llvm-6502/lib/Target/X86/X86InstrInfo.cpp
Nate Begeman f63be7d395 First round of support for doing scalar FP using the SSE2 ISA extension and
XMM registers.  There are many known deficiencies and fixmes, which will be
addressed ASAP.  The major benefit of this work is that it will allow the
LLVM register allocator to allocate FP registers across basic blocks.

The x86 backend will still default to x87 style FP.  To enable this work,
you must pass -enable-sse-scalar-fp and either -sse2 or -sse3 to llc.

An example before and after would be for:
double foo(double *P) { double Sum = 0; int i; for (i = 0; i < 1000; ++i)
                        Sum += P[i]; return Sum; }

The inner loop looks like the following:
x87:
.LBB_foo_1:     # no_exit
        fldl (%esp)
        faddl (%eax,%ecx,8)
        fstpl (%esp)
        incl %ecx
        cmpl $1000, %ecx
        #FP_REG_KILL
        jne .LBB_foo_1  # no_exit

SSE2:
        addsd (%eax,%ecx,8), %xmm0
        incl %ecx
        cmpl $1000, %ecx
        #FP_REG_KILL
        jne .LBB_foo_1  # no_exit


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22340 91177308-0d34-0410-b5e6-96231b3b80d8
2005-07-06 18:59:04 +00:00

186 lines
7.3 KiB
C++

//===- X86InstrInfo.cpp - X86 Instruction Information -----------*- C++ -*-===//
//
// 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 the X86 implementation of the TargetInstrInfo class.
//
//===----------------------------------------------------------------------===//
#include "X86InstrInfo.h"
#include "X86.h"
#include "X86InstrBuilder.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "X86GenInstrInfo.inc"
using namespace llvm;
X86InstrInfo::X86InstrInfo()
: TargetInstrInfo(X86Insts, sizeof(X86Insts)/sizeof(X86Insts[0])) {
}
bool X86InstrInfo::isMoveInstr(const MachineInstr& MI,
unsigned& sourceReg,
unsigned& destReg) const {
MachineOpCode oc = MI.getOpcode();
if (oc == X86::MOV8rr || oc == X86::MOV16rr || oc == X86::MOV32rr ||
oc == X86::FpMOV || oc == X86::MOVAPDrr) {
assert(MI.getNumOperands() == 2 &&
MI.getOperand(0).isRegister() &&
MI.getOperand(1).isRegister() &&
"invalid register-register move instruction");
sourceReg = MI.getOperand(1).getReg();
destReg = MI.getOperand(0).getReg();
return true;
}
return false;
}
/// convertToThreeAddress - This method must be implemented by targets that
/// set the M_CONVERTIBLE_TO_3_ADDR flag. When this flag is set, the target
/// may be able to convert a two-address instruction into a true
/// three-address instruction on demand. This allows the X86 target (for
/// example) to convert ADD and SHL instructions into LEA instructions if they
/// would require register copies due to two-addressness.
///
/// This method returns a null pointer if the transformation cannot be
/// performed, otherwise it returns the new instruction.
///
MachineInstr *X86InstrInfo::convertToThreeAddress(MachineInstr *MI) const {
// All instructions input are two-addr instructions. Get the known operands.
unsigned Dest = MI->getOperand(0).getReg();
unsigned Src = MI->getOperand(1).getReg();
// FIXME: None of these instructions are promotable to LEAs without
// additional information. In particular, LEA doesn't set the flags that
// add and inc do. :(
return 0;
// FIXME: 16-bit LEA's are really slow on Athlons, but not bad on P4's. When
// we have subtarget support, enable the 16-bit LEA generation here.
bool DisableLEA16 = true;
switch (MI->getOpcode()) {
case X86::INC32r:
assert(MI->getNumOperands() == 2 && "Unknown inc instruction!");
return addRegOffset(BuildMI(X86::LEA32r, 5, Dest), Src, 1);
case X86::INC16r:
if (DisableLEA16) return 0;
assert(MI->getNumOperands() == 2 && "Unknown inc instruction!");
return addRegOffset(BuildMI(X86::LEA16r, 5, Dest), Src, 1);
case X86::DEC32r:
assert(MI->getNumOperands() == 2 && "Unknown dec instruction!");
return addRegOffset(BuildMI(X86::LEA32r, 5, Dest), Src, -1);
case X86::DEC16r:
if (DisableLEA16) return 0;
assert(MI->getNumOperands() == 2 && "Unknown dec instruction!");
return addRegOffset(BuildMI(X86::LEA16r, 5, Dest), Src, -1);
case X86::ADD32rr:
assert(MI->getNumOperands() == 3 && "Unknown add instruction!");
return addRegReg(BuildMI(X86::LEA32r, 5, Dest), Src,
MI->getOperand(2).getReg());
case X86::ADD16rr:
if (DisableLEA16) return 0;
assert(MI->getNumOperands() == 3 && "Unknown add instruction!");
return addRegReg(BuildMI(X86::LEA16r, 5, Dest), Src,
MI->getOperand(2).getReg());
case X86::ADD32ri:
assert(MI->getNumOperands() == 3 && "Unknown add instruction!");
if (MI->getOperand(2).isImmediate())
return addRegOffset(BuildMI(X86::LEA32r, 5, Dest), Src,
MI->getOperand(2).getImmedValue());
return 0;
case X86::ADD16ri:
if (DisableLEA16) return 0;
assert(MI->getNumOperands() == 3 && "Unknown add instruction!");
if (MI->getOperand(2).isImmediate())
return addRegOffset(BuildMI(X86::LEA16r, 5, Dest), Src,
MI->getOperand(2).getImmedValue());
break;
case X86::SHL16ri:
if (DisableLEA16) return 0;
case X86::SHL32ri:
assert(MI->getNumOperands() == 3 && MI->getOperand(2).isImmediate() &&
"Unknown shl instruction!");
unsigned ShAmt = MI->getOperand(2).getImmedValue();
if (ShAmt == 1 || ShAmt == 2 || ShAmt == 3) {
X86AddressMode AM;
AM.Scale = 1 << ShAmt;
AM.IndexReg = Src;
unsigned Opc = MI->getOpcode() == X86::SHL32ri ? X86::LEA32r :X86::LEA16r;
return addFullAddress(BuildMI(Opc, 5, Dest), AM);
}
break;
}
return 0;
}
/// commuteInstruction - We have a few instructions that must be hacked on to
/// commute them.
///
MachineInstr *X86InstrInfo::commuteInstruction(MachineInstr *MI) const {
switch (MI->getOpcode()) {
case X86::SHRD16rri8: // A = SHRD16rri8 B, C, I -> A = SHLD16rri8 C, B, (16-I)
case X86::SHLD16rri8: // A = SHLD16rri8 B, C, I -> A = SHRD16rri8 C, B, (16-I)
case X86::SHRD32rri8: // A = SHRD32rri8 B, C, I -> A = SHLD32rri8 C, B, (32-I)
case X86::SHLD32rri8:{// A = SHLD32rri8 B, C, I -> A = SHRD32rri8 C, B, (32-I)
unsigned Opc;
unsigned Size;
switch (MI->getOpcode()) {
default: assert(0 && "Unreachable!");
case X86::SHRD16rri8: Size = 16; Opc = X86::SHLD16rri8; break;
case X86::SHLD16rri8: Size = 16; Opc = X86::SHRD16rri8; break;
case X86::SHRD32rri8: Size = 32; Opc = X86::SHLD32rri8; break;
case X86::SHLD32rri8: Size = 32; Opc = X86::SHRD32rri8; break;
}
unsigned Amt = MI->getOperand(3).getImmedValue();
unsigned A = MI->getOperand(0).getReg();
unsigned B = MI->getOperand(1).getReg();
unsigned C = MI->getOperand(2).getReg();
return BuildMI(Opc, 3, A).addReg(C).addReg(B).addImm(Size-Amt);
}
default:
return TargetInstrInfo::commuteInstruction(MI);
}
}
void X86InstrInfo::insertGoto(MachineBasicBlock& MBB,
MachineBasicBlock& TMBB) const {
BuildMI(MBB, MBB.end(), X86::JMP, 1).addMBB(&TMBB);
}
MachineBasicBlock::iterator
X86InstrInfo::reverseBranchCondition(MachineBasicBlock::iterator MI) const {
unsigned Opcode = MI->getOpcode();
assert(isBranch(Opcode) && "MachineInstr must be a branch");
unsigned ROpcode;
switch (Opcode) {
default: assert(0 && "Cannot reverse unconditional branches!");
case X86::JB: ROpcode = X86::JAE; break;
case X86::JAE: ROpcode = X86::JB; break;
case X86::JE: ROpcode = X86::JNE; break;
case X86::JNE: ROpcode = X86::JE; break;
case X86::JBE: ROpcode = X86::JA; break;
case X86::JA: ROpcode = X86::JBE; break;
case X86::JS: ROpcode = X86::JNS; break;
case X86::JNS: ROpcode = X86::JS; break;
case X86::JP: ROpcode = X86::JNP; break;
case X86::JNP: ROpcode = X86::JP; break;
case X86::JL: ROpcode = X86::JGE; break;
case X86::JGE: ROpcode = X86::JL; break;
case X86::JLE: ROpcode = X86::JG; break;
case X86::JG: ROpcode = X86::JLE; break;
}
MachineBasicBlock* MBB = MI->getParent();
MachineBasicBlock* TMBB = MI->getOperand(0).getMachineBasicBlock();
return BuildMI(*MBB, MBB->erase(MI), ROpcode, 1).addMBB(TMBB);
}