8263227: C2: inconsistent spilling due to dead nodes in exception block #3303

robcasloz · 2021-04-01T10:02:39Z

This change eliminates dead multi-nodes created by call-catch cleanup after GCM. Eliminating all dead code created by call-catch cleanup avoids potential issues when splitting the live range of call result values, see the analysis in the bug report for details. This solution is the least invasive of the three alternatives proposed in the bug report (the other two are constraining global code motion and extending live-range splitting).

The change also extends the control-flow graph verification pass to catch similar live-range splitting issues earlier (with +VerifyRegisterAllocator).

Tested on:

original bug reproducer
hs-tier1-5 (windows-x64, linux-x64, linux-aarch64, and macosx-x64) with +VerifyRegisterAllocator
hs-tier1-3 (windows-x64, linux-x64, linux-aarch64, and macosx-x64) with +VerifyRegisterAllocator and +StressGCM (5 repetitions)

Progress

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

Issue

JDK-8263227: C2: inconsistent spilling due to dead nodes in exception block

Reviewers

Nils Eliasson (@neliasso - Reviewer)
Vladimir Kozlov (@vnkozlov - Reviewer)

Reviewing

Using git

Checkout this PR locally:
$ git fetch https://git.openjdk.java.net/jdk pull/3303/head:pull/3303
$ git checkout pull/3303

Update a local copy of the PR:
$ git checkout pull/3303
$ git pull https://git.openjdk.java.net/jdk pull/3303/head

Using Skara CLI tools

Checkout this PR locally:
$ git pr checkout 3303

View PR using the GUI difftool:
$ git pr show -t 3303

Using diff file

Download this PR as a diff file:
https://git.openjdk.java.net/jdk/pull/3303.diff

bridgekeeper · 2021-04-01T10:03:34Z

👋 Welcome back rcastanedalo! 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 · 2021-04-01T10:05:43Z

@robcasloz 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 · 2021-04-06T07:57:41Z

Webrevs

01: Full - Incremental (808451de)
00: Full (a4973ba0)

neliasso

Looks good!

openjdk · 2021-04-06T09:43:31Z

@robcasloz 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:

8263227: C2: inconsistent spilling due to dead nodes in exception block

Eliminate dead nodes created by call-catch cleanup even if they have multiple
projections. Assert that definitions dominate uses and projections are scheduled
next to their parent nodes.

Reviewed-by: neliasso, kvn

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 510 new commits pushed to the master branch:

a25bae8: 8261445: Use memory_order_relaxed for os::random().
5136643: 8262725: IGV: crash when removing all graphs in a group
79798c6: 8265136: ZGC: Expose GarbageCollectorMXBeans for both pauses and cycles
f1d4ae6: 8263718: unused-result warning happens at os_linux.cpp
787908c: 8264221: Rewrite confusing stream API chain in SnippetMaps
142edd3: 8265152: jpackage cleanup fails on Windows with IOException deleting msi
ab22407: 8265486: ProblemList javax/sound/midi/Sequencer/Recording.java on macosx-aarch64
e0fd5fc: 8265028: JDWP debug agent thread lookup can be made faster
713483c: 8265373: Change to GCC 10.3 for building on Linux at Oracle
3990713: 8265463: ProblemList vmTestbase/vm/mlvm/mixed/stress/regression/b6969574/INDIFY_Test.java on Win-X64 -Xcomp
... and 500 more: https://git.openjdk.java.net/jdk/compare/d1baed677e30d33522033c2205138b33e8d392b6...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.

neliasso · 2021-04-06T09:49:24Z

And thanks for the PDF with the excellent walk through!

robcasloz · 2021-04-06T09:51:41Z

Looks good!

Thanks for reviewing, Nils!

src/hotspot/share/opto/lcm.cpp

vnkozlov

Also where is guarantee that all Proj users are in the same block.

robcasloz · 2021-04-08T11:04:00Z

Also where is guarantee that all Proj users are in the same block.

Isn't this guaranteed by PhaseCFG::schedule_node_into_block()?

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

Lines 48 to 72 in ec599da

    
           //----------------------------schedule_node_into_block------------------------- 
        
           // Insert node n into block b. Look for projections of n and make sure they 
        
           // are in b also. 
        
           void PhaseCFG::schedule_node_into_block( Node *n, Block *b ) { 
        
             // Set basic block of n, Add n to b, 
        
             map_node_to_block(n, b); 
        
             b->add_inst(n); 
        
             // After Matching, nearly any old Node may have projections trailing it. 
        
             // These are usually machine-dependent flags.  In any case, they might 
        
             // float to another block below this one.  Move them up. 
        
             for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) { 
        
               Node*  use  = n->fast_out(i); 
        
               if (use->is_Proj()) { 
        
                 Block* buse = get_block_for_node(use); 
        
                 if (buse != b) {              // In wrong block? 
        
                   if (buse != NULL) { 
        
                     buse->find_remove(use);   // Remove from wrong block 
        
                   } 
        
                   map_node_to_block(use, b); 
        
                   b->add_inst(use); 
        
                 } 
        
               } 
        
             } 
        
           }

I will run some tests with extra assertions in PhaseCFG::verify() to double-check.

vnkozlov · 2021-04-08T17:53:08Z

Also where is guarantee that all Proj users are in the same block.

Isn't this guaranteed by PhaseCFG::schedule_node_into_block()?

I will run some tests with extra assertions in PhaseCFG::verify() to double-check.

Okay.

…larity

robcasloz · 2021-04-12T08:37:28Z

Also where is guarantee that all Proj users are in the same block.

Isn't this guaranteed by PhaseCFG::schedule_node_into_block()?
I will run some tests with extra assertions in PhaseCFG::verify() to double-check.

Okay.

Done, see results in the related discussion (#3303 (comment)).

mlbridge · 2021-04-20T07:23:44Z

Mailing list message from Roberto Casta=c3=b1eda Lozano on hotspot-compiler-dev:

On 2021-04-19 19:10, Vladimir Kozlov wrote:

Marked as reviewed by kvn (Reviewer).

Thanks for reviewing, Vladimir!

mlbridge · 2021-04-20T07:23:46Z

Mailing list message from Roberto Casta=c3=b1eda Lozano on hotspot-compiler-dev:

On 2021-04-19 19:10, Vladimir Kozlov wrote:

Marked as reviewed by kvn (Reviewer).

Thanks for reviewing, Vladimir!

neliasso

Looks good!

robcasloz · 2021-04-20T09:01:54Z

Looks good!

Thanks again, Nils.

robcasloz · 2021-04-20T10:02:35Z

/summary
Eliminate dead nodes created by call-catch cleanup even if they have multiple
projections. Assert that definitions dominate uses and projections are scheduled
next to their parent nodes.

openjdk · 2021-04-20T10:02:56Z

@robcasloz Setting summary to:

Eliminate dead nodes created by call-catch cleanup even if they have multiple
projections. Assert that definitions dominate uses and projections are scheduled
next to their parent nodes.

robcasloz · 2021-04-20T10:05:41Z

/integrate

openjdk · 2021-04-20T10:07:13Z

@robcasloz Since your change was applied there have been 510 commits pushed to the master branch:

a25bae8: 8261445: Use memory_order_relaxed for os::random().
5136643: 8262725: IGV: crash when removing all graphs in a group
79798c6: 8265136: ZGC: Expose GarbageCollectorMXBeans for both pauses and cycles
f1d4ae6: 8263718: unused-result warning happens at os_linux.cpp
787908c: 8264221: Rewrite confusing stream API chain in SnippetMaps
142edd3: 8265152: jpackage cleanup fails on Windows with IOException deleting msi
ab22407: 8265486: ProblemList javax/sound/midi/Sequencer/Recording.java on macosx-aarch64
e0fd5fc: 8265028: JDWP debug agent thread lookup can be made faster
713483c: 8265373: Change to GCC 10.3 for building on Linux at Oracle
3990713: 8265463: ProblemList vmTestbase/vm/mlvm/mixed/stress/regression/b6969574/INDIFY_Test.java on Win-X64 -Xcomp
... and 500 more: https://git.openjdk.java.net/jdk/compare/d1baed677e30d33522033c2205138b33e8d392b6...master

Your commit was automatically rebased without conflicts.

Pushed as commit d81b046.

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

robcasloz added 6 commits March 30, 2021 12:07

Add test case

a80149d

Add basic definition dominance assertion

505a8e0

Document assumptions in live-range splitting

d783a0c

Remove cloned multi-nodes in call-catch cleanup

5b802a3

Exclude MachMerge nodes from dominance assertion

bbdaa6f

Update comments in test case

a4973ba

openjdk bot added the hotspot-compiler hotspot-compiler-dev@openjdk.org label Apr 1, 2021

robcasloz marked this pull request as ready for review April 6, 2021 07:51

openjdk bot added the rfr Pull request is ready for review label Apr 6, 2021

neliasso approved these changes Apr 6, 2021

View reviewed changes

openjdk bot added the ready Pull request is ready to be integrated label Apr 6, 2021

vnkozlov reviewed Apr 6, 2021

View reviewed changes

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

vnkozlov reviewed Apr 6, 2021

View reviewed changes

src/hotspot/share/opto/lcm.cpp Show resolved Hide resolved

vnkozlov reviewed Apr 6, 2021

View reviewed changes

robcasloz added 4 commits April 9, 2021 11:38

Assert that projections are stuck to their parents

8d96589

Simplify loop that removes dead projections, rename and comment for c…

cd77119

…larity

Enforce code style in surrounding 'for' statements

54ec8a0

Complete projection assertions with basic checks

808451d

vnkozlov approved these changes Apr 19, 2021

View reviewed changes

neliasso approved these changes Apr 20, 2021

View reviewed changes

openjdk bot closed this Apr 20, 2021

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 Apr 20, 2021

8263227: C2: inconsistent spilling due to dead nodes in exception block #3303

8263227: C2: inconsistent spilling due to dead nodes in exception block #3303

Uh oh!

Conversation

robcasloz commented Apr 1, 2021 • edited by openjdk bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Progress

Issue

Reviewers

Reviewing

Uh oh!

bridgekeeper bot commented Apr 1, 2021

Uh oh!

openjdk bot commented Apr 1, 2021

Uh oh!

mlbridge bot commented Apr 6, 2021 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Webrevs

Uh oh!

neliasso left a comment

Choose a reason for hiding this comment

Uh oh!

openjdk bot commented Apr 6, 2021 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

neliasso commented Apr 6, 2021

Uh oh!

robcasloz commented Apr 6, 2021

Uh oh!

Uh oh!

Uh oh!

vnkozlov left a comment

Choose a reason for hiding this comment

Uh oh!

robcasloz commented Apr 8, 2021

Uh oh!

vnkozlov commented Apr 8, 2021

Uh oh!

robcasloz commented Apr 12, 2021

Uh oh!

mlbridge bot commented Apr 20, 2021

Uh oh!

mlbridge bot commented Apr 20, 2021

Uh oh!

neliasso left a comment

Choose a reason for hiding this comment

Uh oh!

robcasloz commented Apr 20, 2021

Uh oh!

robcasloz commented Apr 20, 2021

Uh oh!

openjdk bot commented Apr 20, 2021

Uh oh!

robcasloz commented Apr 20, 2021

Uh oh!

openjdk bot commented Apr 20, 2021

Uh oh!

Reviewers

Assignees

Labels

Milestone

Development

Uh oh!

3 participants

robcasloz commented Apr 1, 2021 •

edited by openjdk bot

Loading

mlbridge bot commented Apr 6, 2021 •

edited

Loading

openjdk bot commented Apr 6, 2021 •

edited

Loading