[X86][AVX] Match broadcast loads through a bitcast

AVX1 v8i32/v4i64 shuffles are bitcasted to v8f32/v4f64, this patch peeks through bitcasts to check for a load node to allow broadcasts to occur.

Follow up to D15310

llvm-svn: 257055

1 files changed, 7 insertions, 2 deletions

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index d32146b32cf..06def2cacd8 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -8163,6 +8163,11 @@ static SDValue lowerVectorShuffleAsBroadcast(SDLoc DL, MVT VT, SDValue V,
     break;
   }
 
+  // Peek through any bitcast (only useful for loads).
+  SDValue BC = V;
+  while (BC.getOpcode() == ISD::BITCAST)
+    BC = BC.getOperand(0);
+
   // Check if this is a broadcast of a scalar. We special case lowering
   // for scalars so that we can more effectively fold with loads.
   // First, look through bitcast: if the original value has a larger element
@@ -8182,10 +8187,10 @@ static SDValue lowerVectorShuffleAsBroadcast(SDLoc DL, MVT VT, SDValue V,
     // Only AVX2 has register broadcasts.
     if (!Subtarget->hasAVX2() && !isShuffleFoldableLoad(V))
       return SDValue();
-  } else if (MayFoldLoad(V) && !cast<LoadSDNode>(V)->isVolatile()) {
+  } else if (MayFoldLoad(BC) && !cast<LoadSDNode>(BC)->isVolatile()) {
     // If we are broadcasting a load that is only used by the shuffle
     // then we can reduce the vector load to the broadcasted scalar load.
-    LoadSDNode *Ld = cast<LoadSDNode>(V);
+    LoadSDNode *Ld = cast<LoadSDNode>(BC);
     SDValue BaseAddr = Ld->getOperand(1);
     EVT AddrVT = BaseAddr.getValueType();
     EVT SVT = VT.getScalarType();