return X86::MOV64toSDrr;
}
+ // SrcReg(FR32) -> DestReg(GR32)
+ // SrcReg(GR32) -> DestReg(FR32)
+
+ if (X86::GR32RegClass.contains(DestReg) && X86::FR32RegClass.contains(SrcReg))
+ // Copy from a FR32 register to a GR32 register.
+ return HasAVX ? X86::VMOVSS2DIrr : X86::MOVSS2DIrr;
+
+ if (X86::FR32RegClass.contains(DestReg) && X86::GR32RegClass.contains(SrcReg))
+ // Copy from a GR32 register to a FR32 register.
+ return HasAVX ? X86::VMOVDI2SSrr : X86::MOVDI2SSrr;
+
return 0;
}
// Copying to or from a physical H register on x86-64 requires a NOREX
// move. Otherwise use a normal move.
if ((isHReg(DestReg) || isHReg(SrcReg)) &&
- TM.getSubtarget<X86Subtarget>().is64Bit())
+ TM.getSubtarget<X86Subtarget>().is64Bit()) {
Opc = X86::MOV8rr_NOREX;
- else
+ // Both operands must be encodable without an REX prefix.
+ assert(X86::GR8_NOREXRegClass.contains(SrcReg, DestReg) &&
+ "8-bit H register can not be copied outside GR8_NOREX");
+ } else
Opc = X86::MOV8rr;
} else if (X86::VR128RegClass.contains(DestReg, SrcReg))
Opc = HasAVX ? X86::VMOVAPSrr : X86::MOVAPSrr;
NewMIs.push_back(MIB);
}
+/// Expand2AddrUndef - Expand a single-def pseudo instruction to a two-addr
+/// instruction with two undef reads of the register being defined. This is
+/// used for mapping:
+/// %xmm4 = V_SET0
+/// to:
+/// %xmm4 = PXORrr %xmm4<undef>, %xmm4<undef>
+///
+static bool Expand2AddrUndef(MachineInstr *MI, const MCInstrDesc &Desc) {
+ assert(Desc.getNumOperands() == 3 && "Expected two-addr instruction.");
+ unsigned Reg = MI->getOperand(0).getReg();
+ MI->setDesc(Desc);
+
+ // MachineInstr::addOperand() will insert explicit operands before any
+ // implicit operands.
+ MachineInstrBuilder(MI).addReg(Reg, RegState::Undef)
+ .addReg(Reg, RegState::Undef);
+ // But we don't trust that.
+ assert(MI->getOperand(1).getReg() == Reg &&
+ MI->getOperand(2).getReg() == Reg && "Misplaced operand");
+ return true;
+}
+
+bool X86InstrInfo::expandPostRAPseudo(MachineBasicBlock::iterator MI) const {
+ bool HasAVX = TM.getSubtarget<X86Subtarget>().hasAVX();
+ switch (MI->getOpcode()) {
+ case X86::V_SET0:
+ return Expand2AddrUndef(MI, get(HasAVX ? X86::VPXORrr : X86::PXORrr));
+ case X86::TEST8ri_NOREX:
+ MI->setDesc(get(X86::TEST8ri));
+ return true;
+ }
+ return false;
+}
+
MachineInstr*
X86InstrInfo::emitFrameIndexDebugValue(MachineFunction &MF,
int FrameIx, uint64_t Offset,
return NULL;
}
+/// hasPartialRegUpdate - Return true for all instructions that only update
+/// the first 32 or 64-bits of the destination register and leave the rest
+/// unmodified. This can be used to avoid folding loads if the instructions
+/// only update part of the destination register, and the non-updated part is
+/// not needed. e.g. cvtss2sd, sqrtss. Unfolding the load from these
+/// instructions breaks the partial register dependency and it can improve
+/// performance. e.g.:
+///
+/// movss (%rdi), %xmm0
+/// cvtss2sd %xmm0, %xmm0
+///
+/// Instead of
+/// cvtss2sd (%rdi), %xmm0
+///
+/// FIXME: This should be turned into a TSFlags.
+///
+static bool hasPartialRegUpdate(unsigned Opcode) {
+ switch (Opcode) {
+ case X86::CVTSD2SSrr:
+ case X86::Int_CVTSD2SSrr:
+ case X86::CVTSS2SDrr:
+ case X86::Int_CVTSS2SDrr:
+ case X86::RCPSSr:
+ case X86::RCPSSr_Int:
+ case X86::ROUNDSDr:
+ case X86::ROUNDSSr:
+ case X86::RSQRTSSr:
+ case X86::RSQRTSSr_Int:
+ case X86::SQRTSSr:
+ case X86::SQRTSSr_Int:
+ // AVX encoded versions
+ case X86::VCVTSD2SSrr:
+ case X86::Int_VCVTSD2SSrr:
+ case X86::VCVTSS2SDrr:
+ case X86::Int_VCVTSS2SDrr:
+ case X86::VRCPSSr:
+ case X86::VROUNDSDr:
+ case X86::VROUNDSSr:
+ case X86::VRSQRTSSr:
+ case X86::VSQRTSSr:
+ return true;
+ }
+
+ return false;
+}
MachineInstr* X86InstrInfo::foldMemoryOperandImpl(MachineFunction &MF,
MachineInstr *MI,
// Check switch flag
if (NoFusing) return NULL;
- if (!MF.getFunction()->hasFnAttr(Attribute::OptimizeForSize))
- switch (MI->getOpcode()) {
- case X86::CVTSD2SSrr:
- case X86::Int_CVTSD2SSrr:
- case X86::CVTSS2SDrr:
- case X86::Int_CVTSS2SDrr:
- case X86::RCPSSr:
- case X86::RCPSSr_Int:
- case X86::ROUNDSDr:
- case X86::ROUNDSSr:
- case X86::RSQRTSSr:
- case X86::RSQRTSSr_Int:
- case X86::SQRTSSr:
- case X86::SQRTSSr_Int:
- return 0;
- }
+ // Unless optimizing for size, don't fold to avoid partial
+ // register update stalls
+ if (!MF.getFunction()->hasFnAttr(Attribute::OptimizeForSize) &&
+ hasPartialRegUpdate(MI->getOpcode()))
+ return 0;
const MachineFrameInfo *MFI = MF.getFrameInfo();
unsigned Size = MFI->getObjectSize(FrameIndex);
// Check switch flag
if (NoFusing) return NULL;
- if (!MF.getFunction()->hasFnAttr(Attribute::OptimizeForSize))
- switch (MI->getOpcode()) {
- case X86::CVTSD2SSrr:
- case X86::Int_CVTSD2SSrr:
- case X86::CVTSS2SDrr:
- case X86::Int_CVTSS2SDrr:
- case X86::RCPSSr:
- case X86::RCPSSr_Int:
- case X86::ROUNDSDr:
- case X86::ROUNDSSr:
- case X86::RSQRTSSr:
- case X86::RSQRTSSr_Int:
- case X86::SQRTSSr:
- case X86::SQRTSSr_Int:
- return 0;
- }
+ // Unless optimizing for size, don't fold to avoid partial
+ // register update stalls
+ if (!MF.getFunction()->hasFnAttr(Attribute::OptimizeForSize) &&
+ hasPartialRegUpdate(MI->getOpcode()))
+ return 0;
// Determine the alignment of the load.
unsigned Alignment = 0;
case X86::AVX_SET0PDY:
Alignment = 32;
break;
- case X86::V_SET0PS:
- case X86::V_SET0PD:
- case X86::V_SET0PI:
+ case X86::V_SET0:
case X86::V_SETALLONES:
- case X86::AVX_SET0PS:
- case X86::AVX_SET0PD:
- case X86::AVX_SET0PI:
case X86::AVX_SETALLONES:
Alignment = 16;
break;
SmallVector<MachineOperand,X86::AddrNumOperands> MOs;
switch (LoadMI->getOpcode()) {
- case X86::V_SET0PS:
- case X86::V_SET0PD:
- case X86::V_SET0PI:
+ case X86::V_SET0:
case X86::V_SETALLONES:
- case X86::AVX_SET0PS:
- case X86::AVX_SET0PD:
- case X86::AVX_SET0PI:
case X86::AVX_SET0PSY:
case X86::AVX_SET0PDY:
case X86::AVX_SETALLONES:
case X86::FsFLD0SS:
case X86::VFsFLD0SD:
case X86::VFsFLD0SS: {
- // Folding a V_SET0P? or V_SETALLONES as a load, to ease register pressure.
+ // Folding a V_SET0 or V_SETALLONES as a load, to ease register pressure.
// Create a constant-pool entry and operands to load from it.
// Medium and large mode can't fold loads this way.
case X86::MMX_MOVQ64rm:
case X86::FsMOVAPSrm:
case X86::FsMOVAPDrm:
- case X86::FsVMOVAPSrm:
- case X86::FsVMOVAPDrm:
case X86::MOVAPSrm:
case X86::MOVUPSrm:
case X86::MOVAPDrm:
case X86::MOVDQArm:
case X86::MOVDQUrm:
+ // AVX load instructions
+ case X86::VMOVSSrm:
+ case X86::VMOVSDrm:
+ case X86::FsVMOVAPSrm:
+ case X86::FsVMOVAPDrm:
case X86::VMOVAPSrm:
case X86::VMOVUPSrm:
case X86::VMOVAPDrm:
case X86::MMX_MOVQ64rm:
case X86::FsMOVAPSrm:
case X86::FsMOVAPDrm:
- case X86::FsVMOVAPSrm:
- case X86::FsVMOVAPDrm:
case X86::MOVAPSrm:
case X86::MOVUPSrm:
case X86::MOVAPDrm:
case X86::MOVDQArm:
case X86::MOVDQUrm:
+ // AVX load instructions
+ case X86::VMOVSSrm:
+ case X86::VMOVSDrm:
+ case X86::FsVMOVAPSrm:
+ case X86::FsVMOVAPDrm:
case X86::VMOVAPSrm:
case X86::VMOVUPSrm:
case X86::VMOVAPDrm:
{ X86::ANDPSrr, X86::ANDPDrr, X86::PANDrr },
{ X86::ORPSrm, X86::ORPDrm, X86::PORrm },
{ X86::ORPSrr, X86::ORPDrr, X86::PORrr },
- { X86::V_SET0PS, X86::V_SET0PD, X86::V_SET0PI },
{ X86::XORPSrm, X86::XORPDrm, X86::PXORrm },
{ X86::XORPSrr, X86::XORPDrr, X86::PXORrr },
// AVX 128-bit support
{ X86::VANDPSrr, X86::VANDPDrr, X86::VPANDrr },
{ X86::VORPSrm, X86::VORPDrm, X86::VPORrm },
{ X86::VORPSrr, X86::VORPDrr, X86::VPORrr },
- { X86::AVX_SET0PS, X86::AVX_SET0PD, X86::AVX_SET0PI },
{ X86::VXORPSrm, X86::VXORPDrm, X86::VPXORrm },
{ X86::VXORPSrr, X86::VXORPDrr, X86::VPXORrr },
// AVX 256-bit support
}
std::pair<uint16_t, uint16_t>
-X86InstrInfo::GetSSEDomain(const MachineInstr *MI) const {
+X86InstrInfo::getExecutionDomain(const MachineInstr *MI) const {
uint16_t domain = (MI->getDesc().TSFlags >> X86II::SSEDomainShift) & 3;
return std::make_pair(domain,
domain && lookup(MI->getOpcode(), domain) ? 0xe : 0);
}
-void X86InstrInfo::SetSSEDomain(MachineInstr *MI, unsigned Domain) const {
+void X86InstrInfo::setExecutionDomain(MachineInstr *MI, unsigned Domain) const {
assert(Domain>0 && Domain<4 && "Invalid execution domain");
uint16_t dom = (MI->getDesc().TSFlags >> X86II::SSEDomainShift) & 3;
assert(dom && "Not an SSE instruction");
case X86::SQRTSSm_Int:
case X86::SQRTSSr:
case X86::SQRTSSr_Int:
+ // AVX instructions with high latency
+ case X86::VDIVSDrm:
+ case X86::VDIVSDrm_Int:
+ case X86::VDIVSDrr:
+ case X86::VDIVSDrr_Int:
+ case X86::VDIVSSrm:
+ case X86::VDIVSSrm_Int:
+ case X86::VDIVSSrr:
+ case X86::VDIVSSrr_Int:
+ case X86::VSQRTPDm:
+ case X86::VSQRTPDm_Int:
+ case X86::VSQRTPDr:
+ case X86::VSQRTPDr_Int:
+ case X86::VSQRTPSm:
+ case X86::VSQRTPSm_Int:
+ case X86::VSQRTPSr:
+ case X86::VSQRTPSr_Int:
+ case X86::VSQRTSDm:
+ case X86::VSQRTSDm_Int:
+ case X86::VSQRTSDr:
+ case X86::VSQRTSSm:
+ case X86::VSQRTSSm_Int:
+ case X86::VSQRTSSr:
return true;
}
}