//
// The LLVM Compiler Infrastructure
//
-// This file was developed by Nate Begeman and is distributed under the
-// University of Illinois Open Source License. See LICENSE.TXT for details.
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
#define DEBUG_TYPE "dagcombine"
#include "llvm/CodeGen/SelectionDAG.h"
+#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/Target/TargetData.h"
+#include "llvm/Target/TargetFrameInfo.h"
#include "llvm/Target/TargetLowering.h"
#include "llvm/Target/TargetMachine.h"
#include "llvm/Target/TargetOptions.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/MathExtras.h"
#include <algorithm>
+#include <set>
using namespace llvm;
STATISTIC(NodesCombined , "Number of dag nodes combined");
void AddUsersToWorkList(SDNode *N) {
for (SDNode::use_iterator UI = N->use_begin(), UE = N->use_end();
UI != UE; ++UI)
- AddToWorkList(*UI);
+ AddToWorkList(UI->getUser());
}
- /// removeFromWorkList - remove all instances of N from the worklist.
- ///
- void removeFromWorkList(SDNode *N) {
- WorkList.erase(std::remove(WorkList.begin(), WorkList.end(), N),
- WorkList.end());
- }
-
/// visit - call the node-specific routine that knows how to fold each
/// particular type of node.
SDOperand visit(SDNode *N);
WorkList.push_back(N);
}
- SDOperand CombineTo(SDNode *N, const SDOperand *To, unsigned NumTo,
- bool AddTo = true) {
- assert(N->getNumValues() == NumTo && "Broken CombineTo call!");
- ++NodesCombined;
- DOUT << "\nReplacing.1 "; DEBUG(N->dump(&DAG));
- DOUT << "\nWith: "; DEBUG(To[0].Val->dump(&DAG));
- DOUT << " and " << NumTo-1 << " other values\n";
- std::vector<SDNode*> NowDead;
- DAG.ReplaceAllUsesWith(N, To, &NowDead);
-
- if (AddTo) {
- // Push the new nodes and any users onto the worklist
- for (unsigned i = 0, e = NumTo; i != e; ++i) {
- AddToWorkList(To[i].Val);
- AddUsersToWorkList(To[i].Val);
- }
- }
-
- // Nodes can be reintroduced into the worklist. Make sure we do not
- // process a node that has been replaced.
- removeFromWorkList(N);
- for (unsigned i = 0, e = NowDead.size(); i != e; ++i)
- removeFromWorkList(NowDead[i]);
-
- // Finally, since the node is now dead, remove it from the graph.
- DAG.DeleteNode(N);
- return SDOperand(N, 0);
+ /// removeFromWorkList - remove all instances of N from the worklist.
+ ///
+ void removeFromWorkList(SDNode *N) {
+ WorkList.erase(std::remove(WorkList.begin(), WorkList.end(), N),
+ WorkList.end());
}
+ SDOperand CombineTo(SDNode *N, const SDOperand *To, unsigned NumTo,
+ bool AddTo = true);
+
SDOperand CombineTo(SDNode *N, SDOperand Res, bool AddTo = true) {
return CombineTo(N, &Res, 1, AddTo);
}
SDOperand To[] = { Res0, Res1 };
return CombineTo(N, To, 2, AddTo);
}
+
private:
/// SimplifyDemandedBits - Check the specified integer node value to see if
/// it can be simplified or if things it uses can be simplified by bit
/// propagation. If so, return true.
- bool SimplifyDemandedBits(SDOperand Op, uint64_t Demanded = ~0ULL) {
- TargetLowering::TargetLoweringOpt TLO(DAG, AfterLegalize);
- uint64_t KnownZero, KnownOne;
- Demanded &= MVT::getIntVTBitMask(Op.getValueType());
- if (!TLI.SimplifyDemandedBits(Op, Demanded, KnownZero, KnownOne, TLO))
- return false;
-
- // Revisit the node.
- AddToWorkList(Op.Val);
-
- // Replace the old value with the new one.
- ++NodesCombined;
- DOUT << "\nReplacing.2 "; DEBUG(TLO.Old.Val->dump(&DAG));
- DOUT << "\nWith: "; DEBUG(TLO.New.Val->dump(&DAG));
- DOUT << '\n';
-
- std::vector<SDNode*> NowDead;
- DAG.ReplaceAllUsesOfValueWith(TLO.Old, TLO.New, &NowDead);
-
- // Push the new node and any (possibly new) users onto the worklist.
- AddToWorkList(TLO.New.Val);
- AddUsersToWorkList(TLO.New.Val);
-
- // Nodes can end up on the worklist more than once. Make sure we do
- // not process a node that has been replaced.
- for (unsigned i = 0, e = NowDead.size(); i != e; ++i)
- removeFromWorkList(NowDead[i]);
-
- // Finally, if the node is now dead, remove it from the graph. The node
- // may not be dead if the replacement process recursively simplified to
- // something else needing this node.
- if (TLO.Old.Val->use_empty()) {
- removeFromWorkList(TLO.Old.Val);
-
- // If the operands of this node are only used by the node, they will now
- // be dead. Make sure to visit them first to delete dead nodes early.
- for (unsigned i = 0, e = TLO.Old.Val->getNumOperands(); i != e; ++i)
- if (TLO.Old.Val->getOperand(i).Val->hasOneUse())
- AddToWorkList(TLO.Old.Val->getOperand(i).Val);
-
- DAG.DeleteNode(TLO.Old.Val);
- }
- return true;
+ bool SimplifyDemandedBits(SDOperand Op) {
+ APInt Demanded = APInt::getAllOnesValue(Op.getValueSizeInBits());
+ return SimplifyDemandedBits(Op, Demanded);
}
+ bool SimplifyDemandedBits(SDOperand Op, const APInt &Demanded);
+
bool CombineToPreIndexedLoadStore(SDNode *N);
bool CombineToPostIndexedLoadStore(SDNode *N);
// otherwise - N should be replaced by the returned Operand.
//
SDOperand visitTokenFactor(SDNode *N);
+ SDOperand visitMERGE_VALUES(SDNode *N);
SDOperand visitADD(SDNode *N);
SDOperand visitSUB(SDNode *N);
SDOperand visitADDC(SDNode *N);
SDOperand visitSIGN_EXTEND_INREG(SDNode *N);
SDOperand visitTRUNCATE(SDNode *N);
SDOperand visitBIT_CONVERT(SDNode *N);
+ SDOperand visitBUILD_PAIR(SDNode *N);
SDOperand visitFADD(SDNode *N);
SDOperand visitFSUB(SDNode *N);
SDOperand visitFMUL(SDNode *N);
SDOperand SimplifySelectCC(SDOperand N0, SDOperand N1, SDOperand N2,
SDOperand N3, ISD::CondCode CC,
bool NotExtCompare = false);
- SDOperand SimplifySetCC(MVT::ValueType VT, SDOperand N0, SDOperand N1,
+ SDOperand SimplifySetCC(MVT VT, SDOperand N0, SDOperand N1,
ISD::CondCode Cond, bool foldBooleans = true);
- bool SimplifyNodeWithTwoResults(SDNode *N, unsigned LoOp, unsigned HiOp);
- SDOperand ConstantFoldBIT_CONVERTofBUILD_VECTOR(SDNode *, MVT::ValueType);
+ SDOperand SimplifyNodeWithTwoResults(SDNode *N, unsigned LoOp,
+ unsigned HiOp);
+ SDOperand CombineConsecutiveLoads(SDNode *N, MVT VT);
+ SDOperand ConstantFoldBIT_CONVERTofBUILD_VECTOR(SDNode *, MVT);
SDOperand BuildSDIV(SDNode *N);
SDOperand BuildUDIV(SDNode *N);
SDNode *MatchRotate(SDOperand LHS, SDOperand RHS);
SDOperand ReduceLoadWidth(SDNode *N);
- SDOperand GetDemandedBits(SDOperand V, uint64_t Mask);
+ SDOperand GetDemandedBits(SDOperand V, const APInt &Mask);
/// GatherAllAliases - Walk up chain skipping non-aliasing memory nodes,
/// looking for aliasing nodes and adding them to the Aliases vector.
};
}
+
+namespace {
+/// WorkListRemover - This class is a DAGUpdateListener that removes any deleted
+/// nodes from the worklist.
+class VISIBILITY_HIDDEN WorkListRemover :
+ public SelectionDAG::DAGUpdateListener {
+ DAGCombiner &DC;
+public:
+ explicit WorkListRemover(DAGCombiner &dc) : DC(dc) {}
+
+ virtual void NodeDeleted(SDNode *N, SDNode *E) {
+ DC.removeFromWorkList(N);
+ }
+
+ virtual void NodeUpdated(SDNode *N) {
+ // Ignore updates.
+ }
+};
+}
+
//===----------------------------------------------------------------------===//
// TargetLowering::DAGCombinerInfo implementation
//===----------------------------------------------------------------------===//
/// isNegatibleForFree - Return 1 if we can compute the negated form of the
/// specified expression for the same cost as the expression itself, or 2 if we
/// can compute the negated form more cheaply than the expression itself.
-static char isNegatibleForFree(SDOperand Op, unsigned Depth = 0) {
+static char isNegatibleForFree(SDOperand Op, bool AfterLegalize,
+ unsigned Depth = 0) {
// No compile time optimizations on this type.
if (Op.getValueType() == MVT::ppcf128)
return 0;
switch (Op.getOpcode()) {
default: return false;
case ISD::ConstantFP:
- return 1;
+ // Don't invert constant FP values after legalize. The negated constant
+ // isn't necessarily legal.
+ return AfterLegalize ? 0 : 1;
case ISD::FADD:
// FIXME: determine better conditions for this xform.
if (!UnsafeFPMath) return 0;
// -(A+B) -> -A - B
- if (char V = isNegatibleForFree(Op.getOperand(0), Depth+1))
+ if (char V = isNegatibleForFree(Op.getOperand(0), AfterLegalize, Depth+1))
return V;
// -(A+B) -> -B - A
- return isNegatibleForFree(Op.getOperand(1), Depth+1);
+ return isNegatibleForFree(Op.getOperand(1), AfterLegalize, Depth+1);
case ISD::FSUB:
// We can't turn -(A-B) into B-A when we honor signed zeros.
if (!UnsafeFPMath) return 0;
if (HonorSignDependentRoundingFPMath()) return 0;
// -(X*Y) -> (-X * Y) or (X*-Y)
- if (char V = isNegatibleForFree(Op.getOperand(0), Depth+1))
+ if (char V = isNegatibleForFree(Op.getOperand(0), AfterLegalize, Depth+1))
return V;
- return isNegatibleForFree(Op.getOperand(1), Depth+1);
+ return isNegatibleForFree(Op.getOperand(1), AfterLegalize, Depth+1);
case ISD::FP_EXTEND:
case ISD::FP_ROUND:
case ISD::FSIN:
- return isNegatibleForFree(Op.getOperand(0), Depth+1);
+ return isNegatibleForFree(Op.getOperand(0), AfterLegalize, Depth+1);
}
}
/// GetNegatedExpression - If isNegatibleForFree returns true, this function
/// returns the newly negated expression.
static SDOperand GetNegatedExpression(SDOperand Op, SelectionDAG &DAG,
- unsigned Depth = 0) {
+ bool AfterLegalize, unsigned Depth = 0) {
// fneg is removable even if it has multiple uses.
if (Op.getOpcode() == ISD::FNEG) return Op.getOperand(0);
assert(UnsafeFPMath);
// -(A+B) -> -A - B
- if (isNegatibleForFree(Op.getOperand(0), Depth+1))
+ if (isNegatibleForFree(Op.getOperand(0), AfterLegalize, Depth+1))
return DAG.getNode(ISD::FSUB, Op.getValueType(),
- GetNegatedExpression(Op.getOperand(0), DAG, Depth+1),
+ GetNegatedExpression(Op.getOperand(0), DAG,
+ AfterLegalize, Depth+1),
Op.getOperand(1));
// -(A+B) -> -B - A
return DAG.getNode(ISD::FSUB, Op.getValueType(),
- GetNegatedExpression(Op.getOperand(1), DAG, Depth+1),
+ GetNegatedExpression(Op.getOperand(1), DAG,
+ AfterLegalize, Depth+1),
Op.getOperand(0));
case ISD::FSUB:
// We can't turn -(A-B) into B-A when we honor signed zeros.
assert(!HonorSignDependentRoundingFPMath());
// -(X*Y) -> -X * Y
- if (isNegatibleForFree(Op.getOperand(0), Depth+1))
+ if (isNegatibleForFree(Op.getOperand(0), AfterLegalize, Depth+1))
return DAG.getNode(Op.getOpcode(), Op.getValueType(),
- GetNegatedExpression(Op.getOperand(0), DAG, Depth+1),
+ GetNegatedExpression(Op.getOperand(0), DAG,
+ AfterLegalize, Depth+1),
Op.getOperand(1));
// -(X*Y) -> X * -Y
return DAG.getNode(Op.getOpcode(), Op.getValueType(),
Op.getOperand(0),
- GetNegatedExpression(Op.getOperand(1), DAG, Depth+1));
+ GetNegatedExpression(Op.getOperand(1), DAG,
+ AfterLegalize, Depth+1));
case ISD::FP_EXTEND:
- case ISD::FP_ROUND:
case ISD::FSIN:
return DAG.getNode(Op.getOpcode(), Op.getValueType(),
- GetNegatedExpression(Op.getOperand(0), DAG, Depth+1));
+ GetNegatedExpression(Op.getOperand(0), DAG,
+ AfterLegalize, Depth+1));
+ case ISD::FP_ROUND:
+ return DAG.getNode(ISD::FP_ROUND, Op.getValueType(),
+ GetNegatedExpression(Op.getOperand(0), DAG,
+ AfterLegalize, Depth+1),
+ Op.getOperand(1));
}
}
if (N.getOpcode() == ISD::SELECT_CC &&
N.getOperand(2).getOpcode() == ISD::Constant &&
N.getOperand(3).getOpcode() == ISD::Constant &&
- cast<ConstantSDNode>(N.getOperand(2))->getValue() == 1 &&
+ cast<ConstantSDNode>(N.getOperand(2))->getAPIntValue() == 1 &&
cast<ConstantSDNode>(N.getOperand(3))->isNullValue()) {
LHS = N.getOperand(0);
RHS = N.getOperand(1);
}
SDOperand DAGCombiner::ReassociateOps(unsigned Opc, SDOperand N0, SDOperand N1){
- MVT::ValueType VT = N0.getValueType();
+ MVT VT = N0.getValueType();
// reassoc. (op (op x, c1), y) -> (op (op x, y), c1) iff x+c1 has one use
// reassoc. (op (op x, c1), c2) -> (op x, (op c1, c2))
if (N0.getOpcode() == Opc && isa<ConstantSDNode>(N0.getOperand(1))) {
return SDOperand();
}
+SDOperand DAGCombiner::CombineTo(SDNode *N, const SDOperand *To, unsigned NumTo,
+ bool AddTo) {
+ assert(N->getNumValues() == NumTo && "Broken CombineTo call!");
+ ++NodesCombined;
+ DOUT << "\nReplacing.1 "; DEBUG(N->dump(&DAG));
+ DOUT << "\nWith: "; DEBUG(To[0].Val->dump(&DAG));
+ DOUT << " and " << NumTo-1 << " other values\n";
+ WorkListRemover DeadNodes(*this);
+ DAG.ReplaceAllUsesWith(N, To, &DeadNodes);
+
+ if (AddTo) {
+ // Push the new nodes and any users onto the worklist
+ for (unsigned i = 0, e = NumTo; i != e; ++i) {
+ AddToWorkList(To[i].Val);
+ AddUsersToWorkList(To[i].Val);
+ }
+ }
+
+ // Nodes can be reintroduced into the worklist. Make sure we do not
+ // process a node that has been replaced.
+ removeFromWorkList(N);
+
+ // Finally, since the node is now dead, remove it from the graph.
+ DAG.DeleteNode(N);
+ return SDOperand(N, 0);
+}
+
+/// SimplifyDemandedBits - Check the specified integer node value to see if
+/// it can be simplified or if things it uses can be simplified by bit
+/// propagation. If so, return true.
+bool DAGCombiner::SimplifyDemandedBits(SDOperand Op, const APInt &Demanded) {
+ TargetLowering::TargetLoweringOpt TLO(DAG, AfterLegalize);
+ APInt KnownZero, KnownOne;
+ if (!TLI.SimplifyDemandedBits(Op, Demanded, KnownZero, KnownOne, TLO))
+ return false;
+
+ // Revisit the node.
+ AddToWorkList(Op.Val);
+
+ // Replace the old value with the new one.
+ ++NodesCombined;
+ DOUT << "\nReplacing.2 "; DEBUG(TLO.Old.Val->dump(&DAG));
+ DOUT << "\nWith: "; DEBUG(TLO.New.Val->dump(&DAG));
+ DOUT << '\n';
+
+ // Replace all uses. If any nodes become isomorphic to other nodes and
+ // are deleted, make sure to remove them from our worklist.
+ WorkListRemover DeadNodes(*this);
+ DAG.ReplaceAllUsesOfValueWith(TLO.Old, TLO.New, &DeadNodes);
+
+ // Push the new node and any (possibly new) users onto the worklist.
+ AddToWorkList(TLO.New.Val);
+ AddUsersToWorkList(TLO.New.Val);
+
+ // Finally, if the node is now dead, remove it from the graph. The node
+ // may not be dead if the replacement process recursively simplified to
+ // something else needing this node.
+ if (TLO.Old.Val->use_empty()) {
+ removeFromWorkList(TLO.Old.Val);
+
+ // If the operands of this node are only used by the node, they will now
+ // be dead. Make sure to visit them first to delete dead nodes early.
+ for (unsigned i = 0, e = TLO.Old.Val->getNumOperands(); i != e; ++i)
+ if (TLO.Old.Val->getOperand(i).Val->hasOneUse())
+ AddToWorkList(TLO.Old.Val->getOperand(i).Val);
+
+ DAG.DeleteNode(TLO.Old.Val);
+ }
+ return true;
+}
+
//===----------------------------------------------------------------------===//
// Main DAG Combiner implementation
//===----------------------------------------------------------------------===//
SDOperand RV = combine(N);
- if (RV.Val) {
- ++NodesCombined;
- // If we get back the same node we passed in, rather than a new node or
- // zero, we know that the node must have defined multiple values and
- // CombineTo was used. Since CombineTo takes care of the worklist
- // mechanics for us, we have no work to do in this case.
- if (RV.Val != N) {
- assert(N->getOpcode() != ISD::DELETED_NODE &&
- RV.Val->getOpcode() != ISD::DELETED_NODE &&
- "Node was deleted but visit returned new node!");
-
- DOUT << "\nReplacing.3 "; DEBUG(N->dump(&DAG));
- DOUT << "\nWith: "; DEBUG(RV.Val->dump(&DAG));
- DOUT << '\n';
- std::vector<SDNode*> NowDead;
- if (N->getNumValues() == RV.Val->getNumValues())
- DAG.ReplaceAllUsesWith(N, RV.Val, &NowDead);
- else {
- assert(N->getValueType(0) == RV.getValueType() && "Type mismatch");
- SDOperand OpV = RV;
- DAG.ReplaceAllUsesWith(N, &OpV, &NowDead);
- }
-
- // Push the new node and any users onto the worklist
- AddToWorkList(RV.Val);
- AddUsersToWorkList(RV.Val);
-
- // Nodes can be reintroduced into the worklist. Make sure we do not
- // process a node that has been replaced.
- removeFromWorkList(N);
- for (unsigned i = 0, e = NowDead.size(); i != e; ++i)
- removeFromWorkList(NowDead[i]);
-
- // Finally, since the node is now dead, remove it from the graph.
- DAG.DeleteNode(N);
- }
+ if (RV.Val == 0)
+ continue;
+
+ ++NodesCombined;
+
+ // If we get back the same node we passed in, rather than a new node or
+ // zero, we know that the node must have defined multiple values and
+ // CombineTo was used. Since CombineTo takes care of the worklist
+ // mechanics for us, we have no work to do in this case.
+ if (RV.Val == N)
+ continue;
+
+ assert(N->getOpcode() != ISD::DELETED_NODE &&
+ RV.Val->getOpcode() != ISD::DELETED_NODE &&
+ "Node was deleted but visit returned new node!");
+
+ DOUT << "\nReplacing.3 "; DEBUG(N->dump(&DAG));
+ DOUT << "\nWith: "; DEBUG(RV.Val->dump(&DAG));
+ DOUT << '\n';
+ WorkListRemover DeadNodes(*this);
+ if (N->getNumValues() == RV.Val->getNumValues())
+ DAG.ReplaceAllUsesWith(N, RV.Val, &DeadNodes);
+ else {
+ assert(N->getValueType(0) == RV.getValueType() &&
+ N->getNumValues() == 1 && "Type mismatch");
+ SDOperand OpV = RV;
+ DAG.ReplaceAllUsesWith(N, &OpV, &DeadNodes);
}
+
+ // Push the new node and any users onto the worklist
+ AddToWorkList(RV.Val);
+ AddUsersToWorkList(RV.Val);
+
+ // Add any uses of the old node to the worklist in case this node is the
+ // last one that uses them. They may become dead after this node is
+ // deleted.
+ for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i)
+ AddToWorkList(N->getOperand(i).Val);
+
+ // Nodes can be reintroduced into the worklist. Make sure we do not
+ // process a node that has been replaced.
+ removeFromWorkList(N);
+
+ // Finally, since the node is now dead, remove it from the graph.
+ DAG.DeleteNode(N);
}
// If the root changed (e.g. it was a dead load, update the root).
switch(N->getOpcode()) {
default: break;
case ISD::TokenFactor: return visitTokenFactor(N);
+ case ISD::MERGE_VALUES: return visitMERGE_VALUES(N);
case ISD::ADD: return visitADD(N);
case ISD::SUB: return visitSUB(N);
case ISD::ADDC: return visitADDC(N);
case ISD::SIGN_EXTEND_INREG: return visitSIGN_EXTEND_INREG(N);
case ISD::TRUNCATE: return visitTRUNCATE(N);
case ISD::BIT_CONVERT: return visitBIT_CONVERT(N);
+ case ISD::BUILD_PAIR: return visitBUILD_PAIR(N);
case ISD::FADD: return visitFADD(N);
case ISD::FSUB: return visitFSUB(N);
case ISD::FMUL: return visitFMUL(N);
}
}
+ // If N is a commutative binary node, try commuting it to enable more
+ // sdisel CSE.
+ if (RV.Val == 0 &&
+ SelectionDAG::isCommutativeBinOp(N->getOpcode()) &&
+ N->getNumValues() == 1) {
+ SDOperand N0 = N->getOperand(0);
+ SDOperand N1 = N->getOperand(1);
+ // Constant operands are canonicalized to RHS.
+ if (isa<ConstantSDNode>(N0) || !isa<ConstantSDNode>(N1)) {
+ SDOperand Ops[] = { N1, N0 };
+ SDNode *CSENode = DAG.getNodeIfExists(N->getOpcode(), N->getVTList(),
+ Ops, 2);
+ if (CSENode)
+ return SDOperand(CSENode, 0);
+ }
+ }
+
return RV;
}
// If we've change things around then replace token factor.
if (Changed) {
- if (Ops.size() == 0) {
+ if (Ops.empty()) {
// The entry token is the only possible outcome.
Result = DAG.getEntryNode();
} else {
return Result;
}
+/// MERGE_VALUES can always be eliminated.
+SDOperand DAGCombiner::visitMERGE_VALUES(SDNode *N) {
+ WorkListRemover DeadNodes(*this);
+ for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i)
+ DAG.ReplaceAllUsesOfValueWith(SDOperand(N, i), N->getOperand(i),
+ &DeadNodes);
+ removeFromWorkList(N);
+ DAG.DeleteNode(N);
+ return SDOperand(N, 0); // Return N so it doesn't get rechecked!
+}
+
+
static
SDOperand combineShlAddConstant(SDOperand N0, SDOperand N1, SelectionDAG &DAG) {
- MVT::ValueType VT = N0.getValueType();
+ MVT VT = N0.getValueType();
SDOperand N00 = N0.getOperand(0);
SDOperand N01 = N0.getOperand(1);
ConstantSDNode *N01C = dyn_cast<ConstantSDNode>(N01);
static
SDOperand combineSelectAndUse(SDNode *N, SDOperand Slct, SDOperand OtherOp,
SelectionDAG &DAG) {
- MVT::ValueType VT = N->getValueType(0);
+ MVT VT = N->getValueType(0);
unsigned Opc = N->getOpcode();
bool isSlctCC = Slct.getOpcode() == ISD::SELECT_CC;
SDOperand LHS = isSlctCC ? Slct.getOperand(2) : Slct.getOperand(1);
RHS.getOpcode() == ISD::Constant &&
cast<ConstantSDNode>(RHS)->isNullValue()) {
std::swap(LHS, RHS);
- bool isInt = MVT::isInteger(isSlctCC ? Slct.getOperand(0).getValueType()
- : Slct.getOperand(0).getOperand(0).getValueType());
+ SDOperand Op0 = Slct.getOperand(0);
+ bool isInt = (isSlctCC ? Op0.getValueType() :
+ Op0.getOperand(0).getValueType()).isInteger();
CC = ISD::getSetCCInverse(CC, isInt);
DoXform = true;
InvCC = true;
SDOperand N1 = N->getOperand(1);
ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0);
ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1);
- MVT::ValueType VT = N0.getValueType();
+ MVT VT = N0.getValueType();
// fold vector ops
- if (MVT::isVector(VT)) {
+ if (VT.isVector()) {
SDOperand FoldedVOp = SimplifyVBinOp(N);
if (FoldedVOp.Val) return FoldedVOp;
}
return N1;
// fold (add c1, c2) -> c1+c2
if (N0C && N1C)
- return DAG.getNode(ISD::ADD, VT, N0, N1);
+ return DAG.getConstant(N0C->getAPIntValue() + N1C->getAPIntValue(), VT);
// canonicalize constant to RHS
if (N0C && !N1C)
return DAG.getNode(ISD::ADD, VT, N1, N0);
if (N1C && N0.getOpcode() == ISD::SUB)
if (ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0.getOperand(0)))
return DAG.getNode(ISD::SUB, VT,
- DAG.getConstant(N1C->getValue()+N0C->getValue(), VT),
+ DAG.getConstant(N1C->getAPIntValue()+
+ N0C->getAPIntValue(), VT),
N0.getOperand(1));
// reassociate add
SDOperand RADD = ReassociateOps(ISD::ADD, N0, N1);
if (N1.getOpcode() == ISD::SUB && N0 == N1.getOperand(1))
return N1.getOperand(0);
- if (!MVT::isVector(VT) && SimplifyDemandedBits(SDOperand(N, 0)))
+ if (!VT.isVector() && SimplifyDemandedBits(SDOperand(N, 0)))
return SDOperand(N, 0);
// fold (a+b) -> (a|b) iff a and b share no bits.
- if (MVT::isInteger(VT) && !MVT::isVector(VT)) {
- uint64_t LHSZero, LHSOne;
- uint64_t RHSZero, RHSOne;
- uint64_t Mask = MVT::getIntVTBitMask(VT);
+ if (VT.isInteger() && !VT.isVector()) {
+ APInt LHSZero, LHSOne;
+ APInt RHSZero, RHSOne;
+ APInt Mask = APInt::getAllOnesValue(VT.getSizeInBits());
DAG.ComputeMaskedBits(N0, Mask, LHSZero, LHSOne);
- if (LHSZero) {
+ if (LHSZero.getBoolValue()) {
DAG.ComputeMaskedBits(N1, Mask, RHSZero, RHSOne);
// If all possibly-set bits on the LHS are clear on the RHS, return an OR.
SDOperand N1 = N->getOperand(1);
ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0);
ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1);
- MVT::ValueType VT = N0.getValueType();
+ MVT VT = N0.getValueType();
// If the flag result is dead, turn this into an ADD.
if (N->hasNUsesOfValue(0, 1))
DAG.getNode(ISD::CARRY_FALSE, MVT::Flag));
// canonicalize constant to RHS.
- if (N0C && !N1C) {
- SDOperand Ops[] = { N1, N0 };
- return DAG.getNode(ISD::ADDC, N->getVTList(), Ops, 2);
- }
+ if (N0C && !N1C)
+ return DAG.getNode(ISD::ADDC, N->getVTList(), N1, N0);
// fold (addc x, 0) -> x + no carry out
if (N1C && N1C->isNullValue())
return CombineTo(N, N0, DAG.getNode(ISD::CARRY_FALSE, MVT::Flag));
// fold (addc a, b) -> (or a, b), CARRY_FALSE iff a and b share no bits.
- uint64_t LHSZero, LHSOne;
- uint64_t RHSZero, RHSOne;
- uint64_t Mask = MVT::getIntVTBitMask(VT);
+ APInt LHSZero, LHSOne;
+ APInt RHSZero, RHSOne;
+ APInt Mask = APInt::getAllOnesValue(VT.getSizeInBits());
DAG.ComputeMaskedBits(N0, Mask, LHSZero, LHSOne);
- if (LHSZero) {
+ if (LHSZero.getBoolValue()) {
DAG.ComputeMaskedBits(N1, Mask, RHSZero, RHSOne);
// If all possibly-set bits on the LHS are clear on the RHS, return an OR.
SDOperand CarryIn = N->getOperand(2);
ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0);
ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1);
- //MVT::ValueType VT = N0.getValueType();
+ //MVT VT = N0.getValueType();
// canonicalize constant to RHS
- if (N0C && !N1C) {
- SDOperand Ops[] = { N1, N0, CarryIn };
- return DAG.getNode(ISD::ADDE, N->getVTList(), Ops, 3);
- }
+ if (N0C && !N1C)
+ return DAG.getNode(ISD::ADDE, N->getVTList(), N1, N0, CarryIn);
// fold (adde x, y, false) -> (addc x, y)
- if (CarryIn.getOpcode() == ISD::CARRY_FALSE) {
- SDOperand Ops[] = { N1, N0 };
- return DAG.getNode(ISD::ADDC, N->getVTList(), Ops, 2);
- }
+ if (CarryIn.getOpcode() == ISD::CARRY_FALSE)
+ return DAG.getNode(ISD::ADDC, N->getVTList(), N1, N0);
return SDOperand();
}
SDOperand N1 = N->getOperand(1);
ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0.Val);
ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1.Val);
- MVT::ValueType VT = N0.getValueType();
+ MVT VT = N0.getValueType();
// fold vector ops
- if (MVT::isVector(VT)) {
+ if (VT.isVector()) {
SDOperand FoldedVOp = SimplifyVBinOp(N);
if (FoldedVOp.Val) return FoldedVOp;
}
return DAG.getNode(ISD::SUB, VT, N0, N1);
// fold (sub x, c) -> (add x, -c)
if (N1C)
- return DAG.getNode(ISD::ADD, VT, N0, DAG.getConstant(-N1C->getValue(), VT));
+ return DAG.getNode(ISD::ADD, VT, N0,
+ DAG.getConstant(-N1C->getAPIntValue(), VT));
// fold (A+B)-A -> B
if (N0.getOpcode() == ISD::ADD && N0.getOperand(0) == N1)
return N0.getOperand(1);
SDOperand N1 = N->getOperand(1);
ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0);
ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1);
- MVT::ValueType VT = N0.getValueType();
+ MVT VT = N0.getValueType();
// fold vector ops
- if (MVT::isVector(VT)) {
+ if (VT.isVector()) {
SDOperand FoldedVOp = SimplifyVBinOp(N);
if (FoldedVOp.Val) return FoldedVOp;
}
if (N1C && N1C->isAllOnesValue())
return DAG.getNode(ISD::SUB, VT, DAG.getConstant(0, VT), N0);
// fold (mul x, (1 << c)) -> x << c
- if (N1C && isPowerOf2_64(N1C->getValue()))
+ if (N1C && N1C->getAPIntValue().isPowerOf2())
return DAG.getNode(ISD::SHL, VT, N0,
- DAG.getConstant(Log2_64(N1C->getValue()),
+ DAG.getConstant(N1C->getAPIntValue().logBase2(),
TLI.getShiftAmountTy()));
// fold (mul x, -(1 << c)) -> -(x << c) or (-x) << c
if (N1C && isPowerOf2_64(-N1C->getSignExtended())) {
SDOperand N1 = N->getOperand(1);
ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0.Val);
ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1.Val);
- MVT::ValueType VT = N->getValueType(0);
+ MVT VT = N->getValueType(0);
// fold vector ops
- if (MVT::isVector(VT)) {
+ if (VT.isVector()) {
SDOperand FoldedVOp = SimplifyVBinOp(N);
if (FoldedVOp.Val) return FoldedVOp;
}
return DAG.getNode(ISD::SUB, VT, DAG.getConstant(0, VT), N0);
// If we know the sign bits of both operands are zero, strength reduce to a
// udiv instead. Handles (X&15) /s 4 -> X&15 >> 2
- uint64_t SignBit = 1ULL << (MVT::getSizeInBits(VT)-1);
- if (DAG.MaskedValueIsZero(N1, SignBit) &&
- DAG.MaskedValueIsZero(N0, SignBit))
- return DAG.getNode(ISD::UDIV, N1.getValueType(), N0, N1);
+ if (!VT.isVector()) {
+ if (DAG.SignBitIsZero(N1) && DAG.SignBitIsZero(N0))
+ return DAG.getNode(ISD::UDIV, N1.getValueType(), N0, N1);
+ }
// fold (sdiv X, pow2) -> simple ops after legalize
- if (N1C && N1C->getValue() && !TLI.isIntDivCheap() &&
+ if (N1C && !N1C->isNullValue() && !TLI.isIntDivCheap() &&
(isPowerOf2_64(N1C->getSignExtended()) ||
isPowerOf2_64(-N1C->getSignExtended()))) {
// If dividing by powers of two is cheap, then don't perform the following
unsigned lg2 = Log2_64(abs2);
// Splat the sign bit into the register
SDOperand SGN = DAG.getNode(ISD::SRA, VT, N0,
- DAG.getConstant(MVT::getSizeInBits(VT)-1,
+ DAG.getConstant(VT.getSizeInBits()-1,
TLI.getShiftAmountTy()));
AddToWorkList(SGN.Val);
// Add (N0 < 0) ? abs2 - 1 : 0;
SDOperand SRL = DAG.getNode(ISD::SRL, VT, SGN,
- DAG.getConstant(MVT::getSizeInBits(VT)-lg2,
+ DAG.getConstant(VT.getSizeInBits()-lg2,
TLI.getShiftAmountTy()));
SDOperand ADD = DAG.getNode(ISD::ADD, VT, N0, SRL);
AddToWorkList(SRL.Val);
SDOperand N1 = N->getOperand(1);
ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0.Val);
ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1.Val);
- MVT::ValueType VT = N->getValueType(0);
+ MVT VT = N->getValueType(0);
// fold vector ops
- if (MVT::isVector(VT)) {
+ if (VT.isVector()) {
SDOperand FoldedVOp = SimplifyVBinOp(N);
if (FoldedVOp.Val) return FoldedVOp;
}
if (N0C && N1C && !N1C->isNullValue())
return DAG.getNode(ISD::UDIV, VT, N0, N1);
// fold (udiv x, (1 << c)) -> x >>u c
- if (N1C && isPowerOf2_64(N1C->getValue()))
+ if (N1C && N1C->getAPIntValue().isPowerOf2())
return DAG.getNode(ISD::SRL, VT, N0,
- DAG.getConstant(Log2_64(N1C->getValue()),
+ DAG.getConstant(N1C->getAPIntValue().logBase2(),
TLI.getShiftAmountTy()));
// fold (udiv x, (shl c, y)) -> x >>u (log2(c)+y) iff c is power of 2
if (N1.getOpcode() == ISD::SHL) {
if (ConstantSDNode *SHC = dyn_cast<ConstantSDNode>(N1.getOperand(0))) {
- if (isPowerOf2_64(SHC->getValue())) {
- MVT::ValueType ADDVT = N1.getOperand(1).getValueType();
+ if (SHC->getAPIntValue().isPowerOf2()) {
+ MVT ADDVT = N1.getOperand(1).getValueType();
SDOperand Add = DAG.getNode(ISD::ADD, ADDVT, N1.getOperand(1),
- DAG.getConstant(Log2_64(SHC->getValue()),
+ DAG.getConstant(SHC->getAPIntValue()
+ .logBase2(),
ADDVT));
AddToWorkList(Add.Val);
return DAG.getNode(ISD::SRL, VT, N0, Add);
}
}
// fold (udiv x, c) -> alternate
- if (N1C && N1C->getValue() && !TLI.isIntDivCheap()) {
+ if (N1C && !N1C->isNullValue() && !TLI.isIntDivCheap()) {
SDOperand Op = BuildUDIV(N);
if (Op.Val) return Op;
}
SDOperand N1 = N->getOperand(1);
ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0);
ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1);
- MVT::ValueType VT = N->getValueType(0);
+ MVT VT = N->getValueType(0);
// fold (srem c1, c2) -> c1%c2
if (N0C && N1C && !N1C->isNullValue())
return DAG.getNode(ISD::SREM, VT, N0, N1);
// If we know the sign bits of both operands are zero, strength reduce to a
// urem instead. Handles (X & 0x0FFFFFFF) %s 16 -> X&15
- uint64_t SignBit = 1ULL << (MVT::getSizeInBits(VT)-1);
- if (DAG.MaskedValueIsZero(N1, SignBit) &&
- DAG.MaskedValueIsZero(N0, SignBit))
- return DAG.getNode(ISD::UREM, VT, N0, N1);
+ if (!VT.isVector()) {
+ if (DAG.SignBitIsZero(N1) && DAG.SignBitIsZero(N0))
+ return DAG.getNode(ISD::UREM, VT, N0, N1);
+ }
// If X/C can be simplified by the division-by-constant logic, lower
// X%C to the equivalent of X-X/C*C.
if (N1C && !N1C->isNullValue()) {
SDOperand Div = DAG.getNode(ISD::SDIV, VT, N0, N1);
+ AddToWorkList(Div.Val);
SDOperand OptimizedDiv = combine(Div.Val);
if (OptimizedDiv.Val && OptimizedDiv.Val != Div.Val) {
SDOperand Mul = DAG.getNode(ISD::MUL, VT, OptimizedDiv, N1);
SDOperand N1 = N->getOperand(1);
ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0);
ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1);
- MVT::ValueType VT = N->getValueType(0);
+ MVT VT = N->getValueType(0);
// fold (urem c1, c2) -> c1%c2
if (N0C && N1C && !N1C->isNullValue())
return DAG.getNode(ISD::UREM, VT, N0, N1);
// fold (urem x, pow2) -> (and x, pow2-1)
- if (N1C && !N1C->isNullValue() && isPowerOf2_64(N1C->getValue()))
- return DAG.getNode(ISD::AND, VT, N0, DAG.getConstant(N1C->getValue()-1,VT));
+ if (N1C && !N1C->isNullValue() && N1C->getAPIntValue().isPowerOf2())
+ return DAG.getNode(ISD::AND, VT, N0,
+ DAG.getConstant(N1C->getAPIntValue()-1,VT));
// fold (urem x, (shl pow2, y)) -> (and x, (add (shl pow2, y), -1))
if (N1.getOpcode() == ISD::SHL) {
if (ConstantSDNode *SHC = dyn_cast<ConstantSDNode>(N1.getOperand(0))) {
- if (isPowerOf2_64(SHC->getValue())) {
- SDOperand Add = DAG.getNode(ISD::ADD, VT, N1,DAG.getConstant(~0ULL,VT));
+ if (SHC->getAPIntValue().isPowerOf2()) {
+ SDOperand Add =
+ DAG.getNode(ISD::ADD, VT, N1,
+ DAG.getConstant(APInt::getAllOnesValue(VT.getSizeInBits()),
+ VT));
AddToWorkList(Add.Val);
return DAG.getNode(ISD::AND, VT, N0, Add);
}
SDOperand N0 = N->getOperand(0);
SDOperand N1 = N->getOperand(1);
ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1);
- MVT::ValueType VT = N->getValueType(0);
+ MVT VT = N->getValueType(0);
// fold (mulhs x, 0) -> 0
if (N1C && N1C->isNullValue())
return N1;
// fold (mulhs x, 1) -> (sra x, size(x)-1)
- if (N1C && N1C->getValue() == 1)
+ if (N1C && N1C->getAPIntValue() == 1)
return DAG.getNode(ISD::SRA, N0.getValueType(), N0,
- DAG.getConstant(MVT::getSizeInBits(N0.getValueType())-1,
+ DAG.getConstant(N0.getValueType().getSizeInBits()-1,
TLI.getShiftAmountTy()));
// fold (mulhs x, undef) -> 0
if (N0.getOpcode() == ISD::UNDEF || N1.getOpcode() == ISD::UNDEF)
SDOperand N0 = N->getOperand(0);
SDOperand N1 = N->getOperand(1);
ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1);
- MVT::ValueType VT = N->getValueType(0);
+ MVT VT = N->getValueType(0);
// fold (mulhu x, 0) -> 0
if (N1C && N1C->isNullValue())
return N1;
// fold (mulhu x, 1) -> 0
- if (N1C && N1C->getValue() == 1)
+ if (N1C && N1C->getAPIntValue() == 1)
return DAG.getConstant(0, N0.getValueType());
// fold (mulhu x, undef) -> 0
if (N0.getOpcode() == ISD::UNDEF || N1.getOpcode() == ISD::UNDEF)
/// compute two values. LoOp and HiOp give the opcodes for the two computations
/// that are being performed. Return true if a simplification was made.
///
-bool DAGCombiner::SimplifyNodeWithTwoResults(SDNode *N,
- unsigned LoOp, unsigned HiOp) {
+SDOperand DAGCombiner::SimplifyNodeWithTwoResults(SDNode *N, unsigned LoOp,
+ unsigned HiOp) {
// If the high half is not needed, just compute the low half.
bool HiExists = N->hasAnyUseOfValue(1);
if (!HiExists &&
(!AfterLegalize ||
TLI.isOperationLegal(LoOp, N->getValueType(0)))) {
- DAG.ReplaceAllUsesOfValueWith(SDOperand(N, 0),
- DAG.getNode(LoOp, N->getValueType(0),
- N->op_begin(),
- N->getNumOperands()));
- return true;
+ SDOperand Res = DAG.getNode(LoOp, N->getValueType(0), N->op_begin(),
+ N->getNumOperands());
+ return CombineTo(N, Res, Res);
}
// If the low half is not needed, just compute the high half.
if (!LoExists &&
(!AfterLegalize ||
TLI.isOperationLegal(HiOp, N->getValueType(1)))) {
- DAG.ReplaceAllUsesOfValueWith(SDOperand(N, 1),
- DAG.getNode(HiOp, N->getValueType(1),
- N->op_begin(),
- N->getNumOperands()));
- return true;
+ SDOperand Res = DAG.getNode(HiOp, N->getValueType(1), N->op_begin(),
+ N->getNumOperands());
+ return CombineTo(N, Res, Res);
}
// If both halves are used, return as it is.
if (LoExists && HiExists)
- return false;
+ return SDOperand();
// If the two computed results can be simplified separately, separate them.
- bool RetVal = false;
if (LoExists) {
SDOperand Lo = DAG.getNode(LoOp, N->getValueType(0),
N->op_begin(), N->getNumOperands());
+ AddToWorkList(Lo.Val);
SDOperand LoOpt = combine(Lo.Val);
- if (LoOpt.Val && LoOpt != Lo &&
- TLI.isOperationLegal(LoOpt.getOpcode(), LoOpt.getValueType())) {
- RetVal = true;
- DAG.ReplaceAllUsesOfValueWith(SDOperand(N, 0), LoOpt);
- } else
- DAG.DeleteNode(Lo.Val);
+ if (LoOpt.Val && LoOpt.Val != Lo.Val &&
+ (!AfterLegalize ||
+ TLI.isOperationLegal(LoOpt.getOpcode(), LoOpt.getValueType())))
+ return CombineTo(N, LoOpt, LoOpt);
}
if (HiExists) {
SDOperand Hi = DAG.getNode(HiOp, N->getValueType(1),
N->op_begin(), N->getNumOperands());
+ AddToWorkList(Hi.Val);
SDOperand HiOpt = combine(Hi.Val);
if (HiOpt.Val && HiOpt != Hi &&
- TLI.isOperationLegal(HiOpt.getOpcode(), HiOpt.getValueType())) {
- RetVal = true;
- DAG.ReplaceAllUsesOfValueWith(SDOperand(N, 1), HiOpt);
- } else
- DAG.DeleteNode(Hi.Val);
+ (!AfterLegalize ||
+ TLI.isOperationLegal(HiOpt.getOpcode(), HiOpt.getValueType())))
+ return CombineTo(N, HiOpt, HiOpt);
}
-
- return RetVal;
+ return SDOperand();
}
SDOperand DAGCombiner::visitSMUL_LOHI(SDNode *N) {
-
- if (SimplifyNodeWithTwoResults(N, ISD::MUL, ISD::MULHS))
- return SDOperand();
+ SDOperand Res = SimplifyNodeWithTwoResults(N, ISD::MUL, ISD::MULHS);
+ if (Res.Val) return Res;
return SDOperand();
}
SDOperand DAGCombiner::visitUMUL_LOHI(SDNode *N) {
-
- if (SimplifyNodeWithTwoResults(N, ISD::MUL, ISD::MULHU))
- return SDOperand();
+ SDOperand Res = SimplifyNodeWithTwoResults(N, ISD::MUL, ISD::MULHU);
+ if (Res.Val) return Res;
return SDOperand();
}
SDOperand DAGCombiner::visitSDIVREM(SDNode *N) {
-
- if (SimplifyNodeWithTwoResults(N, ISD::SDIV, ISD::SREM))
- return SDOperand();
+ SDOperand Res = SimplifyNodeWithTwoResults(N, ISD::SDIV, ISD::SREM);
+ if (Res.Val) return Res;
return SDOperand();
}
SDOperand DAGCombiner::visitUDIVREM(SDNode *N) {
-
- if (SimplifyNodeWithTwoResults(N, ISD::UDIV, ISD::UREM))
- return SDOperand();
+ SDOperand Res = SimplifyNodeWithTwoResults(N, ISD::UDIV, ISD::UREM);
+ if (Res.Val) return Res;
return SDOperand();
}
/// two operands of the same opcode, try to simplify it.
SDOperand DAGCombiner::SimplifyBinOpWithSameOpcodeHands(SDNode *N) {
SDOperand N0 = N->getOperand(0), N1 = N->getOperand(1);
- MVT::ValueType VT = N0.getValueType();
+ MVT VT = N0.getValueType();
assert(N0.getOpcode() == N1.getOpcode() && "Bad input!");
// For each of OP in AND/OR/XOR:
SDOperand LL, LR, RL, RR, CC0, CC1;
ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0);
ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1);
- MVT::ValueType VT = N1.getValueType();
+ MVT VT = N1.getValueType();
+ unsigned BitWidth = VT.getSizeInBits();
// fold vector ops
- if (MVT::isVector(VT)) {
+ if (VT.isVector()) {
SDOperand FoldedVOp = SimplifyVBinOp(N);
if (FoldedVOp.Val) return FoldedVOp;
}
if (N1C && N1C->isAllOnesValue())
return N0;
// if (and x, c) is known to be zero, return 0
- if (N1C && DAG.MaskedValueIsZero(SDOperand(N, 0), MVT::getIntVTBitMask(VT)))
+ if (N1C && DAG.MaskedValueIsZero(SDOperand(N, 0),
+ APInt::getAllOnesValue(BitWidth)))
return DAG.getConstant(0, VT);
// reassociate and
SDOperand RAND = ReassociateOps(ISD::AND, N0, N1);
// fold (and (or x, 0xFFFF), 0xFF) -> 0xFF
if (N1C && N0.getOpcode() == ISD::OR)
if (ConstantSDNode *ORI = dyn_cast<ConstantSDNode>(N0.getOperand(1)))
- if ((ORI->getValue() & N1C->getValue()) == N1C->getValue())
+ if ((ORI->getAPIntValue() & N1C->getAPIntValue()) == N1C->getAPIntValue())
return N1;
// fold (and (any_ext V), c) -> (zero_ext V) if 'and' only clears top bits.
if (N1C && N0.getOpcode() == ISD::ANY_EXTEND) {
- unsigned InMask = MVT::getIntVTBitMask(N0.getOperand(0).getValueType());
- if (DAG.MaskedValueIsZero(N0.getOperand(0),
- ~N1C->getValue() & InMask)) {
+ SDOperand N0Op0 = N0.getOperand(0);
+ APInt Mask = ~N1C->getAPIntValue();
+ Mask.trunc(N0Op0.getValueSizeInBits());
+ if (DAG.MaskedValueIsZero(N0Op0, Mask)) {
SDOperand Zext = DAG.getNode(ISD::ZERO_EXTEND, N0.getValueType(),
- N0.getOperand(0));
+ N0Op0);
// Replace uses of the AND with uses of the Zero extend node.
CombineTo(N, Zext);
ISD::CondCode Op1 = cast<CondCodeSDNode>(CC1)->get();
if (LR == RR && isa<ConstantSDNode>(LR) && Op0 == Op1 &&
- MVT::isInteger(LL.getValueType())) {
+ LL.getValueType().isInteger()) {
// fold (X == 0) & (Y == 0) -> (X|Y == 0)
- if (cast<ConstantSDNode>(LR)->getValue() == 0 && Op1 == ISD::SETEQ) {
+ if (cast<ConstantSDNode>(LR)->isNullValue() && Op1 == ISD::SETEQ) {
SDOperand ORNode = DAG.getNode(ISD::OR, LR.getValueType(), LL, RL);
AddToWorkList(ORNode.Val);
return DAG.getSetCC(VT, ORNode, LR, Op1);
std::swap(RL, RR);
}
if (LL == RL && LR == RR) {
- bool isInteger = MVT::isInteger(LL.getValueType());
+ bool isInteger = LL.getValueType().isInteger();
ISD::CondCode Result = ISD::getSetCCAndOperation(Op0, Op1, isInteger);
if (Result != ISD::SETCC_INVALID)
return DAG.getSetCC(N0.getValueType(), LL, LR, Result);
// fold (and (sign_extend_inreg x, i16 to i32), 1) -> (and x, 1)
// fold (and (sra)) -> (and (srl)) when possible.
- if (!MVT::isVector(VT) &&
+ if (!VT.isVector() &&
SimplifyDemandedBits(SDOperand(N, 0)))
return SDOperand(N, 0);
// fold (zext_inreg (extload x)) -> (zextload x)
if (ISD::isEXTLoad(N0.Val) && ISD::isUNINDEXEDLoad(N0.Val)) {
LoadSDNode *LN0 = cast<LoadSDNode>(N0);
- MVT::ValueType EVT = LN0->getLoadedVT();
+ MVT EVT = LN0->getMemoryVT();
// If we zero all the possible extended bits, then we can turn this into
// a zextload if we are running before legalize or the operation is legal.
- if (DAG.MaskedValueIsZero(N1, ~0ULL << MVT::getSizeInBits(EVT)) &&
- (!AfterLegalize || TLI.isLoadXLegal(ISD::ZEXTLOAD, EVT))) {
+ unsigned BitWidth = N1.getValueSizeInBits();
+ if (DAG.MaskedValueIsZero(N1, APInt::getHighBitsSet(BitWidth,
+ BitWidth - EVT.getSizeInBits())) &&
+ ((!AfterLegalize && !LN0->isVolatile()) ||
+ TLI.isLoadXLegal(ISD::ZEXTLOAD, EVT))) {
SDOperand ExtLoad = DAG.getExtLoad(ISD::ZEXTLOAD, VT, LN0->getChain(),
LN0->getBasePtr(), LN0->getSrcValue(),
LN0->getSrcValueOffset(), EVT,
if (ISD::isSEXTLoad(N0.Val) && ISD::isUNINDEXEDLoad(N0.Val) &&
N0.hasOneUse()) {
LoadSDNode *LN0 = cast<LoadSDNode>(N0);
- MVT::ValueType EVT = LN0->getLoadedVT();
+ MVT EVT = LN0->getMemoryVT();
// If we zero all the possible extended bits, then we can turn this into
// a zextload if we are running before legalize or the operation is legal.
- if (DAG.MaskedValueIsZero(N1, ~0ULL << MVT::getSizeInBits(EVT)) &&
- (!AfterLegalize || TLI.isLoadXLegal(ISD::ZEXTLOAD, EVT))) {
+ unsigned BitWidth = N1.getValueSizeInBits();
+ if (DAG.MaskedValueIsZero(N1, APInt::getHighBitsSet(BitWidth,
+ BitWidth - EVT.getSizeInBits())) &&
+ ((!AfterLegalize && !LN0->isVolatile()) ||
+ TLI.isLoadXLegal(ISD::ZEXTLOAD, EVT))) {
SDOperand ExtLoad = DAG.getExtLoad(ISD::ZEXTLOAD, VT, LN0->getChain(),
LN0->getBasePtr(), LN0->getSrcValue(),
LN0->getSrcValueOffset(), EVT,
if (N1C && N0.getOpcode() == ISD::LOAD) {
LoadSDNode *LN0 = cast<LoadSDNode>(N0);
if (LN0->getExtensionType() != ISD::SEXTLOAD &&
- LN0->getAddressingMode() == ISD::UNINDEXED &&
- N0.hasOneUse()) {
- MVT::ValueType EVT, LoadedVT;
- if (N1C->getValue() == 255)
- EVT = MVT::i8;
- else if (N1C->getValue() == 65535)
- EVT = MVT::i16;
- else if (N1C->getValue() == ~0U)
- EVT = MVT::i32;
- else
- EVT = MVT::Other;
-
- LoadedVT = LN0->getLoadedVT();
- if (EVT != MVT::Other && LoadedVT > EVT &&
+ LN0->isUnindexed() && N0.hasOneUse() &&
+ // Do not change the width of a volatile load.
+ !LN0->isVolatile()) {
+ MVT EVT = MVT::Other;
+ uint32_t ActiveBits = N1C->getAPIntValue().getActiveBits();
+ if (ActiveBits > 0 && APIntOps::isMask(ActiveBits, N1C->getAPIntValue()))
+ EVT = MVT::getIntegerVT(ActiveBits);
+
+ MVT LoadedVT = LN0->getMemoryVT();
+ // Do not generate loads of non-round integer types since these can
+ // be expensive (and would be wrong if the type is not byte sized).
+ if (EVT != MVT::Other && LoadedVT.bitsGT(EVT) && EVT.isRound() &&
(!AfterLegalize || TLI.isLoadXLegal(ISD::ZEXTLOAD, EVT))) {
- MVT::ValueType PtrType = N0.getOperand(1).getValueType();
+ MVT PtrType = N0.getOperand(1).getValueType();
// For big endian targets, we need to add an offset to the pointer to
// load the correct bytes. For little endian systems, we merely need to
// read fewer bytes from the same pointer.
- unsigned LVTStoreBytes = MVT::getStoreSizeInBits(LoadedVT)/8;
- unsigned EVTStoreBytes = MVT::getStoreSizeInBits(EVT)/8;
+ unsigned LVTStoreBytes = LoadedVT.getStoreSizeInBits()/8;
+ unsigned EVTStoreBytes = EVT.getStoreSizeInBits()/8;
unsigned PtrOff = LVTStoreBytes - EVTStoreBytes;
unsigned Alignment = LN0->getAlignment();
SDOperand NewPtr = LN0->getBasePtr();
- if (!TLI.isLittleEndian()) {
+ if (TLI.isBigEndian()) {
NewPtr = DAG.getNode(ISD::ADD, PtrType, NewPtr,
DAG.getConstant(PtrOff, PtrType));
Alignment = MinAlign(Alignment, PtrOff);
SDOperand LL, LR, RL, RR, CC0, CC1;
ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0);
ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1);
- MVT::ValueType VT = N1.getValueType();
- unsigned OpSizeInBits = MVT::getSizeInBits(VT);
+ MVT VT = N1.getValueType();
// fold vector ops
- if (MVT::isVector(VT)) {
+ if (VT.isVector()) {
SDOperand FoldedVOp = SimplifyVBinOp(N);
if (FoldedVOp.Val) return FoldedVOp;
}
if (N1C && N1C->isAllOnesValue())
return N1;
// fold (or x, c) -> c iff (x & ~c) == 0
- if (N1C &&
- DAG.MaskedValueIsZero(N0,~N1C->getValue() & (~0ULL>>(64-OpSizeInBits))))
+ if (N1C && DAG.MaskedValueIsZero(N0, ~N1C->getAPIntValue()))
return N1;
// reassociate or
SDOperand ROR = ReassociateOps(ISD::OR, N0, N1);
ConstantSDNode *C1 = cast<ConstantSDNode>(N0.getOperand(1));
return DAG.getNode(ISD::AND, VT, DAG.getNode(ISD::OR, VT, N0.getOperand(0),
N1),
- DAG.getConstant(N1C->getValue() | C1->getValue(), VT));
+ DAG.getConstant(N1C->getAPIntValue() |
+ C1->getAPIntValue(), VT));
}
// fold (or (setcc x), (setcc y)) -> (setcc (or x, y))
if (isSetCCEquivalent(N0, LL, LR, CC0) && isSetCCEquivalent(N1, RL, RR, CC1)){
ISD::CondCode Op1 = cast<CondCodeSDNode>(CC1)->get();
if (LR == RR && isa<ConstantSDNode>(LR) && Op0 == Op1 &&
- MVT::isInteger(LL.getValueType())) {
+ LL.getValueType().isInteger()) {
// fold (X != 0) | (Y != 0) -> (X|Y != 0)
// fold (X < 0) | (Y < 0) -> (X|Y < 0)
- if (cast<ConstantSDNode>(LR)->getValue() == 0 &&
+ if (cast<ConstantSDNode>(LR)->isNullValue() &&
(Op1 == ISD::SETNE || Op1 == ISD::SETLT)) {
SDOperand ORNode = DAG.getNode(ISD::OR, LR.getValueType(), LL, RL);
AddToWorkList(ORNode.Val);
std::swap(RL, RR);
}
if (LL == RL && LR == RR) {
- bool isInteger = MVT::isInteger(LL.getValueType());
+ bool isInteger = LL.getValueType().isInteger();
ISD::CondCode Result = ISD::getSetCCOrOperation(Op0, Op1, isInteger);
if (Result != ISD::SETCC_INVALID)
return DAG.getSetCC(N0.getValueType(), LL, LR, Result);
(N0.Val->hasOneUse() || N1.Val->hasOneUse())) {
// We can only do this xform if we know that bits from X that are set in C2
// but not in C1 are already zero. Likewise for Y.
- uint64_t LHSMask = cast<ConstantSDNode>(N0.getOperand(1))->getValue();
- uint64_t RHSMask = cast<ConstantSDNode>(N1.getOperand(1))->getValue();
+ const APInt &LHSMask =
+ cast<ConstantSDNode>(N0.getOperand(1))->getAPIntValue();
+ const APInt &RHSMask =
+ cast<ConstantSDNode>(N1.getOperand(1))->getAPIntValue();
if (DAG.MaskedValueIsZero(N0.getOperand(0), RHSMask&~LHSMask) &&
DAG.MaskedValueIsZero(N1.getOperand(0), LHSMask&~RHSMask)) {
// idioms for rotate, and if the target supports rotation instructions, generate
// a rot[lr].
SDNode *DAGCombiner::MatchRotate(SDOperand LHS, SDOperand RHS) {
- // Must be a legal type. Expanded an promoted things won't work with rotates.
- MVT::ValueType VT = LHS.getValueType();
+ // Must be a legal type. Expanded 'n promoted things won't work with rotates.
+ MVT VT = LHS.getValueType();
if (!TLI.isTypeLegal(VT)) return 0;
// The target must have at least one rotate flavor.
bool HasROTL = TLI.isOperationLegal(ISD::ROTL, VT);
bool HasROTR = TLI.isOperationLegal(ISD::ROTR, VT);
if (!HasROTL && !HasROTR) return 0;
-
+
// Match "(X shl/srl V1) & V2" where V2 may not be present.
SDOperand LHSShift; // The shift.
SDOperand LHSMask; // AND value if any.
std::swap(LHSMask , RHSMask );
}
- unsigned OpSizeInBits = MVT::getSizeInBits(VT);
+ unsigned OpSizeInBits = VT.getSizeInBits();
SDOperand LHSShiftArg = LHSShift.getOperand(0);
SDOperand LHSShiftAmt = LHSShift.getOperand(1);
SDOperand RHSShiftAmt = RHSShift.getOperand(1);
// If there is an AND of either shifted operand, apply it to the result.
if (LHSMask.Val || RHSMask.Val) {
- uint64_t Mask = MVT::getIntVTBitMask(VT);
+ APInt Mask = APInt::getAllOnesValue(OpSizeInBits);
if (LHSMask.Val) {
- uint64_t RHSBits = (1ULL << LShVal)-1;
- Mask &= cast<ConstantSDNode>(LHSMask)->getValue() | RHSBits;
+ APInt RHSBits = APInt::getLowBitsSet(OpSizeInBits, LShVal);
+ Mask &= cast<ConstantSDNode>(LHSMask)->getAPIntValue() | RHSBits;
}
if (RHSMask.Val) {
- uint64_t LHSBits = ~((1ULL << (OpSizeInBits-RShVal))-1);
- Mask &= cast<ConstantSDNode>(RHSMask)->getValue() | LHSBits;
+ APInt LHSBits = APInt::getHighBitsSet(OpSizeInBits, RShVal);
+ Mask &= cast<ConstantSDNode>(RHSMask)->getAPIntValue() | LHSBits;
}
Rot = DAG.getNode(ISD::AND, VT, Rot, DAG.getConstant(Mask, VT));
LHSShiftAmt == RHSShiftAmt.getOperand(1)) {
if (ConstantSDNode *SUBC =
dyn_cast<ConstantSDNode>(RHSShiftAmt.getOperand(0))) {
- if (SUBC->getValue() == OpSizeInBits)
+ if (SUBC->getAPIntValue() == OpSizeInBits) {
if (HasROTL)
return DAG.getNode(ISD::ROTL, VT, LHSShiftArg, LHSShiftAmt).Val;
else
return DAG.getNode(ISD::ROTR, VT, LHSShiftArg, RHSShiftAmt).Val;
+ }
}
}
RHSShiftAmt == LHSShiftAmt.getOperand(1)) {
if (ConstantSDNode *SUBC =
dyn_cast<ConstantSDNode>(LHSShiftAmt.getOperand(0))) {
- if (SUBC->getValue() == OpSizeInBits)
+ if (SUBC->getAPIntValue() == OpSizeInBits) {
if (HasROTL)
return DAG.getNode(ISD::ROTL, VT, LHSShiftArg, LHSShiftAmt).Val;
else
return DAG.getNode(ISD::ROTR, VT, LHSShiftArg, RHSShiftAmt).Val;
+ }
}
}
// fold (or (shl x, (*ext y)), (srl x, (*ext (sub 32, y)))) ->
// (rotl x, (sub 32, y))
if (ConstantSDNode *SUBC = cast<ConstantSDNode>(RExtOp0.getOperand(0))) {
- if (SUBC->getValue() == OpSizeInBits) {
+ if (SUBC->getAPIntValue() == OpSizeInBits) {
if (HasROTL)
return DAG.getNode(ISD::ROTL, VT, LHSShiftArg, LHSShiftAmt).Val;
else
// fold (or (shl x, (*ext (sub 32, y))), (srl x, (*ext r))) ->
// (rotr x, (sub 32, y))
if (ConstantSDNode *SUBC = cast<ConstantSDNode>(LExtOp0.getOperand(0))) {
- if (SUBC->getValue() == OpSizeInBits) {
+ if (SUBC->getAPIntValue() == OpSizeInBits) {
if (HasROTL)
return DAG.getNode(ISD::ROTL, VT, LHSShiftArg, RHSShiftAmt).Val;
else
SDOperand LHS, RHS, CC;
ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0);
ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1);
- MVT::ValueType VT = N0.getValueType();
+ MVT VT = N0.getValueType();
// fold vector ops
- if (MVT::isVector(VT)) {
+ if (VT.isVector()) {
SDOperand FoldedVOp = SimplifyVBinOp(N);
if (FoldedVOp.Val) return FoldedVOp;
}
+ // fold (xor undef, undef) -> 0. This is a common idiom (misuse).
+ if (N0.getOpcode() == ISD::UNDEF && N1.getOpcode() == ISD::UNDEF)
+ return DAG.getConstant(0, VT);
// fold (xor x, undef) -> undef
if (N0.getOpcode() == ISD::UNDEF)
return N0;
if (RXOR.Val != 0)
return RXOR;
// fold !(x cc y) -> (x !cc y)
- if (N1C && N1C->getValue() == 1 && isSetCCEquivalent(N0, LHS, RHS, CC)) {
- bool isInt = MVT::isInteger(LHS.getValueType());
+ if (N1C && N1C->getAPIntValue() == 1 && isSetCCEquivalent(N0, LHS, RHS, CC)) {
+ bool isInt = LHS.getValueType().isInteger();
ISD::CondCode NotCC = ISD::getSetCCInverse(cast<CondCodeSDNode>(CC)->get(),
isInt);
if (N0.getOpcode() == ISD::SETCC)
abort();
}
// fold (not (zext (setcc x, y))) -> (zext (not (setcc x, y)))
- if (N1C && N1C->getValue() == 1 && N0.getOpcode() == ISD::ZERO_EXTEND &&
+ if (N1C && N1C->getAPIntValue() == 1 && N0.getOpcode() == ISD::ZERO_EXTEND &&
N0.Val->hasOneUse() && isSetCCEquivalent(N0.getOperand(0), LHS, RHS, CC)){
SDOperand V = N0.getOperand(0);
V = DAG.getNode(ISD::XOR, V.getValueType(), V,
}
// fold !(x or y) -> (!x and !y) iff x or y are setcc
- if (N1C && N1C->getValue() == 1 && VT == MVT::i1 &&
+ if (N1C && N1C->getAPIntValue() == 1 && VT == MVT::i1 &&
(N0.getOpcode() == ISD::OR || N0.getOpcode() == ISD::AND)) {
SDOperand LHS = N0.getOperand(0), RHS = N0.getOperand(1);
if (isOneUseSetCC(RHS) || isOneUseSetCC(LHS)) {
ConstantSDNode *N01C = dyn_cast<ConstantSDNode>(N0.getOperand(1));
if (N00C)
return DAG.getNode(ISD::XOR, VT, N0.getOperand(1),
- DAG.getConstant(N1C->getValue()^N00C->getValue(), VT));
+ DAG.getConstant(N1C->getAPIntValue()^
+ N00C->getAPIntValue(), VT));
if (N01C)
return DAG.getNode(ISD::XOR, VT, N0.getOperand(0),
- DAG.getConstant(N1C->getValue()^N01C->getValue(), VT));
+ DAG.getConstant(N1C->getAPIntValue()^
+ N01C->getAPIntValue(), VT));
}
// fold (xor x, x) -> 0
if (N0 == N1) {
- if (!MVT::isVector(VT)) {
+ if (!VT.isVector()) {
return DAG.getConstant(0, VT);
} else if (!AfterLegalize || TLI.isOperationLegal(ISD::BUILD_VECTOR, VT)) {
// Produce a vector of zeros.
- SDOperand El = DAG.getConstant(0, MVT::getVectorElementType(VT));
- std::vector<SDOperand> Ops(MVT::getVectorNumElements(VT), El);
+ SDOperand El = DAG.getConstant(0, VT.getVectorElementType());
+ std::vector<SDOperand> Ops(VT.getVectorNumElements(), El);
return DAG.getNode(ISD::BUILD_VECTOR, VT, &Ops[0], Ops.size());
}
}
}
// Simplify the expression using non-local knowledge.
- if (!MVT::isVector(VT) &&
+ if (!VT.isVector() &&
SimplifyDemandedBits(SDOperand(N, 0)))
return SDOperand(N, 0);
!isa<ConstantSDNode>(BinOpLHSVal->getOperand(1)))
return SDOperand();
- MVT::ValueType VT = N->getValueType(0);
+ MVT VT = N->getValueType(0);
// If this is a signed shift right, and the high bit is modified
// by the logical operation, do not perform the transformation.
// the constant which would cause it to be modified for this
// operation.
if (N->getOpcode() == ISD::SRA) {
- uint64_t BinOpRHSSign = BinOpCst->getValue() >> MVT::getSizeInBits(VT)-1;
- if ((bool)BinOpRHSSign != HighBitSet)
+ bool BinOpRHSSignSet = BinOpCst->getAPIntValue().isNegative();
+ if (BinOpRHSSignSet != HighBitSet)
return SDOperand();
}
SDOperand N1 = N->getOperand(1);
ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0);
ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1);
- MVT::ValueType VT = N0.getValueType();
- unsigned OpSizeInBits = MVT::getSizeInBits(VT);
+ MVT VT = N0.getValueType();
+ unsigned OpSizeInBits = VT.getSizeInBits();
// fold (shl c1, c2) -> c1<<c2
if (N0C && N1C)
if (N1C && N1C->isNullValue())
return N0;
// if (shl x, c) is known to be zero, return 0
- if (DAG.MaskedValueIsZero(SDOperand(N, 0), MVT::getIntVTBitMask(VT)))
+ if (DAG.MaskedValueIsZero(SDOperand(N, 0),
+ APInt::getAllOnesValue(VT.getSizeInBits())))
return DAG.getConstant(0, VT);
if (N1C && SimplifyDemandedBits(SDOperand(N, 0)))
return SDOperand(N, 0);
SDOperand N1 = N->getOperand(1);
ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0);
ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1);
- MVT::ValueType VT = N0.getValueType();
+ MVT VT = N0.getValueType();
// fold (sra c1, c2) -> c1>>c2
if (N0C && N1C)
if (N0C && N0C->isAllOnesValue())
return N0;
// fold (sra x, c >= size(x)) -> undef
- if (N1C && N1C->getValue() >= MVT::getSizeInBits(VT))
+ if (N1C && N1C->getValue() >= VT.getSizeInBits())
return DAG.getNode(ISD::UNDEF, VT);
// fold (sra x, 0) -> x
if (N1C && N1C->isNullValue())
// fold (sra (shl x, c1), c1) -> sext_inreg for some c1 and target supports
// sext_inreg.
if (N1C && N0.getOpcode() == ISD::SHL && N1 == N0.getOperand(1)) {
- unsigned LowBits = MVT::getSizeInBits(VT) - (unsigned)N1C->getValue();
- MVT::ValueType EVT;
- switch (LowBits) {
- default: EVT = MVT::Other; break;
- case 1: EVT = MVT::i1; break;
- case 8: EVT = MVT::i8; break;
- case 16: EVT = MVT::i16; break;
- case 32: EVT = MVT::i32; break;
- }
- if (EVT > MVT::Other && TLI.isOperationLegal(ISD::SIGN_EXTEND_INREG, EVT))
+ unsigned LowBits = VT.getSizeInBits() - (unsigned)N1C->getValue();
+ MVT EVT = MVT::getIntegerVT(LowBits);
+ if (EVT.isSimple() && // TODO: remove when apint codegen support lands.
+ (!AfterLegalize || TLI.isOperationLegal(ISD::SIGN_EXTEND_INREG, EVT)))
return DAG.getNode(ISD::SIGN_EXTEND_INREG, VT, N0.getOperand(0),
DAG.getValueType(EVT));
}
-
+
// fold (sra (sra x, c1), c2) -> (sra x, c1+c2)
if (N1C && N0.getOpcode() == ISD::SRA) {
if (ConstantSDNode *C1 = dyn_cast<ConstantSDNode>(N0.getOperand(1))) {
unsigned Sum = N1C->getValue() + C1->getValue();
- if (Sum >= MVT::getSizeInBits(VT)) Sum = MVT::getSizeInBits(VT)-1;
+ if (Sum >= VT.getSizeInBits()) Sum = VT.getSizeInBits()-1;
return DAG.getNode(ISD::SRA, VT, N0.getOperand(0),
DAG.getConstant(Sum, N1C->getValueType(0)));
}
}
+
+ // fold sra (shl X, m), result_size - n
+ // -> (sign_extend (trunc (shl X, result_size - n - m))) for
+ // result_size - n != m.
+ // If truncate is free for the target sext(shl) is likely to result in better
+ // code.
+ if (N0.getOpcode() == ISD::SHL) {
+ // Get the two constanst of the shifts, CN0 = m, CN = n.
+ const ConstantSDNode *N01C = dyn_cast<ConstantSDNode>(N0.getOperand(1));
+ if (N01C && N1C) {
+ // Determine what the truncate's result bitsize and type would be.
+ unsigned VTValSize = VT.getSizeInBits();
+ MVT TruncVT =
+ MVT::getIntegerVT(VTValSize - N1C->getValue());
+ // Determine the residual right-shift amount.
+ unsigned ShiftAmt = N1C->getValue() - N01C->getValue();
+
+ // If the shift is not a no-op (in which case this should be just a sign
+ // extend already), the truncated to type is legal, sign_extend is legal
+ // on that type, and the the truncate to that type is both legal and free,
+ // perform the transform.
+ if (ShiftAmt &&
+ TLI.isOperationLegal(ISD::SIGN_EXTEND, TruncVT) &&
+ TLI.isOperationLegal(ISD::TRUNCATE, VT) &&
+ TLI.isTruncateFree(VT, TruncVT)) {
+
+ SDOperand Amt = DAG.getConstant(ShiftAmt, TLI.getShiftAmountTy());
+ SDOperand Shift = DAG.getNode(ISD::SRL, VT, N0.getOperand(0), Amt);
+ SDOperand Trunc = DAG.getNode(ISD::TRUNCATE, TruncVT, Shift);
+ return DAG.getNode(ISD::SIGN_EXTEND, N->getValueType(0), Trunc);
+ }
+ }
+ }
// Simplify, based on bits shifted out of the LHS.
if (N1C && SimplifyDemandedBits(SDOperand(N, 0)))
// If the sign bit is known to be zero, switch this to a SRL.
- if (DAG.MaskedValueIsZero(N0, MVT::getIntVTSignBit(VT)))
+ if (DAG.SignBitIsZero(N0))
return DAG.getNode(ISD::SRL, VT, N0, N1);
return N1C ? visitShiftByConstant(N, N1C->getValue()) : SDOperand();
SDOperand N1 = N->getOperand(1);
ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0);
ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1);
- MVT::ValueType VT = N0.getValueType();
- unsigned OpSizeInBits = MVT::getSizeInBits(VT);
+ MVT VT = N0.getValueType();
+ unsigned OpSizeInBits = VT.getSizeInBits();
// fold (srl c1, c2) -> c1 >>u c2
if (N0C && N1C)
if (N1C && N1C->isNullValue())
return N0;
// if (srl x, c) is known to be zero, return 0
- if (N1C && DAG.MaskedValueIsZero(SDOperand(N, 0), ~0ULL >> (64-OpSizeInBits)))
+ if (N1C && DAG.MaskedValueIsZero(SDOperand(N, 0),
+ APInt::getAllOnesValue(OpSizeInBits)))
return DAG.getConstant(0, VT);
// fold (srl (srl x, c1), c2) -> 0 or (srl x, c1+c2)
// fold (srl (anyextend x), c) -> (anyextend (srl x, c))
if (N1C && N0.getOpcode() == ISD::ANY_EXTEND) {
// Shifting in all undef bits?
- MVT::ValueType SmallVT = N0.getOperand(0).getValueType();
- if (N1C->getValue() >= MVT::getSizeInBits(SmallVT))
+ MVT SmallVT = N0.getOperand(0).getValueType();
+ if (N1C->getValue() >= SmallVT.getSizeInBits())
return DAG.getNode(ISD::UNDEF, VT);
SDOperand SmallShift = DAG.getNode(ISD::SRL, SmallVT, N0.getOperand(0), N1);
// fold (srl (sra X, Y), 31) -> (srl X, 31). This srl only looks at the sign
// bit, which is unmodified by sra.
- if (N1C && N1C->getValue()+1 == MVT::getSizeInBits(VT)) {
+ if (N1C && N1C->getValue()+1 == VT.getSizeInBits()) {
if (N0.getOpcode() == ISD::SRA)
return DAG.getNode(ISD::SRL, VT, N0.getOperand(0), N1);
}
// fold (srl (ctlz x), "5") -> x iff x has one bit set (the low bit).
if (N1C && N0.getOpcode() == ISD::CTLZ &&
- N1C->getValue() == Log2_32(MVT::getSizeInBits(VT))) {
- uint64_t KnownZero, KnownOne, Mask = MVT::getIntVTBitMask(VT);
+ N1C->getAPIntValue() == Log2_32(VT.getSizeInBits())) {
+ APInt KnownZero, KnownOne;
+ APInt Mask = APInt::getAllOnesValue(VT.getSizeInBits());
DAG.ComputeMaskedBits(N0.getOperand(0), Mask, KnownZero, KnownOne);
// If any of the input bits are KnownOne, then the input couldn't be all
// zeros, thus the result of the srl will always be zero.
- if (KnownOne) return DAG.getConstant(0, VT);
+ if (KnownOne.getBoolValue()) return DAG.getConstant(0, VT);
// If all of the bits input the to ctlz node are known to be zero, then
// the result of the ctlz is "32" and the result of the shift is one.
- uint64_t UnknownBits = ~KnownZero & Mask;
+ APInt UnknownBits = ~KnownZero & Mask;
if (UnknownBits == 0) return DAG.getConstant(1, VT);
// Otherwise, check to see if there is exactly one bit input to the ctlz.
// could be set on input to the CTLZ node. If this bit is set, the SRL
// will return 0, if it is clear, it returns 1. Change the CTLZ/SRL pair
// to an SRL,XOR pair, which is likely to simplify more.
- unsigned ShAmt = CountTrailingZeros_64(UnknownBits);
+ unsigned ShAmt = UnknownBits.countTrailingZeros();
SDOperand Op = N0.getOperand(0);
if (ShAmt) {
Op = DAG.getNode(ISD::SRL, VT, Op,
SDOperand DAGCombiner::visitCTLZ(SDNode *N) {
SDOperand N0 = N->getOperand(0);
- MVT::ValueType VT = N->getValueType(0);
+ MVT VT = N->getValueType(0);
// fold (ctlz c1) -> c2
if (isa<ConstantSDNode>(N0))
SDOperand DAGCombiner::visitCTTZ(SDNode *N) {
SDOperand N0 = N->getOperand(0);
- MVT::ValueType VT = N->getValueType(0);
+ MVT VT = N->getValueType(0);
// fold (cttz c1) -> c2
if (isa<ConstantSDNode>(N0))
SDOperand DAGCombiner::visitCTPOP(SDNode *N) {
SDOperand N0 = N->getOperand(0);
- MVT::ValueType VT = N->getValueType(0);
+ MVT VT = N->getValueType(0);
// fold (ctpop c1) -> c2
if (isa<ConstantSDNode>(N0))
ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0);
ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1);
ConstantSDNode *N2C = dyn_cast<ConstantSDNode>(N2);
- MVT::ValueType VT = N->getValueType(0);
- MVT::ValueType VT0 = N0.getValueType();
+ MVT VT = N->getValueType(0);
+ MVT VT0 = N0.getValueType();
// fold select C, X, X -> X
if (N1 == N2)
if (N0C && N0C->isNullValue())
return N2;
// fold select C, 1, X -> C | X
- if (MVT::i1 == VT && N1C && N1C->getValue() == 1)
+ if (VT == MVT::i1 && N1C && N1C->getAPIntValue() == 1)
return DAG.getNode(ISD::OR, VT, N0, N2);
// fold select C, 0, 1 -> ~C
- if (MVT::isInteger(VT) && MVT::isInteger(VT0) &&
- N1C && N2C && N1C->isNullValue() && N2C->getValue() == 1) {
+ if (VT.isInteger() && VT0.isInteger() &&
+ N1C && N2C && N1C->isNullValue() && N2C->getAPIntValue() == 1) {
SDOperand XORNode = DAG.getNode(ISD::XOR, VT0, N0, DAG.getConstant(1, VT0));
if (VT == VT0)
return XORNode;
AddToWorkList(XORNode.Val);
- if (MVT::getSizeInBits(VT) > MVT::getSizeInBits(VT0))
+ if (VT.bitsGT(VT0))
return DAG.getNode(ISD::ZERO_EXTEND, VT, XORNode);
return DAG.getNode(ISD::TRUNCATE, VT, XORNode);
}
return DAG.getNode(ISD::AND, VT, XORNode, N2);
}
// fold select C, X, 1 -> ~C | X
- if (VT == VT0 && VT == MVT::i1 && N2C && N2C->getValue() == 1) {
+ if (VT == VT0 && VT == MVT::i1 && N2C && N2C->getAPIntValue() == 1) {
SDOperand XORNode = DAG.getNode(ISD::XOR, VT, N0, DAG.getConstant(1, VT));
AddToWorkList(XORNode.Val);
return DAG.getNode(ISD::OR, VT, XORNode, N1);
}
// fold select C, X, 0 -> C & X
// FIXME: this should check for C type == X type, not i1?
- if (MVT::i1 == VT && N2C && N2C->isNullValue())
+ if (VT == MVT::i1 && N2C && N2C->isNullValue())
return DAG.getNode(ISD::AND, VT, N0, N1);
// fold X ? X : Y --> X ? 1 : Y --> X | Y
- if (MVT::i1 == VT && N0 == N1)
+ if (VT == MVT::i1 && N0 == N1)
return DAG.getNode(ISD::OR, VT, N0, N2);
// fold X ? Y : X --> X ? Y : 0 --> X & Y
- if (MVT::i1 == VT && N0 == N2)
+ if (VT == MVT::i1 && N0 == N2)
return DAG.getNode(ISD::AND, VT, N0, N1);
// If we can fold this based on the true/false value, do so.
if (SimplifySelectOps(N, N1, N2))
return SDOperand(N, 0); // Don't revisit N.
-
+
// fold selects based on a setcc into other things, such as min/max/abs
- if (N0.getOpcode() == ISD::SETCC)
+ if (N0.getOpcode() == ISD::SETCC) {
// FIXME:
// Check against MVT::Other for SELECT_CC, which is a workaround for targets
// having to say they don't support SELECT_CC on every type the DAG knows
N1, N2, N0.getOperand(2));
else
return SimplifySelect(N0, N1, N2);
+ }
return SDOperand();
}
return N2;
// Determine if the condition we're dealing with is constant
- SDOperand SCC = SimplifySetCC(TLI.getSetCCResultTy(), N0, N1, CC, false);
+ SDOperand SCC = SimplifySetCC(TLI.getSetCCResultType(N0), N0, N1, CC, false);
if (SCC.Val) AddToWorkList(SCC.Val);
if (ConstantSDNode *SCCC = dyn_cast_or_null<ConstantSDNode>(SCC.Val)) {
- if (SCCC->getValue())
+ if (!SCCC->isNullValue())
return N2; // cond always true -> true val
else
return N3; // cond always false -> false val
bool isTruncFree = TLI.isTruncateFree(N->getValueType(0), N0.getValueType());
for (SDNode::use_iterator UI = N0.Val->use_begin(), UE = N0.Val->use_end();
UI != UE; ++UI) {
- SDNode *User = *UI;
+ SDNode *User = UI->getUser();
if (User == N)
continue;
// FIXME: Only extend SETCC N, N and SETCC N, c for now.
bool BothLiveOut = false;
for (SDNode::use_iterator UI = N->use_begin(), UE = N->use_end();
UI != UE; ++UI) {
- SDNode *User = *UI;
+ SDNode *User = UI->getUser();
for (unsigned i = 0, e = User->getNumOperands(); i != e; ++i) {
SDOperand UseOp = User->getOperand(i);
if (UseOp.Val == N && UseOp.ResNo == 0) {
SDOperand DAGCombiner::visitSIGN_EXTEND(SDNode *N) {
SDOperand N0 = N->getOperand(0);
- MVT::ValueType VT = N->getValueType(0);
+ MVT VT = N->getValueType(0);
// fold (sext c1) -> c1
if (isa<ConstantSDNode>(N0))
if (N0.getOpcode() == ISD::SIGN_EXTEND || N0.getOpcode() == ISD::ANY_EXTEND)
return DAG.getNode(ISD::SIGN_EXTEND, VT, N0.getOperand(0));
- // fold (sext (truncate (load x))) -> (sext (smaller load x))
- // fold (sext (truncate (srl (load x), c))) -> (sext (smaller load (x+c/n)))
if (N0.getOpcode() == ISD::TRUNCATE) {
+ // fold (sext (truncate (load x))) -> (sext (smaller load x))
+ // fold (sext (truncate (srl (load x), c))) -> (sext (smaller load (x+c/n)))
SDOperand NarrowLoad = ReduceLoadWidth(N0.Val);
if (NarrowLoad.Val) {
if (NarrowLoad.Val != N0.Val)
CombineTo(N0.Val, NarrowLoad);
return DAG.getNode(ISD::SIGN_EXTEND, VT, NarrowLoad);
}
- }
- // See if the value being truncated is already sign extended. If so, just
- // eliminate the trunc/sext pair.
- if (N0.getOpcode() == ISD::TRUNCATE) {
+ // See if the value being truncated is already sign extended. If so, just
+ // eliminate the trunc/sext pair.
SDOperand Op = N0.getOperand(0);
- unsigned OpBits = MVT::getSizeInBits(Op.getValueType());
- unsigned MidBits = MVT::getSizeInBits(N0.getValueType());
- unsigned DestBits = MVT::getSizeInBits(VT);
+ unsigned OpBits = Op.getValueType().getSizeInBits();
+ unsigned MidBits = N0.getValueType().getSizeInBits();
+ unsigned DestBits = VT.getSizeInBits();
unsigned NumSignBits = DAG.ComputeNumSignBits(Op);
if (OpBits == DestBits) {
// fold (sext (truncate x)) -> (sextinreg x).
if (!AfterLegalize || TLI.isOperationLegal(ISD::SIGN_EXTEND_INREG,
N0.getValueType())) {
- if (Op.getValueType() < VT)
+ if (Op.getValueType().bitsLT(VT))
Op = DAG.getNode(ISD::ANY_EXTEND, VT, Op);
- else if (Op.getValueType() > VT)
+ else if (Op.getValueType().bitsGT(VT))
Op = DAG.getNode(ISD::TRUNCATE, VT, Op);
return DAG.getNode(ISD::SIGN_EXTEND_INREG, VT, Op,
DAG.getValueType(N0.getValueType()));
// fold (sext (load x)) -> (sext (truncate (sextload x)))
if (ISD::isNON_EXTLoad(N0.Val) &&
- (!AfterLegalize||TLI.isLoadXLegal(ISD::SEXTLOAD, N0.getValueType()))){
+ ((!AfterLegalize && !cast<LoadSDNode>(N0)->isVolatile()) ||
+ TLI.isLoadXLegal(ISD::SEXTLOAD, N0.getValueType()))) {
bool DoXform = true;
SmallVector<SDNode*, 4> SetCCs;
if (!N0.hasOneUse())
if ((ISD::isSEXTLoad(N0.Val) || ISD::isEXTLoad(N0.Val)) &&
ISD::isUNINDEXEDLoad(N0.Val) && N0.hasOneUse()) {
LoadSDNode *LN0 = cast<LoadSDNode>(N0);
- MVT::ValueType EVT = LN0->getLoadedVT();
- if (!AfterLegalize || TLI.isLoadXLegal(ISD::SEXTLOAD, EVT)) {
+ MVT EVT = LN0->getMemoryVT();
+ if ((!AfterLegalize && !LN0->isVolatile()) ||
+ TLI.isLoadXLegal(ISD::SEXTLOAD, EVT)) {
SDOperand ExtLoad = DAG.getExtLoad(ISD::SEXTLOAD, VT, LN0->getChain(),
LN0->getBasePtr(), LN0->getSrcValue(),
LN0->getSrcValueOffset(), EVT,
if (SCC.Val) return SCC;
}
+ // fold (sext x) -> (zext x) if the sign bit is known zero.
+ if ((!AfterLegalize || TLI.isOperationLegal(ISD::ZERO_EXTEND, VT)) &&
+ DAG.SignBitIsZero(N0))
+ return DAG.getNode(ISD::ZERO_EXTEND, VT, N0);
+
return SDOperand();
}
SDOperand DAGCombiner::visitZERO_EXTEND(SDNode *N) {
SDOperand N0 = N->getOperand(0);
- MVT::ValueType VT = N->getValueType(0);
+ MVT VT = N->getValueType(0);
// fold (zext c1) -> c1
if (isa<ConstantSDNode>(N0))
if (N0.getOpcode() == ISD::TRUNCATE &&
(!AfterLegalize || TLI.isOperationLegal(ISD::AND, VT))) {
SDOperand Op = N0.getOperand(0);
- if (Op.getValueType() < VT) {
+ if (Op.getValueType().bitsLT(VT)) {
Op = DAG.getNode(ISD::ANY_EXTEND, VT, Op);
- } else if (Op.getValueType() > VT) {
+ } else if (Op.getValueType().bitsGT(VT)) {
Op = DAG.getNode(ISD::TRUNCATE, VT, Op);
}
return DAG.getZeroExtendInReg(Op, N0.getValueType());
N0.getOperand(0).getOpcode() == ISD::TRUNCATE &&
N0.getOperand(1).getOpcode() == ISD::Constant) {
SDOperand X = N0.getOperand(0).getOperand(0);
- if (X.getValueType() < VT) {
+ if (X.getValueType().bitsLT(VT)) {
X = DAG.getNode(ISD::ANY_EXTEND, VT, X);
- } else if (X.getValueType() > VT) {
+ } else if (X.getValueType().bitsGT(VT)) {
X = DAG.getNode(ISD::TRUNCATE, VT, X);
}
- uint64_t Mask = cast<ConstantSDNode>(N0.getOperand(1))->getValue();
+ APInt Mask = cast<ConstantSDNode>(N0.getOperand(1))->getAPIntValue();
+ Mask.zext(VT.getSizeInBits());
return DAG.getNode(ISD::AND, VT, X, DAG.getConstant(Mask, VT));
}
// fold (zext (load x)) -> (zext (truncate (zextload x)))
if (ISD::isNON_EXTLoad(N0.Val) &&
- (!AfterLegalize||TLI.isLoadXLegal(ISD::ZEXTLOAD, N0.getValueType()))) {
+ ((!AfterLegalize && !cast<LoadSDNode>(N0)->isVolatile()) ||
+ TLI.isLoadXLegal(ISD::ZEXTLOAD, N0.getValueType()))) {
bool DoXform = true;
SmallVector<SDNode*, 4> SetCCs;
if (!N0.hasOneUse())
if ((ISD::isZEXTLoad(N0.Val) || ISD::isEXTLoad(N0.Val)) &&
ISD::isUNINDEXEDLoad(N0.Val) && N0.hasOneUse()) {
LoadSDNode *LN0 = cast<LoadSDNode>(N0);
- MVT::ValueType EVT = LN0->getLoadedVT();
- SDOperand ExtLoad = DAG.getExtLoad(ISD::ZEXTLOAD, VT, LN0->getChain(),
- LN0->getBasePtr(), LN0->getSrcValue(),
- LN0->getSrcValueOffset(), EVT,
- LN0->isVolatile(),
- LN0->getAlignment());
- CombineTo(N, ExtLoad);
- CombineTo(N0.Val, DAG.getNode(ISD::TRUNCATE, N0.getValueType(), ExtLoad),
- ExtLoad.getValue(1));
- return SDOperand(N, 0); // Return N so it doesn't get rechecked!
+ MVT EVT = LN0->getMemoryVT();
+ if ((!AfterLegalize && !LN0->isVolatile()) ||
+ TLI.isLoadXLegal(ISD::ZEXTLOAD, EVT)) {
+ SDOperand ExtLoad = DAG.getExtLoad(ISD::ZEXTLOAD, VT, LN0->getChain(),
+ LN0->getBasePtr(), LN0->getSrcValue(),
+ LN0->getSrcValueOffset(), EVT,
+ LN0->isVolatile(),
+ LN0->getAlignment());
+ CombineTo(N, ExtLoad);
+ CombineTo(N0.Val, DAG.getNode(ISD::TRUNCATE, N0.getValueType(), ExtLoad),
+ ExtLoad.getValue(1));
+ return SDOperand(N, 0); // Return N so it doesn't get rechecked!
+ }
}
// zext(setcc x,y,cc) -> select_cc x, y, 1, 0, cc
SDOperand DAGCombiner::visitANY_EXTEND(SDNode *N) {
SDOperand N0 = N->getOperand(0);
- MVT::ValueType VT = N->getValueType(0);
+ MVT VT = N->getValueType(0);
// fold (aext c1) -> c1
if (isa<ConstantSDNode>(N0))
SDOperand TruncOp = N0.getOperand(0);
if (TruncOp.getValueType() == VT)
return TruncOp; // x iff x size == zext size.
- if (TruncOp.getValueType() > VT)
+ if (TruncOp.getValueType().bitsGT(VT))
return DAG.getNode(ISD::TRUNCATE, VT, TruncOp);
return DAG.getNode(ISD::ANY_EXTEND, VT, TruncOp);
}
N0.getOperand(0).getOpcode() == ISD::TRUNCATE &&
N0.getOperand(1).getOpcode() == ISD::Constant) {
SDOperand X = N0.getOperand(0).getOperand(0);
- if (X.getValueType() < VT) {
+ if (X.getValueType().bitsLT(VT)) {
X = DAG.getNode(ISD::ANY_EXTEND, VT, X);
- } else if (X.getValueType() > VT) {
+ } else if (X.getValueType().bitsGT(VT)) {
X = DAG.getNode(ISD::TRUNCATE, VT, X);
}
- uint64_t Mask = cast<ConstantSDNode>(N0.getOperand(1))->getValue();
+ APInt Mask = cast<ConstantSDNode>(N0.getOperand(1))->getAPIntValue();
+ Mask.zext(VT.getSizeInBits());
return DAG.getNode(ISD::AND, VT, X, DAG.getConstant(Mask, VT));
}
// fold (aext (load x)) -> (aext (truncate (extload x)))
if (ISD::isNON_EXTLoad(N0.Val) && N0.hasOneUse() &&
- (!AfterLegalize||TLI.isLoadXLegal(ISD::EXTLOAD, N0.getValueType()))) {
+ ((!AfterLegalize && !cast<LoadSDNode>(N0)->isVolatile()) ||
+ TLI.isLoadXLegal(ISD::EXTLOAD, N0.getValueType()))) {
LoadSDNode *LN0 = cast<LoadSDNode>(N0);
SDOperand ExtLoad = DAG.getExtLoad(ISD::EXTLOAD, VT, LN0->getChain(),
LN0->getBasePtr(), LN0->getSrcValue(),
!ISD::isNON_EXTLoad(N0.Val) && ISD::isUNINDEXEDLoad(N0.Val) &&
N0.hasOneUse()) {
LoadSDNode *LN0 = cast<LoadSDNode>(N0);
- MVT::ValueType EVT = LN0->getLoadedVT();
+ MVT EVT = LN0->getMemoryVT();
SDOperand ExtLoad = DAG.getExtLoad(LN0->getExtensionType(), VT,
LN0->getChain(), LN0->getBasePtr(),
LN0->getSrcValue(),
/// GetDemandedBits - See if the specified operand can be simplified with the
/// knowledge that only the bits specified by Mask are used. If so, return the
/// simpler operand, otherwise return a null SDOperand.
-SDOperand DAGCombiner::GetDemandedBits(SDOperand V, uint64_t Mask) {
+SDOperand DAGCombiner::GetDemandedBits(SDOperand V, const APInt &Mask) {
switch (V.getOpcode()) {
default: break;
case ISD::OR:
if (ConstantSDNode *RHSC = dyn_cast<ConstantSDNode>(V.getOperand(1))) {
// See if we can recursively simplify the LHS.
unsigned Amt = RHSC->getValue();
- Mask = (Mask << Amt) & MVT::getIntVTBitMask(V.getValueType());
- SDOperand SimplifyLHS = GetDemandedBits(V.getOperand(0), Mask);
+ APInt NewMask = Mask << Amt;
+ SDOperand SimplifyLHS = GetDemandedBits(V.getOperand(0), NewMask);
if (SimplifyLHS.Val) {
return DAG.getNode(ISD::SRL, V.getValueType(),
SimplifyLHS, V.getOperand(1));
unsigned Opc = N->getOpcode();
ISD::LoadExtType ExtType = ISD::NON_EXTLOAD;
SDOperand N0 = N->getOperand(0);
- MVT::ValueType VT = N->getValueType(0);
- MVT::ValueType EVT = N->getValueType(0);
+ MVT VT = N->getValueType(0);
+ MVT EVT = N->getValueType(0);
// Special case: SIGN_EXTEND_INREG is basically truncating to EVT then
// extended to VT.
return SDOperand();
}
- unsigned EVTBits = MVT::getSizeInBits(EVT);
+ unsigned EVTBits = EVT.getSizeInBits();
unsigned ShAmt = 0;
bool CombineSRL = false;
if (N0.getOpcode() == ISD::SRL && N0.hasOneUse()) {
// Is the shift amount a multiple of size of VT?
if ((ShAmt & (EVTBits-1)) == 0) {
N0 = N0.getOperand(0);
- if (MVT::getSizeInBits(N0.getValueType()) <= EVTBits)
+ if (N0.getValueType().getSizeInBits() <= EVTBits)
return SDOperand();
CombineSRL = true;
}
}
}
- if (ISD::isNON_EXTLoad(N0.Val) && N0.hasOneUse() &&
- // Do not allow folding to i1 here. i1 is implicitly stored in memory in
- // zero extended form: by shrinking the load, we lose track of the fact
- // that it is already zero extended.
- // FIXME: This should be reevaluated.
- VT != MVT::i1) {
- assert(MVT::getSizeInBits(N0.getValueType()) > EVTBits &&
+ // Do not generate loads of non-round integer types since these can
+ // be expensive (and would be wrong if the type is not byte sized).
+ if (ISD::isNON_EXTLoad(N0.Val) && N0.hasOneUse() && VT.isRound() &&
+ // Do not change the width of a volatile load.
+ !cast<LoadSDNode>(N0)->isVolatile()) {
+ assert(N0.getValueType().getSizeInBits() > EVTBits &&
"Cannot truncate to larger type!");
LoadSDNode *LN0 = cast<LoadSDNode>(N0);
- MVT::ValueType PtrType = N0.getOperand(1).getValueType();
+ MVT PtrType = N0.getOperand(1).getValueType();
// For big endian targets, we need to adjust the offset to the pointer to
// load the correct bytes.
- if (!TLI.isLittleEndian()) {
- unsigned LVTStoreBits = MVT::getStoreSizeInBits(N0.getValueType());
- unsigned EVTStoreBits = MVT::getStoreSizeInBits(EVT);
+ if (TLI.isBigEndian()) {
+ unsigned LVTStoreBits = N0.getValueType().getStoreSizeInBits();
+ unsigned EVTStoreBits = EVT.getStoreSizeInBits();
ShAmt = LVTStoreBits - EVTStoreBits - ShAmt;
}
uint64_t PtrOff = ShAmt / 8;
LN0->isVolatile(), NewAlign);
AddToWorkList(N);
if (CombineSRL) {
- DAG.ReplaceAllUsesOfValueWith(N0.getValue(1), Load.getValue(1));
+ WorkListRemover DeadNodes(*this);
+ DAG.ReplaceAllUsesOfValueWith(N0.getValue(1), Load.getValue(1),
+ &DeadNodes);
CombineTo(N->getOperand(0).Val, Load);
} else
CombineTo(N0.Val, Load, Load.getValue(1));
SDOperand DAGCombiner::visitSIGN_EXTEND_INREG(SDNode *N) {
SDOperand N0 = N->getOperand(0);
SDOperand N1 = N->getOperand(1);
- MVT::ValueType VT = N->getValueType(0);
- MVT::ValueType EVT = cast<VTSDNode>(N1)->getVT();
- unsigned EVTBits = MVT::getSizeInBits(EVT);
+ MVT VT = N->getValueType(0);
+ MVT EVT = cast<VTSDNode>(N1)->getVT();
+ unsigned VTBits = VT.getSizeInBits();
+ unsigned EVTBits = EVT.getSizeInBits();
// fold (sext_in_reg c1) -> c1
if (isa<ConstantSDNode>(N0) || N0.getOpcode() == ISD::UNDEF)
return DAG.getNode(ISD::SIGN_EXTEND_INREG, VT, N0, N1);
// If the input is already sign extended, just drop the extension.
- if (DAG.ComputeNumSignBits(N0) >= MVT::getSizeInBits(VT)-EVTBits+1)
+ if (DAG.ComputeNumSignBits(N0) >= VT.getSizeInBits()-EVTBits+1)
return N0;
// fold (sext_in_reg (sext_in_reg x, VT2), VT1) -> (sext_in_reg x, minVT) pt2
if (N0.getOpcode() == ISD::SIGN_EXTEND_INREG &&
- EVT < cast<VTSDNode>(N0.getOperand(1))->getVT()) {
+ EVT.bitsLT(cast<VTSDNode>(N0.getOperand(1))->getVT())) {
return DAG.getNode(ISD::SIGN_EXTEND_INREG, VT, N0.getOperand(0), N1);
}
// fold (sext_in_reg x) -> (zext_in_reg x) if the sign bit is known zero.
- if (DAG.MaskedValueIsZero(N0, 1ULL << (EVTBits-1)))
+ if (DAG.MaskedValueIsZero(N0, APInt::getBitsSet(VTBits, EVTBits-1, EVTBits)))
return DAG.getZeroExtendInReg(N0, EVT);
// fold operands of sext_in_reg based on knowledge that the top bits are not
// We already fold "(sext_in_reg (srl X, 25), i8) -> srl X, 25" above.
if (N0.getOpcode() == ISD::SRL) {
if (ConstantSDNode *ShAmt = dyn_cast<ConstantSDNode>(N0.getOperand(1)))
- if (ShAmt->getValue()+EVTBits <= MVT::getSizeInBits(VT)) {
+ if (ShAmt->getValue()+EVTBits <= VT.getSizeInBits()) {
// We can turn this into an SRA iff the input to the SRL is already sign
// extended enough.
unsigned InSignBits = DAG.ComputeNumSignBits(N0.getOperand(0));
- if (MVT::getSizeInBits(VT)-(ShAmt->getValue()+EVTBits) < InSignBits)
+ if (VT.getSizeInBits()-(ShAmt->getValue()+EVTBits) < InSignBits)
return DAG.getNode(ISD::SRA, VT, N0.getOperand(0), N0.getOperand(1));
}
}
// fold (sext_inreg (extload x)) -> (sextload x)
if (ISD::isEXTLoad(N0.Val) &&
ISD::isUNINDEXEDLoad(N0.Val) &&
- EVT == cast<LoadSDNode>(N0)->getLoadedVT() &&
- (!AfterLegalize || TLI.isLoadXLegal(ISD::SEXTLOAD, EVT))) {
+ EVT == cast<LoadSDNode>(N0)->getMemoryVT() &&
+ ((!AfterLegalize && !cast<LoadSDNode>(N0)->isVolatile()) ||
+ TLI.isLoadXLegal(ISD::SEXTLOAD, EVT))) {
LoadSDNode *LN0 = cast<LoadSDNode>(N0);
SDOperand ExtLoad = DAG.getExtLoad(ISD::SEXTLOAD, VT, LN0->getChain(),
LN0->getBasePtr(), LN0->getSrcValue(),
// fold (sext_inreg (zextload x)) -> (sextload x) iff load has one use
if (ISD::isZEXTLoad(N0.Val) && ISD::isUNINDEXEDLoad(N0.Val) &&
N0.hasOneUse() &&
- EVT == cast<LoadSDNode>(N0)->getLoadedVT() &&
- (!AfterLegalize || TLI.isLoadXLegal(ISD::SEXTLOAD, EVT))) {
+ EVT == cast<LoadSDNode>(N0)->getMemoryVT() &&
+ ((!AfterLegalize && !cast<LoadSDNode>(N0)->isVolatile()) ||
+ TLI.isLoadXLegal(ISD::SEXTLOAD, EVT))) {
LoadSDNode *LN0 = cast<LoadSDNode>(N0);
SDOperand ExtLoad = DAG.getExtLoad(ISD::SEXTLOAD, VT, LN0->getChain(),
LN0->getBasePtr(), LN0->getSrcValue(),
SDOperand DAGCombiner::visitTRUNCATE(SDNode *N) {
SDOperand N0 = N->getOperand(0);
- MVT::ValueType VT = N->getValueType(0);
+ MVT VT = N->getValueType(0);
// noop truncate
if (N0.getValueType() == N->getValueType(0))
// fold (truncate (ext x)) -> (ext x) or (truncate x) or x
if (N0.getOpcode() == ISD::ZERO_EXTEND || N0.getOpcode() == ISD::SIGN_EXTEND||
N0.getOpcode() == ISD::ANY_EXTEND) {
- if (N0.getOperand(0).getValueType() < VT)
+ if (N0.getOperand(0).getValueType().bitsLT(VT))
// if the source is smaller than the dest, we still need an extend
return DAG.getNode(N0.getOpcode(), VT, N0.getOperand(0));
- else if (N0.getOperand(0).getValueType() > VT)
+ else if (N0.getOperand(0).getValueType().bitsGT(VT))
// if the source is larger than the dest, than we just need the truncate
return DAG.getNode(ISD::TRUNCATE, VT, N0.getOperand(0));
else
// See if we can simplify the input to this truncate through knowledge that
// only the low bits are being used. For example "trunc (or (shl x, 8), y)"
// -> trunc y
- SDOperand Shorter = GetDemandedBits(N0, MVT::getIntVTBitMask(VT));
+ SDOperand Shorter =
+ GetDemandedBits(N0, APInt::getLowBitsSet(N0.getValueSizeInBits(),
+ VT.getSizeInBits()));
if (Shorter.Val)
return DAG.getNode(ISD::TRUNCATE, VT, Shorter);
return ReduceLoadWidth(N);
}
+static SDNode *getBuildPairElt(SDNode *N, unsigned i) {
+ SDOperand Elt = N->getOperand(i);
+ if (Elt.getOpcode() != ISD::MERGE_VALUES)
+ return Elt.Val;
+ return Elt.getOperand(Elt.ResNo).Val;
+}
+
+/// CombineConsecutiveLoads - build_pair (load, load) -> load
+/// if load locations are consecutive.
+SDOperand DAGCombiner::CombineConsecutiveLoads(SDNode *N, MVT VT) {
+ assert(N->getOpcode() == ISD::BUILD_PAIR);
+
+ SDNode *LD1 = getBuildPairElt(N, 0);
+ if (!ISD::isNON_EXTLoad(LD1) || !LD1->hasOneUse())
+ return SDOperand();
+ MVT LD1VT = LD1->getValueType(0);
+ SDNode *LD2 = getBuildPairElt(N, 1);
+ const MachineFrameInfo *MFI = DAG.getMachineFunction().getFrameInfo();
+ if (ISD::isNON_EXTLoad(LD2) &&
+ LD2->hasOneUse() &&
+ // If both are volatile this would reduce the number of volatile loads.
+ // If one is volatile it might be ok, but play conservative and bail out.
+ !cast<LoadSDNode>(LD1)->isVolatile() &&
+ !cast<LoadSDNode>(LD2)->isVolatile() &&
+ TLI.isConsecutiveLoad(LD2, LD1, LD1VT.getSizeInBits()/8, 1, MFI)) {
+ LoadSDNode *LD = cast<LoadSDNode>(LD1);
+ unsigned Align = LD->getAlignment();
+ unsigned NewAlign = TLI.getTargetMachine().getTargetData()->
+ getABITypeAlignment(VT.getTypeForMVT());
+ if (NewAlign <= Align &&
+ (!AfterLegalize || TLI.isOperationLegal(ISD::LOAD, VT)))
+ return DAG.getLoad(VT, LD->getChain(), LD->getBasePtr(),
+ LD->getSrcValue(), LD->getSrcValueOffset(),
+ false, Align);
+ }
+ return SDOperand();
+}
+
SDOperand DAGCombiner::visitBIT_CONVERT(SDNode *N) {
SDOperand N0 = N->getOperand(0);
- MVT::ValueType VT = N->getValueType(0);
+ MVT VT = N->getValueType(0);
// If the input is a BUILD_VECTOR with all constant elements, fold this now.
// Only do this before legalize, since afterward the target may be depending
// First check to see if this is all constant.
if (!AfterLegalize &&
N0.getOpcode() == ISD::BUILD_VECTOR && N0.Val->hasOneUse() &&
- MVT::isVector(VT)) {
+ VT.isVector()) {
bool isSimple = true;
for (unsigned i = 0, e = N0.getNumOperands(); i != e; ++i)
if (N0.getOperand(i).getOpcode() != ISD::UNDEF &&
break;
}
- MVT::ValueType DestEltVT = MVT::getVectorElementType(N->getValueType(0));
- assert(!MVT::isVector(DestEltVT) &&
+ MVT DestEltVT = N->getValueType(0).getVectorElementType();
+ assert(!DestEltVT.isVector() &&
"Element type of vector ValueType must not be vector!");
if (isSimple) {
return ConstantFoldBIT_CONVERTofBUILD_VECTOR(N0.Val, DestEltVT);
// fold (conv (load x)) -> (load (conv*)x)
// If the resultant load doesn't need a higher alignment than the original!
if (ISD::isNormalLoad(N0.Val) && N0.hasOneUse() &&
- TLI.isOperationLegal(ISD::LOAD, VT)) {
+ // Do not change the width of a volatile load.
+ !cast<LoadSDNode>(N0)->isVolatile() &&
+ (!AfterLegalize || TLI.isOperationLegal(ISD::LOAD, VT))) {
LoadSDNode *LN0 = cast<LoadSDNode>(N0);
unsigned Align = TLI.getTargetMachine().getTargetData()->
- getABITypeAlignment(MVT::getTypeForValueType(VT));
+ getABITypeAlignment(VT.getTypeForMVT());
unsigned OrigAlign = LN0->getAlignment();
if (Align <= OrigAlign) {
SDOperand Load = DAG.getLoad(VT, LN0->getChain(), LN0->getBasePtr(),
return Load;
}
}
+
+ // Fold bitconvert(fneg(x)) -> xor(bitconvert(x), signbit)
+ // Fold bitconvert(fabs(x)) -> and(bitconvert(x), ~signbit)
+ // This often reduces constant pool loads.
+ if ((N0.getOpcode() == ISD::FNEG || N0.getOpcode() == ISD::FABS) &&
+ N0.Val->hasOneUse() && VT.isInteger() && !VT.isVector()) {
+ SDOperand NewConv = DAG.getNode(ISD::BIT_CONVERT, VT, N0.getOperand(0));
+ AddToWorkList(NewConv.Val);
+
+ APInt SignBit = APInt::getSignBit(VT.getSizeInBits());
+ if (N0.getOpcode() == ISD::FNEG)
+ return DAG.getNode(ISD::XOR, VT, NewConv, DAG.getConstant(SignBit, VT));
+ assert(N0.getOpcode() == ISD::FABS);
+ return DAG.getNode(ISD::AND, VT, NewConv, DAG.getConstant(~SignBit, VT));
+ }
+
+ // Fold bitconvert(fcopysign(cst, x)) -> bitconvert(x)&sign | cst&~sign'
+ // Note that we don't handle copysign(x,cst) because this can always be folded
+ // to an fneg or fabs.
+ if (N0.getOpcode() == ISD::FCOPYSIGN && N0.Val->hasOneUse() &&
+ isa<ConstantFPSDNode>(N0.getOperand(0)) &&
+ VT.isInteger() && !VT.isVector()) {
+ unsigned OrigXWidth = N0.getOperand(1).getValueType().getSizeInBits();
+ SDOperand X = DAG.getNode(ISD::BIT_CONVERT,
+ MVT::getIntegerVT(OrigXWidth),
+ N0.getOperand(1));
+ AddToWorkList(X.Val);
+
+ // If X has a different width than the result/lhs, sext it or truncate it.
+ unsigned VTWidth = VT.getSizeInBits();
+ if (OrigXWidth < VTWidth) {
+ X = DAG.getNode(ISD::SIGN_EXTEND, VT, X);
+ AddToWorkList(X.Val);
+ } else if (OrigXWidth > VTWidth) {
+ // To get the sign bit in the right place, we have to shift it right
+ // before truncating.
+ X = DAG.getNode(ISD::SRL, X.getValueType(), X,
+ DAG.getConstant(OrigXWidth-VTWidth, X.getValueType()));
+ AddToWorkList(X.Val);
+ X = DAG.getNode(ISD::TRUNCATE, VT, X);
+ AddToWorkList(X.Val);
+ }
+
+ APInt SignBit = APInt::getSignBit(VT.getSizeInBits());
+ X = DAG.getNode(ISD::AND, VT, X, DAG.getConstant(SignBit, VT));
+ AddToWorkList(X.Val);
+
+ SDOperand Cst = DAG.getNode(ISD::BIT_CONVERT, VT, N0.getOperand(0));
+ Cst = DAG.getNode(ISD::AND, VT, Cst, DAG.getConstant(~SignBit, VT));
+ AddToWorkList(Cst.Val);
+
+ return DAG.getNode(ISD::OR, VT, X, Cst);
+ }
+
+ // bitconvert(build_pair(ld, ld)) -> ld iff load locations are consecutive.
+ if (N0.getOpcode() == ISD::BUILD_PAIR) {
+ SDOperand CombineLD = CombineConsecutiveLoads(N0.Val, VT);
+ if (CombineLD.Val)
+ return CombineLD;
+ }
return SDOperand();
}
+SDOperand DAGCombiner::visitBUILD_PAIR(SDNode *N) {
+ MVT VT = N->getValueType(0);
+ return CombineConsecutiveLoads(N, VT);
+}
+
/// ConstantFoldBIT_CONVERTofBUILD_VECTOR - We know that BV is a build_vector
/// node with Constant, ConstantFP or Undef operands. DstEltVT indicates the
/// destination element value type.
SDOperand DAGCombiner::
-ConstantFoldBIT_CONVERTofBUILD_VECTOR(SDNode *BV, MVT::ValueType DstEltVT) {
- MVT::ValueType SrcEltVT = BV->getOperand(0).getValueType();
+ConstantFoldBIT_CONVERTofBUILD_VECTOR(SDNode *BV, MVT DstEltVT) {
+ MVT SrcEltVT = BV->getOperand(0).getValueType();
// If this is already the right type, we're done.
if (SrcEltVT == DstEltVT) return SDOperand(BV, 0);
- unsigned SrcBitSize = MVT::getSizeInBits(SrcEltVT);
- unsigned DstBitSize = MVT::getSizeInBits(DstEltVT);
+ unsigned SrcBitSize = SrcEltVT.getSizeInBits();
+ unsigned DstBitSize = DstEltVT.getSizeInBits();
// If this is a conversion of N elements of one type to N elements of another
// type, convert each element. This handles FP<->INT cases.
Ops.push_back(DAG.getNode(ISD::BIT_CONVERT, DstEltVT, BV->getOperand(i)));
AddToWorkList(Ops.back().Val);
}
- MVT::ValueType VT =
- MVT::getVectorType(DstEltVT,
- MVT::getVectorNumElements(BV->getValueType(0)));
+ MVT VT = MVT::getVectorVT(DstEltVT,
+ BV->getValueType(0).getVectorNumElements());
return DAG.getNode(ISD::BUILD_VECTOR, VT, &Ops[0], Ops.size());
}
// Otherwise, we're growing or shrinking the elements. To avoid having to
// handle annoying details of growing/shrinking FP values, we convert them to
// int first.
- if (MVT::isFloatingPoint(SrcEltVT)) {
+ if (SrcEltVT.isFloatingPoint()) {
// Convert the input float vector to a int vector where the elements are the
// same sizes.
assert((SrcEltVT == MVT::f32 || SrcEltVT == MVT::f64) && "Unknown FP VT!");
- MVT::ValueType IntVT = SrcEltVT == MVT::f32 ? MVT::i32 : MVT::i64;
+ MVT IntVT = MVT::getIntegerVT(SrcEltVT.getSizeInBits());
BV = ConstantFoldBIT_CONVERTofBUILD_VECTOR(BV, IntVT).Val;
SrcEltVT = IntVT;
}
// Now we know the input is an integer vector. If the output is a FP type,
// convert to integer first, then to FP of the right size.
- if (MVT::isFloatingPoint(DstEltVT)) {
+ if (DstEltVT.isFloatingPoint()) {
assert((DstEltVT == MVT::f32 || DstEltVT == MVT::f64) && "Unknown FP VT!");
- MVT::ValueType TmpVT = DstEltVT == MVT::f32 ? MVT::i32 : MVT::i64;
+ MVT TmpVT = MVT::getIntegerVT(DstEltVT.getSizeInBits());
SDNode *Tmp = ConstantFoldBIT_CONVERTofBUILD_VECTOR(BV, TmpVT).Val;
// Next, convert to FP elements of the same size.
// Okay, we know the src/dst types are both integers of differing types.
// Handling growing first.
- assert(MVT::isInteger(SrcEltVT) && MVT::isInteger(DstEltVT));
+ assert(SrcEltVT.isInteger() && DstEltVT.isInteger());
if (SrcBitSize < DstBitSize) {
unsigned NumInputsPerOutput = DstBitSize/SrcBitSize;
for (unsigned i = 0, e = BV->getNumOperands(); i != e;
i += NumInputsPerOutput) {
bool isLE = TLI.isLittleEndian();
- uint64_t NewBits = 0;
+ APInt NewBits = APInt(DstBitSize, 0);
bool EltIsUndef = true;
for (unsigned j = 0; j != NumInputsPerOutput; ++j) {
// Shift the previously computed bits over.
if (Op.getOpcode() == ISD::UNDEF) continue;
EltIsUndef = false;
- NewBits |= cast<ConstantSDNode>(Op)->getValue();
+ NewBits |=
+ APInt(cast<ConstantSDNode>(Op)->getAPIntValue()).zext(DstBitSize);
}
if (EltIsUndef)
Ops.push_back(DAG.getConstant(NewBits, DstEltVT));
}
- MVT::ValueType VT = MVT::getVectorType(DstEltVT,
- Ops.size());
+ MVT VT = MVT::getVectorVT(DstEltVT, Ops.size());
return DAG.getNode(ISD::BUILD_VECTOR, VT, &Ops[0], Ops.size());
}
// Finally, this must be the case where we are shrinking elements: each input
// turns into multiple outputs.
+ bool isS2V = ISD::isScalarToVector(BV);
unsigned NumOutputsPerInput = SrcBitSize/DstBitSize;
+ MVT VT = MVT::getVectorVT(DstEltVT, NumOutputsPerInput*BV->getNumOperands());
SmallVector<SDOperand, 8> Ops;
for (unsigned i = 0, e = BV->getNumOperands(); i != e; ++i) {
if (BV->getOperand(i).getOpcode() == ISD::UNDEF) {
Ops.push_back(DAG.getNode(ISD::UNDEF, DstEltVT));
continue;
}
- uint64_t OpVal = cast<ConstantSDNode>(BV->getOperand(i))->getValue();
-
+ APInt OpVal = cast<ConstantSDNode>(BV->getOperand(i))->getAPIntValue();
for (unsigned j = 0; j != NumOutputsPerInput; ++j) {
- unsigned ThisVal = OpVal & ((1ULL << DstBitSize)-1);
- OpVal >>= DstBitSize;
+ APInt ThisVal = APInt(OpVal).trunc(DstBitSize);
Ops.push_back(DAG.getConstant(ThisVal, DstEltVT));
+ if (isS2V && i == 0 && j == 0 && APInt(ThisVal).zext(SrcBitSize) == OpVal)
+ // Simply turn this into a SCALAR_TO_VECTOR of the new type.
+ return DAG.getNode(ISD::SCALAR_TO_VECTOR, VT, Ops[0]);
+ OpVal = OpVal.lshr(DstBitSize);
}
// For big endian targets, swap the order of the pieces of each element.
- if (!TLI.isLittleEndian())
+ if (TLI.isBigEndian())
std::reverse(Ops.end()-NumOutputsPerInput, Ops.end());
}
- MVT::ValueType VT = MVT::getVectorType(DstEltVT, Ops.size());
return DAG.getNode(ISD::BUILD_VECTOR, VT, &Ops[0], Ops.size());
}
SDOperand N1 = N->getOperand(1);
ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0);
ConstantFPSDNode *N1CFP = dyn_cast<ConstantFPSDNode>(N1);
- MVT::ValueType VT = N->getValueType(0);
+ MVT VT = N->getValueType(0);
// fold vector ops
- if (MVT::isVector(VT)) {
+ if (VT.isVector()) {
SDOperand FoldedVOp = SimplifyVBinOp(N);
if (FoldedVOp.Val) return FoldedVOp;
}
if (N0CFP && !N1CFP)
return DAG.getNode(ISD::FADD, VT, N1, N0);
// fold (A + (-B)) -> A-B
- if (isNegatibleForFree(N1) == 2)
- return DAG.getNode(ISD::FSUB, VT, N0, GetNegatedExpression(N1, DAG));
+ if (isNegatibleForFree(N1, AfterLegalize) == 2)
+ return DAG.getNode(ISD::FSUB, VT, N0,
+ GetNegatedExpression(N1, DAG, AfterLegalize));
// fold ((-A) + B) -> B-A
- if (isNegatibleForFree(N0) == 2)
- return DAG.getNode(ISD::FSUB, VT, N1, GetNegatedExpression(N0, DAG));
+ if (isNegatibleForFree(N0, AfterLegalize) == 2)
+ return DAG.getNode(ISD::FSUB, VT, N1,
+ GetNegatedExpression(N0, DAG, AfterLegalize));
// If allowed, fold (fadd (fadd x, c1), c2) -> (fadd x, (fadd c1, c2))
if (UnsafeFPMath && N1CFP && N0.getOpcode() == ISD::FADD &&
SDOperand N1 = N->getOperand(1);
ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0);
ConstantFPSDNode *N1CFP = dyn_cast<ConstantFPSDNode>(N1);
- MVT::ValueType VT = N->getValueType(0);
+ MVT VT = N->getValueType(0);
// fold vector ops
- if (MVT::isVector(VT)) {
+ if (VT.isVector()) {
SDOperand FoldedVOp = SimplifyVBinOp(N);
if (FoldedVOp.Val) return FoldedVOp;
}
return DAG.getNode(ISD::FSUB, VT, N0, N1);
// fold (0-B) -> -B
if (UnsafeFPMath && N0CFP && N0CFP->getValueAPF().isZero()) {
- if (isNegatibleForFree(N1))
- return GetNegatedExpression(N1, DAG);
+ if (isNegatibleForFree(N1, AfterLegalize))
+ return GetNegatedExpression(N1, DAG, AfterLegalize);
return DAG.getNode(ISD::FNEG, VT, N1);
}
// fold (A-(-B)) -> A+B
- if (isNegatibleForFree(N1))
- return DAG.getNode(ISD::FADD, VT, N0, GetNegatedExpression(N1, DAG));
+ if (isNegatibleForFree(N1, AfterLegalize))
+ return DAG.getNode(ISD::FADD, VT, N0,
+ GetNegatedExpression(N1, DAG, AfterLegalize));
return SDOperand();
}
SDOperand N1 = N->getOperand(1);
ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0);
ConstantFPSDNode *N1CFP = dyn_cast<ConstantFPSDNode>(N1);
- MVT::ValueType VT = N->getValueType(0);
+ MVT VT = N->getValueType(0);
// fold vector ops
- if (MVT::isVector(VT)) {
+ if (VT.isVector()) {
SDOperand FoldedVOp = SimplifyVBinOp(N);
if (FoldedVOp.Val) return FoldedVOp;
}
return DAG.getNode(ISD::FNEG, VT, N0);
// -X * -Y -> X*Y
- if (char LHSNeg = isNegatibleForFree(N0)) {
- if (char RHSNeg = isNegatibleForFree(N1)) {
+ if (char LHSNeg = isNegatibleForFree(N0, AfterLegalize)) {
+ if (char RHSNeg = isNegatibleForFree(N1, AfterLegalize)) {
// Both can be negated for free, check to see if at least one is cheaper
// negated.
if (LHSNeg == 2 || RHSNeg == 2)
- return DAG.getNode(ISD::FMUL, VT, GetNegatedExpression(N0, DAG),
- GetNegatedExpression(N1, DAG));
+ return DAG.getNode(ISD::FMUL, VT,
+ GetNegatedExpression(N0, DAG, AfterLegalize),
+ GetNegatedExpression(N1, DAG, AfterLegalize));
}
}
SDOperand N1 = N->getOperand(1);
ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0);
ConstantFPSDNode *N1CFP = dyn_cast<ConstantFPSDNode>(N1);
- MVT::ValueType VT = N->getValueType(0);
+ MVT VT = N->getValueType(0);
// fold vector ops
- if (MVT::isVector(VT)) {
+ if (VT.isVector()) {
SDOperand FoldedVOp = SimplifyVBinOp(N);
if (FoldedVOp.Val) return FoldedVOp;
}
// -X / -Y -> X*Y
- if (char LHSNeg = isNegatibleForFree(N0)) {
- if (char RHSNeg = isNegatibleForFree(N1)) {
+ if (char LHSNeg = isNegatibleForFree(N0, AfterLegalize)) {
+ if (char RHSNeg = isNegatibleForFree(N1, AfterLegalize)) {
// Both can be negated for free, check to see if at least one is cheaper
// negated.
if (LHSNeg == 2 || RHSNeg == 2)
- return DAG.getNode(ISD::FDIV, VT, GetNegatedExpression(N0, DAG),
- GetNegatedExpression(N1, DAG));
+ return DAG.getNode(ISD::FDIV, VT,
+ GetNegatedExpression(N0, DAG, AfterLegalize),
+ GetNegatedExpression(N1, DAG, AfterLegalize));
}
}
SDOperand N1 = N->getOperand(1);
ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0);
ConstantFPSDNode *N1CFP = dyn_cast<ConstantFPSDNode>(N1);
- MVT::ValueType VT = N->getValueType(0);
+ MVT VT = N->getValueType(0);
// fold (frem c1, c2) -> fmod(c1,c2)
if (N0CFP && N1CFP && VT != MVT::ppcf128)
SDOperand N1 = N->getOperand(1);
ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0);
ConstantFPSDNode *N1CFP = dyn_cast<ConstantFPSDNode>(N1);
- MVT::ValueType VT = N->getValueType(0);
+ MVT VT = N->getValueType(0);
if (N0CFP && N1CFP && VT != MVT::ppcf128) // Constant fold
return DAG.getNode(ISD::FCOPYSIGN, VT, N0, N1);
SDOperand DAGCombiner::visitSINT_TO_FP(SDNode *N) {
SDOperand N0 = N->getOperand(0);
ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0);
- MVT::ValueType VT = N->getValueType(0);
+ MVT VT = N->getValueType(0);
// fold (sint_to_fp c1) -> c1fp
if (N0C && N0.getValueType() != MVT::ppcf128)
SDOperand DAGCombiner::visitUINT_TO_FP(SDNode *N) {
SDOperand N0 = N->getOperand(0);
ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0);
- MVT::ValueType VT = N->getValueType(0);
+ MVT VT = N->getValueType(0);
// fold (uint_to_fp c1) -> c1fp
if (N0C && N0.getValueType() != MVT::ppcf128)
SDOperand DAGCombiner::visitFP_TO_SINT(SDNode *N) {
SDOperand N0 = N->getOperand(0);
ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0);
- MVT::ValueType VT = N->getValueType(0);
+ MVT VT = N->getValueType(0);
// fold (fp_to_sint c1fp) -> c1
if (N0CFP)
SDOperand DAGCombiner::visitFP_TO_UINT(SDNode *N) {
SDOperand N0 = N->getOperand(0);
ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0);
- MVT::ValueType VT = N->getValueType(0);
+ MVT VT = N->getValueType(0);
// fold (fp_to_uint c1fp) -> c1
if (N0CFP && VT != MVT::ppcf128)
SDOperand DAGCombiner::visitFP_ROUND(SDNode *N) {
SDOperand N0 = N->getOperand(0);
+ SDOperand N1 = N->getOperand(1);
ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0);
- MVT::ValueType VT = N->getValueType(0);
+ MVT VT = N->getValueType(0);
// fold (fp_round c1fp) -> c1fp
if (N0CFP && N0.getValueType() != MVT::ppcf128)
- return DAG.getNode(ISD::FP_ROUND, VT, N0);
+ return DAG.getNode(ISD::FP_ROUND, VT, N0, N1);
// fold (fp_round (fp_extend x)) -> x
if (N0.getOpcode() == ISD::FP_EXTEND && VT == N0.getOperand(0).getValueType())
return N0.getOperand(0);
+ // fold (fp_round (fp_round x)) -> (fp_round x)
+ if (N0.getOpcode() == ISD::FP_ROUND) {
+ // This is a value preserving truncation if both round's are.
+ bool IsTrunc = N->getConstantOperandVal(1) == 1 &&
+ N0.Val->getConstantOperandVal(1) == 1;
+ return DAG.getNode(ISD::FP_ROUND, VT, N0.getOperand(0),
+ DAG.getIntPtrConstant(IsTrunc));
+ }
+
// fold (fp_round (copysign X, Y)) -> (copysign (fp_round X), Y)
if (N0.getOpcode() == ISD::FCOPYSIGN && N0.Val->hasOneUse()) {
- SDOperand Tmp = DAG.getNode(ISD::FP_ROUND, VT, N0.getOperand(0));
+ SDOperand Tmp = DAG.getNode(ISD::FP_ROUND, VT, N0.getOperand(0), N1);
AddToWorkList(Tmp.Val);
return DAG.getNode(ISD::FCOPYSIGN, VT, Tmp, N0.getOperand(1));
}
SDOperand DAGCombiner::visitFP_ROUND_INREG(SDNode *N) {
SDOperand N0 = N->getOperand(0);
- MVT::ValueType VT = N->getValueType(0);
- MVT::ValueType EVT = cast<VTSDNode>(N->getOperand(1))->getVT();
+ MVT VT = N->getValueType(0);
+ MVT EVT = cast<VTSDNode>(N->getOperand(1))->getVT();
ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0);
// fold (fp_round_inreg c1fp) -> c1fp
SDOperand DAGCombiner::visitFP_EXTEND(SDNode *N) {
SDOperand N0 = N->getOperand(0);
ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0);
- MVT::ValueType VT = N->getValueType(0);
+ MVT VT = N->getValueType(0);
+ // If this is fp_round(fpextend), don't fold it, allow ourselves to be folded.
+ if (N->hasOneUse() &&
+ N->use_begin()->getSDOperand().getOpcode() == ISD::FP_ROUND)
+ return SDOperand();
+
// fold (fp_extend c1fp) -> c1fp
if (N0CFP && VT != MVT::ppcf128)
return DAG.getNode(ISD::FP_EXTEND, VT, N0);
-
- // fold (fpext (load x)) -> (fpext (fpround (extload x)))
+
+ // Turn fp_extend(fp_round(X, 1)) -> x since the fp_round doesn't affect the
+ // value of X.
+ if (N0.getOpcode() == ISD::FP_ROUND && N0.Val->getConstantOperandVal(1) == 1){
+ SDOperand In = N0.getOperand(0);
+ if (In.getValueType() == VT) return In;
+ if (VT.bitsLT(In.getValueType()))
+ return DAG.getNode(ISD::FP_ROUND, VT, In, N0.getOperand(1));
+ return DAG.getNode(ISD::FP_EXTEND, VT, In);
+ }
+
+ // fold (fpext (load x)) -> (fpext (fptrunc (extload x)))
if (ISD::isNON_EXTLoad(N0.Val) && N0.hasOneUse() &&
- (!AfterLegalize||TLI.isLoadXLegal(ISD::EXTLOAD, N0.getValueType()))) {
+ ((!AfterLegalize && !cast<LoadSDNode>(N0)->isVolatile()) ||
+ TLI.isLoadXLegal(ISD::EXTLOAD, N0.getValueType()))) {
LoadSDNode *LN0 = cast<LoadSDNode>(N0);
SDOperand ExtLoad = DAG.getExtLoad(ISD::EXTLOAD, VT, LN0->getChain(),
LN0->getBasePtr(), LN0->getSrcValue(),
LN0->isVolatile(),
LN0->getAlignment());
CombineTo(N, ExtLoad);
- CombineTo(N0.Val, DAG.getNode(ISD::FP_ROUND, N0.getValueType(), ExtLoad),
+ CombineTo(N0.Val, DAG.getNode(ISD::FP_ROUND, N0.getValueType(), ExtLoad,
+ DAG.getIntPtrConstant(1)),
ExtLoad.getValue(1));
return SDOperand(N, 0); // Return N so it doesn't get rechecked!
}
-
-
+
return SDOperand();
}
SDOperand DAGCombiner::visitFNEG(SDNode *N) {
SDOperand N0 = N->getOperand(0);
- if (isNegatibleForFree(N0))
- return GetNegatedExpression(N0, DAG);
-
+ if (isNegatibleForFree(N0, AfterLegalize))
+ return GetNegatedExpression(N0, DAG, AfterLegalize);
+
+ // Transform fneg(bitconvert(x)) -> bitconvert(x^sign) to avoid loading
+ // constant pool values.
+ if (N0.getOpcode() == ISD::BIT_CONVERT && N0.Val->hasOneUse() &&
+ N0.getOperand(0).getValueType().isInteger() &&
+ !N0.getOperand(0).getValueType().isVector()) {
+ SDOperand Int = N0.getOperand(0);
+ MVT IntVT = Int.getValueType();
+ if (IntVT.isInteger() && !IntVT.isVector()) {
+ Int = DAG.getNode(ISD::XOR, IntVT, Int,
+ DAG.getConstant(IntVT.getIntegerVTSignBit(), IntVT));
+ AddToWorkList(Int.Val);
+ return DAG.getNode(ISD::BIT_CONVERT, N->getValueType(0), Int);
+ }
+ }
+
return SDOperand();
}
SDOperand DAGCombiner::visitFABS(SDNode *N) {
SDOperand N0 = N->getOperand(0);
ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0);
- MVT::ValueType VT = N->getValueType(0);
+ MVT VT = N->getValueType(0);
// fold (fabs c1) -> fabs(c1)
if (N0CFP && VT != MVT::ppcf128)
if (N0.getOpcode() == ISD::FNEG || N0.getOpcode() == ISD::FCOPYSIGN)
return DAG.getNode(ISD::FABS, VT, N0.getOperand(0));
+ // Transform fabs(bitconvert(x)) -> bitconvert(x&~sign) to avoid loading
+ // constant pool values.
+ if (N0.getOpcode() == ISD::BIT_CONVERT && N0.Val->hasOneUse() &&
+ N0.getOperand(0).getValueType().isInteger() &&
+ !N0.getOperand(0).getValueType().isVector()) {
+ SDOperand Int = N0.getOperand(0);
+ MVT IntVT = Int.getValueType();
+ if (IntVT.isInteger() && !IntVT.isVector()) {
+ Int = DAG.getNode(ISD::AND, IntVT, Int,
+ DAG.getConstant(~IntVT.getIntegerVTSignBit(), IntVT));
+ AddToWorkList(Int.Val);
+ return DAG.getNode(ISD::BIT_CONVERT, N->getValueType(0), Int);
+ }
+ }
+
return SDOperand();
}
if (N1C && N1C->isNullValue())
return Chain;
// unconditional branch
- if (N1C && N1C->getValue() == 1)
+ if (N1C && N1C->getAPIntValue() == 1)
return DAG.getNode(ISD::BR, MVT::Other, Chain, N2);
// fold a brcond with a setcc condition into a BR_CC node if BR_CC is legal
// on the target.
CondCodeSDNode *CC = cast<CondCodeSDNode>(N->getOperand(1));
SDOperand CondLHS = N->getOperand(2), CondRHS = N->getOperand(3);
- // Use SimplifySetCC to simplify SETCC's.
+ // Use SimplifySetCC to simplify SETCC's.
SDOperand Simp = SimplifySetCC(MVT::i1, CondLHS, CondRHS, CC->get(), false);
if (Simp.Val) AddToWorkList(Simp.Val);
ConstantSDNode *SCCC = dyn_cast_or_null<ConstantSDNode>(Simp.Val);
// fold br_cc true, dest -> br dest (unconditional branch)
- if (SCCC && SCCC->getValue())
+ if (SCCC && !SCCC->isNullValue())
return DAG.getNode(ISD::BR, MVT::Other, N->getOperand(0),
N->getOperand(4));
// fold br_cc false, dest -> unconditional fall through
}
-/// CombineToPreIndexedLoadStore - Try turning a load / store and a
-/// pre-indexed load / store when the base pointer is a add or subtract
+/// CombineToPreIndexedLoadStore - Try turning a load / store into a
+/// pre-indexed load / store when the base pointer is an add or subtract
/// and it has other uses besides the load / store. After the
/// transformation, the new indexed load / store has effectively folded
/// the add / subtract in and all of its other uses are redirected to the
bool isLoad = true;
SDOperand Ptr;
- MVT::ValueType VT;
+ MVT VT;
if (LoadSDNode *LD = dyn_cast<LoadSDNode>(N)) {
- if (LD->getAddressingMode() != ISD::UNINDEXED)
+ if (LD->isIndexed())
return false;
- VT = LD->getLoadedVT();
+ VT = LD->getMemoryVT();
if (!TLI.isIndexedLoadLegal(ISD::PRE_INC, VT) &&
!TLI.isIndexedLoadLegal(ISD::PRE_DEC, VT))
return false;
Ptr = LD->getBasePtr();
} else if (StoreSDNode *ST = dyn_cast<StoreSDNode>(N)) {
- if (ST->getAddressingMode() != ISD::UNINDEXED)
+ if (ST->isIndexed())
return false;
- VT = ST->getStoredVT();
+ VT = ST->getMemoryVT();
if (!TLI.isIndexedStoreLegal(ISD::PRE_INC, VT) &&
!TLI.isIndexedStoreLegal(ISD::PRE_DEC, VT))
return false;
return false;
// Don't create a indexed load / store with zero offset.
if (isa<ConstantSDNode>(Offset) &&
- cast<ConstantSDNode>(Offset)->getValue() == 0)
+ cast<ConstantSDNode>(Offset)->isNullValue())
return false;
// Try turning it into a pre-indexed load / store except when:
// Check #2.
if (!isLoad) {
SDOperand Val = cast<StoreSDNode>(N)->getValue();
- if (Val == BasePtr || BasePtr.Val->isPredecessor(Val.Val))
+ if (Val == BasePtr || BasePtr.Val->isPredecessorOf(Val.Val))
return false;
}
bool RealUse = false;
for (SDNode::use_iterator I = Ptr.Val->use_begin(),
E = Ptr.Val->use_end(); I != E; ++I) {
- SDNode *Use = *I;
+ SDNode *Use = I->getUser();
if (Use == N)
continue;
- if (Use->isPredecessor(N))
+ if (Use->isPredecessorOf(N))
return false;
if (!((Use->getOpcode() == ISD::LOAD &&
cast<LoadSDNode>(Use)->getBasePtr() == Ptr) ||
- (Use->getOpcode() == ISD::STORE) &&
- cast<StoreSDNode>(Use)->getBasePtr() == Ptr))
+ (Use->getOpcode() == ISD::STORE &&
+ cast<StoreSDNode>(Use)->getBasePtr() == Ptr)))
RealUse = true;
}
if (!RealUse)
DOUT << "\nReplacing.4 "; DEBUG(N->dump(&DAG));
DOUT << "\nWith: "; DEBUG(Result.Val->dump(&DAG));
DOUT << '\n';
- std::vector<SDNode*> NowDead;
+ WorkListRemover DeadNodes(*this);
if (isLoad) {
DAG.ReplaceAllUsesOfValueWith(SDOperand(N, 0), Result.getValue(0),
- &NowDead);
+ &DeadNodes);
DAG.ReplaceAllUsesOfValueWith(SDOperand(N, 1), Result.getValue(2),
- &NowDead);
+ &DeadNodes);
} else {
DAG.ReplaceAllUsesOfValueWith(SDOperand(N, 0), Result.getValue(1),
- &NowDead);
+ &DeadNodes);
}
- // Nodes can end up on the worklist more than once. Make sure we do
- // not process a node that has been replaced.
- for (unsigned i = 0, e = NowDead.size(); i != e; ++i)
- removeFromWorkList(NowDead[i]);
// Finally, since the node is now dead, remove it from the graph.
DAG.DeleteNode(N);
// Replace the uses of Ptr with uses of the updated base value.
DAG.ReplaceAllUsesOfValueWith(Ptr, Result.getValue(isLoad ? 1 : 0),
- &NowDead);
+ &DeadNodes);
removeFromWorkList(Ptr.Val);
- for (unsigned i = 0, e = NowDead.size(); i != e; ++i)
- removeFromWorkList(NowDead[i]);
DAG.DeleteNode(Ptr.Val);
return true;
}
-/// CombineToPostIndexedLoadStore - Try combine a load / store with a
+/// CombineToPostIndexedLoadStore - Try to combine a load / store with a
/// add / sub of the base pointer node into a post-indexed load / store.
/// The transformation folded the add / subtract into the new indexed
/// load / store effectively and all of its uses are redirected to the
bool isLoad = true;
SDOperand Ptr;
- MVT::ValueType VT;
+ MVT VT;
if (LoadSDNode *LD = dyn_cast<LoadSDNode>(N)) {
- if (LD->getAddressingMode() != ISD::UNINDEXED)
+ if (LD->isIndexed())
return false;
- VT = LD->getLoadedVT();
+ VT = LD->getMemoryVT();
if (!TLI.isIndexedLoadLegal(ISD::POST_INC, VT) &&
!TLI.isIndexedLoadLegal(ISD::POST_DEC, VT))
return false;
Ptr = LD->getBasePtr();
} else if (StoreSDNode *ST = dyn_cast<StoreSDNode>(N)) {
- if (ST->getAddressingMode() != ISD::UNINDEXED)
+ if (ST->isIndexed())
return false;
- VT = ST->getStoredVT();
+ VT = ST->getMemoryVT();
if (!TLI.isIndexedStoreLegal(ISD::POST_INC, VT) &&
!TLI.isIndexedStoreLegal(ISD::POST_DEC, VT))
return false;
for (SDNode::use_iterator I = Ptr.Val->use_begin(),
E = Ptr.Val->use_end(); I != E; ++I) {
- SDNode *Op = *I;
+ SDNode *Op = I->getUser();
if (Op == N ||
(Op->getOpcode() != ISD::ADD && Op->getOpcode() != ISD::SUB))
continue;
continue;
// Don't create a indexed load / store with zero offset.
if (isa<ConstantSDNode>(Offset) &&
- cast<ConstantSDNode>(Offset)->getValue() == 0)
+ cast<ConstantSDNode>(Offset)->isNullValue())
continue;
// Try turning it into a post-indexed load / store except when
bool TryNext = false;
for (SDNode::use_iterator II = BasePtr.Val->use_begin(),
EE = BasePtr.Val->use_end(); II != EE; ++II) {
- SDNode *Use = *II;
+ SDNode *Use = II->getUser();
if (Use == Ptr.Val)
continue;
bool RealUse = false;
for (SDNode::use_iterator III = Use->use_begin(),
EEE = Use->use_end(); III != EEE; ++III) {
- SDNode *UseUse = *III;
+ SDNode *UseUse = III->getUser();
if (!((UseUse->getOpcode() == ISD::LOAD &&
cast<LoadSDNode>(UseUse)->getBasePtr().Val == Use) ||
- (UseUse->getOpcode() == ISD::STORE) &&
- cast<StoreSDNode>(UseUse)->getBasePtr().Val == Use))
+ (UseUse->getOpcode() == ISD::STORE &&
+ cast<StoreSDNode>(UseUse)->getBasePtr().Val == Use)))
RealUse = true;
}
continue;
// Check for #2
- if (!Op->isPredecessor(N) && !N->isPredecessor(Op)) {
+ if (!Op->isPredecessorOf(N) && !N->isPredecessorOf(Op)) {
SDOperand Result = isLoad
? DAG.getIndexedLoad(SDOperand(N,0), BasePtr, Offset, AM)
: DAG.getIndexedStore(SDOperand(N,0), BasePtr, Offset, AM);
DOUT << "\nReplacing.5 "; DEBUG(N->dump(&DAG));
DOUT << "\nWith: "; DEBUG(Result.Val->dump(&DAG));
DOUT << '\n';
- std::vector<SDNode*> NowDead;
+ WorkListRemover DeadNodes(*this);
if (isLoad) {
DAG.ReplaceAllUsesOfValueWith(SDOperand(N, 0), Result.getValue(0),
- &NowDead);
+ &DeadNodes);
DAG.ReplaceAllUsesOfValueWith(SDOperand(N, 1), Result.getValue(2),
- &NowDead);
+ &DeadNodes);
} else {
DAG.ReplaceAllUsesOfValueWith(SDOperand(N, 0), Result.getValue(1),
- &NowDead);
+ &DeadNodes);
}
- // Nodes can end up on the worklist more than once. Make sure we do
- // not process a node that has been replaced.
- for (unsigned i = 0, e = NowDead.size(); i != e; ++i)
- removeFromWorkList(NowDead[i]);
// Finally, since the node is now dead, remove it from the graph.
DAG.DeleteNode(N);
// Replace the uses of Use with uses of the updated base value.
DAG.ReplaceAllUsesOfValueWith(SDOperand(Op, 0),
Result.getValue(isLoad ? 1 : 0),
- &NowDead);
+ &DeadNodes);
removeFromWorkList(Op);
- for (unsigned i = 0, e = NowDead.size(); i != e; ++i)
- removeFromWorkList(NowDead[i]);
DAG.DeleteNode(Op);
-
return true;
}
}
return false;
}
+/// InferAlignment - If we can infer some alignment information from this
+/// pointer, return it.
+static unsigned InferAlignment(SDOperand Ptr, SelectionDAG &DAG) {
+ // If this is a direct reference to a stack slot, use information about the
+ // stack slot's alignment.
+ int FrameIdx = 1 << 31;
+ int64_t FrameOffset = 0;
+ if (FrameIndexSDNode *FI = dyn_cast<FrameIndexSDNode>(Ptr)) {
+ FrameIdx = FI->getIndex();
+ } else if (Ptr.getOpcode() == ISD::ADD &&
+ isa<ConstantSDNode>(Ptr.getOperand(1)) &&
+ isa<FrameIndexSDNode>(Ptr.getOperand(0))) {
+ FrameIdx = cast<FrameIndexSDNode>(Ptr.getOperand(0))->getIndex();
+ FrameOffset = Ptr.getConstantOperandVal(1);
+ }
+
+ if (FrameIdx != (1 << 31)) {
+ // FIXME: Handle FI+CST.
+ const MachineFrameInfo &MFI = *DAG.getMachineFunction().getFrameInfo();
+ if (MFI.isFixedObjectIndex(FrameIdx)) {
+ int64_t ObjectOffset = MFI.getObjectOffset(FrameIdx);
+
+ // The alignment of the frame index can be determined from its offset from
+ // the incoming frame position. If the frame object is at offset 32 and
+ // the stack is guaranteed to be 16-byte aligned, then we know that the
+ // object is 16-byte aligned.
+ unsigned StackAlign = DAG.getTarget().getFrameInfo()->getStackAlignment();
+ unsigned Align = MinAlign(ObjectOffset, StackAlign);
+
+ // Finally, the frame object itself may have a known alignment. Factor
+ // the alignment + offset into a new alignment. For example, if we know
+ // the FI is 8 byte aligned, but the pointer is 4 off, we really have a
+ // 4-byte alignment of the resultant pointer. Likewise align 4 + 4-byte
+ // offset = 4-byte alignment, align 4 + 1-byte offset = align 1, etc.
+ unsigned FIInfoAlign = MinAlign(MFI.getObjectAlignment(FrameIdx),
+ FrameOffset);
+ return std::max(Align, FIInfoAlign);
+ }
+ }
+
+ return 0;
+}
SDOperand DAGCombiner::visitLOAD(SDNode *N) {
LoadSDNode *LD = cast<LoadSDNode>(N);
SDOperand Chain = LD->getChain();
SDOperand Ptr = LD->getBasePtr();
+
+ // Try to infer better alignment information than the load already has.
+ if (LD->isUnindexed()) {
+ if (unsigned Align = InferAlignment(Ptr, DAG)) {
+ if (Align > LD->getAlignment())
+ return DAG.getExtLoad(LD->getExtensionType(), LD->getValueType(0),
+ Chain, Ptr, LD->getSrcValue(),
+ LD->getSrcValueOffset(), LD->getMemoryVT(),
+ LD->isVolatile(), Align);
+ }
+ }
+
// If load is not volatile and there are no uses of the loaded value (and
// the updated indexed value in case of indexed loads), change uses of the
if (!LD->isVolatile()) {
if (N->getValueType(1) == MVT::Other) {
// Unindexed loads.
- if (N->hasNUsesOfValue(0, 0))
- return CombineTo(N, DAG.getNode(ISD::UNDEF, N->getValueType(0)), Chain);
+ if (N->hasNUsesOfValue(0, 0)) {
+ // It's not safe to use the two value CombineTo variant here. e.g.
+ // v1, chain2 = load chain1, loc
+ // v2, chain3 = load chain2, loc
+ // v3 = add v2, c
+ // Now we replace use of chain2 with chain1. This makes the second load
+ // isomorphic to the one we are deleting, and thus makes this load live.
+ DOUT << "\nReplacing.6 "; DEBUG(N->dump(&DAG));
+ DOUT << "\nWith chain: "; DEBUG(Chain.Val->dump(&DAG));
+ DOUT << "\n";
+ WorkListRemover DeadNodes(*this);
+ DAG.ReplaceAllUsesOfValueWith(SDOperand(N, 1), Chain, &DeadNodes);
+ if (N->use_empty()) {
+ removeFromWorkList(N);
+ DAG.DeleteNode(N);
+ }
+ return SDOperand(N, 0); // Return N so it doesn't get rechecked!
+ }
} else {
// Indexed loads.
assert(N->getValueType(2) == MVT::Other && "Malformed indexed loads?");
if (N->hasNUsesOfValue(0, 0) && N->hasNUsesOfValue(0, 1)) {
- SDOperand Undef0 = DAG.getNode(ISD::UNDEF, N->getValueType(0));
- SDOperand Undef1 = DAG.getNode(ISD::UNDEF, N->getValueType(1));
- SDOperand To[] = { Undef0, Undef1, Chain };
- return CombineTo(N, To, 3);
+ SDOperand Undef = DAG.getNode(ISD::UNDEF, N->getValueType(0));
+ DOUT << "\nReplacing.6 "; DEBUG(N->dump(&DAG));
+ DOUT << "\nWith: "; DEBUG(Undef.Val->dump(&DAG));
+ DOUT << " and 2 other values\n";
+ WorkListRemover DeadNodes(*this);
+ DAG.ReplaceAllUsesOfValueWith(SDOperand(N, 0), Undef, &DeadNodes);
+ DAG.ReplaceAllUsesOfValueWith(SDOperand(N, 1),
+ DAG.getNode(ISD::UNDEF, N->getValueType(1)),
+ &DeadNodes);
+ DAG.ReplaceAllUsesOfValueWith(SDOperand(N, 2), Chain, &DeadNodes);
+ removeFromWorkList(N);
+ DAG.DeleteNode(N);
+ return SDOperand(N, 0); // Return N so it doesn't get rechecked!
}
}
}
// value.
// TODO: Handle store large -> read small portion.
// TODO: Handle TRUNCSTORE/LOADEXT
- if (LD->getExtensionType() == ISD::NON_EXTLOAD) {
+ if (LD->getExtensionType() == ISD::NON_EXTLOAD &&
+ !LD->isVolatile()) {
if (ISD::isNON_TRUNCStore(Chain.Val)) {
StoreSDNode *PrevST = cast<StoreSDNode>(Chain);
if (PrevST->getBasePtr() == Ptr &&
LD->getValueType(0),
BetterChain, Ptr, LD->getSrcValue(),
LD->getSrcValueOffset(),
- LD->getLoadedVT(),
+ LD->getMemoryVT(),
LD->isVolatile(),
LD->getAlignment());
}
return SDOperand();
}
+
SDOperand DAGCombiner::visitSTORE(SDNode *N) {
StoreSDNode *ST = cast<StoreSDNode>(N);
SDOperand Chain = ST->getChain();
SDOperand Value = ST->getValue();
SDOperand Ptr = ST->getBasePtr();
+ // Try to infer better alignment information than the store already has.
+ if (ST->isUnindexed()) {
+ if (unsigned Align = InferAlignment(Ptr, DAG)) {
+ if (Align > ST->getAlignment())
+ return DAG.getTruncStore(Chain, Value, Ptr, ST->getSrcValue(),
+ ST->getSrcValueOffset(), ST->getMemoryVT(),
+ ST->isVolatile(), Align);
+ }
+ }
+
// If this is a store of a bit convert, store the input value if the
// resultant store does not need a higher alignment than the original.
if (Value.getOpcode() == ISD::BIT_CONVERT && !ST->isTruncatingStore() &&
- ST->getAddressingMode() == ISD::UNINDEXED) {
+ ST->isUnindexed()) {
unsigned Align = ST->getAlignment();
- MVT::ValueType SVT = Value.getOperand(0).getValueType();
+ MVT SVT = Value.getOperand(0).getValueType();
unsigned OrigAlign = TLI.getTargetMachine().getTargetData()->
- getABITypeAlignment(MVT::getTypeForValueType(SVT));
- if (Align <= OrigAlign && TLI.isOperationLegal(ISD::STORE, SVT))
+ getABITypeAlignment(SVT.getTypeForMVT());
+ if (Align <= OrigAlign &&
+ ((!AfterLegalize && !ST->isVolatile()) ||
+ TLI.isOperationLegal(ISD::STORE, SVT)))
return DAG.getStore(Chain, Value.getOperand(0), Ptr, ST->getSrcValue(),
ST->getSrcValueOffset(), ST->isVolatile(), Align);
}
-
+
// Turn 'store float 1.0, Ptr' -> 'store int 0x12345678, Ptr'
if (ConstantFPSDNode *CFP = dyn_cast<ConstantFPSDNode>(Value)) {
+ // NOTE: If the original store is volatile, this transform must not increase
+ // the number of stores. For example, on x86-32 an f64 can be stored in one
+ // processor operation but an i64 (which is not legal) requires two. So the
+ // transform should not be done in this case.
if (Value.getOpcode() != ISD::TargetConstantFP) {
SDOperand Tmp;
- switch (CFP->getValueType(0)) {
+ switch (CFP->getValueType(0).getSimpleVT()) {
default: assert(0 && "Unknown FP type");
case MVT::f80: // We don't do this for these yet.
case MVT::f128:
case MVT::ppcf128:
break;
case MVT::f32:
- if (!AfterLegalize || TLI.isTypeLegal(MVT::i32)) {
+ if ((!AfterLegalize && !ST->isVolatile()) ||
+ TLI.isOperationLegal(ISD::STORE, MVT::i32)) {
Tmp = DAG.getConstant((uint32_t)CFP->getValueAPF().
convertToAPInt().getZExtValue(), MVT::i32);
return DAG.getStore(Chain, Tmp, Ptr, ST->getSrcValue(),
}
break;
case MVT::f64:
- if (!AfterLegalize || TLI.isTypeLegal(MVT::i64)) {
+ if ((!AfterLegalize && !ST->isVolatile()) ||
+ TLI.isOperationLegal(ISD::STORE, MVT::i64)) {
Tmp = DAG.getConstant(CFP->getValueAPF().convertToAPInt().
getZExtValue(), MVT::i64);
return DAG.getStore(Chain, Tmp, Ptr, ST->getSrcValue(),
ST->getSrcValueOffset(), ST->isVolatile(),
ST->getAlignment());
- } else if (TLI.isTypeLegal(MVT::i32)) {
+ } else if (!ST->isVolatile() &&
+ TLI.isOperationLegal(ISD::STORE, MVT::i32)) {
// Many FP stores are not made apparent until after legalize, e.g. for
// argument passing. Since this is so common, custom legalize the
// 64-bit integer store into two 32-bit stores.
uint64_t Val = CFP->getValueAPF().convertToAPInt().getZExtValue();
SDOperand Lo = DAG.getConstant(Val & 0xFFFFFFFF, MVT::i32);
SDOperand Hi = DAG.getConstant(Val >> 32, MVT::i32);
- if (!TLI.isLittleEndian()) std::swap(Lo, Hi);
+ if (TLI.isBigEndian()) std::swap(Lo, Hi);
int SVOffset = ST->getSrcValueOffset();
unsigned Alignment = ST->getAlignment();
SDOperand ReplStore;
if (ST->isTruncatingStore()) {
ReplStore = DAG.getTruncStore(BetterChain, Value, Ptr,
- ST->getSrcValue(), ST->getSrcValueOffset(), ST->getStoredVT(),
- ST->isVolatile(), ST->getAlignment());
+ ST->getSrcValue(),ST->getSrcValueOffset(),
+ ST->getMemoryVT(),
+ ST->isVolatile(), ST->getAlignment());
} else {
ReplStore = DAG.getStore(BetterChain, Value, Ptr,
- ST->getSrcValue(), ST->getSrcValueOffset(),
- ST->isVolatile(), ST->getAlignment());
+ ST->getSrcValue(), ST->getSrcValueOffset(),
+ ST->isVolatile(), ST->getAlignment());
}
// Create token to keep both nodes around.
if (CombineToPreIndexedLoadStore(N) || CombineToPostIndexedLoadStore(N))
return SDOperand(N, 0);
- // FIXME: is there such a think as a truncating indexed store?
- if (ST->isTruncatingStore() && ST->getAddressingMode() == ISD::UNINDEXED &&
- MVT::isInteger(Value.getValueType())) {
+ // FIXME: is there such a thing as a truncating indexed store?
+ if (ST->isTruncatingStore() && ST->isUnindexed() &&
+ Value.getValueType().isInteger()) {
// See if we can simplify the input to this truncstore with knowledge that
// only the low bits are being used. For example:
// "truncstore (or (shl x, 8), y), i8" -> "truncstore y, i8"
SDOperand Shorter =
- GetDemandedBits(Value, MVT::getIntVTBitMask(ST->getStoredVT()));
+ GetDemandedBits(Value,
+ APInt::getLowBitsSet(Value.getValueSizeInBits(),
+ ST->getMemoryVT().getSizeInBits()));
AddToWorkList(Value.Val);
if (Shorter.Val)
return DAG.getTruncStore(Chain, Shorter, Ptr, ST->getSrcValue(),
- ST->getSrcValueOffset(), ST->getStoredVT(),
+ ST->getSrcValueOffset(), ST->getMemoryVT(),
ST->isVolatile(), ST->getAlignment());
// Otherwise, see if we can simplify the operation with
// SimplifyDemandedBits, which only works if the value has a single use.
- if (SimplifyDemandedBits(Value, MVT::getIntVTBitMask(ST->getStoredVT())))
+ if (SimplifyDemandedBits(Value,
+ APInt::getLowBitsSet(
+ Value.getValueSizeInBits(),
+ ST->getMemoryVT().getSizeInBits())))
return SDOperand(N, 0);
}
+ // If this is a load followed by a store to the same location, then the store
+ // is dead/noop.
+ if (LoadSDNode *Ld = dyn_cast<LoadSDNode>(Value)) {
+ if (Ld->getBasePtr() == Ptr && ST->getMemoryVT() == Ld->getMemoryVT() &&
+ ST->isUnindexed() && !ST->isVolatile() &&
+ // There can't be any side effects between the load and store, such as
+ // a call or store.
+ Chain.reachesChainWithoutSideEffects(SDOperand(Ld, 1))) {
+ // The store is dead, remove it.
+ return Chain;
+ }
+ }
+
+ // If this is an FP_ROUND or TRUNC followed by a store, fold this into a
+ // truncating store. We can do this even if this is already a truncstore.
+ if ((Value.getOpcode() == ISD::FP_ROUND || Value.getOpcode() == ISD::TRUNCATE)
+ && Value.Val->hasOneUse() && ST->isUnindexed() &&
+ TLI.isTruncStoreLegal(Value.getOperand(0).getValueType(),
+ ST->getMemoryVT())) {
+ return DAG.getTruncStore(Chain, Value.getOperand(0), Ptr, ST->getSrcValue(),
+ ST->getSrcValueOffset(), ST->getMemoryVT(),
+ ST->isVolatile(), ST->getAlignment());
+ }
+
return SDOperand();
}
}
SDOperand DAGCombiner::visitEXTRACT_VECTOR_ELT(SDNode *N) {
+ // (vextract (v4f32 load $addr), c) -> (f32 load $addr+c*size)
+ // (vextract (v4f32 s2v (f32 load $addr)), c) -> (f32 load $addr+c*size)
+ // (vextract (v4f32 shuffle (load $addr), <1,u,u,u>), 0) -> (f32 load $addr)
+
+ // Perform only after legalization to ensure build_vector / vector_shuffle
+ // optimizations have already been done.
+ if (!AfterLegalize) return SDOperand();
+
SDOperand InVec = N->getOperand(0);
SDOperand EltNo = N->getOperand(1);
- // (vextract (v4f32 s2v (f32 load $addr)), 0) -> (f32 load $addr)
- // (vextract (v4i32 bc (v4f32 s2v (f32 load $addr))), 0) -> (i32 load $addr)
if (isa<ConstantSDNode>(EltNo)) {
unsigned Elt = cast<ConstantSDNode>(EltNo)->getValue();
bool NewLoad = false;
- if (Elt == 0) {
- MVT::ValueType VT = InVec.getValueType();
- MVT::ValueType EVT = MVT::getVectorElementType(VT);
- MVT::ValueType LVT = EVT;
- unsigned NumElts = MVT::getVectorNumElements(VT);
- if (InVec.getOpcode() == ISD::BIT_CONVERT) {
- MVT::ValueType BCVT = InVec.getOperand(0).getValueType();
- if (!MVT::isVector(BCVT) ||
- NumElts != MVT::getVectorNumElements(BCVT))
- return SDOperand();
+ MVT VT = InVec.getValueType();
+ MVT EVT = VT.getVectorElementType();
+ MVT LVT = EVT;
+ if (InVec.getOpcode() == ISD::BIT_CONVERT) {
+ MVT BCVT = InVec.getOperand(0).getValueType();
+ if (!BCVT.isVector() || EVT.bitsGT(BCVT.getVectorElementType()))
+ return SDOperand();
+ InVec = InVec.getOperand(0);
+ EVT = BCVT.getVectorElementType();
+ NewLoad = true;
+ }
+
+ LoadSDNode *LN0 = NULL;
+ if (ISD::isNormalLoad(InVec.Val))
+ LN0 = cast<LoadSDNode>(InVec);
+ else if (InVec.getOpcode() == ISD::SCALAR_TO_VECTOR &&
+ InVec.getOperand(0).getValueType() == EVT &&
+ ISD::isNormalLoad(InVec.getOperand(0).Val)) {
+ LN0 = cast<LoadSDNode>(InVec.getOperand(0));
+ } else if (InVec.getOpcode() == ISD::VECTOR_SHUFFLE) {
+ // (vextract (vector_shuffle (load $addr), v2, <1, u, u, u>), 1)
+ // =>
+ // (load $addr+1*size)
+ unsigned Idx = cast<ConstantSDNode>(InVec.getOperand(2).
+ getOperand(Elt))->getValue();
+ unsigned NumElems = InVec.getOperand(2).getNumOperands();
+ InVec = (Idx < NumElems) ? InVec.getOperand(0) : InVec.getOperand(1);
+ if (InVec.getOpcode() == ISD::BIT_CONVERT)
InVec = InVec.getOperand(0);
- EVT = MVT::getVectorElementType(BCVT);
- NewLoad = true;
+ if (ISD::isNormalLoad(InVec.Val)) {
+ LN0 = cast<LoadSDNode>(InVec);
+ Elt = (Idx < NumElems) ? Idx : Idx - NumElems;
}
- if (InVec.getOpcode() == ISD::SCALAR_TO_VECTOR &&
- InVec.getOperand(0).getValueType() == EVT &&
- ISD::isNormalLoad(InVec.getOperand(0).Val) &&
- InVec.getOperand(0).hasOneUse()) {
- LoadSDNode *LN0 = cast<LoadSDNode>(InVec.getOperand(0));
- unsigned Align = LN0->getAlignment();
- if (NewLoad) {
- // Check the resultant load doesn't need a higher alignment than the
- // original load.
- unsigned NewAlign = TLI.getTargetMachine().getTargetData()->
- getABITypeAlignment(MVT::getTypeForValueType(LVT));
- if (!TLI.isOperationLegal(ISD::LOAD, LVT) || NewAlign > Align)
- return SDOperand();
- Align = NewAlign;
- }
+ }
+ if (!LN0 || !LN0->hasOneUse() || LN0->isVolatile())
+ return SDOperand();
- return DAG.getLoad(LVT, LN0->getChain(), LN0->getBasePtr(),
- LN0->getSrcValue(), LN0->getSrcValueOffset(),
- LN0->isVolatile(), Align);
- }
+ unsigned Align = LN0->getAlignment();
+ if (NewLoad) {
+ // Check the resultant load doesn't need a higher alignment than the
+ // original load.
+ unsigned NewAlign = TLI.getTargetMachine().getTargetData()->
+ getABITypeAlignment(LVT.getTypeForMVT());
+ if (NewAlign > Align || !TLI.isOperationLegal(ISD::LOAD, LVT))
+ return SDOperand();
+ Align = NewAlign;
}
+
+ SDOperand NewPtr = LN0->getBasePtr();
+ if (Elt) {
+ unsigned PtrOff = LVT.getSizeInBits() * Elt / 8;
+ MVT PtrType = NewPtr.getValueType();
+ if (TLI.isBigEndian())
+ PtrOff = VT.getSizeInBits() / 8 - PtrOff;
+ NewPtr = DAG.getNode(ISD::ADD, PtrType, NewPtr,
+ DAG.getConstant(PtrOff, PtrType));
+ }
+ return DAG.getLoad(LVT, LN0->getChain(), NewPtr,
+ LN0->getSrcValue(), LN0->getSrcValueOffset(),
+ LN0->isVolatile(), Align);
}
return SDOperand();
}
SDOperand DAGCombiner::visitBUILD_VECTOR(SDNode *N) {
unsigned NumInScalars = N->getNumOperands();
- MVT::ValueType VT = N->getValueType(0);
- unsigned NumElts = MVT::getVectorNumElements(VT);
- MVT::ValueType EltType = MVT::getVectorElementType(VT);
+ MVT VT = N->getValueType(0);
+ unsigned NumElts = VT.getVectorNumElements();
+ MVT EltType = VT.getVectorElementType();
// Check to see if this is a BUILD_VECTOR of a bunch of EXTRACT_VECTOR_ELT
// operations. If so, and if the EXTRACT_VECTOR_ELT vector inputs come from
// Otherwise, use InIdx + VecSize
unsigned Idx = cast<ConstantSDNode>(Extract.getOperand(1))->getValue();
- BuildVecIndices.push_back(DAG.getConstant(Idx+NumInScalars,
- TLI.getPointerTy()));
+ BuildVecIndices.push_back(DAG.getIntPtrConstant(Idx+NumInScalars));
}
// Add count and size info.
- MVT::ValueType BuildVecVT =
- MVT::getVectorType(TLI.getPointerTy(), NumElts);
+ MVT BuildVecVT = MVT::getVectorVT(TLI.getPointerTy(), NumElts);
// Return the new VECTOR_SHUFFLE node.
SDOperand Ops[5];
// look though conversions that change things like v4f32 to v2f64.
if (V->getOpcode() == ISD::BIT_CONVERT) {
SDOperand ConvInput = V->getOperand(0);
- if (MVT::getVectorNumElements(ConvInput.getValueType()) == NumElts)
+ if (ConvInput.getValueType().getVectorNumElements() == NumElts)
V = ConvInput.Val;
}
std::vector<SDOperand> IdxOps;
unsigned NumOps = RHS.getNumOperands();
unsigned NumElts = NumOps;
- MVT::ValueType EVT = MVT::getVectorElementType(RHS.getValueType());
+ MVT EVT = RHS.getValueType().getVectorElementType();
for (unsigned i = 0; i != NumElts; ++i) {
SDOperand Elt = RHS.getOperand(i);
if (!isa<ConstantSDNode>(Elt))
return SDOperand();
// Return the new VECTOR_SHUFFLE node.
- MVT::ValueType VT = MVT::getVectorType(EVT, NumElts);
+ MVT VT = MVT::getVectorVT(EVT, NumElts);
std::vector<SDOperand> Ops;
LHS = DAG.getNode(ISD::BIT_CONVERT, VT, LHS);
Ops.push_back(LHS);
// things. Simplifying them may result in a loss of legality.
if (AfterLegalize) return SDOperand();
- MVT::ValueType VT = N->getValueType(0);
- assert(MVT::isVector(VT) && "SimplifyVBinOp only works on vectors!");
+ MVT VT = N->getValueType(0);
+ assert(VT.isVector() && "SimplifyVBinOp only works on vectors!");
- MVT::ValueType EltType = MVT::getVectorElementType(VT);
+ MVT EltType = VT.getVectorElementType();
SDOperand LHS = N->getOperand(0);
SDOperand RHS = N->getOperand(1);
SDOperand Shuffle = XformToShuffleWithZero(N);
}
if (Ops.size() == LHS.getNumOperands()) {
- MVT::ValueType VT = LHS.getValueType();
+ MVT VT = LHS.getValueType();
return DAG.getNode(ISD::BUILD_VECTOR, VT, &Ops[0], Ops.size());
}
}
// This triggers in things like "select bool X, 10.0, 123.0" after the FP
// constants have been dropped into the constant pool.
if (LHS.getOpcode() == ISD::LOAD &&
+ // Do not let this transformation reduce the number of volatile loads.
+ !cast<LoadSDNode>(LHS)->isVolatile() &&
+ !cast<LoadSDNode>(RHS)->isVolatile() &&
// Token chains must be identical.
LHS.getOperand(0) == RHS.getOperand(0)) {
LoadSDNode *LLD = cast<LoadSDNode>(LHS);
LoadSDNode *RLD = cast<LoadSDNode>(RHS);
// If this is an EXTLOAD, the VT's must match.
- if (LLD->getLoadedVT() == RLD->getLoadedVT()) {
+ if (LLD->getMemoryVT() == RLD->getMemoryVT()) {
// FIXME: this conflates two src values, discarding one. This is not
// the right thing to do, but nothing uses srcvalues now. When they do,
// turn SrcValue into a list of locations.
if (TheSelect->getOpcode() == ISD::SELECT) {
// Check that the condition doesn't reach either load. If so, folding
// this will induce a cycle into the DAG.
- if (!LLD->isPredecessor(TheSelect->getOperand(0).Val) &&
- !RLD->isPredecessor(TheSelect->getOperand(0).Val)) {
+ if (!LLD->isPredecessorOf(TheSelect->getOperand(0).Val) &&
+ !RLD->isPredecessorOf(TheSelect->getOperand(0).Val)) {
Addr = DAG.getNode(ISD::SELECT, LLD->getBasePtr().getValueType(),
TheSelect->getOperand(0), LLD->getBasePtr(),
RLD->getBasePtr());
} else {
// Check that the condition doesn't reach either load. If so, folding
// this will induce a cycle into the DAG.
- if (!LLD->isPredecessor(TheSelect->getOperand(0).Val) &&
- !RLD->isPredecessor(TheSelect->getOperand(0).Val) &&
- !LLD->isPredecessor(TheSelect->getOperand(1).Val) &&
- !RLD->isPredecessor(TheSelect->getOperand(1).Val)) {
+ if (!LLD->isPredecessorOf(TheSelect->getOperand(0).Val) &&
+ !RLD->isPredecessorOf(TheSelect->getOperand(0).Val) &&
+ !LLD->isPredecessorOf(TheSelect->getOperand(1).Val) &&
+ !RLD->isPredecessorOf(TheSelect->getOperand(1).Val)) {
Addr = DAG.getNode(ISD::SELECT_CC, LLD->getBasePtr().getValueType(),
TheSelect->getOperand(0),
TheSelect->getOperand(1),
TheSelect->getValueType(0),
LLD->getChain(), Addr, LLD->getSrcValue(),
LLD->getSrcValueOffset(),
- LLD->getLoadedVT(),
+ LLD->getMemoryVT(),
LLD->isVolatile(),
LLD->getAlignment());
}
SDOperand N2, SDOperand N3,
ISD::CondCode CC, bool NotExtCompare) {
- MVT::ValueType VT = N2.getValueType();
+ MVT VT = N2.getValueType();
ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1.Val);
ConstantSDNode *N2C = dyn_cast<ConstantSDNode>(N2.Val);
ConstantSDNode *N3C = dyn_cast<ConstantSDNode>(N3.Val);
// Determine if the condition we're dealing with is constant
- SDOperand SCC = SimplifySetCC(TLI.getSetCCResultTy(), N0, N1, CC, false);
+ SDOperand SCC = SimplifySetCC(TLI.getSetCCResultType(N0), N0, N1, CC, false);
if (SCC.Val) AddToWorkList(SCC.Val);
ConstantSDNode *SCCC = dyn_cast_or_null<ConstantSDNode>(SCC.Val);
// fold select_cc true, x, y -> x
- if (SCCC && SCCC->getValue())
+ if (SCCC && !SCCC->isNullValue())
return N2;
// fold select_cc false, x, y -> y
- if (SCCC && SCCC->getValue() == 0)
+ if (SCCC && SCCC->isNullValue())
return N3;
// Check to see if we can simplify the select into an fabs node
// Check to see if we can perform the "gzip trick", transforming
// select_cc setlt X, 0, A, 0 -> and (sra X, size(X)-1), A
if (N1C && N3C && N3C->isNullValue() && CC == ISD::SETLT &&
- MVT::isInteger(N0.getValueType()) &&
- MVT::isInteger(N2.getValueType()) &&
- (N1C->isNullValue() || // (a < 0) ? b : 0
- (N1C->getValue() == 1 && N0 == N2))) { // (a < 1) ? a : 0
- MVT::ValueType XType = N0.getValueType();
- MVT::ValueType AType = N2.getValueType();
- if (XType >= AType) {
+ N0.getValueType().isInteger() &&
+ N2.getValueType().isInteger() &&
+ (N1C->isNullValue() || // (a < 0) ? b : 0
+ (N1C->getAPIntValue() == 1 && N0 == N2))) { // (a < 1) ? a : 0
+ MVT XType = N0.getValueType();
+ MVT AType = N2.getValueType();
+ if (XType.bitsGE(AType)) {
// and (sra X, size(X)-1, A) -> "and (srl X, C2), A" iff A is a
// single-bit constant.
- if (N2C && ((N2C->getValue() & (N2C->getValue()-1)) == 0)) {
- unsigned ShCtV = Log2_64(N2C->getValue());
- ShCtV = MVT::getSizeInBits(XType)-ShCtV-1;
+ if (N2C && ((N2C->getAPIntValue() & (N2C->getAPIntValue()-1)) == 0)) {
+ unsigned ShCtV = N2C->getAPIntValue().logBase2();
+ ShCtV = XType.getSizeInBits()-ShCtV-1;
SDOperand ShCt = DAG.getConstant(ShCtV, TLI.getShiftAmountTy());
SDOperand Shift = DAG.getNode(ISD::SRL, XType, N0, ShCt);
AddToWorkList(Shift.Val);
- if (XType > AType) {
+ if (XType.bitsGT(AType)) {
Shift = DAG.getNode(ISD::TRUNCATE, AType, Shift);
AddToWorkList(Shift.Val);
}
return DAG.getNode(ISD::AND, AType, Shift, N2);
}
SDOperand Shift = DAG.getNode(ISD::SRA, XType, N0,
- DAG.getConstant(MVT::getSizeInBits(XType)-1,
+ DAG.getConstant(XType.getSizeInBits()-1,
TLI.getShiftAmountTy()));
AddToWorkList(Shift.Val);
- if (XType > AType) {
+ if (XType.bitsGT(AType)) {
Shift = DAG.getNode(ISD::TRUNCATE, AType, Shift);
AddToWorkList(Shift.Val);
}
}
// fold select C, 16, 0 -> shl C, 4
- if (N2C && N3C && N3C->isNullValue() && isPowerOf2_64(N2C->getValue()) &&
+ if (N2C && N3C && N3C->isNullValue() && N2C->getAPIntValue().isPowerOf2() &&
TLI.getSetCCResultContents() == TargetLowering::ZeroOrOneSetCCResult) {
// If the caller doesn't want us to simplify this into a zext of a compare,
// don't do it.
- if (NotExtCompare && N2C->getValue() == 1)
+ if (NotExtCompare && N2C->getAPIntValue() == 1)
return SDOperand();
// Get a SetCC of the condition
SDOperand Temp, SCC;
// cast from setcc result type to select result type
if (AfterLegalize) {
- SCC = DAG.getSetCC(TLI.getSetCCResultTy(), N0, N1, CC);
- if (N2.getValueType() < SCC.getValueType())
+ SCC = DAG.getSetCC(TLI.getSetCCResultType(N0), N0, N1, CC);
+ if (N2.getValueType().bitsLT(SCC.getValueType()))
Temp = DAG.getZeroExtendInReg(SCC, N2.getValueType());
else
Temp = DAG.getNode(ISD::ZERO_EXTEND, N2.getValueType(), SCC);
AddToWorkList(SCC.Val);
AddToWorkList(Temp.Val);
- if (N2C->getValue() == 1)
+ if (N2C->getAPIntValue() == 1)
return Temp;
// shl setcc result by log2 n2c
return DAG.getNode(ISD::SHL, N2.getValueType(), Temp,
- DAG.getConstant(Log2_64(N2C->getValue()),
+ DAG.getConstant(N2C->getAPIntValue().logBase2(),
TLI.getShiftAmountTy()));
}
// Check to see if this is the equivalent of setcc
// FIXME: Turn all of these into setcc if setcc if setcc is legal
// otherwise, go ahead with the folds.
- if (0 && N3C && N3C->isNullValue() && N2C && (N2C->getValue() == 1ULL)) {
- MVT::ValueType XType = N0.getValueType();
- if (TLI.isOperationLegal(ISD::SETCC, TLI.getSetCCResultTy())) {
- SDOperand Res = DAG.getSetCC(TLI.getSetCCResultTy(), N0, N1, CC);
+ if (0 && N3C && N3C->isNullValue() && N2C && (N2C->getAPIntValue() == 1ULL)) {
+ MVT XType = N0.getValueType();
+ if (!AfterLegalize ||
+ TLI.isOperationLegal(ISD::SETCC, TLI.getSetCCResultType(N0))) {
+ SDOperand Res = DAG.getSetCC(TLI.getSetCCResultType(N0), N0, N1, CC);
if (Res.getValueType() != VT)
Res = DAG.getNode(ISD::ZERO_EXTEND, VT, Res);
return Res;
// seteq X, 0 -> srl (ctlz X, log2(size(X)))
if (N1C && N1C->isNullValue() && CC == ISD::SETEQ &&
- TLI.isOperationLegal(ISD::CTLZ, XType)) {
+ (!AfterLegalize ||
+ TLI.isOperationLegal(ISD::CTLZ, XType))) {
SDOperand Ctlz = DAG.getNode(ISD::CTLZ, XType, N0);
return DAG.getNode(ISD::SRL, XType, Ctlz,
- DAG.getConstant(Log2_32(MVT::getSizeInBits(XType)),
+ DAG.getConstant(Log2_32(XType.getSizeInBits()),
TLI.getShiftAmountTy()));
}
// setgt X, 0 -> srl (and (-X, ~X), size(X)-1)
DAG.getConstant(~0ULL, XType));
return DAG.getNode(ISD::SRL, XType,
DAG.getNode(ISD::AND, XType, NegN0, NotN0),
- DAG.getConstant(MVT::getSizeInBits(XType)-1,
+ DAG.getConstant(XType.getSizeInBits()-1,
TLI.getShiftAmountTy()));
}
// setgt X, -1 -> xor (srl (X, size(X)-1), 1)
if (N1C && N1C->isAllOnesValue() && CC == ISD::SETGT) {
SDOperand Sign = DAG.getNode(ISD::SRL, XType, N0,
- DAG.getConstant(MVT::getSizeInBits(XType)-1,
+ DAG.getConstant(XType.getSizeInBits()-1,
TLI.getShiftAmountTy()));
return DAG.getNode(ISD::XOR, XType, Sign, DAG.getConstant(1, XType));
}
// Y = sra (X, size(X)-1); xor (add (X, Y), Y)
if (N1C && N1C->isNullValue() && (CC == ISD::SETLT || CC == ISD::SETLE) &&
N0 == N3 && N2.getOpcode() == ISD::SUB && N0 == N2.getOperand(1) &&
- N2.getOperand(0) == N1 && MVT::isInteger(N0.getValueType())) {
- MVT::ValueType XType = N0.getValueType();
+ N2.getOperand(0) == N1 && N0.getValueType().isInteger()) {
+ MVT XType = N0.getValueType();
SDOperand Shift = DAG.getNode(ISD::SRA, XType, N0,
- DAG.getConstant(MVT::getSizeInBits(XType)-1,
+ DAG.getConstant(XType.getSizeInBits()-1,
TLI.getShiftAmountTy()));
SDOperand Add = DAG.getNode(ISD::ADD, XType, N0, Shift);
AddToWorkList(Shift.Val);
if (N1C && N1C->isAllOnesValue() && CC == ISD::SETGT &&
N0 == N2 && N3.getOpcode() == ISD::SUB && N0 == N3.getOperand(1)) {
if (ConstantSDNode *SubC = dyn_cast<ConstantSDNode>(N3.getOperand(0))) {
- MVT::ValueType XType = N0.getValueType();
- if (SubC->isNullValue() && MVT::isInteger(XType)) {
+ MVT XType = N0.getValueType();
+ if (SubC->isNullValue() && XType.isInteger()) {
SDOperand Shift = DAG.getNode(ISD::SRA, XType, N0,
- DAG.getConstant(MVT::getSizeInBits(XType)-1,
+ DAG.getConstant(XType.getSizeInBits()-1,
TLI.getShiftAmountTy()));
SDOperand Add = DAG.getNode(ISD::ADD, XType, N0, Shift);
AddToWorkList(Shift.Val);
}
/// SimplifySetCC - This is a stub for TargetLowering::SimplifySetCC.
-SDOperand DAGCombiner::SimplifySetCC(MVT::ValueType VT, SDOperand N0,
+SDOperand DAGCombiner::SimplifySetCC(MVT VT, SDOperand N0,
SDOperand N1, ISD::CondCode Cond,
bool foldBooleans) {
TargetLowering::DAGCombinerInfo
const Value *&SrcValue, int &SrcValueOffset) {
if (LoadSDNode *LD = dyn_cast<LoadSDNode>(N)) {
Ptr = LD->getBasePtr();
- Size = MVT::getSizeInBits(LD->getLoadedVT()) >> 3;
+ Size = LD->getMemoryVT().getSizeInBits() >> 3;
SrcValue = LD->getSrcValue();
SrcValueOffset = LD->getSrcValueOffset();
return true;
} else if (StoreSDNode *ST = dyn_cast<StoreSDNode>(N)) {
Ptr = ST->getBasePtr();
- Size = MVT::getSizeInBits(ST->getStoredVT()) >> 3;
+ Size = ST->getMemoryVT().getSizeInBits() >> 3;
SrcValue = ST->getSrcValue();
SrcValueOffset = ST->getSrcValueOffset();
} else {
projects / oota-llvm.git / blobdiff