#include "X86.h"
#include "X86InstrInfo.h"
-#include "llvm/Pass.h"
#include "llvm/Function.h"
+#include "llvm/Constant.h"
#include "llvm/Target/TargetMachine.h"
-#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/CodeGen/MachineConstantPool.h"
#include "llvm/CodeGen/MachineInstr.h"
#include "Support/Statistic.h"
namespace {
- struct Printer : public FunctionPass {
- TargetMachine &TM;
+ struct Printer : public MachineFunctionPass {
std::ostream &O;
+ unsigned ConstIdx;
+ Printer(std::ostream &o) : O(o), ConstIdx(0) {}
- Printer(TargetMachine &tm, std::ostream &o) : TM(tm), O(o) {}
+ virtual const char *getPassName() const {
+ return "X86 Assembly Printer";
+ }
- bool runOnFunction(Function &F);
+ void printConstantPool(MachineConstantPool *MCP, const TargetData &TD);
+ bool runOnMachineFunction(MachineFunction &F);
};
}
/// the specified stream. This function should work regardless of whether or
/// not the function is in SSA form or not.
///
-Pass *createX86CodePrinterPass(TargetMachine &TM, std::ostream &O) {
- return new Printer(TM, O);
+Pass *createX86CodePrinterPass(std::ostream &O) {
+ return new Printer(O);
}
+// printConstantPool - Print out any constants which have been spilled to
+// memory...
+void Printer::printConstantPool(MachineConstantPool *MCP, const TargetData &TD){
+ const std::vector<Constant*> &CP = MCP->getConstants();
+ 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" << i+ConstIdx << ":\t\t\t\t\t;" << *CP[i] << "\n";
+ O << "\t*Constant output not implemented yet!*\n\n";
+ }
+ ConstIdx += CP.size(); // Don't recycle constant pool index numbers
+}
+
/// runOnFunction - This uses the X86InstructionInfo::print method
/// to print assembly for each instruction.
-bool Printer::runOnFunction (Function & F)
-{
- static unsigned bbnumber = 0;
- MachineFunction & MF = MachineFunction::get (&F);
- const MachineInstrInfo & MII = TM.getInstrInfo ();
+bool Printer::runOnMachineFunction(MachineFunction &MF) {
+ static unsigned BBNumber = 0;
+ const TargetMachine &TM = MF.getTarget();
+ const TargetInstrInfo &TII = TM.getInstrInfo();
- O << "; x86 printing only sorta implemented so far!\n";
+ // Print out constants referenced by the function
+ printConstantPool(MF.getConstantPool(), TM.getTargetData());
// Print out labels for the function.
- O << "\t.globl\t" << F.getName () << "\n";
- O << "\t.type\t" << F.getName () << ", @function\n";
- O << F.getName () << ":\n";
+ O << "\t.text\n";
+ O << "\t.align 16\n";
+ O << "\t.globl\t" << MF.getFunction()->getName() << "\n";
+ O << "\t.type\t" << MF.getFunction()->getName() << ", @function\n";
+ O << MF.getFunction()->getName() << ":\n";
// Print out code for the function.
- for (MachineFunction::const_iterator bb_i = MF.begin (), bb_e = MF.end ();
- bb_i != bb_e; ++bb_i)
- {
- // Print a label for the basic block.
- O << ".BB" << bbnumber++ << ":\n";
- for (MachineBasicBlock::const_iterator i_i = bb_i->begin (), i_e =
- bb_i->end (); i_i != i_e; ++i_i)
- {
- // Print the assembly for the instruction.
- O << "\t";
- MII.print(*i_i, O, TM);
- }
+ for (MachineFunction::const_iterator I = MF.begin(), E = MF.end();
+ I != E; ++I) {
+ // Print a label for the basic block.
+ O << ".BB" << BBNumber++ << ":\n";
+ for (MachineBasicBlock::const_iterator II = I->begin(), E = I->end();
+ II != E; ++II) {
+ // Print the assembly for the instruction.
+ O << "\t";
+ TII.print(*II, O, TM);
}
+ }
// 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();
+}
+
static void printOp(std::ostream &O, const MachineOperand &MO,
const MRegisterInfo &RI) {
switch (MO.getType()) {
case MachineOperand::MO_VirtualRegister:
+ if (Value *V = MO.getVRegValueOrNull()) {
+ O << "<" << V->getName() << ">";
+ return;
+ }
+ // FALLTHROUGH
case MachineOperand::MO_MachineRegister:
if (MO.getReg() < MRegisterInfo::FirstVirtualRegister)
O << RI.get(MO.getReg()).Name;
case MachineOperand::MO_UnextendedImmed:
O << (int)MO.getImmedValue();
return;
+ case MachineOperand::MO_PCRelativeDisp:
+ O << "<" << MO.getVRegValue()->getName() << ">";
+ return;
+ case MachineOperand::MO_GlobalAddress:
+ O << "<" << MO.getGlobal()->getName() << ">";
+ return;
+ case MachineOperand::MO_ExternalSymbol:
+ O << "<" << MO.getSymbolName() << ">";
+ return;
default:
O << "<unknown op ty>"; return;
}
}
-static inline void toHexDigit(std::ostream &O, unsigned char V) {
- if (V >= 10)
- O << (char)('A'+V-10);
- else
- O << (char)('0'+V);
-}
-
-static std::ostream &toHex(std::ostream &O, unsigned char V) {
- toHexDigit(O, V >> 4);
- toHexDigit(O, V & 0xF);
- return O;
-}
-
-static std::ostream &emitConstant(std::ostream &O, unsigned Val, unsigned Size){
- // Output the constant in little endian byte order...
- for (unsigned i = 0; i != Size; ++i) {
- toHex(O, Val) << " ";
- Val >>= 8;
+static const std::string sizePtr(const TargetInstrDescriptor &Desc) {
+ switch (Desc.TSFlags & X86II::ArgMask) {
+ default: assert(0 && "Unknown arg size!");
+ case X86II::Arg8: return "BYTE PTR";
+ case X86II::Arg16: return "WORD PTR";
+ case X86II::Arg32: return "DWORD PTR";
+ case X86II::Arg64: return "QWORD PTR";
+ case X86II::ArgF32: return "DWORD PTR";
+ case X86II::ArgF64: return "QWORD PTR";
+ case X86II::ArgF80: return "XWORD PTR";
}
- return O;
}
+static void printMemReference(std::ostream &O, const MachineInstr *MI,
+ unsigned Op, const MRegisterInfo &RI) {
+ assert(isMem(MI, Op) && "Invalid memory reference!");
-static bool isReg(const MachineOperand &MO) {
- return MO.getType() == MachineOperand::MO_VirtualRegister ||
- MO.getType() == MachineOperand::MO_MachineRegister;
-}
-
-static bool isImmediate(const MachineOperand &MO) {
- return MO.getType() == MachineOperand::MO_SignExtendedImmed ||
- MO.getType() == MachineOperand::MO_UnextendedImmed;
-}
+ 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" << 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();
-// getX86RegNum - This function maps LLVM register identifiers to their X86
-// specific numbering, which is used in various places encoding instructions.
-//
-static unsigned getX86RegNum(unsigned RegNo) {
- switch(RegNo) {
- case X86::EAX: case X86::AX: case X86::AL: return 0;
- case X86::ECX: case X86::CX: case X86::CL: return 1;
- case X86::EDX: case X86::DX: case X86::DL: return 2;
- case X86::EBX: case X86::BX: case X86::BL: return 3;
- case X86::ESP: case X86::SP: case X86::AH: return 4;
- case X86::EBP: case X86::BP: case X86::CH: return 5;
- case X86::ESI: case X86::SI: case X86::DH: return 6;
- case X86::EDI: case X86::DI: case X86::BH: return 7;
- default:
- assert(RegNo >= MRegisterInfo::FirstVirtualRegister &&
- "Unknown physical register!");
- DEBUG(std::cerr << "Register allocator hasn't allocated " << RegNo
- << " correctly yet!\n");
- return 0;
+ O << "[";
+ bool NeedPlus = false;
+ if (BaseReg.getReg()) {
+ printOp(O, BaseReg, RI);
+ NeedPlus = true;
}
-}
-inline static unsigned char ModRMByte(unsigned Mod, unsigned RegOpcode,
- unsigned RM) {
- assert(Mod < 4 && RegOpcode < 8 && RM < 8 && "ModRM Fields out of range!");
- return RM | (RegOpcode << 3) | (Mod << 6);
-}
+ if (IndexReg.getReg()) {
+ if (NeedPlus) O << " + ";
+ if (ScaleVal != 1)
+ O << ScaleVal << "*";
+ printOp(O, IndexReg, RI);
+ NeedPlus = true;
+ }
-static unsigned char regModRMByte(unsigned ModRMReg, unsigned RegOpcodeField) {
- return ModRMByte(3, RegOpcodeField, getX86RegNum(ModRMReg));
+ if (DispVal) {
+ if (NeedPlus)
+ if (DispVal > 0)
+ O << " + ";
+ else {
+ O << " - ";
+ DispVal = -DispVal;
+ }
+ O << DispVal;
+ }
+ O << "]";
}
-
// print - Print out an x86 instruction in intel syntax
void X86InstrInfo::print(const MachineInstr *MI, std::ostream &O,
const TargetMachine &TM) const {
unsigned Opcode = MI->getOpcode();
- const MachineInstrDescriptor &Desc = get(Opcode);
-
- // Print instruction prefixes if neccesary
-
- if (Desc.TSFlags & X86II::OpSize) O << "66 "; // Operand size...
- if (Desc.TSFlags & X86II::TB) O << "0F "; // Two-byte opcode prefix
+ const TargetInstrDescriptor &Desc = get(Opcode);
switch (Desc.TSFlags & X86II::FormMask) {
- case X86II::OtherFrm:
- O << "\t\t\t";
- O << "-"; MI->print(O, TM);
- break;
- case X86II::RawFrm:
- toHex(O, getBaseOpcodeFor(Opcode));
- O << "\n\t\t\t\t";
- O << getName(MI->getOpCode()) << " ";
-
- for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
- if (i) O << ", ";
- printOp(O, MI->getOperand(i), RI);
+ case X86II::Pseudo:
+ if (Opcode == X86::PHI) {
+ printOp(O, MI->getOperand(0), RI);
+ O << " = phi ";
+ for (unsigned i = 1, e = MI->getNumOperands(); i != e; i+=2) {
+ if (i != 1) O << ", ";
+ O << "[";
+ printOp(O, MI->getOperand(i), RI);
+ O << ", ";
+ printOp(O, MI->getOperand(i+1), RI);
+ O << "]";
+ }
+ } else {
+ unsigned i = 0;
+ if (MI->getNumOperands() && (MI->getOperand(0).opIsDefOnly() ||
+ MI->getOperand(0).opIsDefAndUse())) {
+ printOp(O, MI->getOperand(0), RI);
+ O << " = ";
+ ++i;
+ }
+ O << getName(MI->getOpcode());
+
+ for (unsigned e = MI->getNumOperands(); i != e; ++i) {
+ O << " ";
+ if (MI->getOperand(i).opIsDefOnly() ||
+ MI->getOperand(i).opIsDefAndUse()) O << "*";
+ printOp(O, MI->getOperand(i), RI);
+ if (MI->getOperand(i).opIsDefOnly() ||
+ MI->getOperand(i).opIsDefAndUse()) O << "*";
+ }
}
O << "\n";
return;
+ case X86II::RawFrm:
+ // The accepted forms of Raw instructions are:
+ // 1. nop - No operand required
+ // 2. jmp foo - PC relative displacement operand
+ // 3. call bar - GlobalAddress Operand or External Symbol Operand
+ //
+ assert(MI->getNumOperands() == 0 ||
+ (MI->getNumOperands() == 1 &&
+ (MI->getOperand(0).isPCRelativeDisp() ||
+ MI->getOperand(0).isGlobalAddress() ||
+ MI->getOperand(0).isExternalSymbol())) &&
+ "Illegal raw instruction!");
+ O << getName(MI->getOpcode()) << " ";
+
+ if (MI->getNumOperands() == 1) {
+ printOp(O, MI->getOperand(0), RI);
+ }
+ O << "\n";
+ return;
case X86II::AddRegFrm: {
// There are currently two forms of acceptable AddRegFrm instructions.
// Either the instruction JUST takes a single register (like inc, dec, etc),
// or it takes a register and an immediate of the same size as the register
- // (move immediate f.e.).
+ // (move immediate f.e.). Note that this immediate value might be stored as
+ // an LLVM value, to represent, for example, loading the address of a global
+ // into a register. The initial register might be duplicated if this is a
+ // M_2_ADDR_REG instruction
//
- assert(isReg(MI->getOperand(0)) &&
+ assert(MI->getOperand(0).isRegister() &&
(MI->getNumOperands() == 1 ||
- (MI->getNumOperands() == 2 && isImmediate(MI->getOperand(1)))) &&
+ (MI->getNumOperands() == 2 &&
+ (MI->getOperand(1).getVRegValueOrNull() ||
+ MI->getOperand(1).isImmediate() ||
+ MI->getOperand(1).isRegister() ||
+ MI->getOperand(1).isGlobalAddress() ||
+ MI->getOperand(1).isExternalSymbol()))) &&
"Illegal form for AddRegFrm instruction!");
unsigned Reg = MI->getOperand(0).getReg();
- toHex(O, getBaseOpcodeFor(Opcode) + getX86RegNum(Reg)) << " ";
-
- if (MI->getNumOperands() == 2) {
- unsigned Size = 4;
- emitConstant(O, MI->getOperand(1).getImmedValue(), Size);
- }
- O << "\n\t\t\t\t";
O << getName(MI->getOpCode()) << " ";
printOp(O, MI->getOperand(0), RI);
- if (MI->getNumOperands() == 2) {
+ if (MI->getNumOperands() == 2 &&
+ (!MI->getOperand(1).isRegister() ||
+ MI->getOperand(1).getVRegValueOrNull() ||
+ MI->getOperand(1).isGlobalAddress() ||
+ MI->getOperand(1).isExternalSymbol())) {
O << ", ";
- printOp(O, MI->getOperand(MI->getNumOperands()-1), RI);
+ printOp(O, MI->getOperand(1), RI);
}
O << "\n";
return;
}
case X86II::MRMDestReg: {
- // There are two acceptable forms of MRMDestReg instructions, those with 3
- // and 2 operands:
+ // There are two acceptable forms of MRMDestReg instructions, those with 2,
+ // 3 and 4 operands:
+ //
+ // 2 Operands: this is for things like mov that do not read a second input
//
// 3 Operands: in this form, the first two registers (the destination, and
// the first operand) should be the same, post register allocation. The 3rd
// operand is an additional input. This should be for things like add
// instructions.
//
- // 2 Operands: this is for things like mov that do not read a second input
+ // 4 Operands: This form is for instructions which are 3 operands forms, but
+ // have a constant argument as well.
//
- assert(isReg(MI->getOperand(0)) &&
- (MI->getNumOperands() == 2 ||
- (MI->getNumOperands() == 3 && isReg(MI->getOperand(1)))) &&
- isReg(MI->getOperand(MI->getNumOperands()-1))
+ bool isTwoAddr = isTwoAddrInstr(Opcode);
+ assert(MI->getOperand(0).isRegister() &&
+ (MI->getNumOperands() == 2 ||
+ (isTwoAddr && MI->getOperand(1).isRegister() &&
+ MI->getOperand(0).getReg() == MI->getOperand(1).getReg() &&
+ (MI->getNumOperands() == 3 ||
+ (MI->getNumOperands() == 4 && MI->getOperand(3).isImmediate()))))
&& "Bad format for MRMDestReg!");
- if (MI->getNumOperands() == 3 &&
- MI->getOperand(0).getReg() != MI->getOperand(1).getReg())
- O << "**";
-
- toHex(O, getBaseOpcodeFor(Opcode)) << " ";
- unsigned ModRMReg = MI->getOperand(0).getReg();
- unsigned ExtraReg = MI->getOperand(MI->getNumOperands()-1).getReg();
- toHex(O, regModRMByte(ModRMReg, getX86RegNum(ExtraReg)));
- O << "\n\t\t\t\t";
O << getName(MI->getOpCode()) << " ";
printOp(O, MI->getOperand(0), RI);
O << ", ";
- printOp(O, MI->getOperand(MI->getNumOperands()-1), RI);
+ printOp(O, MI->getOperand(1+isTwoAddr), RI);
+ if (MI->getNumOperands() == 4) {
+ O << ", ";
+ printOp(O, MI->getOperand(3), RI);
+ }
+ O << "\n";
+ return;
+ }
+
+ case X86II::MRMDestMem: {
+ // These instructions are the same as MRMDestReg, but instead of having a
+ // register reference for the mod/rm field, it's a memory reference.
+ //
+ assert(isMem(MI, 0) && MI->getNumOperands() == 4+1 &&
+ MI->getOperand(4).isRegister() && "Bad format for MRMDestMem!");
+
+ O << getName(MI->getOpCode()) << " " << sizePtr(Desc) << " ";
+ printMemReference(O, MI, 0, RI);
+ O << ", ";
+ printOp(O, MI->getOperand(4), RI);
O << "\n";
return;
}
+
case X86II::MRMSrcReg: {
// There is a two forms that are acceptable for MRMSrcReg instructions,
// those with 3 and 2 operands:
//
// 2 Operands: this is for things like mov that do not read a second input
//
- assert(isReg(MI->getOperand(0)) &&
- isReg(MI->getOperand(1)) &&
+ assert(MI->getOperand(0).isRegister() &&
+ MI->getOperand(1).isRegister() &&
(MI->getNumOperands() == 2 ||
- (MI->getNumOperands() == 3 && isReg(MI->getOperand(2))))
- && "Bad format for MRMDestReg!");
+ (MI->getNumOperands() == 3 && MI->getOperand(2).isRegister()))
+ && "Bad format for MRMSrcReg!");
if (MI->getNumOperands() == 3 &&
MI->getOperand(0).getReg() != MI->getOperand(1).getReg())
O << "**";
- toHex(O, getBaseOpcodeFor(Opcode)) << " ";
- unsigned ModRMReg = MI->getOperand(MI->getNumOperands()-1).getReg();
- unsigned ExtraReg = MI->getOperand(0).getReg();
- toHex(O, regModRMByte(ModRMReg, getX86RegNum(ExtraReg)));
-
- O << "\n\t\t\t\t";
O << getName(MI->getOpCode()) << " ";
printOp(O, MI->getOperand(0), RI);
O << ", ";
O << "\n";
return;
}
- case X86II::MRMDestMem:
- case X86II::MRMSrcMem:
+
+ case X86II::MRMSrcMem: {
+ // These instructions are the same as MRMSrcReg, but instead of having a
+ // register reference for the mod/rm field, it's a memory reference.
+ //
+ assert(MI->getOperand(0).isRegister() &&
+ (MI->getNumOperands() == 1+4 && isMem(MI, 1)) ||
+ (MI->getNumOperands() == 2+4 && MI->getOperand(1).isRegister() &&
+ isMem(MI, 2))
+ && "Bad format for MRMDestReg!");
+ if (MI->getNumOperands() == 2+4 &&
+ MI->getOperand(0).getReg() != MI->getOperand(1).getReg())
+ O << "**";
+
+ O << getName(MI->getOpCode()) << " ";
+ printOp(O, MI->getOperand(0), RI);
+ O << ", " << sizePtr(Desc) << " ";
+ printMemReference(O, MI, MI->getNumOperands()-4, RI);
+ O << "\n";
+ return;
+ }
+
+ case X86II::MRMS0r: case X86II::MRMS1r:
+ case X86II::MRMS2r: case X86II::MRMS3r:
+ case X86II::MRMS4r: case X86II::MRMS5r:
+ case X86II::MRMS6r: case X86II::MRMS7r: {
+ // In this form, the following are valid formats:
+ // 1. sete r
+ // 2. cmp reg, immediate
+ // 2. shl rdest, rinput <implicit CL or 1>
+ // 3. sbb rdest, rinput, immediate [rdest = rinput]
+ //
+ assert(MI->getNumOperands() > 0 && MI->getNumOperands() < 4 &&
+ MI->getOperand(0).isRegister() && "Bad MRMSxR format!");
+ assert((MI->getNumOperands() != 2 ||
+ MI->getOperand(1).isRegister() || MI->getOperand(1).isImmediate())&&
+ "Bad MRMSxR format!");
+ assert((MI->getNumOperands() < 3 ||
+ (MI->getOperand(1).isRegister() && MI->getOperand(2).isImmediate())) &&
+ "Bad MRMSxR format!");
+
+ if (MI->getNumOperands() > 1 && MI->getOperand(1).isRegister() &&
+ MI->getOperand(0).getReg() != MI->getOperand(1).getReg())
+ O << "**";
+
+ O << getName(MI->getOpCode()) << " ";
+ printOp(O, MI->getOperand(0), RI);
+ if (MI->getOperand(MI->getNumOperands()-1).isImmediate()) {
+ O << ", ";
+ printOp(O, MI->getOperand(MI->getNumOperands()-1), RI);
+ }
+ O << "\n";
+
+ return;
+ }
+
+ case X86II::MRMS0m: case X86II::MRMS1m:
+ case X86II::MRMS2m: case X86II::MRMS3m:
+ case X86II::MRMS4m: case X86II::MRMS5m:
+ case X86II::MRMS6m: case X86II::MRMS7m: {
+ // In this form, the following are valid formats:
+ // 1. sete [m]
+ // 2. cmp [m], immediate
+ // 2. shl [m], rinput <implicit CL or 1>
+ // 3. sbb [m], immediate
+ //
+ assert(MI->getNumOperands() >= 4 && MI->getNumOperands() <= 5 &&
+ isMem(MI, 0) && "Bad MRMSxM format!");
+ assert((MI->getNumOperands() != 5 || MI->getOperand(4).isImmediate()) &&
+ "Bad MRMSxM format!");
+
+ O << getName(MI->getOpCode()) << " ";
+ O << sizePtr(Desc) << " ";
+ printMemReference(O, MI, 0, RI);
+ if (MI->getNumOperands() == 5) {
+ O << ", ";
+ printOp(O, MI->getOperand(4), RI);
+ }
+ O << "\n";
+ return;
+ }
+
default:
- O << "\t\t\t-"; MI->print(O, TM); break;
+ O << "\tUNKNOWN FORM:\t\t-"; MI->print(O, TM); break;
}
}