//===----------------------------------------------------------------------===//
#include "X86.h"
-#include "X86ISelLowering.h"
+#include "X86CallingConv.h"
#include "X86InstrBuilder.h"
+#include "X86MachineFunctionInfo.h"
#include "X86RegisterInfo.h"
#include "X86Subtarget.h"
#include "X86TargetMachine.h"
#include "llvm/CodeGen/MachineConstantPool.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/IR/CallSite.h"
#include "llvm/IR/CallingConv.h"
#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/GetElementPtrTypeIterator.h"
#include "llvm/IR/GlobalAlias.h"
#include "llvm/IR/GlobalVariable.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/IntrinsicInst.h"
#include "llvm/IR/Operator.h"
-#include "llvm/Support/CallSite.h"
#include "llvm/Support/ErrorHandling.h"
-#include "llvm/Support/GetElementPtrTypeIterator.h"
#include "llvm/Target/TargetOptions.h"
using namespace llvm;
namespace {
-class X86FastISel : public FastISel {
+class X86FastISel final : public FastISel {
/// Subtarget - Keep a pointer to the X86Subtarget around so that we can
/// make the right decision when generating code for different targets.
const X86Subtarget *Subtarget;
- /// RegInfo - X86 register info.
- ///
- const X86RegisterInfo *RegInfo;
-
/// X86ScalarSSEf32, X86ScalarSSEf64 - Select between SSE or x87
/// floating point ops.
/// When SSE is available, use it for f32 operations.
Subtarget = &TM.getSubtarget<X86Subtarget>();
X86ScalarSSEf64 = Subtarget->hasSSE2();
X86ScalarSSEf32 = Subtarget->hasSSE1();
- RegInfo = static_cast<const X86RegisterInfo*>(TM.getRegisterInfo());
}
- virtual bool TargetSelectInstruction(const Instruction *I);
+ bool TargetSelectInstruction(const Instruction *I) override;
/// \brief The specified machine instr operand is a vreg, and that
/// vreg is being provided by the specified load instruction. If possible,
/// try to fold the load as an operand to the instruction, returning true if
/// possible.
- virtual bool tryToFoldLoadIntoMI(MachineInstr *MI, unsigned OpNo,
- const LoadInst *LI);
+ bool tryToFoldLoadIntoMI(MachineInstr *MI, unsigned OpNo,
+ const LoadInst *LI) override;
- virtual bool FastLowerArguments();
+ bool FastLowerArguments() override;
#include "X86GenFastISel.inc"
bool X86FastEmitLoad(EVT VT, const X86AddressMode &AM, unsigned &RR);
- bool X86FastEmitStore(EVT VT, const Value *Val, const X86AddressMode &AM);
- bool X86FastEmitStore(EVT VT, unsigned Val, const X86AddressMode &AM);
+ bool X86FastEmitStore(EVT VT, const Value *Val, const X86AddressMode &AM,
+ bool Aligned = false);
+ bool X86FastEmitStore(EVT VT, unsigned ValReg, const X86AddressMode &AM,
+ bool Aligned = false);
bool X86FastEmitExtend(ISD::NodeType Opc, EVT DstVT, unsigned Src, EVT SrcVT,
unsigned &ResultReg);
return static_cast<const X86TargetMachine *>(&TM);
}
- unsigned TargetMaterializeConstant(const Constant *C);
+ bool handleConstantAddresses(const Value *V, X86AddressMode &AM);
+
+ unsigned TargetMaterializeConstant(const Constant *C) override;
- unsigned TargetMaterializeAlloca(const AllocaInst *C);
+ unsigned TargetMaterializeAlloca(const AllocaInst *C) override;
- unsigned TargetMaterializeFloatZero(const ConstantFP *CF);
+ unsigned TargetMaterializeFloatZero(const ConstantFP *CF) override;
/// isScalarFPTypeInSSEReg - Return true if the specified scalar FP type is
/// computed in an SSE register, not on the X87 floating point stack.
unsigned &ResultReg) {
// Get opcode and regclass of the output for the given load instruction.
unsigned Opc = 0;
- const TargetRegisterClass *RC = NULL;
+ const TargetRegisterClass *RC = nullptr;
switch (VT.getSimpleVT().SimpleTy) {
default: return false;
case MVT::i1:
ResultReg = createResultReg(RC);
addFullAddress(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt,
- DL, TII.get(Opc), ResultReg), AM);
+ DbgLoc, TII.get(Opc), ResultReg), AM);
return true;
}
/// and a displacement offset, or a GlobalAddress,
/// i.e. V. Return true if it is possible.
bool
-X86FastISel::X86FastEmitStore(EVT VT, unsigned Val, const X86AddressMode &AM) {
+X86FastISel::X86FastEmitStore(EVT VT, unsigned ValReg,
+ const X86AddressMode &AM, bool Aligned) {
// Get opcode and regclass of the output for the given store instruction.
unsigned Opc = 0;
switch (VT.getSimpleVT().SimpleTy) {
case MVT::i1: {
// Mask out all but lowest bit.
unsigned AndResult = createResultReg(&X86::GR8RegClass);
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
- TII.get(X86::AND8ri), AndResult).addReg(Val).addImm(1);
- Val = AndResult;
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
+ TII.get(X86::AND8ri), AndResult).addReg(ValReg).addImm(1);
+ ValReg = AndResult;
}
// FALLTHROUGH, handling i1 as i8.
case MVT::i8: Opc = X86::MOV8mr; break;
(Subtarget->hasAVX() ? X86::VMOVSDmr : X86::MOVSDmr) : X86::ST_Fp64m;
break;
case MVT::v4f32:
- Opc = X86::MOVAPSmr;
+ if (Aligned)
+ Opc = Subtarget->hasAVX() ? X86::VMOVAPSmr : X86::MOVAPSmr;
+ else
+ Opc = Subtarget->hasAVX() ? X86::VMOVUPSmr : X86::MOVUPSmr;
break;
case MVT::v2f64:
- Opc = X86::MOVAPDmr;
+ if (Aligned)
+ Opc = Subtarget->hasAVX() ? X86::VMOVAPDmr : X86::MOVAPDmr;
+ else
+ Opc = Subtarget->hasAVX() ? X86::VMOVUPDmr : X86::MOVUPDmr;
break;
case MVT::v4i32:
case MVT::v2i64:
case MVT::v8i16:
case MVT::v16i8:
- Opc = X86::MOVDQAmr;
+ if (Aligned)
+ Opc = Subtarget->hasAVX() ? X86::VMOVDQAmr : X86::MOVDQAmr;
+ else
+ Opc = Subtarget->hasAVX() ? X86::VMOVDQUmr : X86::MOVDQUmr;
break;
}
addFullAddress(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt,
- DL, TII.get(Opc)), AM).addReg(Val);
+ DbgLoc, TII.get(Opc)), AM).addReg(ValReg);
return true;
}
bool X86FastISel::X86FastEmitStore(EVT VT, const Value *Val,
- const X86AddressMode &AM) {
+ const X86AddressMode &AM, bool Aligned) {
// Handle 'null' like i32/i64 0.
if (isa<ConstantPointerNull>(Val))
- Val = Constant::getNullValue(TD.getIntPtrType(Val->getContext()));
+ Val = Constant::getNullValue(DL.getIntPtrType(Val->getContext()));
// If this is a store of a simple constant, fold the constant into the store.
if (const ConstantInt *CI = dyn_cast<ConstantInt>(Val)) {
if (Opc) {
addFullAddress(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt,
- DL, TII.get(Opc)), AM)
+ DbgLoc, TII.get(Opc)), AM)
.addImm(Signed ? (uint64_t) CI->getSExtValue() :
CI->getZExtValue());
return true;
if (ValReg == 0)
return false;
- return X86FastEmitStore(VT, ValReg, AM);
+ return X86FastEmitStore(VT, ValReg, AM, Aligned);
}
/// X86FastEmitExtend - Emit a machine instruction to extend a value Src of
return true;
}
+bool X86FastISel::handleConstantAddresses(const Value *V, X86AddressMode &AM) {
+ // Handle constant address.
+ if (const GlobalValue *GV = dyn_cast<GlobalValue>(V)) {
+ // Can't handle alternate code models yet.
+ if (TM.getCodeModel() != CodeModel::Small)
+ return false;
+
+ // Can't handle TLS yet.
+ if (GV->isThreadLocal())
+ return false;
+
+ // RIP-relative addresses can't have additional register operands, so if
+ // we've already folded stuff into the addressing mode, just force the
+ // global value into its own register, which we can use as the basereg.
+ if (!Subtarget->isPICStyleRIPRel() ||
+ (AM.Base.Reg == 0 && AM.IndexReg == 0)) {
+ // Okay, we've committed to selecting this global. Set up the address.
+ AM.GV = GV;
+
+ // Allow the subtarget to classify the global.
+ unsigned char GVFlags = Subtarget->ClassifyGlobalReference(GV, TM);
+
+ // If this reference is relative to the pic base, set it now.
+ if (isGlobalRelativeToPICBase(GVFlags)) {
+ // FIXME: How do we know Base.Reg is free??
+ AM.Base.Reg = getInstrInfo()->getGlobalBaseReg(FuncInfo.MF);
+ }
+
+ // Unless the ABI requires an extra load, return a direct reference to
+ // the global.
+ if (!isGlobalStubReference(GVFlags)) {
+ if (Subtarget->isPICStyleRIPRel()) {
+ // Use rip-relative addressing if we can. Above we verified that the
+ // base and index registers are unused.
+ assert(AM.Base.Reg == 0 && AM.IndexReg == 0);
+ AM.Base.Reg = X86::RIP;
+ }
+ AM.GVOpFlags = GVFlags;
+ return true;
+ }
+
+ // Ok, we need to do a load from a stub. If we've already loaded from
+ // this stub, reuse the loaded pointer, otherwise emit the load now.
+ DenseMap<const Value*, unsigned>::iterator I = LocalValueMap.find(V);
+ unsigned LoadReg;
+ if (I != LocalValueMap.end() && I->second != 0) {
+ LoadReg = I->second;
+ } else {
+ // Issue load from stub.
+ unsigned Opc = 0;
+ const TargetRegisterClass *RC = nullptr;
+ X86AddressMode StubAM;
+ StubAM.Base.Reg = AM.Base.Reg;
+ StubAM.GV = GV;
+ StubAM.GVOpFlags = GVFlags;
+
+ // Prepare for inserting code in the local-value area.
+ SavePoint SaveInsertPt = enterLocalValueArea();
+
+ if (TLI.getPointerTy() == MVT::i64) {
+ Opc = X86::MOV64rm;
+ RC = &X86::GR64RegClass;
+
+ if (Subtarget->isPICStyleRIPRel())
+ StubAM.Base.Reg = X86::RIP;
+ } else {
+ Opc = X86::MOV32rm;
+ RC = &X86::GR32RegClass;
+ }
+
+ LoadReg = createResultReg(RC);
+ MachineInstrBuilder LoadMI =
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc), LoadReg);
+ addFullAddress(LoadMI, StubAM);
+
+ // Ok, back to normal mode.
+ leaveLocalValueArea(SaveInsertPt);
+
+ // Prevent loading GV stub multiple times in same MBB.
+ LocalValueMap[V] = LoadReg;
+ }
+
+ // Now construct the final address. Note that the Disp, Scale,
+ // and Index values may already be set here.
+ AM.Base.Reg = LoadReg;
+ AM.GV = nullptr;
+ return true;
+ }
+ }
+
+ // If all else fails, try to materialize the value in a register.
+ if (!AM.GV || !Subtarget->isPICStyleRIPRel()) {
+ if (AM.Base.Reg == 0) {
+ AM.Base.Reg = getRegForValue(V);
+ return AM.Base.Reg != 0;
+ }
+ if (AM.IndexReg == 0) {
+ assert(AM.Scale == 1 && "Scale with no index!");
+ AM.IndexReg = getRegForValue(V);
+ return AM.IndexReg != 0;
+ }
+ }
+
+ return false;
+}
+
/// X86SelectAddress - Attempt to fill in an address from the given value.
///
bool X86FastISel::X86SelectAddress(const Value *V, X86AddressMode &AM) {
- const User *U = NULL;
+ SmallVector<const Value *, 32> GEPs;
+redo_gep:
+ const User *U = nullptr;
unsigned Opcode = Instruction::UserOp1;
if (const Instruction *I = dyn_cast<Instruction>(V)) {
// Don't walk into other basic blocks; it's possible we haven't
i != e; ++i, ++GTI) {
const Value *Op = *i;
if (StructType *STy = dyn_cast<StructType>(*GTI)) {
- const StructLayout *SL = TD.getStructLayout(STy);
+ const StructLayout *SL = DL.getStructLayout(STy);
Disp += SL->getElementOffset(cast<ConstantInt>(Op)->getZExtValue());
continue;
}
// A array/variable index is always of the form i*S where S is the
// constant scale size. See if we can push the scale into immediates.
- uint64_t S = TD.getTypeAllocSize(GTI.getIndexedType());
+ uint64_t S = DL.getTypeAllocSize(GTI.getIndexedType());
for (;;) {
if (const ConstantInt *CI = dyn_cast<ConstantInt>(Op)) {
// Constant-offset addressing.
Disp += CI->getSExtValue() * S;
break;
}
- if (isa<AddOperator>(Op) &&
- (!isa<Instruction>(Op) ||
- FuncInfo.MBBMap[cast<Instruction>(Op)->getParent()]
- == FuncInfo.MBB) &&
- isa<ConstantInt>(cast<AddOperator>(Op)->getOperand(1))) {
- // An add (in the same block) with a constant operand. Fold the
- // constant.
+ if (canFoldAddIntoGEP(U, Op)) {
+ // A compatible add with a constant operand. Fold the constant.
ConstantInt *CI =
cast<ConstantInt>(cast<AddOperator>(Op)->getOperand(1));
Disp += CI->getSExtValue() * S;
goto unsupported_gep;
}
}
+
// Check for displacement overflow.
if (!isInt<32>(Disp))
break;
- // Ok, the GEP indices were covered by constant-offset and scaled-index
- // addressing. Update the address state and move on to examining the base.
+
AM.IndexReg = IndexReg;
AM.Scale = Scale;
AM.Disp = (uint32_t)Disp;
- if (X86SelectAddress(U->getOperand(0), AM))
+ GEPs.push_back(V);
+
+ if (const GetElementPtrInst *GEP =
+ dyn_cast<GetElementPtrInst>(U->getOperand(0))) {
+ // Ok, the GEP indices were covered by constant-offset and scaled-index
+ // addressing. Update the address state and move on to examining the base.
+ V = GEP;
+ goto redo_gep;
+ } else if (X86SelectAddress(U->getOperand(0), AM)) {
return true;
+ }
// If we couldn't merge the gep value into this addr mode, revert back to
// our address and just match the value instead of completely failing.
AM = SavedAM;
- break;
- unsupported_gep:
- // Ok, the GEP indices weren't all covered.
- break;
- }
- }
- // Handle constant address.
- if (const GlobalValue *GV = dyn_cast<GlobalValue>(V)) {
- // Can't handle alternate code models yet.
- if (TM.getCodeModel() != CodeModel::Small)
- return false;
-
- // Can't handle TLS yet.
- if (const GlobalVariable *GVar = dyn_cast<GlobalVariable>(GV))
- if (GVar->isThreadLocal())
- return false;
-
- // Can't handle TLS yet, part 2 (this is slightly crazy, but this is how
- // it works...).
- if (const GlobalAlias *GA = dyn_cast<GlobalAlias>(GV))
- if (const GlobalVariable *GVar =
- dyn_cast_or_null<GlobalVariable>(GA->resolveAliasedGlobal(false)))
- if (GVar->isThreadLocal())
- return false;
-
- // RIP-relative addresses can't have additional register operands, so if
- // we've already folded stuff into the addressing mode, just force the
- // global value into its own register, which we can use as the basereg.
- if (!Subtarget->isPICStyleRIPRel() ||
- (AM.Base.Reg == 0 && AM.IndexReg == 0)) {
- // Okay, we've committed to selecting this global. Set up the address.
- AM.GV = GV;
-
- // Allow the subtarget to classify the global.
- unsigned char GVFlags = Subtarget->ClassifyGlobalReference(GV, TM);
-
- // If this reference is relative to the pic base, set it now.
- if (isGlobalRelativeToPICBase(GVFlags)) {
- // FIXME: How do we know Base.Reg is free??
- AM.Base.Reg = getInstrInfo()->getGlobalBaseReg(FuncInfo.MF);
- }
-
- // Unless the ABI requires an extra load, return a direct reference to
- // the global.
- if (!isGlobalStubReference(GVFlags)) {
- if (Subtarget->isPICStyleRIPRel()) {
- // Use rip-relative addressing if we can. Above we verified that the
- // base and index registers are unused.
- assert(AM.Base.Reg == 0 && AM.IndexReg == 0);
- AM.Base.Reg = X86::RIP;
- }
- AM.GVOpFlags = GVFlags;
+ for (SmallVectorImpl<const Value *>::reverse_iterator
+ I = GEPs.rbegin(), E = GEPs.rend(); I != E; ++I)
+ if (handleConstantAddresses(*I, AM))
return true;
- }
-
- // Ok, we need to do a load from a stub. If we've already loaded from
- // this stub, reuse the loaded pointer, otherwise emit the load now.
- DenseMap<const Value*, unsigned>::iterator I = LocalValueMap.find(V);
- unsigned LoadReg;
- if (I != LocalValueMap.end() && I->second != 0) {
- LoadReg = I->second;
- } else {
- // Issue load from stub.
- unsigned Opc = 0;
- const TargetRegisterClass *RC = NULL;
- X86AddressMode StubAM;
- StubAM.Base.Reg = AM.Base.Reg;
- StubAM.GV = GV;
- StubAM.GVOpFlags = GVFlags;
-
- // Prepare for inserting code in the local-value area.
- SavePoint SaveInsertPt = enterLocalValueArea();
-
- if (TLI.getPointerTy() == MVT::i64) {
- Opc = X86::MOV64rm;
- RC = &X86::GR64RegClass;
-
- if (Subtarget->isPICStyleRIPRel())
- StubAM.Base.Reg = X86::RIP;
- } else {
- Opc = X86::MOV32rm;
- RC = &X86::GR32RegClass;
- }
-
- LoadReg = createResultReg(RC);
- MachineInstrBuilder LoadMI =
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(Opc), LoadReg);
- addFullAddress(LoadMI, StubAM);
-
- // Ok, back to normal mode.
- leaveLocalValueArea(SaveInsertPt);
-
- // Prevent loading GV stub multiple times in same MBB.
- LocalValueMap[V] = LoadReg;
- }
- // Now construct the final address. Note that the Disp, Scale,
- // and Index values may already be set here.
- AM.Base.Reg = LoadReg;
- AM.GV = 0;
- return true;
- }
+ return false;
+ unsupported_gep:
+ // Ok, the GEP indices weren't all covered.
+ break;
}
-
- // If all else fails, try to materialize the value in a register.
- if (!AM.GV || !Subtarget->isPICStyleRIPRel()) {
- if (AM.Base.Reg == 0) {
- AM.Base.Reg = getRegForValue(V);
- return AM.Base.Reg != 0;
- }
- if (AM.IndexReg == 0) {
- assert(AM.Scale == 1 && "Scale with no index!");
- AM.IndexReg = getRegForValue(V);
- return AM.IndexReg != 0;
- }
}
- return false;
+ return handleConstantAddresses(V, AM);
}
/// X86SelectCallAddress - Attempt to fill in an address from the given value.
///
bool X86FastISel::X86SelectCallAddress(const Value *V, X86AddressMode &AM) {
- const User *U = NULL;
+ const User *U = nullptr;
unsigned Opcode = Instruction::UserOp1;
- if (const Instruction *I = dyn_cast<Instruction>(V)) {
+ const Instruction *I = dyn_cast<Instruction>(V);
+ // Record if the value is defined in the same basic block.
+ //
+ // This information is crucial to know whether or not folding an
+ // operand is valid.
+ // Indeed, FastISel generates or reuses a virtual register for all
+ // operands of all instructions it selects. Obviously, the definition and
+ // its uses must use the same virtual register otherwise the produced
+ // code is incorrect.
+ // Before instruction selection, FunctionLoweringInfo::set sets the virtual
+ // registers for values that are alive across basic blocks. This ensures
+ // that the values are consistently set between across basic block, even
+ // if different instruction selection mechanisms are used (e.g., a mix of
+ // SDISel and FastISel).
+ // For values local to a basic block, the instruction selection process
+ // generates these virtual registers with whatever method is appropriate
+ // for its needs. In particular, FastISel and SDISel do not share the way
+ // local virtual registers are set.
+ // Therefore, this is impossible (or at least unsafe) to share values
+ // between basic blocks unless they use the same instruction selection
+ // method, which is not guarantee for X86.
+ // Moreover, things like hasOneUse could not be used accurately, if we
+ // allow to reference values across basic blocks whereas they are not
+ // alive across basic blocks initially.
+ bool InMBB = true;
+ if (I) {
Opcode = I->getOpcode();
U = I;
+ InMBB = I->getParent() == FuncInfo.MBB->getBasicBlock();
} else if (const ConstantExpr *C = dyn_cast<ConstantExpr>(V)) {
Opcode = C->getOpcode();
U = C;
switch (Opcode) {
default: break;
case Instruction::BitCast:
- // Look past bitcasts.
- return X86SelectCallAddress(U->getOperand(0), AM);
+ // Look past bitcasts if its operand is in the same BB.
+ if (InMBB)
+ return X86SelectCallAddress(U->getOperand(0), AM);
+ break;
case Instruction::IntToPtr:
- // Look past no-op inttoptrs.
- if (TLI.getValueType(U->getOperand(0)->getType()) == TLI.getPointerTy())
+ // Look past no-op inttoptrs if its operand is in the same BB.
+ if (InMBB &&
+ TLI.getValueType(U->getOperand(0)->getType()) == TLI.getPointerTy())
return X86SelectCallAddress(U->getOperand(0), AM);
break;
case Instruction::PtrToInt:
- // Look past no-op ptrtoints.
- if (TLI.getValueType(U->getType()) == TLI.getPointerTy())
+ // Look past no-op ptrtoints if its operand is in the same BB.
+ if (InMBB &&
+ TLI.getValueType(U->getType()) == TLI.getPointerTy())
return X86SelectCallAddress(U->getOperand(0), AM);
break;
}
(AM.Base.Reg != 0 || AM.IndexReg != 0))
return false;
- // Can't handle DLLImport.
- if (GV->hasDLLImportLinkage())
+ // Can't handle DbgLocLImport.
+ if (GV->hasDLLImportStorageClass())
return false;
// Can't handle TLS.
if (S->isAtomic())
return false;
+ unsigned SABIAlignment =
+ DL.getABITypeAlignment(S->getValueOperand()->getType());
+ bool Aligned = S->getAlignment() == 0 || S->getAlignment() >= SABIAlignment;
+
MVT VT;
if (!isTypeLegal(I->getOperand(0)->getType(), VT, /*AllowI1=*/true))
return false;
if (!X86SelectAddress(I->getOperand(1), AM))
return false;
- return X86FastEmitStore(VT, I->getOperand(0), AM);
+ return X86FastEmitStore(VT, I->getOperand(0), AM, Aligned);
}
/// X86SelectRet - Select and emit code to implement ret instructions.
CallingConv::ID CC = F.getCallingConv();
if (CC != CallingConv::C &&
CC != CallingConv::Fast &&
- CC != CallingConv::X86_FastCall)
+ CC != CallingConv::X86_FastCall &&
+ CC != CallingConv::X86_64_SysV)
return false;
- if (Subtarget->isTargetWin64())
+ if (Subtarget->isCallingConvWin64(CC))
return false;
// Don't handle popping bytes on return for now.
// Avoid a cross-class copy. This is very unlikely.
if (!SrcRC->contains(DstReg))
return false;
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(TargetOpcode::COPY),
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(TargetOpcode::COPY),
DstReg).addReg(SrcReg);
// Add register to return instruction.
// a virtual register in the entry block, so now we copy the value out
// and into %rax. We also do the same with %eax for Win32.
if (F.hasStructRetAttr() &&
- (Subtarget->is64Bit() || Subtarget->isTargetWindows())) {
+ (Subtarget->is64Bit() || Subtarget->isTargetKnownWindowsMSVC())) {
unsigned Reg = X86MFInfo->getSRetReturnReg();
assert(Reg &&
"SRetReturnReg should have been set in LowerFormalArguments()!");
unsigned RetReg = Subtarget->is64Bit() ? X86::RAX : X86::EAX;
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(TargetOpcode::COPY),
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(TargetOpcode::COPY),
RetReg).addReg(Reg);
RetRegs.push_back(RetReg);
}
// Now emit the RET.
MachineInstrBuilder MIB =
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(X86::RET));
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Subtarget->is64Bit() ? X86::RETQ : X86::RETL));
for (unsigned i = 0, e = RetRegs.size(); i != e; ++i)
MIB.addReg(RetRegs[i], RegState::Implicit);
return true;
// Handle 'null' like i32/i64 0.
if (isa<ConstantPointerNull>(Op1))
- Op1 = Constant::getNullValue(TD.getIntPtrType(Op0->getContext()));
+ Op1 = Constant::getNullValue(DL.getIntPtrType(Op0->getContext()));
// We have two options: compare with register or immediate. If the RHS of
// the compare is an immediate that we can fold into this compare, use
// CMPri, otherwise use CMPrr.
if (const ConstantInt *Op1C = dyn_cast<ConstantInt>(Op1)) {
if (unsigned CompareImmOpc = X86ChooseCmpImmediateOpcode(VT, Op1C)) {
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(CompareImmOpc))
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(CompareImmOpc))
.addReg(Op0Reg)
.addImm(Op1C->getSExtValue());
return true;
unsigned Op1Reg = getRegForValue(Op1);
if (Op1Reg == 0) return false;
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(CompareOpc))
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(CompareOpc))
.addReg(Op0Reg)
.addReg(Op1Reg);
unsigned EReg = createResultReg(&X86::GR8RegClass);
unsigned NPReg = createResultReg(&X86::GR8RegClass);
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(X86::SETEr), EReg);
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(X86::SETEr), EReg);
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
TII.get(X86::SETNPr), NPReg);
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
TII.get(X86::AND8rr), ResultReg).addReg(NPReg).addReg(EReg);
UpdateValueMap(I, ResultReg);
return true;
unsigned NEReg = createResultReg(&X86::GR8RegClass);
unsigned PReg = createResultReg(&X86::GR8RegClass);
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(X86::SETNEr), NEReg);
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(X86::SETPr), PReg);
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(X86::OR8rr),ResultReg)
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(X86::SETNEr), NEReg);
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(X86::SETPr), PReg);
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(X86::OR8rr),ResultReg)
.addReg(PReg).addReg(NEReg);
UpdateValueMap(I, ResultReg);
return true;
if (!X86FastEmitCompare(Op0, Op1, VT))
return false;
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(SetCCOpc), ResultReg);
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(SetCCOpc), ResultReg);
UpdateValueMap(I, ResultReg);
return true;
}
return false;
// Handle zero-extension from i1 to i8, which is common.
- MVT SrcVT = TLI.getValueType(I->getOperand(0)->getType()).getSimpleVT();
+ MVT SrcVT = TLI.getSimpleValueType(I->getOperand(0)->getType());
if (SrcVT.SimpleTy == MVT::i1) {
// Set the high bits to zero.
ResultReg = FastEmitZExtFromI1(MVT::i8, ResultReg, /*TODO: Kill=*/false);
}
unsigned Result32 = createResultReg(&X86::GR32RegClass);
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(MovInst), Result32)
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(MovInst), Result32)
.addReg(ResultReg);
ResultReg = createResultReg(&X86::GR64RegClass);
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(TargetOpcode::SUBREG_TO_REG),
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(TargetOpcode::SUBREG_TO_REG),
ResultReg)
.addImm(0).addReg(Result32).addImm(X86::sub_32bit);
} else if (DstVT != MVT::i8) {
if (!X86FastEmitCompare(Op0, Op1, VT))
return false;
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(BranchOpc))
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(BranchOpc))
.addMBB(TrueMBB);
if (Predicate == CmpInst::FCMP_UNE) {
// X86 requires a second branch to handle UNE (and OEQ,
// which is mapped to UNE above).
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(X86::JP_4))
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(X86::JP_4))
.addMBB(TrueMBB);
}
- FastEmitBranch(FalseMBB, DL);
+ FastEmitBranch(FalseMBB, DbgLoc);
FuncInfo.MBB->addSuccessor(TrueMBB);
return true;
}
if (TestOpc) {
unsigned OpReg = getRegForValue(TI->getOperand(0));
if (OpReg == 0) return false;
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(TestOpc))
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(TestOpc))
.addReg(OpReg).addImm(1);
unsigned JmpOpc = X86::JNE_4;
JmpOpc = X86::JE_4;
}
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(JmpOpc))
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(JmpOpc))
.addMBB(TrueMBB);
- FastEmitBranch(FalseMBB, DL);
+ FastEmitBranch(FalseMBB, DbgLoc);
FuncInfo.MBB->addSuccessor(TrueMBB);
return true;
}
unsigned OpReg = getRegForValue(BI->getCondition());
if (OpReg == 0) return false;
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(X86::TEST8ri))
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(X86::TEST8ri))
.addReg(OpReg).addImm(1);
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(X86::JNE_4))
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(X86::JNE_4))
.addMBB(TrueMBB);
- FastEmitBranch(FalseMBB, DL);
+ FastEmitBranch(FalseMBB, DbgLoc);
FuncInfo.MBB->addSuccessor(TrueMBB);
return true;
}
bool X86FastISel::X86SelectShift(const Instruction *I) {
unsigned CReg = 0, OpReg = 0;
- const TargetRegisterClass *RC = NULL;
+ const TargetRegisterClass *RC = nullptr;
if (I->getType()->isIntegerTy(8)) {
CReg = X86::CL;
RC = &X86::GR8RegClass;
unsigned Op1Reg = getRegForValue(I->getOperand(1));
if (Op1Reg == 0) return false;
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(TargetOpcode::COPY),
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(TargetOpcode::COPY),
CReg).addReg(Op1Reg);
// The shift instruction uses X86::CL. If we defined a super-register
// of X86::CL, emit a subreg KILL to precisely describe what we're doing here.
if (CReg != X86::CL)
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
TII.get(TargetOpcode::KILL), X86::CL)
.addReg(CReg, RegState::Kill);
unsigned ResultReg = createResultReg(RC);
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(OpReg), ResultReg)
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(OpReg), ResultReg)
.addReg(Op0Reg);
UpdateValueMap(I, ResultReg);
return true;
return false;
// Move op0 into low-order input register.
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
TII.get(OpEntry.OpCopy), TypeEntry.LowInReg).addReg(Op0Reg);
// Zero-extend or sign-extend into high-order input register.
if (OpEntry.OpSignExtend) {
if (OpEntry.IsOpSigned)
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
TII.get(OpEntry.OpSignExtend));
else {
unsigned Zero32 = createResultReg(&X86::GR32RegClass);
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
TII.get(X86::MOV32r0), Zero32);
// Copy the zero into the appropriate sub/super/identical physical
// register. Unfortunately the operations needed are not uniform enough to
// fit neatly into the table above.
if (VT.SimpleTy == MVT::i16) {
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
- TII.get(TargetOpcode::COPY), TypeEntry.HighInReg)
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
+ TII.get(Copy), TypeEntry.HighInReg)
.addReg(Zero32, 0, X86::sub_16bit);
} else if (VT.SimpleTy == MVT::i32) {
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
- TII.get(TargetOpcode::COPY), TypeEntry.HighInReg)
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
+ TII.get(Copy), TypeEntry.HighInReg)
.addReg(Zero32);
} else if (VT.SimpleTy == MVT::i64) {
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
TII.get(TargetOpcode::SUBREG_TO_REG), TypeEntry.HighInReg)
.addImm(0).addReg(Zero32).addImm(X86::sub_32bit);
}
}
}
// Generate the DIV/IDIV instruction.
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
TII.get(OpEntry.OpDivRem)).addReg(Op1Reg);
- // Copy output register into result register.
- unsigned ResultReg = createResultReg(TypeEntry.RC);
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
- TII.get(Copy), ResultReg).addReg(OpEntry.DivRemResultReg);
+ // For i8 remainder, we can't reference AH directly, as we'll end
+ // up with bogus copies like %R9B = COPY %AH. Reference AX
+ // instead to prevent AH references in a REX instruction.
+ //
+ // The current assumption of the fast register allocator is that isel
+ // won't generate explicit references to the GPR8_NOREX registers. If
+ // the allocator and/or the backend get enhanced to be more robust in
+ // that regard, this can be, and should be, removed.
+ unsigned ResultReg = 0;
+ if ((I->getOpcode() == Instruction::SRem ||
+ I->getOpcode() == Instruction::URem) &&
+ OpEntry.DivRemResultReg == X86::AH && Subtarget->is64Bit()) {
+ unsigned SourceSuperReg = createResultReg(&X86::GR16RegClass);
+ unsigned ResultSuperReg = createResultReg(&X86::GR16RegClass);
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
+ TII.get(Copy), SourceSuperReg).addReg(X86::AX);
+
+ // Shift AX right by 8 bits instead of using AH.
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(X86::SHR16ri),
+ ResultSuperReg).addReg(SourceSuperReg).addImm(8);
+
+ // Now reference the 8-bit subreg of the result.
+ ResultReg = FastEmitInst_extractsubreg(MVT::i8, ResultSuperReg,
+ /*Kill=*/true, X86::sub_8bit);
+ }
+ // Copy the result out of the physreg if we haven't already.
+ if (!ResultReg) {
+ ResultReg = createResultReg(TypeEntry.RC);
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Copy), ResultReg)
+ .addReg(OpEntry.DivRemResultReg);
+ }
UpdateValueMap(I, ResultReg);
return true;
if (!Subtarget->hasCMov()) return false;
unsigned Opc = 0;
- const TargetRegisterClass *RC = NULL;
+ const TargetRegisterClass *RC = nullptr;
if (VT == MVT::i16) {
Opc = X86::CMOVE16rr;
RC = &X86::GR16RegClass;
unsigned Op2Reg = getRegForValue(I->getOperand(2));
if (Op2Reg == 0) return false;
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(X86::TEST8rr))
- .addReg(Op0Reg).addReg(Op0Reg);
+ // Selects operate on i1, however, Op0Reg is 8 bits width and may contain
+ // garbage. Indeed, only the less significant bit is supposed to be accurate.
+ // If we read more than the lsb, we may see non-zero values whereas lsb
+ // is zero. Therefore, we have to truncate Op0Reg to i1 for the select.
+ // This is achieved by performing TEST against 1.
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(X86::TEST8ri))
+ .addReg(Op0Reg).addImm(1);
unsigned ResultReg = createResultReg(RC);
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(Opc), ResultReg)
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc), ResultReg)
.addReg(Op1Reg).addReg(Op2Reg);
UpdateValueMap(I, ResultReg);
return true;
unsigned OpReg = getRegForValue(V);
if (OpReg == 0) return false;
unsigned ResultReg = createResultReg(&X86::FR64RegClass);
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
TII.get(X86::CVTSS2SDrr), ResultReg)
.addReg(OpReg);
UpdateValueMap(I, ResultReg);
unsigned OpReg = getRegForValue(V);
if (OpReg == 0) return false;
unsigned ResultReg = createResultReg(&X86::FR32RegClass);
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
TII.get(X86::CVTSD2SSrr), ResultReg)
.addReg(OpReg);
UpdateValueMap(I, ResultReg);
(const TargetRegisterClass*)&X86::GR16_ABCDRegClass :
(const TargetRegisterClass*)&X86::GR32_ABCDRegClass;
unsigned CopyReg = createResultReg(CopyRC);
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(TargetOpcode::COPY),
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(TargetOpcode::COPY),
CopyReg).addReg(InputReg);
InputReg = CopyReg;
}
const Value *Op1 = I.getArgOperand(0); // The guard's value.
const AllocaInst *Slot = cast<AllocaInst>(I.getArgOperand(1));
+ MFI.setStackProtectorIndex(FuncInfo.StaticAllocaMap[Slot]);
+
// Grab the frame index.
X86AddressMode AM;
if (!X86SelectAddress(Slot, AM)) return false;
const MCInstrDesc &II = TII.get(TargetOpcode::DBG_VALUE);
// FIXME may need to add RegState::Debug to any registers produced,
// although ESP/EBP should be the only ones at the moment.
- addFullAddress(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II), AM).
+ addFullAddress(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, II), AM).
addImm(0).addMetadata(DI->getVariable());
return true;
}
case Intrinsic::trap: {
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(X86::TRAP));
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(X86::TRAP));
return true;
}
case Intrinsic::sadd_with_overflow:
// The call to CreateRegs builds two sequential registers, to store the
// both the returned values.
unsigned ResultReg = FuncInfo.CreateRegs(I.getType());
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(OpC), ResultReg)
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(OpC), ResultReg)
.addReg(Reg1).addReg(Reg2);
unsigned Opc = X86::SETBr;
if (I.getIntrinsicID() == Intrinsic::sadd_with_overflow)
Opc = X86::SETOr;
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(Opc), ResultReg+1);
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc),
+ ResultReg + 1);
UpdateValueMap(&I, ResultReg, 2);
return true;
if (!FuncInfo.CanLowerReturn)
return false;
- if (Subtarget->isTargetWin64())
- return false;
-
const Function *F = FuncInfo.Fn;
if (F->isVarArg())
return false;
CallingConv::ID CC = F->getCallingConv();
if (CC != CallingConv::C)
return false;
-
+
+ if (Subtarget->isCallingConvWin64(CC))
+ return false;
+
if (!Subtarget->is64Bit())
return false;
}
}
- static const uint16_t GPR32ArgRegs[] = {
+ static const MCPhysReg GPR32ArgRegs[] = {
X86::EDI, X86::ESI, X86::EDX, X86::ECX, X86::R8D, X86::R9D
};
- static const uint16_t GPR64ArgRegs[] = {
+ static const MCPhysReg GPR64ArgRegs[] = {
X86::RDI, X86::RSI, X86::RDX, X86::RCX, X86::R8 , X86::R9
};
const TargetRegisterClass *RC64 = TLI.getRegClassFor(MVT::i64);
for (Function::const_arg_iterator I = F->arg_begin(), E = F->arg_end();
I != E; ++I, ++Idx) {
- if (I->use_empty())
- continue;
bool is32Bit = TLI.getValueType(I->getType()) == MVT::i32;
const TargetRegisterClass *RC = is32Bit ? RC32 : RC64;
unsigned SrcReg = is32Bit ? GPR32ArgRegs[Idx] : GPR64ArgRegs[Idx];
// Without this, EmitLiveInCopies may eliminate the livein if its only
// use is a bitcast (which isn't turned into an instruction).
unsigned ResultReg = createResultReg(RC);
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(TargetOpcode::COPY),
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
+ TII.get(TargetOpcode::COPY),
ResultReg).addReg(DstReg, getKillRegState(true));
UpdateValueMap(I, ResultReg);
}
if (cast<CallInst>(I)->isTailCall())
return false;
- return DoSelectCall(I, 0);
+ return DoSelectCall(I, nullptr);
}
static unsigned computeBytesPoppedByCallee(const X86Subtarget &Subtarget,
const ImmutableCallSite &CS) {
if (Subtarget.is64Bit())
return 0;
- if (Subtarget.isTargetWindows())
+ if (Subtarget.getTargetTriple().isOSMSVCRT())
return 0;
CallingConv::ID CC = CS.getCallingConv();
if (CC == CallingConv::Fast || CC == CallingConv::GHC)
// Handle only C and fastcc calling conventions for now.
ImmutableCallSite CS(CI);
CallingConv::ID CC = CS.getCallingConv();
+ bool isWin64 = Subtarget->isCallingConvWin64(CC);
if (CC != CallingConv::C && CC != CallingConv::Fast &&
- CC != CallingConv::X86_FastCall)
+ CC != CallingConv::X86_FastCall && CC != CallingConv::X86_64_Win64 &&
+ CC != CallingConv::X86_64_SysV)
return false;
// fastcc with -tailcallopt is intended to provide a guaranteed
// Don't know how to handle Win64 varargs yet. Nothing special needed for
// x86-32. Special handling for x86-64 is implemented.
- if (isVarArg && Subtarget->isTargetWin64())
+ if (isVarArg && isWin64)
+ return false;
+
+ // Don't know about inalloca yet.
+ if (CS.hasInAllocaArgument())
return false;
// Fast-isel doesn't know about callee-pop yet.
if (!X86SelectCallAddress(Callee, CalleeAM))
return false;
unsigned CalleeOp = 0;
- const GlobalValue *GV = 0;
- if (CalleeAM.GV != 0) {
+ const GlobalValue *GV = nullptr;
+ if (CalleeAM.GV != nullptr) {
GV = CalleeAM.GV;
} else if (CalleeAM.Base.Reg != 0) {
CalleeOp = CalleeAM.Base.Reg;
if (CS.paramHasAttr(AttrInd, Attribute::ByVal)) {
PointerType *Ty = cast<PointerType>(ArgVal->getType());
Type *ElementTy = Ty->getElementType();
- unsigned FrameSize = TD.getTypeAllocSize(ElementTy);
+ unsigned FrameSize = DL.getTypeAllocSize(ElementTy);
unsigned FrameAlign = CS.getParamAlignment(AttrInd);
if (!FrameAlign)
FrameAlign = TLI.getByValTypeAlignment(ElementTy);
return false;
if (ArgVT == MVT::x86mmx)
return false;
- unsigned OriginalAlignment = TD.getABITypeAlignment(ArgTy);
+ unsigned OriginalAlignment = DL.getABITypeAlignment(ArgTy);
Flags.setOrigAlign(OriginalAlignment);
Args.push_back(ArgReg);
I->getParent()->getContext());
// Allocate shadow area for Win64
- if (Subtarget->isTargetWin64())
+ if (isWin64)
CCInfo.AllocateStack(32, 8);
CCInfo.AnalyzeCallOperands(ArgVTs, ArgFlags, CC_X86);
// Issue CALLSEQ_START
unsigned AdjStackDown = TII.getCallFrameSetupOpcode();
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(AdjStackDown))
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(AdjStackDown))
.addImm(NumBytes);
// Process argument: walk the register/memloc assignments, inserting
// FIXME: Indirect doesn't need extending, but fast-isel doesn't fully
// support this.
return false;
+ case CCValAssign::FPExt:
+ llvm_unreachable("Unexpected loc info!");
}
if (VA.isRegLoc()) {
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(TargetOpcode::COPY),
- VA.getLocReg()).addReg(Arg);
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
+ TII.get(TargetOpcode::COPY), VA.getLocReg()).addReg(Arg);
RegArgs.push_back(VA.getLocReg());
} else {
unsigned LocMemOffset = VA.getLocMemOffset();
X86AddressMode AM;
+ const X86RegisterInfo *RegInfo = static_cast<const X86RegisterInfo*>(
+ getTargetMachine()->getRegisterInfo());
AM.Base.Reg = RegInfo->getStackRegister();
AM.Disp = LocMemOffset;
const Value *ArgVal = ArgVals[VA.getValNo()];
// GOT pointer.
if (Subtarget->isPICStyleGOT()) {
unsigned Base = getInstrInfo()->getGlobalBaseReg(FuncInfo.MF);
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(TargetOpcode::COPY),
- X86::EBX).addReg(Base);
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
+ TII.get(TargetOpcode::COPY), X86::EBX).addReg(Base);
}
- if (Subtarget->is64Bit() && isVarArg && !Subtarget->isTargetWin64()) {
+ if (Subtarget->is64Bit() && isVarArg && !isWin64) {
// Count the number of XMM registers allocated.
- static const uint16_t XMMArgRegs[] = {
+ static const MCPhysReg XMMArgRegs[] = {
X86::XMM0, X86::XMM1, X86::XMM2, X86::XMM3,
X86::XMM4, X86::XMM5, X86::XMM6, X86::XMM7
};
unsigned NumXMMRegs = CCInfo.getFirstUnallocated(XMMArgRegs, 8);
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(X86::MOV8ri),
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(X86::MOV8ri),
X86::AL).addImm(NumXMMRegs);
}
CallOpc = X86::CALL64r;
else
CallOpc = X86::CALL32r;
- MIB = BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(CallOpc))
+ MIB = BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(CallOpc))
.addReg(CalleeOp);
} else {
}
- MIB = BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(CallOpc));
+ MIB = BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(CallOpc));
if (MemIntName)
MIB.addExternalSymbol(MemIntName, OpFlags);
else
if (Subtarget->isPICStyleGOT())
MIB.addReg(X86::EBX, RegState::Implicit);
- if (Subtarget->is64Bit() && isVarArg && !Subtarget->isTargetWin64())
+ if (Subtarget->is64Bit() && isVarArg && !isWin64)
MIB.addReg(X86::AL, RegState::Implicit);
// Add implicit physical register uses to the call.
// Issue CALLSEQ_END
unsigned AdjStackUp = TII.getCallFrameDestroyOpcode();
const unsigned NumBytesCallee = computeBytesPoppedByCallee(*Subtarget, CS);
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(AdjStackUp))
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(AdjStackUp))
.addImm(NumBytes).addImm(NumBytesCallee);
// Build info for return calling conv lowering code.
CopyVT = MVT::f80;
CopyReg = createResultReg(&X86::RFP80RegClass);
}
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(X86::FpPOP_RETVAL),
- CopyReg);
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
+ TII.get(X86::FpPOP_RETVAL), CopyReg);
} else {
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(TargetOpcode::COPY),
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
+ TII.get(TargetOpcode::COPY),
CopyReg).addReg(RVLocs[i].getLocReg());
UsedRegs.push_back(RVLocs[i].getLocReg());
}
unsigned Opc = ResVT == MVT::f32 ? X86::ST_Fp80m32 : X86::ST_Fp80m64;
unsigned MemSize = ResVT.getSizeInBits()/8;
int FI = MFI.CreateStackObject(MemSize, MemSize, false);
- addFrameReference(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
+ addFrameReference(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
TII.get(Opc)), FI)
.addReg(CopyReg);
Opc = ResVT == MVT::f32 ? X86::MOVSSrm : X86::MOVSDrm;
- addFrameReference(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
+ addFrameReference(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
TII.get(Opc), ResultReg + i), FI);
}
}
// Get opcode and regclass of the output for the given load instruction.
unsigned Opc = 0;
- const TargetRegisterClass *RC = NULL;
+ const TargetRegisterClass *RC = nullptr;
switch (VT.SimpleTy) {
default: return 0;
case MVT::i8:
// If the expression is just a basereg, then we're done, otherwise we need
// to emit an LEA.
if (AM.BaseType == X86AddressMode::RegBase &&
- AM.IndexReg == 0 && AM.Disp == 0 && AM.GV == 0)
+ AM.IndexReg == 0 && AM.Disp == 0 && AM.GV == nullptr)
return AM.Base.Reg;
Opc = TLI.getPointerTy() == MVT::i32 ? X86::LEA32r : X86::LEA64r;
unsigned ResultReg = createResultReg(RC);
- addFullAddress(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
+ addFullAddress(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
TII.get(Opc), ResultReg), AM);
return ResultReg;
}
}
// MachineConstantPool wants an explicit alignment.
- unsigned Align = TD.getPrefTypeAlignment(C->getType());
+ unsigned Align = DL.getPrefTypeAlignment(C->getType());
if (Align == 0) {
// Alignment of vector types. FIXME!
- Align = TD.getTypeAllocSize(C->getType());
+ Align = DL.getTypeAllocSize(C->getType());
}
// x86-32 PIC requires a PIC base register for constant pools.
// Create the load from the constant pool.
unsigned MCPOffset = MCP.getConstantPoolIndex(C, Align);
unsigned ResultReg = createResultReg(RC);
- addConstantPoolReference(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
+ addConstantPoolReference(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
TII.get(Opc), ResultReg),
MCPOffset, PICBase, OpFlag);
// X86SelectAddrss, and TargetMaterializeAlloca.
if (!FuncInfo.StaticAllocaMap.count(C))
return 0;
+ assert(C->isStaticAlloca() && "dynamic alloca in the static alloca map?");
X86AddressMode AM;
if (!X86SelectAddress(C, AM))
unsigned Opc = Subtarget->is64Bit() ? X86::LEA64r : X86::LEA32r;
const TargetRegisterClass* RC = TLI.getRegClassFor(TLI.getPointerTy());
unsigned ResultReg = createResultReg(RC);
- addFullAddress(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
+ addFullAddress(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
TII.get(Opc), ResultReg), AM);
return ResultReg;
}
// Get opcode and regclass for the given zero.
unsigned Opc = 0;
- const TargetRegisterClass *RC = NULL;
+ const TargetRegisterClass *RC = nullptr;
switch (VT.SimpleTy) {
default: return 0;
case MVT::f32:
}
unsigned ResultReg = createResultReg(RC);
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(Opc), ResultReg);
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc), ResultReg);
return ResultReg;
}
const X86InstrInfo &XII = (const X86InstrInfo&)TII;
- unsigned Size = TD.getTypeAllocSize(LI->getType());
+ unsigned Size = DL.getTypeAllocSize(LI->getType());
unsigned Alignment = LI->getAlignment();
SmallVector<MachineOperand, 8> AddrOps;
MachineInstr *Result =
XII.foldMemoryOperandImpl(*FuncInfo.MF, MI, OpNo, AddrOps, Size, Alignment);
- if (Result == 0) return false;
+ if (!Result) return false;
FuncInfo.MBB->insert(FuncInfo.InsertPt, Result);
MI->eraseFromParent();