JDK-8263495: Gather liveness info in the mark phase of G1 full gc #2966

Hamlin-Li · 2021-03-12T12:41:49Z

Gather liveness info in the mark phase of G1 full gc.

Per-region liveness info in the mark phase of G1 full gc is for several purposes:

JDK-8262068 needs it to determine whether compaction of a region should be skipped
JDK-8258431 for a JFR event that prints live set size estimate

so add this functionality.

Progress

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

Issue

JDK-8263495: Gather liveness info in the mark phase of G1 full gc

Reviewers

Thomas Schatzl (@tschatzl - Reviewer)
Stefan Johansson (@kstefanj - Reviewer)

Contributors

Shoubing Ma <mashoubing1@huawei.com>

Download

To checkout this PR locally:
$ git fetch https://git.openjdk.java.net/jdk pull/2966/head:pull/2966
$ git checkout pull/2966

To update a local copy of the PR:
$ git checkout pull/2966
$ git pull https://git.openjdk.java.net/jdk pull/2966/head

bridgekeeper · 2021-03-12T12:42:34Z

👋 Welcome back mli! 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 · 2021-03-12T12:43:29Z

@Hamlin-Li The following label will be automatically applied to this pull request:

hotspot-gc

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 · 2021-03-12T12:49:16Z

Webrevs

Hamlin-Li · 2021-03-15T10:03:33Z

@tschatzl Hi Thomas, the patch is ready, would you mind to review the patch when available? Thanks.

tschatzl

Other than that, initial review looks good. Thanks for extracting this piece of functionality, which makes the follow-up change likely easier to review.

tschatzl · 2021-03-15T12:42:16Z

src/hotspot/share/gc/g1/g1FullCollector.cpp

+    for (uint j = 0; j < _heap->max_regions(); j++) {
+      sum += _live_stats[j]._live_words;
+    }
+    _heap->set_used(sum * HeapWordSize);


used() is different to live() in that used() contains all space that is not allocatable (or should at least). This includes space lost due to fragmentation or just not compacting it. I think this hunk should be removed at this time until it is required.
Also, this value will be overwritten by the call to rebuild_region_sets in G1FullCollector::prepare_heap_for_mutators() as far as I can tell.

tschatzl · 2021-03-15T12:44:04Z

src/hotspot/share/gc/g1/g1FullGCMarker.hpp

+  G1RegionMarkStatsCache _mark_region_cache;
+  // Number of entries in the per-task stats entry. This seems enough to have a very
+  // low cache miss rate.
+  static const uint RegionMarkStatsCacheSize = 1024;


Please move this constant to G1RegionMarkStatsCache and reuse in both places where this constant is used instead of duplicating it.

tschatzl · 2021-03-15T12:44:40Z

src/hotspot/share/gc/g1/g1FullGCMarker.hpp

@@ -74,7 +80,8 @@ class G1FullGCMarker : public CHeapObj<mtGC> {
  inline void follow_array(objArrayOop array);
  inline void follow_array_chunk(objArrayOop array, int index);
 public:
-  G1FullGCMarker(G1FullCollector* collector, uint worker_id, PreservedMarks* preserved_stack);
+  G1FullGCMarker(G1FullCollector* collector, uint worker_id,
+                 PreservedMarks* preserved_stack, G1RegionMarkStats* mark_stats);


I think it is better at this point to put one argument per line.

tschatzl · 2021-03-15T12:52:51Z

src/hotspot/share/gc/g1/g1FullGCAdjustTask.cpp

@@ -31,6 +31,7 @@
 #include "gc/g1/g1FullGCCompactionPoint.hpp"
 #include "gc/g1/g1FullGCMarker.hpp"
 #include "gc/g1/g1FullGCOopClosures.inline.hpp"
+#include "gc/g1/g1RegionMarkStatsCache.inline.hpp"


This should not be needed, the cache is not accessed in this file.

I thought so at the beginning, but pre-submit test report compilation error on mac and linux x86, so I added these includes.
compilation error log is at: https://github.com/Hamlin-Li/jdk/runs/2095446669?check_suite_focus=true

It complains that mark_object() needs it, so it should be enough to add the include to this file: src/hotspot/share/gc/g1/g1FullGCMarker.inline.hpp which is where mark_object() is defined.

tschatzl · 2021-03-15T12:54:45Z

src/hotspot/share/gc/g1/g1FullGCOopClosures.cpp

@@ -27,6 +27,7 @@
 #include "gc/g1/g1FullCollector.hpp"
 #include "gc/g1/g1FullGCMarker.inline.hpp"
 #include "gc/g1/g1FullGCOopClosures.inline.hpp"
+#include "gc/g1/g1RegionMarkStatsCache.inline.hpp"


This file does not seem to use the G1RegionMarkStatsCache so this include seems unnecessary.

tschatzl · 2021-03-15T12:54:54Z

src/hotspot/share/gc/g1/g1FullGCPrepareTask.cpp

@@ -31,6 +31,7 @@
 #include "gc/g1/g1FullGCOopClosures.inline.hpp"
 #include "gc/g1/g1FullGCPrepareTask.hpp"
 #include "gc/g1/g1HotCardCache.hpp"
+#include "gc/g1/g1RegionMarkStatsCache.inline.hpp"


Not needed.

tschatzl · 2021-03-15T12:55:03Z

src/hotspot/share/gc/g1/g1FullGCReferenceProcessorExecutor.cpp

@@ -28,6 +28,7 @@
 #include "gc/g1/g1FullGCMarker.hpp"
 #include "gc/g1/g1FullGCOopClosures.inline.hpp"
 #include "gc/g1/g1FullGCReferenceProcessorExecutor.hpp"
+#include "gc/g1/g1RegionMarkStatsCache.inline.hpp"


Not needed.

tschatzl · 2021-03-15T12:56:15Z

src/hotspot/share/gc/g1/heapRegion.hpp

@@ -27,6 +27,7 @@

 #include "gc/g1/g1BlockOffsetTable.hpp"
 #include "gc/g1/g1HeapRegionTraceType.hpp"
+#include "gc/g1/g1RegionMarkStatsCache.hpp"


This file does not use G1RegionMarkStatsCache, so shouldn't need an include.

Thanks for pointing out.
it compiles successfully on my local env, let's see if it will compile in pre-submit build/test

This typically indicates that the header include is missing elsewhere. One thing that often helps is compiling with configure disabled precompiled headers, ie. --disable-precompiled-headers in configure.

tschatzl · 2021-03-15T13:12:15Z

src/hotspot/share/gc/g1/g1RegionMarkStatsCache.cpp

@@ -44,6 +45,17 @@ G1RegionMarkStatsCache::~G1RegionMarkStatsCache() {
  FREE_C_HEAP_ARRAY(G1RegionMarkStatsCacheEntry, _cache);
 }

+// cache size is equal to or bigger than region size to intialize region_index
+void G1RegionMarkStatsCache::initialize() {


If the change really wants to avoid the initial cache misses (for regions != 0), please merge this with the existing reset() method. I do not think it makes a difference to have this optimization that exploits the simplicity of G1RegionMarkStatsCacheEntry::hash().
(And if the hash() implementation ever changes, this optimization won't work any more).

If you do so, also add a clear comment why the cache is initialized with these values. The comment

// cache size is equal to or bigger than region size to intialize region_index

seems at least misplaced.

The change could, at the assignment (maybe even add an optional parameter to clear()) state:

// Avoid the initial cache miss and eviction by setting the i'th's cache region_idx to the region_idx due to how the hash is calculated.

Note that resetting to constant values in that loop will likely be folded into a memset, while this loop needs to be translated as is, and may be a lot slower (with 100k+ regions...) in the pause.

In a textbook implementation, G1RegionMarkStatsCacheEntry::clear() should probably set the region index to G1_NO_HRM_INDEX (i.e. (uint)-1) to indicate emptiness so that all initial cache lookups fail (i.e. evict the previous, non-existing value). But then the simple implementation of the G1RegionMarkStatsCache::evict method will access the liveness array out of bounds... (that's probably why the initial region_idx values are zero, not as some sort of optimization - we save some conditional in that method).

Thanks Thomas, I will merge reset and initialize, and add the comment you suggested.

BTW, a quick question: is there a limit on the region number in G1?

The limit is 2^32-2, i.e. the range of uint - 1 for that G1_NO_HRM_INDEX which is a special value.

kstefanj

Just a few additional comments to the ones made by Thomas.

kstefanj · 2021-03-16T13:44:39Z

src/hotspot/share/gc/g1/g1FullGCAdjustTask.cpp

@@ -31,6 +31,7 @@
 #include "gc/g1/g1FullGCCompactionPoint.hpp"
 #include "gc/g1/g1FullGCMarker.hpp"
 #include "gc/g1/g1FullGCOopClosures.inline.hpp"
+#include "gc/g1/g1RegionMarkStatsCache.inline.hpp"


It complains that mark_object() needs it, so it should be enough to add the include to this file: src/hotspot/share/gc/g1/g1FullGCMarker.inline.hpp which is where mark_object() is defined.

kstefanj · 2021-03-16T13:48:17Z

src/hotspot/share/gc/g1/g1FullGCMarker.cpp

@@ -25,14 +25,16 @@
 #include "precompiled.hpp"
 #include "classfile/classLoaderData.hpp"
 #include "gc/g1/g1FullGCMarker.inline.hpp"
+#include "gc/g1/g1RegionMarkStatsCache.inline.hpp"


Shouldn't need the inline.hpp here, .hpp should be enough.

Thanks Stefan.

kstefanj

Starting to look good, just a few more cleanups.

kstefanj · 2021-03-16T15:01:27Z

src/hotspot/share/gc/g1/g1FullGCMarkTask.cpp

@@ -30,6 +30,7 @@
 #include "gc/g1/g1FullGCMarkTask.hpp"
 #include "gc/g1/g1FullGCOopClosures.inline.hpp"
 #include "gc/g1/g1FullGCReferenceProcessorExecutor.hpp"
+#include "gc/g1/g1RegionMarkStatsCache.inline.hpp"


Not needed afaict.

kstefanj · 2021-03-16T15:06:30Z

src/hotspot/share/gc/g1/g1FullGCMarker.hpp

+  void flush_mark_region_cache() {
+    _mark_region_cache.evict_all();
+  }


Please move this definition to the cpp-file.

kstefanj · 2021-03-16T15:07:25Z

src/hotspot/share/gc/g1/g1FullGCMarker.hpp

@@ -63,6 +63,9 @@ class G1FullGCMarker : public CHeapObj<mtGC> {
  G1FollowStackClosure _stack_closure;
  CLDToOopClosure      _cld_closure;

+
+  G1RegionMarkStatsCache _mark_region_cache;


Please add the include to this file:
#include "gc/g1/g1RegionMarkStatsCache.hpp"

Thanks Stefan, Just modified as you suggested.

Hamlin-Li · 2021-03-17T04:59:27Z

There are windows build failure, but it seems not related to my patch.

Hamlin-Li · 2021-03-17T05:00:33Z

/contributor add Shoubing Ma mashoubing1@huawei.com

openjdk · 2021-03-17T05:00:52Z

@Hamlin-Li Could not parse Shoubing Ma mashoubing1@huawei.com as a valid contributor.
Syntax: /contributor (add|remove) [@user | openjdk-user | Full Name <email@address>]. For example:

/contributor add @openjdk-bot
/contributor add duke
/contributor add J. Duke <duke@openjdk.org>

kstefanj · 2021-03-17T09:24:05Z

Thanks for addressing the comments, I've kicked off some additional functional testing and will do some manual performance testing as well.

Hamlin-Li · 2021-03-17T09:38:02Z

/contributor add Shoubing Ma mashoubing1@huawei.com

openjdk · 2021-03-17T09:48:21Z

@Hamlin-Li
Contributor Shoubing Ma <mashoubing1@huawei.com> successfully added.

tschatzl

Some additional sweep through the code showed some more minor issues. Sorry for not noticing earlier.

tschatzl · 2021-03-17T09:52:06Z

src/hotspot/share/gc/g1/g1FullGCMarkTask.cpp

@@ -60,6 +60,8 @@ void G1FullGCMarkTask::work(uint worker_id) {

  // Mark stack is populated, now process and drain it.
  marker->complete_marking(collector()->oop_queue_set(), collector()->array_queue_set(), &_terminator);
+  // flush live bytes to regions
+  marker->flush_mark_region_cache();


Please use mark_stats instead of mark_region in all identifier names. It is kind of jarring to have the class called G1Mark*Stats*Cache, all other related variables also with something containing mark_stats and then to use something else in the instances for this for no discernible reason.

I.e. please rename this method and the member from _mark_region_cache to _mark_stats_cache.

Of course, if you have a reason to do so please tell.

tschatzl · 2021-03-17T09:54:15Z

src/hotspot/share/gc/g1/g1FullGCMarkTask.cpp

@@ -60,6 +60,8 @@ void G1FullGCMarkTask::work(uint worker_id) {

  // Mark stack is populated, now process and drain it.
  marker->complete_marking(collector()->oop_queue_set(), collector()->array_queue_set(), &_terminator);
+  // flush live bytes to regions


Comment containing full sentences should start with upper case like the one before and end with a full stop. Also that comment is probably better placed at the declaration in the .hpp file.

tschatzl · 2021-03-17T09:58:15Z

src/hotspot/share/gc/g1/g1RegionMarkStatsCache.cpp

+// Reset liveness of all cache entries to their default values,
+// initialize _region_idx to avoid initial cache miss.


Comments describing the functionality of a method should be at the declaration. Only implementation details should be added at the definition.

tschatzl · 2021-03-17T09:59:32Z

src/hotspot/share/gc/g1/g1RegionMarkStatsCache.hpp

-    void clear() {
-      _region_idx = 0;
+    void clear(uint idx = 0) {
+      _region_idx = idx;


I unfortunately can't comment on the other method is_clear() here - please remove because it is a) wrong now with that special initialization and b) not used anywhere apparently.

Thanks for the detailed review.

tschatzl · 2021-03-17T10:01:35Z

src/hotspot/share/gc/g1/g1RegionMarkStatsCache.hpp

@@ -98,6 +98,10 @@ class G1RegionMarkStatsCache {

  G1RegionMarkStatsCacheEntry* find_for_add(uint region_idx);
 public:
+  // Number of entries in the per-task stats entry. This seems enough to have a very


pre-existing: please add "value" after the "This" to read "This value seems enough to...." (or "The suggested value" or something to that effect.

kstefanj · 2021-03-17T12:40:20Z

Testing showed that the regression I reported in the other PR is still present for this fix. Same small reproducer can be used:

public class SystemGCLarge {
  public static Object[] holder;
  public static void main(String args[]) {
    holder = new Object[128 * 1024 *1024];
    System.gc();
    System.out.println("Done");
  }
}

I've made a change that eliminates the regression by making sure the cache update is not inlined and by doing so allows for other inlining to be done:
kstefanj@abf54fb

Hamlin-Li · 2021-03-18T02:02:14Z

Testing showed that the regression I reported in the other PR is still present for this fix. Same small reproducer can be used:
public class SystemGCLarge {
  public static Object[] holder;
  public static void main(String args[]) {
    holder = new Object[128 * 1024 *1024];
    System.gc();
    System.out.println("Done");
  }
}
I've made a change that eliminates the regression by making sure the cache update is not inlined and by doing so allows for other inlining to be done:
kstefanj@abf54fb

On my local env, it seems make no difference wIth or w/o your patch, for both marking phase and whole full gc pause. But I will apply your patch, Thanks a lot for the benchmarking.

kstefanj · 2021-03-18T08:14:38Z

On my local env, it seems make no difference wIth or w/o your patch, for both marking phase and whole full gc pause. But I will apply your patch, Thanks a lot for the benchmarking.

What inlining gets done will depend a lot on the build env, so it might be that your build don't experience the same issue. Because you mean that you can't reproduce the regression at all right? Or is the regression present even after my patch?

Hamlin-Li · 2021-03-18T08:55:23Z

It's all the same, following data is in the order of master/with/w/o, (I don't "make clean" before make images)

$ java -Xlog:gc+phases=trace SystemGCLarge | grep "Phase 1: Mark live objects"
[0.518s][info ][gc,phases] GC(0) Phase 1: Mark live objects 68.763ms
$ java -Xlog:gc+phases=trace SystemGCLarge | grep "Phase 1: Mark live objects"
[0.499s][info ][gc,phases] GC(0) Phase 1: Mark live objects 69.115ms
$ java -Xlog:gc+phases=trace SystemGCLarge | grep "Phase 1: Mark live objects"
[0.509s][info ][gc,phases] GC(0) Phase 1: Mark live objects 75.952ms
$ java -Xlog:gc+phases=trace SystemGCLarge | grep "Phase 1: Mark live objects"
[0.519s][info ][gc,phases] GC(0) Phase 1: Mark live objects 68.478ms
$ java -Xlog:gc+phases=trace SystemGCLarge | grep "Phase 1: Mark live objects"
[0.529s][info ][gc,phases] GC(0) Phase 1: Mark live objects 69.646ms

$ java -Xlog:gc+phases=trace SystemGCLarge | grep "Phase 1: Mark live objects"
[0.499s][info ][gc,phases] GC(0) Phase 1: Mark live objects 68.603ms
$ java -Xlog:gc+phases=trace SystemGCLarge | grep "Phase 1: Mark live objects"
[0.519s][info ][gc,phases] GC(0) Phase 1: Mark live objects 72.012ms
$ java -Xlog:gc+phases=trace SystemGCLarge | grep "Phase 1: Mark live objects"
[0.515s][info ][gc,phases] GC(0) Phase 1: Mark live objects 69.525ms
$ java -Xlog:gc+phases=trace SystemGCLarge | grep "Phase 1: Mark live objects"
[0.512s][info ][gc,phases] GC(0) Phase 1: Mark live objects 69.656ms
$ java -Xlog:gc+phases=trace SystemGCLarge | grep "Phase 1: Mark live objects"
[0.512s][info ][gc,phases] GC(0) Phase 1: Mark live objects 69.365ms
$ java -Xlog:gc+phases=trace SystemGCLarge | grep "Phase 1: Mark live objects"
[0.517s][info ][gc,phases] GC(0) Phase 1: Mark live objects 69.441ms

$ java -Xlog:gc+phases=trace SystemGCLarge | grep "Phase 1: Mark live objects"
[0.501s][info ][gc,phases] GC(0) Phase 1: Mark live objects 67.338ms
$ java -Xlog:gc+phases=trace SystemGCLarge | grep "Phase 1: Mark live objects"
[0.511s][info ][gc,phases] GC(0) Phase 1: Mark live objects 75.821ms
$ java -Xlog:gc+phases=trace SystemGCLarge | grep "Phase 1: Mark live objects"
[0.508s][info ][gc,phases] GC(0) Phase 1: Mark live objects 68.903ms
$ java -Xlog:gc+phases=trace SystemGCLarge | grep "Phase 1: Mark live objects"
[0.499s][info ][gc,phases] GC(0) Phase 1: Mark live objects 68.870ms
$ java -Xlog:gc+phases=trace SystemGCLarge | grep "Phase 1: Mark live objects"
[0.495s][info ][gc,phases] GC(0) Phase 1: Mark live objects 67.374ms
$ java -Xlog:gc+phases=trace SystemGCLarge | grep "Phase 1: Mark live objects"
[0.501s][info ][gc,phases] GC(0) Phase 1: Mark live objects 69.018ms

kstefanj · 2021-03-18T09:23:40Z

Thanks for verifying.

tschatzl

Lgtm assuming that Windows build failure is just something specific to Github. Thanks.

To answer my own question: it builds internally on Windows. So it's good to go.

openjdk · 2021-03-18T12:45:09Z

@Hamlin-Li 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:

8263495: Gather liveness info in the mark phase of G1 full gc

Co-authored-by: Shoubing Ma <mashoubing1@huawei.com>
Reviewed-by: tschatzl, sjohanss

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 791 new commits pushed to the master branch:

a85dc55: 8263311: Watch registry changes for remote printers update instead of polling
3f31a6b: 8263775: C2: igv_print() crash unexpectedly when called from debugger
63eae8f: 8260605: Various java.lang.invoke cleanups
9cd21b6: 8263590: Rawtypes warnings should be produced for pattern matching in instanceof
ff52f29: 8260716: Assert in MacroAssembler::clear_mem with -XX:-IdealizeClearArrayNode
72b82fd: 8263725: JFR oldobject tests are not run when GCs are specified explicitly
444a80b: 8263455: NMT: assert on registering a region which completely engulfs an existing region
2b93ae0: 8261480: MetaspaceShared::preload_and_dump should check exceptions
81ba578: 8263676: AArch64: one potential bug in C1 LIRGenerator::generate_address()
9225a23: 8263108: Class initialization deadlock in java.lang.constant
... and 781 more: https://git.openjdk.java.net/jdk/compare/35c9da70316da21a6be2fd92a5e5b5f193bdd9d6...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.

kstefanj

I'm good with the current changes as well.

Hamlin-Li · 2021-03-18T14:14:24Z

/integrate

albertnetymk · 2021-03-18T14:14:12Z

src/hotspot/share/gc/g1/g1FullCollector.cpp

+  for (uint j = 0; j < heap->max_regions(); j++) {
+    _live_stats[j].clear();
+  }


Nit: i could be used instead of j, right?

openjdk · 2021-03-18T14:15:34Z

@Hamlin-Li Since your change was applied there have been 791 commits pushed to the master branch:

a85dc55: 8263311: Watch registry changes for remote printers update instead of polling
3f31a6b: 8263775: C2: igv_print() crash unexpectedly when called from debugger
63eae8f: 8260605: Various java.lang.invoke cleanups
9cd21b6: 8263590: Rawtypes warnings should be produced for pattern matching in instanceof
ff52f29: 8260716: Assert in MacroAssembler::clear_mem with -XX:-IdealizeClearArrayNode
72b82fd: 8263725: JFR oldobject tests are not run when GCs are specified explicitly
444a80b: 8263455: NMT: assert on registering a region which completely engulfs an existing region
2b93ae0: 8261480: MetaspaceShared::preload_and_dump should check exceptions
81ba578: 8263676: AArch64: one potential bug in C1 LIRGenerator::generate_address()
9225a23: 8263108: Class initialization deadlock in java.lang.constant
... and 781 more: https://git.openjdk.java.net/jdk/compare/35c9da70316da21a6be2fd92a5e5b5f193bdd9d6...master

Your commit was automatically rebased without conflicts.

Pushed as commit 8c8d1b3.

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

JDK-8263495: Gather liveness info in the mark phase of G1 full gc

c47c30d

openjdk bot added hotspot-gc hotspot-gc-dev@openjdk.org rfr Pull request is ready for review labels Mar 12, 2021

Hamlin-Li added 2 commits March 12, 2021 21:38

fix compilation error

207b77c

fix compilation error

51325ed

tschatzl suggested changes Mar 15, 2021

View reviewed changes

kstefanj suggested changes Mar 16, 2021

View reviewed changes

Hamlin-Li added 2 commits March 16, 2021 22:36

misc improvement.

278326e

refine inline includes.

4dd36cd

kstefanj reviewed Mar 16, 2021

View reviewed changes

refine includes.

7dc9751

tschatzl suggested changes Mar 17, 2021

View reviewed changes

refine names and comments, delete dead code.

d24e7de

tschatzl mentioned this pull request Mar 17, 2021

8258431: Provide a JFR event with live set size estimate #2579

Closed

3 tasks

better inline suggestions.

b133747

Hamlin-Li mentioned this pull request Mar 18, 2021

JDK-8262068: Improve G1 Full GC by skipping compaction for regions with high survival ratio #2760

Closed

3 tasks

tschatzl approved these changes Mar 18, 2021

View reviewed changes

openjdk bot added the ready Pull request is ready to be integrated label Mar 18, 2021

kstefanj approved these changes Mar 18, 2021

View reviewed changes

albertnetymk approved these changes Mar 18, 2021

View reviewed changes

openjdk bot closed this Mar 18, 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 Mar 18, 2021

Hamlin-Li deleted the gather-liveness-g1 branch March 25, 2021 11:37

		// Reset liveness of all cache entries to their default values,
		// initialize _region_idx to avoid initial cache miss.

JDK-8263495: Gather liveness info in the mark phase of G1 full gc #2966

JDK-8263495: Gather liveness info in the mark phase of G1 full gc #2966

Conversation

Hamlin-Li commented Mar 12, 2021 • edited by openjdk bot Loading

Progress

Issue

Reviewers

Contributors

Download

bridgekeeper bot commented Mar 12, 2021

openjdk bot commented Mar 12, 2021

mlbridge bot commented Mar 12, 2021 • edited Loading

Webrevs

Hamlin-Li commented Mar 15, 2021

tschatzl left a 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

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

kstefanj left a 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

kstefanj left a 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

Hamlin-Li Mar 17, 2021 • edited Loading

Choose a reason for hiding this comment

Hamlin-Li commented Mar 17, 2021

Hamlin-Li commented Mar 17, 2021

openjdk bot commented Mar 17, 2021

kstefanj commented Mar 17, 2021

Hamlin-Li commented Mar 17, 2021

openjdk bot commented Mar 17, 2021

tschatzl left a 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

kstefanj commented Mar 17, 2021

Hamlin-Li commented Mar 18, 2021

kstefanj commented Mar 18, 2021

Hamlin-Li commented Mar 18, 2021

kstefanj commented Mar 18, 2021

tschatzl left a comment • edited Loading

Choose a reason for hiding this comment

openjdk bot commented Mar 18, 2021 • edited Loading

kstefanj left a comment

Choose a reason for hiding this comment

Hamlin-Li commented Mar 18, 2021

Choose a reason for hiding this comment

openjdk bot commented Mar 18, 2021

Hamlin-Li commented Mar 12, 2021 •

edited by openjdk bot

Loading

mlbridge bot commented Mar 12, 2021 •

edited

Loading

Hamlin-Li Mar 17, 2021 •

edited

Loading

tschatzl left a comment •

edited

Loading

openjdk bot commented Mar 18, 2021 •

edited

Loading