AArch64: More correctly constrain target vector extend lowering.

The AArch64 target lowering for [zs]ext of vectors is set up to handle
input simple types and expects the generic SDag path to do something reasonable
with anything that's not a simple type. The code, however, was only
checking that the result type was a simple type and assuming that
implied that the source type would also be a simple type. That's not a
valid assumption, as operations like "zext <1 x i1> %0 to <1 x i32>"
demonstrate. The fix is to simply explicitly validate the source type
as well as the result type.

PR20791

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@216689 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/AArch64/AArch64ISelLowering.cpp b/lib/Target/AArch64/AArch64ISelLowering.cpp
index bcc8d05c8c73c238a3c77a43f1ea6b8e8959ade8..60cd78442c0f1e65b36a3f52ba9044930c4027af 100644 (file)
--- a/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -7514,11 +7514,11 @@ static SDValue performExtendCombine(SDNode *N,
   // If the vector type isn't a simple VT, it's beyond the scope of what
   // we're  worried about here. Let legalization do its thing and hope for
   // the best.
-  if (!ResVT.isSimple())
+  SDValue Src = N->getOperand(0);
+  EVT SrcVT = Src->getValueType(0);
+  if (!ResVT.isSimple() || !SrcVT.isSimple())
     return SDValue();
 
-  SDValue Src = N->getOperand(0);
-  MVT SrcVT = Src->getValueType(0).getSimpleVT();
   // If the source VT is a 64-bit vector, we can play games and get the
   // better results we want.
   if (SrcVT.getSizeInBits() != 64)

diff --git a/test/CodeGen/AArch64/arm64-vector-ext.ll b/test/CodeGen/AArch64/arm64-vector-ext.ll
index 650ff1e14f0299a88d179ac3e094b0be0ba36322..5bee1611e6c64d1cb8f9f9cf2d7e38756a28a15b 100644 (file)
--- a/test/CodeGen/AArch64/arm64-vector-ext.ll
+++ b/test/CodeGen/AArch64/arm64-vector-ext.ll
@@ -14,3 +14,14 @@ define void @func30(%T0_30 %v0, %T1_30* %p1) {
   store %T1_30 %r, %T1_30* %p1
   ret void
 }
+
+; Extend from v1i1 was crashing things (PR20791). Make sure we do something
+; sensible instead.
+define <1 x i32> @autogen_SD7918() {
+; CHECK-LABEL: autogen_SD7918
+; CHECK: movi d0, #0000000000000000
+; CHECK-NEXT: ret
+  %I29 = insertelement <1 x i1> zeroinitializer, i1 false, i32 0
+  %ZE = zext <1 x i1> %I29 to <1 x i32>
+  ret <1 x i32> %ZE
+}