8335708: C2: Compile::verify_graph_edges must start at root and safepoints, just like CCP traversal #23977

marc-chevalier · 2025-03-11T09:12:44Z

In CCP, we transform the nodes going up (toward inputs) starting from root and safepoints because infinite loops can be reachable from the root, but not co-reachable from the root, that is one can follow def-use from root to the loop, but not the use-def from root to loop. For more details, see:

jdk/src/hotspot/share/opto/phaseX.cpp

Lines 2063 to 2070 in 4cf6316

    
           // Initialize the traversal. 
        
           // This CCP pass may prove that no exit test for a loop ever succeeds (i.e. the loop is infinite). In that case, 
        
           // the logic below doesn't follow any path from Root to the loop body: there's at least one such path but it's proven 
        
           // never taken (its type is TOP). As a consequence the node on the exit path that's input to Root (let's call it n) is 
        
           // replaced by the top node and the inputs of that node n are not enqueued for further processing. If CCP only works 
        
           // through the graph from Root, this causes the loop body to never be processed here even when it's not dead (that 
        
           // is reachable from Root following its uses). To prevent that issue, transform() starts walking the graph from Root 
        
           // and all safepoints.

Since we are specifically marking nodes as useful if they are above a safepoint, the check that no dead nodes must be there anymore must also consider nodes above a safepoint as alive: the same criterion must apply. We should nevertheless not start from a safepoint killed by CCP.

About the test, I use this trick found in TestInfiniteLoopCCP because I indeed need a really infinite loop, but I want a terminating test. The crash is not deterministic, as it needs StressIGVN, so I did a bit of stats. Using a little helper script, on 100 runs, 69 runs fail as in the JBS ticket and 31 are successful (so 0 fail in another way). After the fix, I find 100 successes.

And thanks to @eme64 who extracted such a concise reproducer.

Progress

Change must be properly reviewed (1 review required, with at least 1 Reviewer)
Change must not contain extraneous whitespace
Commit message must refer to an issue

Issue

JDK-8335708: C2: Compile::verify_graph_edges must start at root and safepoints, just like CCP traversal (Bug - P4)

Reviewers

Emanuel Peter (@eme64 - Reviewer) Review applies to 6f8fce6e
Christian Hagedorn (@chhagedorn - Reviewer)

Reviewing

Using git

Checkout this PR locally:
$ git fetch https://git.openjdk.org/jdk.git pull/23977/head:pull/23977
$ git checkout pull/23977

Update a local copy of the PR:
$ git checkout pull/23977
$ git pull https://git.openjdk.org/jdk.git pull/23977/head

Using Skara CLI tools

Checkout this PR locally:
$ git pr checkout 23977

View PR using the GUI difftool:
$ git pr show -t 23977

Using diff file

Download this PR as a diff file:
https://git.openjdk.org/jdk/pull/23977.diff

Using Webrev

Link to Webrev Comment

bridgekeeper · 2025-03-11T09:13:20Z

👋 Welcome back marc-chevalier! A progress list of the required criteria for merging this PR into master will be added to the body of your pull request. There are additional pull request commands available for use with this pull request.

openjdk · 2025-03-11T09:13:46Z

@marc-chevalier This change now passes all automated pre-integration checks.

ℹ️ This project also has non-automated pre-integration requirements. Please see the file CONTRIBUTING.md for details.

After integration, the commit message for the final commit will be:

8335708: C2: Compile::verify_graph_edges must start at root and safepoints, just like CCP traversal

Reviewed-by: chagedorn, epeter

You can use pull request commands such as /summary, /contributor and /issue to adjust it as needed.

At the time when this comment was updated there had been 182 new commits pushed to the master branch:

2560a63: 8352131: [REDO] C2: Print compilation bailouts with PrintCompilation compile command
73177d9: 8347734: Turning off PerfData logging doesn't work
96305e0: 4466930: JTable.selectAll boundary handling
fb210e3: 8351952: [IR Framework]: allow ignoring methods that are not compilable
3ed010a: 8352020: [CompileFramework] enable compilation for VectorAPI
a5d06a1: 8352249: Remove incidental whitespace in traditional doc comments
74df384: 8352428: GenShen: Old-gen cycles are still looping
fcc2a24: 8350813: Rendering of bulky sound bank from MIDI sequence can cause OutOfMemoryError
ac3ad03: 8350589: Investigate cleaner implementation of AArch64 ML-DSA intrinsic introduced in JDK-8348561
8a1c85e: 8350898: Shenandoah: Eliminate final roots safepoint
... and 172 more: https://git.openjdk.org/jdk/compare/75f028b46b245bdcbde8391af69020befda66b7d...master

As there are no conflicts, your changes will automatically be rebased on top of these commits when integrating. If you prefer to avoid this automatic rebasing, please check the documentation for the /integrate command for further details.

As you do not have Committer status in this project an existing Committer must agree to sponsor your change. Possible candidates are the reviewers of this PR (@eme64, @chhagedorn) but any other Committer may sponsor as well.

➡️ To flag this PR as ready for integration with the above commit message, type /integrate in a new comment. (Afterwards, your sponsor types /sponsor in a new comment to perform the integration).

openjdk · 2025-03-11T09:14:29Z

@marc-chevalier The following label will be automatically applied to this pull request:

hotspot-compiler

When this pull request is ready to be reviewed, an "RFR" email will be sent to the corresponding mailing list. If you would like to change these labels, use the /label pull request command.

mlbridge · 2025-03-14T15:11:13Z

Webrevs

eme64

Nice work, and thanks again for the offline discussion :)

I left a few comments / suggestions below.
Ah, and you could also consider changing the PR name. Maybe something like graph verification must start at root and safepoints, just like CCP traversal. Maybe you have an even better idea ;)

test/hotspot/jtreg/compiler/loopopts/Test8335708.java

eme64 · 2025-03-17T06:39:52Z

test/hotspot/jtreg/compiler/loopopts/Test8335708.java

+ * @run main/othervm
+ *       -XX:-TieredCompilation -XX:+VerifyGraphEdges
+ *       -XX:+StressIGVN -Xcomp
+ *       -XX:CompileCommand=compileonly,compiler.loopopts.Test8335708::mainTest
+ *       compiler.loopopts.Test8335708


Can you please add a run without any flags? Sometimes that allows other bugs to trigger, because it can then be used without any flags, or other flag combinations.

Done. I hope it's the right way.

eme64 · 2025-03-17T06:40:38Z

test/hotspot/jtreg/compiler/loopopts/Test8335708.java

@@ -0,0 +1,69 @@
+/*


Can you rename the test file? I think the new common practice is to give it a descriptive name, rather than just the bug number which is already tracked under @bug 8335708 anyway ;)

Done. I think it's more explicit.

test/hotspot/jtreg/compiler/loopopts/Test8335708.java

eme64 · 2025-03-17T06:45:13Z

src/hotspot/share/opto/compile.cpp

-  Node_List nstack(MAX2(stack_size, (uint)OptoNodeListSize));
-  nstack.push(_root);
+  Node_List nstack(MAX2(stack_size, (uint) OptoNodeListSize));
+  if (root_and_safepoints != nullptr) {


Can you say in which cases we don't have root_and_safepoints? Why is it ok not to also start at SafePoint in those cases?

I think you should also say that we start the traversal from Root and Safepoints, just like during CCP.

I've improved the comment on the declaration of verify_graph_edges: it is the only caller of verify_bidirectional_edges, which acts more like a helper, and verify_graph_edges is the one called a bit everywhere. Also, I think it's not an implementation detail, but a signature/contract thing: when writing a call to verify_graph_edges, I must know what I need to provide in root_and_safepoints, or when I can omit it. So now, I hope it's documented.

marc-chevalier · 2025-03-18T10:30:24Z

you could also consider changing the PR name. Maybe something like graph verification must start at root and safepoints, just like CCP traversal. Maybe you have an even better idea ;)

I don't understand what you mean. the title of the PR must match the JBS ticket. I have very little creative freedom there. No?

eme64 · 2025-03-18T10:37:43Z

you could also consider changing the PR name. Maybe something like graph verification must start at root and safepoints, just like CCP traversal. Maybe you have an even better idea ;)

I don't understand what you mean. the title of the PR must match the JBS ticket. I have very little creative freedom there. No?

You can always change both, so that they match ;)

eme64

@marc-chevalier Looks good to me, except for missing punctuation ;)

src/hotspot/share/opto/compile.hpp

chhagedorn

Some minor comments. Otherwise, looks good to me, too!

src/hotspot/share/opto/compile.cpp

src/hotspot/share/opto/compile.hpp

marc-chevalier · 2025-03-20T14:34:00Z

I applied your fixes. I don't think I need proper review again, but I need re-approval.

EDIT: nevermind, @chhagedorn was faster than me.

marc-chevalier · 2025-03-20T14:36:06Z

/integrate

Thanks for the reviews!

openjdk · 2025-03-20T14:36:41Z

@marc-chevalier
Your change (at version 6a149e2) is now ready to be sponsored by a Committer.

TobiHartmann · 2025-03-20T15:49:05Z

/sponsor

openjdk · 2025-03-20T15:50:23Z

Going to push as commit 2bc4f64.
Since your change was applied there have been 184 commits pushed to the master branch:

9a17a6f: 8350749: Upgrade JLine to 3.29.0
91836e1: 8352276: Skip jtreg tests using native executable with libjvm.so/libjli.so dependencies on static JDK
2560a63: 8352131: [REDO] C2: Print compilation bailouts with PrintCompilation compile command
73177d9: 8347734: Turning off PerfData logging doesn't work
96305e0: 4466930: JTable.selectAll boundary handling
fb210e3: 8351952: [IR Framework]: allow ignoring methods that are not compilable
3ed010a: 8352020: [CompileFramework] enable compilation for VectorAPI
a5d06a1: 8352249: Remove incidental whitespace in traditional doc comments
74df384: 8352428: GenShen: Old-gen cycles are still looping
fcc2a24: 8350813: Rendering of bulky sound bank from MIDI sequence can cause OutOfMemoryError
... and 174 more: https://git.openjdk.org/jdk/compare/75f028b46b245bdcbde8391af69020befda66b7d...master

Your commit was automatically rebased without conflicts.

openjdk · 2025-03-20T15:50:31Z

@TobiHartmann @marc-chevalier Pushed as commit 2bc4f64.

💡 You may see a message that your pull request was closed with unmerged commits. This can be safely ignored.

Start detection of useful nodes also at safepoints

Loading
Loading status checks…

f44411b

openjdk bot added the hotspot-compiler label Mar 11, 2025

Don't start on removed safepoints.

Loading
Loading status checks…

a07dc23

marc-chevalier marked this pull request as ready for review March 14, 2025 15:06

openjdk bot added the rfr label Mar 14, 2025

eme64 suggested changes Mar 17, 2025

View reviewed changes

marc-chevalier changed the title ~~8335708: C2: assert(!dead_nodes) failed: using nodes must be reachable from root~~ 8335708: C2: Compile::verify_graph_edges must start at root and safepoints, just like CCP traversal Mar 18, 2025

various fixes

Loading
Loading status checks…

6f8fce6

eme64 approved these changes Mar 20, 2025

View reviewed changes

src/hotspot/share/opto/compile.hpp Outdated Show resolved Hide resolved

openjdk bot added the ready label Mar 20, 2025

chhagedorn approved these changes Mar 20, 2025

View reviewed changes

src/hotspot/share/opto/compile.cpp Outdated Show resolved Hide resolved

src/hotspot/share/opto/compile.cpp Outdated Show resolved Hide resolved

src/hotspot/share/opto/compile.cpp Outdated Show resolved Hide resolved

src/hotspot/share/opto/compile.hpp Outdated Show resolved Hide resolved

mostly punctuation

Loading
Loading status checks…

6a149e2

openjdk bot removed the ready label Mar 20, 2025

chhagedorn approved these changes Mar 20, 2025

View reviewed changes

openjdk bot added the ready label Mar 20, 2025

openjdk bot added the sponsor label Mar 20, 2025

openjdk bot added the integrated label Mar 20, 2025

openjdk bot closed this Mar 20, 2025

openjdk bot removed the ready label Mar 20, 2025

openjdk bot removed rfr sponsor labels Mar 20, 2025

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

8335708: C2: Compile::verify_graph_edges must start at root and safepoints, just like CCP traversal #23977

8335708: C2: Compile::verify_graph_edges must start at root and safepoints, just like CCP traversal #23977

marc-chevalier commented Mar 11, 2025 •

edited by openjdk bot

Loading

bridgekeeper bot commented Mar 11, 2025

openjdk bot commented Mar 11, 2025 •

edited

Loading

openjdk bot commented Mar 11, 2025

mlbridge bot commented Mar 14, 2025 •

edited

Loading

eme64 left a comment

eme64 Mar 17, 2025

marc-chevalier Mar 18, 2025

eme64 Mar 17, 2025

marc-chevalier Mar 18, 2025

eme64 Mar 17, 2025

eme64 Mar 17, 2025

marc-chevalier Mar 18, 2025

marc-chevalier commented Mar 18, 2025

eme64 commented Mar 18, 2025

eme64 left a comment

chhagedorn left a comment

marc-chevalier commented Mar 20, 2025 •

edited

Loading

marc-chevalier commented Mar 20, 2025

openjdk bot commented Mar 20, 2025

TobiHartmann commented Mar 20, 2025

openjdk bot commented Mar 20, 2025

openjdk bot commented Mar 20, 2025

	// Initialize the traversal.
	// This CCP pass may prove that no exit test for a loop ever succeeds (i.e. the loop is infinite). In that case,
	// the logic below doesn't follow any path from Root to the loop body: there's at least one such path but it's proven
	// never taken (its type is TOP). As a consequence the node on the exit path that's input to Root (let's call it n) is
	// replaced by the top node and the inputs of that node n are not enqueued for further processing. If CCP only works
	// through the graph from Root, this causes the loop body to never be processed here even when it's not dead (that
	// is reachable from Root following its uses). To prevent that issue, transform() starts walking the graph from Root
	// and all safepoints.

8335708: C2: Compile::verify_graph_edges must start at root and safepoints, just like CCP traversal #23977

8335708: C2: Compile::verify_graph_edges must start at root and safepoints, just like CCP traversal #23977

Conversation

marc-chevalier commented Mar 11, 2025 • edited by openjdk bot Loading

Progress

Issue

Reviewers

Reviewing

bridgekeeper bot commented Mar 11, 2025

openjdk bot commented Mar 11, 2025 • edited Loading

openjdk bot commented Mar 11, 2025

mlbridge bot commented Mar 14, 2025 • edited Loading

Webrevs

eme64 left a comment

Choose a reason for hiding this comment

eme64 Mar 17, 2025

Choose a reason for hiding this comment

marc-chevalier Mar 18, 2025

Choose a reason for hiding this comment

eme64 Mar 17, 2025

Choose a reason for hiding this comment

marc-chevalier Mar 18, 2025

Choose a reason for hiding this comment

eme64 Mar 17, 2025

Choose a reason for hiding this comment

eme64 Mar 17, 2025

Choose a reason for hiding this comment

marc-chevalier Mar 18, 2025

Choose a reason for hiding this comment

marc-chevalier commented Mar 18, 2025

eme64 commented Mar 18, 2025

eme64 left a comment

Choose a reason for hiding this comment

chhagedorn left a comment

Choose a reason for hiding this comment

marc-chevalier commented Mar 20, 2025 • edited Loading

marc-chevalier commented Mar 20, 2025

openjdk bot commented Mar 20, 2025

TobiHartmann commented Mar 20, 2025

openjdk bot commented Mar 20, 2025

openjdk bot commented Mar 20, 2025

marc-chevalier commented Mar 11, 2025 •

edited by openjdk bot

Loading

openjdk bot commented Mar 11, 2025 •

edited

Loading

mlbridge bot commented Mar 14, 2025 •

edited

Loading

marc-chevalier commented Mar 20, 2025 •

edited

Loading