mirror of
https://github.com/RPCSX/llvm.git
synced 2024-12-27 14:45:24 +00:00
[AVX-512] Teach X86InstrInfo::copyPhysReg to use a 512-bit move if XMM16-XMM31 or YMM16-YMM31 are the source or dest of the copy and VLX is not supported.
This can happen with SUBREG_TO_REG of ZMM16-ZMM31. Fixes PR30430. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@281959 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
parent
fd8c733710
commit
0acae1f681
@ -4704,11 +4704,31 @@ void X86InstrInfo::copyPhysReg(MachineBasicBlock &MBB,
|
||||
}
|
||||
else if (X86::VR64RegClass.contains(DestReg, SrcReg))
|
||||
Opc = X86::MMX_MOVQ64rr;
|
||||
else if (X86::VR128XRegClass.contains(DestReg, SrcReg))
|
||||
Opc = HasVLX ? X86::VMOVAPSZ128rr : HasAVX ? X86::VMOVAPSrr : X86::MOVAPSrr;
|
||||
else if (X86::VR256XRegClass.contains(DestReg, SrcReg))
|
||||
Opc = HasVLX ? X86::VMOVAPSZ256rr : X86::VMOVAPSYrr;
|
||||
else if (X86::VR512RegClass.contains(DestReg, SrcReg))
|
||||
else if (X86::VR128XRegClass.contains(DestReg, SrcReg)) {
|
||||
if (HasVLX)
|
||||
Opc = X86::VMOVAPSZ128rr;
|
||||
else if (X86::VR128RegClass.contains(DestReg, SrcReg))
|
||||
Opc = HasAVX ? X86::VMOVAPSrr : X86::MOVAPSrr;
|
||||
else {
|
||||
// If this an extended register and we don't have VLX we need to use a
|
||||
// 512-bit move.
|
||||
Opc = X86::VMOVAPSZrr;
|
||||
DestReg = get512BitSuperRegister(DestReg);
|
||||
SrcReg = get512BitSuperRegister(SrcReg);
|
||||
}
|
||||
} else if (X86::VR256XRegClass.contains(DestReg, SrcReg)) {
|
||||
if (HasVLX)
|
||||
Opc = X86::VMOVAPSZ256rr;
|
||||
else if (X86::VR256RegClass.contains(DestReg, SrcReg))
|
||||
Opc = X86::VMOVAPSYrr;
|
||||
else {
|
||||
// If this an extended register and we don't have VLX we need to use a
|
||||
// 512-bit move.
|
||||
Opc = X86::VMOVAPSZrr;
|
||||
DestReg = get512BitSuperRegister(DestReg);
|
||||
SrcReg = get512BitSuperRegister(SrcReg);
|
||||
}
|
||||
} else if (X86::VR512RegClass.contains(DestReg, SrcReg))
|
||||
Opc = X86::VMOVAPSZrr;
|
||||
// All KMASK RegClasses hold the same k registers, can be tested against anyone.
|
||||
else if (X86::VK16RegClass.contains(DestReg, SrcReg))
|
||||
|
@ -691,3 +691,13 @@ X86RegisterInfo::getPtrSizedFrameRegister(const MachineFunction &MF) const {
|
||||
FrameReg = getX86SubSuperRegister(FrameReg, 32);
|
||||
return FrameReg;
|
||||
}
|
||||
|
||||
unsigned llvm::get512BitSuperRegister(unsigned Reg) {
|
||||
if (Reg >= X86::XMM0 && Reg <= X86::XMM31)
|
||||
return X86::ZMM0 + (Reg - X86::XMM0);
|
||||
if (Reg >= X86::YMM0 && Reg <= X86::YMM31)
|
||||
return X86::ZMM0 + (Reg - X86::YMM0);
|
||||
if (Reg >= X86::ZMM0 && Reg <= X86::ZMM31)
|
||||
return Reg;
|
||||
llvm_unreachable("Unexpected SIMD register");
|
||||
}
|
||||
|
@ -137,6 +137,9 @@ public:
|
||||
unsigned getSlotSize() const { return SlotSize; }
|
||||
};
|
||||
|
||||
//get512BitRegister - X86 utility - returns 512-bit super register
|
||||
unsigned get512BitSuperRegister(unsigned Reg);
|
||||
|
||||
} // End llvm namespace
|
||||
|
||||
#endif
|
||||
|
Loading…
Reference in New Issue
Block a user