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/mulhs.ll b/test/CodeGen/PowerPC/mulhs.ll
new file mode 100644
index 0000000..967905d
--- /dev/null
+++ b/test/CodeGen/PowerPC/mulhs.ll
@@ -0,0 +1,18 @@
+; All of these ands and shifts should be folded into rlwimi's
+; RUN: llvm-upgrade < %s | llvm-as | llc -march=ppc32 -o %t -f
+; RUN: not grep mulhwu %t
+; RUN: not grep srawi %t
+; RUN: not grep add %t
+; RUN: grep mulhw %t | wc -l | grep 1
+
+implementation ; Functions:
+
+int %mulhs(int %a, int %b) {
+entry:
+ %tmp.1 = cast int %a to ulong ; <ulong> [#uses=1]
+ %tmp.3 = cast int %b to ulong ; <ulong> [#uses=1]
+ %tmp.4 = mul ulong %tmp.3, %tmp.1 ; <ulong> [#uses=1]
+ %tmp.6 = shr ulong %tmp.4, ubyte 32 ; <ulong> [#uses=1]
+ %tmp.7 = cast ulong %tmp.6 to int ; <int> [#uses=1]
+ ret int %tmp.7
+}