#define DEBUG_TYPE "isel"
#include "SDNodeDbgValue.h"
#include "SelectionDAGBuilder.h"
-#include "FunctionLoweringInfo.h"
#include "llvm/ADT/BitVector.h"
+#include "llvm/ADT/PostOrderIterator.h"
#include "llvm/ADT/SmallSet.h"
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/Analysis/ConstantFolding.h"
#include "llvm/IntrinsicInst.h"
#include "llvm/LLVMContext.h"
#include "llvm/Module.h"
+#include "llvm/CodeGen/Analysis.h"
#include "llvm/CodeGen/FastISel.h"
+#include "llvm/CodeGen/FunctionLoweringInfo.h"
#include "llvm/CodeGen/GCStrategy.h"
#include "llvm/CodeGen/GCMetadata.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/PseudoSourceValue.h"
#include "llvm/CodeGen/SelectionDAG.h"
#include "llvm/Analysis/DebugInfo.h"
-#include "llvm/Target/TargetRegisterInfo.h"
#include "llvm/Target/TargetData.h"
-#include "llvm/Target/TargetFrameInfo.h"
+#include "llvm/Target/TargetFrameLowering.h"
#include "llvm/Target/TargetInstrInfo.h"
#include "llvm/Target/TargetIntrinsicInfo.h"
#include "llvm/Target/TargetLowering.h"
cl::location(LimitFloatPrecision),
cl::init(0));
-namespace {
- /// RegsForValue - This struct represents the registers (physical or virtual)
- /// that a particular set of values is assigned, and the type information
- /// about the value. The most common situation is to represent one value at a
- /// time, but struct or array values are handled element-wise as multiple
- /// values. The splitting of aggregates is performed recursively, so that we
- /// never have aggregate-typed registers. The values at this point do not
- /// necessarily have legal types, so each value may require one or more
- /// registers of some legal type.
- ///
- struct RegsForValue {
- /// TLI - The TargetLowering object.
- ///
- const TargetLowering *TLI;
-
- /// ValueVTs - The value types of the values, which may not be legal, and
- /// may need be promoted or synthesized from one or more registers.
- ///
- SmallVector<EVT, 4> ValueVTs;
-
- /// RegVTs - The value types of the registers. This is the same size as
- /// ValueVTs and it records, for each value, what the type of the assigned
- /// register or registers are. (Individual values are never synthesized
- /// from more than one type of register.)
- ///
- /// With virtual registers, the contents of RegVTs is redundant with TLI's
- /// getRegisterType member function, however when with physical registers
- /// it is necessary to have a separate record of the types.
- ///
- SmallVector<EVT, 4> RegVTs;
-
- /// Regs - This list holds the registers assigned to the values.
- /// Each legal or promoted value requires one register, and each
- /// expanded value requires multiple registers.
- ///
- SmallVector<unsigned, 4> Regs;
-
- RegsForValue() : TLI(0) {}
-
- RegsForValue(const TargetLowering &tli,
- const SmallVector<unsigned, 4> ®s,
- EVT regvt, EVT valuevt)
- : TLI(&tli), ValueVTs(1, valuevt), RegVTs(1, regvt), Regs(regs) {}
- RegsForValue(const TargetLowering &tli,
- const SmallVector<unsigned, 4> ®s,
- const SmallVector<EVT, 4> ®vts,
- const SmallVector<EVT, 4> &valuevts)
- : TLI(&tli), ValueVTs(valuevts), RegVTs(regvts), Regs(regs) {}
- RegsForValue(LLVMContext &Context, const TargetLowering &tli,
- unsigned Reg, const Type *Ty) : TLI(&tli) {
- ComputeValueVTs(tli, Ty, ValueVTs);
-
- for (unsigned Value = 0, e = ValueVTs.size(); Value != e; ++Value) {
- EVT ValueVT = ValueVTs[Value];
- unsigned NumRegs = TLI->getNumRegisters(Context, ValueVT);
- EVT RegisterVT = TLI->getRegisterType(Context, ValueVT);
- for (unsigned i = 0; i != NumRegs; ++i)
- Regs.push_back(Reg + i);
- RegVTs.push_back(RegisterVT);
- Reg += NumRegs;
- }
- }
-
- /// areValueTypesLegal - Return true if types of all the values are legal.
- bool areValueTypesLegal() {
- for (unsigned Value = 0, e = ValueVTs.size(); Value != e; ++Value) {
- EVT RegisterVT = RegVTs[Value];
- if (!TLI->isTypeLegal(RegisterVT))
- return false;
- }
- return true;
- }
-
-
- /// append - Add the specified values to this one.
- void append(const RegsForValue &RHS) {
- TLI = RHS.TLI;
- ValueVTs.append(RHS.ValueVTs.begin(), RHS.ValueVTs.end());
- RegVTs.append(RHS.RegVTs.begin(), RHS.RegVTs.end());
- Regs.append(RHS.Regs.begin(), RHS.Regs.end());
- }
-
-
- /// getCopyFromRegs - Emit a series of CopyFromReg nodes that copies from
- /// this value and returns the result as a ValueVTs value. This uses
- /// Chain/Flag as the input and updates them for the output Chain/Flag.
- /// If the Flag pointer is NULL, no flag is used.
- SDValue getCopyFromRegs(SelectionDAG &DAG, DebugLoc dl,
- SDValue &Chain, SDValue *Flag) const;
-
- /// getCopyToRegs - Emit a series of CopyToReg nodes that copies the
- /// specified value into the registers specified by this object. This uses
- /// Chain/Flag as the input and updates them for the output Chain/Flag.
- /// If the Flag pointer is NULL, no flag is used.
- void getCopyToRegs(SDValue Val, SelectionDAG &DAG, DebugLoc dl,
- SDValue &Chain, SDValue *Flag) const;
-
- /// AddInlineAsmOperands - Add this value to the specified inlineasm node
- /// operand list. This adds the code marker, matching input operand index
- /// (if applicable), and includes the number of values added into it.
- void AddInlineAsmOperands(unsigned Kind,
- bool HasMatching, unsigned MatchingIdx,
- SelectionDAG &DAG,
- std::vector<SDValue> &Ops) const;
- };
-}
+// Limit the width of DAG chains. This is important in general to prevent
+// prevent DAG-based analysis from blowing up. For example, alias analysis and
+// load clustering may not complete in reasonable time. It is difficult to
+// recognize and avoid this situation within each individual analysis, and
+// future analyses are likely to have the same behavior. Limiting DAG width is
+// the safe approach, and will be especially important with global DAGs.
+//
+// MaxParallelChains default is arbitrarily high to avoid affecting
+// optimization, but could be lowered to improve compile time. Any ld-ld-st-st
+// sequence over this should have been converted to llvm.memcpy by the
+// frontend. It easy to induce this behavior with .ll code such as:
+// %buffer = alloca [4096 x i8]
+// %data = load [4096 x i8]* %argPtr
+// store [4096 x i8] %data, [4096 x i8]* %buffer
+static cl::opt<unsigned>
+MaxParallelChains("dag-chain-limit", cl::desc("Max parallel isel dag chains"),
+ cl::init(64), cl::Hidden);
+
+static SDValue getCopyFromPartsVector(SelectionDAG &DAG, DebugLoc DL,
+ const SDValue *Parts, unsigned NumParts,
+ EVT PartVT, EVT ValueVT);
/// getCopyFromParts - Create a value that contains the specified legal parts
/// combined into the value they represent. If the parts combine to a type
/// larger then ValueVT then AssertOp can be used to specify whether the extra
/// bits are known to be zero (ISD::AssertZext) or sign extended from ValueVT
/// (ISD::AssertSext).
-static SDValue getCopyFromParts(SelectionDAG &DAG, DebugLoc dl,
+static SDValue getCopyFromParts(SelectionDAG &DAG, DebugLoc DL,
const SDValue *Parts,
unsigned NumParts, EVT PartVT, EVT ValueVT,
ISD::NodeType AssertOp = ISD::DELETED_NODE) {
+ if (ValueVT.isVector())
+ return getCopyFromPartsVector(DAG, DL, Parts, NumParts, PartVT, ValueVT);
+
assert(NumParts > 0 && "No parts to assemble!");
const TargetLowering &TLI = DAG.getTargetLoweringInfo();
SDValue Val = Parts[0];
if (NumParts > 1) {
// Assemble the value from multiple parts.
- if (!ValueVT.isVector() && ValueVT.isInteger()) {
+ if (ValueVT.isInteger()) {
unsigned PartBits = PartVT.getSizeInBits();
unsigned ValueBits = ValueVT.getSizeInBits();
EVT HalfVT = EVT::getIntegerVT(*DAG.getContext(), RoundBits/2);
if (RoundParts > 2) {
- Lo = getCopyFromParts(DAG, dl, Parts, RoundParts / 2,
+ Lo = getCopyFromParts(DAG, DL, Parts, RoundParts / 2,
PartVT, HalfVT);
- Hi = getCopyFromParts(DAG, dl, Parts + RoundParts / 2,
+ Hi = getCopyFromParts(DAG, DL, Parts + RoundParts / 2,
RoundParts / 2, PartVT, HalfVT);
} else {
- Lo = DAG.getNode(ISD::BIT_CONVERT, dl, HalfVT, Parts[0]);
- Hi = DAG.getNode(ISD::BIT_CONVERT, dl, HalfVT, Parts[1]);
+ Lo = DAG.getNode(ISD::BITCAST, DL, HalfVT, Parts[0]);
+ Hi = DAG.getNode(ISD::BITCAST, DL, HalfVT, Parts[1]);
}
if (TLI.isBigEndian())
std::swap(Lo, Hi);
- Val = DAG.getNode(ISD::BUILD_PAIR, dl, RoundVT, Lo, Hi);
+ Val = DAG.getNode(ISD::BUILD_PAIR, DL, RoundVT, Lo, Hi);
if (RoundParts < NumParts) {
// Assemble the trailing non-power-of-2 part.
unsigned OddParts = NumParts - RoundParts;
EVT OddVT = EVT::getIntegerVT(*DAG.getContext(), OddParts * PartBits);
- Hi = getCopyFromParts(DAG, dl,
+ Hi = getCopyFromParts(DAG, DL,
Parts + RoundParts, OddParts, PartVT, OddVT);
// Combine the round and odd parts.
if (TLI.isBigEndian())
std::swap(Lo, Hi);
EVT TotalVT = EVT::getIntegerVT(*DAG.getContext(), NumParts * PartBits);
- Hi = DAG.getNode(ISD::ANY_EXTEND, dl, TotalVT, Hi);
- Hi = DAG.getNode(ISD::SHL, dl, TotalVT, Hi,
+ Hi = DAG.getNode(ISD::ANY_EXTEND, DL, TotalVT, Hi);
+ Hi = DAG.getNode(ISD::SHL, DL, TotalVT, Hi,
DAG.getConstant(Lo.getValueType().getSizeInBits(),
TLI.getPointerTy()));
- Lo = DAG.getNode(ISD::ZERO_EXTEND, dl, TotalVT, Lo);
- Val = DAG.getNode(ISD::OR, dl, TotalVT, Lo, Hi);
- }
- } else if (ValueVT.isVector()) {
- // Handle a multi-element vector.
- EVT IntermediateVT, RegisterVT;
- unsigned NumIntermediates;
- unsigned NumRegs =
- TLI.getVectorTypeBreakdown(*DAG.getContext(), ValueVT, IntermediateVT,
- NumIntermediates, RegisterVT);
- assert(NumRegs == NumParts
- && "Part count doesn't match vector breakdown!");
- NumParts = NumRegs; // Silence a compiler warning.
- assert(RegisterVT == PartVT
- && "Part type doesn't match vector breakdown!");
- assert(RegisterVT == Parts[0].getValueType() &&
- "Part type doesn't match part!");
-
- // Assemble the parts into intermediate operands.
- SmallVector<SDValue, 8> Ops(NumIntermediates);
- if (NumIntermediates == NumParts) {
- // If the register was not expanded, truncate or copy the value,
- // as appropriate.
- for (unsigned i = 0; i != NumParts; ++i)
- Ops[i] = getCopyFromParts(DAG, dl, &Parts[i], 1,
- PartVT, IntermediateVT);
- } else if (NumParts > 0) {
- // If the intermediate type was expanded, build the intermediate
- // operands from the parts.
- assert(NumParts % NumIntermediates == 0 &&
- "Must expand into a divisible number of parts!");
- unsigned Factor = NumParts / NumIntermediates;
- for (unsigned i = 0; i != NumIntermediates; ++i)
- Ops[i] = getCopyFromParts(DAG, dl, &Parts[i * Factor], Factor,
- PartVT, IntermediateVT);
+ Lo = DAG.getNode(ISD::ZERO_EXTEND, DL, TotalVT, Lo);
+ Val = DAG.getNode(ISD::OR, DL, TotalVT, Lo, Hi);
}
-
- // Build a vector with BUILD_VECTOR or CONCAT_VECTORS from the
- // intermediate operands.
- Val = DAG.getNode(IntermediateVT.isVector() ?
- ISD::CONCAT_VECTORS : ISD::BUILD_VECTOR, dl,
- ValueVT, &Ops[0], NumIntermediates);
} else if (PartVT.isFloatingPoint()) {
// FP split into multiple FP parts (for ppcf128)
assert(ValueVT == EVT(MVT::ppcf128) && PartVT == EVT(MVT::f64) &&
"Unexpected split");
SDValue Lo, Hi;
- Lo = DAG.getNode(ISD::BIT_CONVERT, dl, EVT(MVT::f64), Parts[0]);
- Hi = DAG.getNode(ISD::BIT_CONVERT, dl, EVT(MVT::f64), Parts[1]);
+ Lo = DAG.getNode(ISD::BITCAST, DL, EVT(MVT::f64), Parts[0]);
+ Hi = DAG.getNode(ISD::BITCAST, DL, EVT(MVT::f64), Parts[1]);
if (TLI.isBigEndian())
std::swap(Lo, Hi);
- Val = DAG.getNode(ISD::BUILD_PAIR, dl, ValueVT, Lo, Hi);
+ Val = DAG.getNode(ISD::BUILD_PAIR, DL, ValueVT, Lo, Hi);
} else {
// FP split into integer parts (soft fp)
assert(ValueVT.isFloatingPoint() && PartVT.isInteger() &&
!PartVT.isVector() && "Unexpected split");
EVT IntVT = EVT::getIntegerVT(*DAG.getContext(), ValueVT.getSizeInBits());
- Val = getCopyFromParts(DAG, dl, Parts, NumParts, PartVT, IntVT);
+ Val = getCopyFromParts(DAG, DL, Parts, NumParts, PartVT, IntVT);
}
}
if (PartVT == ValueVT)
return Val;
- if (PartVT.isVector()) {
- assert(ValueVT.isVector() && "Unknown vector conversion!");
- return DAG.getNode(ISD::BIT_CONVERT, dl, ValueVT, Val);
- }
-
- if (ValueVT.isVector()) {
- assert(ValueVT.getVectorElementType() == PartVT &&
- ValueVT.getVectorNumElements() == 1 &&
- "Only trivial scalar-to-vector conversions should get here!");
- return DAG.getNode(ISD::BUILD_VECTOR, dl, ValueVT, Val);
- }
-
- if (PartVT.isInteger() &&
- ValueVT.isInteger()) {
+ if (PartVT.isInteger() && ValueVT.isInteger()) {
if (ValueVT.bitsLT(PartVT)) {
// For a truncate, see if we have any information to
// indicate whether the truncated bits will always be
// zero or sign-extension.
if (AssertOp != ISD::DELETED_NODE)
- Val = DAG.getNode(AssertOp, dl, PartVT, Val,
+ Val = DAG.getNode(AssertOp, DL, PartVT, Val,
DAG.getValueType(ValueVT));
- return DAG.getNode(ISD::TRUNCATE, dl, ValueVT, Val);
- } else {
- return DAG.getNode(ISD::ANY_EXTEND, dl, ValueVT, Val);
+ return DAG.getNode(ISD::TRUNCATE, DL, ValueVT, Val);
}
+ return DAG.getNode(ISD::ANY_EXTEND, DL, ValueVT, Val);
}
if (PartVT.isFloatingPoint() && ValueVT.isFloatingPoint()) {
- if (ValueVT.bitsLT(Val.getValueType())) {
- // FP_ROUND's are always exact here.
- return DAG.getNode(ISD::FP_ROUND, dl, ValueVT, Val,
+ // FP_ROUND's are always exact here.
+ if (ValueVT.bitsLT(Val.getValueType()))
+ return DAG.getNode(ISD::FP_ROUND, DL, ValueVT, Val,
DAG.getIntPtrConstant(1));
- }
- return DAG.getNode(ISD::FP_EXTEND, dl, ValueVT, Val);
+ return DAG.getNode(ISD::FP_EXTEND, DL, ValueVT, Val);
}
if (PartVT.getSizeInBits() == ValueVT.getSizeInBits())
- return DAG.getNode(ISD::BIT_CONVERT, dl, ValueVT, Val);
+ return DAG.getNode(ISD::BITCAST, DL, ValueVT, Val);
llvm_unreachable("Unknown mismatch!");
return SDValue();
}
+/// getCopyFromParts - Create a value that contains the specified legal parts
+/// combined into the value they represent. If the parts combine to a type
+/// larger then ValueVT then AssertOp can be used to specify whether the extra
+/// bits are known to be zero (ISD::AssertZext) or sign extended from ValueVT
+/// (ISD::AssertSext).
+static SDValue getCopyFromPartsVector(SelectionDAG &DAG, DebugLoc DL,
+ const SDValue *Parts, unsigned NumParts,
+ EVT PartVT, EVT ValueVT) {
+ assert(ValueVT.isVector() && "Not a vector value");
+ assert(NumParts > 0 && "No parts to assemble!");
+ const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+ SDValue Val = Parts[0];
+
+ // Handle a multi-element vector.
+ if (NumParts > 1) {
+ EVT IntermediateVT, RegisterVT;
+ unsigned NumIntermediates;
+ unsigned NumRegs =
+ TLI.getVectorTypeBreakdown(*DAG.getContext(), ValueVT, IntermediateVT,
+ NumIntermediates, RegisterVT);
+ assert(NumRegs == NumParts && "Part count doesn't match vector breakdown!");
+ NumParts = NumRegs; // Silence a compiler warning.
+ assert(RegisterVT == PartVT && "Part type doesn't match vector breakdown!");
+ assert(RegisterVT == Parts[0].getValueType() &&
+ "Part type doesn't match part!");
+
+ // Assemble the parts into intermediate operands.
+ SmallVector<SDValue, 8> Ops(NumIntermediates);
+ if (NumIntermediates == NumParts) {
+ // If the register was not expanded, truncate or copy the value,
+ // as appropriate.
+ for (unsigned i = 0; i != NumParts; ++i)
+ Ops[i] = getCopyFromParts(DAG, DL, &Parts[i], 1,
+ PartVT, IntermediateVT);
+ } else if (NumParts > 0) {
+ // If the intermediate type was expanded, build the intermediate
+ // operands from the parts.
+ assert(NumParts % NumIntermediates == 0 &&
+ "Must expand into a divisible number of parts!");
+ unsigned Factor = NumParts / NumIntermediates;
+ for (unsigned i = 0; i != NumIntermediates; ++i)
+ Ops[i] = getCopyFromParts(DAG, DL, &Parts[i * Factor], Factor,
+ PartVT, IntermediateVT);
+ }
+
+ // Build a vector with BUILD_VECTOR or CONCAT_VECTORS from the
+ // intermediate operands.
+ Val = DAG.getNode(IntermediateVT.isVector() ?
+ ISD::CONCAT_VECTORS : ISD::BUILD_VECTOR, DL,
+ ValueVT, &Ops[0], NumIntermediates);
+ }
+
+ // There is now one part, held in Val. Correct it to match ValueVT.
+ PartVT = Val.getValueType();
+
+ if (PartVT == ValueVT)
+ return Val;
+
+ if (PartVT.isVector()) {
+ // If the element type of the source/dest vectors are the same, but the
+ // parts vector has more elements than the value vector, then we have a
+ // vector widening case (e.g. <2 x float> -> <4 x float>). Extract the
+ // elements we want.
+ if (PartVT.getVectorElementType() == ValueVT.getVectorElementType()) {
+ assert(PartVT.getVectorNumElements() > ValueVT.getVectorNumElements() &&
+ "Cannot narrow, it would be a lossy transformation");
+ return DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, ValueVT, Val,
+ DAG.getIntPtrConstant(0));
+ }
+
+ // Vector/Vector bitcast.
+ return DAG.getNode(ISD::BITCAST, DL, ValueVT, Val);
+ }
+
+ assert(ValueVT.getVectorElementType() == PartVT &&
+ ValueVT.getVectorNumElements() == 1 &&
+ "Only trivial scalar-to-vector conversions should get here!");
+ return DAG.getNode(ISD::BUILD_VECTOR, DL, ValueVT, Val);
+}
+
+
+
+
+static void getCopyToPartsVector(SelectionDAG &DAG, DebugLoc dl,
+ SDValue Val, SDValue *Parts, unsigned NumParts,
+ EVT PartVT);
+
/// getCopyToParts - Create a series of nodes that contain the specified value
/// split into legal parts. If the parts contain more bits than Val, then, for
/// integers, ExtendKind can be used to specify how to generate the extra bits.
-static void getCopyToParts(SelectionDAG &DAG, DebugLoc dl,
+static void getCopyToParts(SelectionDAG &DAG, DebugLoc DL,
SDValue Val, SDValue *Parts, unsigned NumParts,
EVT PartVT,
ISD::NodeType ExtendKind = ISD::ANY_EXTEND) {
- const TargetLowering &TLI = DAG.getTargetLoweringInfo();
- EVT PtrVT = TLI.getPointerTy();
EVT ValueVT = Val.getValueType();
+
+ // Handle the vector case separately.
+ if (ValueVT.isVector())
+ return getCopyToPartsVector(DAG, DL, Val, Parts, NumParts, PartVT);
+
+ const TargetLowering &TLI = DAG.getTargetLoweringInfo();
unsigned PartBits = PartVT.getSizeInBits();
unsigned OrigNumParts = NumParts;
assert(TLI.isTypeLegal(PartVT) && "Copying to an illegal type!");
- if (!NumParts)
+ if (NumParts == 0)
return;
- if (!ValueVT.isVector()) {
+ assert(!ValueVT.isVector() && "Vector case handled elsewhere");
+ if (PartVT == ValueVT) {
+ assert(NumParts == 1 && "No-op copy with multiple parts!");
+ Parts[0] = Val;
+ return;
+ }
+
+ if (NumParts * PartBits > ValueVT.getSizeInBits()) {
+ // If the parts cover more bits than the value has, promote the value.
+ if (PartVT.isFloatingPoint() && ValueVT.isFloatingPoint()) {
+ assert(NumParts == 1 && "Do not know what to promote to!");
+ Val = DAG.getNode(ISD::FP_EXTEND, DL, PartVT, Val);
+ } else {
+ assert(PartVT.isInteger() && ValueVT.isInteger() &&
+ "Unknown mismatch!");
+ ValueVT = EVT::getIntegerVT(*DAG.getContext(), NumParts * PartBits);
+ Val = DAG.getNode(ExtendKind, DL, ValueVT, Val);
+ }
+ } else if (PartBits == ValueVT.getSizeInBits()) {
+ // Different types of the same size.
+ assert(NumParts == 1 && PartVT != ValueVT);
+ Val = DAG.getNode(ISD::BITCAST, DL, PartVT, Val);
+ } else if (NumParts * PartBits < ValueVT.getSizeInBits()) {
+ // If the parts cover less bits than value has, truncate the value.
+ assert(PartVT.isInteger() && ValueVT.isInteger() &&
+ "Unknown mismatch!");
+ ValueVT = EVT::getIntegerVT(*DAG.getContext(), NumParts * PartBits);
+ Val = DAG.getNode(ISD::TRUNCATE, DL, ValueVT, Val);
+ }
+
+ // The value may have changed - recompute ValueVT.
+ ValueVT = Val.getValueType();
+ assert(NumParts * PartBits == ValueVT.getSizeInBits() &&
+ "Failed to tile the value with PartVT!");
+
+ if (NumParts == 1) {
+ assert(PartVT == ValueVT && "Type conversion failed!");
+ Parts[0] = Val;
+ return;
+ }
+
+ // Expand the value into multiple parts.
+ if (NumParts & (NumParts - 1)) {
+ // The number of parts is not a power of 2. Split off and copy the tail.
+ assert(PartVT.isInteger() && ValueVT.isInteger() &&
+ "Do not know what to expand to!");
+ unsigned RoundParts = 1 << Log2_32(NumParts);
+ unsigned RoundBits = RoundParts * PartBits;
+ unsigned OddParts = NumParts - RoundParts;
+ SDValue OddVal = DAG.getNode(ISD::SRL, DL, ValueVT, Val,
+ DAG.getIntPtrConstant(RoundBits));
+ getCopyToParts(DAG, DL, OddVal, Parts + RoundParts, OddParts, PartVT);
+
+ if (TLI.isBigEndian())
+ // The odd parts were reversed by getCopyToParts - unreverse them.
+ std::reverse(Parts + RoundParts, Parts + NumParts);
+
+ NumParts = RoundParts;
+ ValueVT = EVT::getIntegerVT(*DAG.getContext(), NumParts * PartBits);
+ Val = DAG.getNode(ISD::TRUNCATE, DL, ValueVT, Val);
+ }
+
+ // The number of parts is a power of 2. Repeatedly bisect the value using
+ // EXTRACT_ELEMENT.
+ Parts[0] = DAG.getNode(ISD::BITCAST, DL,
+ EVT::getIntegerVT(*DAG.getContext(),
+ ValueVT.getSizeInBits()),
+ Val);
+
+ for (unsigned StepSize = NumParts; StepSize > 1; StepSize /= 2) {
+ for (unsigned i = 0; i < NumParts; i += StepSize) {
+ unsigned ThisBits = StepSize * PartBits / 2;
+ EVT ThisVT = EVT::getIntegerVT(*DAG.getContext(), ThisBits);
+ SDValue &Part0 = Parts[i];
+ SDValue &Part1 = Parts[i+StepSize/2];
+
+ Part1 = DAG.getNode(ISD::EXTRACT_ELEMENT, DL,
+ ThisVT, Part0, DAG.getIntPtrConstant(1));
+ Part0 = DAG.getNode(ISD::EXTRACT_ELEMENT, DL,
+ ThisVT, Part0, DAG.getIntPtrConstant(0));
+
+ if (ThisBits == PartBits && ThisVT != PartVT) {
+ Part0 = DAG.getNode(ISD::BITCAST, DL, PartVT, Part0);
+ Part1 = DAG.getNode(ISD::BITCAST, DL, PartVT, Part1);
+ }
+ }
+ }
+
+ if (TLI.isBigEndian())
+ std::reverse(Parts, Parts + OrigNumParts);
+}
+
+
+/// getCopyToPartsVector - Create a series of nodes that contain the specified
+/// value split into legal parts.
+static void getCopyToPartsVector(SelectionDAG &DAG, DebugLoc DL,
+ SDValue Val, SDValue *Parts, unsigned NumParts,
+ EVT PartVT) {
+ EVT ValueVT = Val.getValueType();
+ assert(ValueVT.isVector() && "Not a vector");
+ const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+
+ if (NumParts == 1) {
if (PartVT == ValueVT) {
- assert(NumParts == 1 && "No-op copy with multiple parts!");
- Parts[0] = Val;
- return;
+ // Nothing to do.
+ } else if (PartVT.getSizeInBits() == ValueVT.getSizeInBits()) {
+ // Bitconvert vector->vector case.
+ Val = DAG.getNode(ISD::BITCAST, DL, PartVT, Val);
+ } else if (PartVT.isVector() &&
+ PartVT.getVectorElementType() == ValueVT.getVectorElementType()&&
+ PartVT.getVectorNumElements() > ValueVT.getVectorNumElements()) {
+ EVT ElementVT = PartVT.getVectorElementType();
+ // Vector widening case, e.g. <2 x float> -> <4 x float>. Shuffle in
+ // undef elements.
+ SmallVector<SDValue, 16> Ops;
+ for (unsigned i = 0, e = ValueVT.getVectorNumElements(); i != e; ++i)
+ Ops.push_back(DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL,
+ ElementVT, Val, DAG.getIntPtrConstant(i)));
+
+ for (unsigned i = ValueVT.getVectorNumElements(),
+ e = PartVT.getVectorNumElements(); i != e; ++i)
+ Ops.push_back(DAG.getUNDEF(ElementVT));
+
+ Val = DAG.getNode(ISD::BUILD_VECTOR, DL, PartVT, &Ops[0], Ops.size());
+
+ // FIXME: Use CONCAT for 2x -> 4x.
+
+ //SDValue UndefElts = DAG.getUNDEF(VectorTy);
+ //Val = DAG.getNode(ISD::CONCAT_VECTORS, DL, PartVT, Val, UndefElts);
+ } else {
+ // Vector -> scalar conversion.
+ assert(ValueVT.getVectorElementType() == PartVT &&
+ ValueVT.getVectorNumElements() == 1 &&
+ "Only trivial vector-to-scalar conversions should get here!");
+ Val = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL,
+ PartVT, Val, DAG.getIntPtrConstant(0));
}
- if (NumParts * PartBits > ValueVT.getSizeInBits()) {
- // If the parts cover more bits than the value has, promote the value.
- if (PartVT.isFloatingPoint() && ValueVT.isFloatingPoint()) {
- assert(NumParts == 1 && "Do not know what to promote to!");
- Val = DAG.getNode(ISD::FP_EXTEND, dl, PartVT, Val);
- } else if (PartVT.isInteger() && ValueVT.isInteger()) {
- ValueVT = EVT::getIntegerVT(*DAG.getContext(), NumParts * PartBits);
- Val = DAG.getNode(ExtendKind, dl, ValueVT, Val);
- } else {
- llvm_unreachable("Unknown mismatch!");
+ Parts[0] = Val;
+ return;
+ }
+
+ // Handle a multi-element vector.
+ EVT IntermediateVT, RegisterVT;
+ unsigned NumIntermediates;
+ unsigned NumRegs = TLI.getVectorTypeBreakdown(*DAG.getContext(), ValueVT,
+ IntermediateVT,
+ NumIntermediates, RegisterVT);
+ unsigned NumElements = ValueVT.getVectorNumElements();
+
+ assert(NumRegs == NumParts && "Part count doesn't match vector breakdown!");
+ NumParts = NumRegs; // Silence a compiler warning.
+ assert(RegisterVT == PartVT && "Part type doesn't match vector breakdown!");
+
+ // Split the vector into intermediate operands.
+ SmallVector<SDValue, 8> Ops(NumIntermediates);
+ for (unsigned i = 0; i != NumIntermediates; ++i) {
+ if (IntermediateVT.isVector())
+ Ops[i] = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL,
+ IntermediateVT, Val,
+ DAG.getIntPtrConstant(i * (NumElements / NumIntermediates)));
+ else
+ Ops[i] = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL,
+ IntermediateVT, Val, DAG.getIntPtrConstant(i));
+ }
+
+ // Split the intermediate operands into legal parts.
+ if (NumParts == NumIntermediates) {
+ // If the register was not expanded, promote or copy the value,
+ // as appropriate.
+ for (unsigned i = 0; i != NumParts; ++i)
+ getCopyToParts(DAG, DL, Ops[i], &Parts[i], 1, PartVT);
+ } else if (NumParts > 0) {
+ // If the intermediate type was expanded, split each the value into
+ // legal parts.
+ assert(NumParts % NumIntermediates == 0 &&
+ "Must expand into a divisible number of parts!");
+ unsigned Factor = NumParts / NumIntermediates;
+ for (unsigned i = 0; i != NumIntermediates; ++i)
+ getCopyToParts(DAG, DL, Ops[i], &Parts[i*Factor], Factor, PartVT);
+ }
+}
+
+
+
+
+namespace {
+ /// RegsForValue - This struct represents the registers (physical or virtual)
+ /// that a particular set of values is assigned, and the type information
+ /// about the value. The most common situation is to represent one value at a
+ /// time, but struct or array values are handled element-wise as multiple
+ /// values. The splitting of aggregates is performed recursively, so that we
+ /// never have aggregate-typed registers. The values at this point do not
+ /// necessarily have legal types, so each value may require one or more
+ /// registers of some legal type.
+ ///
+ struct RegsForValue {
+ /// ValueVTs - The value types of the values, which may not be legal, and
+ /// may need be promoted or synthesized from one or more registers.
+ ///
+ SmallVector<EVT, 4> ValueVTs;
+
+ /// RegVTs - The value types of the registers. This is the same size as
+ /// ValueVTs and it records, for each value, what the type of the assigned
+ /// register or registers are. (Individual values are never synthesized
+ /// from more than one type of register.)
+ ///
+ /// With virtual registers, the contents of RegVTs is redundant with TLI's
+ /// getRegisterType member function, however when with physical registers
+ /// it is necessary to have a separate record of the types.
+ ///
+ SmallVector<EVT, 4> RegVTs;
+
+ /// Regs - This list holds the registers assigned to the values.
+ /// Each legal or promoted value requires one register, and each
+ /// expanded value requires multiple registers.
+ ///
+ SmallVector<unsigned, 4> Regs;
+
+ RegsForValue() {}
+
+ RegsForValue(const SmallVector<unsigned, 4> ®s,
+ EVT regvt, EVT valuevt)
+ : ValueVTs(1, valuevt), RegVTs(1, regvt), Regs(regs) {}
+
+ RegsForValue(LLVMContext &Context, const TargetLowering &tli,
+ unsigned Reg, const Type *Ty) {
+ ComputeValueVTs(tli, Ty, ValueVTs);
+
+ for (unsigned Value = 0, e = ValueVTs.size(); Value != e; ++Value) {
+ EVT ValueVT = ValueVTs[Value];
+ unsigned NumRegs = tli.getNumRegisters(Context, ValueVT);
+ EVT RegisterVT = tli.getRegisterType(Context, ValueVT);
+ for (unsigned i = 0; i != NumRegs; ++i)
+ Regs.push_back(Reg + i);
+ RegVTs.push_back(RegisterVT);
+ Reg += NumRegs;
}
- } else if (PartBits == ValueVT.getSizeInBits()) {
- // Different types of the same size.
- assert(NumParts == 1 && PartVT != ValueVT);
- Val = DAG.getNode(ISD::BIT_CONVERT, dl, PartVT, Val);
- } else if (NumParts * PartBits < ValueVT.getSizeInBits()) {
- // If the parts cover less bits than value has, truncate the value.
- if (PartVT.isInteger() && ValueVT.isInteger()) {
- ValueVT = EVT::getIntegerVT(*DAG.getContext(), NumParts * PartBits);
- Val = DAG.getNode(ISD::TRUNCATE, dl, ValueVT, Val);
+ }
+
+ /// areValueTypesLegal - Return true if types of all the values are legal.
+ bool areValueTypesLegal(const TargetLowering &TLI) {
+ for (unsigned Value = 0, e = ValueVTs.size(); Value != e; ++Value) {
+ EVT RegisterVT = RegVTs[Value];
+ if (!TLI.isTypeLegal(RegisterVT))
+ return false;
+ }
+ return true;
+ }
+
+ /// append - Add the specified values to this one.
+ void append(const RegsForValue &RHS) {
+ ValueVTs.append(RHS.ValueVTs.begin(), RHS.ValueVTs.end());
+ RegVTs.append(RHS.RegVTs.begin(), RHS.RegVTs.end());
+ Regs.append(RHS.Regs.begin(), RHS.Regs.end());
+ }
+
+ /// getCopyFromRegs - Emit a series of CopyFromReg nodes that copies from
+ /// this value and returns the result as a ValueVTs value. This uses
+ /// Chain/Flag as the input and updates them for the output Chain/Flag.
+ /// If the Flag pointer is NULL, no flag is used.
+ SDValue getCopyFromRegs(SelectionDAG &DAG, FunctionLoweringInfo &FuncInfo,
+ DebugLoc dl,
+ SDValue &Chain, SDValue *Flag) const;
+
+ /// getCopyToRegs - Emit a series of CopyToReg nodes that copies the
+ /// specified value into the registers specified by this object. This uses
+ /// Chain/Flag as the input and updates them for the output Chain/Flag.
+ /// If the Flag pointer is NULL, no flag is used.
+ void getCopyToRegs(SDValue Val, SelectionDAG &DAG, DebugLoc dl,
+ SDValue &Chain, SDValue *Flag) const;
+
+ /// AddInlineAsmOperands - Add this value to the specified inlineasm node
+ /// operand list. This adds the code marker, matching input operand index
+ /// (if applicable), and includes the number of values added into it.
+ void AddInlineAsmOperands(unsigned Kind,
+ bool HasMatching, unsigned MatchingIdx,
+ SelectionDAG &DAG,
+ std::vector<SDValue> &Ops) const;
+ };
+}
+
+/// getCopyFromRegs - Emit a series of CopyFromReg nodes that copies from
+/// this value and returns the result as a ValueVT value. This uses
+/// Chain/Flag as the input and updates them for the output Chain/Flag.
+/// If the Flag pointer is NULL, no flag is used.
+SDValue RegsForValue::getCopyFromRegs(SelectionDAG &DAG,
+ FunctionLoweringInfo &FuncInfo,
+ DebugLoc dl,
+ SDValue &Chain, SDValue *Flag) const {
+ // A Value with type {} or [0 x %t] needs no registers.
+ if (ValueVTs.empty())
+ return SDValue();
+
+ const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+
+ // Assemble the legal parts into the final values.
+ SmallVector<SDValue, 4> Values(ValueVTs.size());
+ SmallVector<SDValue, 8> Parts;
+ for (unsigned Value = 0, Part = 0, e = ValueVTs.size(); Value != e; ++Value) {
+ // Copy the legal parts from the registers.
+ EVT ValueVT = ValueVTs[Value];
+ unsigned NumRegs = TLI.getNumRegisters(*DAG.getContext(), ValueVT);
+ EVT RegisterVT = RegVTs[Value];
+
+ Parts.resize(NumRegs);
+ for (unsigned i = 0; i != NumRegs; ++i) {
+ SDValue P;
+ if (Flag == 0) {
+ P = DAG.getCopyFromReg(Chain, dl, Regs[Part+i], RegisterVT);
} else {
- llvm_unreachable("Unknown mismatch!");
+ P = DAG.getCopyFromReg(Chain, dl, Regs[Part+i], RegisterVT, *Flag);
+ *Flag = P.getValue(2);
}
- }
- // The value may have changed - recompute ValueVT.
- ValueVT = Val.getValueType();
- assert(NumParts * PartBits == ValueVT.getSizeInBits() &&
- "Failed to tile the value with PartVT!");
+ Chain = P.getValue(1);
+ Parts[i] = P;
- if (NumParts == 1) {
- assert(PartVT == ValueVT && "Type conversion failed!");
- Parts[0] = Val;
- return;
- }
+ // If the source register was virtual and if we know something about it,
+ // add an assert node.
+ if (!TargetRegisterInfo::isVirtualRegister(Regs[Part+i]) ||
+ !RegisterVT.isInteger() || RegisterVT.isVector() ||
+ !FuncInfo.LiveOutRegInfo.inBounds(Regs[Part+i]))
+ continue;
+
+ const FunctionLoweringInfo::LiveOutInfo &LOI =
+ FuncInfo.LiveOutRegInfo[Regs[Part+i]];
+
+ unsigned RegSize = RegisterVT.getSizeInBits();
+ unsigned NumSignBits = LOI.NumSignBits;
+ unsigned NumZeroBits = LOI.KnownZero.countLeadingOnes();
+
+ // FIXME: We capture more information than the dag can represent. For
+ // now, just use the tightest assertzext/assertsext possible.
+ bool isSExt = true;
+ EVT FromVT(MVT::Other);
+ if (NumSignBits == RegSize)
+ isSExt = true, FromVT = MVT::i1; // ASSERT SEXT 1
+ else if (NumZeroBits >= RegSize-1)
+ isSExt = false, FromVT = MVT::i1; // ASSERT ZEXT 1
+ else if (NumSignBits > RegSize-8)
+ isSExt = true, FromVT = MVT::i8; // ASSERT SEXT 8
+ else if (NumZeroBits >= RegSize-8)
+ isSExt = false, FromVT = MVT::i8; // ASSERT ZEXT 8
+ else if (NumSignBits > RegSize-16)
+ isSExt = true, FromVT = MVT::i16; // ASSERT SEXT 16
+ else if (NumZeroBits >= RegSize-16)
+ isSExt = false, FromVT = MVT::i16; // ASSERT ZEXT 16
+ else if (NumSignBits > RegSize-32)
+ isSExt = true, FromVT = MVT::i32; // ASSERT SEXT 32
+ else if (NumZeroBits >= RegSize-32)
+ isSExt = false, FromVT = MVT::i32; // ASSERT ZEXT 32
+ else
+ continue;
- // Expand the value into multiple parts.
- if (NumParts & (NumParts - 1)) {
- // The number of parts is not a power of 2. Split off and copy the tail.
- assert(PartVT.isInteger() && ValueVT.isInteger() &&
- "Do not know what to expand to!");
- unsigned RoundParts = 1 << Log2_32(NumParts);
- unsigned RoundBits = RoundParts * PartBits;
- unsigned OddParts = NumParts - RoundParts;
- SDValue OddVal = DAG.getNode(ISD::SRL, dl, ValueVT, Val,
- DAG.getConstant(RoundBits,
- TLI.getPointerTy()));
- getCopyToParts(DAG, dl, OddVal, Parts + RoundParts,
- OddParts, PartVT);
+ // Add an assertion node.
+ assert(FromVT != MVT::Other);
+ Parts[i] = DAG.getNode(isSExt ? ISD::AssertSext : ISD::AssertZext, dl,
+ RegisterVT, P, DAG.getValueType(FromVT));
+ }
- if (TLI.isBigEndian())
- // The odd parts were reversed by getCopyToParts - unreverse them.
- std::reverse(Parts + RoundParts, Parts + NumParts);
+ Values[Value] = getCopyFromParts(DAG, dl, Parts.begin(),
+ NumRegs, RegisterVT, ValueVT);
+ Part += NumRegs;
+ Parts.clear();
+ }
- NumParts = RoundParts;
- ValueVT = EVT::getIntegerVT(*DAG.getContext(), NumParts * PartBits);
- Val = DAG.getNode(ISD::TRUNCATE, dl, ValueVT, Val);
- }
+ return DAG.getNode(ISD::MERGE_VALUES, dl,
+ DAG.getVTList(&ValueVTs[0], ValueVTs.size()),
+ &Values[0], ValueVTs.size());
+}
- // The number of parts is a power of 2. Repeatedly bisect the value using
- // EXTRACT_ELEMENT.
- Parts[0] = DAG.getNode(ISD::BIT_CONVERT, dl,
- EVT::getIntegerVT(*DAG.getContext(),
- ValueVT.getSizeInBits()),
- Val);
-
- for (unsigned StepSize = NumParts; StepSize > 1; StepSize /= 2) {
- for (unsigned i = 0; i < NumParts; i += StepSize) {
- unsigned ThisBits = StepSize * PartBits / 2;
- EVT ThisVT = EVT::getIntegerVT(*DAG.getContext(), ThisBits);
- SDValue &Part0 = Parts[i];
- SDValue &Part1 = Parts[i+StepSize/2];
-
- Part1 = DAG.getNode(ISD::EXTRACT_ELEMENT, dl,
- ThisVT, Part0,
- DAG.getConstant(1, PtrVT));
- Part0 = DAG.getNode(ISD::EXTRACT_ELEMENT, dl,
- ThisVT, Part0,
- DAG.getConstant(0, PtrVT));
-
- if (ThisBits == PartBits && ThisVT != PartVT) {
- Part0 = DAG.getNode(ISD::BIT_CONVERT, dl,
- PartVT, Part0);
- Part1 = DAG.getNode(ISD::BIT_CONVERT, dl,
- PartVT, Part1);
- }
- }
- }
+/// getCopyToRegs - Emit a series of CopyToReg nodes that copies the
+/// specified value into the registers specified by this object. This uses
+/// Chain/Flag as the input and updates them for the output Chain/Flag.
+/// If the Flag pointer is NULL, no flag is used.
+void RegsForValue::getCopyToRegs(SDValue Val, SelectionDAG &DAG, DebugLoc dl,
+ SDValue &Chain, SDValue *Flag) const {
+ const TargetLowering &TLI = DAG.getTargetLoweringInfo();
- if (TLI.isBigEndian())
- std::reverse(Parts, Parts + OrigNumParts);
+ // Get the list of the values's legal parts.
+ unsigned NumRegs = Regs.size();
+ SmallVector<SDValue, 8> Parts(NumRegs);
+ for (unsigned Value = 0, Part = 0, e = ValueVTs.size(); Value != e; ++Value) {
+ EVT ValueVT = ValueVTs[Value];
+ unsigned NumParts = TLI.getNumRegisters(*DAG.getContext(), ValueVT);
+ EVT RegisterVT = RegVTs[Value];
- return;
+ getCopyToParts(DAG, dl, Val.getValue(Val.getResNo() + Value),
+ &Parts[Part], NumParts, RegisterVT);
+ Part += NumParts;
}
- // Vector ValueVT.
- if (NumParts == 1) {
- if (PartVT != ValueVT) {
- if (PartVT.getSizeInBits() == ValueVT.getSizeInBits()) {
- Val = DAG.getNode(ISD::BIT_CONVERT, dl, PartVT, Val);
- } else {
- assert(ValueVT.getVectorElementType() == PartVT &&
- ValueVT.getVectorNumElements() == 1 &&
- "Only trivial vector-to-scalar conversions should get here!");
- Val = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl,
- PartVT, Val,
- DAG.getConstant(0, PtrVT));
- }
+ // Copy the parts into the registers.
+ SmallVector<SDValue, 8> Chains(NumRegs);
+ for (unsigned i = 0; i != NumRegs; ++i) {
+ SDValue Part;
+ if (Flag == 0) {
+ Part = DAG.getCopyToReg(Chain, dl, Regs[i], Parts[i]);
+ } else {
+ Part = DAG.getCopyToReg(Chain, dl, Regs[i], Parts[i], *Flag);
+ *Flag = Part.getValue(1);
}
- Parts[0] = Val;
- return;
+ Chains[i] = Part.getValue(0);
}
- // Handle a multi-element vector.
- EVT IntermediateVT, RegisterVT;
- unsigned NumIntermediates;
- unsigned NumRegs = TLI.getVectorTypeBreakdown(*DAG.getContext(), ValueVT,
- IntermediateVT, NumIntermediates, RegisterVT);
- unsigned NumElements = ValueVT.getVectorNumElements();
+ if (NumRegs == 1 || Flag)
+ // If NumRegs > 1 && Flag is used then the use of the last CopyToReg is
+ // flagged to it. That is the CopyToReg nodes and the user are considered
+ // a single scheduling unit. If we create a TokenFactor and return it as
+ // chain, then the TokenFactor is both a predecessor (operand) of the
+ // user as well as a successor (the TF operands are flagged to the user).
+ // c1, f1 = CopyToReg
+ // c2, f2 = CopyToReg
+ // c3 = TokenFactor c1, c2
+ // ...
+ // = op c3, ..., f2
+ Chain = Chains[NumRegs-1];
+ else
+ Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, &Chains[0], NumRegs);
+}
- assert(NumRegs == NumParts && "Part count doesn't match vector breakdown!");
- NumParts = NumRegs; // Silence a compiler warning.
- assert(RegisterVT == PartVT && "Part type doesn't match vector breakdown!");
+/// AddInlineAsmOperands - Add this value to the specified inlineasm node
+/// operand list. This adds the code marker and includes the number of
+/// values added into it.
+void RegsForValue::AddInlineAsmOperands(unsigned Code, bool HasMatching,
+ unsigned MatchingIdx,
+ SelectionDAG &DAG,
+ std::vector<SDValue> &Ops) const {
+ const TargetLowering &TLI = DAG.getTargetLoweringInfo();
- // Split the vector into intermediate operands.
- SmallVector<SDValue, 8> Ops(NumIntermediates);
- for (unsigned i = 0; i != NumIntermediates; ++i) {
- if (IntermediateVT.isVector())
- Ops[i] = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl,
- IntermediateVT, Val,
- DAG.getConstant(i * (NumElements / NumIntermediates),
- PtrVT));
- else
- Ops[i] = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl,
- IntermediateVT, Val,
- DAG.getConstant(i, PtrVT));
- }
+ unsigned Flag = InlineAsm::getFlagWord(Code, Regs.size());
+ if (HasMatching)
+ Flag = InlineAsm::getFlagWordForMatchingOp(Flag, MatchingIdx);
+ SDValue Res = DAG.getTargetConstant(Flag, MVT::i32);
+ Ops.push_back(Res);
- // Split the intermediate operands into legal parts.
- if (NumParts == NumIntermediates) {
- // If the register was not expanded, promote or copy the value,
- // as appropriate.
- for (unsigned i = 0; i != NumParts; ++i)
- getCopyToParts(DAG, dl, Ops[i], &Parts[i], 1, PartVT);
- } else if (NumParts > 0) {
- // If the intermediate type was expanded, split each the value into
- // legal parts.
- assert(NumParts % NumIntermediates == 0 &&
- "Must expand into a divisible number of parts!");
- unsigned Factor = NumParts / NumIntermediates;
- for (unsigned i = 0; i != NumIntermediates; ++i)
- getCopyToParts(DAG, dl, Ops[i], &Parts[i*Factor], Factor, PartVT);
+ for (unsigned Value = 0, Reg = 0, e = ValueVTs.size(); Value != e; ++Value) {
+ unsigned NumRegs = TLI.getNumRegisters(*DAG.getContext(), ValueVTs[Value]);
+ EVT RegisterVT = RegVTs[Value];
+ for (unsigned i = 0; i != NumRegs; ++i) {
+ assert(Reg < Regs.size() && "Mismatch in # registers expected");
+ Ops.push_back(DAG.getRegister(Regs[Reg++], RegisterVT));
+ }
}
}
-
void SelectionDAGBuilder::init(GCFunctionInfo *gfi, AliasAnalysis &aa) {
AA = &aa;
GFI = gfi;
/// consumed.
void SelectionDAGBuilder::clear() {
NodeMap.clear();
+ UnusedArgNodeMap.clear();
PendingLoads.clear();
PendingExports.clear();
- EdgeMapping.clear();
- DAG.clear();
+ DanglingDebugInfoMap.clear();
CurDebugLoc = DebugLoc();
HasTailCall = false;
}
}
void SelectionDAGBuilder::visit(const Instruction &I) {
+ // Set up outgoing PHI node register values before emitting the terminator.
+ if (isa<TerminatorInst>(&I))
+ HandlePHINodesInSuccessorBlocks(I.getParent());
+
+ CurDebugLoc = I.getDebugLoc();
+
visit(I.getOpcode(), I);
+
+ if (!isa<TerminatorInst>(&I) && !HasTailCall)
+ CopyToExportRegsIfNeeded(&I);
+
+ CurDebugLoc = DebugLoc();
+}
+
+void SelectionDAGBuilder::visitPHI(const PHINode &) {
+ llvm_unreachable("SelectionDAGBuilder shouldn't visit PHI nodes!");
}
void SelectionDAGBuilder::visit(unsigned Opcode, const User &I) {
}
}
+// resolveDanglingDebugInfo - if we saw an earlier dbg_value referring to V,
+// generate the debug data structures now that we've seen its definition.
+void SelectionDAGBuilder::resolveDanglingDebugInfo(const Value *V,
+ SDValue Val) {
+ DanglingDebugInfo &DDI = DanglingDebugInfoMap[V];
+ if (DDI.getDI()) {
+ const DbgValueInst *DI = DDI.getDI();
+ DebugLoc dl = DDI.getdl();
+ unsigned DbgSDNodeOrder = DDI.getSDNodeOrder();
+ MDNode *Variable = DI->getVariable();
+ uint64_t Offset = DI->getOffset();
+ SDDbgValue *SDV;
+ if (Val.getNode()) {
+ if (!EmitFuncArgumentDbgValue(V, Variable, Offset, Val)) {
+ SDV = DAG.getDbgValue(Variable, Val.getNode(),
+ Val.getResNo(), Offset, dl, DbgSDNodeOrder);
+ DAG.AddDbgValue(SDV, Val.getNode(), false);
+ }
+ } else
+ DEBUG(dbgs() << "Dropping debug info for " << DI);
+ DanglingDebugInfoMap[V] = DanglingDebugInfo();
+ }
+}
+
+// getValue - Return an SDValue for the given Value.
SDValue SelectionDAGBuilder::getValue(const Value *V) {
+ // If we already have an SDValue for this value, use it. It's important
+ // to do this first, so that we don't create a CopyFromReg if we already
+ // have a regular SDValue.
+ SDValue &N = NodeMap[V];
+ if (N.getNode()) return N;
+
+ // If there's a virtual register allocated and initialized for this
+ // value, use it.
+ DenseMap<const Value *, unsigned>::iterator It = FuncInfo.ValueMap.find(V);
+ if (It != FuncInfo.ValueMap.end()) {
+ unsigned InReg = It->second;
+ RegsForValue RFV(*DAG.getContext(), TLI, InReg, V->getType());
+ SDValue Chain = DAG.getEntryNode();
+ N = RFV.getCopyFromRegs(DAG, FuncInfo, getCurDebugLoc(), Chain,NULL);
+ resolveDanglingDebugInfo(V, N);
+ return N;
+ }
+
+ // Otherwise create a new SDValue and remember it.
+ SDValue Val = getValueImpl(V);
+ NodeMap[V] = Val;
+ resolveDanglingDebugInfo(V, Val);
+ return Val;
+}
+
+/// getNonRegisterValue - Return an SDValue for the given Value, but
+/// don't look in FuncInfo.ValueMap for a virtual register.
+SDValue SelectionDAGBuilder::getNonRegisterValue(const Value *V) {
+ // If we already have an SDValue for this value, use it.
SDValue &N = NodeMap[V];
if (N.getNode()) return N;
- if (Constant *C = const_cast<Constant*>(dyn_cast<Constant>(V))) {
+ // Otherwise create a new SDValue and remember it.
+ SDValue Val = getValueImpl(V);
+ NodeMap[V] = Val;
+ resolveDanglingDebugInfo(V, Val);
+ return Val;
+}
+
+/// getValueImpl - Helper function for getValue and getNonRegisterValue.
+/// Create an SDValue for the given value.
+SDValue SelectionDAGBuilder::getValueImpl(const Value *V) {
+ if (const Constant *C = dyn_cast<Constant>(V)) {
EVT VT = TLI.getValueType(V->getType(), true);
- if (ConstantInt *CI = dyn_cast<ConstantInt>(C))
- return N = DAG.getConstant(*CI, VT);
+ if (const ConstantInt *CI = dyn_cast<ConstantInt>(C))
+ return DAG.getConstant(*CI, VT);
- if (GlobalValue *GV = dyn_cast<GlobalValue>(C))
- return N = DAG.getGlobalAddress(GV, VT);
+ if (const GlobalValue *GV = dyn_cast<GlobalValue>(C))
+ return DAG.getGlobalAddress(GV, getCurDebugLoc(), VT);
if (isa<ConstantPointerNull>(C))
- return N = DAG.getConstant(0, TLI.getPointerTy());
+ return DAG.getConstant(0, TLI.getPointerTy());
- if (ConstantFP *CFP = dyn_cast<ConstantFP>(C))
- return N = DAG.getConstantFP(*CFP, VT);
+ if (const ConstantFP *CFP = dyn_cast<ConstantFP>(C))
+ return DAG.getConstantFP(*CFP, VT);
if (isa<UndefValue>(C) && !V->getType()->isAggregateType())
- return N = DAG.getUNDEF(VT);
+ return DAG.getUNDEF(VT);
- if (ConstantExpr *CE = dyn_cast<ConstantExpr>(C)) {
+ if (const ConstantExpr *CE = dyn_cast<ConstantExpr>(C)) {
visit(CE->getOpcode(), *CE);
SDValue N1 = NodeMap[V];
- assert(N1.getNode() && "visit didn't populate the ValueMap!");
+ assert(N1.getNode() && "visit didn't populate the NodeMap!");
return N1;
}
getCurDebugLoc());
}
- if (BlockAddress *BA = dyn_cast<BlockAddress>(C))
+ if (const BlockAddress *BA = dyn_cast<BlockAddress>(C))
return DAG.getBlockAddress(BA, VT);
const VectorType *VecTy = cast<VectorType>(V->getType());
// Now that we know the number and type of the elements, get that number of
// elements into the Ops array based on what kind of constant it is.
SmallVector<SDValue, 16> Ops;
- if (ConstantVector *CP = dyn_cast<ConstantVector>(C)) {
+ if (const ConstantVector *CP = dyn_cast<ConstantVector>(C)) {
for (unsigned i = 0; i != NumElements; ++i)
Ops.push_back(getValue(CP->getOperand(i)));
} else {
return DAG.getFrameIndex(SI->second, TLI.getPointerTy());
}
- unsigned InReg = FuncInfo.ValueMap[V];
- assert(InReg && "Value not in map!");
-
- RegsForValue RFV(*DAG.getContext(), TLI, InReg, V->getType());
- SDValue Chain = DAG.getEntryNode();
- return RFV.getCopyFromRegs(DAG, getCurDebugLoc(), Chain, NULL);
-}
-
-/// Get the EVTs and ArgFlags collections that represent the legalized return
-/// type of the given function. This does not require a DAG or a return value,
-/// and is suitable for use before any DAGs for the function are constructed.
-static void getReturnInfo(const Type* ReturnType,
- Attributes attr, SmallVectorImpl<EVT> &OutVTs,
- SmallVectorImpl<ISD::ArgFlagsTy> &OutFlags,
- TargetLowering &TLI,
- SmallVectorImpl<uint64_t> *Offsets = 0) {
- SmallVector<EVT, 4> ValueVTs;
- ComputeValueVTs(TLI, ReturnType, ValueVTs);
- unsigned NumValues = ValueVTs.size();
- if (NumValues == 0) return;
- unsigned Offset = 0;
-
- for (unsigned j = 0, f = NumValues; j != f; ++j) {
- EVT VT = ValueVTs[j];
- ISD::NodeType ExtendKind = ISD::ANY_EXTEND;
-
- if (attr & Attribute::SExt)
- ExtendKind = ISD::SIGN_EXTEND;
- else if (attr & Attribute::ZExt)
- ExtendKind = ISD::ZERO_EXTEND;
-
- // FIXME: C calling convention requires the return type to be promoted to
- // at least 32-bit. But this is not necessary for non-C calling
- // conventions. The frontend should mark functions whose return values
- // require promoting with signext or zeroext attributes.
- if (ExtendKind != ISD::ANY_EXTEND && VT.isInteger()) {
- EVT MinVT = TLI.getRegisterType(ReturnType->getContext(), MVT::i32);
- if (VT.bitsLT(MinVT))
- VT = MinVT;
- }
-
- unsigned NumParts = TLI.getNumRegisters(ReturnType->getContext(), VT);
- EVT PartVT = TLI.getRegisterType(ReturnType->getContext(), VT);
- unsigned PartSize = TLI.getTargetData()->getTypeAllocSize(
- PartVT.getTypeForEVT(ReturnType->getContext()));
-
- // 'inreg' on function refers to return value
- ISD::ArgFlagsTy Flags = ISD::ArgFlagsTy();
- if (attr & Attribute::InReg)
- Flags.setInReg();
-
- // Propagate extension type if any
- if (attr & Attribute::SExt)
- Flags.setSExt();
- else if (attr & Attribute::ZExt)
- Flags.setZExt();
-
- for (unsigned i = 0; i < NumParts; ++i) {
- OutVTs.push_back(PartVT);
- OutFlags.push_back(Flags);
- if (Offsets)
- {
- Offsets->push_back(Offset);
- Offset += PartSize;
- }
- }
+ // If this is an instruction which fast-isel has deferred, select it now.
+ if (const Instruction *Inst = dyn_cast<Instruction>(V)) {
+ unsigned InReg = FuncInfo.InitializeRegForValue(Inst);
+ RegsForValue RFV(*DAG.getContext(), TLI, InReg, Inst->getType());
+ SDValue Chain = DAG.getEntryNode();
+ return RFV.getCopyFromRegs(DAG, FuncInfo, getCurDebugLoc(), Chain, NULL);
}
+
+ llvm_unreachable("Can't get register for value!");
+ return SDValue();
}
void SelectionDAGBuilder::visitRet(const ReturnInst &I) {
SDValue Chain = getControlRoot();
SmallVector<ISD::OutputArg, 8> Outs;
- FunctionLoweringInfo &FLI = DAG.getFunctionLoweringInfo();
+ SmallVector<SDValue, 8> OutVals;
- if (!FLI.CanLowerReturn) {
- unsigned DemoteReg = FLI.DemoteRegister;
+ if (!FuncInfo.CanLowerReturn) {
+ unsigned DemoteReg = FuncInfo.DemoteRegister;
const Function *F = I.getParent()->getParent();
// Emit a store of the return value through the virtual register.
unsigned NumValues = ValueVTs.size();
SmallVector<SDValue, 4> Chains(NumValues);
- EVT PtrVT = PtrValueVTs[0];
for (unsigned i = 0; i != NumValues; ++i) {
- SDValue Add = DAG.getNode(ISD::ADD, getCurDebugLoc(), PtrVT, RetPtr,
- DAG.getConstant(Offsets[i], PtrVT));
+ SDValue Add = DAG.getNode(ISD::ADD, getCurDebugLoc(),
+ RetPtr.getValueType(), RetPtr,
+ DAG.getIntPtrConstant(Offsets[i]));
Chains[i] =
DAG.getStore(Chain, getCurDebugLoc(),
SDValue(RetOp.getNode(), RetOp.getResNo() + i),
- Add, NULL, Offsets[i], false, false, 0);
+ // FIXME: better loc info would be nice.
+ Add, MachinePointerInfo(), false, false, 0);
}
Chain = DAG.getNode(ISD::TokenFactor, getCurDebugLoc(),
else if (F->paramHasAttr(0, Attribute::ZExt))
Flags.setZExt();
- for (unsigned i = 0; i < NumParts; ++i)
- Outs.push_back(ISD::OutputArg(Flags, Parts[i], /*isfixed=*/true));
+ for (unsigned i = 0; i < NumParts; ++i) {
+ Outs.push_back(ISD::OutputArg(Flags, Parts[i].getValueType(),
+ /*isfixed=*/true));
+ OutVals.push_back(Parts[i]);
+ }
}
}
}
CallingConv::ID CallConv =
DAG.getMachineFunction().getFunction()->getCallingConv();
Chain = TLI.LowerReturn(Chain, CallConv, isVarArg,
- Outs, getCurDebugLoc(), DAG);
+ Outs, OutVals, getCurDebugLoc(), DAG);
// Verify that the target's LowerReturn behaved as expected.
assert(Chain.getNode() && Chain.getValueType() == MVT::Other &&
/// created for it, emit nodes to copy the value into the virtual
/// registers.
void SelectionDAGBuilder::CopyToExportRegsIfNeeded(const Value *V) {
- if (!V->use_empty()) {
- DenseMap<const Value *, unsigned>::iterator VMI = FuncInfo.ValueMap.find(V);
- if (VMI != FuncInfo.ValueMap.end())
- CopyValueToVirtualRegister(V, VMI->second);
+ DenseMap<const Value *, unsigned>::iterator VMI = FuncInfo.ValueMap.find(V);
+ if (VMI != FuncInfo.ValueMap.end()) {
+ assert(!V->use_empty() && "Unused value assigned virtual registers!");
+ CopyValueToVirtualRegister(V, VMI->second);
}
}
SelectionDAGBuilder::EmitBranchForMergedCondition(const Value *Cond,
MachineBasicBlock *TBB,
MachineBasicBlock *FBB,
- MachineBasicBlock *CurBB) {
+ MachineBasicBlock *CurBB,
+ MachineBasicBlock *SwitchBB) {
const BasicBlock *BB = CurBB->getBasicBlock();
// If the leaf of the tree is a comparison, merge the condition into
// The operands of the cmp have to be in this block. We don't know
// how to export them from some other block. If this is the first block
// of the sequence, no exporting is needed.
- if (CurBB == CurMBB ||
+ if (CurBB == SwitchBB ||
(isExportableFromCurrentBlock(BOp->getOperand(0), BB) &&
isExportableFromCurrentBlock(BOp->getOperand(1), BB))) {
ISD::CondCode Condition;
MachineBasicBlock *TBB,
MachineBasicBlock *FBB,
MachineBasicBlock *CurBB,
+ MachineBasicBlock *SwitchBB,
unsigned Opc) {
// If this node is not part of the or/and tree, emit it as a branch.
const Instruction *BOp = dyn_cast<Instruction>(Cond);
BOp->getParent() != CurBB->getBasicBlock() ||
!InBlock(BOp->getOperand(0), CurBB->getBasicBlock()) ||
!InBlock(BOp->getOperand(1), CurBB->getBasicBlock())) {
- EmitBranchForMergedCondition(Cond, TBB, FBB, CurBB);
+ EmitBranchForMergedCondition(Cond, TBB, FBB, CurBB, SwitchBB);
return;
}
//
// Emit the LHS condition.
- FindMergedConditions(BOp->getOperand(0), TBB, TmpBB, CurBB, Opc);
+ FindMergedConditions(BOp->getOperand(0), TBB, TmpBB, CurBB, SwitchBB, Opc);
// Emit the RHS condition into TmpBB.
- FindMergedConditions(BOp->getOperand(1), TBB, FBB, TmpBB, Opc);
+ FindMergedConditions(BOp->getOperand(1), TBB, FBB, TmpBB, SwitchBB, Opc);
} else {
assert(Opc == Instruction::And && "Unknown merge op!");
// Codegen X & Y as:
// This requires creation of TmpBB after CurBB.
// Emit the LHS condition.
- FindMergedConditions(BOp->getOperand(0), TmpBB, FBB, CurBB, Opc);
+ FindMergedConditions(BOp->getOperand(0), TmpBB, FBB, CurBB, SwitchBB, Opc);
// Emit the RHS condition into TmpBB.
- FindMergedConditions(BOp->getOperand(1), TBB, FBB, TmpBB, Opc);
+ FindMergedConditions(BOp->getOperand(1), TBB, FBB, TmpBB, SwitchBB, Opc);
}
}
if (Cases[0].CC == ISD::SETNE && Cases[0].FalseBB == Cases[1].ThisBB)
return false;
}
-
+
return true;
}
void SelectionDAGBuilder::visitBr(const BranchInst &I) {
+ MachineBasicBlock *BrMBB = FuncInfo.MBB;
+
// Update machine-CFG edges.
MachineBasicBlock *Succ0MBB = FuncInfo.MBBMap[I.getSuccessor(0)];
// Figure out which block is immediately after the current one.
MachineBasicBlock *NextBlock = 0;
- MachineFunction::iterator BBI = CurMBB;
+ MachineFunction::iterator BBI = BrMBB;
if (++BBI != FuncInfo.MF->end())
NextBlock = BBI;
if (I.isUnconditional()) {
// Update machine-CFG edges.
- CurMBB->addSuccessor(Succ0MBB);
+ BrMBB->addSuccessor(Succ0MBB);
// If this is not a fall-through branch, emit the branch.
if (Succ0MBB != NextBlock)
// If this is a series of conditions that are or'd or and'd together, emit
// this as a sequence of branches instead of setcc's with and/or operations.
+ // As long as jumps are not expensive, this should improve performance.
// For example, instead of something like:
// cmp A, B
// C = seteq
// jle foo
//
if (const BinaryOperator *BOp = dyn_cast<BinaryOperator>(CondVal)) {
- if (BOp->hasOneUse() &&
+ if (!TLI.isJumpExpensive() &&
+ BOp->hasOneUse() &&
(BOp->getOpcode() == Instruction::And ||
BOp->getOpcode() == Instruction::Or)) {
- FindMergedConditions(BOp, Succ0MBB, Succ1MBB, CurMBB, BOp->getOpcode());
+ FindMergedConditions(BOp, Succ0MBB, Succ1MBB, BrMBB, BrMBB,
+ BOp->getOpcode());
// If the compares in later blocks need to use values not currently
// exported from this block, export them now. This block should always
// be the first entry.
- assert(SwitchCases[0].ThisBB == CurMBB && "Unexpected lowering!");
+ assert(SwitchCases[0].ThisBB == BrMBB && "Unexpected lowering!");
// Allow some cases to be rejected.
if (ShouldEmitAsBranches(SwitchCases)) {
}
// Emit the branch for this block.
- visitSwitchCase(SwitchCases[0]);
+ visitSwitchCase(SwitchCases[0], BrMBB);
SwitchCases.erase(SwitchCases.begin());
return;
}
// Create a CaseBlock record representing this branch.
CaseBlock CB(ISD::SETEQ, CondVal, ConstantInt::getTrue(*DAG.getContext()),
- NULL, Succ0MBB, Succ1MBB, CurMBB);
+ NULL, Succ0MBB, Succ1MBB, BrMBB);
// Use visitSwitchCase to actually insert the fast branch sequence for this
// cond branch.
- visitSwitchCase(CB);
+ visitSwitchCase(CB, BrMBB);
}
/// visitSwitchCase - Emits the necessary code to represent a single node in
/// the binary search tree resulting from lowering a switch instruction.
-void SelectionDAGBuilder::visitSwitchCase(CaseBlock &CB) {
+void SelectionDAGBuilder::visitSwitchCase(CaseBlock &CB,
+ MachineBasicBlock *SwitchBB) {
SDValue Cond;
SDValue CondLHS = getValue(CB.CmpLHS);
DebugLoc dl = getCurDebugLoc();
}
// Update successor info
- CurMBB->addSuccessor(CB.TrueBB);
- CurMBB->addSuccessor(CB.FalseBB);
+ SwitchBB->addSuccessor(CB.TrueBB);
+ SwitchBB->addSuccessor(CB.FalseBB);
// Set NextBlock to be the MBB immediately after the current one, if any.
// This is used to avoid emitting unnecessary branches to the next block.
MachineBasicBlock *NextBlock = 0;
- MachineFunction::iterator BBI = CurMBB;
+ MachineFunction::iterator BBI = SwitchBB;
if (++BBI != FuncInfo.MF->end())
NextBlock = BBI;
MVT::Other, getControlRoot(), Cond,
DAG.getBasicBlock(CB.TrueBB));
- // If the branch was constant folded, fix up the CFG.
- if (BrCond.getOpcode() == ISD::BR) {
- CurMBB->removeSuccessor(CB.FalseBB);
- } else {
- // Otherwise, go ahead and insert the false branch.
- if (BrCond == getControlRoot())
- CurMBB->removeSuccessor(CB.TrueBB);
-
- if (CB.FalseBB != NextBlock)
- BrCond = DAG.getNode(ISD::BR, dl, MVT::Other, BrCond,
- DAG.getBasicBlock(CB.FalseBB));
- }
+ // Insert the false branch. Do this even if it's a fall through branch,
+ // this makes it easier to do DAG optimizations which require inverting
+ // the branch condition.
+ BrCond = DAG.getNode(ISD::BR, dl, MVT::Other, BrCond,
+ DAG.getBasicBlock(CB.FalseBB));
DAG.setRoot(BrCond);
}
/// visitJumpTableHeader - This function emits necessary code to produce index
/// in the JumpTable from switch case.
void SelectionDAGBuilder::visitJumpTableHeader(JumpTable &JT,
- JumpTableHeader &JTH) {
+ JumpTableHeader &JTH,
+ MachineBasicBlock *SwitchBB) {
// Subtract the lowest switch case value from the value being switched on and
// conditional branch to default mbb if the result is greater than the
// difference between smallest and largest cases.
// therefore require extension or truncating.
SwitchOp = DAG.getZExtOrTrunc(Sub, getCurDebugLoc(), TLI.getPointerTy());
- unsigned JumpTableReg = FuncInfo.MakeReg(TLI.getPointerTy());
+ unsigned JumpTableReg = FuncInfo.CreateReg(TLI.getPointerTy());
SDValue CopyTo = DAG.getCopyToReg(getControlRoot(), getCurDebugLoc(),
JumpTableReg, SwitchOp);
JT.Reg = JumpTableReg;
// Set NextBlock to be the MBB immediately after the current one, if any.
// This is used to avoid emitting unnecessary branches to the next block.
MachineBasicBlock *NextBlock = 0;
- MachineFunction::iterator BBI = CurMBB;
+ MachineFunction::iterator BBI = SwitchBB;
if (++BBI != FuncInfo.MF->end())
NextBlock = BBI;
/// visitBitTestHeader - This function emits necessary code to produce value
/// suitable for "bit tests"
-void SelectionDAGBuilder::visitBitTestHeader(BitTestBlock &B) {
+void SelectionDAGBuilder::visitBitTestHeader(BitTestBlock &B,
+ MachineBasicBlock *SwitchBB) {
// Subtract the minimum value
SDValue SwitchOp = getValue(B.SValue);
EVT VT = SwitchOp.getValueType();
Sub, DAG.getConstant(B.Range, VT),
ISD::SETUGT);
- SDValue ShiftOp = DAG.getZExtOrTrunc(Sub, getCurDebugLoc(),
- TLI.getPointerTy());
+ // Determine the type of the test operands.
+ bool UsePtrType = false;
+ if (!TLI.isTypeLegal(VT))
+ UsePtrType = true;
+ else {
+ for (unsigned i = 0, e = B.Cases.size(); i != e; ++i)
+ if ((uint64_t)((int64_t)B.Cases[i].Mask >> VT.getSizeInBits()) + 1 >= 2) {
+ // Switch table case range are encoded into series of masks.
+ // Just use pointer type, it's guaranteed to fit.
+ UsePtrType = true;
+ break;
+ }
+ }
+ if (UsePtrType) {
+ VT = TLI.getPointerTy();
+ Sub = DAG.getZExtOrTrunc(Sub, getCurDebugLoc(), VT);
+ }
- B.Reg = FuncInfo.MakeReg(TLI.getPointerTy());
+ B.RegVT = VT;
+ B.Reg = FuncInfo.CreateReg(VT);
SDValue CopyTo = DAG.getCopyToReg(getControlRoot(), getCurDebugLoc(),
- B.Reg, ShiftOp);
+ B.Reg, Sub);
// Set NextBlock to be the MBB immediately after the current one, if any.
// This is used to avoid emitting unnecessary branches to the next block.
MachineBasicBlock *NextBlock = 0;
- MachineFunction::iterator BBI = CurMBB;
+ MachineFunction::iterator BBI = SwitchBB;
if (++BBI != FuncInfo.MF->end())
NextBlock = BBI;
MachineBasicBlock* MBB = B.Cases[0].ThisBB;
- CurMBB->addSuccessor(B.Default);
- CurMBB->addSuccessor(MBB);
+ SwitchBB->addSuccessor(B.Default);
+ SwitchBB->addSuccessor(MBB);
SDValue BrRange = DAG.getNode(ISD::BRCOND, getCurDebugLoc(),
MVT::Other, CopyTo, RangeCmp,
}
/// visitBitTestCase - this function produces one "bit test"
-void SelectionDAGBuilder::visitBitTestCase(MachineBasicBlock* NextMBB,
+void SelectionDAGBuilder::visitBitTestCase(BitTestBlock &BB,
+ MachineBasicBlock* NextMBB,
unsigned Reg,
- BitTestCase &B) {
- // Make desired shift
- SDValue ShiftOp = DAG.getCopyFromReg(getControlRoot(), getCurDebugLoc(), Reg,
- TLI.getPointerTy());
- SDValue SwitchVal = DAG.getNode(ISD::SHL, getCurDebugLoc(),
- TLI.getPointerTy(),
- DAG.getConstant(1, TLI.getPointerTy()),
- ShiftOp);
-
- // Emit bit tests and jumps
- SDValue AndOp = DAG.getNode(ISD::AND, getCurDebugLoc(),
- TLI.getPointerTy(), SwitchVal,
- DAG.getConstant(B.Mask, TLI.getPointerTy()));
- SDValue AndCmp = DAG.getSetCC(getCurDebugLoc(),
- TLI.getSetCCResultType(AndOp.getValueType()),
- AndOp, DAG.getConstant(0, TLI.getPointerTy()),
- ISD::SETNE);
-
- CurMBB->addSuccessor(B.TargetBB);
- CurMBB->addSuccessor(NextMBB);
+ BitTestCase &B,
+ MachineBasicBlock *SwitchBB) {
+ EVT VT = BB.RegVT;
+ SDValue ShiftOp = DAG.getCopyFromReg(getControlRoot(), getCurDebugLoc(),
+ Reg, VT);
+ SDValue Cmp;
+ if (CountPopulation_64(B.Mask) == 1) {
+ // Testing for a single bit; just compare the shift count with what it
+ // would need to be to shift a 1 bit in that position.
+ Cmp = DAG.getSetCC(getCurDebugLoc(),
+ TLI.getSetCCResultType(VT),
+ ShiftOp,
+ DAG.getConstant(CountTrailingZeros_64(B.Mask), VT),
+ ISD::SETEQ);
+ } else {
+ // Make desired shift
+ SDValue SwitchVal = DAG.getNode(ISD::SHL, getCurDebugLoc(), VT,
+ DAG.getConstant(1, VT), ShiftOp);
+
+ // Emit bit tests and jumps
+ SDValue AndOp = DAG.getNode(ISD::AND, getCurDebugLoc(),
+ VT, SwitchVal, DAG.getConstant(B.Mask, VT));
+ Cmp = DAG.getSetCC(getCurDebugLoc(),
+ TLI.getSetCCResultType(VT),
+ AndOp, DAG.getConstant(0, VT),
+ ISD::SETNE);
+ }
+
+ SwitchBB->addSuccessor(B.TargetBB);
+ SwitchBB->addSuccessor(NextMBB);
SDValue BrAnd = DAG.getNode(ISD::BRCOND, getCurDebugLoc(),
MVT::Other, getControlRoot(),
- AndCmp, DAG.getBasicBlock(B.TargetBB));
+ Cmp, DAG.getBasicBlock(B.TargetBB));
// Set NextBlock to be the MBB immediately after the current one, if any.
// This is used to avoid emitting unnecessary branches to the next block.
MachineBasicBlock *NextBlock = 0;
- MachineFunction::iterator BBI = CurMBB;
+ MachineFunction::iterator BBI = SwitchBB;
if (++BBI != FuncInfo.MF->end())
NextBlock = BBI;
}
void SelectionDAGBuilder::visitInvoke(const InvokeInst &I) {
+ MachineBasicBlock *InvokeMBB = FuncInfo.MBB;
+
// Retrieve successors.
MachineBasicBlock *Return = FuncInfo.MBBMap[I.getSuccessor(0)];
MachineBasicBlock *LandingPad = FuncInfo.MBBMap[I.getSuccessor(1)];
CopyToExportRegsIfNeeded(&I);
// Update successor info
- CurMBB->addSuccessor(Return);
- CurMBB->addSuccessor(LandingPad);
+ InvokeMBB->addSuccessor(Return);
+ InvokeMBB->addSuccessor(LandingPad);
// Drop into normal successor.
DAG.setRoot(DAG.getNode(ISD::BR, getCurDebugLoc(),
bool SelectionDAGBuilder::handleSmallSwitchRange(CaseRec& CR,
CaseRecVector& WorkList,
const Value* SV,
- MachineBasicBlock* Default) {
+ MachineBasicBlock *Default,
+ MachineBasicBlock *SwitchBB) {
Case& BackCase = *(CR.Range.second-1);
// Size is the number of Cases represented by this range.
if (++BBI != FuncInfo.MF->end())
NextBlock = BBI;
- // TODO: If any two of the cases has the same destination, and if one value
+ // If any two of the cases has the same destination, and if one value
// is the same as the other, but has one bit unset that the other has set,
// use bit manipulation to do two compares at once. For example:
// "if (X == 6 || X == 4)" -> "if ((X|2) == 6)"
+ // TODO: This could be extended to merge any 2 cases in switches with 3 cases.
+ // TODO: Handle cases where CR.CaseBB != SwitchBB.
+ if (Size == 2 && CR.CaseBB == SwitchBB) {
+ Case &Small = *CR.Range.first;
+ Case &Big = *(CR.Range.second-1);
+
+ if (Small.Low == Small.High && Big.Low == Big.High && Small.BB == Big.BB) {
+ const APInt& SmallValue = cast<ConstantInt>(Small.Low)->getValue();
+ const APInt& BigValue = cast<ConstantInt>(Big.Low)->getValue();
+
+ // Check that there is only one bit different.
+ if (BigValue.countPopulation() == SmallValue.countPopulation() + 1 &&
+ (SmallValue | BigValue) == BigValue) {
+ // Isolate the common bit.
+ APInt CommonBit = BigValue & ~SmallValue;
+ assert((SmallValue | CommonBit) == BigValue &&
+ CommonBit.countPopulation() == 1 && "Not a common bit?");
+
+ SDValue CondLHS = getValue(SV);
+ EVT VT = CondLHS.getValueType();
+ DebugLoc DL = getCurDebugLoc();
+
+ SDValue Or = DAG.getNode(ISD::OR, DL, VT, CondLHS,
+ DAG.getConstant(CommonBit, VT));
+ SDValue Cond = DAG.getSetCC(DL, MVT::i1,
+ Or, DAG.getConstant(BigValue, VT),
+ ISD::SETEQ);
+
+ // Update successor info.
+ SwitchBB->addSuccessor(Small.BB);
+ SwitchBB->addSuccessor(Default);
+
+ // Insert the true branch.
+ SDValue BrCond = DAG.getNode(ISD::BRCOND, DL, MVT::Other,
+ getControlRoot(), Cond,
+ DAG.getBasicBlock(Small.BB));
+
+ // Insert the false branch.
+ BrCond = DAG.getNode(ISD::BR, DL, MVT::Other, BrCond,
+ DAG.getBasicBlock(Default));
+
+ DAG.setRoot(BrCond);
+ return true;
+ }
+ }
+ }
// Rearrange the case blocks so that the last one falls through if possible.
if (NextBlock && Default != NextBlock && BackCase.BB != NextBlock) {
// code into the current block. Otherwise, push the CaseBlock onto the
// vector to be later processed by SDISel, and insert the node's MBB
// before the next MBB.
- if (CurBlock == CurMBB)
- visitSwitchCase(CB);
+ if (CurBlock == SwitchBB)
+ visitSwitchCase(CB, SwitchBB);
else
SwitchCases.push_back(CB);
}
static APInt ComputeRange(const APInt &First, const APInt &Last) {
- APInt LastExt(Last), FirstExt(First);
uint32_t BitWidth = std::max(Last.getBitWidth(), First.getBitWidth()) + 1;
- LastExt.sext(BitWidth); FirstExt.sext(BitWidth);
+ APInt LastExt = Last.sext(BitWidth), FirstExt = First.sext(BitWidth);
return (LastExt - FirstExt + 1ULL);
}
bool SelectionDAGBuilder::handleJTSwitchCase(CaseRec& CR,
CaseRecVector& WorkList,
const Value* SV,
- MachineBasicBlock* Default) {
+ MachineBasicBlock* Default,
+ MachineBasicBlock *SwitchBB) {
Case& FrontCase = *CR.Range.first;
Case& BackCase = *(CR.Range.second-1);
// Set the jump table information so that we can codegen it as a second
// MachineBasicBlock
JumpTable JT(-1U, JTI, JumpTableBB, Default);
- JumpTableHeader JTH(First, Last, SV, CR.CaseBB, (CR.CaseBB == CurMBB));
- if (CR.CaseBB == CurMBB)
- visitJumpTableHeader(JT, JTH);
+ JumpTableHeader JTH(First, Last, SV, CR.CaseBB, (CR.CaseBB == SwitchBB));
+ if (CR.CaseBB == SwitchBB)
+ visitJumpTableHeader(JT, JTH, SwitchBB);
JTCases.push_back(JumpTableBlock(JTH, JT));
bool SelectionDAGBuilder::handleBTSplitSwitchCase(CaseRec& CR,
CaseRecVector& WorkList,
const Value* SV,
- MachineBasicBlock* Default) {
+ MachineBasicBlock *Default,
+ MachineBasicBlock *SwitchBB) {
// Get the MachineFunction which holds the current MBB. This is used when
// inserting any additional MBBs necessary to represent the switch.
MachineFunction *CurMF = FuncInfo.MF;
// Otherwise, branch to LHS.
CaseBlock CB(ISD::SETLT, SV, C, NULL, TrueBB, FalseBB, CR.CaseBB);
- if (CR.CaseBB == CurMBB)
- visitSwitchCase(CB);
+ if (CR.CaseBB == SwitchBB)
+ visitSwitchCase(CB, SwitchBB);
else
SwitchCases.push_back(CB);
bool SelectionDAGBuilder::handleBitTestsSwitchCase(CaseRec& CR,
CaseRecVector& WorkList,
const Value* SV,
- MachineBasicBlock* Default){
+ MachineBasicBlock* Default,
+ MachineBasicBlock *SwitchBB){
EVT PTy = TLI.getPointerTy();
unsigned IntPtrBits = PTy.getSizeInBits();
}
BitTestBlock BTB(lowBound, cmpRange, SV,
- -1U, (CR.CaseBB == CurMBB),
+ -1U, MVT::Other, (CR.CaseBB == SwitchBB),
CR.CaseBB, Default, BTC);
- if (CR.CaseBB == CurMBB)
- visitBitTestHeader(BTB);
+ if (CR.CaseBB == SwitchBB)
+ visitBitTestHeader(BTB, SwitchBB);
BitTestCases.push_back(BTB);
if (Cases.size() >= 2)
// Must recompute end() each iteration because it may be
// invalidated by erase if we hold on to it
- for (CaseItr I = Cases.begin(), J = ++(Cases.begin()); J != Cases.end(); ) {
+ for (CaseItr I = Cases.begin(), J = llvm::next(Cases.begin());
+ J != Cases.end(); ) {
const APInt& nextValue = cast<ConstantInt>(J->Low)->getValue();
const APInt& currentValue = cast<ConstantInt>(I->High)->getValue();
MachineBasicBlock* nextBB = J->BB;
return numCmps;
}
+void SelectionDAGBuilder::UpdateSplitBlock(MachineBasicBlock *First,
+ MachineBasicBlock *Last) {
+ // Update JTCases.
+ for (unsigned i = 0, e = JTCases.size(); i != e; ++i)
+ if (JTCases[i].first.HeaderBB == First)
+ JTCases[i].first.HeaderBB = Last;
+
+ // Update BitTestCases.
+ for (unsigned i = 0, e = BitTestCases.size(); i != e; ++i)
+ if (BitTestCases[i].Parent == First)
+ BitTestCases[i].Parent = Last;
+}
+
void SelectionDAGBuilder::visitSwitch(const SwitchInst &SI) {
+ MachineBasicBlock *SwitchMBB = FuncInfo.MBB;
+
// Figure out which block is immediately after the current one.
MachineBasicBlock *NextBlock = 0;
MachineBasicBlock *Default = FuncInfo.MBBMap[SI.getDefaultDest()];
// Update machine-CFG edges.
// If this is not a fall-through branch, emit the branch.
- CurMBB->addSuccessor(Default);
+ SwitchMBB->addSuccessor(Default);
if (Default != NextBlock)
DAG.setRoot(DAG.getNode(ISD::BR, getCurDebugLoc(),
MVT::Other, getControlRoot(),
// Push the initial CaseRec onto the worklist
CaseRecVector WorkList;
- WorkList.push_back(CaseRec(CurMBB,0,0,CaseRange(Cases.begin(),Cases.end())));
+ WorkList.push_back(CaseRec(SwitchMBB,0,0,
+ CaseRange(Cases.begin(),Cases.end())));
while (!WorkList.empty()) {
// Grab a record representing a case range to process off the worklist
CaseRec CR = WorkList.back();
WorkList.pop_back();
- if (handleBitTestsSwitchCase(CR, WorkList, SV, Default))
+ if (handleBitTestsSwitchCase(CR, WorkList, SV, Default, SwitchMBB))
continue;
// If the range has few cases (two or less) emit a series of specific
// tests.
- if (handleSmallSwitchRange(CR, WorkList, SV, Default))
+ if (handleSmallSwitchRange(CR, WorkList, SV, Default, SwitchMBB))
continue;
// If the switch has more than 5 blocks, and at least 40% dense, and the
// target supports indirect branches, then emit a jump table rather than
// lowering the switch to a binary tree of conditional branches.
- if (handleJTSwitchCase(CR, WorkList, SV, Default))
+ if (handleJTSwitchCase(CR, WorkList, SV, Default, SwitchMBB))
continue;
// Emit binary tree. We need to pick a pivot, and push left and right ranges
// onto the worklist. Leafs are handled via handleSmallSwitchRange() call.
- handleBTSplitSwitchCase(CR, WorkList, SV, Default);
+ handleBTSplitSwitchCase(CR, WorkList, SV, Default, SwitchMBB);
}
}
void SelectionDAGBuilder::visitIndirectBr(const IndirectBrInst &I) {
+ MachineBasicBlock *IndirectBrMBB = FuncInfo.MBB;
+
// Update machine-CFG edges with unique successors.
SmallVector<BasicBlock*, 32> succs;
succs.reserve(I.getNumSuccessors());
array_pod_sort(succs.begin(), succs.end());
succs.erase(std::unique(succs.begin(), succs.end()), succs.end());
for (unsigned i = 0, e = succs.size(); i != e; ++i)
- CurMBB->addSuccessor(FuncInfo.MBBMap[succs[i]]);
+ IndirectBrMBB->addSuccessor(FuncInfo.MBBMap[succs[i]]);
DAG.setRoot(DAG.getNode(ISD::BRIND, getCurDebugLoc(),
MVT::Other, getControlRoot(),
void SelectionDAGBuilder::visitFSub(const User &I) {
// -0.0 - X --> fneg
const Type *Ty = I.getType();
- if (Ty->isVectorTy()) {
- if (ConstantVector *CV = dyn_cast<ConstantVector>(I.getOperand(0))) {
- const VectorType *DestTy = cast<VectorType>(I.getType());
- const Type *ElTy = DestTy->getElementType();
- unsigned VL = DestTy->getNumElements();
- std::vector<Constant*> NZ(VL, ConstantFP::getNegativeZero(ElTy));
- Constant *CNZ = ConstantVector::get(&NZ[0], NZ.size());
- if (CV == CNZ) {
- SDValue Op2 = getValue(I.getOperand(1));
- setValue(&I, DAG.getNode(ISD::FNEG, getCurDebugLoc(),
- Op2.getValueType(), Op2));
- return;
- }
- }
+ if (isa<Constant>(I.getOperand(0)) &&
+ I.getOperand(0) == ConstantFP::getZeroValueForNegation(Ty)) {
+ SDValue Op2 = getValue(I.getOperand(1));
+ setValue(&I, DAG.getNode(ISD::FNEG, getCurDebugLoc(),
+ Op2.getValueType(), Op2));
+ return;
}
- if (ConstantFP *CFP = dyn_cast<ConstantFP>(I.getOperand(0)))
- if (CFP->isExactlyValue(ConstantFP::getNegativeZero(Ty)->getValueAPF())) {
- SDValue Op2 = getValue(I.getOperand(1));
- setValue(&I, DAG.getNode(ISD::FNEG, getCurDebugLoc(),
- Op2.getValueType(), Op2));
- return;
- }
-
visitBinary(I, ISD::FSUB);
}
void SelectionDAGBuilder::visitShift(const User &I, unsigned Opcode) {
SDValue Op1 = getValue(I.getOperand(0));
SDValue Op2 = getValue(I.getOperand(1));
- if (!I.getType()->isVectorTy() &&
- Op2.getValueType() != TLI.getShiftAmountTy()) {
+
+ MVT ShiftTy = TLI.getShiftAmountTy();
+
+ // Coerce the shift amount to the right type if we can.
+ if (!I.getType()->isVectorTy() && Op2.getValueType() != ShiftTy) {
+ unsigned ShiftSize = ShiftTy.getSizeInBits();
+ unsigned Op2Size = Op2.getValueType().getSizeInBits();
+ DebugLoc DL = getCurDebugLoc();
+
// If the operand is smaller than the shift count type, promote it.
- EVT PTy = TLI.getPointerTy();
- EVT STy = TLI.getShiftAmountTy();
- if (STy.bitsGT(Op2.getValueType()))
- Op2 = DAG.getNode(ISD::ANY_EXTEND, getCurDebugLoc(),
- TLI.getShiftAmountTy(), Op2);
+ if (ShiftSize > Op2Size)
+ Op2 = DAG.getNode(ISD::ZERO_EXTEND, DL, ShiftTy, Op2);
+
// If the operand is larger than the shift count type but the shift
// count type has enough bits to represent any shift value, truncate
// it now. This is a common case and it exposes the truncate to
// optimization early.
- else if (STy.getSizeInBits() >=
- Log2_32_Ceil(Op2.getValueType().getSizeInBits()))
- Op2 = DAG.getNode(ISD::TRUNCATE, getCurDebugLoc(),
- TLI.getShiftAmountTy(), Op2);
- // Otherwise we'll need to temporarily settle for some other
- // convenient type; type legalization will make adjustments as
- // needed.
- else if (PTy.bitsLT(Op2.getValueType()))
- Op2 = DAG.getNode(ISD::TRUNCATE, getCurDebugLoc(),
- TLI.getPointerTy(), Op2);
- else if (PTy.bitsGT(Op2.getValueType()))
- Op2 = DAG.getNode(ISD::ANY_EXTEND, getCurDebugLoc(),
- TLI.getPointerTy(), Op2);
+ else if (ShiftSize >= Log2_32_Ceil(Op2.getValueType().getSizeInBits()))
+ Op2 = DAG.getNode(ISD::TRUNCATE, DL, ShiftTy, Op2);
+ // Otherwise we'll need to temporarily settle for some other convenient
+ // type. Type legalization will make adjustments once the shiftee is split.
+ else
+ Op2 = DAG.getZExtOrTrunc(Op2, DL, MVT::i32);
}
setValue(&I, DAG.getNode(Opcode, getCurDebugLoc(),
// What to do depends on the size of the integer and the size of the pointer.
// We can either truncate, zero extend, or no-op, accordingly.
SDValue N = getValue(I.getOperand(0));
- EVT SrcVT = N.getValueType();
EVT DestVT = TLI.getValueType(I.getType());
setValue(&I, DAG.getZExtOrTrunc(N, getCurDebugLoc(), DestVT));
}
// What to do depends on the size of the integer and the size of the pointer.
// We can either truncate, zero extend, or no-op, accordingly.
SDValue N = getValue(I.getOperand(0));
- EVT SrcVT = N.getValueType();
EVT DestVT = TLI.getValueType(I.getType());
setValue(&I, DAG.getZExtOrTrunc(N, getCurDebugLoc(), DestVT));
}
EVT DestVT = TLI.getValueType(I.getType());
// BitCast assures us that source and destination are the same size so this is
- // either a BIT_CONVERT or a no-op.
+ // either a BITCAST or a no-op.
if (DestVT != N.getValueType())
- setValue(&I, DAG.getNode(ISD::BIT_CONVERT, getCurDebugLoc(),
+ setValue(&I, DAG.getNode(ISD::BITCAST, getCurDebugLoc(),
DestVT, N)); // convert types.
else
setValue(&I, N); // noop cast.
} else {
StartIdx[Input] = (MinRange[Input]/MaskNumElts)*MaskNumElts;
if (MaxRange[Input] - StartIdx[Input] < (int)MaskNumElts &&
- StartIdx[Input] + MaskNumElts < SrcNumElts)
+ StartIdx[Input] + MaskNumElts <= SrcNumElts)
RangeUse[Input] = 1; // Extract from a multiple of the mask length.
}
}
bool IntoUndef = isa<UndefValue>(Op0);
bool FromUndef = isa<UndefValue>(Op1);
- unsigned LinearIndex = ComputeLinearIndex(TLI, AggTy,
- I.idx_begin(), I.idx_end());
+ unsigned LinearIndex = ComputeLinearIndex(AggTy, I.idx_begin(), I.idx_end());
SmallVector<EVT, 4> AggValueVTs;
ComputeValueVTs(TLI, AggTy, AggValueVTs);
const Type *ValTy = I.getType();
bool OutOfUndef = isa<UndefValue>(Op0);
- unsigned LinearIndex = ComputeLinearIndex(TLI, AggTy,
- I.idx_begin(), I.idx_end());
+ unsigned LinearIndex = ComputeLinearIndex(AggTy, I.idx_begin(), I.idx_end());
SmallVector<EVT, 4> ValValueVTs;
ComputeValueVTs(TLI, ValTy, ValValueVTs);
}
Ty = StTy->getElementType(Field);
- } else if (const UnionType *UnTy = dyn_cast<UnionType>(Ty)) {
- unsigned Field = cast<ConstantInt>(Idx)->getZExtValue();
-
- // Offset canonically 0 for unions, but type changes
- Ty = UnTy->getElementType(Field);
} else {
Ty = cast<SequentialType>(Ty)->getElementType();
// If this is a constant subscript, handle it quickly.
if (const ConstantInt *CI = dyn_cast<ConstantInt>(Idx)) {
- if (CI->getZExtValue() == 0) continue;
+ if (CI->isZero()) continue;
uint64_t Offs =
TD->getTypeAllocSize(Ty)*cast<ConstantInt>(CI)->getSExtValue();
SDValue OffsVal;
SDValue AllocSize = getValue(I.getArraySize());
- AllocSize = DAG.getNode(ISD::MUL, getCurDebugLoc(), AllocSize.getValueType(),
- AllocSize,
- DAG.getConstant(TySize, AllocSize.getValueType()));
-
EVT IntPtr = TLI.getPointerTy();
- AllocSize = DAG.getZExtOrTrunc(AllocSize, getCurDebugLoc(), IntPtr);
+ if (AllocSize.getValueType() != IntPtr)
+ AllocSize = DAG.getZExtOrTrunc(AllocSize, getCurDebugLoc(), IntPtr);
+
+ AllocSize = DAG.getNode(ISD::MUL, getCurDebugLoc(), IntPtr,
+ AllocSize,
+ DAG.getConstant(TySize, IntPtr));
// Handle alignment. If the requested alignment is less than or equal to
// the stack alignment, ignore it. If the size is greater than or equal to
// the stack alignment, we note this in the DYNAMIC_STACKALLOC node.
- unsigned StackAlign =
- TLI.getTargetMachine().getFrameInfo()->getStackAlignment();
+ unsigned StackAlign = TM.getFrameLowering()->getStackAlignment();
if (Align <= StackAlign)
Align = 0;
// Inform the Frame Information that we have just allocated a variable-sized
// object.
- FuncInfo.MF->getFrameInfo()->CreateVariableSizedObject();
+ FuncInfo.MF->getFrameInfo()->CreateVariableSizedObject(Align ? Align : 1);
}
void SelectionDAGBuilder::visitLoad(const LoadInst &I) {
bool isVolatile = I.isVolatile();
bool isNonTemporal = I.getMetadata("nontemporal") != 0;
unsigned Alignment = I.getAlignment();
+ const MDNode *TBAAInfo = I.getMetadata(LLVMContext::MD_tbaa);
SmallVector<EVT, 4> ValueVTs;
SmallVector<uint64_t, 4> Offsets;
SDValue Root;
bool ConstantMemory = false;
- if (I.isVolatile())
+ if (I.isVolatile() || NumValues > MaxParallelChains)
// Serialize volatile loads with other side effects.
Root = getRoot();
- else if (AA->pointsToConstantMemory(SV)) {
+ else if (AA->pointsToConstantMemory(
+ AliasAnalysis::Location(SV, AA->getTypeStoreSize(Ty), TBAAInfo))) {
// Do not serialize (non-volatile) loads of constant memory with anything.
Root = DAG.getEntryNode();
ConstantMemory = true;
}
SmallVector<SDValue, 4> Values(NumValues);
- SmallVector<SDValue, 4> Chains(NumValues);
+ SmallVector<SDValue, 4> Chains(std::min(unsigned(MaxParallelChains),
+ NumValues));
EVT PtrVT = Ptr.getValueType();
- for (unsigned i = 0; i != NumValues; ++i) {
+ unsigned ChainI = 0;
+ for (unsigned i = 0; i != NumValues; ++i, ++ChainI) {
+ // Serializing loads here may result in excessive register pressure, and
+ // TokenFactor places arbitrary choke points on the scheduler. SD scheduling
+ // could recover a bit by hoisting nodes upward in the chain by recognizing
+ // they are side-effect free or do not alias. The optimizer should really
+ // avoid this case by converting large object/array copies to llvm.memcpy
+ // (MaxParallelChains should always remain as failsafe).
+ if (ChainI == MaxParallelChains) {
+ assert(PendingLoads.empty() && "PendingLoads must be serialized first");
+ SDValue Chain = DAG.getNode(ISD::TokenFactor, getCurDebugLoc(),
+ MVT::Other, &Chains[0], ChainI);
+ Root = Chain;
+ ChainI = 0;
+ }
SDValue A = DAG.getNode(ISD::ADD, getCurDebugLoc(),
PtrVT, Ptr,
DAG.getConstant(Offsets[i], PtrVT));
SDValue L = DAG.getLoad(ValueVTs[i], getCurDebugLoc(), Root,
- A, SV, Offsets[i], isVolatile,
- isNonTemporal, Alignment);
+ A, MachinePointerInfo(SV, Offsets[i]), isVolatile,
+ isNonTemporal, Alignment, TBAAInfo);
Values[i] = L;
- Chains[i] = L.getValue(1);
+ Chains[ChainI] = L.getValue(1);
}
if (!ConstantMemory) {
SDValue Chain = DAG.getNode(ISD::TokenFactor, getCurDebugLoc(),
- MVT::Other, &Chains[0], NumValues);
+ MVT::Other, &Chains[0], ChainI);
if (isVolatile)
DAG.setRoot(Chain);
else
SDValue Ptr = getValue(PtrV);
SDValue Root = getRoot();
- SmallVector<SDValue, 4> Chains(NumValues);
+ SmallVector<SDValue, 4> Chains(std::min(unsigned(MaxParallelChains),
+ NumValues));
EVT PtrVT = Ptr.getValueType();
bool isVolatile = I.isVolatile();
bool isNonTemporal = I.getMetadata("nontemporal") != 0;
unsigned Alignment = I.getAlignment();
-
- for (unsigned i = 0; i != NumValues; ++i) {
+ const MDNode *TBAAInfo = I.getMetadata(LLVMContext::MD_tbaa);
+
+ unsigned ChainI = 0;
+ for (unsigned i = 0; i != NumValues; ++i, ++ChainI) {
+ // See visitLoad comments.
+ if (ChainI == MaxParallelChains) {
+ SDValue Chain = DAG.getNode(ISD::TokenFactor, getCurDebugLoc(),
+ MVT::Other, &Chains[0], ChainI);
+ Root = Chain;
+ ChainI = 0;
+ }
SDValue Add = DAG.getNode(ISD::ADD, getCurDebugLoc(), PtrVT, Ptr,
DAG.getConstant(Offsets[i], PtrVT));
- Chains[i] = DAG.getStore(Root, getCurDebugLoc(),
- SDValue(Src.getNode(), Src.getResNo() + i),
- Add, PtrV, Offsets[i], isVolatile,
- isNonTemporal, Alignment);
- }
-
- DAG.setRoot(DAG.getNode(ISD::TokenFactor, getCurDebugLoc(),
- MVT::Other, &Chains[0], NumValues));
+ SDValue St = DAG.getStore(Root, getCurDebugLoc(),
+ SDValue(Src.getNode(), Src.getResNo() + i),
+ Add, MachinePointerInfo(PtrV, Offsets[i]),
+ isVolatile, isNonTemporal, Alignment, TBAAInfo);
+ Chains[ChainI] = St;
+ }
+
+ SDValue StoreNode = DAG.getNode(ISD::TokenFactor, getCurDebugLoc(),
+ MVT::Other, &Chains[0], ChainI);
+ ++SDNodeOrder;
+ AssignOrderingToNode(StoreNode.getNode());
+ DAG.setRoot(StoreNode);
}
/// visitTargetIntrinsic - Lower a call of a target intrinsic to an INTRINSIC
bool IsTgtIntrinsic = TLI.getTgtMemIntrinsic(Info, I, Intrinsic);
// Add the intrinsic ID as an integer operand if it's not a target intrinsic.
- if (!IsTgtIntrinsic)
+ if (!IsTgtIntrinsic || Info.opc == ISD::INTRINSIC_VOID ||
+ Info.opc == ISD::INTRINSIC_W_CHAIN)
Ops.push_back(DAG.getConstant(Intrinsic, TLI.getPointerTy()));
// Add all operands of the call to the operand list.
- for (unsigned i = 1, e = I.getNumOperands(); i != e; ++i) {
- SDValue Op = getValue(I.getOperand(i));
+ for (unsigned i = 0, e = I.getNumArgOperands(); i != e; ++i) {
+ SDValue Op = getValue(I.getArgOperand(i));
assert(TLI.isTypeLegal(Op.getValueType()) &&
"Intrinsic uses a non-legal type?");
Ops.push_back(Op);
// This is target intrinsic that touches memory
Result = DAG.getMemIntrinsicNode(Info.opc, getCurDebugLoc(),
VTs, &Ops[0], Ops.size(),
- Info.memVT, Info.ptrVal, Info.offset,
+ Info.memVT,
+ MachinePointerInfo(Info.ptrVal, Info.offset),
Info.align, Info.vol,
Info.readMem, Info.writeMem);
} else if (!HasChain) {
if (!I.getType()->isVoidTy()) {
if (const VectorType *PTy = dyn_cast<VectorType>(I.getType())) {
EVT VT = TLI.getValueType(PTy);
- Result = DAG.getNode(ISD::BIT_CONVERT, getCurDebugLoc(), VT, Result);
+ Result = DAG.getNode(ISD::BITCAST, getCurDebugLoc(), VT, Result);
}
setValue(&I, Result);
DAG.getConstant(0x007fffff, MVT::i32));
SDValue t2 = DAG.getNode(ISD::OR, dl, MVT::i32, t1,
DAG.getConstant(0x3f800000, MVT::i32));
- return DAG.getNode(ISD::BIT_CONVERT, dl, MVT::f32, t2);
+ return DAG.getNode(ISD::BITCAST, dl, MVT::f32, t2);
}
/// GetExponent - Get the exponent:
SDValue Root = getRoot();
SDValue L =
DAG.getAtomic(Op, getCurDebugLoc(),
- getValue(I.getOperand(2)).getValueType().getSimpleVT(),
+ getValue(I.getArgOperand(1)).getValueType().getSimpleVT(),
Root,
- getValue(I.getOperand(1)),
- getValue(I.getOperand(2)),
- I.getOperand(1));
+ getValue(I.getArgOperand(0)),
+ getValue(I.getArgOperand(1)),
+ I.getArgOperand(0));
setValue(&I, L);
DAG.setRoot(L.getValue(1));
return 0;
// implVisitAluOverflow - Lower arithmetic overflow instrinsics.
const char *
SelectionDAGBuilder::implVisitAluOverflow(const CallInst &I, ISD::NodeType Op) {
- SDValue Op1 = getValue(I.getOperand(1));
- SDValue Op2 = getValue(I.getOperand(2));
+ SDValue Op1 = getValue(I.getArgOperand(0));
+ SDValue Op2 = getValue(I.getArgOperand(1));
SDVTList VTs = DAG.getVTList(Op1.getValueType(), MVT::i1);
setValue(&I, DAG.getNode(Op, getCurDebugLoc(), VTs, Op1, Op2));
SDValue result;
DebugLoc dl = getCurDebugLoc();
- if (getValue(I.getOperand(1)).getValueType() == MVT::f32 &&
+ if (getValue(I.getArgOperand(0)).getValueType() == MVT::f32 &&
LimitFloatPrecision > 0 && LimitFloatPrecision <= 18) {
- SDValue Op = getValue(I.getOperand(1));
+ SDValue Op = getValue(I.getArgOperand(0));
// Put the exponent in the right bit position for later addition to the
// final result:
SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X);
SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4,
getF32Constant(DAG, 0x3f7f5e7e));
- SDValue TwoToFracPartOfX = DAG.getNode(ISD::BIT_CONVERT, dl,MVT::i32, t5);
+ SDValue TwoToFracPartOfX = DAG.getNode(ISD::BITCAST, dl,MVT::i32, t5);
// Add the exponent into the result in integer domain.
SDValue t6 = DAG.getNode(ISD::ADD, dl, MVT::i32,
TwoToFracPartOfX, IntegerPartOfX);
- result = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::f32, t6);
+ result = DAG.getNode(ISD::BITCAST, dl, MVT::f32, t6);
} else if (LimitFloatPrecision > 6 && LimitFloatPrecision <= 12) {
// For floating-point precision of 12:
//
SDValue t6 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t5, X);
SDValue t7 = DAG.getNode(ISD::FADD, dl, MVT::f32, t6,
getF32Constant(DAG, 0x3f7ff8fd));
- SDValue TwoToFracPartOfX = DAG.getNode(ISD::BIT_CONVERT, dl,MVT::i32, t7);
+ SDValue TwoToFracPartOfX = DAG.getNode(ISD::BITCAST, dl,MVT::i32, t7);
// Add the exponent into the result in integer domain.
SDValue t8 = DAG.getNode(ISD::ADD, dl, MVT::i32,
TwoToFracPartOfX, IntegerPartOfX);
- result = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::f32, t8);
+ result = DAG.getNode(ISD::BITCAST, dl, MVT::f32, t8);
} else { // LimitFloatPrecision > 12 && LimitFloatPrecision <= 18
// For floating-point precision of 18:
//
SDValue t12 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t11, X);
SDValue t13 = DAG.getNode(ISD::FADD, dl, MVT::f32, t12,
getF32Constant(DAG, 0x3f800000));
- SDValue TwoToFracPartOfX = DAG.getNode(ISD::BIT_CONVERT, dl,
+ SDValue TwoToFracPartOfX = DAG.getNode(ISD::BITCAST, dl,
MVT::i32, t13);
// Add the exponent into the result in integer domain.
SDValue t14 = DAG.getNode(ISD::ADD, dl, MVT::i32,
TwoToFracPartOfX, IntegerPartOfX);
- result = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::f32, t14);
+ result = DAG.getNode(ISD::BITCAST, dl, MVT::f32, t14);
}
} else {
// No special expansion.
result = DAG.getNode(ISD::FEXP, dl,
- getValue(I.getOperand(1)).getValueType(),
- getValue(I.getOperand(1)));
+ getValue(I.getArgOperand(0)).getValueType(),
+ getValue(I.getArgOperand(0)));
}
setValue(&I, result);
SDValue result;
DebugLoc dl = getCurDebugLoc();
- if (getValue(I.getOperand(1)).getValueType() == MVT::f32 &&
+ if (getValue(I.getArgOperand(0)).getValueType() == MVT::f32 &&
LimitFloatPrecision > 0 && LimitFloatPrecision <= 18) {
- SDValue Op = getValue(I.getOperand(1));
- SDValue Op1 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i32, Op);
+ SDValue Op = getValue(I.getArgOperand(0));
+ SDValue Op1 = DAG.getNode(ISD::BITCAST, dl, MVT::i32, Op);
// Scale the exponent by log(2) [0.69314718f].
SDValue Exp = GetExponent(DAG, Op1, TLI, dl);
} else {
// No special expansion.
result = DAG.getNode(ISD::FLOG, dl,
- getValue(I.getOperand(1)).getValueType(),
- getValue(I.getOperand(1)));
+ getValue(I.getArgOperand(0)).getValueType(),
+ getValue(I.getArgOperand(0)));
}
setValue(&I, result);
SDValue result;
DebugLoc dl = getCurDebugLoc();
- if (getValue(I.getOperand(1)).getValueType() == MVT::f32 &&
+ if (getValue(I.getArgOperand(0)).getValueType() == MVT::f32 &&
LimitFloatPrecision > 0 && LimitFloatPrecision <= 18) {
- SDValue Op = getValue(I.getOperand(1));
- SDValue Op1 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i32, Op);
+ SDValue Op = getValue(I.getArgOperand(0));
+ SDValue Op1 = DAG.getNode(ISD::BITCAST, dl, MVT::i32, Op);
// Get the exponent.
SDValue LogOfExponent = GetExponent(DAG, Op1, TLI, dl);
} else {
// No special expansion.
result = DAG.getNode(ISD::FLOG2, dl,
- getValue(I.getOperand(1)).getValueType(),
- getValue(I.getOperand(1)));
+ getValue(I.getArgOperand(0)).getValueType(),
+ getValue(I.getArgOperand(0)));
}
setValue(&I, result);
SDValue result;
DebugLoc dl = getCurDebugLoc();
- if (getValue(I.getOperand(1)).getValueType() == MVT::f32 &&
+ if (getValue(I.getArgOperand(0)).getValueType() == MVT::f32 &&
LimitFloatPrecision > 0 && LimitFloatPrecision <= 18) {
- SDValue Op = getValue(I.getOperand(1));
- SDValue Op1 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i32, Op);
+ SDValue Op = getValue(I.getArgOperand(0));
+ SDValue Op1 = DAG.getNode(ISD::BITCAST, dl, MVT::i32, Op);
// Scale the exponent by log10(2) [0.30102999f].
SDValue Exp = GetExponent(DAG, Op1, TLI, dl);
} else {
// No special expansion.
result = DAG.getNode(ISD::FLOG10, dl,
- getValue(I.getOperand(1)).getValueType(),
- getValue(I.getOperand(1)));
+ getValue(I.getArgOperand(0)).getValueType(),
+ getValue(I.getArgOperand(0)));
}
setValue(&I, result);
SDValue result;
DebugLoc dl = getCurDebugLoc();
- if (getValue(I.getOperand(1)).getValueType() == MVT::f32 &&
+ if (getValue(I.getArgOperand(0)).getValueType() == MVT::f32 &&
LimitFloatPrecision > 0 && LimitFloatPrecision <= 18) {
- SDValue Op = getValue(I.getOperand(1));
+ SDValue Op = getValue(I.getArgOperand(0));
SDValue IntegerPartOfX = DAG.getNode(ISD::FP_TO_SINT, dl, MVT::i32, Op);
SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X);
SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4,
getF32Constant(DAG, 0x3f7f5e7e));
- SDValue t6 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i32, t5);
+ SDValue t6 = DAG.getNode(ISD::BITCAST, dl, MVT::i32, t5);
SDValue TwoToFractionalPartOfX =
DAG.getNode(ISD::ADD, dl, MVT::i32, t6, IntegerPartOfX);
- result = DAG.getNode(ISD::BIT_CONVERT, dl,
+ result = DAG.getNode(ISD::BITCAST, dl,
MVT::f32, TwoToFractionalPartOfX);
} else if (LimitFloatPrecision > 6 && LimitFloatPrecision <= 12) {
// For floating-point precision of 12:
SDValue t6 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t5, X);
SDValue t7 = DAG.getNode(ISD::FADD, dl, MVT::f32, t6,
getF32Constant(DAG, 0x3f7ff8fd));
- SDValue t8 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i32, t7);
+ SDValue t8 = DAG.getNode(ISD::BITCAST, dl, MVT::i32, t7);
SDValue TwoToFractionalPartOfX =
DAG.getNode(ISD::ADD, dl, MVT::i32, t8, IntegerPartOfX);
- result = DAG.getNode(ISD::BIT_CONVERT, dl,
+ result = DAG.getNode(ISD::BITCAST, dl,
MVT::f32, TwoToFractionalPartOfX);
} else { // LimitFloatPrecision > 12 && LimitFloatPrecision <= 18
// For floating-point precision of 18:
SDValue t12 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t11, X);
SDValue t13 = DAG.getNode(ISD::FADD, dl, MVT::f32, t12,
getF32Constant(DAG, 0x3f800000));
- SDValue t14 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i32, t13);
+ SDValue t14 = DAG.getNode(ISD::BITCAST, dl, MVT::i32, t13);
SDValue TwoToFractionalPartOfX =
DAG.getNode(ISD::ADD, dl, MVT::i32, t14, IntegerPartOfX);
- result = DAG.getNode(ISD::BIT_CONVERT, dl,
+ result = DAG.getNode(ISD::BITCAST, dl,
MVT::f32, TwoToFractionalPartOfX);
}
} else {
// No special expansion.
result = DAG.getNode(ISD::FEXP2, dl,
- getValue(I.getOperand(1)).getValueType(),
- getValue(I.getOperand(1)));
+ getValue(I.getArgOperand(0)).getValueType(),
+ getValue(I.getArgOperand(0)));
}
setValue(&I, result);
void
SelectionDAGBuilder::visitPow(const CallInst &I) {
SDValue result;
- const Value *Val = I.getOperand(1);
+ const Value *Val = I.getArgOperand(0);
DebugLoc dl = getCurDebugLoc();
bool IsExp10 = false;
if (getValue(Val).getValueType() == MVT::f32 &&
- getValue(I.getOperand(2)).getValueType() == MVT::f32 &&
+ getValue(I.getArgOperand(1)).getValueType() == MVT::f32 &&
LimitFloatPrecision > 0 && LimitFloatPrecision <= 18) {
if (Constant *C = const_cast<Constant*>(dyn_cast<Constant>(Val))) {
if (ConstantFP *CFP = dyn_cast<ConstantFP>(C)) {
}
if (IsExp10 && LimitFloatPrecision > 0 && LimitFloatPrecision <= 18) {
- SDValue Op = getValue(I.getOperand(2));
+ SDValue Op = getValue(I.getArgOperand(1));
// Put the exponent in the right bit position for later addition to the
// final result:
SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X);
SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4,
getF32Constant(DAG, 0x3f7f5e7e));
- SDValue t6 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i32, t5);
+ SDValue t6 = DAG.getNode(ISD::BITCAST, dl, MVT::i32, t5);
SDValue TwoToFractionalPartOfX =
DAG.getNode(ISD::ADD, dl, MVT::i32, t6, IntegerPartOfX);
- result = DAG.getNode(ISD::BIT_CONVERT, dl,
+ result = DAG.getNode(ISD::BITCAST, dl,
MVT::f32, TwoToFractionalPartOfX);
} else if (LimitFloatPrecision > 6 && LimitFloatPrecision <= 12) {
// For floating-point precision of 12:
SDValue t6 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t5, X);
SDValue t7 = DAG.getNode(ISD::FADD, dl, MVT::f32, t6,
getF32Constant(DAG, 0x3f7ff8fd));
- SDValue t8 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i32, t7);
+ SDValue t8 = DAG.getNode(ISD::BITCAST, dl, MVT::i32, t7);
SDValue TwoToFractionalPartOfX =
DAG.getNode(ISD::ADD, dl, MVT::i32, t8, IntegerPartOfX);
- result = DAG.getNode(ISD::BIT_CONVERT, dl,
+ result = DAG.getNode(ISD::BITCAST, dl,
MVT::f32, TwoToFractionalPartOfX);
} else { // LimitFloatPrecision > 12 && LimitFloatPrecision <= 18
// For floating-point precision of 18:
SDValue t12 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t11, X);
SDValue t13 = DAG.getNode(ISD::FADD, dl, MVT::f32, t12,
getF32Constant(DAG, 0x3f800000));
- SDValue t14 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i32, t13);
+ SDValue t14 = DAG.getNode(ISD::BITCAST, dl, MVT::i32, t13);
SDValue TwoToFractionalPartOfX =
DAG.getNode(ISD::ADD, dl, MVT::i32, t14, IntegerPartOfX);
- result = DAG.getNode(ISD::BIT_CONVERT, dl,
+ result = DAG.getNode(ISD::BITCAST, dl,
MVT::f32, TwoToFractionalPartOfX);
}
} else {
// No special expansion.
result = DAG.getNode(ISD::FPOW, dl,
- getValue(I.getOperand(1)).getValueType(),
- getValue(I.getOperand(1)),
- getValue(I.getOperand(2)));
+ getValue(I.getArgOperand(0)).getValueType(),
+ getValue(I.getArgOperand(0)),
+ getValue(I.getArgOperand(1)));
}
setValue(&I, result);
return DAG.getNode(ISD::FPOWI, DL, LHS.getValueType(), LHS, RHS);
}
+/// EmitFuncArgumentDbgValue - If the DbgValueInst is a dbg_value of a function
+/// argument, create the corresponding DBG_VALUE machine instruction for it now.
+/// At the end of instruction selection, they will be inserted to the entry BB.
+bool
+SelectionDAGBuilder::EmitFuncArgumentDbgValue(const Value *V, MDNode *Variable,
+ int64_t Offset,
+ const SDValue &N) {
+ const Argument *Arg = dyn_cast<Argument>(V);
+ if (!Arg)
+ return false;
+
+ MachineFunction &MF = DAG.getMachineFunction();
+ const TargetInstrInfo *TII = DAG.getTarget().getInstrInfo();
+ const TargetRegisterInfo *TRI = DAG.getTarget().getRegisterInfo();
+
+ // Ignore inlined function arguments here.
+ DIVariable DV(Variable);
+ if (DV.isInlinedFnArgument(MF.getFunction()))
+ return false;
+
+ MachineBasicBlock *MBB = FuncInfo.MBB;
+ if (MBB != &MF.front())
+ return false;
+
+ unsigned Reg = 0;
+ if (Arg->hasByValAttr()) {
+ // Byval arguments' frame index is recorded during argument lowering.
+ // Use this info directly.
+ Reg = TRI->getFrameRegister(MF);
+ Offset = FuncInfo.getByValArgumentFrameIndex(Arg);
+ // If byval argument ofset is not recorded then ignore this.
+ if (!Offset)
+ Reg = 0;
+ }
+
+ if (N.getNode() && N.getOpcode() == ISD::CopyFromReg) {
+ Reg = cast<RegisterSDNode>(N.getOperand(1))->getReg();
+ if (TargetRegisterInfo::isVirtualRegister(Reg)) {
+ MachineRegisterInfo &RegInfo = MF.getRegInfo();
+ unsigned PR = RegInfo.getLiveInPhysReg(Reg);
+ if (PR)
+ Reg = PR;
+ }
+ }
+
+ if (!Reg) {
+ // Check if ValueMap has reg number.
+ DenseMap<const Value *, unsigned>::iterator VMI = FuncInfo.ValueMap.find(V);
+ if (VMI != FuncInfo.ValueMap.end())
+ Reg = VMI->second;
+ }
+
+ if (!Reg && N.getNode()) {
+ // Check if frame index is available.
+ if (LoadSDNode *LNode = dyn_cast<LoadSDNode>(N.getNode()))
+ if (FrameIndexSDNode *FINode =
+ dyn_cast<FrameIndexSDNode>(LNode->getBasePtr().getNode())) {
+ Reg = TRI->getFrameRegister(MF);
+ Offset = FINode->getIndex();
+ }
+ }
+
+ if (!Reg)
+ return false;
+
+ MachineInstrBuilder MIB = BuildMI(MF, getCurDebugLoc(),
+ TII->get(TargetOpcode::DBG_VALUE))
+ .addReg(Reg, RegState::Debug).addImm(Offset).addMetadata(Variable);
+ FuncInfo.ArgDbgValues.push_back(&*MIB);
+ return true;
+}
+
+// VisualStudio defines setjmp as _setjmp
+#if defined(_MSC_VER) && defined(setjmp) && \
+ !defined(setjmp_undefined_for_msvc)
+# pragma push_macro("setjmp")
+# undef setjmp
+# define setjmp_undefined_for_msvc
+#endif
/// visitIntrinsicCall - Lower the call to the specified intrinsic function. If
/// we want to emit this as a call to a named external function, return the name
case Intrinsic::vacopy: visitVACopy(I); return 0;
case Intrinsic::returnaddress:
setValue(&I, DAG.getNode(ISD::RETURNADDR, dl, TLI.getPointerTy(),
- getValue(I.getOperand(1))));
+ getValue(I.getArgOperand(0))));
return 0;
case Intrinsic::frameaddress:
setValue(&I, DAG.getNode(ISD::FRAMEADDR, dl, TLI.getPointerTy(),
- getValue(I.getOperand(1))));
+ getValue(I.getArgOperand(0))));
return 0;
case Intrinsic::setjmp:
return "_setjmp"+!TLI.usesUnderscoreSetJmp();
case Intrinsic::memcpy: {
// Assert for address < 256 since we support only user defined address
// spaces.
- assert(cast<PointerType>(I.getOperand(1)->getType())->getAddressSpace()
+ assert(cast<PointerType>(I.getArgOperand(0)->getType())->getAddressSpace()
< 256 &&
- cast<PointerType>(I.getOperand(2)->getType())->getAddressSpace()
+ cast<PointerType>(I.getArgOperand(1)->getType())->getAddressSpace()
< 256 &&
"Unknown address space");
- SDValue Op1 = getValue(I.getOperand(1));
- SDValue Op2 = getValue(I.getOperand(2));
- SDValue Op3 = getValue(I.getOperand(3));
- unsigned Align = cast<ConstantInt>(I.getOperand(4))->getZExtValue();
- bool isVol = cast<ConstantInt>(I.getOperand(5))->getZExtValue();
+ SDValue Op1 = getValue(I.getArgOperand(0));
+ SDValue Op2 = getValue(I.getArgOperand(1));
+ SDValue Op3 = getValue(I.getArgOperand(2));
+ unsigned Align = cast<ConstantInt>(I.getArgOperand(3))->getZExtValue();
+ bool isVol = cast<ConstantInt>(I.getArgOperand(4))->getZExtValue();
DAG.setRoot(DAG.getMemcpy(getRoot(), dl, Op1, Op2, Op3, Align, isVol, false,
- I.getOperand(1), 0, I.getOperand(2), 0));
+ MachinePointerInfo(I.getArgOperand(0)),
+ MachinePointerInfo(I.getArgOperand(1))));
return 0;
}
case Intrinsic::memset: {
// Assert for address < 256 since we support only user defined address
// spaces.
- assert(cast<PointerType>(I.getOperand(1)->getType())->getAddressSpace()
+ assert(cast<PointerType>(I.getArgOperand(0)->getType())->getAddressSpace()
< 256 &&
"Unknown address space");
- SDValue Op1 = getValue(I.getOperand(1));
- SDValue Op2 = getValue(I.getOperand(2));
- SDValue Op3 = getValue(I.getOperand(3));
- unsigned Align = cast<ConstantInt>(I.getOperand(4))->getZExtValue();
- bool isVol = cast<ConstantInt>(I.getOperand(5))->getZExtValue();
+ SDValue Op1 = getValue(I.getArgOperand(0));
+ SDValue Op2 = getValue(I.getArgOperand(1));
+ SDValue Op3 = getValue(I.getArgOperand(2));
+ unsigned Align = cast<ConstantInt>(I.getArgOperand(3))->getZExtValue();
+ bool isVol = cast<ConstantInt>(I.getArgOperand(4))->getZExtValue();
DAG.setRoot(DAG.getMemset(getRoot(), dl, Op1, Op2, Op3, Align, isVol,
- I.getOperand(1), 0));
+ MachinePointerInfo(I.getArgOperand(0))));
return 0;
}
case Intrinsic::memmove: {
// Assert for address < 256 since we support only user defined address
// spaces.
- assert(cast<PointerType>(I.getOperand(1)->getType())->getAddressSpace()
+ assert(cast<PointerType>(I.getArgOperand(0)->getType())->getAddressSpace()
< 256 &&
- cast<PointerType>(I.getOperand(2)->getType())->getAddressSpace()
+ cast<PointerType>(I.getArgOperand(1)->getType())->getAddressSpace()
< 256 &&
"Unknown address space");
- SDValue Op1 = getValue(I.getOperand(1));
- SDValue Op2 = getValue(I.getOperand(2));
- SDValue Op3 = getValue(I.getOperand(3));
- unsigned Align = cast<ConstantInt>(I.getOperand(4))->getZExtValue();
- bool isVol = cast<ConstantInt>(I.getOperand(5))->getZExtValue();
-
- // If the source and destination are known to not be aliases, we can
- // lower memmove as memcpy.
- uint64_t Size = -1ULL;
- if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op3))
- Size = C->getZExtValue();
- if (AA->alias(I.getOperand(1), Size, I.getOperand(2), Size) ==
- AliasAnalysis::NoAlias) {
- DAG.setRoot(DAG.getMemcpy(getRoot(), dl, Op1, Op2, Op3, Align, isVol,
- false, I.getOperand(1), 0, I.getOperand(2), 0));
- return 0;
- }
-
+ SDValue Op1 = getValue(I.getArgOperand(0));
+ SDValue Op2 = getValue(I.getArgOperand(1));
+ SDValue Op3 = getValue(I.getArgOperand(2));
+ unsigned Align = cast<ConstantInt>(I.getArgOperand(3))->getZExtValue();
+ bool isVol = cast<ConstantInt>(I.getArgOperand(4))->getZExtValue();
DAG.setRoot(DAG.getMemmove(getRoot(), dl, Op1, Op2, Op3, Align, isVol,
- I.getOperand(1), 0, I.getOperand(2), 0));
+ MachinePointerInfo(I.getArgOperand(0)),
+ MachinePointerInfo(I.getArgOperand(1))));
return 0;
}
case Intrinsic::dbg_declare: {
- // FIXME: currently, we get here only if OptLevel != CodeGenOpt::None.
- // The real handling of this intrinsic is in FastISel.
- if (OptLevel != CodeGenOpt::None)
- // FIXME: Variable debug info is not supported here.
- return 0;
const DbgDeclareInst &DI = cast<DbgDeclareInst>(I);
- if (!DIDescriptor::ValidDebugInfo(DI.getVariable(), CodeGenOpt::None))
- return 0;
-
MDNode *Variable = DI.getVariable();
const Value *Address = DI.getAddress();
- if (!Address)
- return 0;
- if (const BitCastInst *BCI = dyn_cast<BitCastInst>(Address))
- Address = BCI->getOperand(0);
- const AllocaInst *AI = dyn_cast<AllocaInst>(Address);
- // Don't handle byval struct arguments or VLAs, for example.
- if (!AI)
+ if (!Address || !DIVariable(DI.getVariable()).Verify())
return 0;
- DenseMap<const AllocaInst*, int>::iterator SI =
- FuncInfo.StaticAllocaMap.find(AI);
- if (SI == FuncInfo.StaticAllocaMap.end())
- return 0; // VLAs.
- int FI = SI->second;
- MachineModuleInfo &MMI = DAG.getMachineFunction().getMMI();
- if (!DI.getDebugLoc().isUnknown() && MMI.hasDebugInfo())
- MMI.setVariableDbgInfo(Variable, FI, DI.getDebugLoc());
+ // Build an entry in DbgOrdering. Debug info input nodes get an SDNodeOrder
+ // but do not always have a corresponding SDNode built. The SDNodeOrder
+ // absolute, but not relative, values are different depending on whether
+ // debug info exists.
+ ++SDNodeOrder;
+
+ // Check if address has undef value.
+ if (isa<UndefValue>(Address) ||
+ (Address->use_empty() && !isa<Argument>(Address))) {
+ DEBUG(dbgs() << "Dropping debug info for " << DI);
+ return 0;
+ }
+
+ SDValue &N = NodeMap[Address];
+ if (!N.getNode() && isa<Argument>(Address))
+ // Check unused arguments map.
+ N = UnusedArgNodeMap[Address];
+ SDDbgValue *SDV;
+ if (N.getNode()) {
+ // Parameters are handled specially.
+ bool isParameter =
+ DIVariable(Variable).getTag() == dwarf::DW_TAG_arg_variable;
+ if (const BitCastInst *BCI = dyn_cast<BitCastInst>(Address))
+ Address = BCI->getOperand(0);
+ const AllocaInst *AI = dyn_cast<AllocaInst>(Address);
+
+ if (isParameter && !AI) {
+ FrameIndexSDNode *FINode = dyn_cast<FrameIndexSDNode>(N.getNode());
+ if (FINode)
+ // Byval parameter. We have a frame index at this point.
+ SDV = DAG.getDbgValue(Variable, FINode->getIndex(),
+ 0, dl, SDNodeOrder);
+ else {
+ // Can't do anything with other non-AI cases yet. This might be a
+ // parameter of a callee function that got inlined, for example.
+ DEBUG(dbgs() << "Dropping debug info for " << DI);
+ return 0;
+ }
+ } else if (AI)
+ SDV = DAG.getDbgValue(Variable, N.getNode(), N.getResNo(),
+ 0, dl, SDNodeOrder);
+ else {
+ // Can't do anything with other non-AI cases yet.
+ DEBUG(dbgs() << "Dropping debug info for " << DI);
+ return 0;
+ }
+ DAG.AddDbgValue(SDV, N.getNode(), isParameter);
+ } else {
+ // If Address is an argument then try to emit its dbg value using
+ // virtual register info from the FuncInfo.ValueMap.
+ if (!EmitFuncArgumentDbgValue(Address, Variable, 0, N)) {
+ // If variable is pinned by a alloca in dominating bb then
+ // use StaticAllocaMap.
+ if (const AllocaInst *AI = dyn_cast<AllocaInst>(Address)) {
+ if (AI->getParent() != DI.getParent()) {
+ DenseMap<const AllocaInst*, int>::iterator SI =
+ FuncInfo.StaticAllocaMap.find(AI);
+ if (SI != FuncInfo.StaticAllocaMap.end()) {
+ SDV = DAG.getDbgValue(Variable, SI->second,
+ 0, dl, SDNodeOrder);
+ DAG.AddDbgValue(SDV, 0, false);
+ return 0;
+ }
+ }
+ }
+ DEBUG(dbgs() << "Dropping debug info for " << DI);
+ }
+ }
return 0;
}
case Intrinsic::dbg_value: {
const DbgValueInst &DI = cast<DbgValueInst>(I);
- if (!DIDescriptor::ValidDebugInfo(DI.getVariable(), CodeGenOpt::None))
+ if (!DIVariable(DI.getVariable()).Verify())
return 0;
MDNode *Variable = DI.getVariable();
// absolute, but not relative, values are different depending on whether
// debug info exists.
++SDNodeOrder;
+ SDDbgValue *SDV;
if (isa<ConstantInt>(V) || isa<ConstantFP>(V)) {
- DAG.AddDbgValue(DAG.getDbgValue(Variable, V, Offset, dl, SDNodeOrder));
+ SDV = DAG.getDbgValue(Variable, V, Offset, dl, SDNodeOrder);
+ DAG.AddDbgValue(SDV, 0, false);
} else {
- SDValue &N = NodeMap[V];
- if (N.getNode())
- DAG.AddDbgValue(DAG.getDbgValue(Variable, N.getNode(),
- N.getResNo(), Offset, dl, SDNodeOrder),
- N.getNode());
- else
+ // Do not use getValue() in here; we don't want to generate code at
+ // this point if it hasn't been done yet.
+ SDValue N = NodeMap[V];
+ if (!N.getNode() && isa<Argument>(V))
+ // Check unused arguments map.
+ N = UnusedArgNodeMap[V];
+ if (N.getNode()) {
+ if (!EmitFuncArgumentDbgValue(V, Variable, Offset, N)) {
+ SDV = DAG.getDbgValue(Variable, N.getNode(),
+ N.getResNo(), Offset, dl, SDNodeOrder);
+ DAG.AddDbgValue(SDV, N.getNode(), false);
+ }
+ } else if (!V->use_empty() ) {
+ // Do not call getValue(V) yet, as we don't want to generate code.
+ // Remember it for later.
+ DanglingDebugInfo DDI(&DI, dl, SDNodeOrder);
+ DanglingDebugInfoMap[V] = DDI;
+ } else {
// We may expand this to cover more cases. One case where we have no
- // data available is an unreferenced parameter; we need this fallback.
- DAG.AddDbgValue(DAG.getDbgValue(Variable,
- UndefValue::get(V->getType()),
- Offset, dl, SDNodeOrder));
+ // data available is an unreferenced parameter.
+ DEBUG(dbgs() << "Dropping debug info for " << DI);
+ }
}
// Build a debug info table entry.
if (SI == FuncInfo.StaticAllocaMap.end())
return 0; // VLAs.
int FI = SI->second;
-
+
MachineModuleInfo &MMI = DAG.getMachineFunction().getMMI();
if (!DI.getDebugLoc().isUnknown() && MMI.hasDebugInfo())
MMI.setVariableDbgInfo(Variable, FI, DI.getDebugLoc());
}
case Intrinsic::eh_exception: {
// Insert the EXCEPTIONADDR instruction.
- assert(CurMBB->isLandingPad() &&"Call to eh.exception not in landing pad!");
+ assert(FuncInfo.MBB->isLandingPad() &&
+ "Call to eh.exception not in landing pad!");
SDVTList VTs = DAG.getVTList(TLI.getPointerTy(), MVT::Other);
SDValue Ops[1];
Ops[0] = DAG.getRoot();
}
case Intrinsic::eh_selector: {
+ MachineBasicBlock *CallMBB = FuncInfo.MBB;
MachineModuleInfo &MMI = DAG.getMachineFunction().getMMI();
- if (CurMBB->isLandingPad())
- AddCatchInfo(I, &MMI, CurMBB);
+ if (CallMBB->isLandingPad())
+ AddCatchInfo(I, &MMI, CallMBB);
else {
#ifndef NDEBUG
FuncInfo.CatchInfoLost.insert(&I);
#endif
// FIXME: Mark exception selector register as live in. Hack for PR1508.
unsigned Reg = TLI.getExceptionSelectorRegister();
- if (Reg) CurMBB->addLiveIn(Reg);
+ if (Reg) FuncInfo.MBB->addLiveIn(Reg);
}
// Insert the EHSELECTION instruction.
SDVTList VTs = DAG.getVTList(TLI.getPointerTy(), MVT::Other);
SDValue Ops[2];
- Ops[0] = getValue(I.getOperand(1));
+ Ops[0] = getValue(I.getArgOperand(0));
Ops[1] = getRoot();
SDValue Op = DAG.getNode(ISD::EHSELECTION, dl, VTs, Ops, 2);
DAG.setRoot(Op.getValue(1));
case Intrinsic::eh_typeid_for: {
// Find the type id for the given typeinfo.
- GlobalVariable *GV = ExtractTypeInfo(I.getOperand(1));
+ GlobalVariable *GV = ExtractTypeInfo(I.getArgOperand(0));
unsigned TypeID = DAG.getMachineFunction().getMMI().getTypeIDFor(GV);
Res = DAG.getConstant(TypeID, MVT::i32);
setValue(&I, Res);
DAG.setRoot(DAG.getNode(ISD::EH_RETURN, dl,
MVT::Other,
getControlRoot(),
- getValue(I.getOperand(1)),
- getValue(I.getOperand(2))));
+ getValue(I.getArgOperand(0)),
+ getValue(I.getArgOperand(1))));
return 0;
case Intrinsic::eh_unwind_init:
DAG.getMachineFunction().getMMI().setCallsUnwindInit(true);
return 0;
case Intrinsic::eh_dwarf_cfa: {
- EVT VT = getValue(I.getOperand(1)).getValueType();
- SDValue CfaArg = DAG.getSExtOrTrunc(getValue(I.getOperand(1)), dl,
+ SDValue CfaArg = DAG.getSExtOrTrunc(getValue(I.getArgOperand(0)), dl,
TLI.getPointerTy());
SDValue Offset = DAG.getNode(ISD::ADD, dl,
TLI.getPointerTy(),
}
case Intrinsic::eh_sjlj_callsite: {
MachineModuleInfo &MMI = DAG.getMachineFunction().getMMI();
- ConstantInt *CI = dyn_cast<ConstantInt>(I.getOperand(1));
+ ConstantInt *CI = dyn_cast<ConstantInt>(I.getArgOperand(0));
assert(CI && "Non-constant call site value in eh.sjlj.callsite!");
assert(MMI.getCurrentCallSite() == 0 && "Overlapping call sites!");
MMI.setCurrentCallSite(CI->getZExtValue());
return 0;
}
+ case Intrinsic::eh_sjlj_setjmp: {
+ setValue(&I, DAG.getNode(ISD::EH_SJLJ_SETJMP, dl, MVT::i32, getRoot(),
+ getValue(I.getArgOperand(0))));
+ return 0;
+ }
+ case Intrinsic::eh_sjlj_longjmp: {
+ DAG.setRoot(DAG.getNode(ISD::EH_SJLJ_LONGJMP, dl, MVT::Other,
+ getRoot(), getValue(I.getArgOperand(0))));
+ return 0;
+ }
+ case Intrinsic::eh_sjlj_dispatch_setup: {
+ DAG.setRoot(DAG.getNode(ISD::EH_SJLJ_DISPATCHSETUP, dl, MVT::Other,
+ getRoot(), getValue(I.getArgOperand(0))));
+ return 0;
+ }
+
+ case Intrinsic::x86_mmx_pslli_w:
+ case Intrinsic::x86_mmx_pslli_d:
+ case Intrinsic::x86_mmx_pslli_q:
+ case Intrinsic::x86_mmx_psrli_w:
+ case Intrinsic::x86_mmx_psrli_d:
+ case Intrinsic::x86_mmx_psrli_q:
+ case Intrinsic::x86_mmx_psrai_w:
+ case Intrinsic::x86_mmx_psrai_d: {
+ SDValue ShAmt = getValue(I.getArgOperand(1));
+ if (isa<ConstantSDNode>(ShAmt)) {
+ visitTargetIntrinsic(I, Intrinsic);
+ return 0;
+ }
+ unsigned NewIntrinsic = 0;
+ EVT ShAmtVT = MVT::v2i32;
+ switch (Intrinsic) {
+ case Intrinsic::x86_mmx_pslli_w:
+ NewIntrinsic = Intrinsic::x86_mmx_psll_w;
+ break;
+ case Intrinsic::x86_mmx_pslli_d:
+ NewIntrinsic = Intrinsic::x86_mmx_psll_d;
+ break;
+ case Intrinsic::x86_mmx_pslli_q:
+ NewIntrinsic = Intrinsic::x86_mmx_psll_q;
+ break;
+ case Intrinsic::x86_mmx_psrli_w:
+ NewIntrinsic = Intrinsic::x86_mmx_psrl_w;
+ break;
+ case Intrinsic::x86_mmx_psrli_d:
+ NewIntrinsic = Intrinsic::x86_mmx_psrl_d;
+ break;
+ case Intrinsic::x86_mmx_psrli_q:
+ NewIntrinsic = Intrinsic::x86_mmx_psrl_q;
+ break;
+ case Intrinsic::x86_mmx_psrai_w:
+ NewIntrinsic = Intrinsic::x86_mmx_psra_w;
+ break;
+ case Intrinsic::x86_mmx_psrai_d:
+ NewIntrinsic = Intrinsic::x86_mmx_psra_d;
+ break;
+ default: llvm_unreachable("Impossible intrinsic"); // Can't reach here.
+ }
+ // The vector shift intrinsics with scalars uses 32b shift amounts but
+ // the sse2/mmx shift instructions reads 64 bits. Set the upper 32 bits
+ // to be zero.
+ // We must do this early because v2i32 is not a legal type.
+ DebugLoc dl = getCurDebugLoc();
+ SDValue ShOps[2];
+ ShOps[0] = ShAmt;
+ ShOps[1] = DAG.getConstant(0, MVT::i32);
+ ShAmt = DAG.getNode(ISD::BUILD_VECTOR, dl, ShAmtVT, &ShOps[0], 2);
+ EVT DestVT = TLI.getValueType(I.getType());
+ ShAmt = DAG.getNode(ISD::BITCAST, dl, DestVT, ShAmt);
+ Res = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, DestVT,
+ DAG.getConstant(NewIntrinsic, MVT::i32),
+ getValue(I.getArgOperand(0)), ShAmt);
+ setValue(&I, Res);
+ return 0;
+ }
case Intrinsic::convertff:
case Intrinsic::convertfsi:
case Intrinsic::convertfui:
case Intrinsic::convertuu: Code = ISD::CVT_UU; break;
}
EVT DestVT = TLI.getValueType(I.getType());
- const Value *Op1 = I.getOperand(1);
+ const Value *Op1 = I.getArgOperand(0);
Res = DAG.getConvertRndSat(DestVT, getCurDebugLoc(), getValue(Op1),
DAG.getValueType(DestVT),
DAG.getValueType(getValue(Op1).getValueType()),
- getValue(I.getOperand(2)),
- getValue(I.getOperand(3)),
+ getValue(I.getArgOperand(1)),
+ getValue(I.getArgOperand(2)),
Code);
setValue(&I, Res);
return 0;
}
case Intrinsic::sqrt:
setValue(&I, DAG.getNode(ISD::FSQRT, dl,
- getValue(I.getOperand(1)).getValueType(),
- getValue(I.getOperand(1))));
+ getValue(I.getArgOperand(0)).getValueType(),
+ getValue(I.getArgOperand(0))));
return 0;
case Intrinsic::powi:
- setValue(&I, ExpandPowI(dl, getValue(I.getOperand(1)),
- getValue(I.getOperand(2)), DAG));
+ setValue(&I, ExpandPowI(dl, getValue(I.getArgOperand(0)),
+ getValue(I.getArgOperand(1)), DAG));
return 0;
case Intrinsic::sin:
setValue(&I, DAG.getNode(ISD::FSIN, dl,
- getValue(I.getOperand(1)).getValueType(),
- getValue(I.getOperand(1))));
+ getValue(I.getArgOperand(0)).getValueType(),
+ getValue(I.getArgOperand(0))));
return 0;
case Intrinsic::cos:
setValue(&I, DAG.getNode(ISD::FCOS, dl,
- getValue(I.getOperand(1)).getValueType(),
- getValue(I.getOperand(1))));
+ getValue(I.getArgOperand(0)).getValueType(),
+ getValue(I.getArgOperand(0))));
return 0;
case Intrinsic::log:
visitLog(I);
return 0;
case Intrinsic::convert_to_fp16:
setValue(&I, DAG.getNode(ISD::FP32_TO_FP16, dl,
- MVT::i16, getValue(I.getOperand(1))));
+ MVT::i16, getValue(I.getArgOperand(0))));
return 0;
case Intrinsic::convert_from_fp16:
setValue(&I, DAG.getNode(ISD::FP16_TO_FP32, dl,
- MVT::f32, getValue(I.getOperand(1))));
+ MVT::f32, getValue(I.getArgOperand(0))));
return 0;
case Intrinsic::pcmarker: {
- SDValue Tmp = getValue(I.getOperand(1));
+ SDValue Tmp = getValue(I.getArgOperand(0));
DAG.setRoot(DAG.getNode(ISD::PCMARKER, dl, MVT::Other, getRoot(), Tmp));
return 0;
}
}
case Intrinsic::bswap:
setValue(&I, DAG.getNode(ISD::BSWAP, dl,
- getValue(I.getOperand(1)).getValueType(),
- getValue(I.getOperand(1))));
+ getValue(I.getArgOperand(0)).getValueType(),
+ getValue(I.getArgOperand(0))));
return 0;
case Intrinsic::cttz: {
- SDValue Arg = getValue(I.getOperand(1));
+ SDValue Arg = getValue(I.getArgOperand(0));
EVT Ty = Arg.getValueType();
setValue(&I, DAG.getNode(ISD::CTTZ, dl, Ty, Arg));
return 0;
}
case Intrinsic::ctlz: {
- SDValue Arg = getValue(I.getOperand(1));
+ SDValue Arg = getValue(I.getArgOperand(0));
EVT Ty = Arg.getValueType();
setValue(&I, DAG.getNode(ISD::CTLZ, dl, Ty, Arg));
return 0;
}
case Intrinsic::ctpop: {
- SDValue Arg = getValue(I.getOperand(1));
+ SDValue Arg = getValue(I.getArgOperand(0));
EVT Ty = Arg.getValueType();
setValue(&I, DAG.getNode(ISD::CTPOP, dl, Ty, Arg));
return 0;
return 0;
}
case Intrinsic::stackrestore: {
- Res = getValue(I.getOperand(1));
+ Res = getValue(I.getArgOperand(0));
DAG.setRoot(DAG.getNode(ISD::STACKRESTORE, dl, MVT::Other, getRoot(), Res));
return 0;
}
MachineFrameInfo *MFI = MF.getFrameInfo();
EVT PtrTy = TLI.getPointerTy();
- SDValue Src = getValue(I.getOperand(1)); // The guard's value.
- AllocaInst *Slot = cast<AllocaInst>(I.getOperand(2));
+ SDValue Src = getValue(I.getArgOperand(0)); // The guard's value.
+ AllocaInst *Slot = cast<AllocaInst>(I.getArgOperand(1));
int FI = FuncInfo.StaticAllocaMap[Slot];
MFI->setStackProtectorIndex(FI);
// Store the stack protector onto the stack.
Res = DAG.getStore(getRoot(), getCurDebugLoc(), Src, FIN,
- PseudoSourceValue::getFixedStack(FI),
- 0, true, false, 0);
+ MachinePointerInfo::getFixedStack(FI),
+ true, false, 0);
setValue(&I, Res);
DAG.setRoot(Res);
return 0;
}
case Intrinsic::objectsize: {
// If we don't know by now, we're never going to know.
- ConstantInt *CI = dyn_cast<ConstantInt>(I.getOperand(2));
+ ConstantInt *CI = dyn_cast<ConstantInt>(I.getArgOperand(1));
assert(CI && "Non-constant type in __builtin_object_size?");
- SDValue Arg = getValue(I.getOperand(0));
+ SDValue Arg = getValue(I.getCalledValue());
EVT Ty = Arg.getValueType();
- if (CI->getZExtValue() == 0)
+ if (CI->isZero())
Res = DAG.getConstant(-1ULL, Ty);
else
Res = DAG.getConstant(0, Ty);
return 0;
case Intrinsic::init_trampoline: {
- const Function *F = cast<Function>(I.getOperand(2)->stripPointerCasts());
+ const Function *F = cast<Function>(I.getArgOperand(1)->stripPointerCasts());
SDValue Ops[6];
Ops[0] = getRoot();
- Ops[1] = getValue(I.getOperand(1));
- Ops[2] = getValue(I.getOperand(2));
- Ops[3] = getValue(I.getOperand(3));
- Ops[4] = DAG.getSrcValue(I.getOperand(1));
+ Ops[1] = getValue(I.getArgOperand(0));
+ Ops[2] = getValue(I.getArgOperand(1));
+ Ops[3] = getValue(I.getArgOperand(2));
+ Ops[4] = DAG.getSrcValue(I.getArgOperand(0));
Ops[5] = DAG.getSrcValue(F);
Res = DAG.getNode(ISD::TRAMPOLINE, dl,
}
case Intrinsic::gcroot:
if (GFI) {
- const Value *Alloca = I.getOperand(1);
- const Constant *TypeMap = cast<Constant>(I.getOperand(2));
+ const Value *Alloca = I.getArgOperand(0);
+ const Constant *TypeMap = cast<Constant>(I.getArgOperand(1));
FrameIndexSDNode *FI = cast<FrameIndexSDNode>(getValue(Alloca).getNode());
GFI->addStackRoot(FI->getIndex(), TypeMap);
case Intrinsic::prefetch: {
SDValue Ops[4];
+ unsigned rw = cast<ConstantInt>(I.getArgOperand(1))->getZExtValue();
Ops[0] = getRoot();
- Ops[1] = getValue(I.getOperand(1));
- Ops[2] = getValue(I.getOperand(2));
- Ops[3] = getValue(I.getOperand(3));
- DAG.setRoot(DAG.getNode(ISD::PREFETCH, dl, MVT::Other, &Ops[0], 4));
+ Ops[1] = getValue(I.getArgOperand(0));
+ Ops[2] = getValue(I.getArgOperand(1));
+ Ops[3] = getValue(I.getArgOperand(2));
+ DAG.setRoot(DAG.getMemIntrinsicNode(ISD::PREFETCH, dl,
+ DAG.getVTList(MVT::Other),
+ &Ops[0], 4,
+ EVT::getIntegerVT(*Context, 8),
+ MachinePointerInfo(I.getArgOperand(0)),
+ 0, /* align */
+ false, /* volatile */
+ rw==0, /* read */
+ rw==1)); /* write */
return 0;
}
-
case Intrinsic::memory_barrier: {
SDValue Ops[6];
Ops[0] = getRoot();
for (int x = 1; x < 6; ++x)
- Ops[x] = getValue(I.getOperand(x));
+ Ops[x] = getValue(I.getArgOperand(x - 1));
DAG.setRoot(DAG.getNode(ISD::MEMBARRIER, dl, MVT::Other, &Ops[0], 6));
return 0;
SDValue Root = getRoot();
SDValue L =
DAG.getAtomic(ISD::ATOMIC_CMP_SWAP, getCurDebugLoc(),
- getValue(I.getOperand(2)).getValueType().getSimpleVT(),
+ getValue(I.getArgOperand(1)).getValueType().getSimpleVT(),
Root,
- getValue(I.getOperand(1)),
- getValue(I.getOperand(2)),
- getValue(I.getOperand(3)),
- I.getOperand(1));
+ getValue(I.getArgOperand(0)),
+ getValue(I.getArgOperand(1)),
+ getValue(I.getArgOperand(2)),
+ MachinePointerInfo(I.getArgOperand(0)));
setValue(&I, L);
DAG.setRoot(L.getValue(1));
return 0;
}
}
-/// Test if the given instruction is in a position to be optimized
-/// with a tail-call. This roughly means that it's in a block with
-/// a return and there's nothing that needs to be scheduled
-/// between it and the return.
-///
-/// This function only tests target-independent requirements.
-static bool
-isInTailCallPosition(ImmutableCallSite CS, Attributes CalleeRetAttr,
- const TargetLowering &TLI) {
- const Instruction *I = CS.getInstruction();
- const BasicBlock *ExitBB = I->getParent();
- const TerminatorInst *Term = ExitBB->getTerminator();
- const ReturnInst *Ret = dyn_cast<ReturnInst>(Term);
- const Function *F = ExitBB->getParent();
-
- // The block must end in a return statement or unreachable.
- //
- // FIXME: Decline tailcall if it's not guaranteed and if the block ends in
- // an unreachable, for now. The way tailcall optimization is currently
- // implemented means it will add an epilogue followed by a jump. That is
- // not profitable. Also, if the callee is a special function (e.g.
- // longjmp on x86), it can end up causing miscompilation that has not
- // been fully understood.
- if (!Ret &&
- (!GuaranteedTailCallOpt || !isa<UnreachableInst>(Term))) return false;
-
- // If I will have a chain, make sure no other instruction that will have a
- // chain interposes between I and the return.
- if (I->mayHaveSideEffects() || I->mayReadFromMemory() ||
- !I->isSafeToSpeculativelyExecute())
- for (BasicBlock::const_iterator BBI = prior(prior(ExitBB->end())); ;
- --BBI) {
- if (&*BBI == I)
- break;
- // Debug info intrinsics do not get in the way of tail call optimization.
- if (isa<DbgInfoIntrinsic>(BBI))
- continue;
- if (BBI->mayHaveSideEffects() || BBI->mayReadFromMemory() ||
- !BBI->isSafeToSpeculativelyExecute())
- return false;
- }
-
- // If the block ends with a void return or unreachable, it doesn't matter
- // what the call's return type is.
- if (!Ret || Ret->getNumOperands() == 0) return true;
-
- // If the return value is undef, it doesn't matter what the call's
- // return type is.
- if (isa<UndefValue>(Ret->getOperand(0))) return true;
-
- // Conservatively require the attributes of the call to match those of
- // the return. Ignore noalias because it doesn't affect the call sequence.
- unsigned CallerRetAttr = F->getAttributes().getRetAttributes();
- if ((CalleeRetAttr ^ CallerRetAttr) & ~Attribute::NoAlias)
- return false;
-
- // It's not safe to eliminate the sign / zero extension of the return value.
- if ((CallerRetAttr & Attribute::ZExt) || (CallerRetAttr & Attribute::SExt))
- return false;
-
- // Otherwise, make sure the unmodified return value of I is the return value.
- for (const Instruction *U = dyn_cast<Instruction>(Ret->getOperand(0)); ;
- U = dyn_cast<Instruction>(U->getOperand(0))) {
- if (!U)
- return false;
- if (!U->hasOneUse())
- return false;
- if (U == I)
- break;
- // Check for a truly no-op truncate.
- if (isa<TruncInst>(U) &&
- TLI.isTruncateFree(U->getOperand(0)->getType(), U->getType()))
- continue;
- // Check for a truly no-op bitcast.
- if (isa<BitCastInst>(U) &&
- (U->getOperand(0)->getType() == U->getType() ||
- (U->getOperand(0)->getType()->isPointerTy() &&
- U->getType()->isPointerTy())))
- continue;
- // Otherwise it's not a true no-op.
- return false;
- }
-
- return true;
-}
-
void SelectionDAGBuilder::LowerCallTo(ImmutableCallSite CS, SDValue Callee,
bool isTailCall,
MachineBasicBlock *LandingPad) {
Args.reserve(CS.arg_size());
// Check whether the function can return without sret-demotion.
- SmallVector<EVT, 4> OutVTs;
- SmallVector<ISD::ArgFlagsTy, 4> OutsFlags;
+ SmallVector<ISD::OutputArg, 4> Outs;
SmallVector<uint64_t, 4> Offsets;
- getReturnInfo(RetTy, CS.getAttributes().getRetAttributes(),
- OutVTs, OutsFlags, TLI, &Offsets);
+ GetReturnInfo(RetTy, CS.getAttributes().getRetAttributes(),
+ Outs, TLI, &Offsets);
bool CanLowerReturn = TLI.CanLowerReturn(CS.getCallingConv(),
- FTy->isVarArg(), OutVTs, OutsFlags, DAG);
+ FTy->isVarArg(), Outs, FTy->getContext());
SDValue DemoteStackSlot;
+ int DemoteStackIdx = -100;
if (!CanLowerReturn) {
uint64_t TySize = TLI.getTargetData()->getTypeAllocSize(
unsigned Align = TLI.getTargetData()->getPrefTypeAlignment(
FTy->getReturnType());
MachineFunction &MF = DAG.getMachineFunction();
- int SSFI = MF.getFrameInfo()->CreateStackObject(TySize, Align, false);
+ DemoteStackIdx = MF.getFrameInfo()->CreateStackObject(TySize, Align, false);
const Type *StackSlotPtrType = PointerType::getUnqual(FTy->getReturnType());
- DemoteStackSlot = DAG.getFrameIndex(SSFI, TLI.getPointerTy());
+ DemoteStackSlot = DAG.getFrameIndex(DemoteStackIdx, TLI.getPointerTy());
Entry.Node = DemoteStackSlot;
Entry.Ty = StackSlotPtrType;
Entry.isSExt = false;
!isInTailCallPosition(CS, CS.getAttributes().getRetAttributes(), TLI))
isTailCall = false;
+ // If there's a possibility that fast-isel has already selected some amount
+ // of the current basic block, don't emit a tail call.
+ if (isTailCall && EnableFastISel)
+ isTailCall = false;
+
std::pair<SDValue,SDValue> Result =
TLI.LowerCallTo(getRoot(), RetTy,
CS.paramHasAttr(0, Attribute::SExt),
ComputeValueVTs(TLI, PtrRetTy, PVTs);
assert(PVTs.size() == 1 && "Pointers should fit in one register");
EVT PtrVT = PVTs[0];
- unsigned NumValues = OutVTs.size();
+ unsigned NumValues = Outs.size();
SmallVector<SDValue, 4> Values(NumValues);
SmallVector<SDValue, 4> Chains(NumValues);
SDValue Add = DAG.getNode(ISD::ADD, getCurDebugLoc(), PtrVT,
DemoteStackSlot,
DAG.getConstant(Offsets[i], PtrVT));
- SDValue L = DAG.getLoad(OutVTs[i], getCurDebugLoc(), Result.second,
- Add, NULL, Offsets[i], false, false, 1);
+ SDValue L = DAG.getLoad(Outs[i].VT, getCurDebugLoc(), Result.second,
+ Add,
+ MachinePointerInfo::getFixedStack(DemoteStackIdx, Offsets[i]),
+ false, false, 1);
Values[i] = L;
Chains[i] = L.getValue(1);
}
SDValue Chain = DAG.getNode(ISD::TokenFactor, getCurDebugLoc(),
MVT::Other, &Chains[0], NumValues);
PendingLoads.push_back(Chain);
-
+
// Collect the legal value parts into potentially illegal values
// that correspond to the original function's return values.
SmallVector<EVT, 4> RetTys;
EVT VT = RetTys[I];
EVT RegisterVT = TLI.getRegisterType(RetTy->getContext(), VT);
unsigned NumRegs = TLI.getNumRegisters(RetTy->getContext(), VT);
-
+
SDValue ReturnValue =
getCopyFromParts(DAG, getCurDebugLoc(), &Values[CurReg], NumRegs,
RegisterVT, VT, AssertOp);
SDValue Ptr = Builder.getValue(PtrVal);
SDValue LoadVal = Builder.DAG.getLoad(LoadVT, Builder.getCurDebugLoc(), Root,
- Ptr, PtrVal /*SrcValue*/, 0/*SVOffset*/,
+ Ptr, MachinePointerInfo(PtrVal),
false /*volatile*/,
false /*nontemporal*/, 1 /* align=1 */);
/// lowered like a normal call.
bool SelectionDAGBuilder::visitMemCmpCall(const CallInst &I) {
// Verify that the prototype makes sense. int memcmp(void*,void*,size_t)
- if (I.getNumOperands() != 4)
+ if (I.getNumArgOperands() != 3)
return false;
- const Value *LHS = I.getOperand(1), *RHS = I.getOperand(2);
+ const Value *LHS = I.getArgOperand(0), *RHS = I.getArgOperand(1);
if (!LHS->getType()->isPointerTy() || !RHS->getType()->isPointerTy() ||
- !I.getOperand(3)->getType()->isIntegerTy() ||
+ !I.getArgOperand(2)->getType()->isIntegerTy() ||
!I.getType()->isIntegerTy())
return false;
- const ConstantInt *Size = dyn_cast<ConstantInt>(I.getOperand(3));
+ const ConstantInt *Size = dyn_cast<ConstantInt>(I.getArgOperand(2));
// memcmp(S1,S2,2) != 0 -> (*(short*)LHS != *(short*)RHS) != 0
// memcmp(S1,S2,4) != 0 -> (*(int*)LHS != *(int*)RHS) != 0
ActuallyDoIt = false;
}
- if (ActuallyDoIt) {
- SDValue LHSVal = getMemCmpLoad(LHS, LoadVT, LoadTy, *this);
- SDValue RHSVal = getMemCmpLoad(RHS, LoadVT, LoadTy, *this);
-
- SDValue Res = DAG.getSetCC(getCurDebugLoc(), MVT::i1, LHSVal, RHSVal,
- ISD::SETNE);
- EVT CallVT = TLI.getValueType(I.getType(), true);
- setValue(&I, DAG.getZExtOrTrunc(Res, getCurDebugLoc(), CallVT));
- return true;
- }
- }
-
-
- return false;
-}
-
-
-void SelectionDAGBuilder::visitCall(const CallInst &I) {
- const char *RenameFn = 0;
- if (Function *F = I.getCalledFunction()) {
- if (F->isDeclaration()) {
- const TargetIntrinsicInfo *II = TLI.getTargetMachine().getIntrinsicInfo();
- if (II) {
- if (unsigned IID = II->getIntrinsicID(F)) {
- RenameFn = visitIntrinsicCall(I, IID);
- if (!RenameFn)
- return;
- }
- }
- if (unsigned IID = F->getIntrinsicID()) {
- RenameFn = visitIntrinsicCall(I, IID);
- if (!RenameFn)
- return;
- }
- }
-
- // Check for well-known libc/libm calls. If the function is internal, it
- // can't be a library call.
- if (!F->hasLocalLinkage() && F->hasName()) {
- StringRef Name = F->getName();
- if (Name == "copysign" || Name == "copysignf" || Name == "copysignl") {
- if (I.getNumOperands() == 3 && // Basic sanity checks.
- I.getOperand(1)->getType()->isFloatingPointTy() &&
- I.getType() == I.getOperand(1)->getType() &&
- I.getType() == I.getOperand(2)->getType()) {
- SDValue LHS = getValue(I.getOperand(1));
- SDValue RHS = getValue(I.getOperand(2));
- setValue(&I, DAG.getNode(ISD::FCOPYSIGN, getCurDebugLoc(),
- LHS.getValueType(), LHS, RHS));
- return;
- }
- } else if (Name == "fabs" || Name == "fabsf" || Name == "fabsl") {
- if (I.getNumOperands() == 2 && // Basic sanity checks.
- I.getOperand(1)->getType()->isFloatingPointTy() &&
- I.getType() == I.getOperand(1)->getType()) {
- SDValue Tmp = getValue(I.getOperand(1));
- setValue(&I, DAG.getNode(ISD::FABS, getCurDebugLoc(),
- Tmp.getValueType(), Tmp));
- return;
- }
- } else if (Name == "sin" || Name == "sinf" || Name == "sinl") {
- if (I.getNumOperands() == 2 && // Basic sanity checks.
- I.getOperand(1)->getType()->isFloatingPointTy() &&
- I.getType() == I.getOperand(1)->getType() &&
- I.onlyReadsMemory()) {
- SDValue Tmp = getValue(I.getOperand(1));
- setValue(&I, DAG.getNode(ISD::FSIN, getCurDebugLoc(),
- Tmp.getValueType(), Tmp));
- return;
- }
- } else if (Name == "cos" || Name == "cosf" || Name == "cosl") {
- if (I.getNumOperands() == 2 && // Basic sanity checks.
- I.getOperand(1)->getType()->isFloatingPointTy() &&
- I.getType() == I.getOperand(1)->getType() &&
- I.onlyReadsMemory()) {
- SDValue Tmp = getValue(I.getOperand(1));
- setValue(&I, DAG.getNode(ISD::FCOS, getCurDebugLoc(),
- Tmp.getValueType(), Tmp));
- return;
- }
- } else if (Name == "sqrt" || Name == "sqrtf" || Name == "sqrtl") {
- if (I.getNumOperands() == 2 && // Basic sanity checks.
- I.getOperand(1)->getType()->isFloatingPointTy() &&
- I.getType() == I.getOperand(1)->getType() &&
- I.onlyReadsMemory()) {
- SDValue Tmp = getValue(I.getOperand(1));
- setValue(&I, DAG.getNode(ISD::FSQRT, getCurDebugLoc(),
- Tmp.getValueType(), Tmp));
- return;
- }
- } else if (Name == "memcmp") {
- if (visitMemCmpCall(I))
- return;
- }
- }
- } else if (isa<InlineAsm>(I.getOperand(0))) {
- visitInlineAsm(&I);
- return;
- }
-
- SDValue Callee;
- if (!RenameFn)
- Callee = getValue(I.getOperand(0));
- else
- Callee = DAG.getExternalSymbol(RenameFn, TLI.getPointerTy());
-
- // Check if we can potentially perform a tail call. More detailed checking is
- // be done within LowerCallTo, after more information about the call is known.
- LowerCallTo(&I, Callee, I.isTailCall());
-}
-
-/// getCopyFromRegs - Emit a series of CopyFromReg nodes that copies from
-/// this value and returns the result as a ValueVT value. This uses
-/// Chain/Flag as the input and updates them for the output Chain/Flag.
-/// If the Flag pointer is NULL, no flag is used.
-SDValue RegsForValue::getCopyFromRegs(SelectionDAG &DAG, DebugLoc dl,
- SDValue &Chain, SDValue *Flag) const {
- // Assemble the legal parts into the final values.
- SmallVector<SDValue, 4> Values(ValueVTs.size());
- SmallVector<SDValue, 8> Parts;
- for (unsigned Value = 0, Part = 0, e = ValueVTs.size(); Value != e; ++Value) {
- // Copy the legal parts from the registers.
- EVT ValueVT = ValueVTs[Value];
- unsigned NumRegs = TLI->getNumRegisters(*DAG.getContext(), ValueVT);
- EVT RegisterVT = RegVTs[Value];
-
- Parts.resize(NumRegs);
- for (unsigned i = 0; i != NumRegs; ++i) {
- SDValue P;
- if (Flag == 0) {
- P = DAG.getCopyFromReg(Chain, dl, Regs[Part+i], RegisterVT);
- } else {
- P = DAG.getCopyFromReg(Chain, dl, Regs[Part+i], RegisterVT, *Flag);
- *Flag = P.getValue(2);
- }
-
- Chain = P.getValue(1);
-
- // If the source register was virtual and if we know something about it,
- // add an assert node.
- if (TargetRegisterInfo::isVirtualRegister(Regs[Part+i]) &&
- RegisterVT.isInteger() && !RegisterVT.isVector()) {
- unsigned SlotNo = Regs[Part+i]-TargetRegisterInfo::FirstVirtualRegister;
- FunctionLoweringInfo &FLI = DAG.getFunctionLoweringInfo();
- if (FLI.LiveOutRegInfo.size() > SlotNo) {
- FunctionLoweringInfo::LiveOutInfo &LOI = FLI.LiveOutRegInfo[SlotNo];
-
- unsigned RegSize = RegisterVT.getSizeInBits();
- unsigned NumSignBits = LOI.NumSignBits;
- unsigned NumZeroBits = LOI.KnownZero.countLeadingOnes();
-
- // FIXME: We capture more information than the dag can represent. For
- // now, just use the tightest assertzext/assertsext possible.
- bool isSExt = true;
- EVT FromVT(MVT::Other);
- if (NumSignBits == RegSize)
- isSExt = true, FromVT = MVT::i1; // ASSERT SEXT 1
- else if (NumZeroBits >= RegSize-1)
- isSExt = false, FromVT = MVT::i1; // ASSERT ZEXT 1
- else if (NumSignBits > RegSize-8)
- isSExt = true, FromVT = MVT::i8; // ASSERT SEXT 8
- else if (NumZeroBits >= RegSize-8)
- isSExt = false, FromVT = MVT::i8; // ASSERT ZEXT 8
- else if (NumSignBits > RegSize-16)
- isSExt = true, FromVT = MVT::i16; // ASSERT SEXT 16
- else if (NumZeroBits >= RegSize-16)
- isSExt = false, FromVT = MVT::i16; // ASSERT ZEXT 16
- else if (NumSignBits > RegSize-32)
- isSExt = true, FromVT = MVT::i32; // ASSERT SEXT 32
- else if (NumZeroBits >= RegSize-32)
- isSExt = false, FromVT = MVT::i32; // ASSERT ZEXT 32
-
- if (FromVT != MVT::Other)
- P = DAG.getNode(isSExt ? ISD::AssertSext : ISD::AssertZext, dl,
- RegisterVT, P, DAG.getValueType(FromVT));
- }
- }
-
- Parts[i] = P;
- }
-
- Values[Value] = getCopyFromParts(DAG, dl, Parts.begin(),
- NumRegs, RegisterVT, ValueVT);
- Part += NumRegs;
- Parts.clear();
- }
-
- return DAG.getNode(ISD::MERGE_VALUES, dl,
- DAG.getVTList(&ValueVTs[0], ValueVTs.size()),
- &Values[0], ValueVTs.size());
-}
-
-/// getCopyToRegs - Emit a series of CopyToReg nodes that copies the
-/// specified value into the registers specified by this object. This uses
-/// Chain/Flag as the input and updates them for the output Chain/Flag.
-/// If the Flag pointer is NULL, no flag is used.
-void RegsForValue::getCopyToRegs(SDValue Val, SelectionDAG &DAG, DebugLoc dl,
- SDValue &Chain, SDValue *Flag) const {
- // Get the list of the values's legal parts.
- unsigned NumRegs = Regs.size();
- SmallVector<SDValue, 8> Parts(NumRegs);
- for (unsigned Value = 0, Part = 0, e = ValueVTs.size(); Value != e; ++Value) {
- EVT ValueVT = ValueVTs[Value];
- unsigned NumParts = TLI->getNumRegisters(*DAG.getContext(), ValueVT);
- EVT RegisterVT = RegVTs[Value];
-
- getCopyToParts(DAG, dl,
- Val.getValue(Val.getResNo() + Value),
- &Parts[Part], NumParts, RegisterVT);
- Part += NumParts;
- }
-
- // Copy the parts into the registers.
- SmallVector<SDValue, 8> Chains(NumRegs);
- for (unsigned i = 0; i != NumRegs; ++i) {
- SDValue Part;
- if (Flag == 0) {
- Part = DAG.getCopyToReg(Chain, dl, Regs[i], Parts[i]);
- } else {
- Part = DAG.getCopyToReg(Chain, dl, Regs[i], Parts[i], *Flag);
- *Flag = Part.getValue(1);
- }
-
- Chains[i] = Part.getValue(0);
- }
-
- if (NumRegs == 1 || Flag)
- // If NumRegs > 1 && Flag is used then the use of the last CopyToReg is
- // flagged to it. That is the CopyToReg nodes and the user are considered
- // a single scheduling unit. If we create a TokenFactor and return it as
- // chain, then the TokenFactor is both a predecessor (operand) of the
- // user as well as a successor (the TF operands are flagged to the user).
- // c1, f1 = CopyToReg
- // c2, f2 = CopyToReg
- // c3 = TokenFactor c1, c2
- // ...
- // = op c3, ..., f2
- Chain = Chains[NumRegs-1];
- else
- Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, &Chains[0], NumRegs);
-}
-
-/// AddInlineAsmOperands - Add this value to the specified inlineasm node
-/// operand list. This adds the code marker and includes the number of
-/// values added into it.
-void RegsForValue::AddInlineAsmOperands(unsigned Code, bool HasMatching,
- unsigned MatchingIdx,
- SelectionDAG &DAG,
- std::vector<SDValue> &Ops) const {
- unsigned Flag = InlineAsm::getFlagWord(Code, Regs.size());
- if (HasMatching)
- Flag = InlineAsm::getFlagWordForMatchingOp(Flag, MatchingIdx);
- SDValue Res = DAG.getTargetConstant(Flag, MVT::i32);
- Ops.push_back(Res);
+ if (ActuallyDoIt) {
+ SDValue LHSVal = getMemCmpLoad(LHS, LoadVT, LoadTy, *this);
+ SDValue RHSVal = getMemCmpLoad(RHS, LoadVT, LoadTy, *this);
- for (unsigned Value = 0, Reg = 0, e = ValueVTs.size(); Value != e; ++Value) {
- unsigned NumRegs = TLI->getNumRegisters(*DAG.getContext(), ValueVTs[Value]);
- EVT RegisterVT = RegVTs[Value];
- for (unsigned i = 0; i != NumRegs; ++i) {
- assert(Reg < Regs.size() && "Mismatch in # registers expected");
- Ops.push_back(DAG.getRegister(Regs[Reg++], RegisterVT));
+ SDValue Res = DAG.getSetCC(getCurDebugLoc(), MVT::i1, LHSVal, RHSVal,
+ ISD::SETNE);
+ EVT CallVT = TLI.getValueType(I.getType(), true);
+ setValue(&I, DAG.getZExtOrTrunc(Res, getCurDebugLoc(), CallVT));
+ return true;
}
}
+
+
+ return false;
}
-/// isAllocatableRegister - If the specified register is safe to allocate,
-/// i.e. it isn't a stack pointer or some other special register, return the
-/// register class for the register. Otherwise, return null.
-static const TargetRegisterClass *
-isAllocatableRegister(unsigned Reg, MachineFunction &MF,
- const TargetLowering &TLI,
- const TargetRegisterInfo *TRI) {
- EVT FoundVT = MVT::Other;
- const TargetRegisterClass *FoundRC = 0;
- for (TargetRegisterInfo::regclass_iterator RCI = TRI->regclass_begin(),
- E = TRI->regclass_end(); RCI != E; ++RCI) {
- EVT ThisVT = MVT::Other;
- const TargetRegisterClass *RC = *RCI;
- // If none of the value types for this register class are valid, we
- // can't use it. For example, 64-bit reg classes on 32-bit targets.
- for (TargetRegisterClass::vt_iterator I = RC->vt_begin(), E = RC->vt_end();
- I != E; ++I) {
- if (TLI.isTypeLegal(*I)) {
- // If we have already found this register in a different register class,
- // choose the one with the largest VT specified. For example, on
- // PowerPC, we favor f64 register classes over f32.
- if (FoundVT == MVT::Other || FoundVT.bitsLT(*I)) {
- ThisVT = *I;
- break;
- }
+void SelectionDAGBuilder::visitCall(const CallInst &I) {
+ // Handle inline assembly differently.
+ if (isa<InlineAsm>(I.getCalledValue())) {
+ visitInlineAsm(&I);
+ return;
+ }
+
+ // See if any floating point values are being passed to this function. This is
+ // used to emit an undefined reference to fltused on Windows.
+ const FunctionType *FT =
+ cast<FunctionType>(I.getCalledValue()->getType()->getContainedType(0));
+ MachineModuleInfo &MMI = DAG.getMachineFunction().getMMI();
+ if (FT->isVarArg() &&
+ !MMI.callsExternalVAFunctionWithFloatingPointArguments()) {
+ for (unsigned i = 0, e = I.getNumArgOperands(); i != e; ++i) {
+ const Type* T = I.getArgOperand(i)->getType();
+ for (po_iterator<const Type*> i = po_begin(T), e = po_end(T);
+ i != e; ++i) {
+ if (!i->isFloatingPointTy()) continue;
+ MMI.setCallsExternalVAFunctionWithFloatingPointArguments(true);
+ break;
}
}
+ }
- if (ThisVT == MVT::Other) continue;
+ const char *RenameFn = 0;
+ if (Function *F = I.getCalledFunction()) {
+ if (F->isDeclaration()) {
+ if (const TargetIntrinsicInfo *II = TM.getIntrinsicInfo()) {
+ if (unsigned IID = II->getIntrinsicID(F)) {
+ RenameFn = visitIntrinsicCall(I, IID);
+ if (!RenameFn)
+ return;
+ }
+ }
+ if (unsigned IID = F->getIntrinsicID()) {
+ RenameFn = visitIntrinsicCall(I, IID);
+ if (!RenameFn)
+ return;
+ }
+ }
- // NOTE: This isn't ideal. In particular, this might allocate the
- // frame pointer in functions that need it (due to them not being taken
- // out of allocation, because a variable sized allocation hasn't been seen
- // yet). This is a slight code pessimization, but should still work.
- for (TargetRegisterClass::iterator I = RC->allocation_order_begin(MF),
- E = RC->allocation_order_end(MF); I != E; ++I)
- if (*I == Reg) {
- // We found a matching register class. Keep looking at others in case
- // we find one with larger registers that this physreg is also in.
- FoundRC = RC;
- FoundVT = ThisVT;
- break;
+ // Check for well-known libc/libm calls. If the function is internal, it
+ // can't be a library call.
+ if (!F->hasLocalLinkage() && F->hasName()) {
+ StringRef Name = F->getName();
+ if (Name == "copysign" || Name == "copysignf" || Name == "copysignl") {
+ if (I.getNumArgOperands() == 2 && // Basic sanity checks.
+ I.getArgOperand(0)->getType()->isFloatingPointTy() &&
+ I.getType() == I.getArgOperand(0)->getType() &&
+ I.getType() == I.getArgOperand(1)->getType()) {
+ SDValue LHS = getValue(I.getArgOperand(0));
+ SDValue RHS = getValue(I.getArgOperand(1));
+ setValue(&I, DAG.getNode(ISD::FCOPYSIGN, getCurDebugLoc(),
+ LHS.getValueType(), LHS, RHS));
+ return;
+ }
+ } else if (Name == "fabs" || Name == "fabsf" || Name == "fabsl") {
+ if (I.getNumArgOperands() == 1 && // Basic sanity checks.
+ I.getArgOperand(0)->getType()->isFloatingPointTy() &&
+ I.getType() == I.getArgOperand(0)->getType()) {
+ SDValue Tmp = getValue(I.getArgOperand(0));
+ setValue(&I, DAG.getNode(ISD::FABS, getCurDebugLoc(),
+ Tmp.getValueType(), Tmp));
+ return;
+ }
+ } else if (Name == "sin" || Name == "sinf" || Name == "sinl") {
+ if (I.getNumArgOperands() == 1 && // Basic sanity checks.
+ I.getArgOperand(0)->getType()->isFloatingPointTy() &&
+ I.getType() == I.getArgOperand(0)->getType() &&
+ I.onlyReadsMemory()) {
+ SDValue Tmp = getValue(I.getArgOperand(0));
+ setValue(&I, DAG.getNode(ISD::FSIN, getCurDebugLoc(),
+ Tmp.getValueType(), Tmp));
+ return;
+ }
+ } else if (Name == "cos" || Name == "cosf" || Name == "cosl") {
+ if (I.getNumArgOperands() == 1 && // Basic sanity checks.
+ I.getArgOperand(0)->getType()->isFloatingPointTy() &&
+ I.getType() == I.getArgOperand(0)->getType() &&
+ I.onlyReadsMemory()) {
+ SDValue Tmp = getValue(I.getArgOperand(0));
+ setValue(&I, DAG.getNode(ISD::FCOS, getCurDebugLoc(),
+ Tmp.getValueType(), Tmp));
+ return;
+ }
+ } else if (Name == "sqrt" || Name == "sqrtf" || Name == "sqrtl") {
+ if (I.getNumArgOperands() == 1 && // Basic sanity checks.
+ I.getArgOperand(0)->getType()->isFloatingPointTy() &&
+ I.getType() == I.getArgOperand(0)->getType() &&
+ I.onlyReadsMemory()) {
+ SDValue Tmp = getValue(I.getArgOperand(0));
+ setValue(&I, DAG.getNode(ISD::FSQRT, getCurDebugLoc(),
+ Tmp.getValueType(), Tmp));
+ return;
+ }
+ } else if (Name == "memcmp") {
+ if (visitMemCmpCall(I))
+ return;
}
+ }
}
- return FoundRC;
-}
+ SDValue Callee;
+ if (!RenameFn)
+ Callee = getValue(I.getCalledValue());
+ else
+ Callee = DAG.getExternalSymbol(RenameFn, TLI.getPointerTy());
+
+ // Check if we can potentially perform a tail call. More detailed checking is
+ // be done within LowerCallTo, after more information about the call is known.
+ LowerCallTo(&I, Callee, I.isTailCall());
+}
namespace llvm {
+
/// AsmOperandInfo - This contains information for each constraint that we are
/// lowering.
-class VISIBILITY_HIDDEN SDISelAsmOperandInfo :
+class LLVM_LIBRARY_VISIBILITY SDISelAsmOperandInfo :
public TargetLowering::AsmOperandInfo {
public:
/// CallOperand - If this is the result output operand or a clobber
/// contains the set of register corresponding to the operand.
RegsForValue AssignedRegs;
- explicit SDISelAsmOperandInfo(const InlineAsm::ConstraintInfo &info)
+ explicit SDISelAsmOperandInfo(const TargetLowering::AsmOperandInfo &info)
: TargetLowering::AsmOperandInfo(info), CallOperand(0,0) {
}
Regs.insert(*Aliases);
}
};
+
+typedef SmallVector<SDISelAsmOperandInfo,16> SDISelAsmOperandInfoVector;
+
} // end llvm namespace.
+/// isAllocatableRegister - If the specified register is safe to allocate,
+/// i.e. it isn't a stack pointer or some other special register, return the
+/// register class for the register. Otherwise, return null.
+static const TargetRegisterClass *
+isAllocatableRegister(unsigned Reg, MachineFunction &MF,
+ const TargetLowering &TLI,
+ const TargetRegisterInfo *TRI) {
+ EVT FoundVT = MVT::Other;
+ const TargetRegisterClass *FoundRC = 0;
+ for (TargetRegisterInfo::regclass_iterator RCI = TRI->regclass_begin(),
+ E = TRI->regclass_end(); RCI != E; ++RCI) {
+ EVT ThisVT = MVT::Other;
+
+ const TargetRegisterClass *RC = *RCI;
+ // If none of the value types for this register class are valid, we
+ // can't use it. For example, 64-bit reg classes on 32-bit targets.
+ for (TargetRegisterClass::vt_iterator I = RC->vt_begin(), E = RC->vt_end();
+ I != E; ++I) {
+ if (TLI.isTypeLegal(*I)) {
+ // If we have already found this register in a different register class,
+ // choose the one with the largest VT specified. For example, on
+ // PowerPC, we favor f64 register classes over f32.
+ if (FoundVT == MVT::Other || FoundVT.bitsLT(*I)) {
+ ThisVT = *I;
+ break;
+ }
+ }
+ }
+
+ if (ThisVT == MVT::Other) continue;
+
+ // NOTE: This isn't ideal. In particular, this might allocate the
+ // frame pointer in functions that need it (due to them not being taken
+ // out of allocation, because a variable sized allocation hasn't been seen
+ // yet). This is a slight code pessimization, but should still work.
+ for (TargetRegisterClass::iterator I = RC->allocation_order_begin(MF),
+ E = RC->allocation_order_end(MF); I != E; ++I)
+ if (*I == Reg) {
+ // We found a matching register class. Keep looking at others in case
+ // we find one with larger registers that this physreg is also in.
+ FoundRC = RC;
+ FoundVT = ThisVT;
+ break;
+ }
+ }
+ return FoundRC;
+}
/// GetRegistersForValue - Assign registers (virtual or physical) for the
/// specified operand. We prefer to assign virtual registers, to allow the
// vector types).
EVT RegVT = *PhysReg.second->vt_begin();
if (RegVT.getSizeInBits() == OpInfo.ConstraintVT.getSizeInBits()) {
- OpInfo.CallOperand = DAG.getNode(ISD::BIT_CONVERT, getCurDebugLoc(),
+ OpInfo.CallOperand = DAG.getNode(ISD::BITCAST, getCurDebugLoc(),
RegVT, OpInfo.CallOperand);
OpInfo.ConstraintVT = RegVT;
} else if (RegVT.isInteger() && OpInfo.ConstraintVT.isFloatingPoint()) {
// machine.
RegVT = EVT::getIntegerVT(Context,
OpInfo.ConstraintVT.getSizeInBits());
- OpInfo.CallOperand = DAG.getNode(ISD::BIT_CONVERT, getCurDebugLoc(),
+ OpInfo.CallOperand = DAG.getNode(ISD::BITCAST, getCurDebugLoc(),
RegVT, OpInfo.CallOperand);
OpInfo.ConstraintVT = RegVT;
}
}
}
- OpInfo.AssignedRegs = RegsForValue(TLI, Regs, RegVT, ValueVT);
+ OpInfo.AssignedRegs = RegsForValue(Regs, RegVT, ValueVT);
const TargetRegisterInfo *TRI = DAG.getTarget().getRegisterInfo();
OpInfo.MarkAllocatedRegs(isOutReg, isInReg, OutputRegs, InputRegs, *TRI);
return;
for (; NumRegs; --NumRegs)
Regs.push_back(RegInfo.createVirtualRegister(RC));
- OpInfo.AssignedRegs = RegsForValue(TLI, Regs, RegVT, ValueVT);
+ OpInfo.AssignedRegs = RegsForValue(Regs, RegVT, ValueVT);
return;
}
for (unsigned i = RegStart; i != RegEnd; ++i)
Regs.push_back(RegClassRegs[i]);
- OpInfo.AssignedRegs = RegsForValue(TLI, Regs, *RC->vt_begin(),
+ OpInfo.AssignedRegs = RegsForValue(Regs, *RC->vt_begin(),
OpInfo.ConstraintVT);
OpInfo.MarkAllocatedRegs(isOutReg, isInReg, OutputRegs, InputRegs, *TRI);
return;
const InlineAsm *IA = cast<InlineAsm>(CS.getCalledValue());
/// ConstraintOperands - Information about all of the constraints.
- std::vector<SDISelAsmOperandInfo> ConstraintOperands;
+ SDISelAsmOperandInfoVector ConstraintOperands;
std::set<unsigned> OutputRegs, InputRegs;
- // Do a prepass over the constraints, canonicalizing them, and building up the
- // ConstraintOperands list.
- std::vector<InlineAsm::ConstraintInfo>
- ConstraintInfos = IA->ParseConstraints();
-
- bool hasMemory = hasInlineAsmMemConstraint(ConstraintInfos, TLI);
-
- SDValue Chain, Flag;
-
- // We won't need to flush pending loads if this asm doesn't touch
- // memory and is nonvolatile.
- if (hasMemory || IA->hasSideEffects())
- Chain = getRoot();
- else
- Chain = DAG.getRoot();
+ TargetLowering::AsmOperandInfoVector TargetConstraints = TLI.ParseConstraints(CS);
+ bool hasMemory = false;
unsigned ArgNo = 0; // ArgNo - The argument of the CallInst.
unsigned ResNo = 0; // ResNo - The result number of the next output.
- for (unsigned i = 0, e = ConstraintInfos.size(); i != e; ++i) {
- ConstraintOperands.push_back(SDISelAsmOperandInfo(ConstraintInfos[i]));
+ for (unsigned i = 0, e = TargetConstraints.size(); i != e; ++i) {
+ ConstraintOperands.push_back(SDISelAsmOperandInfo(TargetConstraints[i]));
SDISelAsmOperandInfo &OpInfo = ConstraintOperands.back();
EVT OpVT = MVT::Other;
// If this is an input or an indirect output, process the call argument.
// BasicBlocks are labels, currently appearing only in asm's.
if (OpInfo.CallOperandVal) {
- // Strip bitcasts, if any. This mostly comes up for functions.
- OpInfo.CallOperandVal = OpInfo.CallOperandVal->stripPointerCasts();
-
if (const BasicBlock *BB = dyn_cast<BasicBlock>(OpInfo.CallOperandVal)) {
OpInfo.CallOperand = DAG.getBasicBlock(FuncInfo.MBBMap[BB]);
} else {
}
OpInfo.ConstraintVT = OpVT;
+
+ // Indirect operand accesses access memory.
+ if (OpInfo.isIndirect)
+ hasMemory = true;
+ else {
+ for (unsigned j = 0, ee = OpInfo.Codes.size(); j != ee; ++j) {
+ TargetLowering::ConstraintType CType = TLI.getConstraintType(OpInfo.Codes[j]);
+ if (CType == TargetLowering::C_Memory) {
+ hasMemory = true;
+ break;
+ }
+ }
+ }
}
+ SDValue Chain, Flag;
+
+ // We won't need to flush pending loads if this asm doesn't touch
+ // memory and is nonvolatile.
+ if (hasMemory || IA->hasSideEffects())
+ Chain = getRoot();
+ else
+ Chain = DAG.getRoot();
+
// Second pass over the constraints: compute which constraint option to use
// and assign registers to constraints that want a specific physreg.
- for (unsigned i = 0, e = ConstraintInfos.size(); i != e; ++i) {
+ for (unsigned i = 0, e = ConstraintOperands.size(); i != e; ++i) {
SDISelAsmOperandInfo &OpInfo = ConstraintOperands[i];
// If this is an output operand with a matching input operand, look up the
// error.
if (OpInfo.hasMatchingInput()) {
SDISelAsmOperandInfo &Input = ConstraintOperands[OpInfo.MatchingInput];
-
+
if (OpInfo.ConstraintVT != Input.ConstraintVT) {
if ((OpInfo.ConstraintVT.isInteger() !=
Input.ConstraintVT.isInteger()) ||
}
// Compute the constraint code and ConstraintType to use.
- TLI.ComputeConstraintToUse(OpInfo, OpInfo.CallOperand, hasMemory, &DAG);
+ TLI.ComputeConstraintToUse(OpInfo, OpInfo.CallOperand, &DAG);
// If this is a memory input, and if the operand is not indirect, do what we
// need to to provide an address for the memory input.
if (OpInfo.ConstraintType == TargetLowering::C_Memory &&
!OpInfo.isIndirect) {
- assert(OpInfo.Type == InlineAsm::isInput &&
+ assert((OpInfo.isMultipleAlternative || (OpInfo.Type == InlineAsm::isInput)) &&
"Can only indirectify direct input operands!");
// Memory operands really want the address of the value. If we don't have
int SSFI = MF.getFrameInfo()->CreateStackObject(TySize, Align, false);
SDValue StackSlot = DAG.getFrameIndex(SSFI, TLI.getPointerTy());
Chain = DAG.getStore(Chain, getCurDebugLoc(),
- OpInfo.CallOperand, StackSlot, NULL, 0,
+ OpInfo.CallOperand, StackSlot,
+ MachinePointerInfo::getFixedStack(SSFI),
false, false, 0);
OpInfo.CallOperand = StackSlot;
}
GetRegistersForValue(OpInfo, OutputRegs, InputRegs);
}
- ConstraintInfos.clear();
-
// Second pass - Loop over all of the operands, assigning virtual or physregs
// to register class operands.
for (unsigned i = 0, e = ConstraintOperands.size(); i != e; ++i) {
const MDNode *SrcLoc = CS.getInstruction()->getMetadata("srcloc");
AsmNodeOperands.push_back(DAG.getMDNode(SrcLoc));
+ // Remember the HasSideEffect and AlignStack bits as operand 3.
+ unsigned ExtraInfo = 0;
+ if (IA->hasSideEffects())
+ ExtraInfo |= InlineAsm::Extra_HasSideEffects;
+ if (IA->isAlignStack())
+ ExtraInfo |= InlineAsm::Extra_IsAlignStack;
+ AsmNodeOperands.push_back(DAG.getTargetConstant(ExtraInfo,
+ TLI.getPointerTy()));
+
// Loop over all of the inputs, copying the operand values into the
// appropriate registers and processing the output regs.
RegsForValue RetValRegs;
// Add (OpFlag&0xffff)>>3 registers to MatchedRegs.
if (OpInfo.isIndirect) {
// This happens on gcc/testsuite/gcc.dg/pr8788-1.c
- LLVMContext &Ctx = CurMBB->getParent()->getFunction()->getContext();
+ LLVMContext &Ctx = *DAG.getContext();
Ctx.emitError(CS.getInstruction(), "inline asm not supported yet:"
" don't know how to handle tied "
"indirect register inputs");
}
-
+
RegsForValue MatchedRegs;
- MatchedRegs.TLI = &TLI;
MatchedRegs.ValueVTs.push_back(InOperandVal.getValueType());
EVT RegVT = AsmNodeOperands[CurOp+1].getValueType();
MatchedRegs.RegVTs.push_back(RegVT);
DAG, AsmNodeOperands);
break;
}
-
+
assert(InlineAsm::isMemKind(OpFlag) && "Unknown matching constraint!");
assert(InlineAsm::getNumOperandRegisters(OpFlag) == 1 &&
"Unexpected number of operands");
break;
}
- if (OpInfo.ConstraintType == TargetLowering::C_Other) {
- assert(!OpInfo.isIndirect &&
- "Don't know how to handle indirect other inputs yet!");
+ // Treat indirect 'X' constraint as memory.
+ if (OpInfo.ConstraintType == TargetLowering::C_Other &&
+ OpInfo.isIndirect)
+ OpInfo.ConstraintType = TargetLowering::C_Memory;
+ if (OpInfo.ConstraintType == TargetLowering::C_Other) {
std::vector<SDValue> Ops;
TLI.LowerAsmOperandForConstraint(InOperandVal, OpInfo.ConstraintCode[0],
- hasMemory, Ops, DAG);
+ Ops, DAG);
if (Ops.empty())
report_fatal_error("Invalid operand for inline asm constraint '" +
Twine(OpInfo.ConstraintCode) + "'!");
AsmNodeOperands.insert(AsmNodeOperands.end(), Ops.begin(), Ops.end());
break;
}
-
+
if (OpInfo.ConstraintType == TargetLowering::C_Memory) {
assert(OpInfo.isIndirect && "Operand must be indirect to be a mem!");
assert(InOperandVal.getValueType() == TLI.getPointerTy() &&
// Copy the input into the appropriate registers.
if (OpInfo.AssignedRegs.Regs.empty() ||
- !OpInfo.AssignedRegs.areValueTypesLegal())
+ !OpInfo.AssignedRegs.areValueTypesLegal(TLI))
report_fatal_error("Couldn't allocate input reg for constraint '" +
Twine(OpInfo.ConstraintCode) + "'!");
}
// Finish up input operands. Set the input chain and add the flag last.
- AsmNodeOperands[0] = Chain;
+ AsmNodeOperands[InlineAsm::Op_InputChain] = Chain;
if (Flag.getNode()) AsmNodeOperands.push_back(Flag);
Chain = DAG.getNode(ISD::INLINEASM, getCurDebugLoc(),
- DAG.getVTList(MVT::Other, MVT::Flag),
+ DAG.getVTList(MVT::Other, MVT::Glue),
&AsmNodeOperands[0], AsmNodeOperands.size());
Flag = Chain.getValue(1);
// If this asm returns a register value, copy the result from that register
// and set it as the value of the call.
if (!RetValRegs.Regs.empty()) {
- SDValue Val = RetValRegs.getCopyFromRegs(DAG, getCurDebugLoc(),
+ SDValue Val = RetValRegs.getCopyFromRegs(DAG, FuncInfo, getCurDebugLoc(),
Chain, &Flag);
// FIXME: Why don't we do this for inline asms with MRVs?
// not have the same VT as was expected. Convert it to the right type
// with bit_convert.
if (ResultType != Val.getValueType() && Val.getValueType().isVector()) {
- Val = DAG.getNode(ISD::BIT_CONVERT, getCurDebugLoc(),
+ Val = DAG.getNode(ISD::BITCAST, getCurDebugLoc(),
ResultType, Val);
} else if (ResultType != Val.getValueType() &&
for (unsigned i = 0, e = IndirectStoresToEmit.size(); i != e; ++i) {
RegsForValue &OutRegs = IndirectStoresToEmit[i].first;
const Value *Ptr = IndirectStoresToEmit[i].second;
- SDValue OutVal = OutRegs.getCopyFromRegs(DAG, getCurDebugLoc(),
+ SDValue OutVal = OutRegs.getCopyFromRegs(DAG, FuncInfo, getCurDebugLoc(),
Chain, &Flag);
StoresToEmit.push_back(std::make_pair(OutVal, Ptr));
}
SDValue Val = DAG.getStore(Chain, getCurDebugLoc(),
StoresToEmit[i].first,
getValue(StoresToEmit[i].second),
- StoresToEmit[i].second, 0,
+ MachinePointerInfo(StoresToEmit[i].second),
false, false, 0);
OutChains.push_back(Val);
}
void SelectionDAGBuilder::visitVAStart(const CallInst &I) {
DAG.setRoot(DAG.getNode(ISD::VASTART, getCurDebugLoc(),
MVT::Other, getRoot(),
- getValue(I.getOperand(1)),
- DAG.getSrcValue(I.getOperand(1))));
+ getValue(I.getArgOperand(0)),
+ DAG.getSrcValue(I.getArgOperand(0))));
}
void SelectionDAGBuilder::visitVAArg(const VAArgInst &I) {
+ const TargetData &TD = *TLI.getTargetData();
SDValue V = DAG.getVAArg(TLI.getValueType(I.getType()), getCurDebugLoc(),
getRoot(), getValue(I.getOperand(0)),
- DAG.getSrcValue(I.getOperand(0)));
+ DAG.getSrcValue(I.getOperand(0)),
+ TD.getABITypeAlignment(I.getType()));
setValue(&I, V);
DAG.setRoot(V.getValue(1));
}
void SelectionDAGBuilder::visitVAEnd(const CallInst &I) {
DAG.setRoot(DAG.getNode(ISD::VAEND, getCurDebugLoc(),
MVT::Other, getRoot(),
- getValue(I.getOperand(1)),
- DAG.getSrcValue(I.getOperand(1))));
+ getValue(I.getArgOperand(0)),
+ DAG.getSrcValue(I.getArgOperand(0))));
}
void SelectionDAGBuilder::visitVACopy(const CallInst &I) {
DAG.setRoot(DAG.getNode(ISD::VACOPY, getCurDebugLoc(),
MVT::Other, getRoot(),
- getValue(I.getOperand(1)),
- getValue(I.getOperand(2)),
- DAG.getSrcValue(I.getOperand(1)),
- DAG.getSrcValue(I.getOperand(2))));
+ getValue(I.getArgOperand(0)),
+ getValue(I.getArgOperand(1)),
+ DAG.getSrcValue(I.getArgOperand(0)),
+ DAG.getSrcValue(I.getArgOperand(1))));
}
/// TargetLowering::LowerCallTo - This is the default LowerCallTo
CallingConv::ID CallConv, bool isTailCall,
bool isReturnValueUsed,
SDValue Callee,
- ArgListTy &Args, SelectionDAG &DAG, DebugLoc dl) {
+ ArgListTy &Args, SelectionDAG &DAG,
+ DebugLoc dl) const {
// Handle all of the outgoing arguments.
SmallVector<ISD::OutputArg, 32> Outs;
+ SmallVector<SDValue, 32> OutVals;
for (unsigned i = 0, e = Args.size(); i != e; ++i) {
SmallVector<EVT, 4> ValueVTs;
ComputeValueVTs(*this, Args[i].Ty, ValueVTs);
for (unsigned j = 0; j != NumParts; ++j) {
// if it isn't first piece, alignment must be 1
- ISD::OutputArg MyFlags(Flags, Parts[j], i < NumFixedArgs);
+ ISD::OutputArg MyFlags(Flags, Parts[j].getValueType(),
+ i < NumFixedArgs);
if (NumParts > 1 && j == 0)
MyFlags.Flags.setSplit();
else if (j != 0)
MyFlags.Flags.setOrigAlign(1);
Outs.push_back(MyFlags);
+ OutVals.push_back(Parts[j]);
}
}
}
unsigned NumRegs = getNumRegisters(RetTy->getContext(), VT);
for (unsigned i = 0; i != NumRegs; ++i) {
ISD::InputArg MyFlags;
- MyFlags.VT = RegisterVT;
+ MyFlags.VT = RegisterVT.getSimpleVT();
MyFlags.Used = isReturnValueUsed;
if (RetSExt)
MyFlags.Flags.setSExt();
SmallVector<SDValue, 4> InVals;
Chain = LowerCall(Chain, Callee, CallConv, isVarArg, isTailCall,
- Outs, Ins, dl, DAG, InVals);
+ Outs, OutVals, Ins, dl, DAG, InVals);
// Verify that the target's LowerCall behaved as expected.
assert(Chain.getNode() && Chain.getValueType() == MVT::Other &&
DEBUG(for (unsigned i = 0, e = Ins.size(); i != e; ++i) {
assert(InVals[i].getNode() &&
"LowerCall emitted a null value!");
- assert(Ins[i].VT == InVals[i].getValueType() &&
+ assert(EVT(Ins[i].VT) == InVals[i].getValueType() &&
"LowerCall emitted a value with the wrong type!");
});
void TargetLowering::LowerOperationWrapper(SDNode *N,
SmallVectorImpl<SDValue> &Results,
- SelectionDAG &DAG) {
+ SelectionDAG &DAG) const {
SDValue Res = LowerOperation(SDValue(N, 0), DAG);
if (Res.getNode())
Results.push_back(Res);
}
-SDValue TargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) {
+SDValue TargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
llvm_unreachable("LowerOperation not implemented for this target!");
return SDValue();
}
void
SelectionDAGBuilder::CopyValueToVirtualRegister(const Value *V, unsigned Reg) {
- SDValue Op = getValue(V);
+ SDValue Op = getNonRegisterValue(V);
assert((Op.getOpcode() != ISD::CopyFromReg ||
cast<RegisterSDNode>(Op.getOperand(1))->getReg() != Reg) &&
"Copy from a reg to the same reg!");
// If this is the entry block, emit arguments.
const Function &F = *LLVMBB->getParent();
SelectionDAG &DAG = SDB->DAG;
- SDValue OldRoot = DAG.getRoot();
DebugLoc dl = SDB->getCurDebugLoc();
const TargetData *TD = TLI.getTargetData();
SmallVector<ISD::InputArg, 16> Ins;
// Check whether the function can return without sret-demotion.
- SmallVector<EVT, 4> OutVTs;
- SmallVector<ISD::ArgFlagsTy, 4> OutsFlags;
- getReturnInfo(F.getReturnType(), F.getAttributes().getRetAttributes(),
- OutVTs, OutsFlags, TLI);
- FunctionLoweringInfo &FLI = DAG.getFunctionLoweringInfo();
-
- FLI.CanLowerReturn = TLI.CanLowerReturn(F.getCallingConv(), F.isVarArg(),
- OutVTs, OutsFlags, DAG);
- if (!FLI.CanLowerReturn) {
+ SmallVector<ISD::OutputArg, 4> Outs;
+ GetReturnInfo(F.getReturnType(), F.getAttributes().getRetAttributes(),
+ Outs, TLI);
+
+ if (!FuncInfo->CanLowerReturn) {
// Put in an sret pointer parameter before all the other parameters.
SmallVector<EVT, 1> ValueVTs;
ComputeValueVTs(TLI, PointerType::getUnqual(F.getReturnType()), ValueVTs);
// or one register.
ISD::ArgFlagsTy Flags;
Flags.setSRet();
- EVT RegisterVT = TLI.getRegisterType(*CurDAG->getContext(), ValueVTs[0]);
+ EVT RegisterVT = TLI.getRegisterType(*DAG.getContext(), ValueVTs[0]);
ISD::InputArg RetArg(Flags, RegisterVT, true);
Ins.push_back(RetArg);
}
for (unsigned i = 0, e = Ins.size(); i != e; ++i) {
assert(InVals[i].getNode() &&
"LowerFormalArguments emitted a null value!");
- assert(Ins[i].VT == InVals[i].getValueType() &&
+ assert(EVT(Ins[i].VT) == InVals[i].getValueType() &&
"LowerFormalArguments emitted a value with the wrong type!");
}
});
// Set up the argument values.
unsigned i = 0;
Idx = 1;
- if (!FLI.CanLowerReturn) {
+ if (!FuncInfo->CanLowerReturn) {
// Create a virtual register for the sret pointer, and put in a copy
// from the sret argument into it.
SmallVector<EVT, 1> ValueVTs;
MachineFunction& MF = SDB->DAG.getMachineFunction();
MachineRegisterInfo& RegInfo = MF.getRegInfo();
unsigned SRetReg = RegInfo.createVirtualRegister(TLI.getRegClassFor(RegVT));
- FLI.DemoteRegister = SRetReg;
+ FuncInfo->DemoteRegister = SRetReg;
NewRoot = SDB->DAG.getCopyToReg(NewRoot, SDB->getCurDebugLoc(),
SRetReg, ArgValue);
DAG.setRoot(NewRoot);
SmallVector<EVT, 4> ValueVTs;
ComputeValueVTs(TLI, I->getType(), ValueVTs);
unsigned NumValues = ValueVTs.size();
+
+ // If this argument is unused then remember its value. It is used to generate
+ // debugging information.
+ if (I->use_empty() && NumValues)
+ SDB->setUnusedArgValue(I, InVals[i]);
+
for (unsigned Value = 0; Value != NumValues; ++Value) {
EVT VT = ValueVTs[Value];
EVT PartVT = TLI.getRegisterType(*CurDAG->getContext(), VT);
i += NumParts;
}
+ // Note down frame index for byval arguments.
+ if (I->hasByValAttr() && !ArgValues.empty())
+ if (FrameIndexSDNode *FI =
+ dyn_cast<FrameIndexSDNode>(ArgValues[0].getNode()))
+ FuncInfo->setByValArgumentFrameIndex(I, FI->getIndex());
+
if (!I->use_empty()) {
SDValue Res;
if (!ArgValues.empty())
/// the end.
///
void
-SelectionDAGISel::HandlePHINodesInSuccessorBlocks(const BasicBlock *LLVMBB) {
+SelectionDAGBuilder::HandlePHINodesInSuccessorBlocks(const BasicBlock *LLVMBB) {
const TerminatorInst *TI = LLVMBB->getTerminator();
SmallPtrSet<MachineBasicBlock *, 4> SuccsHandled;
for (unsigned succ = 0, e = TI->getNumSuccessors(); succ != e; ++succ) {
const BasicBlock *SuccBB = TI->getSuccessor(succ);
if (!isa<PHINode>(SuccBB->begin())) continue;
- MachineBasicBlock *SuccMBB = FuncInfo->MBBMap[SuccBB];
+ MachineBasicBlock *SuccMBB = FuncInfo.MBBMap[SuccBB];
// If this terminator has multiple identical successors (common for
// switches), only handle each succ once.
const Value *PHIOp = PN->getIncomingValueForBlock(LLVMBB);
if (const Constant *C = dyn_cast<Constant>(PHIOp)) {
- unsigned &RegOut = SDB->ConstantsOut[C];
+ unsigned &RegOut = ConstantsOut[C];
if (RegOut == 0) {
- RegOut = FuncInfo->CreateRegForValue(C);
- SDB->CopyValueToVirtualRegister(C, RegOut);
+ RegOut = FuncInfo.CreateRegs(C->getType());
+ CopyValueToVirtualRegister(C, RegOut);
}
Reg = RegOut;
} else {
- Reg = FuncInfo->ValueMap[PHIOp];
- if (Reg == 0) {
+ DenseMap<const Value *, unsigned>::iterator I =
+ FuncInfo.ValueMap.find(PHIOp);
+ if (I != FuncInfo.ValueMap.end())
+ Reg = I->second;
+ else {
assert(isa<AllocaInst>(PHIOp) &&
- FuncInfo
->StaticAllocaMap.count(cast<AllocaInst>(PHIOp)) &&
+ FuncInfo
.StaticAllocaMap.count(cast<AllocaInst>(PHIOp)) &&
"Didn't codegen value into a register!??");
- Reg = FuncInfo->CreateRegForValue(PHIOp);
- SDB->CopyValueToVirtualRegister(PHIOp, Reg);
+ Reg = FuncInfo.CreateRegs(PHIOp->getType());
+ CopyValueToVirtualRegister(PHIOp, Reg);
}
}
ComputeValueVTs(TLI, PN->getType(), ValueVTs);
for (unsigned vti = 0, vte = ValueVTs.size(); vti != vte; ++vti) {
EVT VT = ValueVTs[vti];
- unsigned NumRegisters = TLI.getNumRegisters(*CurDAG->getContext(), VT);
+ unsigned NumRegisters = TLI.getNumRegisters(*DAG.getContext(), VT);
for (unsigned i = 0, e = NumRegisters; i != e; ++i)
- SDB->PHINodesToUpdate.push_back(std::make_pair(MBBI++, Reg+i));
+ FuncInfo.PHINodesToUpdate.push_back(std::make_pair(MBBI++, Reg+i));
Reg += NumRegisters;
}
}
}
- SDB->ConstantsOut.clear();
-}
-
-/// This is the Fast-ISel version of HandlePHINodesInSuccessorBlocks. It only
-/// supports legal types, and it emits MachineInstrs directly instead of
-/// creating SelectionDAG nodes.
-///
-bool
-SelectionDAGISel::HandlePHINodesInSuccessorBlocksFast(const BasicBlock *LLVMBB,
- FastISel *F) {
- const TerminatorInst *TI = LLVMBB->getTerminator();
-
- SmallPtrSet<MachineBasicBlock *, 4> SuccsHandled;
- unsigned OrigNumPHINodesToUpdate = SDB->PHINodesToUpdate.size();
-
- // Check successor nodes' PHI nodes that expect a constant to be available
- // from this block.
- for (unsigned succ = 0, e = TI->getNumSuccessors(); succ != e; ++succ) {
- const BasicBlock *SuccBB = TI->getSuccessor(succ);
- if (!isa<PHINode>(SuccBB->begin())) continue;
- MachineBasicBlock *SuccMBB = FuncInfo->MBBMap[SuccBB];
-
- // If this terminator has multiple identical successors (common for
- // switches), only handle each succ once.
- if (!SuccsHandled.insert(SuccMBB)) continue;
-
- MachineBasicBlock::iterator MBBI = SuccMBB->begin();
-
- // At this point we know that there is a 1-1 correspondence between LLVM PHI
- // nodes and Machine PHI nodes, but the incoming operands have not been
- // emitted yet.
- for (BasicBlock::const_iterator I = SuccBB->begin();
- const PHINode *PN = dyn_cast<PHINode>(I); ++I) {
- // Ignore dead phi's.
- if (PN->use_empty()) continue;
-
- // Only handle legal types. Two interesting things to note here. First,
- // by bailing out early, we may leave behind some dead instructions,
- // since SelectionDAG's HandlePHINodesInSuccessorBlocks will insert its
- // own moves. Second, this check is necessary becuase FastISel doesn't
- // use CreateRegForValue to create registers, so it always creates
- // exactly one register for each non-void instruction.
- EVT VT = TLI.getValueType(PN->getType(), /*AllowUnknown=*/true);
- if (VT == MVT::Other || !TLI.isTypeLegal(VT)) {
- // Promote MVT::i1.
- if (VT == MVT::i1)
- VT = TLI.getTypeToTransformTo(*CurDAG->getContext(), VT);
- else {
- SDB->PHINodesToUpdate.resize(OrigNumPHINodesToUpdate);
- return false;
- }
- }
-
- const Value *PHIOp = PN->getIncomingValueForBlock(LLVMBB);
-
- unsigned Reg = F->getRegForValue(PHIOp);
- if (Reg == 0) {
- SDB->PHINodesToUpdate.resize(OrigNumPHINodesToUpdate);
- return false;
- }
- SDB->PHINodesToUpdate.push_back(std::make_pair(MBBI++, Reg));
- }
- }
-
- return true;
+ ConstantsOut.clear();
}
projects / oota-llvm.git / blobdiff