Revert r66920. It was causing failures in the self-hosting buildbot (in release
[oota-llvm.git] / lib / Bitcode / Writer / BitcodeWriter.cpp
1 //===--- Bitcode/Writer/BitcodeWriter.cpp - Bitcode Writer ----------------===//
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 // Bitcode writer implementation.
11 //
12 //===----------------------------------------------------------------------===//
13
14 #include "llvm/Bitcode/ReaderWriter.h"
15 #include "llvm/Bitcode/BitstreamWriter.h"
16 #include "llvm/Bitcode/LLVMBitCodes.h"
17 #include "ValueEnumerator.h"
18 #include "llvm/Constants.h"
19 #include "llvm/DerivedTypes.h"
20 #include "llvm/InlineAsm.h"
21 #include "llvm/Instructions.h"
22 #include "llvm/Module.h"
23 #include "llvm/TypeSymbolTable.h"
24 #include "llvm/ValueSymbolTable.h"
25 #include "llvm/Support/MathExtras.h"
26 #include "llvm/Support/Streams.h"
27 #include "llvm/Support/raw_ostream.h"
28 #include "llvm/System/Program.h"
29 using namespace llvm;
30
31 /// These are manifest constants used by the bitcode writer. They do not need to
32 /// be kept in sync with the reader, but need to be consistent within this file.
33 enum {
34   CurVersion = 0,
35   
36   // VALUE_SYMTAB_BLOCK abbrev id's.
37   VST_ENTRY_8_ABBREV = bitc::FIRST_APPLICATION_ABBREV,
38   VST_ENTRY_7_ABBREV,
39   VST_ENTRY_6_ABBREV,
40   VST_BBENTRY_6_ABBREV,
41   
42   // CONSTANTS_BLOCK abbrev id's.
43   CONSTANTS_SETTYPE_ABBREV = bitc::FIRST_APPLICATION_ABBREV,
44   CONSTANTS_INTEGER_ABBREV,
45   CONSTANTS_CE_CAST_Abbrev,
46   CONSTANTS_NULL_Abbrev,
47   
48   // FUNCTION_BLOCK abbrev id's.
49   FUNCTION_INST_LOAD_ABBREV = bitc::FIRST_APPLICATION_ABBREV,
50   FUNCTION_INST_BINOP_ABBREV,
51   FUNCTION_INST_CAST_ABBREV,
52   FUNCTION_INST_RET_VOID_ABBREV,
53   FUNCTION_INST_RET_VAL_ABBREV,
54   FUNCTION_INST_UNREACHABLE_ABBREV
55 };
56
57
58 static unsigned GetEncodedCastOpcode(unsigned Opcode) {
59   switch (Opcode) {
60   default: assert(0 && "Unknown cast instruction!");
61   case Instruction::Trunc   : return bitc::CAST_TRUNC;
62   case Instruction::ZExt    : return bitc::CAST_ZEXT;
63   case Instruction::SExt    : return bitc::CAST_SEXT;
64   case Instruction::FPToUI  : return bitc::CAST_FPTOUI;
65   case Instruction::FPToSI  : return bitc::CAST_FPTOSI;
66   case Instruction::UIToFP  : return bitc::CAST_UITOFP;
67   case Instruction::SIToFP  : return bitc::CAST_SITOFP;
68   case Instruction::FPTrunc : return bitc::CAST_FPTRUNC;
69   case Instruction::FPExt   : return bitc::CAST_FPEXT;
70   case Instruction::PtrToInt: return bitc::CAST_PTRTOINT;
71   case Instruction::IntToPtr: return bitc::CAST_INTTOPTR;
72   case Instruction::BitCast : return bitc::CAST_BITCAST;
73   }
74 }
75
76 static unsigned GetEncodedBinaryOpcode(unsigned Opcode) {
77   switch (Opcode) {
78   default: assert(0 && "Unknown binary instruction!");
79   case Instruction::Add:  return bitc::BINOP_ADD;
80   case Instruction::Sub:  return bitc::BINOP_SUB;
81   case Instruction::Mul:  return bitc::BINOP_MUL;
82   case Instruction::UDiv: return bitc::BINOP_UDIV;
83   case Instruction::FDiv:
84   case Instruction::SDiv: return bitc::BINOP_SDIV;
85   case Instruction::URem: return bitc::BINOP_UREM;
86   case Instruction::FRem:
87   case Instruction::SRem: return bitc::BINOP_SREM;
88   case Instruction::Shl:  return bitc::BINOP_SHL;
89   case Instruction::LShr: return bitc::BINOP_LSHR;
90   case Instruction::AShr: return bitc::BINOP_ASHR;
91   case Instruction::And:  return bitc::BINOP_AND;
92   case Instruction::Or:   return bitc::BINOP_OR;
93   case Instruction::Xor:  return bitc::BINOP_XOR;
94   }
95 }
96
97
98
99 static void WriteStringRecord(unsigned Code, const std::string &Str, 
100                               unsigned AbbrevToUse, BitstreamWriter &Stream) {
101   SmallVector<unsigned, 64> Vals;
102   
103   // Code: [strchar x N]
104   for (unsigned i = 0, e = Str.size(); i != e; ++i)
105     Vals.push_back(Str[i]);
106     
107   // Emit the finished record.
108   Stream.EmitRecord(Code, Vals, AbbrevToUse);
109 }
110
111 // Emit information about parameter attributes.
112 static void WriteAttributeTable(const ValueEnumerator &VE, 
113                                 BitstreamWriter &Stream) {
114   const std::vector<AttrListPtr> &Attrs = VE.getAttributes();
115   if (Attrs.empty()) return;
116   
117   Stream.EnterSubblock(bitc::PARAMATTR_BLOCK_ID, 3);
118
119   SmallVector<uint64_t, 64> Record;
120   for (unsigned i = 0, e = Attrs.size(); i != e; ++i) {
121     const AttrListPtr &A = Attrs[i];
122     for (unsigned i = 0, e = A.getNumSlots(); i != e; ++i) {
123       const AttributeWithIndex &PAWI = A.getSlot(i);
124       Record.push_back(PAWI.Index);
125
126       // FIXME: remove in LLVM 3.0
127       // Store the alignment in the bitcode as a 16-bit raw value instead of a
128       // 5-bit log2 encoded value. Shift the bits above the alignment up by
129       // 11 bits.
130       uint64_t FauxAttr = PAWI.Attrs & 0xffff;
131       if (PAWI.Attrs & Attribute::Alignment)
132         FauxAttr |= (1ull<<16)<<(((PAWI.Attrs & Attribute::Alignment)-1) >> 16);
133       FauxAttr |= (PAWI.Attrs & (0x3FFull << 21)) << 11;
134
135       Record.push_back(FauxAttr);
136     }
137     
138     Stream.EmitRecord(bitc::PARAMATTR_CODE_ENTRY, Record);
139     Record.clear();
140   }
141   
142   Stream.ExitBlock();
143 }
144
145 /// WriteTypeTable - Write out the type table for a module.
146 static void WriteTypeTable(const ValueEnumerator &VE, BitstreamWriter &Stream) {
147   const ValueEnumerator::TypeList &TypeList = VE.getTypes();
148   
149   Stream.EnterSubblock(bitc::TYPE_BLOCK_ID, 4 /*count from # abbrevs */);
150   SmallVector<uint64_t, 64> TypeVals;
151   
152   // Abbrev for TYPE_CODE_POINTER.
153   BitCodeAbbrev *Abbv = new BitCodeAbbrev();
154   Abbv->Add(BitCodeAbbrevOp(bitc::TYPE_CODE_POINTER));
155   Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed,
156                             Log2_32_Ceil(VE.getTypes().size()+1)));
157   Abbv->Add(BitCodeAbbrevOp(0));  // Addrspace = 0
158   unsigned PtrAbbrev = Stream.EmitAbbrev(Abbv);
159   
160   // Abbrev for TYPE_CODE_FUNCTION.
161   Abbv = new BitCodeAbbrev();
162   Abbv->Add(BitCodeAbbrevOp(bitc::TYPE_CODE_FUNCTION));
163   Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 1));  // isvararg
164   Abbv->Add(BitCodeAbbrevOp(0));  // FIXME: DEAD value, remove in LLVM 3.0
165   Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Array));
166   Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed,
167                             Log2_32_Ceil(VE.getTypes().size()+1)));
168   unsigned FunctionAbbrev = Stream.EmitAbbrev(Abbv);
169   
170   // Abbrev for TYPE_CODE_STRUCT.
171   Abbv = new BitCodeAbbrev();
172   Abbv->Add(BitCodeAbbrevOp(bitc::TYPE_CODE_STRUCT));
173   Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 1));  // ispacked
174   Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Array));
175   Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed,
176                             Log2_32_Ceil(VE.getTypes().size()+1)));
177   unsigned StructAbbrev = Stream.EmitAbbrev(Abbv);
178  
179   // Abbrev for TYPE_CODE_ARRAY.
180   Abbv = new BitCodeAbbrev();
181   Abbv->Add(BitCodeAbbrevOp(bitc::TYPE_CODE_ARRAY));
182   Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::VBR, 8));   // size
183   Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed,
184                             Log2_32_Ceil(VE.getTypes().size()+1)));
185   unsigned ArrayAbbrev = Stream.EmitAbbrev(Abbv);
186   
187   // Emit an entry count so the reader can reserve space.
188   TypeVals.push_back(TypeList.size());
189   Stream.EmitRecord(bitc::TYPE_CODE_NUMENTRY, TypeVals);
190   TypeVals.clear();
191   
192   // Loop over all of the types, emitting each in turn.
193   for (unsigned i = 0, e = TypeList.size(); i != e; ++i) {
194     const Type *T = TypeList[i].first;
195     int AbbrevToUse = 0;
196     unsigned Code = 0;
197     
198     switch (T->getTypeID()) {
199     default: assert(0 && "Unknown type!");
200     case Type::VoidTyID:   Code = bitc::TYPE_CODE_VOID;   break;
201     case Type::FloatTyID:  Code = bitc::TYPE_CODE_FLOAT;  break;
202     case Type::DoubleTyID: Code = bitc::TYPE_CODE_DOUBLE; break;
203     case Type::X86_FP80TyID: Code = bitc::TYPE_CODE_X86_FP80; break;
204     case Type::FP128TyID: Code = bitc::TYPE_CODE_FP128; break;
205     case Type::PPC_FP128TyID: Code = bitc::TYPE_CODE_PPC_FP128; break;
206     case Type::LabelTyID:  Code = bitc::TYPE_CODE_LABEL;  break;
207     case Type::OpaqueTyID: Code = bitc::TYPE_CODE_OPAQUE; break;
208     case Type::IntegerTyID:
209       // INTEGER: [width]
210       Code = bitc::TYPE_CODE_INTEGER;
211       TypeVals.push_back(cast<IntegerType>(T)->getBitWidth());
212       break;
213     case Type::PointerTyID: {
214       const PointerType *PTy = cast<PointerType>(T);
215       // POINTER: [pointee type, address space]
216       Code = bitc::TYPE_CODE_POINTER;
217       TypeVals.push_back(VE.getTypeID(PTy->getElementType()));
218       unsigned AddressSpace = PTy->getAddressSpace();
219       TypeVals.push_back(AddressSpace);
220       if (AddressSpace == 0) AbbrevToUse = PtrAbbrev;
221       break;
222     }
223     case Type::FunctionTyID: {
224       const FunctionType *FT = cast<FunctionType>(T);
225       // FUNCTION: [isvararg, attrid, retty, paramty x N]
226       Code = bitc::TYPE_CODE_FUNCTION;
227       TypeVals.push_back(FT->isVarArg());
228       TypeVals.push_back(0);  // FIXME: DEAD: remove in llvm 3.0
229       TypeVals.push_back(VE.getTypeID(FT->getReturnType()));
230       for (unsigned i = 0, e = FT->getNumParams(); i != e; ++i)
231         TypeVals.push_back(VE.getTypeID(FT->getParamType(i)));
232       AbbrevToUse = FunctionAbbrev;
233       break;
234     }
235     case Type::StructTyID: {
236       const StructType *ST = cast<StructType>(T);
237       // STRUCT: [ispacked, eltty x N]
238       Code = bitc::TYPE_CODE_STRUCT;
239       TypeVals.push_back(ST->isPacked());
240       // Output all of the element types.
241       for (StructType::element_iterator I = ST->element_begin(),
242            E = ST->element_end(); I != E; ++I)
243         TypeVals.push_back(VE.getTypeID(*I));
244       AbbrevToUse = StructAbbrev;
245       break;
246     }
247     case Type::ArrayTyID: {
248       const ArrayType *AT = cast<ArrayType>(T);
249       // ARRAY: [numelts, eltty]
250       Code = bitc::TYPE_CODE_ARRAY;
251       TypeVals.push_back(AT->getNumElements());
252       TypeVals.push_back(VE.getTypeID(AT->getElementType()));
253       AbbrevToUse = ArrayAbbrev;
254       break;
255     }
256     case Type::VectorTyID: {
257       const VectorType *VT = cast<VectorType>(T);
258       // VECTOR [numelts, eltty]
259       Code = bitc::TYPE_CODE_VECTOR;
260       TypeVals.push_back(VT->getNumElements());
261       TypeVals.push_back(VE.getTypeID(VT->getElementType()));
262       break;
263     }
264     }
265
266     // Emit the finished record.
267     Stream.EmitRecord(Code, TypeVals, AbbrevToUse);
268     TypeVals.clear();
269   }
270   
271   Stream.ExitBlock();
272 }
273
274 static unsigned getEncodedLinkage(const GlobalValue *GV) {
275   switch (GV->getLinkage()) {
276   default: assert(0 && "Invalid linkage!");
277   case GlobalValue::GhostLinkage:  // Map ghost linkage onto external.
278   case GlobalValue::ExternalLinkage:     return 0;
279   case GlobalValue::WeakAnyLinkage:      return 1;
280   case GlobalValue::AppendingLinkage:    return 2;
281   case GlobalValue::InternalLinkage:     return 3;
282   case GlobalValue::LinkOnceAnyLinkage:  return 4;
283   case GlobalValue::DLLImportLinkage:    return 5;
284   case GlobalValue::DLLExportLinkage:    return 6;
285   case GlobalValue::ExternalWeakLinkage: return 7;
286   case GlobalValue::CommonLinkage:       return 8;
287   case GlobalValue::PrivateLinkage:      return 9;
288   case GlobalValue::WeakODRLinkage:      return 10;
289   case GlobalValue::LinkOnceODRLinkage:  return 11;
290   }
291 }
292
293 static unsigned getEncodedVisibility(const GlobalValue *GV) {
294   switch (GV->getVisibility()) {
295   default: assert(0 && "Invalid visibility!");
296   case GlobalValue::DefaultVisibility:   return 0;
297   case GlobalValue::HiddenVisibility:    return 1;
298   case GlobalValue::ProtectedVisibility: return 2;
299   }
300 }
301
302 // Emit top-level description of module, including target triple, inline asm,
303 // descriptors for global variables, and function prototype info.
304 static void WriteModuleInfo(const Module *M, const ValueEnumerator &VE,
305                             BitstreamWriter &Stream) {
306   // Emit the list of dependent libraries for the Module.
307   for (Module::lib_iterator I = M->lib_begin(), E = M->lib_end(); I != E; ++I)
308     WriteStringRecord(bitc::MODULE_CODE_DEPLIB, *I, 0/*TODO*/, Stream);
309
310   // Emit various pieces of data attached to a module.
311   if (!M->getTargetTriple().empty())
312     WriteStringRecord(bitc::MODULE_CODE_TRIPLE, M->getTargetTriple(),
313                       0/*TODO*/, Stream);
314   if (!M->getDataLayout().empty())
315     WriteStringRecord(bitc::MODULE_CODE_DATALAYOUT, M->getDataLayout(),
316                       0/*TODO*/, Stream);
317   if (!M->getModuleInlineAsm().empty())
318     WriteStringRecord(bitc::MODULE_CODE_ASM, M->getModuleInlineAsm(),
319                       0/*TODO*/, Stream);
320
321   // Emit information about sections and GC, computing how many there are. Also
322   // compute the maximum alignment value.
323   std::map<std::string, unsigned> SectionMap;
324   std::map<std::string, unsigned> GCMap;
325   unsigned MaxAlignment = 0;
326   unsigned MaxGlobalType = 0;
327   for (Module::const_global_iterator GV = M->global_begin(),E = M->global_end();
328        GV != E; ++GV) {
329     MaxAlignment = std::max(MaxAlignment, GV->getAlignment());
330     MaxGlobalType = std::max(MaxGlobalType, VE.getTypeID(GV->getType()));
331     
332     if (!GV->hasSection()) continue;
333     // Give section names unique ID's.
334     unsigned &Entry = SectionMap[GV->getSection()];
335     if (Entry != 0) continue;
336     WriteStringRecord(bitc::MODULE_CODE_SECTIONNAME, GV->getSection(),
337                       0/*TODO*/, Stream);
338     Entry = SectionMap.size();
339   }
340   for (Module::const_iterator F = M->begin(), E = M->end(); F != E; ++F) {
341     MaxAlignment = std::max(MaxAlignment, F->getAlignment());
342     if (F->hasSection()) {
343       // Give section names unique ID's.
344       unsigned &Entry = SectionMap[F->getSection()];
345       if (!Entry) {
346         WriteStringRecord(bitc::MODULE_CODE_SECTIONNAME, F->getSection(),
347                           0/*TODO*/, Stream);
348         Entry = SectionMap.size();
349       }
350     }
351     if (F->hasGC()) {
352       // Same for GC names.
353       unsigned &Entry = GCMap[F->getGC()];
354       if (!Entry) {
355         WriteStringRecord(bitc::MODULE_CODE_GCNAME, F->getGC(),
356                           0/*TODO*/, Stream);
357         Entry = GCMap.size();
358       }
359     }
360   }
361   
362   // Emit abbrev for globals, now that we know # sections and max alignment.
363   unsigned SimpleGVarAbbrev = 0;
364   if (!M->global_empty()) { 
365     // Add an abbrev for common globals with no visibility or thread localness.
366     BitCodeAbbrev *Abbv = new BitCodeAbbrev();
367     Abbv->Add(BitCodeAbbrevOp(bitc::MODULE_CODE_GLOBALVAR));
368     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed,
369                               Log2_32_Ceil(MaxGlobalType+1)));
370     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 1));      // Constant.
371     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::VBR, 6));        // Initializer.
372     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 4));      // Linkage.
373     if (MaxAlignment == 0)                                      // Alignment.
374       Abbv->Add(BitCodeAbbrevOp(0));
375     else {
376       unsigned MaxEncAlignment = Log2_32(MaxAlignment)+1;
377       Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed,
378                                Log2_32_Ceil(MaxEncAlignment+1)));
379     }
380     if (SectionMap.empty())                                    // Section.
381       Abbv->Add(BitCodeAbbrevOp(0));
382     else
383       Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed,
384                                Log2_32_Ceil(SectionMap.size()+1)));
385     // Don't bother emitting vis + thread local.
386     SimpleGVarAbbrev = Stream.EmitAbbrev(Abbv);
387   }
388   
389   // Emit the global variable information.
390   SmallVector<unsigned, 64> Vals;
391   for (Module::const_global_iterator GV = M->global_begin(),E = M->global_end();
392        GV != E; ++GV) {
393     unsigned AbbrevToUse = 0;
394
395     // GLOBALVAR: [type, isconst, initid, 
396     //             linkage, alignment, section, visibility, threadlocal]
397     Vals.push_back(VE.getTypeID(GV->getType()));
398     Vals.push_back(GV->isConstant());
399     Vals.push_back(GV->isDeclaration() ? 0 :
400                    (VE.getValueID(GV->getInitializer()) + 1));
401     Vals.push_back(getEncodedLinkage(GV));
402     Vals.push_back(Log2_32(GV->getAlignment())+1);
403     Vals.push_back(GV->hasSection() ? SectionMap[GV->getSection()] : 0);
404     if (GV->isThreadLocal() || 
405         GV->getVisibility() != GlobalValue::DefaultVisibility) {
406       Vals.push_back(getEncodedVisibility(GV));
407       Vals.push_back(GV->isThreadLocal());
408     } else {
409       AbbrevToUse = SimpleGVarAbbrev;
410     }
411     
412     Stream.EmitRecord(bitc::MODULE_CODE_GLOBALVAR, Vals, AbbrevToUse);
413     Vals.clear();
414   }
415
416   // Emit the function proto information.
417   for (Module::const_iterator F = M->begin(), E = M->end(); F != E; ++F) {
418     // FUNCTION:  [type, callingconv, isproto, paramattr,
419     //             linkage, alignment, section, visibility, gc]
420     Vals.push_back(VE.getTypeID(F->getType()));
421     Vals.push_back(F->getCallingConv());
422     Vals.push_back(F->isDeclaration());
423     Vals.push_back(getEncodedLinkage(F));
424     Vals.push_back(VE.getAttributeID(F->getAttributes()));
425     Vals.push_back(Log2_32(F->getAlignment())+1);
426     Vals.push_back(F->hasSection() ? SectionMap[F->getSection()] : 0);
427     Vals.push_back(getEncodedVisibility(F));
428     Vals.push_back(F->hasGC() ? GCMap[F->getGC()] : 0);
429     
430     unsigned AbbrevToUse = 0;
431     Stream.EmitRecord(bitc::MODULE_CODE_FUNCTION, Vals, AbbrevToUse);
432     Vals.clear();
433   }
434   
435   
436   // Emit the alias information.
437   for (Module::const_alias_iterator AI = M->alias_begin(), E = M->alias_end();
438        AI != E; ++AI) {
439     Vals.push_back(VE.getTypeID(AI->getType()));
440     Vals.push_back(VE.getValueID(AI->getAliasee()));
441     Vals.push_back(getEncodedLinkage(AI));
442     Vals.push_back(getEncodedVisibility(AI));
443     unsigned AbbrevToUse = 0;
444     Stream.EmitRecord(bitc::MODULE_CODE_ALIAS, Vals, AbbrevToUse);
445     Vals.clear();
446   }
447 }
448
449
450 static void WriteConstants(unsigned FirstVal, unsigned LastVal,
451                            const ValueEnumerator &VE,
452                            BitstreamWriter &Stream, bool isGlobal) {
453   if (FirstVal == LastVal) return;
454   
455   Stream.EnterSubblock(bitc::CONSTANTS_BLOCK_ID, 4);
456
457   unsigned AggregateAbbrev = 0;
458   unsigned String8Abbrev = 0;
459   unsigned CString7Abbrev = 0;
460   unsigned CString6Abbrev = 0;
461   // If this is a constant pool for the module, emit module-specific abbrevs.
462   if (isGlobal) {
463     // Abbrev for CST_CODE_AGGREGATE.
464     BitCodeAbbrev *Abbv = new BitCodeAbbrev();
465     Abbv->Add(BitCodeAbbrevOp(bitc::CST_CODE_AGGREGATE));
466     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Array));
467     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, Log2_32_Ceil(LastVal+1)));
468     AggregateAbbrev = Stream.EmitAbbrev(Abbv);
469
470     // Abbrev for CST_CODE_STRING.
471     Abbv = new BitCodeAbbrev();
472     Abbv->Add(BitCodeAbbrevOp(bitc::CST_CODE_STRING));
473     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Array));
474     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 8));
475     String8Abbrev = Stream.EmitAbbrev(Abbv);
476     // Abbrev for CST_CODE_CSTRING.
477     Abbv = new BitCodeAbbrev();
478     Abbv->Add(BitCodeAbbrevOp(bitc::CST_CODE_CSTRING));
479     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Array));
480     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 7));
481     CString7Abbrev = Stream.EmitAbbrev(Abbv);
482     // Abbrev for CST_CODE_CSTRING.
483     Abbv = new BitCodeAbbrev();
484     Abbv->Add(BitCodeAbbrevOp(bitc::CST_CODE_CSTRING));
485     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Array));
486     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Char6));
487     CString6Abbrev = Stream.EmitAbbrev(Abbv);
488   }  
489   
490   SmallVector<uint64_t, 64> Record;
491
492   const ValueEnumerator::ValueList &Vals = VE.getValues();
493   const Type *LastTy = 0;
494   for (unsigned i = FirstVal; i != LastVal; ++i) {
495     const Value *V = Vals[i].first;
496     // If we need to switch types, do so now.
497     if (V->getType() != LastTy) {
498       LastTy = V->getType();
499       Record.push_back(VE.getTypeID(LastTy));
500       Stream.EmitRecord(bitc::CST_CODE_SETTYPE, Record,
501                         CONSTANTS_SETTYPE_ABBREV);
502       Record.clear();
503     }
504     
505     if (const InlineAsm *IA = dyn_cast<InlineAsm>(V)) {
506       Record.push_back(unsigned(IA->hasSideEffects()));
507       
508       // Add the asm string.
509       const std::string &AsmStr = IA->getAsmString();
510       Record.push_back(AsmStr.size());
511       for (unsigned i = 0, e = AsmStr.size(); i != e; ++i)
512         Record.push_back(AsmStr[i]);
513       
514       // Add the constraint string.
515       const std::string &ConstraintStr = IA->getConstraintString();
516       Record.push_back(ConstraintStr.size());
517       for (unsigned i = 0, e = ConstraintStr.size(); i != e; ++i)
518         Record.push_back(ConstraintStr[i]);
519       Stream.EmitRecord(bitc::CST_CODE_INLINEASM, Record);
520       Record.clear();
521       continue;
522     }
523     const Constant *C = cast<Constant>(V);
524     unsigned Code = -1U;
525     unsigned AbbrevToUse = 0;
526     if (C->isNullValue()) {
527       Code = bitc::CST_CODE_NULL;
528     } else if (isa<UndefValue>(C)) {
529       Code = bitc::CST_CODE_UNDEF;
530     } else if (const ConstantInt *IV = dyn_cast<ConstantInt>(C)) {
531       if (IV->getBitWidth() <= 64) {
532         int64_t V = IV->getSExtValue();
533         if (V >= 0)
534           Record.push_back(V << 1);
535         else
536           Record.push_back((-V << 1) | 1);
537         Code = bitc::CST_CODE_INTEGER;
538         AbbrevToUse = CONSTANTS_INTEGER_ABBREV;
539       } else {                             // Wide integers, > 64 bits in size.
540         // We have an arbitrary precision integer value to write whose 
541         // bit width is > 64. However, in canonical unsigned integer 
542         // format it is likely that the high bits are going to be zero.
543         // So, we only write the number of active words.
544         unsigned NWords = IV->getValue().getActiveWords(); 
545         const uint64_t *RawWords = IV->getValue().getRawData();
546         for (unsigned i = 0; i != NWords; ++i) {
547           int64_t V = RawWords[i];
548           if (V >= 0)
549             Record.push_back(V << 1);
550           else
551             Record.push_back((-V << 1) | 1);
552         }
553         Code = bitc::CST_CODE_WIDE_INTEGER;
554       }
555     } else if (const ConstantFP *CFP = dyn_cast<ConstantFP>(C)) {
556       Code = bitc::CST_CODE_FLOAT;
557       const Type *Ty = CFP->getType();
558       if (Ty == Type::FloatTy || Ty == Type::DoubleTy) {
559         Record.push_back(CFP->getValueAPF().bitcastToAPInt().getZExtValue());
560       } else if (Ty == Type::X86_FP80Ty) {
561         // api needed to prevent premature destruction
562         APInt api = CFP->getValueAPF().bitcastToAPInt();
563         const uint64_t *p = api.getRawData();
564         Record.push_back(p[0]);
565         Record.push_back((uint16_t)p[1]);
566       } else if (Ty == Type::FP128Ty || Ty == Type::PPC_FP128Ty) {
567         APInt api = CFP->getValueAPF().bitcastToAPInt();
568         const uint64_t *p = api.getRawData();
569         Record.push_back(p[0]);
570         Record.push_back(p[1]);
571       } else {
572         assert (0 && "Unknown FP type!");
573       }
574     } else if (isa<ConstantArray>(C) && cast<ConstantArray>(C)->isString()) {
575       // Emit constant strings specially.
576       unsigned NumOps = C->getNumOperands();
577       // If this is a null-terminated string, use the denser CSTRING encoding.
578       if (C->getOperand(NumOps-1)->isNullValue()) {
579         Code = bitc::CST_CODE_CSTRING;
580         --NumOps;  // Don't encode the null, which isn't allowed by char6.
581       } else {
582         Code = bitc::CST_CODE_STRING;
583         AbbrevToUse = String8Abbrev;
584       }
585       bool isCStr7 = Code == bitc::CST_CODE_CSTRING;
586       bool isCStrChar6 = Code == bitc::CST_CODE_CSTRING;
587       for (unsigned i = 0; i != NumOps; ++i) {
588         unsigned char V = cast<ConstantInt>(C->getOperand(i))->getZExtValue();
589         Record.push_back(V);
590         isCStr7 &= (V & 128) == 0;
591         if (isCStrChar6) 
592           isCStrChar6 = BitCodeAbbrevOp::isChar6(V);
593       }
594       
595       if (isCStrChar6)
596         AbbrevToUse = CString6Abbrev;
597       else if (isCStr7)
598         AbbrevToUse = CString7Abbrev;
599     } else if (isa<ConstantArray>(C) || isa<ConstantStruct>(V) ||
600                isa<ConstantVector>(V)) {
601       Code = bitc::CST_CODE_AGGREGATE;
602       for (unsigned i = 0, e = C->getNumOperands(); i != e; ++i)
603         Record.push_back(VE.getValueID(C->getOperand(i)));
604       AbbrevToUse = AggregateAbbrev;
605     } else if (const ConstantExpr *CE = dyn_cast<ConstantExpr>(C)) {
606       switch (CE->getOpcode()) {
607       default:
608         if (Instruction::isCast(CE->getOpcode())) {
609           Code = bitc::CST_CODE_CE_CAST;
610           Record.push_back(GetEncodedCastOpcode(CE->getOpcode()));
611           Record.push_back(VE.getTypeID(C->getOperand(0)->getType()));
612           Record.push_back(VE.getValueID(C->getOperand(0)));
613           AbbrevToUse = CONSTANTS_CE_CAST_Abbrev;
614         } else {
615           assert(CE->getNumOperands() == 2 && "Unknown constant expr!");
616           Code = bitc::CST_CODE_CE_BINOP;
617           Record.push_back(GetEncodedBinaryOpcode(CE->getOpcode()));
618           Record.push_back(VE.getValueID(C->getOperand(0)));
619           Record.push_back(VE.getValueID(C->getOperand(1)));
620         }
621         break;
622       case Instruction::GetElementPtr:
623         Code = bitc::CST_CODE_CE_GEP;
624         for (unsigned i = 0, e = CE->getNumOperands(); i != e; ++i) {
625           Record.push_back(VE.getTypeID(C->getOperand(i)->getType()));
626           Record.push_back(VE.getValueID(C->getOperand(i)));
627         }
628         break;
629       case Instruction::Select:
630         Code = bitc::CST_CODE_CE_SELECT;
631         Record.push_back(VE.getValueID(C->getOperand(0)));
632         Record.push_back(VE.getValueID(C->getOperand(1)));
633         Record.push_back(VE.getValueID(C->getOperand(2)));
634         break;
635       case Instruction::ExtractElement:
636         Code = bitc::CST_CODE_CE_EXTRACTELT;
637         Record.push_back(VE.getTypeID(C->getOperand(0)->getType()));
638         Record.push_back(VE.getValueID(C->getOperand(0)));
639         Record.push_back(VE.getValueID(C->getOperand(1)));
640         break;
641       case Instruction::InsertElement:
642         Code = bitc::CST_CODE_CE_INSERTELT;
643         Record.push_back(VE.getValueID(C->getOperand(0)));
644         Record.push_back(VE.getValueID(C->getOperand(1)));
645         Record.push_back(VE.getValueID(C->getOperand(2)));
646         break;
647       case Instruction::ShuffleVector:
648         // If the return type and argument types are the same, this is a
649         // standard shufflevector instruction.  If the types are different,
650         // then the shuffle is widening or truncating the input vectors, and
651         // the argument type must also be encoded.
652         if (C->getType() == C->getOperand(0)->getType()) {
653           Code = bitc::CST_CODE_CE_SHUFFLEVEC;
654         } else {
655           Code = bitc::CST_CODE_CE_SHUFVEC_EX;
656           Record.push_back(VE.getTypeID(C->getOperand(0)->getType()));
657         }
658         Record.push_back(VE.getValueID(C->getOperand(0)));
659         Record.push_back(VE.getValueID(C->getOperand(1)));
660         Record.push_back(VE.getValueID(C->getOperand(2)));
661         break;
662       case Instruction::ICmp:
663       case Instruction::FCmp:
664       case Instruction::VICmp:
665       case Instruction::VFCmp:
666         if (isa<VectorType>(C->getOperand(0)->getType())
667             && (CE->getOpcode() == Instruction::ICmp
668                 || CE->getOpcode() == Instruction::FCmp)) {
669           // compare returning vector of Int1Ty
670           assert(0 && "Unsupported constant!");
671         } else {
672           Code = bitc::CST_CODE_CE_CMP;
673         }
674         Record.push_back(VE.getTypeID(C->getOperand(0)->getType()));
675         Record.push_back(VE.getValueID(C->getOperand(0)));
676         Record.push_back(VE.getValueID(C->getOperand(1)));
677         Record.push_back(CE->getPredicate());
678         break;
679       }
680     } else {
681       assert(0 && "Unknown constant!");
682     }
683     Stream.EmitRecord(Code, Record, AbbrevToUse);
684     Record.clear();
685   }
686
687   Stream.ExitBlock();
688 }
689
690 static void WriteModuleConstants(const ValueEnumerator &VE,
691                                  BitstreamWriter &Stream) {
692   const ValueEnumerator::ValueList &Vals = VE.getValues();
693   
694   // Find the first constant to emit, which is the first non-globalvalue value.
695   // We know globalvalues have been emitted by WriteModuleInfo.
696   for (unsigned i = 0, e = Vals.size(); i != e; ++i) {
697     if (!isa<GlobalValue>(Vals[i].first)) {
698       WriteConstants(i, Vals.size(), VE, Stream, true);
699       return;
700     }
701   }
702 }
703
704 /// PushValueAndType - The file has to encode both the value and type id for
705 /// many values, because we need to know what type to create for forward
706 /// references.  However, most operands are not forward references, so this type
707 /// field is not needed.
708 ///
709 /// This function adds V's value ID to Vals.  If the value ID is higher than the
710 /// instruction ID, then it is a forward reference, and it also includes the
711 /// type ID.
712 static bool PushValueAndType(const Value *V, unsigned InstID,
713                              SmallVector<unsigned, 64> &Vals, 
714                              ValueEnumerator &VE) {
715   unsigned ValID = VE.getValueID(V);
716   Vals.push_back(ValID);
717   if (ValID >= InstID) {
718     Vals.push_back(VE.getTypeID(V->getType()));
719     return true;
720   }
721   return false;
722 }
723
724 /// WriteInstruction - Emit an instruction to the specified stream.
725 static void WriteInstruction(const Instruction &I, unsigned InstID,
726                              ValueEnumerator &VE, BitstreamWriter &Stream,
727                              SmallVector<unsigned, 64> &Vals) {
728   unsigned Code = 0;
729   unsigned AbbrevToUse = 0;
730   switch (I.getOpcode()) {
731   default:
732     if (Instruction::isCast(I.getOpcode())) {
733       Code = bitc::FUNC_CODE_INST_CAST;
734       if (!PushValueAndType(I.getOperand(0), InstID, Vals, VE))
735         AbbrevToUse = FUNCTION_INST_CAST_ABBREV;
736       Vals.push_back(VE.getTypeID(I.getType()));
737       Vals.push_back(GetEncodedCastOpcode(I.getOpcode()));
738     } else {
739       assert(isa<BinaryOperator>(I) && "Unknown instruction!");
740       Code = bitc::FUNC_CODE_INST_BINOP;
741       if (!PushValueAndType(I.getOperand(0), InstID, Vals, VE))
742         AbbrevToUse = FUNCTION_INST_BINOP_ABBREV;
743       Vals.push_back(VE.getValueID(I.getOperand(1)));
744       Vals.push_back(GetEncodedBinaryOpcode(I.getOpcode()));
745     }
746     break;
747
748   case Instruction::GetElementPtr:
749     Code = bitc::FUNC_CODE_INST_GEP;
750     for (unsigned i = 0, e = I.getNumOperands(); i != e; ++i)
751       PushValueAndType(I.getOperand(i), InstID, Vals, VE);
752     break;
753   case Instruction::ExtractValue: {
754     Code = bitc::FUNC_CODE_INST_EXTRACTVAL;
755     PushValueAndType(I.getOperand(0), InstID, Vals, VE);
756     const ExtractValueInst *EVI = cast<ExtractValueInst>(&I);
757     for (const unsigned *i = EVI->idx_begin(), *e = EVI->idx_end(); i != e; ++i)
758       Vals.push_back(*i);
759     break;
760   }
761   case Instruction::InsertValue: {
762     Code = bitc::FUNC_CODE_INST_INSERTVAL;
763     PushValueAndType(I.getOperand(0), InstID, Vals, VE);
764     PushValueAndType(I.getOperand(1), InstID, Vals, VE);
765     const InsertValueInst *IVI = cast<InsertValueInst>(&I);
766     for (const unsigned *i = IVI->idx_begin(), *e = IVI->idx_end(); i != e; ++i)
767       Vals.push_back(*i);
768     break;
769   }
770   case Instruction::Select:
771     Code = bitc::FUNC_CODE_INST_VSELECT;
772     PushValueAndType(I.getOperand(1), InstID, Vals, VE);
773     Vals.push_back(VE.getValueID(I.getOperand(2)));
774     PushValueAndType(I.getOperand(0), InstID, Vals, VE);
775     break;
776   case Instruction::ExtractElement:
777     Code = bitc::FUNC_CODE_INST_EXTRACTELT;
778     PushValueAndType(I.getOperand(0), InstID, Vals, VE);
779     Vals.push_back(VE.getValueID(I.getOperand(1)));
780     break;
781   case Instruction::InsertElement:
782     Code = bitc::FUNC_CODE_INST_INSERTELT;
783     PushValueAndType(I.getOperand(0), InstID, Vals, VE);
784     Vals.push_back(VE.getValueID(I.getOperand(1)));
785     Vals.push_back(VE.getValueID(I.getOperand(2)));
786     break;
787   case Instruction::ShuffleVector:
788     Code = bitc::FUNC_CODE_INST_SHUFFLEVEC;
789     PushValueAndType(I.getOperand(0), InstID, Vals, VE);
790     Vals.push_back(VE.getValueID(I.getOperand(1)));
791     Vals.push_back(VE.getValueID(I.getOperand(2)));
792     break;
793   case Instruction::ICmp:
794   case Instruction::FCmp:
795   case Instruction::VICmp:
796   case Instruction::VFCmp:
797     if (I.getOpcode() == Instruction::ICmp
798         || I.getOpcode() == Instruction::FCmp) {
799       // compare returning Int1Ty or vector of Int1Ty
800       Code = bitc::FUNC_CODE_INST_CMP2;
801     } else {
802       Code = bitc::FUNC_CODE_INST_CMP;
803     }
804     PushValueAndType(I.getOperand(0), InstID, Vals, VE);
805     Vals.push_back(VE.getValueID(I.getOperand(1)));
806     Vals.push_back(cast<CmpInst>(I).getPredicate());
807     break;
808
809   case Instruction::Ret: 
810     {
811       Code = bitc::FUNC_CODE_INST_RET;
812       unsigned NumOperands = I.getNumOperands();
813       if (NumOperands == 0)
814         AbbrevToUse = FUNCTION_INST_RET_VOID_ABBREV;
815       else if (NumOperands == 1) {
816         if (!PushValueAndType(I.getOperand(0), InstID, Vals, VE))
817           AbbrevToUse = FUNCTION_INST_RET_VAL_ABBREV;
818       } else {
819         for (unsigned i = 0, e = NumOperands; i != e; ++i)
820           PushValueAndType(I.getOperand(i), InstID, Vals, VE);
821       }
822     }
823     break;
824   case Instruction::Br:
825     {
826       Code = bitc::FUNC_CODE_INST_BR;
827       BranchInst &II(cast<BranchInst>(I));
828       Vals.push_back(VE.getValueID(II.getSuccessor(0)));
829       if (II.isConditional()) {
830         Vals.push_back(VE.getValueID(II.getSuccessor(1)));
831         Vals.push_back(VE.getValueID(II.getCondition()));
832       }
833     }
834     break;
835   case Instruction::Switch:
836     Code = bitc::FUNC_CODE_INST_SWITCH;
837     Vals.push_back(VE.getTypeID(I.getOperand(0)->getType()));
838     for (unsigned i = 0, e = I.getNumOperands(); i != e; ++i)
839       Vals.push_back(VE.getValueID(I.getOperand(i)));
840     break;
841   case Instruction::Invoke: {
842     const InvokeInst *II = cast<InvokeInst>(&I);
843     const Value *Callee(II->getCalledValue());
844     const PointerType *PTy = cast<PointerType>(Callee->getType());
845     const FunctionType *FTy = cast<FunctionType>(PTy->getElementType());
846     Code = bitc::FUNC_CODE_INST_INVOKE;
847     
848     Vals.push_back(VE.getAttributeID(II->getAttributes()));
849     Vals.push_back(II->getCallingConv());
850     Vals.push_back(VE.getValueID(II->getNormalDest()));
851     Vals.push_back(VE.getValueID(II->getUnwindDest()));
852     PushValueAndType(Callee, InstID, Vals, VE);
853     
854     // Emit value #'s for the fixed parameters.
855     for (unsigned i = 0, e = FTy->getNumParams(); i != e; ++i)
856       Vals.push_back(VE.getValueID(I.getOperand(i+3)));  // fixed param.
857
858     // Emit type/value pairs for varargs params.
859     if (FTy->isVarArg()) {
860       for (unsigned i = 3+FTy->getNumParams(), e = I.getNumOperands();
861            i != e; ++i)
862         PushValueAndType(I.getOperand(i), InstID, Vals, VE); // vararg
863     }
864     break;
865   }
866   case Instruction::Unwind:
867     Code = bitc::FUNC_CODE_INST_UNWIND;
868     break;
869   case Instruction::Unreachable:
870     Code = bitc::FUNC_CODE_INST_UNREACHABLE;
871     AbbrevToUse = FUNCTION_INST_UNREACHABLE_ABBREV;
872     break;
873   
874   case Instruction::PHI:
875     Code = bitc::FUNC_CODE_INST_PHI;
876     Vals.push_back(VE.getTypeID(I.getType()));
877     for (unsigned i = 0, e = I.getNumOperands(); i != e; ++i)
878       Vals.push_back(VE.getValueID(I.getOperand(i)));
879     break;
880     
881   case Instruction::Malloc:
882     Code = bitc::FUNC_CODE_INST_MALLOC;
883     Vals.push_back(VE.getTypeID(I.getType()));
884     Vals.push_back(VE.getValueID(I.getOperand(0))); // size.
885     Vals.push_back(Log2_32(cast<MallocInst>(I).getAlignment())+1);
886     break;
887     
888   case Instruction::Free:
889     Code = bitc::FUNC_CODE_INST_FREE;
890     PushValueAndType(I.getOperand(0), InstID, Vals, VE);
891     break;
892     
893   case Instruction::Alloca:
894     Code = bitc::FUNC_CODE_INST_ALLOCA;
895     Vals.push_back(VE.getTypeID(I.getType()));
896     Vals.push_back(VE.getValueID(I.getOperand(0))); // size.
897     Vals.push_back(Log2_32(cast<AllocaInst>(I).getAlignment())+1);
898     break;
899     
900   case Instruction::Load:
901     Code = bitc::FUNC_CODE_INST_LOAD;
902     if (!PushValueAndType(I.getOperand(0), InstID, Vals, VE))  // ptr
903       AbbrevToUse = FUNCTION_INST_LOAD_ABBREV;
904       
905     Vals.push_back(Log2_32(cast<LoadInst>(I).getAlignment())+1);
906     Vals.push_back(cast<LoadInst>(I).isVolatile());
907     break;
908   case Instruction::Store:
909     Code = bitc::FUNC_CODE_INST_STORE2;
910     PushValueAndType(I.getOperand(1), InstID, Vals, VE);  // ptrty + ptr
911     Vals.push_back(VE.getValueID(I.getOperand(0)));       // val.
912     Vals.push_back(Log2_32(cast<StoreInst>(I).getAlignment())+1);
913     Vals.push_back(cast<StoreInst>(I).isVolatile());
914     break;
915   case Instruction::Call: {
916     const PointerType *PTy = cast<PointerType>(I.getOperand(0)->getType());
917     const FunctionType *FTy = cast<FunctionType>(PTy->getElementType());
918
919     Code = bitc::FUNC_CODE_INST_CALL;
920     
921     const CallInst *CI = cast<CallInst>(&I);
922     Vals.push_back(VE.getAttributeID(CI->getAttributes()));
923     Vals.push_back((CI->getCallingConv() << 1) | unsigned(CI->isTailCall()));
924     PushValueAndType(CI->getOperand(0), InstID, Vals, VE);  // Callee
925     
926     // Emit value #'s for the fixed parameters.
927     for (unsigned i = 0, e = FTy->getNumParams(); i != e; ++i)
928       Vals.push_back(VE.getValueID(I.getOperand(i+1)));  // fixed param.
929       
930     // Emit type/value pairs for varargs params.
931     if (FTy->isVarArg()) {
932       unsigned NumVarargs = I.getNumOperands()-1-FTy->getNumParams();
933       for (unsigned i = I.getNumOperands()-NumVarargs, e = I.getNumOperands();
934            i != e; ++i)
935         PushValueAndType(I.getOperand(i), InstID, Vals, VE);  // varargs
936     }
937     break;
938   }
939   case Instruction::VAArg:
940     Code = bitc::FUNC_CODE_INST_VAARG;
941     Vals.push_back(VE.getTypeID(I.getOperand(0)->getType()));   // valistty
942     Vals.push_back(VE.getValueID(I.getOperand(0))); // valist.
943     Vals.push_back(VE.getTypeID(I.getType())); // restype.
944     break;
945   }
946   
947   Stream.EmitRecord(Code, Vals, AbbrevToUse);
948   Vals.clear();
949 }
950
951 // Emit names for globals/functions etc.
952 static void WriteValueSymbolTable(const ValueSymbolTable &VST,
953                                   const ValueEnumerator &VE,
954                                   BitstreamWriter &Stream) {
955   if (VST.empty()) return;
956   Stream.EnterSubblock(bitc::VALUE_SYMTAB_BLOCK_ID, 4);
957
958   // FIXME: Set up the abbrev, we know how many values there are!
959   // FIXME: We know if the type names can use 7-bit ascii.
960   SmallVector<unsigned, 64> NameVals;
961   
962   for (ValueSymbolTable::const_iterator SI = VST.begin(), SE = VST.end();
963        SI != SE; ++SI) {
964     
965     const ValueName &Name = *SI;
966     
967     // Figure out the encoding to use for the name.
968     bool is7Bit = true;
969     bool isChar6 = true;
970     for (const char *C = Name.getKeyData(), *E = C+Name.getKeyLength();
971          C != E; ++C) {
972       if (isChar6) 
973         isChar6 = BitCodeAbbrevOp::isChar6(*C);
974       if ((unsigned char)*C & 128) {
975         is7Bit = false;
976         break;  // don't bother scanning the rest.
977       }
978     }
979     
980     unsigned AbbrevToUse = VST_ENTRY_8_ABBREV;
981     
982     // VST_ENTRY:   [valueid, namechar x N]
983     // VST_BBENTRY: [bbid, namechar x N]
984     unsigned Code;
985     if (isa<BasicBlock>(SI->getValue())) {
986       Code = bitc::VST_CODE_BBENTRY;
987       if (isChar6)
988         AbbrevToUse = VST_BBENTRY_6_ABBREV;
989     } else {
990       Code = bitc::VST_CODE_ENTRY;
991       if (isChar6)
992         AbbrevToUse = VST_ENTRY_6_ABBREV;
993       else if (is7Bit)
994         AbbrevToUse = VST_ENTRY_7_ABBREV;
995     }
996     
997     NameVals.push_back(VE.getValueID(SI->getValue()));
998     for (const char *P = Name.getKeyData(),
999          *E = Name.getKeyData()+Name.getKeyLength(); P != E; ++P)
1000       NameVals.push_back((unsigned char)*P);
1001     
1002     // Emit the finished record.
1003     Stream.EmitRecord(Code, NameVals, AbbrevToUse);
1004     NameVals.clear();
1005   }
1006   Stream.ExitBlock();
1007 }
1008
1009 /// WriteFunction - Emit a function body to the module stream.
1010 static void WriteFunction(const Function &F, ValueEnumerator &VE, 
1011                           BitstreamWriter &Stream) {
1012   Stream.EnterSubblock(bitc::FUNCTION_BLOCK_ID, 4);
1013   VE.incorporateFunction(F);
1014
1015   SmallVector<unsigned, 64> Vals;
1016   
1017   // Emit the number of basic blocks, so the reader can create them ahead of
1018   // time.
1019   Vals.push_back(VE.getBasicBlocks().size());
1020   Stream.EmitRecord(bitc::FUNC_CODE_DECLAREBLOCKS, Vals);
1021   Vals.clear();
1022   
1023   // If there are function-local constants, emit them now.
1024   unsigned CstStart, CstEnd;
1025   VE.getFunctionConstantRange(CstStart, CstEnd);
1026   WriteConstants(CstStart, CstEnd, VE, Stream, false);
1027   
1028   // Keep a running idea of what the instruction ID is. 
1029   unsigned InstID = CstEnd;
1030   
1031   // Finally, emit all the instructions, in order.
1032   for (Function::const_iterator BB = F.begin(), E = F.end(); BB != E; ++BB)
1033     for (BasicBlock::const_iterator I = BB->begin(), E = BB->end();
1034          I != E; ++I) {
1035       WriteInstruction(*I, InstID, VE, Stream, Vals);
1036       if (I->getType() != Type::VoidTy)
1037         ++InstID;
1038     }
1039   
1040   // Emit names for all the instructions etc.
1041   WriteValueSymbolTable(F.getValueSymbolTable(), VE, Stream);
1042     
1043   VE.purgeFunction();
1044   Stream.ExitBlock();
1045 }
1046
1047 /// WriteTypeSymbolTable - Emit a block for the specified type symtab.
1048 static void WriteTypeSymbolTable(const TypeSymbolTable &TST,
1049                                  const ValueEnumerator &VE,
1050                                  BitstreamWriter &Stream) {
1051   if (TST.empty()) return;
1052   
1053   Stream.EnterSubblock(bitc::TYPE_SYMTAB_BLOCK_ID, 3);
1054   
1055   // 7-bit fixed width VST_CODE_ENTRY strings.
1056   BitCodeAbbrev *Abbv = new BitCodeAbbrev();
1057   Abbv->Add(BitCodeAbbrevOp(bitc::VST_CODE_ENTRY));
1058   Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed,
1059                             Log2_32_Ceil(VE.getTypes().size()+1)));
1060   Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Array));
1061   Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 7));
1062   unsigned V7Abbrev = Stream.EmitAbbrev(Abbv);
1063   
1064   SmallVector<unsigned, 64> NameVals;
1065   
1066   for (TypeSymbolTable::const_iterator TI = TST.begin(), TE = TST.end(); 
1067        TI != TE; ++TI) {
1068     // TST_ENTRY: [typeid, namechar x N]
1069     NameVals.push_back(VE.getTypeID(TI->second));
1070     
1071     const std::string &Str = TI->first;
1072     bool is7Bit = true;
1073     for (unsigned i = 0, e = Str.size(); i != e; ++i) {
1074       NameVals.push_back((unsigned char)Str[i]);
1075       if (Str[i] & 128)
1076         is7Bit = false;
1077     }
1078     
1079     // Emit the finished record.
1080     Stream.EmitRecord(bitc::VST_CODE_ENTRY, NameVals, is7Bit ? V7Abbrev : 0);
1081     NameVals.clear();
1082   }
1083   
1084   Stream.ExitBlock();
1085 }
1086
1087 // Emit blockinfo, which defines the standard abbreviations etc.
1088 static void WriteBlockInfo(const ValueEnumerator &VE, BitstreamWriter &Stream) {
1089   // We only want to emit block info records for blocks that have multiple
1090   // instances: CONSTANTS_BLOCK, FUNCTION_BLOCK and VALUE_SYMTAB_BLOCK.  Other
1091   // blocks can defined their abbrevs inline.
1092   Stream.EnterBlockInfoBlock(2);
1093   
1094   { // 8-bit fixed-width VST_ENTRY/VST_BBENTRY strings.
1095     BitCodeAbbrev *Abbv = new BitCodeAbbrev();
1096     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 3));
1097     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::VBR, 8));
1098     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Array));
1099     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 8));
1100     if (Stream.EmitBlockInfoAbbrev(bitc::VALUE_SYMTAB_BLOCK_ID, 
1101                                    Abbv) != VST_ENTRY_8_ABBREV)
1102       assert(0 && "Unexpected abbrev ordering!");
1103   }
1104   
1105   { // 7-bit fixed width VST_ENTRY strings.
1106     BitCodeAbbrev *Abbv = new BitCodeAbbrev();
1107     Abbv->Add(BitCodeAbbrevOp(bitc::VST_CODE_ENTRY));
1108     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::VBR, 8));
1109     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Array));
1110     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 7));
1111     if (Stream.EmitBlockInfoAbbrev(bitc::VALUE_SYMTAB_BLOCK_ID,
1112                                    Abbv) != VST_ENTRY_7_ABBREV)
1113       assert(0 && "Unexpected abbrev ordering!");
1114   }
1115   { // 6-bit char6 VST_ENTRY strings.
1116     BitCodeAbbrev *Abbv = new BitCodeAbbrev();
1117     Abbv->Add(BitCodeAbbrevOp(bitc::VST_CODE_ENTRY));
1118     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::VBR, 8));
1119     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Array));
1120     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Char6));
1121     if (Stream.EmitBlockInfoAbbrev(bitc::VALUE_SYMTAB_BLOCK_ID,
1122                                    Abbv) != VST_ENTRY_6_ABBREV)
1123       assert(0 && "Unexpected abbrev ordering!");
1124   }
1125   { // 6-bit char6 VST_BBENTRY strings.
1126     BitCodeAbbrev *Abbv = new BitCodeAbbrev();
1127     Abbv->Add(BitCodeAbbrevOp(bitc::VST_CODE_BBENTRY));
1128     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::VBR, 8));
1129     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Array));
1130     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Char6));
1131     if (Stream.EmitBlockInfoAbbrev(bitc::VALUE_SYMTAB_BLOCK_ID,
1132                                    Abbv) != VST_BBENTRY_6_ABBREV)
1133       assert(0 && "Unexpected abbrev ordering!");
1134   }
1135   
1136   
1137   
1138   { // SETTYPE abbrev for CONSTANTS_BLOCK.
1139     BitCodeAbbrev *Abbv = new BitCodeAbbrev();
1140     Abbv->Add(BitCodeAbbrevOp(bitc::CST_CODE_SETTYPE));
1141     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed,
1142                               Log2_32_Ceil(VE.getTypes().size()+1)));
1143     if (Stream.EmitBlockInfoAbbrev(bitc::CONSTANTS_BLOCK_ID,
1144                                    Abbv) != CONSTANTS_SETTYPE_ABBREV)
1145       assert(0 && "Unexpected abbrev ordering!");
1146   }
1147   
1148   { // INTEGER abbrev for CONSTANTS_BLOCK.
1149     BitCodeAbbrev *Abbv = new BitCodeAbbrev();
1150     Abbv->Add(BitCodeAbbrevOp(bitc::CST_CODE_INTEGER));
1151     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::VBR, 8));
1152     if (Stream.EmitBlockInfoAbbrev(bitc::CONSTANTS_BLOCK_ID,
1153                                    Abbv) != CONSTANTS_INTEGER_ABBREV)
1154       assert(0 && "Unexpected abbrev ordering!");
1155   }
1156   
1157   { // CE_CAST abbrev for CONSTANTS_BLOCK.
1158     BitCodeAbbrev *Abbv = new BitCodeAbbrev();
1159     Abbv->Add(BitCodeAbbrevOp(bitc::CST_CODE_CE_CAST));
1160     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 4));  // cast opc
1161     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed,       // typeid
1162                               Log2_32_Ceil(VE.getTypes().size()+1)));
1163     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::VBR, 8));    // value id
1164
1165     if (Stream.EmitBlockInfoAbbrev(bitc::CONSTANTS_BLOCK_ID,
1166                                    Abbv) != CONSTANTS_CE_CAST_Abbrev)
1167       assert(0 && "Unexpected abbrev ordering!");
1168   }
1169   { // NULL abbrev for CONSTANTS_BLOCK.
1170     BitCodeAbbrev *Abbv = new BitCodeAbbrev();
1171     Abbv->Add(BitCodeAbbrevOp(bitc::CST_CODE_NULL));
1172     if (Stream.EmitBlockInfoAbbrev(bitc::CONSTANTS_BLOCK_ID,
1173                                    Abbv) != CONSTANTS_NULL_Abbrev)
1174       assert(0 && "Unexpected abbrev ordering!");
1175   }
1176   
1177   // FIXME: This should only use space for first class types!
1178  
1179   { // INST_LOAD abbrev for FUNCTION_BLOCK.
1180     BitCodeAbbrev *Abbv = new BitCodeAbbrev();
1181     Abbv->Add(BitCodeAbbrevOp(bitc::FUNC_CODE_INST_LOAD));
1182     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::VBR, 6)); // Ptr
1183     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::VBR, 4)); // Align
1184     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 1)); // volatile
1185     if (Stream.EmitBlockInfoAbbrev(bitc::FUNCTION_BLOCK_ID,
1186                                    Abbv) != FUNCTION_INST_LOAD_ABBREV)
1187       assert(0 && "Unexpected abbrev ordering!");
1188   }
1189   { // INST_BINOP abbrev for FUNCTION_BLOCK.
1190     BitCodeAbbrev *Abbv = new BitCodeAbbrev();
1191     Abbv->Add(BitCodeAbbrevOp(bitc::FUNC_CODE_INST_BINOP));
1192     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::VBR, 6)); // LHS
1193     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::VBR, 6)); // RHS
1194     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 4)); // opc
1195     if (Stream.EmitBlockInfoAbbrev(bitc::FUNCTION_BLOCK_ID,
1196                                    Abbv) != FUNCTION_INST_BINOP_ABBREV)
1197       assert(0 && "Unexpected abbrev ordering!");
1198   }
1199   { // INST_CAST abbrev for FUNCTION_BLOCK.
1200     BitCodeAbbrev *Abbv = new BitCodeAbbrev();
1201     Abbv->Add(BitCodeAbbrevOp(bitc::FUNC_CODE_INST_CAST));
1202     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::VBR, 6));    // OpVal
1203     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed,       // dest ty
1204                               Log2_32_Ceil(VE.getTypes().size()+1)));
1205     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 4));  // opc
1206     if (Stream.EmitBlockInfoAbbrev(bitc::FUNCTION_BLOCK_ID,
1207                                    Abbv) != FUNCTION_INST_CAST_ABBREV)
1208       assert(0 && "Unexpected abbrev ordering!");
1209   }
1210   
1211   { // INST_RET abbrev for FUNCTION_BLOCK.
1212     BitCodeAbbrev *Abbv = new BitCodeAbbrev();
1213     Abbv->Add(BitCodeAbbrevOp(bitc::FUNC_CODE_INST_RET));
1214     if (Stream.EmitBlockInfoAbbrev(bitc::FUNCTION_BLOCK_ID,
1215                                    Abbv) != FUNCTION_INST_RET_VOID_ABBREV)
1216       assert(0 && "Unexpected abbrev ordering!");
1217   }
1218   { // INST_RET abbrev for FUNCTION_BLOCK.
1219     BitCodeAbbrev *Abbv = new BitCodeAbbrev();
1220     Abbv->Add(BitCodeAbbrevOp(bitc::FUNC_CODE_INST_RET));
1221     Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::VBR, 6)); // ValID
1222     if (Stream.EmitBlockInfoAbbrev(bitc::FUNCTION_BLOCK_ID,
1223                                    Abbv) != FUNCTION_INST_RET_VAL_ABBREV)
1224       assert(0 && "Unexpected abbrev ordering!");
1225   }
1226   { // INST_UNREACHABLE abbrev for FUNCTION_BLOCK.
1227     BitCodeAbbrev *Abbv = new BitCodeAbbrev();
1228     Abbv->Add(BitCodeAbbrevOp(bitc::FUNC_CODE_INST_UNREACHABLE));
1229     if (Stream.EmitBlockInfoAbbrev(bitc::FUNCTION_BLOCK_ID,
1230                                    Abbv) != FUNCTION_INST_UNREACHABLE_ABBREV)
1231       assert(0 && "Unexpected abbrev ordering!");
1232   }
1233   
1234   Stream.ExitBlock();
1235 }
1236
1237
1238 /// WriteModule - Emit the specified module to the bitstream.
1239 static void WriteModule(const Module *M, BitstreamWriter &Stream) {
1240   Stream.EnterSubblock(bitc::MODULE_BLOCK_ID, 3);
1241   
1242   // Emit the version number if it is non-zero.
1243   if (CurVersion) {
1244     SmallVector<unsigned, 1> Vals;
1245     Vals.push_back(CurVersion);
1246     Stream.EmitRecord(bitc::MODULE_CODE_VERSION, Vals);
1247   }
1248   
1249   // Analyze the module, enumerating globals, functions, etc.
1250   ValueEnumerator VE(M);
1251
1252   // Emit blockinfo, which defines the standard abbreviations etc.
1253   WriteBlockInfo(VE, Stream);
1254   
1255   // Emit information about parameter attributes.
1256   WriteAttributeTable(VE, Stream);
1257   
1258   // Emit information describing all of the types in the module.
1259   WriteTypeTable(VE, Stream);
1260   
1261   // Emit top-level description of module, including target triple, inline asm,
1262   // descriptors for global variables, and function prototype info.
1263   WriteModuleInfo(M, VE, Stream);
1264   
1265   // Emit constants.
1266   WriteModuleConstants(VE, Stream);
1267   
1268   // If we have any aggregate values in the value table, purge them - these can
1269   // only be used to initialize global variables.  Doing so makes the value
1270   // namespace smaller for code in functions.
1271   int NumNonAggregates = VE.PurgeAggregateValues();
1272   if (NumNonAggregates != -1) {
1273     SmallVector<unsigned, 1> Vals;
1274     Vals.push_back(NumNonAggregates);
1275     Stream.EmitRecord(bitc::MODULE_CODE_PURGEVALS, Vals);
1276   }
1277   
1278   // Emit function bodies.
1279   for (Module::const_iterator I = M->begin(), E = M->end(); I != E; ++I)
1280     if (!I->isDeclaration())
1281       WriteFunction(*I, VE, Stream);
1282   
1283   // Emit the type symbol table information.
1284   WriteTypeSymbolTable(M->getTypeSymbolTable(), VE, Stream);
1285   
1286   // Emit names for globals/functions etc.
1287   WriteValueSymbolTable(M->getValueSymbolTable(), VE, Stream);
1288   
1289   Stream.ExitBlock();
1290 }
1291
1292 /// EmitDarwinBCHeader - If generating a bc file on darwin, we have to emit a
1293 /// header and trailer to make it compatible with the system archiver.  To do
1294 /// this we emit the following header, and then emit a trailer that pads the
1295 /// file out to be a multiple of 16 bytes.
1296 /// 
1297 /// struct bc_header {
1298 ///   uint32_t Magic;         // 0x0B17C0DE
1299 ///   uint32_t Version;       // Version, currently always 0.
1300 ///   uint32_t BitcodeOffset; // Offset to traditional bitcode file.
1301 ///   uint32_t BitcodeSize;   // Size of traditional bitcode file.
1302 ///   uint32_t CPUType;       // CPU specifier.
1303 ///   ... potentially more later ...
1304 /// };
1305 enum {
1306   DarwinBCSizeFieldOffset = 3*4, // Offset to bitcode_size.
1307   DarwinBCHeaderSize = 5*4
1308 };
1309
1310 static void EmitDarwinBCHeader(BitstreamWriter &Stream,
1311                                const std::string &TT) {
1312   unsigned CPUType = ~0U;
1313   
1314   // Match x86_64-*, i[3-9]86-*, powerpc-*, powerpc64-*.  The CPUType is a
1315   // magic number from /usr/include/mach/machine.h.  It is ok to reproduce the
1316   // specific constants here because they are implicitly part of the Darwin ABI.
1317   enum {
1318     DARWIN_CPU_ARCH_ABI64      = 0x01000000,
1319     DARWIN_CPU_TYPE_X86        = 7,
1320     DARWIN_CPU_TYPE_POWERPC    = 18
1321   };
1322   
1323   if (TT.find("x86_64-") == 0)
1324     CPUType = DARWIN_CPU_TYPE_X86 | DARWIN_CPU_ARCH_ABI64;
1325   else if (TT.size() >= 5 && TT[0] == 'i' && TT[2] == '8' && TT[3] == '6' &&
1326            TT[4] == '-' && TT[1] - '3' < 6)
1327     CPUType = DARWIN_CPU_TYPE_X86;
1328   else if (TT.find("powerpc-") == 0)
1329     CPUType = DARWIN_CPU_TYPE_POWERPC;
1330   else if (TT.find("powerpc64-") == 0)
1331     CPUType = DARWIN_CPU_TYPE_POWERPC | DARWIN_CPU_ARCH_ABI64;
1332   
1333   // Traditional Bitcode starts after header.
1334   unsigned BCOffset = DarwinBCHeaderSize;
1335   
1336   Stream.Emit(0x0B17C0DE, 32);
1337   Stream.Emit(0         , 32);  // Version.
1338   Stream.Emit(BCOffset  , 32);
1339   Stream.Emit(0         , 32);  // Filled in later.
1340   Stream.Emit(CPUType   , 32);
1341 }
1342
1343 /// EmitDarwinBCTrailer - Emit the darwin epilog after the bitcode file and
1344 /// finalize the header.
1345 static void EmitDarwinBCTrailer(BitstreamWriter &Stream, unsigned BufferSize) {
1346   // Update the size field in the header.
1347   Stream.BackpatchWord(DarwinBCSizeFieldOffset, BufferSize-DarwinBCHeaderSize);
1348   
1349   // If the file is not a multiple of 16 bytes, insert dummy padding.
1350   while (BufferSize & 15) {
1351     Stream.Emit(0, 8);
1352     ++BufferSize;
1353   }
1354 }
1355
1356
1357 /// WriteBitcodeToFile - Write the specified module to the specified output
1358 /// stream.
1359 void llvm::WriteBitcodeToFile(const Module *M, std::ostream &Out) {
1360   raw_os_ostream RawOut(Out);
1361   // If writing to stdout, set binary mode.
1362   if (llvm::cout == Out)
1363     sys::Program::ChangeStdoutToBinary();
1364   WriteBitcodeToFile(M, RawOut);
1365 }
1366
1367 /// WriteBitcodeToFile - Write the specified module to the specified output
1368 /// stream.
1369 void llvm::WriteBitcodeToFile(const Module *M, raw_ostream &Out) {
1370   std::vector<unsigned char> Buffer;
1371   BitstreamWriter Stream(Buffer);
1372   
1373   Buffer.reserve(256*1024);
1374
1375   WriteBitcodeToStream( M, Stream );
1376   
1377   // If writing to stdout, set binary mode.
1378   if (&llvm::outs() == &Out)
1379     sys::Program::ChangeStdoutToBinary();
1380
1381   // Write the generated bitstream to "Out".
1382   Out.write((char*)&Buffer.front(), Buffer.size());
1383   
1384   // Make sure it hits disk now.
1385   Out.flush();
1386 }
1387
1388 /// WriteBitcodeToStream - Write the specified module to the specified output
1389 /// stream.
1390 void llvm::WriteBitcodeToStream(const Module *M, BitstreamWriter &Stream) {
1391   // If this is darwin, emit a file header and trailer if needed.
1392   bool isDarwin = M->getTargetTriple().find("-darwin") != std::string::npos;
1393   if (isDarwin)
1394     EmitDarwinBCHeader(Stream, M->getTargetTriple());
1395   
1396   // Emit the file header.
1397   Stream.Emit((unsigned)'B', 8);
1398   Stream.Emit((unsigned)'C', 8);
1399   Stream.Emit(0x0, 4);
1400   Stream.Emit(0xC, 4);
1401   Stream.Emit(0xE, 4);
1402   Stream.Emit(0xD, 4);
1403
1404   // Emit the module.
1405   WriteModule(M, Stream);
1406
1407   if (isDarwin)
1408     EmitDarwinBCTrailer(Stream, Stream.getBuffer().size());
1409 }