8361608: C2: assert(opaq->outcnt() == 1 && opaq->in(1) == limit) failed

Co-authored-by: Christian Hagedorn <chagedorn@openjdk.org>
Reviewed-by: chagedorn, rcastanedalo
This commit is contained in:
Marc Chevalier 2025-10-31 11:00:06 +00:00
parent 8e3620a344
commit 02f8874c2d
3 changed files with 225 additions and 2 deletions

View File

@ -1970,8 +1970,19 @@ void PhaseIdealLoop::do_unroll(IdealLoopTree *loop, Node_List &old_new, bool adj
if (opaq == nullptr) {
return;
}
// Zero-trip test uses an 'opaque' node which is not shared.
assert(opaq->outcnt() == 1 && opaq->in(1) == limit, "");
// Zero-trip test uses an 'opaque' node which is not shared, otherwise bail out.
if (opaq->outcnt() != 1 || opaq->in(1) != limit) {
#ifdef ASSERT
// In rare cases, loop cloning (as for peeling, for instance) can break this by replacing
// limit and the input of opaq by equivalent but distinct phis.
// Next IGVN should clean it up. Let's try to detect we are in such a case.
Unique_Node_List& worklist = loop->_phase->_igvn._worklist;
assert(C->major_progress(), "The operation that replaced limit and opaq->in(1) (e.g. peeling) should have set major_progress");
assert(opaq->in(1)->is_Phi() && limit->is_Phi(), "Nodes limit and opaq->in(1) should have been replaced by PhiNodes by fix_data_uses from clone_loop.");
assert(worklist.member(opaq->in(1)) && worklist.member(limit), "Nodes limit and opaq->in(1) differ and should have been recorded for IGVN.");
#endif
return;
}
}
C->set_major_progress();

View File

@ -4765,6 +4765,24 @@ void PhaseIdealLoop::eliminate_useless_zero_trip_guard() {
Node* opaque = head->is_canonical_loop_entry();
if (opaque != nullptr) {
useful_zero_trip_guard_opaques_nodes.push(opaque);
#ifdef ASSERT
// See PhaseIdealLoop::do_unroll
// This property is required in do_unroll, but it may not hold after cloning a loop.
// In such a case, we bail out from unrolling, and rely on IGVN to clean up the graph.
// We are here before loop cloning (before iteration_split), so if this property
// does not hold, it must come from the previous round of loop optimizations, meaning
// that IGVN failed to clean it: we will catch that here.
// On the other hand, if this assert passes, a bailout in do_unroll means that
// this property was broken in the current round of loop optimization (between here
// and do_unroll), so we give a chance to IGVN to make the property true again.
if (head->is_main_loop()) {
assert(opaque->outcnt() == 1, "opaque node should not be shared");
assert(opaque->in(1) == head->limit(), "After IGVN cleanup, input of opaque node must be the limit.");
}
if (head->is_post_loop()) {
assert(opaque->outcnt() == 1, "opaque node should not be shared");
}
#endif
}
}
}

View File

@ -0,0 +1,194 @@
/*
* 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 8361608
* @summary crash during unrolling in some rare cases where loop cloning
* (typically from peeling) breaks an invariant in do_unroll
*
* @run main/othervm -XX:+IgnoreUnrecognizedVMOptions
* -XX:+StressLoopPeeling
* -XX:CompileCommand=compileonly,compiler.loopopts.TooStrictAssertForUnrollAfterPeeling::test1
* -XX:-TieredCompilation
* -Xbatch
* -XX:PerMethodTrapLimit=0
* compiler.loopopts.TooStrictAssertForUnrollAfterPeeling
*
* @run main/othervm -XX:CompileOnly=compiler.loopopts.TooStrictAssertForUnrollAfterPeeling::test2
* -XX:-TieredCompilation
* -Xbatch
* -XX:PerMethodTrapLimit=0
* -XX:CompileCommand=inline,compiler.loopopts.TooStrictAssertForUnrollAfterPeeling::foo2
* compiler.loopopts.TooStrictAssertForUnrollAfterPeeling
*
* @run main/othervm -XX:CompileOnly=compiler.loopopts.TooStrictAssertForUnrollAfterPeeling::test3
* -XX:-TieredCompilation
* -Xbatch
* -XX:PerMethodTrapLimit=0
* -XX:CompileCommand=inline,compiler.loopopts.TooStrictAssertForUnrollAfterPeeling::foo3
* -XX:-RangeCheckElimination
* compiler.loopopts.TooStrictAssertForUnrollAfterPeeling
*
* @run main compiler.loopopts.TooStrictAssertForUnrollAfterPeeling
*/
package compiler.loopopts;
/* Cases 2 and 3 can use the additional flags
* -XX:+UnlockDiagnosticVMOptions
* -XX:-LoopMultiversioning
* -XX:-RangeCheckElimination
* -XX:-SplitIfBlocks
* -XX:-UseOnStackReplacement
* -XX:LoopMaxUnroll=2
* to disable more optimizations and give a simpler graph, while still reproducing. It can be useful to debug, investigate...
*/
public class TooStrictAssertForUnrollAfterPeeling {
static int iArr[] = new int[400];
static boolean flag;
public static void main(String[] args) {
run1();
run2();
run3();
}
// Case 1
public static void run1() {
for (int i = 1; i < 1000; i++) {
test1();
}
}
static long test1() {
int s = 0;
int iArr[] = new int[400];
for (int i = 0; i < 70; i++) {}
for (int i = 0; i < 36; i++) {
for (int j = 0; j < 3; j++) {
s += iArr[0] = 7;
if (s != 0) {
return s + foo1(iArr);
}
}
}
return 0;
}
public static long foo1(int[] a) {
long sum = 0;
for (int i = 0; i < a.length; i++) {
sum += a[i];
}
return sum;
}
// Case 2
public static void run2() {
for (int i = 1; i < 10000; i++) {
test2();
}
}
// Lx: Optimized in loop opts round x.
static int test2() {
int x = 5;
for (int i = 1; i < 37; i++) { // L3: Peeled
for (int a = 0; a < 2; a++) { // L2: Max unrolled
for (int b = 0; b < 300; b++) {
} // L1: Empty -> removed
}
int j = 1;
x *= 12;
while (++j < 5) { // L1: Max unrolled: peel + unroll
iArr[0] += 2;
if (iArr[0] > 0) {
// foo(): everything outside loop.
return foo2(iArr);
}
}
}
return 3;
}
public static int foo2(int[] a) {
int sum = 0;
for (int i = 0; i < a.length; i++) { // L2: Pre/main/post, L3: Unrolled -> hit assert!
for (int j = 0; j < 34; j++) {
} // L1: Empty -> removed
if (flag) {
// Ensure not directly unrolled in L2 but only in L3.
return 3;
}
sum += a[i];
}
return sum;
}
public static void run3() {
for (int i = 1; i < 10000; i++) {
test3();
}
}
// Case 3
static int test3() {
int x = 5;
for (int i = 1; i < 37; i++) { // L3: Peeled
for (int a = 0; a < 2; a++) { // L2: Max unrolled
for (int b = 0; b < 300; b++) {
} // L1: Empty -> removed
}
int j = 1;
x *= 12;
while (++j < 5) { // L1: Max unrolled: peel + unroll
iArr[0] += 2;
if (iArr[0] > 0) {
// foo(): everything outside loop.
return foo3(iArr, x);
}
}
}
return 3;
}
public static int foo3(int[] a, int limit) {
int sum = 0;
for (int i = 0; i < limit; i++) { // L2: Pre/main/post, L3: Unrolled -> hit assert!
for (int j = 0; j < 34; j++) {
} // L1: Empty -> removed
if (flag) {
// Ensure not directly unrolled in L2 but only in L3.
return 3;
}
sum += a[i];
}
return sum;
}
}