6502/llvm-6502
1
0
Fork 0
mirror of https://github.com/c64scene-ar/llvm-6502.git synced 2025-01-06 04:31:08 +00:00
Code Issues Projects Releases Wiki Activity
6845dfe07e
llvm-6502/lib/Target/X86/X86AsmPrinter.cpp
Chris Lattner 8e61d82528 Remove a dead method 
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@15659 91177308-0d34-0410-b5e6-96231b3b80d8
2004-08-11 07:07:14 +00:00

679 lines
23 KiB
C++
Raw Blame History

//===-- X86AsmPrinter.cpp - Convert X86 LLVM code to Intel assembly -------===//
//
// The LLVM Compiler Infrastructure
//
// This file was developed by the LLVM research group and is distributed under
// the University of Illinois Open Source License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file contains a printer that converts from our internal representation
// of machine-dependent LLVM code to Intel-format assembly language. This
// printer is the output mechanism used by `llc' and `lli -print-machineinstrs'
// on X86.
//
//===----------------------------------------------------------------------===//
#include "X86.h"
#include "X86InstrInfo.h"
#include "X86TargetMachine.h"
#include "llvm/Constants.h"
#include "llvm/DerivedTypes.h"
#include "llvm/Module.h"
#include "llvm/Assembly/Writer.h"
#include "llvm/CodeGen/MachineCodeEmitter.h"
#include "llvm/CodeGen/MachineConstantPool.h"
#include "llvm/CodeGen/MachineFunctionPass.h"
#include "llvm/CodeGen/MachineInstr.h"
#include "llvm/CodeGen/ValueTypes.h"
#include "llvm/Target/TargetMachine.h"
#include "llvm/Support/Mangler.h"
#include "Support/Statistic.h"
#include "Support/StringExtras.h"
#include "Support/CommandLine.h"
using namespace llvm;
namespace {
Statistic<> EmittedInsts("asm-printer", "Number of machine instrs printed");
// FIXME: This should be automatically picked up by autoconf from the C
// frontend
cl::opt<bool> EmitCygwin("enable-cygwin-compatible-output", cl::Hidden,
cl::desc("Emit X86 assembly code suitable for consumption by cygwin"));
struct GasBugWorkaroundEmitter : public MachineCodeEmitter {
GasBugWorkaroundEmitter(std::ostream& o)
: O(o), OldFlags(O.flags()), firstByte(true) {
O << std::hex;
}
~GasBugWorkaroundEmitter() {
O.flags(OldFlags);
}
virtual void emitByte(unsigned char B) {
if (!firstByte) O << "\n\t";
firstByte = false;
O << ".byte 0x" << (unsigned) B;
}
// These should never be called
virtual void emitWord(unsigned W) { assert(0); }
virtual uint64_t getGlobalValueAddress(GlobalValue *V) { abort(); }
virtual uint64_t getGlobalValueAddress(const std::string &Name) { abort(); }
virtual uint64_t getConstantPoolEntryAddress(unsigned Index) { abort(); }
virtual uint64_t getCurrentPCValue() { abort(); }
virtual uint64_t forceCompilationOf(Function *F) { abort(); }
private:
std::ostream& O;
std::ios::fmtflags OldFlags;
bool firstByte;
};
struct X86AsmPrinter : public MachineFunctionPass {
/// Output stream on which we're printing assembly code.
///
std::ostream &O;
/// Target machine description which we query for reg. names, data
/// layout, etc.
///
TargetMachine &TM;
/// Name-mangler for global names.
///
Mangler *Mang;
X86AsmPrinter(std::ostream &o, TargetMachine &tm) : O(o), TM(tm) { }
/// Cache of mangled name for current function. This is
/// recalculated at the beginning of each call to
/// runOnMachineFunction().
///
std::string CurrentFnName;
virtual const char *getPassName() const {
return "X86 Assembly Printer";
}
/// printInstruction - This method is automatically generated by tablegen
/// from the instruction set description. This method returns true if the
/// machine instruction was sufficiently described to print it, otherwise it
/// returns false.
bool printInstruction(const MachineInstr *MI);
// This method is used by the tablegen'erated instruction printer.
void printOperand(const MachineInstr *MI, unsigned OpNo, MVT::ValueType VT) {
const MachineOperand &MO = MI->getOperand(OpNo);
if (MO.getType() == MachineOperand::MO_MachineRegister) {
assert(MRegisterInfo::isPhysicalRegister(MO.getReg())&&"Not physref??");
// Bug Workaround: See note in Printer::doInitialization about %.
O << "%" << TM.getRegisterInfo()->get(MO.getReg()).Name;
} else {
printOp(MO);
}
}
void printCallOperand(const MachineInstr *MI, unsigned OpNo, MVT::ValueType VT) {
printOp(MI->getOperand(OpNo), true); // Don't print "OFFSET".
}
void printMemoryOperand(const MachineInstr *MI, unsigned OpNo,
MVT::ValueType VT) {
switch (VT) {
default: assert(0 && "Unknown arg size!");
case MVT::i8: O << "BYTE PTR "; break;
case MVT::i16: O << "WORD PTR "; break;
case MVT::i32:
case MVT::f32: O << "DWORD PTR "; break;
case MVT::i64:
case MVT::f64: O << "QWORD PTR "; break;
case MVT::f80: O << "XWORD PTR "; break;
}
printMemReference(MI, OpNo);
}
void printMachineInstruction(const MachineInstr *MI);
void printOp(const MachineOperand &MO, bool elideOffsetKeyword = false);
void printMemReference(const MachineInstr *MI, unsigned Op);
void printConstantPool(MachineConstantPool *MCP);
bool runOnMachineFunction(MachineFunction &F);
bool doInitialization(Module &M);
bool doFinalization(Module &M);
void emitGlobalConstant(const Constant* CV);
void emitConstantValueOnly(const Constant *CV);
};
} // end of anonymous namespace
/// createX86CodePrinterPass - Returns a pass that prints the X86
/// assembly code for a MachineFunction to the given output stream,
/// using the given target machine description. This should work
/// regardless of whether the function is in SSA form.
///
FunctionPass *llvm::createX86CodePrinterPass(std::ostream &o,TargetMachine &tm){
return new X86AsmPrinter(o, tm);
}
// Include the auto-generated portion of the assembly writer.
#include "X86GenAsmWriter.inc"
/// toOctal - Convert the low order bits of X into an octal digit.
///
static inline char toOctal(int X) {
return (X&7)+'0';
}
/// getAsCString - Return the specified array as a C compatible
/// string, only if the predicate isStringCompatible is true.
///
static void printAsCString(std::ostream &O, const ConstantArray *CVA) {
assert(CVA->isString() && "Array is not string compatible!");
O << "\"";
for (unsigned i = 0; i != CVA->getNumOperands(); ++i) {
unsigned char C = cast<ConstantInt>(CVA->getOperand(i))->getRawValue();
if (C == '"') {
O << "\\\"";
} else if (C == '\\') {
O << "\\\\";
} else if (isprint(C)) {
O << C;
} else {
switch(C) {
case '\b': O << "\\b"; break;
case '\f': O << "\\f"; break;
case '\n': O << "\\n"; break;
case '\r': O << "\\r"; break;
case '\t': O << "\\t"; break;
default:
O << '\\';
O << toOctal(C >> 6);
O << toOctal(C >> 3);
O << toOctal(C >> 0);
break;
}
}
}
O << "\"";
}
// Print out the specified constant, without a storage class. Only the
// constants valid in constant expressions can occur here.
void X86AsmPrinter::emitConstantValueOnly(const Constant *CV) {
if (CV->isNullValue())
O << "0";
else if (const ConstantBool *CB = dyn_cast<ConstantBool>(CV)) {
assert(CB == ConstantBool::True);
O << "1";
} else if (const ConstantSInt *CI = dyn_cast<ConstantSInt>(CV))
if (((CI->getValue() << 32) >> 32) == CI->getValue())
O << CI->getValue();
else
O << (unsigned long long)CI->getValue();
else if (const ConstantUInt *CI = dyn_cast<ConstantUInt>(CV))
O << CI->getValue();
else if (const GlobalValue *GV = dyn_cast<GlobalValue>(CV))
// This is a constant address for a global variable or function. Use the
// name of the variable or function as the address value.
O << Mang->getValueName(GV);
else if (const ConstantExpr *CE = dyn_cast<ConstantExpr>(CV)) {
const TargetData &TD = TM.getTargetData();
switch(CE->getOpcode()) {
case Instruction::GetElementPtr: {
// generate a symbolic expression for the byte address
const Constant *ptrVal = CE->getOperand(0);
std::vector<Value*> idxVec(CE->op_begin()+1, CE->op_end());
if (unsigned Offset = TD.getIndexedOffset(ptrVal->getType(), idxVec)) {
O << "(";
emitConstantValueOnly(ptrVal);
O << ") + " << Offset;
} else {
emitConstantValueOnly(ptrVal);
}
break;
}
case Instruction::Cast: {
// Support only non-converting or widening casts for now, that is, ones
// that do not involve a change in value. This assertion is really gross,
// and may not even be a complete check.
Constant *Op = CE->getOperand(0);
const Type *OpTy = Op->getType(), *Ty = CE->getType();
// Remember, kids, pointers on x86 can be losslessly converted back and
// forth into 32-bit or wider integers, regardless of signedness. :-P
assert(((isa<PointerType>(OpTy)
&& (Ty == Type::LongTy || Ty == Type::ULongTy
|| Ty == Type::IntTy || Ty == Type::UIntTy))
|| (isa<PointerType>(Ty)
&& (OpTy == Type::LongTy || OpTy == Type::ULongTy
|| OpTy == Type::IntTy || OpTy == Type::UIntTy))
|| (((TD.getTypeSize(Ty) >= TD.getTypeSize(OpTy))
&& OpTy->isLosslesslyConvertibleTo(Ty))))
&& "FIXME: Don't yet support this kind of constant cast expr");
O << "(";
emitConstantValueOnly(Op);
O << ")";
break;
}
case Instruction::Add:
O << "(";
emitConstantValueOnly(CE->getOperand(0));
O << ") + (";
emitConstantValueOnly(CE->getOperand(1));
O << ")";
break;
default:
assert(0 && "Unsupported operator!");
}
} else {
assert(0 && "Unknown constant value!");
}
}
// Print a constant value or values, with the appropriate storage class as a
// prefix.
void X86AsmPrinter::emitGlobalConstant(const Constant *CV) {
const TargetData &TD = TM.getTargetData();
if (CV->isNullValue()) {
O << "\t.zero\t " << TD.getTypeSize(CV->getType()) << "\n";
return;
} else if (const ConstantArray *CVA = dyn_cast<ConstantArray>(CV)) {
if (CVA->isString()) {
O << "\t.ascii\t";
printAsCString(O, CVA);
O << "\n";
} else { // Not a string. Print the values in successive locations
for (unsigned i = 0, e = CVA->getNumOperands(); i != e; ++i)
emitGlobalConstant(CVA->getOperand(i));
}
return;
} else if (const ConstantStruct *CVS = dyn_cast<ConstantStruct>(CV)) {
// Print the fields in successive locations. Pad to align if needed!
const StructLayout *cvsLayout = TD.getStructLayout(CVS->getType());
unsigned sizeSoFar = 0;
for (unsigned i = 0, e = CVS->getNumOperands(); i != e; ++i) {
const Constant* field = CVS->getOperand(i);
// Check if padding is needed and insert one or more 0s.
unsigned fieldSize = TD.getTypeSize(field->getType());
unsigned padSize = ((i == e-1? cvsLayout->StructSize
: cvsLayout->MemberOffsets[i+1])
- cvsLayout->MemberOffsets[i]) - fieldSize;
sizeSoFar += fieldSize + padSize;
// Now print the actual field value
emitGlobalConstant(field);
// Insert the field padding unless it's zero bytes...
if (padSize)
O << "\t.zero\t " << padSize << "\n";
}
assert(sizeSoFar == cvsLayout->StructSize &&
"Layout of constant struct may be incorrect!");
return;
} else if (const ConstantFP *CFP = dyn_cast<ConstantFP>(CV)) {
// FP Constants are printed as integer constants to avoid losing
// precision...
double Val = CFP->getValue();
switch (CFP->getType()->getTypeID()) {
default: assert(0 && "Unknown floating point type!");
case Type::FloatTyID: {
union FU { // Abide by C TBAA rules
float FVal;
unsigned UVal;
} U;
U.FVal = Val;
O << ".long\t" << U.UVal << "\t# float " << Val << "\n";
return;
}
case Type::DoubleTyID: {
union DU { // Abide by C TBAA rules
double FVal;
uint64_t UVal;
} U;
U.FVal = Val;
O << ".quad\t" << U.UVal << "\t# double " << Val << "\n";
return;
}
}
}
const Type *type = CV->getType();
O << "\t";
switch (type->getTypeID()) {
case Type::BoolTyID: case Type::UByteTyID: case Type::SByteTyID:
O << ".byte";
break;
case Type::UShortTyID: case Type::ShortTyID:
O << ".word";
break;
case Type::FloatTyID: case Type::PointerTyID:
case Type::UIntTyID: case Type::IntTyID:
O << ".long";
break;
case Type::DoubleTyID:
case Type::ULongTyID: case Type::LongTyID:
O << ".quad";
break;
default:
assert (0 && "Can't handle printing this type of thing");
break;
}
O << "\t";
emitConstantValueOnly(CV);
O << "\n";
}
/// printConstantPool - Print to the current output stream assembly
/// representations of the constants in the constant pool MCP. This is
/// used to print out constants which have been "spilled to memory" by
/// the code generator.
///
void X86AsmPrinter::printConstantPool(MachineConstantPool *MCP) {
const std::vector<Constant*> &CP = MCP->getConstants();
const TargetData &TD = TM.getTargetData();
if (CP.empty()) return;
for (unsigned i = 0, e = CP.size(); i != e; ++i) {
O << "\t.section .rodata\n";
O << "\t.align " << (unsigned)TD.getTypeAlignment(CP[i]->getType())
<< "\n";
O << ".CPI" << CurrentFnName << "_" << i << ":\t\t\t\t\t#"
<< *CP[i] << "\n";
emitGlobalConstant(CP[i]);
}
}
/// runOnMachineFunction - This uses the printMachineInstruction()
/// method to print assembly for each instruction.
///
bool X86AsmPrinter::runOnMachineFunction(MachineFunction &MF) {
O << "\n\n";
// What's my mangled name?
CurrentFnName = Mang->getValueName(MF.getFunction());
// Print out constants referenced by the function
printConstantPool(MF.getConstantPool());
// Print out labels for the function.
O << "\t.text\n";
O << "\t.align 16\n";
O << "\t.globl\t" << CurrentFnName << "\n";
if (!EmitCygwin)
O << "\t.type\t" << CurrentFnName << ", @function\n";
O << CurrentFnName << ":\n";
// Print out code for the function.
for (MachineFunction::const_iterator I = MF.begin(), E = MF.end();
I != E; ++I) {
// Print a label for the basic block.
O << ".LBB" << CurrentFnName << "_" << I->getNumber() << ":\t# "
<< I->getBasicBlock()->getName() << "\n";
for (MachineBasicBlock::const_iterator II = I->begin(), E = I->end();
II != E; ++II) {
// Print the assembly for the instruction.
O << "\t";
printMachineInstruction(II);
}
}
// We didn't modify anything.
return false;
}
static bool isScale(const MachineOperand &MO) {
return MO.isImmediate() &&
(MO.getImmedValue() == 1 || MO.getImmedValue() == 2 ||
MO.getImmedValue() == 4 || MO.getImmedValue() == 8);
}
static bool isMem(const MachineInstr *MI, unsigned Op) {
if (MI->getOperand(Op).isFrameIndex()) return true;
if (MI->getOperand(Op).isConstantPoolIndex()) return true;
return Op+4 <= MI->getNumOperands() &&
MI->getOperand(Op ).isRegister() && isScale(MI->getOperand(Op+1)) &&
MI->getOperand(Op+2).isRegister() && MI->getOperand(Op+3).isImmediate();
}
void X86AsmPrinter::printOp(const MachineOperand &MO,
bool elideOffsetKeyword /* = false */) {
const MRegisterInfo &RI = *TM.getRegisterInfo();
switch (MO.getType()) {
case MachineOperand::MO_VirtualRegister:
if (Value *V = MO.getVRegValueOrNull()) {
O << "<" << V->getName() << ">";
return;
}
// FALLTHROUGH
case MachineOperand::MO_MachineRegister:
if (MRegisterInfo::isPhysicalRegister(MO.getReg()))
// Bug Workaround: See note in Printer::doInitialization about %.
O << "%" << RI.get(MO.getReg()).Name;
else
O << "%reg" << MO.getReg();
return;
case MachineOperand::MO_SignExtendedImmed:
case MachineOperand::MO_UnextendedImmed:
O << (int)MO.getImmedValue();
return;
case MachineOperand::MO_MachineBasicBlock: {
MachineBasicBlock *MBBOp = MO.getMachineBasicBlock();
O << ".LBB" << Mang->getValueName(MBBOp->getParent()->getFunction())
<< "_" << MBBOp->getNumber () << "\t# "
<< MBBOp->getBasicBlock ()->getName ();
return;
}
case MachineOperand::MO_PCRelativeDisp:
std::cerr << "Shouldn't use addPCDisp() when building X86 MachineInstrs";
abort ();
return;
case MachineOperand::MO_GlobalAddress:
if (!elideOffsetKeyword)
O << "OFFSET ";
O << Mang->getValueName(MO.getGlobal());
return;
case MachineOperand::MO_ExternalSymbol:
O << MO.getSymbolName();
return;
default:
O << "<unknown operand type>"; return;
}
}
void X86AsmPrinter::printMemReference(const MachineInstr *MI, unsigned Op) {
assert(isMem(MI, Op) && "Invalid memory reference!");
if (MI->getOperand(Op).isFrameIndex()) {
O << "[frame slot #" << MI->getOperand(Op).getFrameIndex();
if (MI->getOperand(Op+3).getImmedValue())
O << " + " << MI->getOperand(Op+3).getImmedValue();
O << "]";
return;
} else if (MI->getOperand(Op).isConstantPoolIndex()) {
O << "[.CPI" << CurrentFnName << "_"
<< MI->getOperand(Op).getConstantPoolIndex();
if (MI->getOperand(Op+3).getImmedValue())
O << " + " << MI->getOperand(Op+3).getImmedValue();
O << "]";
return;
}
const MachineOperand &BaseReg = MI->getOperand(Op);
int ScaleVal = MI->getOperand(Op+1).getImmedValue();
const MachineOperand &IndexReg = MI->getOperand(Op+2);
int DispVal = MI->getOperand(Op+3).getImmedValue();
O << "[";
bool NeedPlus = false;
if (BaseReg.getReg()) {
printOp(BaseReg);
NeedPlus = true;
}
if (IndexReg.getReg()) {
if (NeedPlus) O << " + ";
if (ScaleVal != 1)
O << ScaleVal << "*";
printOp(IndexReg);
NeedPlus = true;
}
if (DispVal) {
if (NeedPlus)
if (DispVal > 0)
O << " + ";
else {
O << " - ";
DispVal = -DispVal;
}
O << DispVal;
}
O << "]";
}
/// printMachineInstruction -- Print out a single X86 LLVM instruction
/// MI in Intel syntax to the current output stream.
///
void X86AsmPrinter::printMachineInstruction(const MachineInstr *MI) {
++EmittedInsts;
// gas bugs:
//
// The 80-bit FP store-pop instruction "fstp XWORD PTR [...]" is misassembled
// by gas in intel_syntax mode as its 32-bit equivalent "fstp DWORD PTR
// [...]". Workaround: Output the raw opcode bytes instead of the instruction.
//
// The 80-bit FP load instruction "fld XWORD PTR [...]" is misassembled by gas
// in intel_syntax mode as its 32-bit equivalent "fld DWORD PTR
// [...]". Workaround: Output the raw opcode bytes instead of the instruction.
//
// gas intel_syntax mode treats "fild QWORD PTR [...]" as an invalid opcode,
// saying "64 bit operations are only supported in 64 bit modes." libopcodes
// disassembles it as "fild DWORD PTR [...]", which is wrong. Workaround:
// Output the raw opcode bytes instead of the instruction.
//
// gas intel_syntax mode treats "fistp QWORD PTR [...]" as an invalid opcode,
// saying "64 bit operations are only supported in 64 bit modes." libopcodes
// disassembles it as "fistpll DWORD PTR [...]", which is wrong. Workaround:
// Output the raw opcode bytes instead of the instruction.
switch (MI->getOpcode()) {
case X86::FSTP80m:
case X86::FLD80m:
case X86::FILD64m:
case X86::FISTP64m:
GasBugWorkaroundEmitter gwe(O);
X86::emitInstruction(gwe, (X86InstrInfo&)*TM.getInstrInfo(), *MI);
O << "\t# ";
}
// Call the autogenerated instruction printer routines.
bool Handled = printInstruction(MI);
if (!Handled) {
MI->dump();
assert(0 && "Do not know how to print this instruction!");
abort();
}
}
bool X86AsmPrinter::doInitialization(Module &M) {
// Tell gas we are outputting Intel syntax (not AT&T syntax) assembly.
//
// Bug: gas in `intel_syntax noprefix' mode interprets the symbol `Sp' in an
// instruction as a reference to the register named sp, and if you try to
// reference a symbol `Sp' (e.g. `mov ECX, OFFSET Sp') then it gets lowercased
// before being looked up in the symbol table. This creates spurious
// `undefined symbol' errors when linking. Workaround: Do not use `noprefix'
// mode, and decorate all register names with percent signs.
O << "\t.intel_syntax\n";
Mang = new Mangler(M, EmitCygwin);
return false; // success
}
// SwitchSection - Switch to the specified section of the executable if we are
// not already in it!
//
static void SwitchSection(std::ostream &OS, std::string &CurSection,
const char *NewSection) {
if (CurSection != NewSection) {
CurSection = NewSection;
if (!CurSection.empty())
OS << "\t" << NewSection << "\n";
}
}
bool X86AsmPrinter::doFinalization(Module &M) {
const TargetData &TD = TM.getTargetData();
std::string CurSection;
// Print out module-level global variables here.
for (Module::const_giterator I = M.gbegin(), E = M.gend(); I != E; ++I)
if (I->hasInitializer()) { // External global require no code
O << "\n\n";
std::string name = Mang->getValueName(I);
Constant *C = I->getInitializer();
unsigned Size = TD.getTypeSize(C->getType());
unsigned Align = TD.getTypeAlignment(C->getType());
if (C->isNullValue() &&
(I->hasLinkOnceLinkage() || I->hasInternalLinkage() ||
I->hasWeakLinkage() /* FIXME: Verify correct */)) {
SwitchSection(O, CurSection, ".data");
if (I->hasInternalLinkage())
O << "\t.local " << name << "\n";
O << "\t.comm " << name << "," << TD.getTypeSize(C->getType())
<< "," << (unsigned)TD.getTypeAlignment(C->getType());
O << "\t\t# ";
WriteAsOperand(O, I, true, true, &M);
O << "\n";
} else {
switch (I->getLinkage()) {
case GlobalValue::LinkOnceLinkage:
case GlobalValue::WeakLinkage: // FIXME: Verify correct for weak.
// Nonnull linkonce -> weak
O << "\t.weak " << name << "\n";
SwitchSection(O, CurSection, "");
O << "\t.section\t.llvm.linkonce.d." << name << ",\"aw\",@progbits\n";
break;
case GlobalValue::AppendingLinkage:
// FIXME: appending linkage variables should go into a section of
// their name or something. For now, just emit them as external.
case GlobalValue::ExternalLinkage:
// If external or appending, declare as a global symbol
O << "\t.globl " << name << "\n";
// FALL THROUGH
case GlobalValue::InternalLinkage:
if (C->isNullValue())
SwitchSection(O, CurSection, ".bss");
else
SwitchSection(O, CurSection, ".data");
break;
}
O << "\t.align " << Align << "\n";
O << "\t.type " << name << ",@object\n";
O << "\t.size " << name << "," << Size << "\n";
O << name << ":\t\t\t\t# ";
WriteAsOperand(O, I, true, true, &M);
O << " = ";
WriteAsOperand(O, C, false, false, &M);
O << "\n";
emitGlobalConstant(C);
}
}
delete Mang;
return false; // success
}
Reference in New Issue View Git Blame Copy Permalink