[X86] Remove dead code from LowerVectorBroadcast

[zvi]

Summary

Splat vectors are canonicalized to BUILD_VECTOR's so the code can be simplified. NFC-ish.

[zvi]

ping

[RKSimon]

LGTM

[andreadb]

lib/Target/X86/X86ISelLowering.cpp

6030

Would it be possible to change this function so that 'Op' is a BuildVectorSDNode rather than a generic SDValue?
The expicit cast at line 5859 would become redundant then.

It looks like this function is only used by method 'LowerBUILD_VECTOR' at around line 6646.

[zvi]

lib/Target/X86/X86ISelLowering.cpp

6030

Sure.

[zvi]

Applied @andreadb's suggestion, and we could do the same to lowerBuildVectorToBitOp in a follow-up commit.
I also eliminated the 'Splat' variable, so the diff may look a bit larger.

[andreadb]

LGTM. Thanks Zvi!

lib/Target/X86/X86ISelLowering.cpp

if (TLI.isTypeLegal(VecVT)) {
 
       return DAG.getBitcast(VT, ResNode);
     }
   }
 
   return SDValue();
 }
 
-/// Attempt to use the vbroadcast instruction to generate a splat value for the
-/// following cases:
-/// 1. A splat BUILD_VECTOR which uses a single scalar load, or a constant.
-/// 2. A splat shuffle which uses a scalar_to_vector node which comes from
-/// a scalar load, or a constant.
+/// Attempt to use the vbroadcast instruction to generate a splat value for a
+/// splat BUILD_VECTOR which uses a single scalar load, or a constant.
 /// The VBROADCAST node is returned when a pattern is found,
 /// or SDValue() otherwise.
-static SDValue LowerVectorBroadcast(SDValue Op, const X86Subtarget &Subtarget,
+static SDValue LowerVectorBroadcast(BuildVectorSDNode *BVOp, const X86Subtarget &Subtarget,
                                     SelectionDAG &DAG) {
   // VBROADCAST requires AVX.
   // TODO: Splats could be generated for non-AVX CPUs using SSE
   // instructions, but there's less potential gain for only 128-bit vectors.
   if (!Subtarget.hasAVX())
     return SDValue();
 
-  MVT VT = Op.getSimpleValueType();
-  SDLoc dl(Op);
+  MVT VT = BVOp->getSimpleValueType(0);
+  SDLoc dl(BVOp);
 
   assert((VT.is128BitVector() || VT.is256BitVector() || VT.is512BitVector()) &&
          "Unsupported vector type for broadcast.");
 
-  SDValue Ld;
-  bool ConstSplatVal;
-
-  switch (Op.getOpcode()) {
-    default:
-      // Unknown pattern found.
-      return SDValue();
-
-    case ISD::BUILD_VECTOR: {
-      auto *BVOp = cast<BuildVectorSDNode>(Op.getNode());
-      BitVector UndefElements;
-      SDValue Splat = BVOp->getSplatValue(&UndefElements);
+  SDValue Ld = BVOp->getSplatValue(&UndefElements);
 
-      // We need a splat of a single value to use broadcast, and it doesn't
-      // make any sense if the value is only in one element of the vector.
-      if (!Splat || (VT.getVectorNumElements() - UndefElements.count()) <= 1)
-        return SDValue();
+  if (!Ld || (VT.getVectorNumElements() - UndefElements.count()) <= 1)
 
-      Ld = Splat;
-      ConstSplatVal = (Ld.getOpcode() == ISD::Constant ||
-                       Ld.getOpcode() == ISD::ConstantFP);
+  bool ConstSplatVal =
+      (Ld.getOpcode() == ISD::Constant || Ld.getOpcode() == ISD::ConstantFP);
 
-      // Make sure that all of the users of a non-constant load are from the
-      // BUILD_VECTOR node.
-      if (!ConstSplatVal && !BVOp->isOnlyUserOf(Ld.getNode()))
-        return SDValue();
-      break;
-    }
-
-    case ISD::VECTOR_SHUFFLE: {
-      ShuffleVectorSDNode *SVOp = cast<ShuffleVectorSDNode>(Op);
-
-      // Shuffles must have a splat mask where the first element is
-      // broadcasted.
-      if ((!SVOp->isSplat()) || SVOp->getMaskElt(0) != 0)
-        return SDValue();
-
-      SDValue Sc = Op.getOperand(0);
-      if (Sc.getOpcode() != ISD::SCALAR_TO_VECTOR &&
-          Sc.getOpcode() != ISD::BUILD_VECTOR) {
-
-        if (!Subtarget.hasInt256())
-          return SDValue();
-
-        // Use the register form of the broadcast instruction available on AVX2.
-        if (VT.getSizeInBits() >= 256)
-          Sc = extract128BitVector(Sc, 0, DAG, dl);
-        return DAG.getNode(X86ISD::VBROADCAST, dl, VT, Sc);
-      }
-
-      Ld = Sc.getOperand(0);
-      ConstSplatVal = (Ld.getOpcode() == ISD::Constant ||
-                       Ld.getOpcode() == ISD::ConstantFP);
-
-      // The scalar_to_vector node and the suspected
-      // load node must have exactly one user.
-      // Constants may have multiple users.
-
-      // AVX-512 has register version of the broadcast
-      bool hasRegVer = Subtarget.hasAVX512() && VT.is512BitVector() &&
-                       Ld.getValueSizeInBits() >= 32;
-      if (!ConstSplatVal && ((!Sc.hasOneUse() || !Ld.hasOneUse()) &&
-          !hasRegVer))
-        return SDValue();
-      break;
-    }
-  }
 
   unsigned ScalarSize = Ld.getValueSizeInBits();
   bool IsGE256 = (VT.getSizeInBits() >= 256);
 
   // When optimizing for size, generate up to 5 extra bytes for a broadcast
   // instruction to save 8 or more bytes of constant pool data.
   // TODO: If multiple splats are generated to load the same constant,
   // it may be detrimental to overall size. There needs to be a way to detect
X86TargetLowering::LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG) const {
   if (SDValue VectorConstant = materializeVectorConstant(Op, DAG, Subtarget))
     return VectorConstant;
 
   BuildVectorSDNode *BV = cast<BuildVectorSDNode>(Op.getNode());
   if (SDValue AddSub = LowerToAddSub(BV, Subtarget, DAG))
     return AddSub;
   if (SDValue HorizontalOp = LowerToHorizontalOp(BV, Subtarget, DAG))
     return HorizontalOp;
-  if (SDValue Broadcast = LowerVectorBroadcast(Op, Subtarget, DAG))
+  if (SDValue Broadcast = LowerVectorBroadcast(BV, Subtarget, DAG))
     return Broadcast;
   if (SDValue BitOp = lowerBuildVectorToBitOp(Op, DAG))
     return BitOp;
 
   unsigned EVTBits = ExtVT.getSizeInBits();
 
   unsigned NumZero  = 0;
   unsigned NumNonZero = 0;