1 //===-- Support/DJB.cpp ---DJB Hash -----------------------------*- C++ -*-===//
2 //
3 //                     The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // This file contains support for the DJ Bernstein hash function.
11 //
12 //===----------------------------------------------------------------------===//
13 
14 #include "llvm/Support/DJB.h"
15 #include "llvm/ADT/ArrayRef.h"
16 #include "llvm/Support/Compiler.h"
17 #include "llvm/Support/ConvertUTF.h"
18 #include "llvm/Support/Unicode.h"
19 
20 using namespace llvm;
21 
22 static UTF32 chopOneUTF32(StringRef &Buffer) {
23   UTF32 C;
24   const UTF8 *const Begin8Const =
25       reinterpret_cast<const UTF8 *>(Buffer.begin());
26   const UTF8 *Begin8 = Begin8Const;
27   UTF32 *Begin32 = &C;
28 
29   // In lenient mode we will always end up with a "reasonable" value in C for
30   // non-empty input.
31   assert(!Buffer.empty());
32   ConvertUTF8toUTF32(&Begin8, reinterpret_cast<const UTF8 *>(Buffer.end()),
33                      &Begin32, &C + 1, lenientConversion);
34   Buffer = Buffer.drop_front(Begin8 - Begin8Const);
35   return C;
36 }
37 
38 static StringRef toUTF8(UTF32 C, MutableArrayRef<UTF8> Storage) {
39   const UTF32 *Begin32 = &C;
40   UTF8 *Begin8 = Storage.begin();
41 
42   // The case-folded output should always be a valid unicode character, so use
43   // strict mode here.
44   ConversionResult CR = ConvertUTF32toUTF8(&Begin32, &C + 1, &Begin8,
45                                            Storage.end(), strictConversion);
46   assert(CR == conversionOK && "Case folding produced invalid char?");
47   (void)CR;
48   return StringRef(reinterpret_cast<char *>(Storage.begin()),
49                    Begin8 - Storage.begin());
50 }
51 
52 static UTF32 foldCharDwarf(UTF32 C) {
53   // DWARF v5 addition to the unicode folding rules.
54   // Fold "Latin Small Letter Dotless I" and "Latin Capital Letter I With Dot
55   // Above" into "i".
56   if (C == 0x130 || C == 0x131)
57     return 'i';
58   return sys::unicode::foldCharSimple(C);
59 }
60 
61 static uint32_t caseFoldingDjbHashCharSlow(StringRef &Buffer, uint32_t H) {
62   UTF32 C = chopOneUTF32(Buffer);
63 
64   C = foldCharDwarf(C);
65 
66   std::array<UTF8, UNI_MAX_UTF8_BYTES_PER_CODE_POINT> Storage;
67   StringRef Folded = toUTF8(C, Storage);
68   return djbHash(Folded, H);
69 }
70 
71 uint32_t llvm::caseFoldingDjbHash(StringRef Buffer, uint32_t H) {
72   while (!Buffer.empty()) {
73     unsigned char C = Buffer.front();
74     if (LLVM_LIKELY(C <= 0x7f)) {
75       // US-ASCII, encoded as one character in utf-8.
76       // This is by far the most common case, so handle this specially.
77       if (C >= 'A' && C <= 'Z')
78         C = 'a' + (C - 'A'); // fold uppercase into lowercase
79       H = (H << 5) + H + C;
80       Buffer = Buffer.drop_front();
81       continue;
82     }
83     H = caseFoldingDjbHashCharSlow(Buffer, H);
84   }
85   return H;
86 }
87