It's not necessary to do rounding for alloca operations when the requested
alignment is equal to the stack alignment.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@40004 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/test/CodeGen/ARM/bx_fold.ll b/test/CodeGen/ARM/bx_fold.ll
new file mode 100644
index 0000000..437b318
--- /dev/null
+++ b/test/CodeGen/ARM/bx_fold.ll
@@ -0,0 +1,30 @@
+; RUN: llvm-as < %s | llc -march=arm
+; RUN: llvm-as < %s | llc -march=arm | not grep bx
+
+define void @test(i32 %Ptr, i8* %L) {
+entry:
+ br label %bb1
+
+bb: ; preds = %bb1
+ %gep.upgrd.1 = zext i32 %indvar to i64 ; <i64> [#uses=1]
+ %tmp7 = getelementptr i8* %L, i64 %gep.upgrd.1 ; <i8*> [#uses=1]
+ store i8 0, i8* %tmp7
+ %indvar.next = add i32 %indvar, 1 ; <i32> [#uses=1]
+ br label %bb1
+
+bb1: ; preds = %bb, %entry
+ %indvar = phi i32 [ 0, %entry ], [ %indvar.next, %bb ] ; <i32> [#uses=3]
+ %i.0 = bitcast i32 %indvar to i32 ; <i32> [#uses=2]
+ %tmp = tail call i32 (...)* @bar( ) ; <i32> [#uses=1]
+ %tmp2 = add i32 %i.0, %tmp ; <i32> [#uses=1]
+ %Ptr_addr.0 = sub i32 %Ptr, %tmp2 ; <i32> [#uses=0]
+ %tmp12 = icmp eq i32 %i.0, %Ptr ; <i1> [#uses=1]
+ %tmp12.not = xor i1 %tmp12, true ; <i1> [#uses=1]
+ %bothcond = and i1 %tmp12.not, false ; <i1> [#uses=1]
+ br i1 %bothcond, label %bb, label %bb18
+
+bb18: ; preds = %bb1
+ ret void
+}
+
+declare i32 @bar(...)