1 //===- StringTableBuilder.cpp - String table building utility -------------===//
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 #include "llvm/MC/StringTableBuilder.h"
11 #include "llvm/ADT/CachedHashString.h"
12 #include "llvm/ADT/SmallString.h"
13 #include "llvm/ADT/StringRef.h"
14 #include "llvm/BinaryFormat/COFF.h"
15 #include "llvm/Support/Endian.h"
16 #include "llvm/Support/MathExtras.h"
17 #include "llvm/Support/raw_ostream.h"
18 #include <cassert>
19 #include <cstddef>
20 #include <cstdint>
21 #include <cstring>
22 #include <utility>
23 #include <vector>
24 
25 using namespace llvm;
26 
27 StringTableBuilder::~StringTableBuilder() = default;
28 
29 void StringTableBuilder::initSize() {
30   // Account for leading bytes in table so that offsets returned from add are
31   // correct.
32   switch (K) {
33   case RAW:
34     Size = 0;
35     break;
36   case MachO:
37   case ELF:
38     // Start the table with a NUL byte.
39     Size = 1;
40     break;
41   case WinCOFF:
42     // Make room to write the table size later.
43     Size = 4;
44     break;
45   }
46 }
47 
48 StringTableBuilder::StringTableBuilder(Kind K, unsigned Alignment)
49     : K(K), Alignment(Alignment) {
50   initSize();
51 }
52 
53 void StringTableBuilder::write(raw_ostream &OS) const {
54   assert(isFinalized());
55   SmallString<0> Data;
56   Data.resize(getSize());
57   write((uint8_t *)Data.data());
58   OS << Data;
59 }
60 
61 using StringPair = std::pair<CachedHashStringRef, size_t>;
62 
63 void StringTableBuilder::write(uint8_t *Buf) const {
64   assert(isFinalized());
65   for (const StringPair &P : StringIndexMap) {
66     StringRef Data = P.first.val();
67     if (!Data.empty())
68       memcpy(Buf + P.second, Data.data(), Data.size());
69   }
70   if (K != WinCOFF)
71     return;
72   support::endian::write32le(Buf, Size);
73 }
74 
75 // Returns the character at Pos from end of a string.
76 static int charTailAt(StringPair *P, size_t Pos) {
77   StringRef S = P->first.val();
78   if (Pos >= S.size())
79     return -1;
80   return (unsigned char)S[S.size() - Pos - 1];
81 }
82 
83 // Three-way radix quicksort. This is much faster than std::sort with strcmp
84 // because it does not compare characters that we already know the same.
85 static void multikey_qsort(std::vector<StringPair *> &Vec, size_t Begin,
86                            size_t End, int Pos) {
87 tailcall:
88   if (End - Begin <= 1)
89     return;
90 
91   // Partition items so that items in [Begin, P) are greater than the pivot,
92   // [P, Q) are the same as the pivot, and [Q, End) are less than the pivot.
93   int Pivot = charTailAt(Vec[Begin], Pos);
94   size_t P = Begin;
95   size_t Q = End;
96   for (size_t R = Begin + 1; R < Q;) {
97     int C = charTailAt(Vec[R], Pos);
98     if (C > Pivot)
99       std::swap(Vec[P++], Vec[R++]);
100     else if (C < Pivot)
101       std::swap(Vec[--Q], Vec[R]);
102     else
103       R++;
104   }
105 
106   multikey_qsort(Vec, Begin, P, Pos);
107   multikey_qsort(Vec, Q, End, Pos);
108   if (Pivot != -1) {
109     // qsort(P, Q, Pos + 1), but with tail call optimization.
110     Begin = P;
111     End = Q;
112     ++Pos;
113     goto tailcall;
114   }
115 }
116 
117 void StringTableBuilder::finalize() {
118   finalizeStringTable(/*Optimize=*/true);
119 }
120 
121 void StringTableBuilder::finalizeInOrder() {
122   finalizeStringTable(/*Optimize=*/false);
123 }
124 
125 void StringTableBuilder::finalizeStringTable(bool Optimize) {
126   Finalized = true;
127 
128   if (Optimize) {
129     std::vector<StringPair *> Strings;
130     Strings.reserve(StringIndexMap.size());
131     for (StringPair &P : StringIndexMap)
132       Strings.push_back(&P);
133 
134     if (!Strings.empty()) {
135       // If we're optimizing, sort by name. If not, sort by previously assigned
136       // offset.
137       multikey_qsort(Strings, 0, Strings.size(), 0);
138     }
139 
140     initSize();
141 
142     StringRef Previous;
143     for (StringPair *P : Strings) {
144       StringRef S = P->first.val();
145       if (Previous.endswith(S)) {
146         size_t Pos = Size - S.size() - (K != RAW);
147         if (!(Pos & (Alignment - 1))) {
148           P->second = Pos;
149           continue;
150         }
151       }
152 
153       Size = alignTo(Size, Alignment);
154       P->second = Size;
155 
156       Size += S.size();
157       if (K != RAW)
158         ++Size;
159       Previous = S;
160     }
161   }
162 
163   if (K == MachO)
164     Size = alignTo(Size, 4); // Pad to multiple of 4.
165 }
166 
167 void StringTableBuilder::clear() {
168   Finalized = false;
169   StringIndexMap.clear();
170 }
171 
172 size_t StringTableBuilder::getOffset(CachedHashStringRef S) const {
173   assert(isFinalized());
174   auto I = StringIndexMap.find(S);
175   assert(I != StringIndexMap.end() && "String is not in table!");
176   return I->second;
177 }
178 
179 size_t StringTableBuilder::add(CachedHashStringRef S) {
180   if (K == WinCOFF)
181     assert(S.size() > COFF::NameSize && "Short string in COFF string table!");
182 
183   assert(!isFinalized());
184   auto P = StringIndexMap.insert(std::make_pair(S, 0));
185   if (P.second) {
186     size_t Start = alignTo(Size, Alignment);
187     P.first->second = Start;
188     Size = Start + S.size() + (K != RAW);
189   }
190   return P.first->second;
191 }
192