[DAG] Move DAGCombiner::GetDemandedBits to SelectionDAG

[RKSimon]

Summary

This patch moves the DAGCombiner::GetDemandedBits function to SelectionDAG::GetDemandedBits as a first step towards making it easier for targets to get to the source of any demanded bits without the limitations of SimplifyDemandedBits.

[niravd]

LGTM.

llvm/trunk/include/llvm/CodeGen/SelectionDAG.h

public:
   SDValue FoldConstantVectorArithmetic(unsigned Opcode, const SDLoc &DL, EVT VT,
                                        ArrayRef<SDValue> Ops,
                                        const SDNodeFlags Flags = SDNodeFlags());
 
   /// Constant fold a setcc to true or false.
   SDValue FoldSetCC(EVT VT, SDValue N1, SDValue N2, ISD::CondCode Cond,
                     const SDLoc &dl);
 
+  /// 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 SDValue.
+  ///
+  /// (This exists alongside SimplifyDemandedBits because GetDemandedBits can
+  /// simplify nodes with multiple uses more aggressively.)
+  SDValue GetDemandedBits(SDValue V, const APInt &Mask);
+
   /// Return true if the sign bit of Op is known to be zero.
   /// We use this predicate to simplify operations downstream.
   bool SignBitIsZero(SDValue Op, unsigned Depth = 0) const;
 
   /// Return true if 'Op & Mask' is known to be zero.  We
   /// use this predicate to simplify operations downstream.  Op and Mask are
   /// known to be the same type.
   bool MaskedValueIsZero(SDValue Op, const APInt &Mask, unsigned Depth = 0)

llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

private:
     SDValue reduceBuildVecConvertToConvertBuildVec(SDNode *N);
     SDValue reduceBuildVecToShuffle(SDNode *N);
     SDValue reduceBuildVecToTrunc(SDNode *N);
     SDValue createBuildVecShuffle(const SDLoc &DL, SDNode *N,
                                   ArrayRef<int> VectorMask, SDValue VecIn1,
                                   SDValue VecIn2, unsigned LeftIdx);
     SDValue matchVSelectOpSizesWithSetCC(SDNode *N);
 
-    SDValue GetDemandedBits(SDValue V, const APInt &Mask);
-
     /// Walk up chain skipping non-aliasing memory nodes,
     /// looking for aliasing nodes and adding them to the Aliases vector.
     void GatherAllAliases(SDNode *N, SDValue OriginalChain,
                           SmallVectorImpl<SDValue> &Aliases);
 
     /// Return true if there is any possibility that the two addresses overlap.
     bool isAlias(LSBaseSDNode *Op0, LSBaseSDNode *Op1) const;
 
SDValue DAGCombiner::visitAssertZext(SDNode *N) {
   // fold (assertzext (assertzext x, vt), vt) -> (assertzext x, vt)
   if (N0.getOpcode() == ISD::AssertZext &&
       EVT == cast<VTSDNode>(N0.getOperand(1))->getVT())
     return N0;
 
   return SDValue();
 }
 
-/// 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 SDValue.
-///
-/// (This exists alongside SimplifyDemandedBits because GetDemandedBits can
-/// simplify nodes with multiple uses more aggressively.)
-SDValue DAGCombiner::GetDemandedBits(SDValue V, const APInt &Mask) {
-  switch (V.getOpcode()) {
-  default: break;
-  case ISD::Constant: {
-    const ConstantSDNode *CV = cast<ConstantSDNode>(V.getNode());
-    assert(CV && "Const value should be ConstSDNode.");
-    const APInt &CVal = CV->getAPIntValue();
-    APInt NewVal = CVal & Mask;
-    if (NewVal != CVal)
-      return DAG.getConstant(NewVal, SDLoc(V), V.getValueType());
-    break;
-  }
-  case ISD::OR:
-  case ISD::XOR:
-    // If the LHS or RHS don't contribute bits to the or, drop them.
-    if (DAG.MaskedValueIsZero(V.getOperand(0), Mask))
-      return V.getOperand(1);
-    if (DAG.MaskedValueIsZero(V.getOperand(1), Mask))
-      return V.getOperand(0);
-    break;
-  case ISD::SRL:
-    // Only look at single-use SRLs.
-    if (!V.getNode()->hasOneUse())
-      break;
-    if (ConstantSDNode *RHSC = getAsNonOpaqueConstant(V.getOperand(1))) {
-      // See if we can recursively simplify the LHS.
-      unsigned Amt = RHSC->getZExtValue();
-
-      // Watch out for shift count overflow though.
-      if (Amt >= Mask.getBitWidth()) break;
-      APInt NewMask = Mask << Amt;
-      if (SDValue SimplifyLHS = GetDemandedBits(V.getOperand(0), NewMask))
-        return DAG.getNode(ISD::SRL, SDLoc(V), V.getValueType(),
-                           SimplifyLHS, V.getOperand(1));
-    }
-    break;
-  case ISD::AND: {
-    // X & -1 -> X (ignoring bits which aren't demanded).
-    ConstantSDNode *AndVal = isConstOrConstSplat(V.getOperand(1));
-    if (AndVal && Mask.isSubsetOf(AndVal->getAPIntValue()))
-      return V.getOperand(0);
-    break;
-  }
-  }
-  return SDValue();
-}
-
 /// If the result of a wider load is shifted to right of N  bits and then
 /// truncated to a narrower type and where N is a multiple of number of bits of
 /// the narrower type, transform it to a narrower load from address + N / num of
 /// bits of new type. If the result is to be extended, also fold the extension
 /// to form a extending load.
 SDValue DAGCombiner::ReduceLoadWidth(SDNode *N) {
   unsigned Opc = N->getOpcode();
 
SDValue DAGCombiner::visitTRUNCATE(SDNode *N) {
   }
 
   // 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
   // Currently we only perform this optimization on scalars because vectors
   // may have different active low bits.
   if (!VT.isVector()) {
-    if (SDValue Shorter =
-            GetDemandedBits(N0, APInt::getLowBitsSet(N0.getValueSizeInBits(),
-                                                     VT.getSizeInBits())))
+    APInt Mask =
+        APInt::getLowBitsSet(N0.getValueSizeInBits(), VT.getSizeInBits());
+    if (SDValue Shorter = DAG.GetDemandedBits(N0, Mask))
       return DAG.getNode(ISD::TRUNCATE, SDLoc(N), VT, Shorter);
   }
 
   // fold (truncate (load x)) -> (smaller load x)
   // fold (truncate (srl (load x), c)) -> (smaller load (x+c/evtbits))
   if (!LegalTypes || TLI.isTypeDesirableForOp(N0.getOpcode(), VT)) {
     if (SDValue Reduced = ReduceLoadWidth(N))
       return Reduced;
SDValue DAGCombiner::visitSTORE(SDNode *N) {
   }
 
   // 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"
-    SDValue Shorter = GetDemandedBits(
+    SDValue Shorter = DAG.GetDemandedBits(
         Value, APInt::getLowBitsSet(Value.getScalarValueSizeInBits(),
                                     ST->getMemoryVT().getScalarSizeInBits()));
     AddToWorklist(Value.getNode());
     if (Shorter.getNode())
       return DAG.getTruncStore(Chain, SDLoc(N), Shorter,
                                Ptr, ST->getMemoryVT(), ST->getMemOperand());
 
     // Otherwise, see if we can simplify the operation with

llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAG.cpp

if (ConstantFPSDNode *N2C = dyn_cast<ConstantFPSDNode>(N2)) {
       return getSetCC(dl, VT, N2, N1, SwappedCond);
     }
   }
 
   // Could not fold it.
   return SDValue();
 }
 
+/// See if the specified operand can be simplified with the knowledge that only
+/// the bits specified by Mask are used.
+SDValue SelectionDAG::GetDemandedBits(SDValue V, const APInt &Mask) {
+  switch (V.getOpcode()) {
+  default:
+    break;
+  case ISD::Constant: {
+    const ConstantSDNode *CV = cast<ConstantSDNode>(V.getNode());
+    assert(CV && "Const value should be ConstSDNode.");
+    const APInt &CVal = CV->getAPIntValue();
+    APInt NewVal = CVal & Mask;
+    if (NewVal != CVal)
+      return getConstant(NewVal, SDLoc(V), V.getValueType());
+    break;
+  }
+  case ISD::OR:
+  case ISD::XOR:
+    // If the LHS or RHS don't contribute bits to the or, drop them.
+    if (MaskedValueIsZero(V.getOperand(0), Mask))
+      return V.getOperand(1);
+    if (MaskedValueIsZero(V.getOperand(1), Mask))
+      return V.getOperand(0);
+    break;
+  case ISD::SRL:
+    // Only look at single-use SRLs.
+    if (!V.getNode()->hasOneUse())
+      break;
+    if (ConstantSDNode *RHSC = dyn_cast<ConstantSDNode>(V.getOperand(1))) {
+      // See if we can recursively simplify the LHS.
+      unsigned Amt = RHSC->getZExtValue();
+
+      // Watch out for shift count overflow though.
+      if (Amt >= Mask.getBitWidth())
+        break;
+      APInt NewMask = Mask << Amt;
+      if (SDValue SimplifyLHS = GetDemandedBits(V.getOperand(0), NewMask))
+        return getNode(ISD::SRL, SDLoc(V), V.getValueType(), SimplifyLHS,
+                       V.getOperand(1));
+    }
+    break;
+  case ISD::AND: {
+    // X & -1 -> X (ignoring bits which aren't demanded).
+    ConstantSDNode *AndVal = isConstOrConstSplat(V.getOperand(1));
+    if (AndVal && Mask.isSubsetOf(AndVal->getAPIntValue()))
+      return V.getOperand(0);
+    break;
+  }
+  }
+  return SDValue();
+}
+
 /// SignBitIsZero - Return true if the sign bit of Op is known to be zero.  We
 /// use this predicate to simplify operations downstream.
 bool SelectionDAG::SignBitIsZero(SDValue Op, unsigned Depth) const {
   unsigned BitWidth = Op.getScalarValueSizeInBits();
   return MaskedValueIsZero(Op, APInt::getSignMask(BitWidth), Depth);
 }
 
 /// MaskedValueIsZero - Return true if 'V & Mask' is known to be zero.  We use
SDNode *SelectionDAG::MorphNodeTo(SDNode *N, unsigned Opc,
   return N;
 }
 
 SDNode* SelectionDAG::mutateStrictFPToFP(SDNode *Node) {
   unsigned OrigOpc = Node->getOpcode();
   unsigned NewOpc;
   bool IsUnary = false;
   switch (OrigOpc) {
-  default: 
+  default:
     llvm_unreachable("mutateStrictFPToFP called with unexpected opcode!");
   case ISD::STRICT_FADD: NewOpc = ISD::FADD; break;
   case ISD::STRICT_FSUB: NewOpc = ISD::FSUB; break;
   case ISD::STRICT_FMUL: NewOpc = ISD::FMUL; break;
   case ISD::STRICT_FDIV: NewOpc = ISD::FDIV; break;
   case ISD::STRICT_FREM: NewOpc = ISD::FREM; break;
   case ISD::STRICT_FSQRT: NewOpc = ISD::FSQRT; IsUnary = true; break;
   case ISD::STRICT_FPOW: NewOpc = ISD::FPOW; break;
SDNode* SelectionDAG::mutateStrictFPToFP(SDNode *Node) {
 
   SDVTList VTs = getVTList(Node->getOperand(1).getValueType());
   SDNode *Res = nullptr;
   if (IsUnary)
     Res = MorphNodeTo(Node, NewOpc, VTs, { Node->getOperand(1) });
   else
     Res = MorphNodeTo(Node, NewOpc, VTs, { Node->getOperand(1),
                                            Node->getOperand(2) });
-  
   // MorphNodeTo can operate in two ways: if an existing node with the
   // specified operands exists, it can just return it.  Otherwise, it
   // updates the node in place to have the requested operands.
   if (Res == Node) {
     // If we updated the node in place, reset the node ID.  To the isel,
     // this should be just like a newly allocated machine node.
     Res->setNodeId(-1);
   } else {
     ReplaceAllUsesWith(Node, Res);
     RemoveDeadNode(Node);
   }
 
-  return Res; 
+  return Res;
 }
 
 /// getMachineNode - These are used for target selectors to create a new node
 /// with specified return type(s), MachineInstr opcode, and operands.
 ///
 /// Note that getMachineNode returns the resultant node.  If there is already a
 /// node of the specified opcode and operands, it returns that node instead of
 /// the current one.