_neither_aligned(src.add(offset), dst.add(offset));
                    offset += SIMD_STRIDE_SIZE;
                    // This is `offset > len - stride` which means we always have at least `stride` elements to munch next time.
                    if offset > len_minus_stride {
                        break;
                    }
                }
            }
            while offset < len {
                // Safety: Uses len invariant here
                let code_unit = *(src.add(offset));
                *(dst.add(offset)) = code_unit as $dst_unit;
                offset += 1;
            }
        }
    };
}

#[allow(unused_macros)]
macro_rules! ascii_to_ascii_simd_stride {
    // Safety: load/store must be valid for 16 bytes of read/write, which may be unaligned. (candidates: `(load|store)(16|8)_(unaligned|aligned)` functions)
    ($name:ident, $load:ident, $store:ident) => {
        /// Safety: src and dst must be valid for 16 bytes of read/write according to
        /// the $load/$store fn, which may allow for unaligned reads/writes or require
        /// alignment to either 16x8 or u8x16.
        #[inline(always)]
        pub unsafe fn $name(src: *const u8, dst: *mut u8) -> bool {
            let simd = $load(src);
            if !simd_is_ascii(simd) {
                return false;
            }
            $store(dst, simd);
            true
        }
    };
}

#[allow(unused_macros)]
macro_rules! ascii_to_ascii_simd_double_stride {
    // Safety: store must be valid for 32 bytes of write, which may be unaligned (candidates: `store(8|16)_(aligned|unaligned)`)
    ($name:ident, $store:ident) => {
        /// Safety: src must be valid for 32 bytes of aligned u8x16 read
        /// dst must be valid for 32 bytes of unaligned write according to
        /// the $store fn, which may allow for unaligned writes or require
        /// alignment to either 16x8 or u8x16.
        ///
        /// Safety-usable invariant: Returns Some(index) if the element at `index` is invalid ASCII
        #[inline(always)]
        pub unsafe fn $name(src: *const u8, dst: *mut u8) -> Option<usize> {
            let first = load16_aligned(src);
            let second = load16_aligned(src.add(SIMD_STRIDE_SIZE));
            $store(dst, first);
            if unlikely(!simd_is_ascii(first | second)) {
                // Safety: mask_ascii produces a mask of all the high bits.
                let mask_first = mask_ascii(first);
                if mask_first != 0 {
                    // Safety: on little endian systems this will be the number of ascii bytes
                    // before the first non-ascii, i.e. valid for indexing src
                    // TODO SAFETY: What about big-endian systems?
                    return Some(mask_first.trailing_zeros() as usize);
                }
                $store(dst.add(SIMD_STRIDE_SIZE), second);
                let mask_second = mask_ascii(second);
                // Safety: on little endian systems this will be the number of ascii bytes
                // before the first non-ascii, i.e. valid for indexing src
                return Some(SIMD_STRIDE_SIZE + mask_second.trailing_zeros() as usize);
            }
            $store(dst.add(SIMD_STRIDE_SIZE), second);
            None
        }
    };
}

#[allow(unused_macros)]
macro_rules! ascii_to_basic_latin_simd_stride {
    // Safety: load/store must be valid for 16 bytes of read/write, which may be unaligned. (candidates: `(load|store)(16|8)_(unaligned|aligned)` functions)
    ($name:ident, $load:ident, $store:ident) => {
        /// Safety: src and dst must be valid for 16/32 bytes of read/write according to
        /// the $load/$store fn, which may allow for unaligned reads/writes or require
        /// alignment to either 16x8 or u8x16.
        #[inline(always)]
        pub unsafe fn $name(src: *const u8, dst: *mut u16) -> bool {
            let simd = $load(src);
            if !simd_is_ascii(simd) {
                return false;
            }
            let (first, second) = simd_unpack(simd);
            $store(dst, first);
            $store(dst.add(8), second);
            true
        }
    };
}

#[allow(unused_macros)]
macro_rules! ascii_to_basic_latin_simd_double_stride {
    // Safety: store must be valid for 16 bytes of write, which may be unaligned
    ($name:ident, $store:ident) => {
        /// Safety: src must be valid for 2*SIMD_STRIDE_SIZE bytes of aligned reads,
        /// aligned to either 16x8 or u8x16.
        /// dst must be valid for 2*SIMD_STRIDE_SIZE bytes of aligned or unaligned reads
        #[inline(always)]
        pub unsafe fn $name(src: *const u8, dst: *mut u16) -> Option<usize> {
            let first = load16_aligned(src);
            let second = load16_aligned(src.add(SIMD_STRIDE_SIZE));
            let (a, b) = simd_unpack(first);
            $store(dst, a);
            // Safety: divide by 2 since it's a u16 pointer
            $store(dst.add(SIMD_STRIDE_SIZE / 2), b);
            if unlikely(!simd_is_ascii(first | second)) {
                let mask_first = mask_ascii(first);
                if mask_first != 0 {
                    return Some(mask_first.trailing_zeros() as usize);
                }
                let (c, d) = simd_unpack(second);
                $store(dst.add(SIMD_STRIDE_SIZE), c);
                $store(dst.add(SIMD_STRIDE_SIZE + (SIMD_STRIDE_SIZE / 2)), d);
                let mask_second = mask_ascii(second);
                return Some(SIMD_STRIDE_SIZE + mask_second.trailing_zeros() as usize);
            }
            let (c, d) = simd_unpack(second);
            $store(dst.add(SIMD_STRIDE_SIZE), c);
            $store(dst.add(SIMD_STRIDE_SIZE + (SIMD_STRIDE_SIZE / 2)), d);
            None
        }
    };
}

#[allow(unused_macros)]
macro_rules! unpack_simd_stride {
    // Safety: load/store must be valid for 16 bytes of read/write, which may be unaligned. (candidates: `(load|store)(16|8)_(unaligned|aligned)` functions)
    ($name:ident, $load:ident, $store:ident) => {
        /// Safety: src and dst must be valid for 16 bytes of read/write according to
        /// the $load/$store fn, which may allow for unaligned reads/writes or require
        /// alignment to either 16x8 or u8x16.
        #[inline(always)]
        pub unsafe fn $name(src: *const u8, dst: *mut u16) {
            let simd = $load(src);
            let (first, second) = simd_unpack(simd);
            $store(dst, first);
            $store(dst.add(8), second);
        }
    };
}

#[allow(unused_macros)]
macro_rules! basic_latin_to_ascii_simd_stride {
    // Safety: load/store must be valid for 16 bytes of read/write, which may be unaligned. (candidates: `(load|store)(16|8)_(unaligned|aligned)` functions)
    ($name:ident, $load:ident, $store:ident) => {
        /// Safety: src and dst must be valid for 32/16 bytes of read/write according to
        /// the $load/$store fn, which may allow for unaligned reads/writes or require
        /// alignment to either 16x8 or u8x16.
        #[inline(always)]
        pub unsafe fn $name(src: *const u16, dst: *mut u8) -> bool {
            let first = $load(src);
            let second = $load(src.add(8));
            if simd_is_basic_latin(first | second) {
                $store(dst, simd_pack(first, second));
                true
            } else {
                false
            }
        }
    };
}

#[allow(unused_macros)]
macro_rules! pack_simd_stride {
    // Safety: load/store must be valid for 16 bytes of read/write, which may be unaligned. (candidates: `(load|store)(16|8)_(unaligned|aligned)` functions)
    ($name:ident, $load:ident, $store:ident) => {
        /// Safety: src and dst must be valid for 32/16 bytes of read/write according to
        /// the $load/$store fn, which may allow for unaligned reads/writes or require
        /// alignment to either 16x8 or u8x16.
        #[inline(always)]
        pub unsafe fn $name(src: *const u16, dst: *mut u8) {
            let first = $load(src);
            let second = $load(src.add(8));
            $store(dst, simd_pack(first, second));
        }
    };
}

cfg_if! {
    if #[cfg(all(feature = "simd-accel", target_endian = "little", target_arch = "aarch64"))] {
        // SIMD with the same instructions for aligned and unaligned loads and stores

        pub const SIMD_STRIDE_SIZE: usize = 16;

        pub const MAX_STRIDE_SIZE: usize = 16;

//        pub const ALIGNMENT: usize = 8;

        pub const ALU_STRIDE_SIZE: usize = 16;

        pub const ALU_ALIGNMENT: usize = 8;

        pub const ALU_ALIGNMENT_MASK: usize = 7;

        // Safety for stride macros: We stick to the load8_aligned/etc family of functions. We consistently produce
        // neither_unaligned variants using only unaligned inputs.
        ascii_to_ascii_simd_stride!(ascii_to_ascii_stride_neither_aligned, load16_unaligned, store16_unaligned);

        ascii_to_basic_latin_simd_stride!(ascii_to_basic_latin_stride_neither_aligned, load16_unaligned, store8_unaligned);
        unpack_simd_stride!(unpack_stride_neither_aligned, load16_unaligned, store8_unaligned);

        basic_latin_to_ascii_simd_stride!(basic_latin_to_ascii_stride_neither_aligned, load8_unaligned, store16_unaligned);
        pack_simd_stride!(pack_stride_neither_aligned, load8_unaligned, store16_unaligned);

        // Safety for conversion macros: We use the unalign macro with unalign functions above. All stride functions were produced
        // by stride macros that universally munch a single SIMD_STRIDE_SIZE worth of elements.
        ascii_simd_unalign!(ascii_to_ascii, u8, u8, ascii_to_ascii_stride_neither_aligned);
        ascii_simd_unalign!(ascii_to_basic_latin, u8, u16, ascii_to_basic_latin_stride_neither_aligned);
        ascii_simd_unalign!(basic_latin_to_ascii, u16, u8, basic_latin_to_ascii_stride_neither_aligned);
        latin1_simd_unalign!(unpack_latin1, u8, u16, unpack_stride_neither_aligned);
        latin1_simd_unalign!(pack_latin1, u16, u8, pack_stride_neither_aligned);
    } else if #[cfg(all(feature = "simd-accel", target_endian = "little", target_feature = "neon"))] {
        // SIMD with different instructions for aligned and unaligned loads and stores.
        //
        // Newer microarchitectures are not supposed to have a performance difference between
        // aligned and unaligned SSE2 loads and stores when the address is actually aligned,
        // but the benchmark results I see don't agree.

        pub const SIMD_STRIDE_SIZE: usize = 16;

        pub const MAX_STRIDE_SIZE: usize = 16;

        pub const SIMD_ALIGNMENT_MASK: usize = 15;

        // Safety for stride macros: We stick to the load8_aligned/etc family of functions. We consistently name
        // aligned/unaligned functions according to src/dst being aligned/unaligned

        ascii_to_ascii_simd_stride!(ascii_to_ascii_stride_both_aligned, load16_aligned, store16_aligned);
        ascii_to_ascii_simd_stride!(ascii_to_ascii_stride_src_aligned, load16_aligned, store16_unaligned);
        ascii_to_ascii_simd_stride!(ascii_to_ascii_stride_dst_aligned, load16_unaligned, store16_aligned);
        ascii_to_ascii_simd_stride!(ascii_to_ascii_stride_neither_aligned, load16_unaligned, store16_unaligned);

        ascii_to_basic_latin_simd_stride!(ascii_to_basic_latin_stride_both_aligned, load16_aligned, store8_aligned);
        ascii_to_basic_latin_simd_stride!(ascii_to_basic_latin_stride_src_aligned, load16_aligned, store8_unaligned);
        ascii_to_basic_latin_simd_stride!(ascii_to_basic_latin_stride_dst_aligned, load16_unaligned, store8_aligned);
        ascii_to_basic_latin_simd_stride!(ascii_to_basic_latin_stride_neither_aligned, load16_unaligned, store8_unaligned);

        unpack_simd_stride!(unpack_stride_both_aligned, load16_aligned, store8_aligned);
        unpack_simd_stride!(unpack_stride_src_aligned, load16_aligned, store8_unaligned);
        unpack_simd_stride!(unpack_stride_dst_aligned, load16_unaligned, store8_aligned);
        unpack_simd_stride!(unpack_stride_neither_aligned, load16_unaligned, store8_unaligned);

        basic_latin_to_ascii_simd_stride!(basic_latin_to_ascii_stride_both_aligned, load8_aligned, store16_aligned);
        basic_latin_to_ascii_simd_stride!(basic_latin_to_ascii_stride_src_aligned, load8_aligned, store16_unaligned);
        basic_latin_to_ascii_simd_stride!(basic_latin_to_ascii_stride_dst_aligned, load8_unaligned, store16_aligned);
        basic_latin_to_ascii_simd_stride!(basic_latin_to_ascii_stride_neither_aligned, load8_unaligned, store16_unaligned);

        pack_simd_stride!(pack_stride_both_aligned, load8_aligned, store16_aligned);
        pack_simd_stride!(pack_stride_src_aligned, load8_aligned, store16_unaligned);
        pack_simd_stride!(pack_stride_dst_aligned, load8_unaligned, store16_aligned);
        pack_simd_stride!(pack_stride_neither_aligned, load8_unaligned, store16_unaligned);

        // Safety for conversion macros: We use the correct pattern of both/src/dst/neither here. All stride functions were produced
        // by stride macros that universally munch a single SIMD_STRIDE_SIZE worth of elements.

        ascii_simd_check_align!(ascii_to_ascii, u8, u8, ascii_to_ascii_stride_both_aligned, ascii_to_ascii_stride_src_aligned, ascii_to_ascii_stride_dst_aligned, ascii_to_ascii_stride_neither_aligned);
        ascii_simd_check_align!(ascii_to_basic_latin, u8, u16, ascii_to_basic_latin_stride_both_aligned, ascii_to_basic_latin_stride_src_aligned, ascii_to_basic_latin_stride_dst_aligned, ascii_to_basic_latin_stride_neither_aligned);
        ascii_simd_check_align!(basic_latin_to_ascii, u16, u8, basic_latin_to_ascii_stride_both_aligned, basic_latin_to_ascii_stride_src_aligned, basic_latin_to_ascii_stride_dst_aligned, basic_latin_to_ascii_stride_neither_aligned);
        latin1_simd_check_align!(unpack_latin1, u8, u16, unpack_stride_both_aligned, unpack_stride_src_aligned, unpack_stride_dst_aligned, unpack_stride_neither_aligned);
        latin1_simd_check_align!(pack_latin1, u16, u8, pack_stride_both_aligned, pack_stride_src_aligned, pack_stride_dst_aligned, pack_stride_neither_aligned);
    } else if #[cfg(all(feature = "simd-accel", target_feature = "sse2"))] {
        // SIMD with different instructions for aligned and unaligned loads and stores.
        //
        // Newer microarchitectures are not supposed to have a performance difference between
        // aligned and unaligned SSE2 loads and stores when the address is actually aligned,
        // but the benchmark results I see don't agree.

        pub const SIMD_STRIDE_SIZE: usize = 16;

        /// Safety-usable invariant: This should be identical to SIMD_STRIDE_SIZE (used by ascii_simd_check_align_unrolled)
        pub const SIMD_ALIGNMENT: usize = 16;

        pub const MAX_STRIDE_SIZE: usize = 16;

        pub const SIMD_ALIGNMENT_MASK: usize = 15;

        // Safety for stride macros: We stick to the load8_aligned/etc family of functions. We consistently name
        // aligned/unaligned functions according to src/dst being aligned/unaligned

        ascii_to_ascii_simd_double_stride!(ascii_to_ascii_simd_double_stride_both_aligned, store16_aligned);
        ascii_to_ascii_simd_double_stride!(ascii_to_ascii_simd_double_stride_src_aligned, store16_unaligned);

        ascii_to_basic_latin_simd_double_stride!(ascii_to_basic_latin_simd_double_stride_both_aligned, store8_aligned);
        ascii_to_basic_latin_simd_double_stride!(ascii_to_basic_latin_simd_double_stride_src_aligned, store8_unaligned);

        ascii_to_ascii_simd_stride!(ascii_to_ascii_stride_both_aligned, load16_aligned, store16_aligned);
        ascii_to_ascii_simd_stride!(ascii_to_ascii_stride_src_aligned, load16_aligned, store16_unaligned);
        ascii_to_ascii_simd_stride!(ascii_to_ascii_stride_neither_aligned, load16_unaligned, store16_unaligned);

        ascii_to_basic_latin_simd_stride!(ascii_to_basic_latin_stride_both_aligned, load16_aligned, store8_aligned);
        ascii_to_basic_latin_simd_stride!(ascii_to_basic_latin_stride_src_aligned, load16_aligned, store8_unaligned);
        ascii_to_basic_latin_simd_stride!(ascii_to_basic_latin_stride_neither_aligned, load16_unaligned, store8_unaligned);

        unpack_simd_stride!(unpack_stride_both_aligned, load16_aligned, store8_aligned);
        unpack_simd_stride!(unpack_stride_src_aligned, load16_aligned, store8_unaligned);

        basic_latin_to_ascii_simd_stride!(basic_latin_to_ascii_stride_both_aligned, load8_aligned, store16_aligned);
        basic_latin_to_ascii_simd_stride!(basic_latin_to_ascii_stride_src_aligned, load8_aligned, store16_unaligned);
        basic_latin_to_ascii_simd_stride!(basic_latin_to_ascii_stride_dst_aligned, load8_unaligned, store16_aligned);
        basic_latin_to_ascii_simd_stride!(basic_latin_to_ascii_stride_neither_aligned, load8_unaligned, store16_unaligned);

        pack_simd_stride!(pack_stride_both_aligned, load8_aligned, store16_aligned);
        pack_simd_stride!(pack_stride_src_aligned, load8_aligned, store16_unaligned);

        // Safety for conversion macros: We use the correct pattern of both/src/dst/neither/double_both/double_src here. All stride functions were produced
        // by stride macros that universally munch a single SIMD_STRIDE_SIZE worth of elements.

        ascii_simd_check_align_unrolled!(ascii_to_ascii, u8, u8, ascii_to_ascii_stride_both_aligned, ascii_to_ascii_stride_src_aligned, ascii_to_ascii_stride_neither_aligned, ascii_to_ascii_simd_double_stride_both_aligned, ascii_to_ascii_simd_double_stride_src_aligned);
        ascii_simd_check_align_unrolled!(ascii_to_basic_latin, u8, u16, ascii_to_basic_latin_stride_both_aligned, ascii_to_basic_latin_stride_src_aligned, ascii_to_basic_latin_stride_neither_aligned, ascii_to_basic_latin_simd_double_stride_both_aligned, ascii_to_basic_latin_simd_double_stride_src_aligned);

        ascii_simd_check_align!(basic_latin_to_ascii, u16, u8, basic_latin_to_ascii_stride_both_aligned, basic_latin_to_ascii_stride_src_aligned, basic_latin_to_ascii_stride_dst_aligned, basic_latin_to_ascii_stride_neither_aligned);
        latin1_simd_check_align_unrolled!(unpack_latin1, u8, u16, unpack_stride_both_aligned, unpack_stride_src_aligned, unpack_stride_dst_aligned, unpack_stride_neither_aligned);
        latin1_simd_check_align_unrolled!(pack_latin1, u16, u8, pack_stride_both_aligned, pack_stride_src_aligned, pack_stride_dst_aligned, pack_stride_neither_aligned);
    } else if #[cfg(all(target_endian = "little", target_pointer_width = "64"))] {
        // Aligned ALU word, little-endian, 64-bit

        /// Safety invariant: this is the amount of bytes consumed by
        /// unpack_alu. This will be twice the pointer width, as it consumes two usizes.
        /// This is also the number of bytes produced by pack_alu.
        /// This is also the number of u16 code units produced/consumed by unpack_alu/pack_alu respectively.
        pub const ALU_STRIDE_SIZE: usize = 16;

        pub const MAX_STRIDE_SIZE: usize = 16;

        // Safety invariant: this is the pointer width in bytes
        pub const ALU_ALIGNMENT: usize = 8;

        // Safety invariant: this is a mask for getting the bits of a pointer not aligned to ALU_ALIGNMENT
        pub const ALU_ALIGNMENT_MASK: usize = 7;

        /// Safety: dst must point to valid space for writing four `usize`s
        #[inline(always)]
        unsafe fn unpack_alu(word: usize, second_word: usize, dst: *mut usize) {
            let first = ((0x0000_0000_FF00_0000usize & word) << 24) |
                        ((0x0000_0000_00FF_0000usize & word) << 16) |
                        ((0x0000_0000_0000_FF00usize & word) << 8) |
                        (0x0000_0000_0000_00FFusize & word);
            let second = ((0xFF00_0000_0000_0000usize & word) >> 8) |
                         ((0x00FF_0000_0000_0000usize & word) >> 16) |
                         ((0x0000_FF00_0000_0000usize & word) >> 24) |
                         ((0x0000_00FF_0000_0000usize & word) >> 32);
            let third = ((0x0000_0000_FF00_0000usize & second_word) << 24) |
                        ((0x0000_0000_00FF_0000usize & second_word) << 16) |
                        ((0x0000_0000_0000_FF00usize & second_word) << 8) |
                        (0x0000_0000_0000_00FFusize & second_word);
            let fourth = ((0xFF00_0000_0000_0000usize & second_word) >> 8) |
                         ((0x00FF_0000_0000_0000usize & second_word) >> 16) |
                         ((0x0000_FF00_0000_0000usize & second_word) >> 24) |
                         ((0x0000_00FF_0000_0000usize & second_word) >> 32);
            // Safety: fn invariant used here
            *dst = first;
            *(dst.add(1)) = second;
            *(dst.add(2)) = third;
            *(dst.add(3)) = fourth;
        }

        /// Safety: dst must point to valid space for writing two `usize`s
        #[inline(always)]
        unsafe fn pack_alu(first: usize, second: usize, third: usize, fourth: usize, dst: *mut usize) {
            let word = ((0x00FF_0000_0000_0000usize & second) << 8) |
                       ((0x0000_00FF_0000_0000usize & second) << 16) |
                       ((0x0000_0000_00FF_0000usize & second) << 24) |
                       ((0x0000_0000_0000_00FFusize & second) << 32) |
                       ((0x00FF_0000_0000_0000usize & first) >> 24) |
                       ((0x0000_00FF_0000_0000usize & first) >> 16) |
                       ((0x0000_0000_00FF_0000usize & first) >> 8) |
                       (0x0000_0000_0000_00FFusize & first);
            let second_word = ((0x00FF_0000_0000_0000usize & fourth) << 8) |
                              ((0x0000_00FF_0000_0000usize & fourth) << 16) |
                              ((0x0000_0000_00FF_0000usize & fourth) << 24) |
                              ((0x0000_0000_0000_00FFusize & fourth) << 32) |
                              ((0x00FF_0000_0000_0000usize & third) >> 24) |
                              ((0x0000_00FF_0000_0000usize & third) >> 16) |
                              ((0x0000_0000_00FF_0000usize & third) >> 8) |
                              (0x0000_0000_0000_00FFusize & third);
            // Safety: fn invariant used here
            *dst = word;
            *(dst.add(1)) = second_word;
        }
    } else if #[cfg(all(target_endian = "little", target_pointer_width = "32"))] {
        // Aligned ALU word, little-endian, 32-bit

        /// Safety invariant: this is the amount of bytes consumed by
        /// unpack_alu. This will be twice the pointer width, as it consumes two usizes.
        /// This is also the number of bytes produced by pack_alu.
        /// This is also the number of u16 code units produced/consumed by unpack_alu/pack_alu respectively.
        pub const ALU_STRIDE_SIZE: usize = 8;

        pub const MAX_STRIDE_SIZE: usize = 8;

        // Safety invariant: this is the pointer width in bytes
        pub const ALU_ALIGNMENT: usize = 4;

        // Safety invariant: this is a mask for getting the bits of a pointer not aligned to ALU_ALIGNMENT
        pub const ALU_ALIGNMENT_MASK: usize = 3;

        /// Safety: dst must point to valid space for writing four `usize`s
        #[inline(always)]
        unsafe fn unpack_alu(word: usize, second_word: usize, dst: *mut usize) {
            let first = ((0x0000_FF00usize & word) << 8) |
                        (0x0000_00FFusize & word);
            let second = ((0xFF00_0000usize & word) >> 8) |
                         ((0x00FF_0000usize & word) >> 16);
            let third = ((0x0000_FF00usize & second_word) << 8) |
                        (0x0000_00FFusize & second_word);
            let fourth = ((0xFF00_0000usize & second_word) >> 8) |
                         ((0x00FF_0000usize & second_word) >> 16);
            // Safety: fn invariant used here
            *dst = first;
            *(dst.add(1)) = second;
            *(dst.add(2)) = third;
            *(dst.add(3)) = fourth;
        }

        /// Safety: dst must point to valid space for writing two `usize`s
        #[inline(always)]
        unsafe fn pack_alu(first: usize, second: usize, third: usize, fourth: usize, dst: *mut usize) {
            let word = ((0x00FF_0000usize & second) << 8) |
                       ((0x0000_00FFusize & second) << 16) |
                       ((0x00FF_0000usize & first) >> 8) |
                       (0x0000_00FFusize & first);
            let second_word = ((0x00FF_0000usize & fourth) << 8) |
                              ((0x0000_00FFusize & fourth) << 16) |
                              ((0x00FF_0000usize & third) >> 8) |
                              (0x0000_00FFusize & third);
            // Safety: fn invariant used here
            *dst = word;
            *(dst.add(1)) = second_word;
        }
    } else if #[cfg(all(target_endian = "big", target_pointer_width = "64"))] {
        // Aligned ALU word, big-endian, 64-bit

        /// Safety invariant: this is the amount of bytes consumed by
        /// unpack_alu. This will be twice the pointer width, as it consumes two usizes.
        /// This is also the number of bytes produced by pack_alu.
        /// This is also the number of u16 code units produced/consumed by unpack_alu/pack_alu respectively.
        pub const ALU_STRIDE_SIZE: usize = 16;

        pub const MAX_STRIDE_SIZE: usize = 16;

        // Safety invariant: this is the pointer width in bytes
        pub const ALU_ALIGNMENT: usize = 8;

        // Safety invariant: this is a mask for getting the bits of a pointer not aligned to ALU_ALIGNMENT
        pub const ALU_ALIGNMENT_MASK: usize = 7;

        /// Safety: dst must point to valid space for writing four `usize`s
        #[inline(always)]
        unsafe fn unpack_alu(word: usize, second_word: usize, dst: *mut usize) {
            let first = ((0xFF00_0000_0000_0000usize & word) >> 8) |
                         ((0x00FF_0000_0000_0000usize & word) >> 16) |
                         ((0x0000_FF00_0000_0000usize & word) >> 24) |
                         ((0x0000_00FF_0000_0000usize & word) >> 32);
            let second = ((0x0000_0000_FF00_0000usize & word) << 24) |
                        ((0x0000_0000_00FF_0000usize & word) << 16) |
                        ((0x0000_0000_0000_FF00usize & word) << 8) |
                        (0x0000_0000_0000_00FFusize & word);
            let third = ((0xFF00_0000_0000_0000usize & second_word) >> 8) |
                         ((0x00FF_0000_0000_0000usize & second_word) >> 16) |
                         ((0x0000_FF00_0000_0000usize & second_word) >> 24) |
                         ((0x0000_00FF_0000_0000usize & second_word) >> 32);
            let fourth = ((0x0000_0000_FF00_0000usize & second_word) << 24) |
                        ((0x0000_0000_00FF_0000usize & second_word) << 16) |
                        ((0x0000_0000_0000_FF00usize & second_word) << 8) |
                        (0x0000_0000_0000_00FFusize & second_word);
            // Safety: fn invariant used here
            *dst = first;
            *(dst.add(1)) = second;
            *(dst.add(2)) = third;
            *(dst.add(3)) = fourth;
        }

        /// Safety: dst must point to valid space for writing two `usize`s
        #[inline(always)]
        unsafe fn pack_alu(first: usize, second: usize, third: usize, fourth: usize, dst: *mut usize) {
            let word = ((0x00FF0000_00000000usize & first) << 8) |
                       ((0x000000FF_00000000usize & first) << 16) |
                       ((0x00000000_00FF0000usize & first) << 24) |
                       ((0x00000000_000000FFusize & first) << 32) |
                       ((0x00FF0000_00000000usize & second) >> 24) |
                       ((0x000000FF_00000000usize & second) >> 16) |
                       ((0x00000000_00FF0000usize & second) >> 8) |
                       (0x00000000_000000FFusize & second);
            let second_word = ((0x00FF0000_00000000usize & third) << 8) |
                              ((0x000000FF_00000000usize & third) << 16) |
                              ((0x00000000_00FF0000usize & third) << 24) |
                              ((0x00000000_000000FFusize & third) << 32) |
                              ((0x00FF0000_00000000usize & fourth) >> 24) |
                              ((0x000000FF_00000000usize & fourth) >> 16) |
                              ((0x00000000_00FF0000usize & fourth) >> 8) |
                              (0x00000000_000000FFusize &  fourth);
            // Safety: fn invariant used here
            *dst = word;
            *(dst.add(1)) = second_word;
        }
    } else if #[cfg(all(target_endian = "big", target_pointer_width = "32"))] {
        // Aligned ALU word, big-endian, 32-bit

        /// Safety invariant: this is the amount of bytes consumed by
        /// unpack_alu. This will be twice the pointer width, as it consumes two usizes.
        /// This is also the number of bytes produced by pack_alu.
        /// This is also the number of u16 code units produced/consumed by unpack_alu/pack_alu respectively.
        pub const ALU_STRIDE_SIZE: usize = 8;

        pub const MAX_STRIDE_SIZE: usize = 8;

        // Safety invariant: this is the pointer width in bytes
        pub const ALU_ALIGNMENT: usize = 4;

        // Safety invariant: this is a mask for getting the bits of a pointer not aligned to ALU_ALIGNMENT
        pub const ALU_ALIGNMENT_MASK: usize = 3;

        /// Safety: dst must point to valid space for writing four `usize`s
        #[inline(always)]
        unsafe fn unpack_alu(word: usize, second_word: usize, dst: *mut usize) {
            let first = ((0xFF00_0000usize & word) >> 8) |
                         ((0x00FF_0000usize & word) >> 16);
            let second = ((0x0000_FF00usize & word) << 8) |
                        (0x0000_00FFusize & word);
            let third = ((0xFF00_0000usize & second_word) >> 8) |
                         ((0x00FF_0000usize & second_word) >> 16);
            let fourth = ((0x0000_FF00usize & second_word) << 8) |
                        (0x0000_00FFusize & second_word);
            // Safety: fn invariant used here
            *dst = first;
            *(dst.add(1)) = second;
            *(dst.add(2)) = third;
            *(dst.add(3)) = fourth;
        }

        /// Safety: dst must point to valid space for writing two `usize`s
        #[inline(always)]
        unsafe fn pack_alu(first: usize, second: usize, third: usize, fourth: usize, dst: *mut usize) {
            let word = ((0x00FF_0000usize & first) << 8) |
                       ((0x0000_00FFusize & first) << 16) |
                       ((0x00FF_0000usize & second) >> 8) |
                       (0x0000_00FFusize & second);
            let second_word = ((0x00FF_0000usize & third) << 8) |
                              ((0x0000_00FFusize & third) << 16) |
                              ((0x00FF_0000usize & fourth) >> 8) |
                              (0x0000_00FFusize & fourth);
            // Safety: fn invariant used here
            *dst = word;
            *(dst.add(1)) = second_word;
        }
    } else {
        ascii_naive!(ascii_to_ascii, u8, u8);
        ascii_naive!(ascii_to_basic_latin, u8, u16);
        ascii_naive!(basic_latin_to_ascii, u16, u8);
    }
}

cfg_if! {
    // Safety-usable invariant: this counts the zeroes from the "first byte" of utf-8 data packed into a usize
    // with the target endianness
    if #[cfg(target_endian = "little")] {
        #[allow(dead_code)]
        #[inline(always)]
        fn count_zeros(word: usize) -> u32 {
            word.trailing_zeros()
        }
    } else {
        #[allow(dead_code)]
        #[inline(always)]
        fn count_zeros(word: usize) -> u32 {
            word.leading_zeros()
        }
    }
}

cfg_if! {
    if #[cfg(all(feature = "simd-accel", target_endian = "little", target_arch = "disabled"))] {
        /// Safety-usable invariant: Will return the value and position of the first non-ASCII byte in the slice in a Some if found.
        /// In other words, the first element of the Some is always `> 127`
        #[inline(always)]
        pub fn validate_ascii(slice: &[u8]) -> Option<(u8, usize)> {
            let src = slice.as_ptr();
            let len = slice.len();
            let mut offset = 0usize;
            // Safety: if this check succeeds we're valid for reading/writing at least `stride` elements.
            if SIMD_STRIDE_SIZE <= len {
                let len_minus_stride = len - SIMD_STRIDE_SIZE;
                loop {
                    // Safety: src at offset is valid for a `SIMD_STRIDE_SIZE` read
                    let simd = unsafe { load16_unaligned(src.add(offset)) };
                    if !simd_is_ascii(simd) {
                        break;
                    }
                    offset += SIMD_STRIDE_SIZE;
                    // This is `offset > len - SIMD_STRIDE_SIZE` which means we always have at least `SIMD_STRIDE_SIZE` elements to munch next time.
                    if offset > len_minus_stride {
                        break;
                    }
                }
            }
            while offset < len {
                let code_unit = slice[offset];
                if code_unit > 127 {
                    // Safety: Safety-usable invariant upheld here
                    return Some((code_unit, offset));
                }
                offset += 1;
            }
            None
        }
    } else if #[cfg(all(feature = "simd-accel", target_feature = "sse2"))] {
        /// Safety-usable invariant: will return Some() when it encounters non-ASCII, with the first element in the Some being
        /// guaranteed to be non-ASCII (> 127), and the second being the offset where it is found
        #[inline(always)]
        pub fn validate_ascii(slice: &[u8]) -> Option<(u8, usize)> {
            let src = slice.as_ptr();
            let len = slice.len();
            let mut offset = 0usize;
            // Safety: if this check succeeds we're valid for reading at least `stride` elements.
            if SIMD_STRIDE_SIZE <= len {
                // First, process one unaligned vector
                // Safety: src is valid for a `SIMD_STRIDE_SIZE` read
                let simd = unsafe { load16_unaligned(src) };
                let mask = mask_ascii(simd);
                if mask != 0 {
                    offset = mask.trailing_zeros() as usize;
                    let non_ascii = unsafe { *src.add(offset) };
                    return Some((non_ascii, offset));
                }
                offset = SIMD_STRIDE_SIZE;
                // Safety: Now that offset has changed we don't yet know how much it is valid for

                // We have now seen 16 ASCII bytes. Let's guess that
                // there will be enough more to justify more expense
                // in the case of non-ASCII.
                // Use aligned reads for the sake of old microachitectures.
                // Safety: this correctly calculates the number of src_units that need to be read before the remaining list is aligned.
                // This is by definition less than SIMD_ALIGNMENT, which is defined to be equal to SIMD_STRIDE_SIZE.
                let until_alignment = unsafe { (SIMD_ALIGNMENT - ((src.add(offset) as usize) & SIMD_ALIGNMENT_MASK)) & SIMD_ALIGNMENT_MASK };
                // This addition won't overflow, because even in the 32-bit PAE case the
                // address space holds enough code that the slice length can't be that
                // close to address space size.
                // offset now equals SIMD_STRIDE_SIZE, hence times 3 below.
                //
                // Safety: if this check succeeds we're valid for reading at least `2 * SIMD_STRIDE_SIZE` elements plus `until_alignment`.
                // The extra SIMD_STRIDE_SIZE in the condition is because `offset` is already `SIMD_STRIDE_SIZE`.
                if until_alignment + (SIMD_STRIDE_SIZE * 3) <= len {
                    if until_alignment != 0 {
                        // Safety: this is safe to call since we're valid for this read (and more), and don't care about alignment
                        // This will copy over bytes that get decoded twice since it's not incrementing `offset` by SIMD_STRIDE_SIZE. This is fine.
                        let simd = unsafe { load16_unaligned(src.add(offset)) };
                        let mask = mask_ascii(simd);
                        if mask != 0 {
                            offset += mask.trailing_zeros() as usize;
                            let non_ascii = unsafe { *src.add(offset) };
                            return Some((non_ascii, offset));
                        }
                        offset += until_alignment;
                    }
                    // Safety: At this point we're valid for reading 2*SIMD_STRIDE_SIZE elements
                    // Safety: Now `offset` is aligned for `src`
                    let len_minus_stride_times_two = len - (SIMD_STRIDE_SIZE * 2);
                    loop {
                        // Safety: We were valid for this read, and were aligned.
                        let first = unsafe { load16_aligned(src.add(offset)) };
                        let second = unsafe { load16_aligned(src.add(offset + SIMD_STRIDE_SIZE)) };
                        if !simd_is_ascii(first | second) {
                            // Safety: mask_ascii produces a mask of all the high bits.
                            let mask_first = mask_ascii(first);
                            if mask_first != 0 {
                                // Safety: on little endian systems this will be the number of ascii bytes
                                // before the first non-ascii, i.e. valid for indexing src
                                // TODO SAFETY: What about big-endian systems?
                                offset += mask_first.trailing_zeros() as usize;
                            } else {
                                let mask_second = mask_ascii(second);
                                // Safety: on little endian systems this will be the number of ascii bytes
                                // before the first non-ascii, i.e. valid for indexing src
                                offset += SIMD_STRIDE_SIZE + mask_second.trailing_zeros() as usize;
                            }
                            // Safety: We know this is non-ASCII, and can uphold the safety-usable invariant here
                            let non_ascii = unsafe { *src.add(offset) };

                            return Some((non_ascii, offset));
                        }
                        offset += SIMD_STRIDE_SIZE * 2;
                        // Safety: This is `offset > len - 2 * SIMD_STRIDE_SIZE` which means we always have at least `2 * SIMD_STRIDE_SIZE` elements to munch next time.
                        if offset > len_minus_stride_times_two {
                            break;
                        }
                    }
                    // Safety: if this check succeeds we're valid for reading at least `SIMD_STRIDE_SIZE`
                    if offset + SIMD_STRIDE_SIZE <= len {
                        // Safety: We were valid for this read, and were aligned.
                        let simd = unsafe { load16_aligned(src.add(offset)) };
                        // Safety: mask_ascii produces a mask of all the high bits.
                        let mask = mask_ascii(simd);
                        if mask != 0 {
                            // Safety: on little endian systems this will be the number of ascii bytes
                            // before the first non-ascii, i.e. valid for indexing src
                            offset += mask.trailing_zeros() as usize;
                            let non_ascii = unsafe { *src.add(offset) };
                            // Safety: We know this is non-ASCII, and can uphold the safety-usable invariant here
                            return Some((non_ascii, offset));
                        }
                        offset += SIMD_STRIDE_SIZE;
                    }
                } else {
                    // Safety: this is the unaligned branch
                    // At most two iterations, so unroll
                    // Safety: if this check succeeds we're valid for reading at least `SIMD_STRIDE_SIZE`
                    if offset + SIMD_STRIDE_SIZE <= len {
                        // Safety: We're valid for this read but must use an unaligned read
                        let simd = unsafe { load16_unaligned(src.add(offset)) };
                        let mask = mask_ascii(simd);
                        if mask != 0 {
                            offset += mask.trailing_zeros() as usize;
                            let non_ascii = unsafe { *src.add(offset) };
                            // Safety-usable invariant upheld here (same as above)
                            return Some((non_ascii, offset));
                        }
                        offset += SIMD_STRIDE_SIZE;
                        // Safety: if this check succeeds we're valid for reading at least `SIMD_STRIDE_SIZE`
                        if offset + SIMD_STRIDE_SIZE <= len {
                            // Safety: We're valid for this read but must use an unaligned read
                             let simd = unsafe { load16_unaligned(src.add(offset)) };
                             let mask = mask_ascii(simd);
                            if mask != 0 {
                                offset += mask.trailing_zeros() as usize;
                                let non_ascii = unsafe { *src.add(offset) };
                                // Safety-usable invariant upheld here (same as above)
                                return Some((non_ascii, offset));
                            }
                            offset += SIMD_STRIDE_SIZE;
                        }
                    }
                }
            }
            while offset < len {
                // Safety: relies straightforwardly on the `len` invariant
                let code_unit = unsafe { *(src.add(offset)) };
                if code_unit > 127 {
                    // Safety-usable invariant upheld here
                    return Some((code_unit, offset));
                }
                offset += 1;
            }
            None
        }
    } else {
        // Safety-usable invariant: returns byte index of first non-ascii byte
        #[inline(always)]
        fn find_non_ascii(word: usize, second_word: usize) -> Option<usize> {
            let word_masked = word & ASCII_MASK;
            let second_masked = second_word & ASCII_MASK;
            if (word_masked | second_masked) == 0 {
                // Both are ascii, invariant upheld
                return None;
            }
            if word_masked != 0 {
                let zeros = count_zeros(word_masked);
                // `zeros` now contains 0 to 7 (for the seven bits of masked ASCII in little endian,
                // or up to 7 bits of non-ASCII in big endian if the first byte is non-ASCII)
                // plus 8 times the number of ASCII in text order before the
                // non-ASCII byte in the little-endian case or 8 times the number of ASCII in
                // text order before the non-ASCII byte in the big-endian case.
                let num_ascii = (zeros >> 3) as usize;
                // Safety-usable invariant upheld here
                return Some(num_ascii);
            }
            let zeros = count_zeros(second_masked);
            // `zeros` now contains 0 to 7 (for the seven bits of masked ASCII in little endian,
            // or up to 7 bits of non-ASCII in big endian if the first byte is non-ASCII)
            // plus 8 times the number of ASCII in text order before the
            // non-ASCII byte in the little-endian case or 8 times the number of ASCII in
            // text order before the non-ASCII byte in the big-endian case.
            let num_ascii = (zeros >> 3) as usize;
            // Safety-usable invariant upheld here
            Some(ALU_ALIGNMENT + num_ascii)
        }

        /// Safety: `src` must be valid for the reads of two `usize`s
        ///
        /// Safety-usable invariant: will return byte index of first non-ascii byte
        #[inline(always)]
        unsafe fn validate_ascii_stride(src: *const usize) -> Option<usize> {
            let word = *src;
            let second_word = *(src.add(1));
            find_non_ascii(word, second_word)
        }

        /// Safety-usable invariant: will return Some() when it encounters non-ASCII, with the first element in the Some being
        /// guaranteed to be non-ASCII (> 127), and the second being the offset where it is found
        #[cfg_attr(feature = "cargo-clippy", allow(cast_ptr_alignment))]
        #[inline(always)]
        pub fn validate_ascii(slice: &[u8]) -> Option<(u8, usize)> {
            let src = slice.as_ptr();
            let len = slice.len();
            let mut offset = 0usize;
            let mut until_alignment = (ALU_ALIGNMENT - ((src as usize) & ALU_ALIGNMENT_MASK)) & ALU_ALIGNMENT_MASK;
            // Safety: If this check fails we're valid to read `until_alignment + ALU_STRIDE_SIZE` elements
            if until_alignment + ALU_STRIDE_SIZE <= len {
                while until_alignment != 0 {
                    let code_unit = slice[offset];
                    if code_unit > 127 {
                        // Safety-usable invairant upheld here
                        return Some((code_unit, offset));
                    }
                    offset += 1;
                    until_alignment -= 1;
                }
                // Safety: At this point we have read until_alignment elements and
                // are valid for `ALU_STRIDE_SIZE` more.
                let len_minus_stride = len - ALU_STRIDE_SIZE;
                loop {
                    // Safety: we were valid for this read
                    let ptr = unsafe { src.add(offset) as *const usize };
                    if let Some(num_ascii) = unsafe { validate_ascii_stride(ptr) } {
                        offset += num_ascii;
                        // Safety-usable invairant upheld here using the invariant from validate_ascii_stride()
                        return Some((unsafe { *(src.add(offset)) }, offset));
                    }
                    offset += ALU_STRIDE_SIZE;
                    // Safety: This is `offset > ALU_STRIDE_SIZE` which means we always have at least `2 * ALU_STRIDE_SIZE` elements to munch next time.
                    if offset > len_minus_stride {
                        break;
                    }
                }
            }
            while offset < len {
                let code_unit = slice[offset];
                if code_unit > 127 {
                    // Safety-usable invairant upheld here
                    return Some((code_unit, offset));
                }
                offset += 1;
           }
           None
        }

    }
}

cfg_if! {
    if #[cfg(all(feature = "simd-accel", any(target_feature = "sse2", all(target_endian = "little", target_arch = "aarch64"))))] {

    } else if #[cfg(all(feature = "simd-accel", target_endian = "little", target_feature = "neon"))] {
        // Even with NEON enabled, we use the ALU path for ASCII validation, because testing
        // on Exynos 5 indicated that using NEON isn't worthwhile where there are only
        // vector reads without vector writes.

        pub const ALU_STRIDE_SIZE: usize = 8;

        pub const ALU_ALIGNMENT: usize = 4;

        pub const ALU_ALIGNMENT_MASK: usize = 3;
    } else {
        // Safety: src points to two valid `usize`s, dst points to four valid `usize`s
        #[inline(always)]
        unsafe fn unpack_latin1_stride_alu(src: *const usize, dst: *mut usize) {
            // Safety: src safety invariant used here
            let word = *src;
            let second_word = *(src.add(1));
            // Safety: dst safety invariant passed down
            unpack_alu(word, second_word, dst);
        }

        // Safety: src points to four valid `usize`s, dst points to two valid `usize`s
        #[inline(always)]
        unsafe fn pack_latin1_stride_alu(src: *const usize, dst: *mut usize) {
            // Safety: src safety invariant used here
            let first = *src;
            let second = *(src.add(1));
            let third = *(src.add(2));
            let fourth = *(src.add(3));
            // Safety: dst safety invariant passed down
            pack_alu(first, second, third, fourth, dst);
        }

        // Safety: src points to two valid `usize`s, dst points to four valid `usize`s
        #[inline(always)]
        unsafe fn ascii_to_basic_latin_stride_alu(src: *const usize, dst: *mut usize) -> bool {
            // Safety: src safety invariant used here
            let word = *src;
            let second_word = *(src.add(1));
            // Check if the words contains non-ASCII
            if (word & ASCII_MASK) | (second_word & ASCII_MASK) != 0 {
                return false;
            }
            // Safety: dst safety invariant passed down
            unpack_alu(word, second_word, dst);
            true
        }

        // Safety: src points four valid `usize`s, dst points to two valid `usize`s
        #[inline(always)]
        unsafe fn basic_latin_to_ascii_stride_alu(src: *const usize, dst: *mut usize) -> bool {
            // Safety: src safety invariant used here
            let first = *src;
            let second = *(src.add(1));
            let third = *(src.add(2));
            let fourth = *(src.add(3));
            if (first & BASIC_LATIN_MASK) | (second & BASIC_LATIN_MASK) | (third & BASIC_LATIN_MASK) | (fourth & BASIC_LATIN_MASK) != 0 {
                return false;
            }
            // Safety: dst safety invariant passed down
            pack_alu(first, second, third, fourth, dst);
            true
        }

        // Safety: src, dst both point to two valid `usize`s each
        // Safety-usable invariant: Will return byte index of first non-ascii byte.
        #[inline(always)]
        unsafe fn ascii_to_ascii_stride(src: *const usize, dst: *mut usize) -> Option<usize> {
            // Safety: src safety invariant used here
            let word = *src;
            let second_word = *(src.add(1));
            // Safety: src safety invariant used here
            *dst = word;
            *(dst.add(1)) = second_word;
            // Relies on safety-usable invariant here
            find_non_ascii(word, second_word)
        }

        basic_latin_alu!(ascii_to_basic_latin, u8, u16, ascii_to_basic_latin_stride_alu);
        basic_latin_alu!(basic_latin_to_ascii, u16, u8, basic_latin_to_ascii_stride_alu);
        latin1_alu!(unpack_latin1, u8, u16, unpack_latin1_stride_alu);
        latin1_alu!(pack_latin1, u16, u8, pack_latin1_stride_alu);
        // Safety invariant upheld: ascii_to_ascii_stride will return byte index of first non-ascii if found
        ascii_alu!(ascii_to_ascii, u8, u8, ascii_to_ascii_stride);
    }
}

pub fn ascii_valid_up_to(bytes: &[u8]) -> usize {
    match validate_ascii(bytes) {
        None => bytes.len(),
        Some((_, num_valid)) => num_valid,
    }
}

pub fn iso_2022_jp_ascii_valid_up_to(bytes: &[u8]) -> usize {
    for (i, b_ref) in bytes.iter().enumerate() {
        let b = *b_ref;
        if b >= 0x80 || b == 0x1B || b == 0x0E || b == 0x0F {
            return i;
        }
    }
    bytes.len()
}

// Any copyright to the test code below this comment is dedicated to the
// Public Domain. http://creativecommons.org/publicdomain/zero/1.0/

#[cfg(all(test, feature = "alloc"))]
mod tests {
    use super::*;
    use alloc::vec::Vec;

    macro_rules! test_ascii {
        ($test_name:ident, $fn_tested:ident, $src_unit:ty, $dst_unit:ty) => {
            #[test]
            fn $test_name() {
                let mut src: Vec<$src_unit> = Vec::with_capacity(32);
                let mut dst: Vec<$dst_unit> = Vec::with_capacity(32);
                for i in 0..32 {
                    src.clear();
                    dst.clear();
                    dst.resize(32, 0);
                    for j in 0..32 {
                        let c = if i == j { 0xAA } else { j + 0x40 };
                        src.push(c as $src_unit);
                    }
                    match unsafe { $fn_tested(src.as_ptr(), dst.as_mut_ptr(), 32) } {
                        None => unreachable!("Should always find non-ASCII"),
                        Some((non_ascii, num_ascii)) => {
                            assert_eq!(non_ascii, 0xAA);
                            assert_eq!(num_ascii, i);
                            for j in 0..i {
                                assert_eq!(dst[j], (j + 0x40) as $dst_unit);
                            }
                        }
                    }
                }
            }
        };
    }

    test_ascii!(test_ascii_to_ascii, ascii_to_ascii, u8, u8);
    test_ascii!(test_ascii_to_basic_latin, ascii_to_basic_latin, u8, u16);
    test_ascii!(test_basic_latin_to_ascii, basic_latin_to_ascii, u16, u8);
}