1 //===- StringTableBuilder.cpp - String table building utility -------------===//
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 #include "llvm/MC/StringTableBuilder.h"
10 #include "llvm/ADT/ArrayRef.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   case DWARF:
35     Size = 0;
36     break;
37   case MachOLinked:
38   case MachO64Linked:
39     Size = 2;
40     break;
41   case MachO:
42   case MachO64:
43   case ELF:
44     // Start the table with a NUL byte.
45     Size = 1;
46     break;
47   case XCOFF:
48   case WinCOFF:
49     // Make room to write the table size later.
50     Size = 4;
51     break;
52   }
53 }
54 
55 StringTableBuilder::StringTableBuilder(Kind K, unsigned Alignment)
56     : K(K), Alignment(Alignment) {
57   initSize();
58 }
59 
60 void StringTableBuilder::write(raw_ostream &OS) const {
61   assert(isFinalized());
62   SmallString<0> Data;
63   Data.resize(getSize());
64   write((uint8_t *)Data.data());
65   OS << Data;
66 }
67 
68 using StringPair = std::pair<CachedHashStringRef, size_t>;
69 
70 void StringTableBuilder::write(uint8_t *Buf) const {
71   assert(isFinalized());
72   for (const StringPair &P : StringIndexMap) {
73     StringRef Data = P.first.val();
74     if (!Data.empty())
75       memcpy(Buf + P.second, Data.data(), Data.size());
76   }
77   // The COFF formats store the size of the string table in the first 4 bytes.
78   // For Windows, the format is little-endian; for AIX, it is big-endian.
79   if (K == WinCOFF)
80     support::endian::write32le(Buf, Size);
81   else if (K == XCOFF)
82     support::endian::write32be(Buf, Size);
83 }
84 
85 // Returns the character at Pos from end of a string.
86 static int charTailAt(StringPair *P, size_t Pos) {
87   StringRef S = P->first.val();
88   if (Pos >= S.size())
89     return -1;
90   return (unsigned char)S[S.size() - Pos - 1];
91 }
92 
93 // Three-way radix quicksort. This is much faster than std::sort with strcmp
94 // because it does not compare characters that we already know the same.
95 static void multikeySort(MutableArrayRef<StringPair *> Vec, int Pos) {
96 tailcall:
97   if (Vec.size() <= 1)
98     return;
99 
100   // Partition items so that items in [0, I) are greater than the pivot,
101   // [I, J) are the same as the pivot, and [J, Vec.size()) are less than
102   // the pivot.
103   int Pivot = charTailAt(Vec[0], Pos);
104   size_t I = 0;
105   size_t J = Vec.size();
106   for (size_t K = 1; K < J;) {
107     int C = charTailAt(Vec[K], Pos);
108     if (C > Pivot)
109       std::swap(Vec[I++], Vec[K++]);
110     else if (C < Pivot)
111       std::swap(Vec[--J], Vec[K]);
112     else
113       K++;
114   }
115 
116   multikeySort(Vec.slice(0, I), Pos);
117   multikeySort(Vec.slice(J), Pos);
118 
119   // multikeySort(Vec.slice(I, J - I), Pos + 1), but with
120   // tail call optimization.
121   if (Pivot != -1) {
122     Vec = Vec.slice(I, J - I);
123     ++Pos;
124     goto tailcall;
125   }
126 }
127 
128 void StringTableBuilder::finalize() {
129   assert(K != DWARF);
130   finalizeStringTable(/*Optimize=*/true);
131 }
132 
133 void StringTableBuilder::finalizeInOrder() {
134   finalizeStringTable(/*Optimize=*/false);
135 }
136 
137 void StringTableBuilder::finalizeStringTable(bool Optimize) {
138   Finalized = true;
139 
140   if (Optimize) {
141     std::vector<StringPair *> Strings;
142     Strings.reserve(StringIndexMap.size());
143     for (StringPair &P : StringIndexMap)
144       Strings.push_back(&P);
145 
146     multikeySort(Strings, 0);
147     initSize();
148 
149     StringRef Previous;
150     for (StringPair *P : Strings) {
151       StringRef S = P->first.val();
152       if (Previous.endswith(S)) {
153         size_t Pos = Size - S.size() - (K != RAW);
154         if (!(Pos & (Alignment - 1))) {
155           P->second = Pos;
156           continue;
157         }
158       }
159 
160       Size = alignTo(Size, Alignment);
161       P->second = Size;
162 
163       Size += S.size();
164       if (K != RAW)
165         ++Size;
166       Previous = S;
167     }
168   }
169 
170   if (K == MachO || K == MachOLinked)
171     Size = alignTo(Size, 4); // Pad to multiple of 4.
172   if (K == MachO64 || K == MachO64Linked)
173     Size = alignTo(Size, 8); // Pad to multiple of 8.
174 
175   // According to ld64 the string table of a final linked Mach-O binary starts
176   // with " ", i.e. the first byte is ' ' and the second byte is zero. In
177   // 'initSize()' we reserved the first two bytes for holding this string.
178   if (K == MachOLinked || K == MachO64Linked)
179     StringIndexMap[CachedHashStringRef(" ")] = 0;
180 
181   // The first byte in an ELF string table must be null, according to the ELF
182   // specification. In 'initSize()' we reserved the first byte to hold null for
183   // this purpose and here we actually add the string to allow 'getOffset()' to
184   // be called on an empty string.
185   if (K == ELF)
186     StringIndexMap[CachedHashStringRef("")] = 0;
187 }
188 
189 void StringTableBuilder::clear() {
190   Finalized = false;
191   StringIndexMap.clear();
192 }
193 
194 size_t StringTableBuilder::getOffset(CachedHashStringRef S) const {
195   assert(isFinalized());
196   auto I = StringIndexMap.find(S);
197   assert(I != StringIndexMap.end() && "String is not in table!");
198   return I->second;
199 }
200 
201 size_t StringTableBuilder::add(CachedHashStringRef S) {
202   if (K == WinCOFF)
203     assert(S.size() > COFF::NameSize && "Short string in COFF string table!");
204 
205   assert(!isFinalized());
206   auto P = StringIndexMap.insert(std::make_pair(S, 0));
207   if (P.second) {
208     size_t Start = alignTo(Size, Alignment);
209     P.first->second = Start;
210     Size = Start + S.size() + (K != RAW);
211   }
212   return P.first->second;
213 }
214