8369898: C2 SuperWord: assert(has_ctrl(i)) failed: should be control, not loop

Reviewed-by: chagedorn, kvn

src/hotspot/share/opto/loopUnswitch.cpp

@@ -522,7 +522,9 @@ IfTrueNode* PhaseIdealLoop::create_new_if_for_multiversion(IfTrueNode* multivers
 
   // Hook region into slow_path, in stead of the multiversion_slow_proj.
   // This also moves all other dependencies of the multiversion_slow_proj to the region.
-  _igvn.replace_node(multiversion_slow_proj, region);
+  // The lazy_replace ensures that any get_ctrl that used to have multiversion_slow_proj
+  // as their control are forwarded to the new region node as their control.
+  lazy_replace(multiversion_slow_proj, region);
 
   return new_if_true;
 }

test/hotspot/jtreg/compiler/loopopts/superword/TestMultiversionSlowProjReplacementAndGetCtrl.java

@@ -0,0 +1,95 @@
+/*
+ * Copyright (c) 2025, Oracle and/or its affiliates. All rights reserved.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This code is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.
+ *
+ * This code is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+ * or visit www.oracle.com if you need additional information or have any
+ * questions.
+ */
+
+/*
+ * @test
+ * @bug 8369898
+ * @summary Bug in PhaseIdealLoop::create_new_if_for_multiversion, that messed up the
+ *          _loop_or_ctrl data structure while doing SuperWord for a first loop, and
+ *          then get_ctrl asserted for a second loop that was also SuperWord-ed in the
+ *          same loop-opts-phase.
+ * @run main/othervm
+ *      -XX:CompileCommand=compileonly,*TestMultiversionSlowProjReplacementAndGetCtrl::test
+ *      -XX:CompileCommand=exclude,*TestMultiversionSlowProjReplacementAndGetCtrl::dontinline
+ *      -XX:-TieredCompilation
+ *      -Xbatch
+ *      compiler.loopopts.superword.TestMultiversionSlowProjReplacementAndGetCtrl
+ * @run main compiler.loopopts.superword.TestMultiversionSlowProjReplacementAndGetCtrl
+ */
+
+package compiler.loopopts.superword;
+
+public class TestMultiversionSlowProjReplacementAndGetCtrl {
+    static final int N = 400;
+
+    static void dontinline() {}
+
+    static long test() {
+        int x = 0;
+        int arrayI[] = new int[N];
+        byte[] arrayB = new byte[N];
+        dontinline();
+        // CallStaticJava for dontinline
+        // -> memory Proj
+        // -> it is used in both the k-indexed and j-indexed loops by their loads/stores.
+        for (int k = 8; k < 92; ++k) {
+            // Loop here is multiversioned, and eventually we insert an aliasing runtime check.
+            // This means that a StoreN (with mem input Proj from above) has its ctrl changed
+            // from the old multiversion_if_proj to a new region. We have to be careful to update
+            // the _loop_or_ctrl side-table so that get_ctrl for StoreN is sane.
+            //
+            // Below is some nested loop material I could not reduce further. Maybe because
+            // of loop-opts phase timing. Because we have to SuperWord the k-indexed loop
+            // above in the same loop-opts-phase as the j-indexed loop below, so that they
+            // have a shared _loop_or_ctrl data structure.
+            int y = 6;
+            while (--y > 0) {}
+            for (long i = 1; i < 6; i++) {
+                // I suspect that it is the two array references below that are SuperWord-ed,
+                // and since we do not manage to statically prove they cannot overlap, we add
+                // a speculative runtime check, i.e. multiversioning in this case.
+                arrayI[0] += 1;
+                arrayI[k] = 0;
+                try {
+                    x = 2 / k % y;
+                } catch (ArithmeticException a_e) {
+                }
+            }
+        }
+        long sum = 0;
+        for (int j = 0; j < arrayB.length; j++) {
+            // Load below has mem input from Proj below dontinline
+            // We look up to the mem input (Proj), and down to uses
+            // that are Stores, checking in_bb on them, which calls
+            // get_ctrl on that StoreN from the other loop above.
+            sum += arrayB[j];
+        }
+        return sum;
+    }
+
+    public static void main(String[] strArr) {
+        for (int i = 0; i < 1_000; i++) {
+            test();
+        }
+    }
+}