void X86InstrInfo::anchor() {}
X86InstrInfo::X86InstrInfo(X86Subtarget &STI)
- : X86GenInstrInfo(
- (STI.isTarget64BitLP64() ? X86::ADJCALLSTACKDOWN64 : X86::ADJCALLSTACKDOWN32),
- (STI.isTarget64BitLP64() ? X86::ADJCALLSTACKUP64 : X86::ADJCALLSTACKUP32)),
+ : X86GenInstrInfo((STI.isTarget64BitLP64() ? X86::ADJCALLSTACKDOWN64
+ : X86::ADJCALLSTACKDOWN32),
+ (STI.isTarget64BitLP64() ? X86::ADJCALLSTACKUP64
+ : X86::ADJCALLSTACKUP32),
+ X86::CATCHRET),
Subtarget(STI), RI(STI.getTargetTriple()) {
static const X86MemoryFoldTableEntry MemoryFoldTable2Addr[] = {
static const X86MemoryFoldTableEntry MemoryFoldTable3[] = {
// FMA foldable instructions
{ X86::VFMADDSSr231r, X86::VFMADDSSr231m, TB_ALIGN_NONE },
+ { X86::VFMADDSSr231r_Int, X86::VFMADDSSr231m_Int, TB_ALIGN_NONE },
{ X86::VFMADDSDr231r, X86::VFMADDSDr231m, TB_ALIGN_NONE },
+ { X86::VFMADDSDr231r_Int, X86::VFMADDSDr231m_Int, TB_ALIGN_NONE },
{ X86::VFMADDSSr132r, X86::VFMADDSSr132m, TB_ALIGN_NONE },
+ { X86::VFMADDSSr132r_Int, X86::VFMADDSSr132m_Int, TB_ALIGN_NONE },
{ X86::VFMADDSDr132r, X86::VFMADDSDr132m, TB_ALIGN_NONE },
+ { X86::VFMADDSDr132r_Int, X86::VFMADDSDr132m_Int, TB_ALIGN_NONE },
{ X86::VFMADDSSr213r, X86::VFMADDSSr213m, TB_ALIGN_NONE },
+ { X86::VFMADDSSr213r_Int, X86::VFMADDSSr213m_Int, TB_ALIGN_NONE },
{ X86::VFMADDSDr213r, X86::VFMADDSDr213m, TB_ALIGN_NONE },
+ { X86::VFMADDSDr213r_Int, X86::VFMADDSDr213m_Int, TB_ALIGN_NONE },
{ X86::VFMADDPSr231r, X86::VFMADDPSr231m, TB_ALIGN_NONE },
{ X86::VFMADDPDr231r, X86::VFMADDPDr231m, TB_ALIGN_NONE },
{ X86::VFMADDPDr213rY, X86::VFMADDPDr213mY, TB_ALIGN_NONE },
{ X86::VFNMADDSSr231r, X86::VFNMADDSSr231m, TB_ALIGN_NONE },
+ { X86::VFNMADDSSr231r_Int, X86::VFNMADDSSr231m_Int, TB_ALIGN_NONE },
{ X86::VFNMADDSDr231r, X86::VFNMADDSDr231m, TB_ALIGN_NONE },
+ { X86::VFNMADDSDr231r_Int, X86::VFNMADDSDr231m_Int, TB_ALIGN_NONE },
{ X86::VFNMADDSSr132r, X86::VFNMADDSSr132m, TB_ALIGN_NONE },
+ { X86::VFNMADDSSr132r_Int, X86::VFNMADDSSr132m_Int, TB_ALIGN_NONE },
{ X86::VFNMADDSDr132r, X86::VFNMADDSDr132m, TB_ALIGN_NONE },
+ { X86::VFNMADDSDr132r_Int, X86::VFNMADDSDr132m_Int, TB_ALIGN_NONE },
{ X86::VFNMADDSSr213r, X86::VFNMADDSSr213m, TB_ALIGN_NONE },
+ { X86::VFNMADDSSr213r_Int, X86::VFNMADDSSr213m_Int, TB_ALIGN_NONE },
{ X86::VFNMADDSDr213r, X86::VFNMADDSDr213m, TB_ALIGN_NONE },
+ { X86::VFNMADDSDr213r_Int, X86::VFNMADDSDr213m_Int, TB_ALIGN_NONE },
{ X86::VFNMADDPSr231r, X86::VFNMADDPSr231m, TB_ALIGN_NONE },
{ X86::VFNMADDPDr231r, X86::VFNMADDPDr231m, TB_ALIGN_NONE },
{ X86::VFNMADDPDr213rY, X86::VFNMADDPDr213mY, TB_ALIGN_NONE },
{ X86::VFMSUBSSr231r, X86::VFMSUBSSr231m, TB_ALIGN_NONE },
+ { X86::VFMSUBSSr231r_Int, X86::VFMSUBSSr231m_Int, TB_ALIGN_NONE },
{ X86::VFMSUBSDr231r, X86::VFMSUBSDr231m, TB_ALIGN_NONE },
+ { X86::VFMSUBSDr231r_Int, X86::VFMSUBSDr231m_Int, TB_ALIGN_NONE },
{ X86::VFMSUBSSr132r, X86::VFMSUBSSr132m, TB_ALIGN_NONE },
+ { X86::VFMSUBSSr132r_Int, X86::VFMSUBSSr132m_Int, TB_ALIGN_NONE },
{ X86::VFMSUBSDr132r, X86::VFMSUBSDr132m, TB_ALIGN_NONE },
+ { X86::VFMSUBSDr132r_Int, X86::VFMSUBSDr132m_Int, TB_ALIGN_NONE },
{ X86::VFMSUBSSr213r, X86::VFMSUBSSr213m, TB_ALIGN_NONE },
+ { X86::VFMSUBSSr213r_Int, X86::VFMSUBSSr213m_Int, TB_ALIGN_NONE },
{ X86::VFMSUBSDr213r, X86::VFMSUBSDr213m, TB_ALIGN_NONE },
+ { X86::VFMSUBSDr213r_Int, X86::VFMSUBSDr213m_Int, TB_ALIGN_NONE },
{ X86::VFMSUBPSr231r, X86::VFMSUBPSr231m, TB_ALIGN_NONE },
{ X86::VFMSUBPDr231r, X86::VFMSUBPDr231m, TB_ALIGN_NONE },
{ X86::VFMSUBPDr213rY, X86::VFMSUBPDr213mY, TB_ALIGN_NONE },
{ X86::VFNMSUBSSr231r, X86::VFNMSUBSSr231m, TB_ALIGN_NONE },
+ { X86::VFNMSUBSSr231r_Int, X86::VFNMSUBSSr231m_Int, TB_ALIGN_NONE },
{ X86::VFNMSUBSDr231r, X86::VFNMSUBSDr231m, TB_ALIGN_NONE },
+ { X86::VFNMSUBSDr231r_Int, X86::VFNMSUBSDr231m_Int, TB_ALIGN_NONE },
{ X86::VFNMSUBSSr132r, X86::VFNMSUBSSr132m, TB_ALIGN_NONE },
+ { X86::VFNMSUBSSr132r_Int, X86::VFNMSUBSSr132m_Int, TB_ALIGN_NONE },
{ X86::VFNMSUBSDr132r, X86::VFNMSUBSDr132m, TB_ALIGN_NONE },
+ { X86::VFNMSUBSDr132r_Int, X86::VFNMSUBSDr132m_Int, TB_ALIGN_NONE },
{ X86::VFNMSUBSSr213r, X86::VFNMSUBSSr213m, TB_ALIGN_NONE },
+ { X86::VFNMSUBSSr213r_Int, X86::VFNMSUBSSr213m_Int, TB_ALIGN_NONE },
{ X86::VFNMSUBSDr213r, X86::VFNMSUBSDr213m, TB_ALIGN_NONE },
+ { X86::VFNMSUBSDr213r_Int, X86::VFNMSUBSDr213m_Int, TB_ALIGN_NONE },
{ X86::VFNMSUBPSr231r, X86::VFNMSUBPSr231m, TB_ALIGN_NONE },
{ X86::VFNMSUBPDr231r, X86::VFNMSUBPDr231m, TB_ALIGN_NONE },
case X86::FsVMOVAPSrm:
case X86::FsVMOVAPDrm:
case X86::FsMOVAPSrm:
- case X86::FsMOVAPDrm: {
+ case X86::FsMOVAPDrm:
+ // AVX-512
+ case X86::VMOVAPDZ128rm:
+ case X86::VMOVAPDZ256rm:
+ case X86::VMOVAPDZrm:
+ case X86::VMOVAPSZ128rm:
+ case X86::VMOVAPSZ256rm:
+ case X86::VMOVAPSZrm:
+ case X86::VMOVDQA32Z128rm:
+ case X86::VMOVDQA32Z256rm:
+ case X86::VMOVDQA32Zrm:
+ case X86::VMOVDQA64Z128rm:
+ case X86::VMOVDQA64Z256rm:
+ case X86::VMOVDQA64Zrm:
+ case X86::VMOVDQU16Z128rm:
+ case X86::VMOVDQU16Z256rm:
+ case X86::VMOVDQU16Zrm:
+ case X86::VMOVDQU32Z128rm:
+ case X86::VMOVDQU32Z256rm:
+ case X86::VMOVDQU32Zrm:
+ case X86::VMOVDQU64Z128rm:
+ case X86::VMOVDQU64Z256rm:
+ case X86::VMOVDQU64Zrm:
+ case X86::VMOVDQU8Z128rm:
+ case X86::VMOVDQU8Z256rm:
+ case X86::VMOVDQU8Zrm:
+ case X86::VMOVUPSZ128rm:
+ case X86::VMOVUPSZ256rm:
+ case X86::VMOVUPSZrm: {
// Loads from constant pools are trivially rematerializable.
if (MI->getOperand(1+X86::AddrBaseReg).isReg() &&
MI->getOperand(1+X86::AddrScaleAmt).isImm() &&
}
/// True if MI has a condition code def, e.g. EFLAGS, that is not marked dead.
-static bool hasLiveCondCodeDef(MachineInstr *MI) {
+bool X86InstrInfo::hasLiveCondCodeDef(MachineInstr *MI) const {
for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
MachineOperand &MO = MI->getOperand(i);
if (MO.isReg() && MO.isDef() &&
return NewMI;
}
-/// We have a few instructions that must be hacked on to commute them.
-///
-MachineInstr *
-X86InstrInfo::commuteInstruction(MachineInstr *MI, bool NewMI) const {
+/// Returns true if the given instruction opcode is FMA3.
+/// Otherwise, returns false.
+static bool isFMA3(unsigned Opcode) {
+ switch (Opcode) {
+ case X86::VFMADDSDr132r: case X86::VFMADDSDr132m:
+ case X86::VFMADDSSr132r: case X86::VFMADDSSr132m:
+ case X86::VFMSUBSDr132r: case X86::VFMSUBSDr132m:
+ case X86::VFMSUBSSr132r: case X86::VFMSUBSSr132m:
+ case X86::VFNMADDSDr132r: case X86::VFNMADDSDr132m:
+ case X86::VFNMADDSSr132r: case X86::VFNMADDSSr132m:
+ case X86::VFNMSUBSDr132r: case X86::VFNMSUBSDr132m:
+ case X86::VFNMSUBSSr132r: case X86::VFNMSUBSSr132m:
+
+ case X86::VFMADDSDr213r: case X86::VFMADDSDr213m:
+ case X86::VFMADDSSr213r: case X86::VFMADDSSr213m:
+ case X86::VFMSUBSDr213r: case X86::VFMSUBSDr213m:
+ case X86::VFMSUBSSr213r: case X86::VFMSUBSSr213m:
+ case X86::VFNMADDSDr213r: case X86::VFNMADDSDr213m:
+ case X86::VFNMADDSSr213r: case X86::VFNMADDSSr213m:
+ case X86::VFNMSUBSDr213r: case X86::VFNMSUBSDr213m:
+ case X86::VFNMSUBSSr213r: case X86::VFNMSUBSSr213m:
+
+ case X86::VFMADDSDr231r: case X86::VFMADDSDr231m:
+ case X86::VFMADDSSr231r: case X86::VFMADDSSr231m:
+ case X86::VFMSUBSDr231r: case X86::VFMSUBSDr231m:
+ case X86::VFMSUBSSr231r: case X86::VFMSUBSSr231m:
+ case X86::VFNMADDSDr231r: case X86::VFNMADDSDr231m:
+ case X86::VFNMADDSSr231r: case X86::VFNMADDSSr231m:
+ case X86::VFNMSUBSDr231r: case X86::VFNMSUBSDr231m:
+ case X86::VFNMSUBSSr231r: case X86::VFNMSUBSSr231m:
+
+ case X86::VFMADDSUBPDr132r: case X86::VFMADDSUBPDr132m:
+ case X86::VFMADDSUBPSr132r: case X86::VFMADDSUBPSr132m:
+ case X86::VFMSUBADDPDr132r: case X86::VFMSUBADDPDr132m:
+ case X86::VFMSUBADDPSr132r: case X86::VFMSUBADDPSr132m:
+ case X86::VFMADDSUBPDr132rY: case X86::VFMADDSUBPDr132mY:
+ case X86::VFMADDSUBPSr132rY: case X86::VFMADDSUBPSr132mY:
+ case X86::VFMSUBADDPDr132rY: case X86::VFMSUBADDPDr132mY:
+ case X86::VFMSUBADDPSr132rY: case X86::VFMSUBADDPSr132mY:
+
+ case X86::VFMADDPDr132r: case X86::VFMADDPDr132m:
+ case X86::VFMADDPSr132r: case X86::VFMADDPSr132m:
+ case X86::VFMSUBPDr132r: case X86::VFMSUBPDr132m:
+ case X86::VFMSUBPSr132r: case X86::VFMSUBPSr132m:
+ case X86::VFNMADDPDr132r: case X86::VFNMADDPDr132m:
+ case X86::VFNMADDPSr132r: case X86::VFNMADDPSr132m:
+ case X86::VFNMSUBPDr132r: case X86::VFNMSUBPDr132m:
+ case X86::VFNMSUBPSr132r: case X86::VFNMSUBPSr132m:
+ case X86::VFMADDPDr132rY: case X86::VFMADDPDr132mY:
+ case X86::VFMADDPSr132rY: case X86::VFMADDPSr132mY:
+ case X86::VFMSUBPDr132rY: case X86::VFMSUBPDr132mY:
+ case X86::VFMSUBPSr132rY: case X86::VFMSUBPSr132mY:
+ case X86::VFNMADDPDr132rY: case X86::VFNMADDPDr132mY:
+ case X86::VFNMADDPSr132rY: case X86::VFNMADDPSr132mY:
+ case X86::VFNMSUBPDr132rY: case X86::VFNMSUBPDr132mY:
+ case X86::VFNMSUBPSr132rY: case X86::VFNMSUBPSr132mY:
+
+ case X86::VFMADDSUBPDr213r: case X86::VFMADDSUBPDr213m:
+ case X86::VFMADDSUBPSr213r: case X86::VFMADDSUBPSr213m:
+ case X86::VFMSUBADDPDr213r: case X86::VFMSUBADDPDr213m:
+ case X86::VFMSUBADDPSr213r: case X86::VFMSUBADDPSr213m:
+ case X86::VFMADDSUBPDr213rY: case X86::VFMADDSUBPDr213mY:
+ case X86::VFMADDSUBPSr213rY: case X86::VFMADDSUBPSr213mY:
+ case X86::VFMSUBADDPDr213rY: case X86::VFMSUBADDPDr213mY:
+ case X86::VFMSUBADDPSr213rY: case X86::VFMSUBADDPSr213mY:
+
+ case X86::VFMADDPDr213r: case X86::VFMADDPDr213m:
+ case X86::VFMADDPSr213r: case X86::VFMADDPSr213m:
+ case X86::VFMSUBPDr213r: case X86::VFMSUBPDr213m:
+ case X86::VFMSUBPSr213r: case X86::VFMSUBPSr213m:
+ case X86::VFNMADDPDr213r: case X86::VFNMADDPDr213m:
+ case X86::VFNMADDPSr213r: case X86::VFNMADDPSr213m:
+ case X86::VFNMSUBPDr213r: case X86::VFNMSUBPDr213m:
+ case X86::VFNMSUBPSr213r: case X86::VFNMSUBPSr213m:
+ case X86::VFMADDPDr213rY: case X86::VFMADDPDr213mY:
+ case X86::VFMADDPSr213rY: case X86::VFMADDPSr213mY:
+ case X86::VFMSUBPDr213rY: case X86::VFMSUBPDr213mY:
+ case X86::VFMSUBPSr213rY: case X86::VFMSUBPSr213mY:
+ case X86::VFNMADDPDr213rY: case X86::VFNMADDPDr213mY:
+ case X86::VFNMADDPSr213rY: case X86::VFNMADDPSr213mY:
+ case X86::VFNMSUBPDr213rY: case X86::VFNMSUBPDr213mY:
+ case X86::VFNMSUBPSr213rY: case X86::VFNMSUBPSr213mY:
+
+ case X86::VFMADDSUBPDr231r: case X86::VFMADDSUBPDr231m:
+ case X86::VFMADDSUBPSr231r: case X86::VFMADDSUBPSr231m:
+ case X86::VFMSUBADDPDr231r: case X86::VFMSUBADDPDr231m:
+ case X86::VFMSUBADDPSr231r: case X86::VFMSUBADDPSr231m:
+ case X86::VFMADDSUBPDr231rY: case X86::VFMADDSUBPDr231mY:
+ case X86::VFMADDSUBPSr231rY: case X86::VFMADDSUBPSr231mY:
+ case X86::VFMSUBADDPDr231rY: case X86::VFMSUBADDPDr231mY:
+ case X86::VFMSUBADDPSr231rY: case X86::VFMSUBADDPSr231mY:
+
+ case X86::VFMADDPDr231r: case X86::VFMADDPDr231m:
+ case X86::VFMADDPSr231r: case X86::VFMADDPSr231m:
+ case X86::VFMSUBPDr231r: case X86::VFMSUBPDr231m:
+ case X86::VFMSUBPSr231r: case X86::VFMSUBPSr231m:
+ case X86::VFNMADDPDr231r: case X86::VFNMADDPDr231m:
+ case X86::VFNMADDPSr231r: case X86::VFNMADDPSr231m:
+ case X86::VFNMSUBPDr231r: case X86::VFNMSUBPDr231m:
+ case X86::VFNMSUBPSr231r: case X86::VFNMSUBPSr231m:
+ case X86::VFMADDPDr231rY: case X86::VFMADDPDr231mY:
+ case X86::VFMADDPSr231rY: case X86::VFMADDPSr231mY:
+ case X86::VFMSUBPDr231rY: case X86::VFMSUBPDr231mY:
+ case X86::VFMSUBPSr231rY: case X86::VFMSUBPSr231mY:
+ case X86::VFNMADDPDr231rY: case X86::VFNMADDPDr231mY:
+ case X86::VFNMADDPSr231rY: case X86::VFNMADDPSr231mY:
+ case X86::VFNMSUBPDr231rY: case X86::VFNMSUBPDr231mY:
+ case X86::VFNMSUBPSr231rY: case X86::VFNMSUBPSr231mY:
+ return true;
+ default:
+ return false;
+ }
+ llvm_unreachable("Opcode not handled by the switch");
+}
+
+MachineInstr *X86InstrInfo::commuteInstructionImpl(MachineInstr *MI,
+ bool NewMI,
+ unsigned OpIdx1,
+ unsigned OpIdx2) 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)
}
MI->setDesc(get(Opc));
MI->getOperand(3).setImm(Size-Amt);
- return TargetInstrInfo::commuteInstruction(MI, NewMI);
+ return TargetInstrInfo::commuteInstructionImpl(MI, NewMI, OpIdx1, OpIdx2);
}
case X86::BLENDPDrri:
case X86::BLENDPSrri:
NewMI = false;
}
MI->getOperand(3).setImm(Mask ^ Imm);
- return TargetInstrInfo::commuteInstruction(MI, NewMI);
+ return TargetInstrInfo::commuteInstructionImpl(MI, NewMI, OpIdx1, OpIdx2);
}
case X86::PCLMULQDQrr:
case X86::VPCLMULQDQrr:{
NewMI = false;
}
MI->getOperand(3).setImm((Src1Hi << 4) | (Src2Hi >> 4));
- return TargetInstrInfo::commuteInstruction(MI, NewMI);
+ return TargetInstrInfo::commuteInstructionImpl(MI, NewMI, OpIdx1, OpIdx2);
}
case X86::CMPPDrri:
case X86::CMPPSrri:
MI = MF.CloneMachineInstr(MI);
NewMI = false;
}
- return TargetInstrInfo::commuteInstruction(MI, NewMI);
+ return TargetInstrInfo::commuteInstructionImpl(MI, NewMI, OpIdx1, OpIdx2);
default:
return nullptr;
}
NewMI = false;
}
MI->getOperand(3).setImm(Imm);
- return TargetInstrInfo::commuteInstruction(MI, NewMI);
+ return TargetInstrInfo::commuteInstructionImpl(MI, NewMI, OpIdx1, OpIdx2);
}
case X86::CMOVB16rr: case X86::CMOVB32rr: case X86::CMOVB64rr:
case X86::CMOVAE16rr: case X86::CMOVAE32rr: case X86::CMOVAE64rr:
// Fallthrough intended.
}
default:
- return TargetInstrInfo::commuteInstruction(MI, NewMI);
+ if (isFMA3(MI->getOpcode())) {
+ unsigned Opc = getFMA3OpcodeToCommuteOperands(MI, OpIdx1, OpIdx2);
+ if (Opc == 0)
+ return nullptr;
+ if (NewMI) {
+ MachineFunction &MF = *MI->getParent()->getParent();
+ MI = MF.CloneMachineInstr(MI);
+ NewMI = false;
+ }
+ MI->setDesc(get(Opc));
+ }
+ return TargetInstrInfo::commuteInstructionImpl(MI, NewMI, OpIdx1, OpIdx2);
}
}
-bool X86InstrInfo::findCommutedOpIndices(MachineInstr *MI, unsigned &SrcOpIdx1,
+bool X86InstrInfo::findFMA3CommutedOpIndices(MachineInstr *MI,
+ unsigned &SrcOpIdx1,
+ unsigned &SrcOpIdx2) const {
+
+ unsigned RegOpsNum = isMem(MI, 3) ? 2 : 3;
+
+ // Only the first RegOpsNum operands are commutable.
+ // Also, the value 'CommuteAnyOperandIndex' is valid here as it means
+ // that the operand is not specified/fixed.
+ if (SrcOpIdx1 != CommuteAnyOperandIndex &&
+ (SrcOpIdx1 < 1 || SrcOpIdx1 > RegOpsNum))
+ return false;
+ if (SrcOpIdx2 != CommuteAnyOperandIndex &&
+ (SrcOpIdx2 < 1 || SrcOpIdx2 > RegOpsNum))
+ return false;
+
+ // Look for two different register operands assumed to be commutable
+ // regardless of the FMA opcode. The FMA opcode is adjusted later.
+ if (SrcOpIdx1 == CommuteAnyOperandIndex ||
+ SrcOpIdx2 == CommuteAnyOperandIndex) {
+ unsigned CommutableOpIdx1 = SrcOpIdx1;
+ unsigned CommutableOpIdx2 = SrcOpIdx2;
+
+ // At least one of operands to be commuted is not specified and
+ // this method is free to choose appropriate commutable operands.
+ if (SrcOpIdx1 == SrcOpIdx2)
+ // Both of operands are not fixed. By default set one of commutable
+ // operands to the last register operand of the instruction.
+ CommutableOpIdx2 = RegOpsNum;
+ else if (SrcOpIdx2 == CommuteAnyOperandIndex)
+ // Only one of operands is not fixed.
+ CommutableOpIdx2 = SrcOpIdx1;
+
+ // CommutableOpIdx2 is well defined now. Let's choose another commutable
+ // operand and assign its index to CommutableOpIdx1.
+ unsigned Op2Reg = MI->getOperand(CommutableOpIdx2).getReg();
+ for (CommutableOpIdx1 = RegOpsNum; CommutableOpIdx1 > 0; CommutableOpIdx1--) {
+ // The commuted operands must have different registers.
+ // Otherwise, the commute transformation does not change anything and
+ // is useless then.
+ if (Op2Reg != MI->getOperand(CommutableOpIdx1).getReg())
+ break;
+ }
+
+ // No appropriate commutable operands were found.
+ if (CommutableOpIdx1 == 0)
+ return false;
+
+ // Assign the found pair of commutable indices to SrcOpIdx1 and SrcOpidx2
+ // to return those values.
+ if (!fixCommutedOpIndices(SrcOpIdx1, SrcOpIdx2,
+ CommutableOpIdx1, CommutableOpIdx2))
+ return false;
+ }
+
+ // Check if we can adjust the opcode to preserve the semantics when
+ // commute the register operands.
+ return getFMA3OpcodeToCommuteOperands(MI, SrcOpIdx1, SrcOpIdx2) != 0;
+}
+
+unsigned X86InstrInfo::getFMA3OpcodeToCommuteOperands(MachineInstr *MI,
+ unsigned SrcOpIdx1,
+ unsigned SrcOpIdx2) const {
+ unsigned Opc = MI->getOpcode();
+
+ // Define the array that holds FMA opcodes in groups
+ // of 3 opcodes(132, 213, 231) in each group.
+ static const unsigned OpcodeGroups[][3] = {
+ { X86::VFMADDSSr132r, X86::VFMADDSSr213r, X86::VFMADDSSr231r },
+ { X86::VFMADDSDr132r, X86::VFMADDSDr213r, X86::VFMADDSDr231r },
+ { X86::VFMADDPSr132r, X86::VFMADDPSr213r, X86::VFMADDPSr231r },
+ { X86::VFMADDPDr132r, X86::VFMADDPDr213r, X86::VFMADDPDr231r },
+ { X86::VFMADDPSr132rY, X86::VFMADDPSr213rY, X86::VFMADDPSr231rY },
+ { X86::VFMADDPDr132rY, X86::VFMADDPDr213rY, X86::VFMADDPDr231rY },
+ { X86::VFMADDSSr132m, X86::VFMADDSSr213m, X86::VFMADDSSr231m },
+ { X86::VFMADDSDr132m, X86::VFMADDSDr213m, X86::VFMADDSDr231m },
+ { X86::VFMADDPSr132m, X86::VFMADDPSr213m, X86::VFMADDPSr231m },
+ { X86::VFMADDPDr132m, X86::VFMADDPDr213m, X86::VFMADDPDr231m },
+ { X86::VFMADDPSr132mY, X86::VFMADDPSr213mY, X86::VFMADDPSr231mY },
+ { X86::VFMADDPDr132mY, X86::VFMADDPDr213mY, X86::VFMADDPDr231mY },
+
+ { X86::VFMSUBSSr132r, X86::VFMSUBSSr213r, X86::VFMSUBSSr231r },
+ { X86::VFMSUBSDr132r, X86::VFMSUBSDr213r, X86::VFMSUBSDr231r },
+ { X86::VFMSUBPSr132r, X86::VFMSUBPSr213r, X86::VFMSUBPSr231r },
+ { X86::VFMSUBPDr132r, X86::VFMSUBPDr213r, X86::VFMSUBPDr231r },
+ { X86::VFMSUBPSr132rY, X86::VFMSUBPSr213rY, X86::VFMSUBPSr231rY },
+ { X86::VFMSUBPDr132rY, X86::VFMSUBPDr213rY, X86::VFMSUBPDr231rY },
+ { X86::VFMSUBSSr132m, X86::VFMSUBSSr213m, X86::VFMSUBSSr231m },
+ { X86::VFMSUBSDr132m, X86::VFMSUBSDr213m, X86::VFMSUBSDr231m },
+ { X86::VFMSUBPSr132m, X86::VFMSUBPSr213m, X86::VFMSUBPSr231m },
+ { X86::VFMSUBPDr132m, X86::VFMSUBPDr213m, X86::VFMSUBPDr231m },
+ { X86::VFMSUBPSr132mY, X86::VFMSUBPSr213mY, X86::VFMSUBPSr231mY },
+ { X86::VFMSUBPDr132mY, X86::VFMSUBPDr213mY, X86::VFMSUBPDr231mY },
+
+ { X86::VFNMADDSSr132r, X86::VFNMADDSSr213r, X86::VFNMADDSSr231r },
+ { X86::VFNMADDSDr132r, X86::VFNMADDSDr213r, X86::VFNMADDSDr231r },
+ { X86::VFNMADDPSr132r, X86::VFNMADDPSr213r, X86::VFNMADDPSr231r },
+ { X86::VFNMADDPDr132r, X86::VFNMADDPDr213r, X86::VFNMADDPDr231r },
+ { X86::VFNMADDPSr132rY, X86::VFNMADDPSr213rY, X86::VFNMADDPSr231rY },
+ { X86::VFNMADDPDr132rY, X86::VFNMADDPDr213rY, X86::VFNMADDPDr231rY },
+ { X86::VFNMADDSSr132m, X86::VFNMADDSSr213m, X86::VFNMADDSSr231m },
+ { X86::VFNMADDSDr132m, X86::VFNMADDSDr213m, X86::VFNMADDSDr231m },
+ { X86::VFNMADDPSr132m, X86::VFNMADDPSr213m, X86::VFNMADDPSr231m },
+ { X86::VFNMADDPDr132m, X86::VFNMADDPDr213m, X86::VFNMADDPDr231m },
+ { X86::VFNMADDPSr132mY, X86::VFNMADDPSr213mY, X86::VFNMADDPSr231mY },
+ { X86::VFNMADDPDr132mY, X86::VFNMADDPDr213mY, X86::VFNMADDPDr231mY },
+
+ { X86::VFNMSUBSSr132r, X86::VFNMSUBSSr213r, X86::VFNMSUBSSr231r },
+ { X86::VFNMSUBSDr132r, X86::VFNMSUBSDr213r, X86::VFNMSUBSDr231r },
+ { X86::VFNMSUBPSr132r, X86::VFNMSUBPSr213r, X86::VFNMSUBPSr231r },
+ { X86::VFNMSUBPDr132r, X86::VFNMSUBPDr213r, X86::VFNMSUBPDr231r },
+ { X86::VFNMSUBPSr132rY, X86::VFNMSUBPSr213rY, X86::VFNMSUBPSr231rY },
+ { X86::VFNMSUBPDr132rY, X86::VFNMSUBPDr213rY, X86::VFNMSUBPDr231rY },
+ { X86::VFNMSUBSSr132m, X86::VFNMSUBSSr213m, X86::VFNMSUBSSr231m },
+ { X86::VFNMSUBSDr132m, X86::VFNMSUBSDr213m, X86::VFNMSUBSDr231m },
+ { X86::VFNMSUBPSr132m, X86::VFNMSUBPSr213m, X86::VFNMSUBPSr231m },
+ { X86::VFNMSUBPDr132m, X86::VFNMSUBPDr213m, X86::VFNMSUBPDr231m },
+ { X86::VFNMSUBPSr132mY, X86::VFNMSUBPSr213mY, X86::VFNMSUBPSr231mY },
+ { X86::VFNMSUBPDr132mY, X86::VFNMSUBPDr213mY, X86::VFNMSUBPDr231mY },
+
+ { X86::VFMADDSUBPSr132r, X86::VFMADDSUBPSr213r, X86::VFMADDSUBPSr231r },
+ { X86::VFMADDSUBPDr132r, X86::VFMADDSUBPDr213r, X86::VFMADDSUBPDr231r },
+ { X86::VFMADDSUBPSr132rY, X86::VFMADDSUBPSr213rY, X86::VFMADDSUBPSr231rY },
+ { X86::VFMADDSUBPDr132rY, X86::VFMADDSUBPDr213rY, X86::VFMADDSUBPDr231rY },
+ { X86::VFMADDSUBPSr132m, X86::VFMADDSUBPSr213m, X86::VFMADDSUBPSr231m },
+ { X86::VFMADDSUBPDr132m, X86::VFMADDSUBPDr213m, X86::VFMADDSUBPDr231m },
+ { X86::VFMADDSUBPSr132mY, X86::VFMADDSUBPSr213mY, X86::VFMADDSUBPSr231mY },
+ { X86::VFMADDSUBPDr132mY, X86::VFMADDSUBPDr213mY, X86::VFMADDSUBPDr231mY },
+
+ { X86::VFMSUBADDPSr132r, X86::VFMSUBADDPSr213r, X86::VFMSUBADDPSr231r },
+ { X86::VFMSUBADDPDr132r, X86::VFMSUBADDPDr213r, X86::VFMSUBADDPDr231r },
+ { X86::VFMSUBADDPSr132rY, X86::VFMSUBADDPSr213rY, X86::VFMSUBADDPSr231rY },
+ { X86::VFMSUBADDPDr132rY, X86::VFMSUBADDPDr213rY, X86::VFMSUBADDPDr231rY },
+ { X86::VFMSUBADDPSr132m, X86::VFMSUBADDPSr213m, X86::VFMSUBADDPSr231m },
+ { X86::VFMSUBADDPDr132m, X86::VFMSUBADDPDr213m, X86::VFMSUBADDPDr231m },
+ { X86::VFMSUBADDPSr132mY, X86::VFMSUBADDPSr213mY, X86::VFMSUBADDPSr231mY },
+ { X86::VFMSUBADDPDr132mY, X86::VFMSUBADDPDr213mY, X86::VFMSUBADDPDr231mY }
+ };
+ const unsigned Form132Index = 0;
+ const unsigned Form213Index = 1;
+ const unsigned Form231Index = 2;
+ const unsigned FormsNum = 3;
+
+ // Look for the input opcode in the OpcodeGroups table.
+ unsigned OpcodeGroupsNum = sizeof(OpcodeGroups) / sizeof(OpcodeGroups[0]);
+ unsigned GroupIndex = 0, FormIndex = FormsNum;
+ for (; GroupIndex < OpcodeGroupsNum && FormIndex == FormsNum; GroupIndex++) {
+ for (FormIndex = 0; FormIndex < FormsNum; FormIndex++) {
+ if (OpcodeGroups[GroupIndex][FormIndex] == Opc)
+ break;
+ }
+ }
+ // Input opcode does not match with any of the opcodes from the table.
+ if (FormIndex == FormsNum)
+ return 0;
+ // Do not forget to fix the GroupIndex after the loop.
+ GroupIndex--;
+
+ // Put the lowest index to SrcOpIdx1 to simplify the checks below.
+ if (SrcOpIdx1 > SrcOpIdx2)
+ std::swap(SrcOpIdx1, SrcOpIdx2);
+
+ unsigned Case;
+ if (SrcOpIdx1 == 1 && SrcOpIdx2 == 2)
+ Case = 0;
+ else if (SrcOpIdx1 == 1 && SrcOpIdx2 == 3)
+ Case = 1;
+ else if (SrcOpIdx1 == 2 && SrcOpIdx2 == 3)
+ Case = 2;
+ else
+ return 0;
+
+ // Define the FMA forms mapping array that helps to map input FMA form
+ // to output FMA form to preserve the operation semantics after
+ // commuting the operands.
+ static const unsigned FormMapping[][3] = {
+ // 0: SrcOpIdx1 == 1 && SrcOpIdx2 == 2;
+ // FMA132 A, C, b; ==> FMA231 C, A, b;
+ // FMA213 B, A, c; ==> FMA213 A, B, c;
+ // FMA231 C, A, b; ==> FMA132 A, C, b;
+ { Form231Index, Form213Index, Form132Index },
+ // 1: SrcOpIdx1 == 1 && SrcOpIdx2 == 3;
+ // FMA132 A, c, B; ==> FMA132 B, c, A;
+ // FMA213 B, a, C; ==> FMA231 C, a, B;
+ // FMA231 C, a, B; ==> FMA213 B, a, C;
+ { Form132Index, Form231Index, Form213Index },
+ // 2: SrcOpIdx1 == 2 && SrcOpIdx2 == 3;
+ // FMA132 a, C, B; ==> FMA213 a, B, C;
+ // FMA213 b, A, C; ==> FMA132 b, C, A;
+ // FMA231 c, A, B; ==> FMA231 c, B, A;
+ { Form213Index, Form132Index, Form231Index }
+ };
+
+ // Everything is ready, just adjust the FMA opcode and return it.
+ FormIndex = FormMapping[Case][FormIndex];
+ return OpcodeGroups[GroupIndex][FormIndex];
+}
+
+bool X86InstrInfo::findCommutedOpIndices(MachineInstr *MI,
+ unsigned &SrcOpIdx1,
unsigned &SrcOpIdx2) const {
switch (MI->getOpcode()) {
case X86::CMPPDrri:
// Ordered/Unordered/Equal/NotEqual tests
unsigned Imm = MI->getOperand(3).getImm() & 0x7;
switch (Imm) {
- case 0x00: // EQUAL
- case 0x03: // UNORDERED
- case 0x04: // NOT EQUAL
- case 0x07: // ORDERED
- SrcOpIdx1 = 1;
- SrcOpIdx2 = 2;
- return true;
+ case 0x00: // EQUAL
+ case 0x03: // UNORDERED
+ case 0x04: // NOT EQUAL
+ case 0x07: // ORDERED
+ // The indices of the commutable operands are 1 and 2.
+ // Assign them to the returned operand indices here.
+ return fixCommutedOpIndices(SrcOpIdx1, SrcOpIdx2, 1, 2);
}
return false;
}
- case X86::VFMADDPDr231r:
- case X86::VFMADDPSr231r:
- case X86::VFMADDSDr231r:
- case X86::VFMADDSSr231r:
- case X86::VFMSUBPDr231r:
- case X86::VFMSUBPSr231r:
- case X86::VFMSUBSDr231r:
- case X86::VFMSUBSSr231r:
- case X86::VFNMADDPDr231r:
- case X86::VFNMADDPSr231r:
- case X86::VFNMADDSDr231r:
- case X86::VFNMADDSSr231r:
- case X86::VFNMSUBPDr231r:
- case X86::VFNMSUBPSr231r:
- case X86::VFNMSUBSDr231r:
- case X86::VFNMSUBSSr231r:
- case X86::VFMADDPDr231rY:
- case X86::VFMADDPSr231rY:
- case X86::VFMSUBPDr231rY:
- case X86::VFMSUBPSr231rY:
- case X86::VFNMADDPDr231rY:
- case X86::VFNMADDPSr231rY:
- case X86::VFNMSUBPDr231rY:
- case X86::VFNMSUBPSr231rY:
- SrcOpIdx1 = 2;
- SrcOpIdx2 = 3;
- return true;
default:
+ if (isFMA3(MI->getOpcode()))
+ return findFMA3CommutedOpIndices(MI, SrcOpIdx1, SrcOpIdx2);
return TargetInstrInfo::findCommutedOpIndices(MI, SrcOpIdx1, SrcOpIdx2);
}
+ return false;
}
static X86::CondCode getCondFromBranchOpc(unsigned BrOpc) {
return true;
case X86::KSET0B:
case X86::KSET0W: return Expand2AddrUndef(MIB, get(X86::KXORWrr));
+ case X86::KSET0D: return Expand2AddrUndef(MIB, get(X86::KXORDrr));
+ case X86::KSET0Q: return Expand2AddrUndef(MIB, get(X86::KXORQrr));
case X86::KSET1B:
case X86::KSET1W: return Expand2AddrUndef(MIB, get(X86::KXNORWrr));
+ case X86::KSET1D: return Expand2AddrUndef(MIB, get(X86::KXNORDrr));
+ case X86::KSET1Q: return Expand2AddrUndef(MIB, get(X86::KXNORQrr));
case TargetOpcode::LOAD_STACK_GUARD:
expandLoadStackGuard(MIB, *this);
return true;
return false;
}
-static void addOperands(MachineInstrBuilder &MIB, ArrayRef<MachineOperand> MOs) {
+static void addOperands(MachineInstrBuilder &MIB, ArrayRef<MachineOperand> MOs,
+ int PtrOffset = 0) {
unsigned NumAddrOps = MOs.size();
- for (unsigned i = 0; i != NumAddrOps; ++i)
- MIB.addOperand(MOs[i]);
- if (NumAddrOps < 4) // FrameIndex only
- addOffset(MIB, 0);
+
+ if (NumAddrOps < 4) {
+ // FrameIndex only - add an immediate offset (whether its zero or not).
+ for (unsigned i = 0; i != NumAddrOps; ++i)
+ MIB.addOperand(MOs[i]);
+ addOffset(MIB, PtrOffset);
+ } else {
+ // General Memory Addressing - we need to add any offset to an existing
+ // offset.
+ assert(MOs.size() == 5 && "Unexpected memory operand list length");
+ for (unsigned i = 0; i != NumAddrOps; ++i) {
+ const MachineOperand &MO = MOs[i];
+ if (i == 3 && PtrOffset != 0) {
+ assert((MO.isImm() || MO.isGlobal()) &&
+ "Unexpected memory operand type");
+ if (MO.isImm()) {
+ MIB.addImm(MO.getImm() + PtrOffset);
+ } else {
+ MIB.addGlobalAddress(MO.getGlobal(), MO.getOffset() + PtrOffset,
+ MO.getTargetFlags());
+ }
+ } else {
+ MIB.addOperand(MO);
+ }
+ }
+ }
}
static MachineInstr *FuseTwoAddrInst(MachineFunction &MF, unsigned Opcode,
static MachineInstr *FuseInst(MachineFunction &MF, unsigned Opcode,
unsigned OpNo, ArrayRef<MachineOperand> MOs,
MachineBasicBlock::iterator InsertPt,
- MachineInstr *MI, const TargetInstrInfo &TII) {
+ MachineInstr *MI, const TargetInstrInfo &TII,
+ int PtrOffset = 0) {
// Omit the implicit operands, something BuildMI can't do.
MachineInstr *NewMI = MF.CreateMachineInstr(TII.get(Opcode),
MI->getDebugLoc(), true);
MachineOperand &MO = MI->getOperand(i);
if (i == OpNo) {
assert(MO.isReg() && "Expected to fold into reg operand!");
- addOperands(MIB, MOs);
+ addOperands(MIB, MOs, PtrOffset);
} else {
MIB.addOperand(MO);
}
return MIB.addImm(0);
}
+MachineInstr *X86InstrInfo::foldMemoryOperandCustom(
+ MachineFunction &MF, MachineInstr *MI, unsigned OpNum,
+ ArrayRef<MachineOperand> MOs, MachineBasicBlock::iterator InsertPt,
+ unsigned Size, unsigned Align) const {
+ switch (MI->getOpcode()) {
+ case X86::INSERTPSrr:
+ case X86::VINSERTPSrr:
+ // Attempt to convert the load of inserted vector into a fold load
+ // of a single float.
+ if (OpNum == 2) {
+ unsigned Imm = MI->getOperand(MI->getNumOperands() - 1).getImm();
+ unsigned ZMask = Imm & 15;
+ unsigned DstIdx = (Imm >> 4) & 3;
+ unsigned SrcIdx = (Imm >> 6) & 3;
+
+ unsigned RCSize = getRegClass(MI->getDesc(), OpNum, &RI, MF)->getSize();
+ if (Size <= RCSize && 4 <= Align) {
+ int PtrOffset = SrcIdx * 4;
+ unsigned NewImm = (DstIdx << 4) | ZMask;
+ unsigned NewOpCode =
+ (MI->getOpcode() == X86::VINSERTPSrr ? X86::VINSERTPSrm
+ : X86::INSERTPSrm);
+ MachineInstr *NewMI =
+ FuseInst(MF, NewOpCode, OpNum, MOs, InsertPt, MI, *this, PtrOffset);
+ NewMI->getOperand(NewMI->getNumOperands() - 1).setImm(NewImm);
+ return NewMI;
+ }
+ }
+ break;
+ };
+
+ return nullptr;
+}
+
MachineInstr *X86InstrInfo::foldMemoryOperandImpl(
MachineFunction &MF, MachineInstr *MI, unsigned OpNum,
ArrayRef<MachineOperand> MOs, MachineBasicBlock::iterator InsertPt,
return nullptr;
MachineInstr *NewMI = nullptr;
+
+ // Attempt to fold any custom cases we have.
+ if (MachineInstr *CustomMI =
+ foldMemoryOperandCustom(MF, MI, OpNum, MOs, InsertPt, Size, Align))
+ return CustomMI;
+
// Folding a memory location into the two-address part of a two-address
// instruction is different than folding it other places. It requires
// replacing the *two* registers with the memory location.
// If the instruction and target operand are commutable, commute the
// instruction and try again.
if (AllowCommute) {
- unsigned OriginalOpIdx = OpNum, CommuteOpIdx1, CommuteOpIdx2;
+ unsigned CommuteOpIdx1 = OpNum, CommuteOpIdx2 = CommuteAnyOperandIndex;
if (findCommutedOpIndices(MI, CommuteOpIdx1, CommuteOpIdx2)) {
bool HasDef = MI->getDesc().getNumDefs();
unsigned Reg0 = HasDef ? MI->getOperand(0).getReg() : 0;
unsigned Reg1 = MI->getOperand(CommuteOpIdx1).getReg();
unsigned Reg2 = MI->getOperand(CommuteOpIdx2).getReg();
- bool Tied0 =
- 0 == MI->getDesc().getOperandConstraint(CommuteOpIdx1, MCOI::TIED_TO);
bool Tied1 =
+ 0 == MI->getDesc().getOperandConstraint(CommuteOpIdx1, MCOI::TIED_TO);
+ bool Tied2 =
0 == MI->getDesc().getOperandConstraint(CommuteOpIdx2, MCOI::TIED_TO);
// If either of the commutable operands are tied to the destination
// then we can not commute + fold.
- if ((HasDef && Reg0 == Reg1 && Tied0) ||
- (HasDef && Reg0 == Reg2 && Tied1))
+ if ((HasDef && Reg0 == Reg1 && Tied1) ||
+ (HasDef && Reg0 == Reg2 && Tied2))
return nullptr;
- if ((CommuteOpIdx1 == OriginalOpIdx) ||
- (CommuteOpIdx2 == OriginalOpIdx)) {
- MachineInstr *CommutedMI = commuteInstruction(MI, false);
- if (!CommutedMI) {
- // Unable to commute.
- return nullptr;
- }
- if (CommutedMI != MI) {
- // New instruction. We can't fold from this.
- CommutedMI->eraseFromParent();
- return nullptr;
- }
+ MachineInstr *CommutedMI =
+ commuteInstruction(MI, false, CommuteOpIdx1, CommuteOpIdx2);
+ if (!CommutedMI) {
+ // Unable to commute.
+ return nullptr;
+ }
+ if (CommutedMI != MI) {
+ // New instruction. We can't fold from this.
+ CommutedMI->eraseFromParent();
+ return nullptr;
+ }
- // Attempt to fold with the commuted version of the instruction.
- unsigned CommuteOp =
- (CommuteOpIdx1 == OriginalOpIdx ? CommuteOpIdx2 : CommuteOpIdx1);
- NewMI =
- foldMemoryOperandImpl(MF, MI, CommuteOp, MOs, InsertPt, Size, Align,
- /*AllowCommute=*/false);
- if (NewMI)
- return NewMI;
-
- // Folding failed again - undo the commute before returning.
- MachineInstr *UncommutedMI = commuteInstruction(MI, false);
- if (!UncommutedMI) {
- // Unable to commute.
- return nullptr;
- }
- if (UncommutedMI != MI) {
- // New instruction. It doesn't need to be kept.
- UncommutedMI->eraseFromParent();
- return nullptr;
- }
+ // Attempt to fold with the commuted version of the instruction.
+ NewMI = foldMemoryOperandImpl(MF, MI, CommuteOpIdx2, MOs, InsertPt,
+ Size, Align, /*AllowCommute=*/false);
+ if (NewMI)
+ return NewMI;
- // Return here to prevent duplicate fuse failure report.
+ // Folding failed again - undo the commute before returning.
+ MachineInstr *UncommutedMI =
+ commuteInstruction(MI, false, CommuteOpIdx1, CommuteOpIdx2);
+ if (!UncommutedMI) {
+ // Unable to commute.
+ return nullptr;
+ }
+ if (UncommutedMI != MI) {
+ // New instruction. It doesn't need to be kept.
+ UncommutedMI->eraseFromParent();
return nullptr;
}
+
+ // Return here to prevent duplicate fuse failure report.
+ return nullptr;
}
}
const MCInstrDesc &MCID = get(Opc);
const TargetRegisterClass *RC = getRegClass(MCID, Index, &RI, MF);
+ // TODO: Check if 32-byte or greater accesses are slow too?
if (!MI->hasOneMemOperand() &&
RC == &X86::VR128RegClass &&
- !Subtarget.isUnalignedMemAccessFast())
+ Subtarget.isUnalignedMem16Slow())
// Without memoperands, loadRegFromAddr and storeRegToStackSlot will
// conservatively assume the address is unaligned. That's bad for
// performance.
cast<MachineSDNode>(N)->memoperands_end());
if (!(*MMOs.first) &&
RC == &X86::VR128RegClass &&
- !Subtarget.isUnalignedMemAccessFast())
+ Subtarget.isUnalignedMem16Slow())
// Do not introduce a slow unaligned load.
return false;
+ // FIXME: If a VR128 can have size 32, we should be checking if a 32-byte
+ // memory access is slow above.
unsigned Alignment = RC->getSize() == 32 ? 32 : 16;
bool isAligned = (*MMOs.first) &&
(*MMOs.first)->getAlignment() >= Alignment;
cast<MachineSDNode>(N)->memoperands_end());
if (!(*MMOs.first) &&
RC == &X86::VR128RegClass &&
- !Subtarget.isUnalignedMemAccessFast())
+ Subtarget.isUnalignedMem16Slow())
// Do not introduce a slow unaligned store.
return false;
+ // FIXME: If a VR128 can have size 32, we should be checking if a 32-byte
+ // memory access is slow above.
unsigned Alignment = RC->getSize() == 32 ? 32 : 16;
bool isAligned = (*MMOs.first) &&
(*MMOs.first)->getAlignment() >= Alignment;
return isHighLatencyDef(DefMI->getOpcode());
}
-static bool hasReassociableOperands(const MachineInstr &Inst,
- const MachineBasicBlock *MBB) {
+bool X86InstrInfo::hasReassociableOperands(const MachineInstr &Inst,
+ const MachineBasicBlock *MBB) const {
assert((Inst.getNumOperands() == 3 || Inst.getNumOperands() == 4) &&
"Reassociation needs binary operators");
- const MachineOperand &Op1 = Inst.getOperand(1);
- const MachineOperand &Op2 = Inst.getOperand(2);
- const MachineRegisterInfo &MRI = MBB->getParent()->getRegInfo();
// Integer binary math/logic instructions have a third source operand:
// the EFLAGS register. That operand must be both defined here and never
if (!Inst.getOperand(3).isDead())
return false;
}
-
- // We need virtual register definitions for the operands that we will
- // reassociate.
- MachineInstr *MI1 = nullptr;
- MachineInstr *MI2 = nullptr;
- if (Op1.isReg() && TargetRegisterInfo::isVirtualRegister(Op1.getReg()))
- MI1 = MRI.getUniqueVRegDef(Op1.getReg());
- if (Op2.isReg() && TargetRegisterInfo::isVirtualRegister(Op2.getReg()))
- MI2 = MRI.getUniqueVRegDef(Op2.getReg());
-
- // And they need to be in the trace (otherwise, they won't have a depth).
- if (MI1 && MI2 && MI1->getParent() == MBB && MI2->getParent() == MBB)
- return true;
- return false;
-}
-
-static bool hasReassociableSibling(const MachineInstr &Inst, bool &Commuted) {
- const MachineBasicBlock *MBB = Inst.getParent();
- const MachineRegisterInfo &MRI = MBB->getParent()->getRegInfo();
- MachineInstr *MI1 = MRI.getUniqueVRegDef(Inst.getOperand(1).getReg());
- MachineInstr *MI2 = MRI.getUniqueVRegDef(Inst.getOperand(2).getReg());
- unsigned AssocOpcode = Inst.getOpcode();
-
- // If only one operand has the same opcode and it's the second source operand,
- // the operands must be commuted.
- Commuted = MI1->getOpcode() != AssocOpcode && MI2->getOpcode() == AssocOpcode;
- if (Commuted)
- std::swap(MI1, MI2);
-
- // 1. The previous instruction must be the same type as Inst.
- // 2. The previous instruction must have virtual register definitions for its
- // operands in the same basic block as Inst.
- // 3. The previous instruction's result must only be used by Inst.
- if (MI1->getOpcode() == AssocOpcode &&
- hasReassociableOperands(*MI1, MBB) &&
- MRI.hasOneNonDBGUse(MI1->getOperand(0).getReg()))
- return true;
-
- return false;
+ return TargetInstrInfo::hasReassociableOperands(Inst, MBB);
}
// TODO: There are many more machine instruction opcodes to match:
// 1. Other data types (integer, vectors)
-// 2. Other math / logic operations (and, or)
+// 2. Other math / logic operations (xor, or)
// 3. Other forms of the same operation (intrinsics and other variants)
-static bool isAssociativeAndCommutative(const MachineInstr &Inst) {
+bool X86InstrInfo::isAssociativeAndCommutative(const MachineInstr &Inst) const {
switch (Inst.getOpcode()) {
+ case X86::AND8rr:
+ case X86::AND16rr:
+ case X86::AND32rr:
+ case X86::AND64rr:
+ case X86::OR8rr:
+ case X86::OR16rr:
+ case X86::OR32rr:
+ case X86::OR64rr:
+ case X86::XOR8rr:
+ case X86::XOR16rr:
+ case X86::XOR32rr:
+ case X86::XOR64rr:
case X86::IMUL16rr:
case X86::IMUL32rr:
case X86::IMUL64rr:
+ case X86::PANDrr:
+ case X86::PORrr:
+ case X86::PXORrr:
+ case X86::VPANDrr:
+ case X86::VPANDYrr:
+ case X86::VPORrr:
+ case X86::VPORYrr:
+ case X86::VPXORrr:
+ case X86::VPXORYrr:
// Normal min/max instructions are not commutative because of NaN and signed
// zero semantics, but these are. Thus, there's no need to check for global
// relaxed math; the instructions themselves have the properties we need.
+ case X86::MAXCPDrr:
+ case X86::MAXCPSrr:
+ case X86::MAXCSDrr:
+ case X86::MAXCSSrr:
+ case X86::MINCPDrr:
+ case X86::MINCPSrr:
+ case X86::MINCSDrr:
case X86::MINCSSrr:
+ case X86::VMAXCPDrr:
+ case X86::VMAXCPSrr:
+ case X86::VMAXCPDYrr:
+ case X86::VMAXCPSYrr:
+ case X86::VMAXCSDrr:
+ case X86::VMAXCSSrr:
+ case X86::VMINCPDrr:
+ case X86::VMINCPSrr:
+ case X86::VMINCPDYrr:
+ case X86::VMINCPSYrr:
+ case X86::VMINCSDrr:
case X86::VMINCSSrr:
return true;
case X86::ADDPDrr:
}
}
-/// Return true if the input instruction is part of a chain of dependent ops
-/// that are suitable for reassociation, otherwise return false.
-/// If the instruction's operands must be commuted to have a previous
-/// instruction of the same type define the first source operand, Commuted will
-/// be set to true.
-static bool isReassociationCandidate(const MachineInstr &Inst, bool &Commuted) {
- // 1. The operation must be associative and commutative.
- // 2. The instruction must have virtual register definitions for its
- // operands in the same basic block.
- // 3. The instruction must have a reassociable sibling.
- if (isAssociativeAndCommutative(Inst) &&
- hasReassociableOperands(Inst, Inst.getParent()) &&
- hasReassociableSibling(Inst, Commuted))
- return true;
-
- return false;
-}
-
-// FIXME: This has the potential to be expensive (compile time) while not
-// improving the code at all. Some ways to limit the overhead:
-// 1. Track successful transforms; bail out if hit rate gets too low.
-// 2. Only enable at -O3 or some other non-default optimization level.
-// 3. Pre-screen pattern candidates here: if an operand of the previous
-// instruction is known to not increase the critical path, then don't match
-// that pattern.
-bool X86InstrInfo::getMachineCombinerPatterns(MachineInstr &Root,
- SmallVectorImpl<MachineCombinerPattern::MC_PATTERN> &Patterns) const {
- // TODO: There is nothing x86-specific here except the instruction type.
- // This logic could be hoisted into the machine combiner pass itself.
-
- // Look for this reassociation pattern:
- // B = A op X (Prev)
- // C = B op Y (Root)
-
- bool Commute;
- if (isReassociationCandidate(Root, Commute)) {
- // We found a sequence of instructions that may be suitable for a
- // reassociation of operands to increase ILP. Specify each commutation
- // possibility for the Prev instruction in the sequence and let the
- // machine combiner decide if changing the operands is worthwhile.
- if (Commute) {
- Patterns.push_back(MachineCombinerPattern::MC_REASSOC_AX_YB);
- Patterns.push_back(MachineCombinerPattern::MC_REASSOC_XA_YB);
- } else {
- Patterns.push_back(MachineCombinerPattern::MC_REASSOC_AX_BY);
- Patterns.push_back(MachineCombinerPattern::MC_REASSOC_XA_BY);
- }
- return true;
- }
-
- return false;
-}
-
/// This is an architecture-specific helper function of reassociateOps.
/// Set special operand attributes for new instructions after reassociation.
-static void setSpecialOperandAttr(MachineInstr &OldMI1, MachineInstr &OldMI2,
- MachineInstr &NewMI1, MachineInstr &NewMI2) {
+void X86InstrInfo::setSpecialOperandAttr(MachineInstr &OldMI1,
+ MachineInstr &OldMI2,
+ MachineInstr &NewMI1,
+ MachineInstr &NewMI2) const {
// Integer instructions define an implicit EFLAGS source register operand as
// the third source (fourth total) operand.
if (OldMI1.getNumOperands() != 4 || OldMI2.getNumOperands() != 4)
assert(NewMI1.getNumOperands() == 4 && NewMI2.getNumOperands() == 4 &&
"Unexpected instruction type for reassociation");
-
+
MachineOperand &OldOp1 = OldMI1.getOperand(3);
MachineOperand &OldOp2 = OldMI2.getOperand(3);
MachineOperand &NewOp1 = NewMI1.getOperand(3);
NewOp2.setIsDead();
}
-/// Attempt the following reassociation to reduce critical path length:
-/// B = A op X (Prev)
-/// C = B op Y (Root)
-/// ===>
-/// B = X op Y
-/// C = A op B
-static void reassociateOps(MachineInstr &Root, MachineInstr &Prev,
- MachineCombinerPattern::MC_PATTERN Pattern,
- SmallVectorImpl<MachineInstr *> &InsInstrs,
- SmallVectorImpl<MachineInstr *> &DelInstrs,
- DenseMap<unsigned, unsigned> &InstrIdxForVirtReg) {
- MachineFunction *MF = Root.getParent()->getParent();
- MachineRegisterInfo &MRI = MF->getRegInfo();
- const TargetInstrInfo *TII = MF->getSubtarget().getInstrInfo();
- const TargetRegisterInfo *TRI = MF->getSubtarget().getRegisterInfo();
- const TargetRegisterClass *RC = Root.getRegClassConstraint(0, TII, TRI);
-
- // This array encodes the operand index for each parameter because the
- // operands may be commuted. Each row corresponds to a pattern value,
- // and each column specifies the index of A, B, X, Y.
- unsigned OpIdx[4][4] = {
- { 1, 1, 2, 2 },
- { 1, 2, 2, 1 },
- { 2, 1, 1, 2 },
- { 2, 2, 1, 1 }
- };
-
- MachineOperand &OpA = Prev.getOperand(OpIdx[Pattern][0]);
- MachineOperand &OpB = Root.getOperand(OpIdx[Pattern][1]);
- MachineOperand &OpX = Prev.getOperand(OpIdx[Pattern][2]);
- MachineOperand &OpY = Root.getOperand(OpIdx[Pattern][3]);
- MachineOperand &OpC = Root.getOperand(0);
-
- unsigned RegA = OpA.getReg();
- unsigned RegB = OpB.getReg();
- unsigned RegX = OpX.getReg();
- unsigned RegY = OpY.getReg();
- unsigned RegC = OpC.getReg();
-
- if (TargetRegisterInfo::isVirtualRegister(RegA))
- MRI.constrainRegClass(RegA, RC);
- if (TargetRegisterInfo::isVirtualRegister(RegB))
- MRI.constrainRegClass(RegB, RC);
- if (TargetRegisterInfo::isVirtualRegister(RegX))
- MRI.constrainRegClass(RegX, RC);
- if (TargetRegisterInfo::isVirtualRegister(RegY))
- MRI.constrainRegClass(RegY, RC);
- if (TargetRegisterInfo::isVirtualRegister(RegC))
- MRI.constrainRegClass(RegC, RC);
-
- // Create a new virtual register for the result of (X op Y) instead of
- // recycling RegB because the MachineCombiner's computation of the critical
- // path requires a new register definition rather than an existing one.
- unsigned NewVR = MRI.createVirtualRegister(RC);
- InstrIdxForVirtReg.insert(std::make_pair(NewVR, 0));
-
- unsigned Opcode = Root.getOpcode();
- bool KillA = OpA.isKill();
- bool KillX = OpX.isKill();
- bool KillY = OpY.isKill();
-
- // Create new instructions for insertion.
- MachineInstrBuilder MIB1 =
- BuildMI(*MF, Prev.getDebugLoc(), TII->get(Opcode), NewVR)
- .addReg(RegX, getKillRegState(KillX))
- .addReg(RegY, getKillRegState(KillY));
- MachineInstrBuilder MIB2 =
- BuildMI(*MF, Root.getDebugLoc(), TII->get(Opcode), RegC)
- .addReg(RegA, getKillRegState(KillA))
- .addReg(NewVR, getKillRegState(true));
-
- setSpecialOperandAttr(Root, Prev, *MIB1, *MIB2);
-
- // Record new instructions for insertion and old instructions for deletion.
- InsInstrs.push_back(MIB1);
- InsInstrs.push_back(MIB2);
- DelInstrs.push_back(&Prev);
- DelInstrs.push_back(&Root);
-}
-
-void X86InstrInfo::genAlternativeCodeSequence(
- MachineInstr &Root,
- MachineCombinerPattern::MC_PATTERN Pattern,
- SmallVectorImpl<MachineInstr *> &InsInstrs,
- SmallVectorImpl<MachineInstr *> &DelInstrs,
- DenseMap<unsigned, unsigned> &InstIdxForVirtReg) const {
- MachineRegisterInfo &MRI = Root.getParent()->getParent()->getRegInfo();
-
- // Select the previous instruction in the sequence based on the input pattern.
- MachineInstr *Prev = nullptr;
- switch (Pattern) {
- case MachineCombinerPattern::MC_REASSOC_AX_BY:
- case MachineCombinerPattern::MC_REASSOC_XA_BY:
- Prev = MRI.getUniqueVRegDef(Root.getOperand(1).getReg());
- break;
- case MachineCombinerPattern::MC_REASSOC_AX_YB:
- case MachineCombinerPattern::MC_REASSOC_XA_YB:
- Prev = MRI.getUniqueVRegDef(Root.getOperand(2).getReg());
- }
- assert(Prev && "Unknown pattern for machine combiner");
-
- reassociateOps(Root, *Prev, Pattern, InsInstrs, DelInstrs, InstIdxForVirtReg);
- return;
-}
-
std::pair<unsigned, unsigned>
X86InstrInfo::decomposeMachineOperandsTargetFlags(unsigned TF) const {
return std::make_pair(TF, 0u);
ArrayRef<std::pair<unsigned, const char *>>
X86InstrInfo::getSerializableDirectMachineOperandTargetFlags() const {
using namespace X86II;
- static std::pair<unsigned, const char *> TargetFlags[] = {
+ static const std::pair<unsigned, const char *> TargetFlags[] = {
{MO_GOT_ABSOLUTE_ADDRESS, "x86-got-absolute-address"},
{MO_PIC_BASE_OFFSET, "x86-pic-base-offset"},
{MO_GOT, "x86-got"},