Handle Neon v2f64 and v2i64 vector shuffles as register copies.

This fixes the remaining issue with pr7167.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@104257 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/ARM/ARMISelLowering.cpp b/lib/Target/ARM/ARMISelLowering.cpp
index 62852b176857d3cabca3aca9ee8648650235e68e..48f3bbfac90507d0be568ab41c6b7539eb8443bc 100644 (file)
--- a/lib/Target/ARM/ARMISelLowering.cpp
+++ b/lib/Target/ARM/ARMISelLowering.cpp
@@ -3022,6 +3022,24 @@ static SDValue LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) {
       return GeneratePerfectShuffle(PFEntry, V1, V2, DAG, dl);
   }
 
+  // v2f64 and v2i64 shuffles are just register copies.
+  if (VT == MVT::v2f64 || VT == MVT::v2i64) {
+    // Do the expansion as f64 since i64 is not legal.
+    V1 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v2f64, V1);
+    V2 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v2f64, V2);
+    SDValue Val = DAG.getUNDEF(MVT::v2f64);
+    for (unsigned i = 0; i < 2; ++i) {
+      if (ShuffleMask[i] < 0)
+        continue;
+      SDValue Elt = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::f64,
+                                ShuffleMask[i] < 2 ? V1 : V2,
+                                DAG.getConstant(ShuffleMask[i] & 1, MVT::i32));
+      Val = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, MVT::v2f64, Val,
+                        Elt, DAG.getConstant(i, MVT::i32));
+    }
+    return DAG.getNode(ISD::BIT_CONVERT, dl, VT, Val);
+  }
+
   return SDValue();
 }
 

diff --git a/test/CodeGen/ARM/2010-05-19-Shuffles.ll b/test/CodeGen/ARM/2010-05-19-Shuffles.ll
index 588937e2f8ea2520abe094e1c4d48c1b0a259cfa..587c0afcb7147b6a8006129b4b82a4dcef76ce61 100644 (file)
--- a/test/CodeGen/ARM/2010-05-19-Shuffles.ll
+++ b/test/CodeGen/ARM/2010-05-19-Shuffles.ll
@@ -12,3 +12,10 @@ define <8 x i8> @f2(<8 x i8> %x) nounwind {
        <8 x i32> <i32 1, i32 2, i32 0, i32 5, i32 3, i32 6, i32 7, i32 4>
   ret <8 x i8> %y
 }
+
+define void @f3(<4 x i64>* %xp) nounwind {
+  %x = load <4 x i64>* %xp
+  %y = shufflevector <4 x i64> %x, <4 x i64> undef, <4 x i32> <i32 0, i32 3, i32 2, i32 1>
+  store <4 x i64> %y, <4 x i64>* %xp
+  ret void
+}