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/X86/mmx-insert-element.ll b/test/CodeGen/X86/mmx-insert-element.ll
new file mode 100644
index 0000000..3f2e402
--- /dev/null
+++ b/test/CodeGen/X86/mmx-insert-element.ll
@@ -0,0 +1,23 @@
+; RUN: llvm-as < %s | llc -march=x86 -mattr=+mmx | grep movq | wc -l | grep 3
+
+; FIXME: This code outputs:
+;
+; subl $28, %esp
+; movl 32(%esp), %eax
+; movd %eax, %mm0
+; movq %mm0, (%esp)
+; movl (%esp), %eax
+; movl %eax, 20(%esp)
+; movq %mm0, 8(%esp)
+; movl 12(%esp), %eax
+; movl %eax, 16(%esp)
+; movq 16(%esp), %mm0
+; addl $28, %esp
+;
+; Which is ugly. We need to fix this.
+
+define <2 x i32> @qux(i32 %A) {
+entry:
+ %tmp3 = insertelement <2 x i32> < i32 0, i32 undef >, i32 %A, i32 1 ; <<2 x i32>> [#uses=1]
+ ret <2 x i32> %tmp3
+}