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/PowerPC/2006-10-11-combiner-aa-regression.ll b/test/CodeGen/PowerPC/2006-10-11-combiner-aa-regression.ll
new file mode 100644
index 0000000..992e52a
--- /dev/null
+++ b/test/CodeGen/PowerPC/2006-10-11-combiner-aa-regression.ll
@@ -0,0 +1,26 @@
+; RUN: llvm-upgrade < %s | llvm-as | \
+; RUN: llc -march=ppc32 -combiner-alias-analysis | grep f5
+
+target endian = big
+target pointersize = 32
+target triple = "powerpc-apple-darwin8.2.0"
+ %struct.Point = type { double, double, double }
+
+implementation ; Functions:
+
+void %offset(%struct.Point* %pt, double %x, double %y, double %z) {
+entry:
+ %tmp = getelementptr %struct.Point* %pt, int 0, uint 0 ; <double*> [#uses=2]
+ %tmp = load double* %tmp ; <double> [#uses=1]
+ %tmp2 = add double %tmp, %x ; <double> [#uses=1]
+ store double %tmp2, double* %tmp
+ %tmp6 = getelementptr %struct.Point* %pt, int 0, uint 1 ; <double*> [#uses=2]
+ %tmp7 = load double* %tmp6 ; <double> [#uses=1]
+ %tmp9 = add double %tmp7, %y ; <double> [#uses=1]
+ store double %tmp9, double* %tmp6
+ %tmp13 = getelementptr %struct.Point* %pt, int 0, uint 2 ; <double*> [#uses=2]
+ %tmp14 = load double* %tmp13 ; <double> [#uses=1]
+ %tmp16 = add double %tmp14, %z ; <double> [#uses=1]
+ store double %tmp16, double* %tmp13
+ ret void
+}