xref: /freebsd/contrib/llvm-project/llvm/lib/Support/DJB.cpp (revision 77a1348b3c1cfe8547be49a121b56299a1e18b69)
1 //===-- Support/DJB.cpp ---DJB Hash -----------------------------*- C++ -*-===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // This file contains support for the DJ Bernstein hash function.
10 //
11 //===----------------------------------------------------------------------===//
12 
13 #include "llvm/Support/DJB.h"
14 #include "llvm/ADT/ArrayRef.h"
15 #include "llvm/Support/Compiler.h"
16 #include "llvm/Support/ConvertUTF.h"
17 #include "llvm/Support/Unicode.h"
18 
19 using namespace llvm;
20 
21 static UTF32 chopOneUTF32(StringRef &Buffer) {
22   UTF32 C;
23   const UTF8 *const Begin8Const =
24       reinterpret_cast<const UTF8 *>(Buffer.begin());
25   const UTF8 *Begin8 = Begin8Const;
26   UTF32 *Begin32 = &C;
27 
28   // In lenient mode we will always end up with a "reasonable" value in C for
29   // non-empty input.
30   assert(!Buffer.empty());
31   ConvertUTF8toUTF32(&Begin8, reinterpret_cast<const UTF8 *>(Buffer.end()),
32                      &Begin32, &C + 1, lenientConversion);
33   Buffer = Buffer.drop_front(Begin8 - Begin8Const);
34   return C;
35 }
36 
37 static StringRef toUTF8(UTF32 C, MutableArrayRef<UTF8> Storage) {
38   const UTF32 *Begin32 = &C;
39   UTF8 *Begin8 = Storage.begin();
40 
41   // The case-folded output should always be a valid unicode character, so use
42   // strict mode here.
43   ConversionResult CR = ConvertUTF32toUTF8(&Begin32, &C + 1, &Begin8,
44                                            Storage.end(), strictConversion);
45   assert(CR == conversionOK && "Case folding produced invalid char?");
46   (void)CR;
47   return StringRef(reinterpret_cast<char *>(Storage.begin()),
48                    Begin8 - Storage.begin());
49 }
50 
51 static UTF32 foldCharDwarf(UTF32 C) {
52   // DWARF v5 addition to the unicode folding rules.
53   // Fold "Latin Small Letter Dotless I" and "Latin Capital Letter I With Dot
54   // Above" into "i".
55   if (C == 0x130 || C == 0x131)
56     return 'i';
57   return sys::unicode::foldCharSimple(C);
58 }
59 
60 static Optional<uint32_t> fastCaseFoldingDjbHash(StringRef Buffer, uint32_t H) {
61   bool AllASCII = true;
62   for (unsigned char C : Buffer) {
63     H = H * 33 + ('A' <= C && C <= 'Z' ? C - 'A' + 'a' : C);
64     AllASCII &= C <= 0x7f;
65   }
66   if (AllASCII)
67     return H;
68   return None;
69 }
70 
71 uint32_t llvm::caseFoldingDjbHash(StringRef Buffer, uint32_t H) {
72   if (Optional<uint32_t> Result = fastCaseFoldingDjbHash(Buffer, H))
73     return *Result;
74 
75   std::array<UTF8, UNI_MAX_UTF8_BYTES_PER_CODE_POINT> Storage;
76   while (!Buffer.empty()) {
77     UTF32 C = foldCharDwarf(chopOneUTF32(Buffer));
78     StringRef Folded = toUTF8(C, Storage);
79     H = djbHash(Folded, H);
80   }
81   return H;
82 }
83