8251925: C2: RenaissanceStressTest fails with assert(!had_error): bad dominance #954

chhagedorn · 2020-10-30T11:28:10Z

The dominance failures start to occur after the fix for JDK-8249749 which enabled the method SWPointer::scaled_iv_plus_offset to call itself recursively and walk the graph to match more instead of stopping immediately (no recursion):
092389e#diff-8f29dd005a0f949d108687dabb7379c73dfd85cd782da453509dc9b6cb8c9f81L3789-R3812

We check in SWPointer::offset_plus_k if a node is invariant and if so then we choose it as invariant. However, we now have cases in the Renaissance benchmarks where we select an invariant that is pinned to a CastIINode between the main and pre loop. An example is shown in the attached image. 5913 SubI is found as an invariant with the improved recursive search enabled by JDK-8249749. The control of 5913 SubI (with get_ctrl) is 5298 CastII. The problem is now that we use the invariant 5913 SubI in the pre loop limit check of 5281 CountedLoopEnd (done in SuperWord::align_initial_loop_index) because we assume that since the invariant is not part of the main loop, it can float into the pre loop. But this is prevented by 5298 CastII. This results in the dominance assertion failure when checking if the earliest control of 5270 Bool in the pre loop (5297 IfTrue because of 5913 SubI that is used by 5270 Bool) dominates the LCA of 5270 Bool (the pre loop header node).

My suggestion is to improve the invariant check in SWPointer::offset_plus_k to also check if the found invariant is not dominated by the pre loop end node. Repeated testing of the RenaissanceStressTest has not resulted in any dominance failures anymore.

Progress

Change must not contain extraneous whitespace
Commit message must refer to an issue
Change must be properly reviewed

Testing

	Linux aarch64	Linux arm	Linux ppc64le	Linux s390x	Linux x64	Linux x86	Windows x64	macOS x64
Build	✔️ (1/1 passed)	✔️ (1/1 passed)	✔️ (1/1 passed)	✔️ (1/1 passed)	✔️ (6/6 passed)	✔️ (2/2 passed)	✔️ (2/2 passed)	✔️ (2/2 passed)
Test (tier1)					✔️ (9/9 passed)	✔️ (9/9 passed)	✔️ (9/9 passed)	✔️ (9/9 passed)

Issue

JDK-8251925: C2: RenaissanceStressTest fails with assert(!had_error): bad dominance

Reviewers

Vladimir Kozlov (@vnkozlov - Reviewer)
Tobias Hartmann (@TobiHartmann - Reviewer)

Download

$ git fetch https://git.openjdk.java.net/jdk pull/954/head:pull/954
$ git checkout pull/954

bridgekeeper · 2020-10-30T11:29:36Z

👋 Welcome back chagedorn! 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 · 2020-10-30T11:31:24Z

@chhagedorn 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 · 2020-10-30T11:35:56Z

Webrevs

03: Full (bcab625)
02: Full - Incremental (6eff991c79010bf71cc6b230997732d19449f207)
01: Full - Incremental (e9c99dcc85e6ebd0e542da62037fcfc1a96fcdcd)
00: Full (2a05f6c6f5fbd246d931d4ce78cd20260f19e52c)

vnkozlov

I would also suggest to run locally next jtreg command and compare number of created vector nodes to make sure your changes did not affect common cases:

$ jtreg -testjdk:/my_jdk/ -va -javaoptions:'-server -Xbatch -XX:-TieredCompilation -XX:CICompilerCount=1 -XX:+TraceNewVectors' -J-Djavatest.maxOutputSize=1000000 compiler/c2/cr6340864/ compiler/codegen/ compiler/loopopts/superword/ compiler/vectorization >new_vects.log

$ grep "new Vector node:" new_vects.log|wc

vnkozlov · 2020-10-30T16:21:49Z

src/hotspot/share/opto/superword.cpp

-  CountedLoopEndNode* pre_end = get_pre_loop_end(main_head);
+  assert(lp()->is_main_loop(), "");
+  CountedLoopEndNode* pre_end = cached_pre_loop_end();
+  assert(get_pre_loop_end(lp()), "pre loop end must still be found");


You already have similar assert check inside cached_pre_loop_end().

vnkozlov · 2020-10-30T16:22:07Z

src/hotspot/share/opto/superword.cpp

@@ -911,7 +917,8 @@ bool SuperWord::ref_is_alignable(SWPointer& p) {
  if (!p.has_iv()) {
    return true;   // no induction variable
  }
-  CountedLoopEndNode* pre_end = get_pre_loop_end(lp()->as_CountedLoop());
+  CountedLoopEndNode* pre_end = cached_pre_loop_end();
+  assert(get_pre_loop_end(lp()), "pre loop end must still be found");


You already have similar assert check inside cached_pre_loop_end().

vnkozlov · 2020-10-30T16:23:54Z

src/hotspot/share/opto/superword.hpp

+  CountedLoopEndNode* cached_pre_loop_end() {
+#ifdef ASSERT
+    Node* pre_end = get_pre_loop_end(_lp);
+    assert(_lp != NULL && (pre_end == NULL || pre_end == _cached_pre_loop_end) , "real CLE either not found anymore (NULL) or unchanged");


Check _lp != NULL should be before it use in previous line.

vnkozlov · 2020-10-30T16:24:32Z

src/hotspot/share/opto/superword.hpp

+#ifdef ASSERT
+    Node* pre_end = get_pre_loop_end(_lp);
+    assert(_lp != NULL && (pre_end == NULL || pre_end == _cached_pre_loop_end) , "real CLE either not found anymore (NULL) or unchanged");
+    assert(_cached_pre_loop_end != NULL, "should be set when fetched");


This should be very first assert in this method.

vnkozlov · 2020-10-30T16:26:28Z

src/hotspot/share/opto/superword.hpp

+    _cached_pre_loop_end = cached_pre_loop_end;
+  }
+
+  int      iv_stride()             { return lp()->stride_con(); }


Can you move this method right after set_lp() as originally?

vnkozlov · 2020-10-30T16:37:43Z

src/hotspot/share/opto/superword.cpp

  NOT_PRODUCT(_tracer.invariant_1(n, n_c);)
  return !lpt()->is_member(phase()->get_loop(n_c));
 }

+bool SWPointer::invariant_not_dominated_by_pre_loop_end(Node* n) {


I think we should have only one invariant() method which does this additional check.
And have separate method is_main_loop_member() where currently we check !invariant().

That's a good idea, fixed.

vnkozlov · 2020-10-30T16:44:04Z

src/hotspot/share/opto/superword.cpp

+    // Check that pre loop end node does not dominate n_c. If it does, then we cannot use n as invariant in the pre loop.
+    // This happens, for example, when n_c is a CastII node that prevents data nodes to flow above the main loop and into
+    // the pre loop. Use the cached version as the real pre loop end might not be found anymore with get_pre_loop_end().
+    return !phase()->is_dominator(_slp->cached_pre_loop_end(), n_c);


I think it is not enough. We don't want invariant be inside pre-loop.
Invariant should be node outside original loop (before splitting and unrolling).
We should check that n_c dominates pre-loop head.
What do you think?

I think that makes sense. I updated that. Tests still pass.

chhagedorn · 2020-11-03T10:23:41Z

@vnkozlov Thanks for your review!

$ jtreg -testjdk:/my_jdk/ -va -javaoptions:'-server -Xbatch -XX:-TieredCompilation -XX:CICompilerCount=1 -XX:+TraceNewVectors' -J-Djavatest.maxOutputSize=1000000 compiler/c2/cr6340864/ compiler/codegen/ compiler/loopopts/superword/ compiler/vectorization >new_vects.log

$ grep "new Vector node:" new_vects.log|wc

I ran that with the new commit - there is no difference between the fix and base without fix:
$ grep "new Vector node:" new_vects_fix.log | wc
9439 227498 1941126
$ grep "new Vector node:" new_vects_base.log | wc
9437 227521 1941437

vnkozlov · 2020-11-03T18:20:57Z

@vnkozlov Thanks for your review!

$ jtreg -testjdk:/my_jdk/ -va -javaoptions:'-server -Xbatch -XX:-TieredCompilation -XX:CICompilerCount=1 -XX:+TraceNewVectors' -J-Djavatest.maxOutputSize=1000000 compiler/c2/cr6340864/ compiler/codegen/ compiler/loopopts/superword/ compiler/vectorization >new_vects.log
$ grep "new Vector node:" new_vects.log|wc

I ran that with the new commit - there is no difference between the fix and base without fix:
$ grep "new Vector node:" new_vects_fix.log | wc
9439 227498 1941126
$ grep "new Vector node:" new_vects_base.log | wc
9437 227521 1941437

You have 2 more new vectors with fix! 9439 vs 9437
It is actually more than I tested last time: 9421

Good - no regression.

vnkozlov · 2020-11-03T18:30:10Z

src/hotspot/share/opto/superword.cpp

@@ -3970,20 +3986,24 @@ bool SWPointer::offset_plus_k(Node* n, bool negate) {
      return true;
    }
  }
-  if (invariant(n)) {
+  if (!is_main_loop_member(n)) {


Please, add comment here explaining why !is_main_loop_member() is used in this case instead of invariant().
Other places looks good.

I updated it and also moved the following invariant() check into the if block as it can only hold if !is_main_loop_member() is true.

vnkozlov

Updated changes looks good

openjdk · 2020-11-06T16:16:35Z

@chhagedorn 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:

8251925: C2: RenaissanceStressTest fails with assert(!had_error): bad dominance

Reviewed-by: kvn, thartmann

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 42 new commits pushed to the master branch:

5fedb69: 8250888: nsk/jvmti/scenarios/general_functions/GF08/gf08t001/TestDriver.java fails
02adaa5: 8255885: Metaspace: freelist commit counter is not updated when purging
fa240f2: 8256594: Unexpected warning: SIGSEGV handler flags expected:SA_RESTART|SA_SIGINFO found:SA_RESTART|SA_SIGINFO
4c09525: 8256108: Create implementation for NSAccessibilityElement protocol peer
6813889: 8251317: Support for CLDR version 38
c816464: 4916923: In MetalRootPaneUI, MetalRootLayout does not correctly calculate minimumsize
fae68ff: 8256640: assert(!m->is_old() || ik()->is_being_redefined()) failed: old methods should not be in vtable
c140773: 8256692: Zero: remove obsolete block from ZeroInterpreter::native_entry
080c707: 8253459: Formatter treats index, width and precision > Integer.MAX_VALUE incorrectly
b9db002: 8256682: JDK-8202343 is incomplete
... and 32 more: https://git.openjdk.java.net/jdk/compare/f504f419d3b377f0ccfd458026a2b57a9704bdff...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.

➡️ To integrate this PR with the above commit message to the master branch, type /integrate in a new comment.

chhagedorn · 2020-11-10T08:15:46Z

@vnkozlov Thanks a lot for the careful review!

… dominance

TobiHartmann

Otherwise looks good to me.

TobiHartmann · 2020-11-19T11:33:48Z

src/hotspot/share/opto/superword.cpp

@@ -69,7 +69,9 @@ SuperWord::SuperWord(PhaseIdealLoop* phase) :
  _nlist(arena(), 8, 0, NULL),            // scratch list of nodes
  _stk(arena(), 8, 0, NULL),              // scratch stack of nodes
  _lpt(NULL),                             // loop tree node
-  _lp(NULL),                              // LoopNode
+  _lp(NULL),                              // CountedLoopNode
+  _pre_loop_head(NULL),                   // Pre loop CountedLoopNode


Is this field really needed? Can't we just retrieve the pre-loop head via _pre_loop_end->loopnode()?

You're right, it is not really needed. I removed it.

chhagedorn · 2020-11-19T17:03:22Z

@TobiHartmann Thanks for your review!

vnkozlov

Unfortunately with forced push I can't see incremental update.
Based on last update comments changes are reasonable.

chhagedorn · 2020-11-20T07:13:23Z

@vnkozlov Thanks for reviewing again! Yes, that's unfortunate, I wasn't aware of that. I will try to do a merge next time.

chhagedorn · 2020-11-20T07:19:43Z

This was the new commit: bcab625, the other commits are identical.

TobiHartmann

Thanks for making these changes. Looks good to me!

chhagedorn · 2020-11-23T08:35:41Z

@TobiHartmann Thanks for your review!

chhagedorn · 2020-11-23T08:35:50Z

/integrate

openjdk · 2020-11-23T08:37:33Z

@chhagedorn Since your change was applied there have been 79 commits pushed to the master branch:

1f32c11: 8256740: ZGC: Move closures out of zOopClosure files
659aec8: 8256719: C1 flags that should have expired are still present
e06a683: 8256497: Zero: enable G1 and Shenandoah GCs
037e49c: 8256670: Zero: enable compressed oops support back
d46f6f5: 8256523: Streamline Java SHA2 implementation
1aa90ac: 8256822: runtime/logging/RedefineClasses.java fails with "Error: VM option 'Verbose' is develop and is available only in debug version of VM."
edf72f0: 8256824: test/langtools/tools/javac/diags/examples/InnerClassCantHaveStatic.java has a bad copyright
9a19eb6: 8254105: allow static nested declarations
14de791: 8255934: JConsole 14 and greater fails to connect to older JVM
86f3602: 8256806: Shenandoah: optimize shenandoah/jni/TestPinnedGarbage.java test
... and 69 more: https://git.openjdk.java.net/jdk/compare/f504f419d3b377f0ccfd458026a2b57a9704bdff...master

Your commit was automatically rebased without conflicts.

Pushed as commit e4a32be.

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

… dominance Reviewed-by: kvn, thartmann

openjdk bot added the rfr Pull request is ready for review label Oct 30, 2020

openjdk bot added the hotspot-compiler hotspot-compiler-dev@openjdk.org label Oct 30, 2020

vnkozlov suggested changes Oct 30, 2020

View reviewed changes

vnkozlov suggested changes Nov 3, 2020

View reviewed changes

vnkozlov approved these changes Nov 6, 2020

View reviewed changes

openjdk bot added the ready Pull request is ready to be integrated label Nov 6, 2020

chhagedorn added 3 commits November 18, 2020 13:38

8251925: C2: RenaissanceStressTest fails with assert(!had_error): bad…

4d04615

… dominance

Check dominance with pre loop head instead of tail

614bb3e

Update comments and invariant selection in offset_plus_k

b381c48

TobiHartmann approved these changes Nov 19, 2020

View reviewed changes

Remove _pre_loop_head field

bcab625

chhagedorn force-pushed the JDK-8251925 branch from 6eff991 to bcab625 Compare November 19, 2020 17:02

vnkozlov approved these changes Nov 19, 2020

View reviewed changes

TobiHartmann approved these changes Nov 20, 2020

View reviewed changes

openjdk bot closed this Nov 23, 2020

openjdk bot added integrated Pull request has been integrated and removed ready Pull request is ready to be integrated rfr Pull request is ready for review labels Nov 23, 2020

openjdk-notifier bot referenced this pull request Nov 23, 2020

8251925: C2: RenaissanceStressTest fails with assert(!had_error): bad…

e4a32be

… dominance Reviewed-by: kvn, thartmann

chhagedorn deleted the JDK-8251925 branch November 23, 2020 16:36

8251925: C2: RenaissanceStressTest fails with assert(!had_error): bad dominance #954

8251925: C2: RenaissanceStressTest fails with assert(!had_error): bad dominance #954

chhagedorn commented Oct 30, 2020 •

edited by openjdk bot

bridgekeeper bot commented Oct 30, 2020

openjdk bot commented Oct 30, 2020

mlbridge bot commented Oct 30, 2020 •

edited

vnkozlov left a comment •

edited

vnkozlov Oct 30, 2020

vnkozlov Oct 30, 2020

vnkozlov Oct 30, 2020

vnkozlov Oct 30, 2020

vnkozlov Oct 30, 2020

vnkozlov Oct 30, 2020

chhagedorn Nov 3, 2020

vnkozlov Oct 30, 2020

chhagedorn Nov 3, 2020

chhagedorn commented Nov 3, 2020 •

edited

vnkozlov commented Nov 3, 2020

vnkozlov Nov 3, 2020

chhagedorn Nov 4, 2020

vnkozlov left a comment

openjdk bot commented Nov 6, 2020 •

edited

chhagedorn commented Nov 10, 2020

TobiHartmann left a comment

TobiHartmann Nov 19, 2020 •

edited

chhagedorn Nov 19, 2020

chhagedorn commented Nov 19, 2020

vnkozlov left a comment

chhagedorn commented Nov 20, 2020

chhagedorn commented Nov 20, 2020

TobiHartmann left a comment

chhagedorn commented Nov 23, 2020

chhagedorn commented Nov 23, 2020

openjdk bot commented Nov 23, 2020

8251925: C2: RenaissanceStressTest fails with assert(!had_error): bad dominance #954

8251925: C2: RenaissanceStressTest fails with assert(!had_error): bad dominance #954

Conversation

chhagedorn commented Oct 30, 2020 • edited by openjdk bot

Progress

Testing

Issue

Reviewers

Download

bridgekeeper bot commented Oct 30, 2020

openjdk bot commented Oct 30, 2020

mlbridge bot commented Oct 30, 2020 • edited

Webrevs

vnkozlov left a comment • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

chhagedorn commented Nov 3, 2020 • edited

vnkozlov commented Nov 3, 2020

Choose a reason for hiding this comment

Choose a reason for hiding this comment

vnkozlov left a comment

Choose a reason for hiding this comment

openjdk bot commented Nov 6, 2020 • edited

chhagedorn commented Nov 10, 2020

TobiHartmann left a comment

Choose a reason for hiding this comment

TobiHartmann Nov 19, 2020 • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

chhagedorn commented Nov 19, 2020

vnkozlov left a comment

Choose a reason for hiding this comment

chhagedorn commented Nov 20, 2020

chhagedorn commented Nov 20, 2020

TobiHartmann left a comment

Choose a reason for hiding this comment

chhagedorn commented Nov 23, 2020

chhagedorn commented Nov 23, 2020

openjdk bot commented Nov 23, 2020

chhagedorn commented Oct 30, 2020 •

edited by openjdk bot

mlbridge bot commented Oct 30, 2020 •

edited

vnkozlov left a comment •

edited

chhagedorn commented Nov 3, 2020 •

edited

openjdk bot commented Nov 6, 2020 •

edited

TobiHartmann Nov 19, 2020 •

edited