1//===-- Mangler.cpp - Self-contained c/asm llvm name mangler --------------===//
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// Unified name mangler for assembly backends.
11//
12//===----------------------------------------------------------------------===//
13
14#include "llvm/Target/Mangler.h"
15#include "llvm/DerivedTypes.h"
16#include "llvm/Function.h"
17#include "llvm/Target/TargetData.h"
18#include "llvm/MC/MCAsmInfo.h"
19#include "llvm/MC/MCContext.h"
20#include "llvm/Support/raw_ostream.h"
21#include "llvm/ADT/SmallString.h"
22#include "llvm/ADT/Twine.h"
23using namespace llvm;
24
25static bool isAcceptableChar(char C, bool AllowPeriod, bool AllowUTF8) {
26  if ((C < 'a' || C > 'z') &&
27      (C < 'A' || C > 'Z') &&
28      (C < '0' || C > '9') &&
29      C != '_' && C != '$' && C != '@' &&
30      !(AllowPeriod && C == '.') &&
31      !(AllowUTF8 && (C & 0x80)))
32    return false;
33  return true;
34}
35
36static char HexDigit(int V) {
37  return V < 10 ? V+'0' : V+'A'-10;
38}
39
40static void MangleLetter(SmallVectorImpl<char> &OutName, unsigned char C) {
41  OutName.push_back('_');
42  OutName.push_back(HexDigit(C >> 4));
43  OutName.push_back(HexDigit(C & 15));
44  OutName.push_back('_');
45}
46
47/// NameNeedsEscaping - Return true if the identifier \p Str needs quotes
48/// for this assembler.
49static bool NameNeedsEscaping(StringRef Str, const MCAsmInfo &MAI) {
50  assert(!Str.empty() && "Cannot create an empty MCSymbol");
51
52  // If the first character is a number and the target does not allow this, we
53  // need quotes.
54  if (!MAI.doesAllowNameToStartWithDigit() && Str[0] >= '0' && Str[0] <= '9')
55    return true;
56
57  // If any of the characters in the string is an unacceptable character, force
58  // quotes.
59  bool AllowPeriod = MAI.doesAllowPeriodsInName();
60  bool AllowUTF8 = MAI.doesAllowUTF8();
61  for (unsigned i = 0, e = Str.size(); i != e; ++i)
62    if (!isAcceptableChar(Str[i], AllowPeriod, AllowUTF8))
63      return true;
64  return false;
65}
66
67/// appendMangledName - Add the specified string in mangled form if it uses
68/// any unusual characters.
69static void appendMangledName(SmallVectorImpl<char> &OutName, StringRef Str,
70                              const MCAsmInfo &MAI) {
71  // The first character is not allowed to be a number unless the target
72  // explicitly allows it.
73  if (!MAI.doesAllowNameToStartWithDigit() && Str[0] >= '0' && Str[0] <= '9') {
74    MangleLetter(OutName, Str[0]);
75    Str = Str.substr(1);
76  }
77
78  bool AllowPeriod = MAI.doesAllowPeriodsInName();
79  bool AllowUTF8 = MAI.doesAllowUTF8();
80  for (unsigned i = 0, e = Str.size(); i != e; ++i) {
81    if (!isAcceptableChar(Str[i], AllowPeriod, AllowUTF8))
82      MangleLetter(OutName, Str[i]);
83    else
84      OutName.push_back(Str[i]);
85  }
86}
87
88
89/// appendMangledQuotedName - On systems that support quoted symbols, we still
90/// have to escape some (obscure) characters like " and \n which would break the
91/// assembler's lexing.
92static void appendMangledQuotedName(SmallVectorImpl<char> &OutName,
93                                   StringRef Str) {
94  for (unsigned i = 0, e = Str.size(); i != e; ++i) {
95    if (Str[i] == '"' || Str[i] == '\n')
96      MangleLetter(OutName, Str[i]);
97    else
98      OutName.push_back(Str[i]);
99  }
100}
101
102
103/// getNameWithPrefix - Fill OutName with the name of the appropriate prefix
104/// and the specified name as the global variable name.  GVName must not be
105/// empty.
106void Mangler::getNameWithPrefix(SmallVectorImpl<char> &OutName,
107                                const Twine &GVName, ManglerPrefixTy PrefixTy) {
108  SmallString<256> TmpData;
109  StringRef Name = GVName.toStringRef(TmpData);
110  assert(!Name.empty() && "getNameWithPrefix requires non-empty name");
111
112  const MCAsmInfo &MAI = Context.getAsmInfo();
113
114  // If the global name is not led with \1, add the appropriate prefixes.
115  if (Name[0] == '\1') {
116    Name = Name.substr(1);
117  } else {
118    if (PrefixTy == Mangler::Private) {
119      const char *Prefix = MAI.getPrivateGlobalPrefix();
120      OutName.append(Prefix, Prefix+strlen(Prefix));
121    } else if (PrefixTy == Mangler::LinkerPrivate) {
122      const char *Prefix = MAI.getLinkerPrivateGlobalPrefix();
123      OutName.append(Prefix, Prefix+strlen(Prefix));
124    }
125
126    const char *Prefix = MAI.getGlobalPrefix();
127    if (Prefix[0] == 0)
128      ; // Common noop, no prefix.
129    else if (Prefix[1] == 0)
130      OutName.push_back(Prefix[0]);  // Common, one character prefix.
131    else
132      OutName.append(Prefix, Prefix+strlen(Prefix)); // Arbitrary length prefix.
133  }
134
135  // If this is a simple string that doesn't need escaping, just append it.
136  if (!NameNeedsEscaping(Name, MAI) ||
137      // If quotes are supported, they can be used unless the string contains
138      // a quote or newline.
139      (MAI.doesAllowQuotesInName() &&
140       Name.find_first_of("\n\"") == StringRef::npos)) {
141    OutName.append(Name.begin(), Name.end());
142    return;
143  }
144
145  // On systems that do not allow quoted names, we need to mangle most
146  // strange characters.
147  if (!MAI.doesAllowQuotesInName())
148    return appendMangledName(OutName, Name, MAI);
149
150  // Okay, the system allows quoted strings.  We can quote most anything, the
151  // only characters that need escaping are " and \n.
152  assert(Name.find_first_of("\n\"") != StringRef::npos);
153  return appendMangledQuotedName(OutName, Name);
154}
155
156/// AddFastCallStdCallSuffix - Microsoft fastcall and stdcall functions require
157/// a suffix on their name indicating the number of words of arguments they
158/// take.
159static void AddFastCallStdCallSuffix(SmallVectorImpl<char> &OutName,
160                                     const Function *F, const TargetData &TD) {
161  // Calculate arguments size total.
162  unsigned ArgWords = 0;
163  for (Function::const_arg_iterator AI = F->arg_begin(), AE = F->arg_end();
164       AI != AE; ++AI) {
165    Type *Ty = AI->getType();
166    // 'Dereference' type in case of byval parameter attribute
167    if (AI->hasByValAttr())
168      Ty = cast<PointerType>(Ty)->getElementType();
169    // Size should be aligned to DWORD boundary
170    ArgWords += ((TD.getTypeAllocSize(Ty) + 3)/4)*4;
171  }
172
173  raw_svector_ostream(OutName) << '@' << ArgWords;
174}
175
176
177/// getNameWithPrefix - Fill OutName with the name of the appropriate prefix
178/// and the specified global variable's name.  If the global variable doesn't
179/// have a name, this fills in a unique name for the global.
180void Mangler::getNameWithPrefix(SmallVectorImpl<char> &OutName,
181                                const GlobalValue *GV,
182                                bool isImplicitlyPrivate) {
183  ManglerPrefixTy PrefixTy = Mangler::Default;
184  if (GV->hasPrivateLinkage() || isImplicitlyPrivate)
185    PrefixTy = Mangler::Private;
186  else if (GV->hasLinkerPrivateLinkage() || GV->hasLinkerPrivateWeakLinkage())
187    PrefixTy = Mangler::LinkerPrivate;
188
189  // If this global has a name, handle it simply.
190  if (GV->hasName()) {
191    getNameWithPrefix(OutName, GV->getName(), PrefixTy);
192  } else {
193    // Get the ID for the global, assigning a new one if we haven't got one
194    // already.
195    unsigned &ID = AnonGlobalIDs[GV];
196    if (ID == 0) ID = NextAnonGlobalID++;
197
198    // Must mangle the global into a unique ID.
199    getNameWithPrefix(OutName, "__unnamed_" + Twine(ID), PrefixTy);
200  }
201
202  // If we are supposed to add a microsoft-style suffix for stdcall/fastcall,
203  // add it.
204  if (Context.getAsmInfo().hasMicrosoftFastStdCallMangling()) {
205    if (const Function *F = dyn_cast<Function>(GV)) {
206      CallingConv::ID CC = F->getCallingConv();
207
208      // fastcall functions need to start with @.
209      // FIXME: This logic seems unlikely to be right.
210      if (CC == CallingConv::X86_FastCall) {
211        if (OutName[0] == '_')
212          OutName[0] = '@';
213        else
214          OutName.insert(OutName.begin(), '@');
215      }
216
217      // fastcall and stdcall functions usually need @42 at the end to specify
218      // the argument info.
219      FunctionType *FT = F->getFunctionType();
220      if ((CC == CallingConv::X86_FastCall || CC == CallingConv::X86_StdCall) &&
221          // "Pure" variadic functions do not receive @0 suffix.
222          (!FT->isVarArg() || FT->getNumParams() == 0 ||
223           (FT->getNumParams() == 1 && F->hasStructRetAttr())))
224        AddFastCallStdCallSuffix(OutName, F, TD);
225    }
226  }
227}
228
229/// getSymbol - Return the MCSymbol for the specified global value.  This
230/// symbol is the main label that is the address of the global.
231MCSymbol *Mangler::getSymbol(const GlobalValue *GV) {
232  SmallString<60> NameStr;
233  getNameWithPrefix(NameStr, GV, false);
234  return Context.GetOrCreateSymbol(NameStr.str());
235}
236
237
238