blob: b7bfbbf7bd88bebc2518070e48574ee890f1e63c [file] [log] [blame]
Jonas Devlieghere92ac9d32018-01-28 11:05:10 +00001//===-- 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"
Pavel Labath3b17b842018-02-21 22:36:31 +000015#include "llvm/ADT/ArrayRef.h"
16#include "llvm/Support/Compiler.h"
17#include "llvm/Support/ConvertUTF.h"
18#include "llvm/Support/Unicode.h"
19
20using namespace llvm;
21
22static inline uint32_t djbHashChar(unsigned char C, uint32_t H) {
23 return (H << 5) + H + C;
24}
Jonas Devlieghere92ac9d32018-01-28 11:05:10 +000025
26uint32_t llvm::djbHash(StringRef Buffer, uint32_t H) {
Pavel Labath3b17b842018-02-21 22:36:31 +000027 for (unsigned char C : Buffer.bytes())
28 H = djbHashChar(C, H);
29 return H;
30}
31
32static UTF32 chopOneUTF32(StringRef &Buffer) {
33 UTF32 C;
34 const UTF8 *const Begin8Const =
35 reinterpret_cast<const UTF8 *>(Buffer.begin());
36 const UTF8 *Begin8 = Begin8Const;
37 UTF32 *Begin32 = &C;
38
39 // In lenient mode we will always end up with a "reasonable" value in C for
40 // non-empty input.
41 assert(!Buffer.empty());
42 ConvertUTF8toUTF32(&Begin8, reinterpret_cast<const UTF8 *>(Buffer.end()),
43 &Begin32, &C + 1, lenientConversion);
44 Buffer = Buffer.drop_front(Begin8 - Begin8Const);
45 return C;
46}
47
48static StringRef toUTF8(UTF32 C, MutableArrayRef<UTF8> Storage) {
49 const UTF32 *Begin32 = &C;
50 UTF8 *Begin8 = Storage.begin();
51
52 // The case-folded output should always be a valid unicode character, so use
53 // strict mode here.
54 ConversionResult CR = ConvertUTF32toUTF8(&Begin32, &C + 1, &Begin8,
55 Storage.end(), strictConversion);
56 assert(CR == conversionOK && "Case folding produced invalid char?");
57 (void)CR;
58 return StringRef(reinterpret_cast<char *>(Storage.begin()),
59 Begin8 - Storage.begin());
60}
61
62static UTF32 foldCharDwarf(UTF32 C) {
63 // DWARF v5 addition to the unicode folding rules.
64 // Fold "Latin Small Letter Dotless I" and "Latin Capital Letter I With Dot
65 // Above" into "i".
66 if (C == 0x130 || C == 0x131)
67 return 'i';
68 return sys::unicode::foldCharSimple(C);
69}
70
71static uint32_t caseFoldingDjbHashCharSlow(StringRef &Buffer, uint32_t H) {
72 UTF32 C = chopOneUTF32(Buffer);
73
74 C = foldCharDwarf(C);
75
76 std::array<UTF8, UNI_MAX_UTF8_BYTES_PER_CODE_POINT> Storage;
77 StringRef Folded = toUTF8(C, Storage);
78 return djbHash(Folded, H);
79}
80
81uint32_t llvm::caseFoldingDjbHash(StringRef Buffer, uint32_t H) {
82 while (!Buffer.empty()) {
83 unsigned char C = Buffer.front();
84 if (LLVM_LIKELY(C <= 0x7f)) {
85 // US-ASCII, encoded as one character in utf-8.
86 // This is by far the most common case, so handle this specially.
87 if (C >= 'A' && C <= 'Z')
88 C = 'a' + (C - 'A'); // fold uppercase into lowercase
89 H = djbHashChar(C, H);
90 Buffer = Buffer.drop_front();
91 continue;
92 }
93 H = caseFoldingDjbHashCharSlow(Buffer, H);
94 }
Jonas Devlieghere92ac9d32018-01-28 11:05:10 +000095 return H;
96}