1 //===-- Writer.cpp - Library for writing VM bytecode files -------*- C++ -*--=//
3 // This library implements the functionality defined in llvm/Bytecode/Writer.h
5 // This library uses the Analysis library to figure out offsets for
6 // variables in the method tables...
8 // Note that this file uses an unusual technique of outputting all the bytecode
9 // to a deque of unsigned char's, then copies the deque to an ostream. The
10 // reason for this is that we must do "seeking" in the stream to do back-
11 // patching, and some very important ostreams that we want to support (like
12 // pipes) do not support seeking. :( :( :(
14 // The choice of the deque data structure is influenced by the extremely fast
15 // "append" speed, plus the free "seek"/replace in the middle of the stream. I
16 // didn't use a vector because the stream could end up very large and copying
17 // the whole thing to reallocate would be kinda silly.
19 // Note that the performance of this library is not terribly important, because
20 // it shouldn't be used by JIT type applications... so it is not a huge focus
23 //===----------------------------------------------------------------------===//
25 #include "WriterInternals.h"
26 #include "llvm/Module.h"
27 #include "llvm/GlobalVariable.h"
28 #include "llvm/Method.h"
29 #include "llvm/BasicBlock.h"
30 #include "llvm/ConstPoolVals.h"
31 #include "llvm/SymbolTable.h"
32 #include "llvm/DerivedTypes.h"
33 #include "llvm/Support/STLExtras.h"
37 BytecodeWriter::BytecodeWriter(deque<unsigned char> &o, const Module *M)
38 : Out(o), Table(M, false) {
42 // Emit the top level CLASS block.
43 BytecodeBlock ModuleBlock(BytecodeFormat::Module, Out);
45 // Output the ID of first "derived" type:
46 output_vbr((unsigned)Type::FirstDerivedTyID, Out);
49 // Output module level constants, including types used by the method protos
50 outputConstants(false);
52 // The ModuleInfoBlock follows directly after the Module constant pool
53 outputModuleInfoBlock(M);
55 // Do the whole module now! Process each method at a time...
56 for_each(M->begin(), M->end(),
57 bind_obj(this, &BytecodeWriter::processMethod));
59 // If needed, output the symbol table for the module...
60 if (M->hasSymbolTable())
61 outputSymbolTable(*M->getSymbolTable());
65 #include "llvm/Assembly/Writer.h"
67 void BytecodeWriter::outputConstants(bool isMethod) {
68 BytecodeBlock CPool(BytecodeFormat::ConstantPool, Out);
70 unsigned NumPlanes = Table.getNumPlanes();
71 for (unsigned pno = 0; pno < NumPlanes; pno++) {
72 const vector<const Value*> &Plane = Table.getPlane(pno);
73 if (Plane.empty()) continue; // Skip empty type planes...
76 if (isMethod) // Don't reemit module constants
77 ValNo = Table.getModuleLevel(pno);
78 else if (pno == Type::TypeTyID)
79 ValNo = Type::FirstDerivedTyID; // Start emitting at the derived types...
81 // Scan through and ignore method arguments...
82 for (; ValNo < Plane.size() && isa<MethodArgument>(Plane[ValNo]); ValNo++)
85 unsigned NC = ValNo; // Number of constants
86 for (; NC < Plane.size() &&
87 (isa<ConstPoolVal>(Plane[NC]) ||
88 isa<Type>(Plane[NC])); NC++) /*empty*/;
89 NC -= ValNo; // Convert from index into count
90 if (NC == 0) continue; // Skip empty type planes...
92 // Output type header: [num entries][type id number]
96 // Output the Type ID Number...
97 int Slot = Table.getValSlot(Plane.front()->getType());
98 assert (Slot != -1 && "Type in constant pool but not in method!!");
99 output_vbr((unsigned)Slot, Out);
101 //cout << "Emitting " << NC << " constants of type '"
102 // << Plane.front()->getType()->getName() << "' = Slot #" << Slot << endl;
104 for (unsigned i = ValNo; i < ValNo+NC; ++i) {
105 const Value *V = Plane[i];
106 if (const ConstPoolVal *CPV = dyn_cast<ConstPoolVal>(V)) {
107 //cerr << "Serializing value: <" << V->getType() << ">: "
108 // << ((const ConstPoolVal*)V)->getStrValue() << ":"
109 // << Out.size() << "\n";
112 const Type *Ty = cast<const Type>(V);
119 void BytecodeWriter::outputModuleInfoBlock(const Module *M) {
120 BytecodeBlock ModuleInfoBlock(BytecodeFormat::ModuleGlobalInfo, Out);
122 // Output the types for the global variables in the module...
123 for (Module::const_giterator I = M->gbegin(), End = M->gend(); I != End;++I) {
124 const GlobalVariable *GV = *I;
125 int Slot = Table.getValSlot(GV->getType());
126 assert(Slot != -1 && "Module global vars is broken!");
128 // Fields: bit0 = isConstant, bit1 = hasInitializer, bit2+ = slot#
129 unsigned oSlot = ((unsigned)Slot << 2) | (GV->hasInitializer() << 1) |
130 isa<ConstPoolVal>(GV);
131 output_vbr(oSlot, Out);
133 // If we have an initialized, output it now.
134 if (GV->hasInitializer()) {
135 Slot = Table.getValSlot(GV->getInitializer());
136 assert(Slot != -1 && "No slot for global var initializer!");
137 output_vbr((unsigned)Slot, Out);
140 output_vbr((unsigned)Table.getValSlot(Type::VoidTy), Out);
142 // Output the types of the methods in this module...
143 for (Module::const_iterator I = M->begin(), End = M->end(); I != End; ++I) {
144 int Slot = Table.getValSlot((*I)->getType());
145 assert(Slot != -1 && "Module const pool is broken!");
146 assert(Slot >= Type::FirstDerivedTyID && "Derived type not in range!");
147 output_vbr((unsigned)Slot, Out);
149 output_vbr((unsigned)Table.getValSlot(Type::VoidTy), Out);
155 void BytecodeWriter::processMethod(const Method *M) {
156 BytecodeBlock MethodBlock(BytecodeFormat::Method, Out);
158 // Only output the constant pool and other goodies if needed...
159 if (!M->isExternal()) {
160 // Get slot information about the method...
161 Table.incorporateMethod(M);
163 // Output information about the constants in the method...
164 outputConstants(true);
166 // Output basic block nodes...
167 for_each(M->begin(), M->end(),
168 bind_obj(this, &BytecodeWriter::processBasicBlock));
170 // If needed, output the symbol table for the method...
171 if (M->hasSymbolTable())
172 outputSymbolTable(*M->getSymbolTable());
179 void BytecodeWriter::processBasicBlock(const BasicBlock *BB) {
180 BytecodeBlock MethodBlock(BytecodeFormat::BasicBlock, Out);
181 // Process all the instructions in the bb...
182 for_each(BB->begin(), BB->end(),
183 bind_obj(this, &BytecodeWriter::processInstruction));
186 void BytecodeWriter::outputSymbolTable(const SymbolTable &MST) {
187 BytecodeBlock MethodBlock(BytecodeFormat::SymbolTable, Out);
189 for (SymbolTable::const_iterator TI = MST.begin(); TI != MST.end(); ++TI) {
190 SymbolTable::type_const_iterator I = MST.type_begin(TI->first);
191 SymbolTable::type_const_iterator End = MST.type_end(TI->first);
194 if (I == End) continue; // Don't mess with an absent type...
196 // Symtab block header: [num entries][type id number]
197 output_vbr(MST.type_size(TI->first), Out);
199 Slot = Table.getValSlot(TI->first);
200 assert(Slot != -1 && "Type in symtab, but not in table!");
201 output_vbr((unsigned)Slot, Out);
203 for (; I != End; ++I) {
204 // Symtab entry: [def slot #][name]
205 Slot = Table.getValSlot(I->second);
206 assert(Slot != -1 && "Value in symtab but has no slot number!!");
207 output_vbr((unsigned)Slot, Out);
208 output(I->first, Out, false); // Don't force alignment...
213 void WriteBytecodeToFile(const Module *C, ostream &Out) {
214 assert(C && "You can't write a null module!!");
216 deque<unsigned char> Buffer;
218 // This object populates buffer for us...
219 BytecodeWriter BCW(Buffer, C);
221 // Okay, write the deque out to the ostream now... the deque is not
222 // sequential in memory, however, so write out as much as possible in big
223 // chunks, until we're done.
225 deque<unsigned char>::const_iterator I = Buffer.begin(), E = Buffer.end();
226 while (I != E) { // Loop until it's all written
227 // Scan to see how big this chunk is...
228 const unsigned char *ChunkPtr = &*I;
229 const unsigned char *LastPtr = ChunkPtr;
231 const unsigned char *ThisPtr = &*++I;
232 if (LastPtr+1 != ThisPtr) break;// Advanced by more than a byte of memory?
236 // Write out the chunk...
237 Out.write(ChunkPtr, LastPtr-ChunkPtr+(I != E));