[X86] Canonicalize (concat_vectors X, zero) -> (insert_subvector zero, X, 0).

In a future patch, I plan to teach isel to use a small vector move with implicit zeroing of the upper elements when it sees the (insert_subvector zero, X, 0) pattern.

llvm-svn: 312448

1 files changed, 10 insertions, 0 deletions

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index e0c17a2c27f..0193f2bfd58 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -35728,6 +35728,16 @@ static SDValue combineInsertSubvector(SDNode *N, SelectionDAG &DAG,
         return DAG.getNode(X86ISD::SUBV_BROADCAST, dl, OpVT,
                            SubVec.getOperand(0));
       }
+
+      // If we're inserting all zeros into the upper half, change this to
+      // an insert into an all zeros vector. We will match this to a move
+      // with implicit upper bit zeroing during isel.
+      if (ISD::isBuildVectorAllZeros(SubVec.getNode())) {
+        return DAG.getNode(ISD::INSERT_SUBVECTOR, dl, OpVT,
+                           getZeroVector(OpVT, Subtarget, DAG, dl), SubVec2,
+                           Vec.getOperand(2));
+
+      }
     }
   }