maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

Use movaps instead of movups to spill 16-byte vector values when default alignment is >= 16. This fixes some massive performance regressions. 

llvm-svn: 53844
This commit is contained in:
Evan Cheng 2008-07-21 06:34:17 +00:00
parent f1c149db9f
commit e001643358
1 changed files with 19 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

28
llvm/lib/Target/X86/X86InstrInfo.cpp
View File

@ -1739,7 +1739,9 @@ void X86InstrInfo::storeRegToStackSlot(MachineBasicBlock &MBB,
unsigned SrcReg, bool isKill, int FrameIdx, unsigned SrcReg, bool isKill, int FrameIdx,
const TargetRegisterClass *RC) const { const TargetRegisterClass *RC) const {
const MachineFunction &MF = *MBB.getParent(); const MachineFunction &MF = *MBB.getParent();
unsigned Opc = getStoreRegOpcode(RC, RI.needsStackRealignment(MF)); bool isAligned = (RI.getStackAlignment() >= 16) ||
RI.needsStackRealignment(MF);
unsigned Opc = getStoreRegOpcode(RC, isAligned);
addFrameReference(BuildMI(MBB, MI, get(Opc)), FrameIdx) addFrameReference(BuildMI(MBB, MI, get(Opc)), FrameIdx)
.addReg(SrcReg, false, false, isKill); .addReg(SrcReg, false, false, isKill);
} }
@ -1749,7 +1751,9 @@ void X86InstrInfo::storeRegToAddr(MachineFunction &MF, unsigned SrcReg,
SmallVectorImpl<MachineOperand> &Addr, SmallVectorImpl<MachineOperand> &Addr,
const TargetRegisterClass *RC, const TargetRegisterClass *RC,
SmallVectorImpl<MachineInstr*> &NewMIs) const { SmallVectorImpl<MachineInstr*> &NewMIs) const {
unsigned Opc = getStoreRegOpcode(RC, RI.needsStackRealignment(MF)); bool isAligned = (RI.getStackAlignment() >= 16) ||
RI.needsStackRealignment(MF);
unsigned Opc = getStoreRegOpcode(RC, isAligned);
MachineInstrBuilder MIB = BuildMI(MF, get(Opc)); MachineInstrBuilder MIB = BuildMI(MF, get(Opc));
for (unsigned i = 0, e = Addr.size(); i != e; ++i) for (unsigned i = 0, e = Addr.size(); i != e; ++i)
MIB = X86InstrAddOperand(MIB, Addr[i]); MIB = X86InstrAddOperand(MIB, Addr[i]);
@ -1800,7 +1804,9 @@ void X86InstrInfo::loadRegFromStackSlot(MachineBasicBlock &MBB,
unsigned DestReg, int FrameIdx, unsigned DestReg, int FrameIdx,
const TargetRegisterClass *RC) const{ const TargetRegisterClass *RC) const{
const MachineFunction &MF = *MBB.getParent(); const MachineFunction &MF = *MBB.getParent();
unsigned Opc = getLoadRegOpcode(RC, RI.needsStackRealignment(MF)); bool isAligned = (RI.getStackAlignment() >= 16) ||
RI.needsStackRealignment(MF);
unsigned Opc = getLoadRegOpcode(RC, isAligned);
addFrameReference(BuildMI(MBB, MI, get(Opc), DestReg), FrameIdx); addFrameReference(BuildMI(MBB, MI, get(Opc), DestReg), FrameIdx);
} }
@ -1808,7 +1814,9 @@ void X86InstrInfo::loadRegFromAddr(MachineFunction &MF, unsigned DestReg,
SmallVectorImpl<MachineOperand> &Addr, SmallVectorImpl<MachineOperand> &Addr,
const TargetRegisterClass *RC, const TargetRegisterClass *RC,
SmallVectorImpl<MachineInstr*> &NewMIs) const { SmallVectorImpl<MachineInstr*> &NewMIs) const {
unsigned Opc = getLoadRegOpcode(RC, RI.needsStackRealignment(MF)); bool isAligned = (RI.getStackAlignment() >= 16) ||
RI.needsStackRealignment(MF);
unsigned Opc = getLoadRegOpcode(RC, isAligned);
MachineInstrBuilder MIB = BuildMI(MF, get(Opc), DestReg); MachineInstrBuilder MIB = BuildMI(MF, get(Opc), DestReg);
for (unsigned i = 0, e = Addr.size(); i != e; ++i) for (unsigned i = 0, e = Addr.size(); i != e; ++i)
MIB = X86InstrAddOperand(MIB, Addr[i]); MIB = X86InstrAddOperand(MIB, Addr[i]);
@ -2275,7 +2283,9 @@ X86InstrInfo::unfoldMemoryOperand(SelectionDAG &DAG, SDNode *N,
const MachineFunction &MF = DAG.getMachineFunction(); const MachineFunction &MF = DAG.getMachineFunction();
if (FoldedLoad) { if (FoldedLoad) {
MVT VT = *RC->vt_begin(); MVT VT = *RC->vt_begin();
Load = DAG.getTargetNode(getLoadRegOpcode(RC, RI.needsStackRealignment(MF)), bool isAligned = (RI.getStackAlignment() >= 16) ||
RI.needsStackRealignment(MF);
Load = DAG.getTargetNode(getLoadRegOpcode(RC, isAligned),
VT, MVT::Other, VT, MVT::Other,
&AddrOps[0], AddrOps.size()); &AddrOps[0], AddrOps.size());
NewNodes.push_back(Load); NewNodes.push_back(Load);
@ -2306,10 +2316,10 @@ X86InstrInfo::unfoldMemoryOperand(SelectionDAG &DAG, SDNode *N,
AddrOps.pop_back(); AddrOps.pop_back();
AddrOps.push_back(SDOperand(NewNode, 0)); AddrOps.push_back(SDOperand(NewNode, 0));
AddrOps.push_back(Chain); AddrOps.push_back(Chain);
SDNode *Store = bool isAligned = (RI.getStackAlignment() >= 16) ||
DAG.getTargetNode(getStoreRegOpcode(DstRC, RI.needsStackRealignment(MF);
RI.needsStackRealignment(MF)), SDNode *Store = DAG.getTargetNode(getStoreRegOpcode(DstRC, isAligned),
MVT::Other, &AddrOps[0], AddrOps.size()); MVT::Other, &AddrOps[0], AddrOps.size());
NewNodes.push_back(Store); NewNodes.push_back(Store);
} }