Finally I'm able to distill a testcase for a problem I'm seeing!

llvm-svn: 4063

test/Regression/Transforms/CorrelatedExprs/2002-10-07-DominatorProblem.ll

@@ -0,0 +1,24 @@
+
+
+implementation
+
+int %test(int %A, int %B, bool %c0) {
+Start:
+	%c1 = seteq int %A, %B
+	br bool %c1, label %Eq, label %Loop
+
+Eq:	; In this block we know that A == B
+	br label %Loop    ; This should be modified to branch to "Forwarded".
+
+Loop:        ;; Merge point, nothing is known here...
+	%Z = phi int [%A, %Start], [%B, %Eq], [%Z, %Bottom]
+	%c2 = setgt int %A, %B
+	br bool %c2, label %Forwarded, label %Bottom
+
+Forwarded:
+	call int %test(int 0, int %Z, bool true)
+	br label %Bottom
+
+Bottom:
+	br label %Loop
+}