[TargetLowering] Add SimplifyDemandedBitsForTargetNode callback + add PMULDQ/PMULUDQ handling

[RKSimon]

Summary

This patch is in 2 parts (which would be committed separately):

1 - Add the SimplifyDemandedBitsForTargetNode callback to handle target nodes.

2 - Add X86 SimplifyDemandedBitsForTargetNode and use it to simplify PMULDQ/PMULUDQ target nodes. This enables us to repeatedly simplify the node's arguments after the previous approach had to be reverted due to PR39398.

[craig.topper]

LGTM

llvm/trunk/include/llvm/CodeGen/TargetLowering.h

virtual unsigned ComputeNumSignBitsForTargetNode(SDValue Op,
                                                    const APInt &DemandedElts,
                                                    const SelectionDAG &DAG,
                                                    unsigned Depth = 0) const;
 
   /// Attempt to simplify any target nodes based on the demanded vector
   /// elements, returning true on success. Otherwise, analyze the expression and
   /// return a mask of KnownUndef and KnownZero elements for the expression
   /// (used to simplify the caller). The KnownUndef/Zero elements may only be
-  /// accurate for those bits in the DemandedMask
+  /// accurate for those bits in the DemandedMask.
   virtual bool SimplifyDemandedVectorEltsForTargetNode(
       SDValue Op, const APInt &DemandedElts, APInt &KnownUndef,
       APInt &KnownZero, TargetLoweringOpt &TLO, unsigned Depth = 0) const;
 
+  /// Attempt to simplify any target nodes based on the demanded bits,
+  /// returning true on success. Otherwise, analyze the
+  /// expression and return a mask of KnownOne and KnownZero bits for the
+  /// expression (used to simplify the caller).  The KnownZero/One bits may only
+  /// be accurate for those bits in the DemandedMask.
+  virtual bool SimplifyDemandedBitsForTargetNode(SDValue Op,
+                                                 const APInt &DemandedBits,
+                                                 KnownBits &Known,
+                                                 TargetLoweringOpt &TLO,
+                                                 unsigned Depth = 0) const;
+
   /// If \p SNaN is false, \returns true if \p Op is known to never be any
   /// NaN. If \p sNaN is true, returns if \p Op is known to never be a signaling
   /// NaN.
   virtual bool isKnownNeverNaNForTargetNode(SDValue Op,
                                             const SelectionDAG &DAG,
                                             bool SNaN = false,
                                             unsigned Depth = 0) const;
   struct DAGCombinerInfo {

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

if (C && !C->isAllOnesValue() && !C->isOne() &&
       Flags.setNoUnsignedWrap(false);
       SDValue NewOp = TLO.DAG.getNode(Op.getOpcode(), dl, VT, Op0, Neg1, Flags);
       return TLO.CombineTo(Op, NewOp);
     }
 
     LLVM_FALLTHROUGH;
   }
   default:
+    if (Op.getOpcode() >= ISD::BUILTIN_OP_END) {
+      if (SimplifyDemandedBitsForTargetNode(Op, DemandedBits, Known, TLO,
+                                            Depth))
+        return true;
+      break;
+    }
+
     // Just use computeKnownBits to compute output bits.
     TLO.DAG.computeKnownBits(Op, Known, Depth);
     break;
   }
 
   // If we know the value of all of the demanded bits, return this as a
   // constant.
   if (DemandedBits.isSubsetOf(Known.Zero | Known.One)) {
assert((Op.getOpcode() >= ISD::BUILTIN_OP_END ||
           Op.getOpcode() == ISD::INTRINSIC_WO_CHAIN ||
           Op.getOpcode() == ISD::INTRINSIC_W_CHAIN ||
           Op.getOpcode() == ISD::INTRINSIC_VOID) &&
          "Should use SimplifyDemandedVectorElts if you don't know whether Op"
          " is a target node!");
   return false;
 }
 
+bool TargetLowering::SimplifyDemandedBitsForTargetNode(
+    SDValue Op, const APInt &DemandedBits, KnownBits &Known,
+    TargetLoweringOpt &TLO, unsigned Depth) const {
+  assert((Op.getOpcode() >= ISD::BUILTIN_OP_END ||
+          Op.getOpcode() == ISD::INTRINSIC_WO_CHAIN ||
+          Op.getOpcode() == ISD::INTRINSIC_W_CHAIN ||
+          Op.getOpcode() == ISD::INTRINSIC_VOID) &&
+         "Should use SimplifyDemandedBits if you don't know whether Op"
+         " is a target node!");
+  EVT VT = Op.getValueType();
+  APInt DemandedElts = VT.isVector()
+                           ? APInt::getAllOnesValue(VT.getVectorNumElements())
+                           : APInt(1, 1);
+  computeKnownBitsForTargetNode(Op, Known, DemandedElts, TLO.DAG, Depth);
+  return false;
+}
+
 bool TargetLowering::isKnownNeverNaNForTargetNode(SDValue Op,
                                                   const SelectionDAG &DAG,
                                                   bool SNaN,
                                                   unsigned Depth) const {
   assert((Op.getOpcode() >= ISD::BUILTIN_OP_END ||
           Op.getOpcode() == ISD::INTRINSIC_WO_CHAIN ||
           Op.getOpcode() == ISD::INTRINSIC_W_CHAIN ||
           Op.getOpcode() == ISD::INTRINSIC_VOID) &&