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

1 files changed, 27 insertions, 0 deletions

diff --git a/test/CodeGen/PowerPC/2006-09-28-shift_64.ll b/test/CodeGen/PowerPC/2006-09-28-shift_64.ll
new file mode 100644
index 0000000..58d1f26
--- /dev/null
+++ b/test/CodeGen/PowerPC/2006-09-28-shift_64.ll
@@ -0,0 +1,27 @@
+; RUN: llvm-upgrade < %s | llvm-as | llc -march=ppc64
+target endian = big
+target pointersize = 64
+target triple = "powerpc64-apple-darwin8"
+
+implementation   ; Functions:
+
+void %glArrayElement_CompExec() {
+entry:
+	%tmp3 = and ulong 0, 18446744073701163007		; <ulong> [#uses=1]
+	br label %cond_true24
+
+cond_false:		; preds = %cond_true24
+	ret void
+
+cond_true24:		; preds = %cond_true24, %entry
+	%indvar.ph = phi uint [ 0, %entry ], [ %indvar.next, %cond_true24 ]		; <uint> [#uses=1]
+	%indvar = add uint 0, %indvar.ph		; <uint> [#uses=2]
+	%code.0 = cast uint %indvar to ubyte		; <ubyte> [#uses=1]
+	%tmp5 = add ubyte %code.0, 16		; <ubyte> [#uses=1]
+	%tmp7 = shr ulong %tmp3, ubyte %tmp5		; <ulong> [#uses=1]
+	%tmp7 = cast ulong %tmp7 to int		; <int> [#uses=1]
+	%tmp8 = and int %tmp7, 1		; <int> [#uses=1]
+	%tmp8 = seteq int %tmp8, 0		; <bool> [#uses=1]
+	%indvar.next = add uint %indvar, 1		; <uint> [#uses=1]
+	br bool %tmp8, label %cond_false, label %cond_true24
+}