summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorSilviu Baranga <silviu.baranga@arm.com>2013-04-25 09:32:33 +0000
committerSilviu Baranga <silviu.baranga@arm.com>2013-04-25 09:32:33 +0000
commit02066838b5cdf17277267e79ffbc9459a58cdd59 (patch)
treed070dd6d9e4fa25659f34273243412bd6a3b0036
parent06fd5bf4ea122960f0435449c2391ea4fe1ea2ca (diff)
Fix constant folding for one lane vector types. Constant folding one lane vector types not returns a vector instead of a scalar.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180254 91177308-0d34-0410-b5e6-96231b3b80d8
-rw-r--r--lib/CodeGen/SelectionDAG/SelectionDAG.cpp2
-rw-r--r--test/CodeGen/ARM/v1-constant-fold.ll18
2 files changed, 19 insertions, 1 deletions
diff --git a/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
index 3082ee7caae..5e30c251e75 100644
--- a/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ b/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -2785,7 +2785,7 @@ SDValue SelectionDAG::FoldConstantArithmetic(unsigned Opcode, EVT VT,
}
// Handle the scalar case first.
- if (Outputs.size() == 1)
+ if (Scalar1 && Scalar2)
return Outputs.back();
// Otherwise build a big vector out of the scalar elements we generated.
diff --git a/test/CodeGen/ARM/v1-constant-fold.ll b/test/CodeGen/ARM/v1-constant-fold.ll
new file mode 100644
index 00000000000..b86d5db29c4
--- /dev/null
+++ b/test/CodeGen/ARM/v1-constant-fold.ll
@@ -0,0 +1,18 @@
+; RUN: llc < %s -mtriple=armv7-none-linux-gnueabi -mattr=+v7,+vfp3,-neon | FileCheck %s
+
+; PR15611. Check that we don't crash when constant folding v1i32 types.
+
+; CHECK: foo:
+define void @foo(i32 %arg) {
+bb:
+ %tmp = insertelement <4 x i32> undef, i32 %arg, i32 0
+ %tmp1 = insertelement <4 x i32> %tmp, i32 0, i32 1
+ %tmp2 = insertelement <4 x i32> %tmp1, i32 0, i32 2
+ %tmp3 = insertelement <4 x i32> %tmp2, i32 0, i32 3
+ %tmp4 = add <4 x i32> %tmp3, <i32 -1, i32 -1, i32 -1, i32 -1>
+; CHECK: bl bar
+ tail call void @bar(<4 x i32> %tmp4)
+ ret void
+}
+
+declare void @bar(<4 x i32>)