//
//===----------------------------------------------------------------------===//
+#define DEBUG_TYPE "bytecodewriter"
#include "WriterInternals.h"
#include "llvm/Bytecode/WriteBytecodePass.h"
#include "llvm/CallingConv.h"
#include "llvm/InlineAsm.h"
#include "llvm/Instructions.h"
#include "llvm/Module.h"
-#include "llvm/SymbolTable.h"
+#include "llvm/TypeSymbolTable.h"
+#include "llvm/ValueSymbolTable.h"
#include "llvm/Support/GetElementPtrTypeIterator.h"
#include "llvm/Support/Compressor.h"
#include "llvm/Support/MathExtras.h"
static RegisterPass<WriteBytecodePass> X("emitbytecode", "Bytecode Writer");
-static Statistic
-BytesWritten("bytecodewriter", "Number of bytecode bytes written");
+STATISTIC(BytesWritten, "Number of bytecode bytes written");
//===----------------------------------------------------------------------===//
//=== Output Primitives ===//
//===----------------------------------------------------------------------===//
void BytecodeWriter::outputType(const Type *T) {
- output_vbr((unsigned)T->getTypeID());
+ const StructType* STy = dyn_cast<StructType>(T);
+ if(STy && STy->isPacked())
+ output_vbr((unsigned)Type::PackedStructTyID);
+ else
+ output_vbr((unsigned)T->getTypeID());
// That's all there is to handling primitive types...
- if (T->isPrimitiveType()) {
+ if (T->isPrimitiveType())
return; // We might do this if we alias a prim type: %x = type int
- }
switch (T->getTypeID()) { // Handle derived types now.
+ case Type::IntegerTyID:
+ output_vbr(cast<IntegerType>(T)->getBitWidth());
+ break;
case Type::FunctionTyID: {
const FunctionType *MT = cast<FunctionType>(T);
int Slot = Table.getSlot(MT->getReturnType());
assert(Slot != -1 && "Type used but not available!!");
output_typeid((unsigned)Slot);
+ output_vbr(unsigned(MT->getParamAttrs(0)));
// Output the number of arguments to function (+1 if varargs):
output_vbr((unsigned)MT->getNumParams()+MT->isVarArg());
// Output all of the arguments...
FunctionType::param_iterator I = MT->param_begin();
+ unsigned Idx = 1;
for (; I != MT->param_end(); ++I) {
Slot = Table.getSlot(*I);
assert(Slot != -1 && "Type used but not available!!");
output_typeid((unsigned)Slot);
+ output_vbr(unsigned(MT->getParamAttrs(Idx)));
+ Idx++;
}
// Terminate list with VoidTy if we are a varargs function...
break;
}
-
case Type::StructTyID: {
const StructType *ST = cast<StructType>(T);
-
// Output all of the element types...
for (StructType::element_iterator I = ST->element_begin(),
E = ST->element_end(); I != E; ++I) {
break;
default:
- llvm_cerr << __FILE__ << ":" << __LINE__ << ": Don't know how to serialize"
- << " Type '" << T->getDescription() << "'\n";
+ cerr << __FILE__ << ":" << __LINE__ << ": Don't know how to serialize"
+ << " Type '" << T->getDescription() << "'\n";
break;
}
}
void BytecodeWriter::outputConstant(const Constant *CPV) {
- assert((CPV->getType()->isPrimitiveType() || !CPV->isNullValue()) &&
- "Shouldn't output null constants!");
+ assert(((CPV->getType()->isPrimitiveType() || CPV->getType()->isInteger()) ||
+ !CPV->isNullValue()) && "Shouldn't output null constants!");
// We must check for a ConstantExpr before switching by type because
// a ConstantExpr can be of any type, and has no explicit value.
}
switch (CPV->getType()->getTypeID()) {
- case Type::BoolTyID: // Boolean Types
- if (cast<ConstantBool>(CPV)->getValue())
- output_vbr(1U);
- else
- output_vbr(0U);
- break;
-
- case Type::UByteTyID: // Unsigned integer types...
- case Type::UShortTyID:
- case Type::UIntTyID:
- case Type::ULongTyID:
- output_vbr(cast<ConstantInt>(CPV)->getZExtValue());
- break;
-
- case Type::SByteTyID: // Signed integer types...
- case Type::ShortTyID:
- case Type::IntTyID:
- case Type::LongTyID:
- output_vbr(cast<ConstantInt>(CPV)->getSExtValue());
+ case Type::IntegerTyID: { // Integer types...
+ unsigned NumBits = cast<IntegerType>(CPV->getType())->getBitWidth();
+ if (NumBits <= 32)
+ output_vbr(uint32_t(cast<ConstantInt>(CPV)->getZExtValue()));
+ else if (NumBits <= 64)
+ output_vbr(uint64_t(cast<ConstantInt>(CPV)->getZExtValue()));
+ else
+ assert("Integer types > 64 bits not supported.");
break;
+ }
case Type::ArrayTyID: {
const ConstantArray *CPA = cast<ConstantArray>(CPV);
case Type::VoidTyID:
case Type::LabelTyID:
default:
- llvm_cerr << __FILE__ << ":" << __LINE__ << ": Don't know how to serialize"
- << " type '" << *CPV->getType() << "'\n";
+ cerr << __FILE__ << ":" << __LINE__ << ": Don't know how to serialize"
+ << " type '" << *CPV->getType() << "'\n";
break;
}
return;
assert(Slot >= 0 && "No slot number for value!?!?");
if (isa<SequentialType>(*TI)) {
- unsigned IdxId;
- switch (I->getOperand(Idx)->getType()->getTypeID()) {
- default: assert(0 && "Unknown index type!");
- case Type::UIntTyID: IdxId = 0; break;
- case Type::IntTyID: IdxId = 1; break;
- case Type::ULongTyID: IdxId = 2; break;
- case Type::LongTyID: IdxId = 3; break;
- }
- Slot = (Slot << 2) | IdxId;
+ // These should be either 32-bits or 64-bits, however, with bit
+ // accurate types we just distinguish between less than or equal to
+ // 32-bits or greater than 32-bits.
+ unsigned BitWidth =
+ cast<IntegerType>(I->getOperand(Idx)->getType())->getBitWidth();
+ assert(BitWidth == 32 || BitWidth == 64 &&
+ "Invalid bitwidth for GEP index");
+ unsigned IdxId = BitWidth == 32 ? 0 : 1;
+ Slot = (Slot << 1) | IdxId;
}
output_vbr(unsigned(Slot));
}
for (gep_type_iterator I = gep_type_begin(GEP), E = gep_type_end(GEP);
I != E; ++I, ++Idx)
if (isa<SequentialType>(*I)) {
- unsigned IdxId;
- switch (GEP->getOperand(Idx)->getType()->getTypeID()) {
- default: assert(0 && "Unknown index type!");
- case Type::UIntTyID: IdxId = 0; break;
- case Type::IntTyID: IdxId = 1; break;
- case Type::ULongTyID: IdxId = 2; break;
- case Type::LongTyID: IdxId = 3; break;
- }
- Slots[Idx] = (Slots[Idx] << 2) | IdxId;
+ // These should be either 32-bits or 64-bits, however, with bit
+ // accurate types we just distinguish between less than or equal to
+ // 32-bits or greater than 32-bits.
+ unsigned BitWidth =
+ cast<IntegerType>(GEP->getOperand(Idx)->getType())->getBitWidth();
+ assert(BitWidth == 32 || BitWidth == 64 &&
+ "Invalid bitwidth for GEP index");
+ unsigned IdxId = BitWidth == 32 ? 0 : 1;
+ Slots[Idx] = (Slots[Idx] << 1) | IdxId;
if (Slots[Idx] > MaxOpSlot) MaxOpSlot = Slots[Idx];
}
} else if (Opcode == 58) {
// Emit the top level CLASS block.
BytecodeBlock ModuleBlock(BytecodeFormat::ModuleBlockID, *this, false, true);
- bool isBigEndian = M->getEndianness() == Module::BigEndian;
- bool hasLongPointers = M->getPointerSize() == Module::Pointer64;
- bool hasNoEndianness = M->getEndianness() == Module::AnyEndianness;
- bool hasNoPointerSize = M->getPointerSize() == Module::AnyPointerSize;
-
- // Output the version identifier and other information.
- unsigned Version = (BCVersionNum << 4) |
- (unsigned)isBigEndian | (hasLongPointers << 1) |
- (hasNoEndianness << 2) |
- (hasNoPointerSize << 3);
- output_vbr(Version);
+ // Output the version identifier
+ output_vbr(BCVersionNum);
// The Global type plane comes first
{
for (Module::const_iterator I = M->begin(), E = M->end(); I != E; ++I)
outputFunction(I);
- // If needed, output the symbol table for the module...
- outputSymbolTable(M->getSymbolTable());
+ // Output the symbole table for types
+ outputTypeSymbolTable(M->getTypeSymbolTable());
+
+ // Output the symbol table for values
+ outputValueSymbolTable(M->getValueSymbolTable());
}
void BytecodeWriter::outputTypes(unsigned TypeNum) {
}
}
+static unsigned getEncodedVisibility(const GlobalValue *GV) {
+ switch (GV->getVisibility()) {
+ default: assert(0 && "Invalid visibility!");
+ case GlobalValue::DefaultVisibility: return 0;
+ case GlobalValue::HiddenVisibility: return 1;
+ }
+}
+
void BytecodeWriter::outputModuleInfoBlock(const Module *M) {
BytecodeBlock ModuleInfoBlock(BytecodeFormat::ModuleGlobalInfoBlockID, *this);
// Fields: bit0 = isConstant, bit1 = hasInitializer, bit2-4=Linkage,
// bit5+ = Slot # for type.
- bool HasExtensionWord = (I->getAlignment() != 0) || I->hasSection();
+ bool HasExtensionWord = (I->getAlignment() != 0) ||
+ I->hasSection() ||
+ (I->getVisibility() != GlobalValue::DefaultVisibility);
// If we need to use the extension byte, set linkage=3(internal) and
// initializer = 0 (impossible!).
output_vbr(oSlot);
// The extension word has this format: bit 0 = has initializer, bit 1-3 =
- // linkage, bit 4-8 = alignment (log2), bit 9 = has SectionID,
- // bits 10+ = future use.
+ // linkage, bit 4-8 = alignment (log2), bit 9 = has SectionID,
+ // bits 10-12 = visibility, bits 13+ = future use.
unsigned ExtWord = (unsigned)I->hasInitializer() |
(getEncodedLinkage(I) << 1) |
((Log2_32(I->getAlignment())+1) << 4) |
- ((unsigned)I->hasSection() << 9);
+ ((unsigned)I->hasSection() << 9) |
+ (getEncodedVisibility(I) << 10);
output_vbr(ExtWord);
if (I->hasSection()) {
// Give section names unique ID's.
unsigned CC = I->getCallingConv()+1;
unsigned ID = (Slot << 5) | (CC & 15);
- if (I->isExternal()) // If external, we don't have an FunctionInfo block.
+ if (I->isDeclaration()) // If external, we don't have an FunctionInfo block.
ID |= 1 << 4;
if (I->getAlignment() || I->hasSection() || (CC & ~15) != 0 ||
- (I->isExternal() && I->hasDLLImportLinkage()) ||
- (I->isExternal() && I->hasExternalWeakLinkage())
+ (I->isDeclaration() && I->hasDLLImportLinkage()) ||
+ (I->isDeclaration() && I->hasExternalWeakLinkage())
)
ID |= 1 << 31; // Do we need an extension word?
// convention, bit 10 = hasSectionID., bits 11-12 = external linkage type
unsigned extLinkage = 0;
- if (I->isExternal()) {
+ if (I->isDeclaration()) {
if (I->hasDLLImportLinkage()) {
extLinkage = 1;
} else if (I->hasExternalWeakLinkage()) {
// Output the target triple from the module
output(M->getTargetTriple());
+
+ // Output the data layout from the module
+ output(M->getDataLayout());
// Emit the table of section names.
output_vbr((unsigned)SectionNames.size());
void BytecodeWriter::outputFunction(const Function *F) {
// If this is an external function, there is nothing else to emit!
- if (F->isExternal()) return;
+ if (F->isDeclaration()) return;
BytecodeBlock FunctionBlock(BytecodeFormat::FunctionBlockID, *this);
- output_vbr(getEncodedLinkage(F));
+ unsigned rWord = (getEncodedVisibility(F) << 16) | getEncodedLinkage(F);
+ output_vbr(rWord);
// Get slot information about the function...
Table.incorporateFunction(F);
- if (Table.getCompactionTable().empty()) {
- // Output information about the constants in the function if the compaction
- // table is not being used.
- outputConstants(true);
- } else {
- // Otherwise, emit the compaction table.
- outputCompactionTable();
- }
+ outputConstants(true);
// Output all of the instructions in the body of the function
outputInstructions(F);
// If needed, output the symbol table for the function...
- outputSymbolTable(F->getSymbolTable());
+ outputValueSymbolTable(F->getValueSymbolTable());
Table.purgeFunction();
}
-void BytecodeWriter::outputCompactionTablePlane(unsigned PlaneNo,
- const std::vector<const Value*> &Plane,
- unsigned StartNo) {
- unsigned End = Table.getModuleLevel(PlaneNo);
- if (Plane.empty() || StartNo == End || End == 0) return; // Nothing to emit
- assert(StartNo < End && "Cannot emit negative range!");
- assert(StartNo < Plane.size() && End <= Plane.size());
-
- // Do not emit the null initializer!
- ++StartNo;
-
- // Figure out which encoding to use. By far the most common case we have is
- // to emit 0-2 entries in a compaction table plane.
- switch (End-StartNo) {
- case 0: // Avoid emitting two vbr's if possible.
- case 1:
- case 2:
- output_vbr((PlaneNo << 2) | End-StartNo);
- break;
- default:
- // Output the number of things.
- output_vbr((unsigned(End-StartNo) << 2) | 3);
- output_typeid(PlaneNo); // Emit the type plane this is
- break;
- }
-
- for (unsigned i = StartNo; i != End; ++i)
- output_vbr(Table.getGlobalSlot(Plane[i]));
-}
-
-void BytecodeWriter::outputCompactionTypes(unsigned StartNo) {
- // Get the compaction type table from the slot calculator
- const std::vector<const Type*> &CTypes = Table.getCompactionTypes();
-
- // The compaction types may have been uncompactified back to the
- // global types. If so, we just write an empty table
- if (CTypes.size() == 0) {
- output_vbr(0U);
- return;
- }
-
- assert(CTypes.size() >= StartNo && "Invalid compaction types start index");
-
- // Determine how many types to write
- unsigned NumTypes = CTypes.size() - StartNo;
-
- // Output the number of types.
- output_vbr(NumTypes);
-
- for (unsigned i = StartNo; i < StartNo+NumTypes; ++i)
- output_typeid(Table.getGlobalSlot(CTypes[i]));
-}
-
-void BytecodeWriter::outputCompactionTable() {
- // Avoid writing the compaction table at all if there is no content.
- if (Table.getCompactionTypes().size() >= Type::FirstDerivedTyID ||
- (!Table.CompactionTableIsEmpty())) {
- BytecodeBlock CTB(BytecodeFormat::CompactionTableBlockID, *this,
- true/*ElideIfEmpty*/);
- const std::vector<std::vector<const Value*> > &CT =
- Table.getCompactionTable();
-
- // First things first, emit the type compaction table if there is one.
- outputCompactionTypes(Type::FirstDerivedTyID);
- for (unsigned i = 0, e = CT.size(); i != e; ++i)
- outputCompactionTablePlane(i, CT[i], 0);
- }
-}
-
-void BytecodeWriter::outputSymbolTable(const SymbolTable &MST) {
- // Do not output the Bytecode block for an empty symbol table, it just wastes
+void BytecodeWriter::outputTypeSymbolTable(const TypeSymbolTable &TST) {
+ // Do not output the block for an empty symbol table, it just wastes
// space!
- if (MST.isEmpty()) return;
+ if (TST.empty()) return;
- BytecodeBlock SymTabBlock(BytecodeFormat::SymbolTableBlockID, *this,
+ // Create a header for the symbol table
+ BytecodeBlock SymTabBlock(BytecodeFormat::TypeSymbolTableBlockID, *this,
true/*ElideIfEmpty*/);
-
// Write the number of types
- output_vbr(MST.num_types());
+ output_vbr(TST.size());
// Write each of the types
- for (SymbolTable::type_const_iterator TI = MST.type_begin(),
- TE = MST.type_end(); TI != TE; ++TI) {
+ for (TypeSymbolTable::const_iterator TI = TST.begin(), TE = TST.end();
+ TI != TE; ++TI) {
// Symtab entry:[def slot #][name]
output_typeid((unsigned)Table.getSlot(TI->second));
output(TI->first);
}
+}
- // Now do each of the type planes in order.
- for (SymbolTable::plane_const_iterator PI = MST.plane_begin(),
- PE = MST.plane_end(); PI != PE; ++PI) {
- SymbolTable::value_const_iterator I = MST.value_begin(PI->first);
- SymbolTable::value_const_iterator End = MST.value_end(PI->first);
+void BytecodeWriter::outputValueSymbolTable(const ValueSymbolTable &VST) {
+ // Do not output the Bytecode block for an empty symbol table, it just wastes
+ // space!
+ if (VST.empty()) return;
+
+ BytecodeBlock SymTabBlock(BytecodeFormat::ValueSymbolTableBlockID, *this,
+ true/*ElideIfEmpty*/);
+
+ // Organize the symbol table by type
+ typedef std::pair<std::string, const Value*> PlaneMapEntry;
+ typedef std::vector<PlaneMapEntry> PlaneMapVector;
+ typedef std::map<const Type*, PlaneMapVector > PlaneMap;
+ PlaneMap Planes;
+ for (ValueSymbolTable::const_iterator SI = VST.begin(), SE = VST.end();
+ SI != SE; ++SI)
+ Planes[SI->second->getType()].push_back(
+ std::make_pair(SI->first,SI->second));
+
+ for (PlaneMap::const_iterator PI = Planes.begin(), PE = Planes.end();
+ PI != PE; ++PI) {
int Slot;
+ PlaneMapVector::const_iterator I = PI->second.begin();
+ PlaneMapVector::const_iterator End = PI->second.end();
+
if (I == End) continue; // Don't mess with an absent type...
// Write the number of values in this plane
}
}
-void llvm::WriteBytecodeToFile(const Module *M, llvm_ostream &Out,
+void llvm::WriteBytecodeToFile(const Module *M, OStream &Out,
bool compress) {
assert(M && "You can't write a null module!!");
// Make sure that std::cout is put into binary mode for systems
// that care.
- if (Out == llvm_cout)
+ if (Out == cout)
sys::Program::ChangeStdoutToBinary();
// Create a vector of unsigned char for the bytecode output. We