Javascript:SpiderMonkey:ProjectGenerationGarbageCollection: Difference between revisions

Mark as Outdated
(Created page with "Objective: Implement Generational Garbage collection in Spider Monkey. <Some goal based off V8 performance delta % on Earley Boyer benchmark> Accountable: Naveed Responsible: Te...")
 
(Mark as Outdated)
 
(61 intermediate revisions by 9 users not shown)
Line 1: Line 1:
Objective: Implement Generational Garbage collection in Spider Monkey. <Some goal based off V8 performance delta % on Earley Boyer benchmark>
{{outdated}}


Accountable: Naveed
<p>Wiki conversion of original Google Document: https://docs.google.com/a/jamsni.com/document/d/1-GZ8F0ZabvdpCnRQaecId0yuu4QFYr5pAvnRC_KFchc/edit
Responsible: Terrence, Jon, and Steve, evilpie Ms2ger
</p>
Consulted: bhackett billm
Informed: Product Marketing


Generational GC - Bug 619558
== Objective ==
Steps
<p>Implement Generational Garbage collection in Spider Monkey. &lt;Some goal based off V8 performance delta&#160;% on Earley Boyer benchmark&gt;</p>
Exact Rooting done in JS Shell [Bug 753203] - 12 weeks started
<b>Accountable</b>: Naveed<br />
Rooting Analysis on TinderBox (sfink) - 3 weeks
<b>Responsible</b>: Terrence, Steve, Jon, Nicholas, evilpie and Ms2ger<br />
patch is ready for review, would fix the problem with some pushes not getting builds
<b>Consulted</b>: bhackett billm<br />
fix up other stuff broken by above change
<b>Informed</b>: Product Marketing<br />
patch would fix last remaining problem with getting it on Try server
<br/>
Unhide it for inbound - keep it green
<b>Tracking Bug:</b> [meta] Implement generational garbage collection - <span class="fck_mw_template"><span class="fck_mw_template"><span class="fck_mw_template">{{bug|619558}}</span></span></span>
Get the root analysis build green - two weeks?
Get jit-tests green [Bug 745742] - 2 days
Get js ref tests green
Get jsapi-tests green [Bug 831376]
Fuzzer coverage of code [Bug 773746]
rely on static analysis to make this not be a whack-a-mole game
Remove E4X
Do something about JSD
Add exact roots to stack structures - 2 weeks
AddRoot/RemoveRoot for Heap structures - 4 weeks
Static code analysis [Bug 831409]
Fix all discovered rooting hazards (~800) (sfink,jonco) - 4 weeks
Optimize all discovered over-rooting (~100)
Automate static analysis (sfink)
Need a server (dm-sixgill01?)
http://people.mozilla.org/~bhackett/rootingHazards.html
http://people.mozilla.org/~bhackett/gcFunctions.html


JIT Integration with post barriers - 4 weeks, parallelized
== Milestones ==
IonMonkey [Bug 831506] - 1 week (bhackett)
JaegerMonkey [Bug 764876] - 2 days (bhackett)
Baseline JIT [Bug 831507] - <ask jandem/djvj for an estimate>


Generation Garbage Collection in the Shell
=== Endgame schedule ===
Implement prototype algorithm (terrence) - 1 week
This is a potential milestone: we would need to build it and test it on TBPL similar to how |r| works now -- |GGCJS|.


Exactly Root the Browser [Bug 831379] - 10 weeks
* Burndown of rooting hazards and unsafe references: https://people.mozilla.com/~sfink/analysis
This is a potential milestone: we would turn on exact rooting for release FF at this point. We would not get a performance boost from this (necessarily), but it would lock in our work to this point.
* Burndown of bugs blocking Exact Rooting: [https://metrics.mozilla.com/bugzilla-analysis/Dashboard-FinalBurndown.html#esfilter=%257B%22or%22%3A%255B%0A%09%257B%22term%22%3A%257B%22bug_id%22%3A753203%257D%257D,%0A%09%257B%22term%22%3A%257B%22blocked_by%22%3A745742%257D%257D,%0A%09%257B%22term%22%3A%257B%22blocked_by%22%3A773686%257D%257D,%0A%09%257B%22term%22%3A%257B%22blocked_by%22%3A789551%257D%257D,%0A%09%257B%22term%22%3A%257B%22blocked_by%22%3A791022%257D%257D,%0A%09%257B%22term%22%3A%257B%22blocked_by%22%3A906940%257D%257D,%0A%09%257B%22term%22%3A%257B%22blocked_by%22%3A813244%257D%257D,%0A%09%257B%22term%22%3A%257B%22blocked_by%22%3A848151%257D%257D,%0A%09%257B%22term%22%3A%257B%22blocked_by%22%3A831379%257D%257D,%0A%09%257B%22term%22%3A%257B%22blocked_by%22%3A868799%257D%257D,%0A%09%257B%22term%22%3A%257B%22blocked_by%22%3A877658%257D%257D,%0A%09%257B%22term%22%3A%257B%22blocked_by%22%3A884617%257D%257D,%0A%09%257B%22term%22%3A%257B%22blocked_by%22%3A898220%257D%257D,%0A%09%257B%22term%22%3A%257B%22blocked_by%22%3A898606%257D%257D,%0A%09%257B%22term%22%3A%257B%22blocked_by%22%3A791062%257D%257D,%0A%09%257B%22term%22%3A%257B%22blocked_by%22%3A834909%257D%257D,%0A%09%257B%22term%22%3A%257B%22blocked_by%22%3A831409%257D%257D,%0A%09%257B%22term%22%3A%257B%22blocked_by%22%3A898554%257D%257D,%0A%09%257B%22term%22%3A%257B%22blocked_by%22%3A901731%257D%257D,%0A%09%257B%22term%22%3A%257B%22blocked_by%22%3A898970%257D%257D,%0A%09%257B%22term%22%3A%257B%22blocked_by%22%3A864785%257D%257D,%0A%09%257B%22term%22%3A%257B%22blocked_by%22%3A868483%257D%257D,%0A%09%257B%22term%22%3A%257B%22blocked_by%22%3A877699%257D%257D,%0A%09%257B%22term%22%3A%257B%22blocked_by%22%3A898815%257D%257D,%0A%09%257B%22term%22%3A%257B%22blocked_by%22%3A900144%257D%257D,%0A%09%257B%22term%22%3A%257B%22blocked_by%22%3A899251%257D%257D,%0A%09%257B%22term%22%3A%257B%22blocked_by%22%3A899251%257D%257D,%0A%09%257B%22term%22%3A%257B%22blocked_by%22%3A900998%257D%257D,%0A%09%257B%22term%22%3A%257B%22blocked_by%22%3A898608%257D%257D,%0A%09%257B%22term%22%3A%257B%22blocked_by%22%3A900713%257D%257D%0A%255D%257D&forTitle=Exact+Rooting&dueDate=2013-10-22&sampleMin=2013-08-01 https://metrics.mozilla.com/...]


Performance Tuning - 4 weeks (This should probably start asap)
{| class="wikitable"
Test against V8 Earley-Boyer benchmark.
! Milestone
Also v8 deltablue, raytrace
! Bug
Implement a Nursery [Bug 706885] - 1 week
! Old Estimated Date
We may need to Implement Bill’s Pools/Zones idea to get the nursery to the requisite perf - 6 weeks
! Current Estimated Date
Refactor code to avoid rooting on hot paths and keep rooter overhead acceptable - ??? weeks (start early) (bhackett)
! Actual Date
! Nightly Version
|-
| ...
|
|
|
|
|-
| style="background:#def;" | [[RapidRelease/Calendar|Nightly 26 (B2G 1.2) is uplifted to Aurora]]
| style="background:#def;" |
| style="background:#def;" | 2013-09-16
| style="background:#def;" |
| style="background:#def;" |
| style="background:#def;" |
|-
| All unsafe references found by static analysis have been fixed
| [https://bugzilla.mozilla.org/show_bug.cgi?id=898608 898608]
| 2013-09-23
| 2013-10-02
|
| 27
|-
| Static analysis of unsafe references is unhidden on TBPL
| [https://bugzilla.mozilla.org/show_bug.cgi?id=834909 834909]
| 2013-09-25
| 2013-10-03
|
| 27
|-
| All rooting hazards found by static analysis have been fixed
| [https://bugzilla.mozilla.org/show_bug.cgi?id=898606 898606]
| 2013-10-18
| 2013-10-07
|
| 27
|-
| Static analysis of rooting hazards is unhidden on TBPL
| [https://bugzilla.mozilla.org/show_bug.cgi?id=834909 834909]
| 2013-10-22
| 2013-10-08
|
| 27
|-
| style="background:#def;" | [[RapidRelease/Calendar|Nightly 27 is uplifted to Aurora]]
| style="background:#def;" |
| style="background:#def;" | 2013-10-28
| style="background:#def;" |
| style="background:#def;" |
| style="background:#def;" |
|-
| ...
|
|
|
|
| 28
|-
| All GGC crashes found by fuzzers have been fixed
| [https://bugzilla.mozilla.org/show_bug.cgi?id=877471 877471]
| 2013-11-__
| 2013-10-__
|
| 28
|-
| AWFY with GGC >= AWFY without GGC
| [https://bugzilla.mozilla.org/show_bug.cgi?id=875863 875863]
| 2013-11-__
|
|
| 28
|-
| GGC is enabled for desktop and Android
| [https://bugzilla.mozilla.org/show_bug.cgi?id=619558 619558]
| 2013-11-__
|
|
| 28
|-
| GGC is enabled for B2G 1.3
| [https://bugzilla.mozilla.org/show_bug.cgi?id=619558 619558]
| 2013-11-__
|
|
| 28
|-
| ...
|
|
|
|
| 28
|-
| style="background:#def;" | [[RapidRelease/Calendar|Nightly 28 (B2G 1.3) is uplifted to Aurora]]
| style="background:#def;" |
| style="background:#def;" | 2013-12-09
| style="background:#def;" |
| style="background:#def;" |
| style="background:#def;" |
|-
| ...
|
|
|
|
| 29
|-
| style="background:#def;" | [[RapidRelease/Calendar|Nightly 29 is uplifted to Aurora]]
| style="background:#def;" |
| style="background:#def;" | 2014-01-20
| style="background:#def;" |
| style="background:#def;" |
| style="background:#def;" |
|-
| ...
|
|
|
|
| 30
|}


Re-implement HashTable rekeying
=== Original schedule ===
We undid this code because it was a perf regression and was still a bit buggy. (terrence) - 3 weeks


Make the Post Barrier Verifier Green in the browser [Bug 764882]
{| class="wikitable"
Investigate how long it will take to do generational barriers - 1 week
! Milestone
! Task
! Deliverable
! Assigned
! Estimated Weeks
! Expected Date
! Actual Date
|-
| 1
| Rooting analysis in the shell (x64)
| green on mozilla-inbound
| sfink, jonco, evilpie, Ms2ger
| 2
| 3/7/2013
| 3/2/2013
|-
| 2
| Rooting analysis in the shell (arm)
| green on mozilla-inbound
| sfink
| 2
|
| pending
|-
| 3
| Exactly rooted shell (x64)
| green on mozilla-inbound
| sfink, jonco, evilpie, Ms2ger
| 2
| 3/14/2013
| done?
|-
| 4
| Exactly rooted shell (arm)
| green on mozilla-inbound
| sfink
| 2
|
| pending
|-
| 4.1
| Exactly rooted js/src
| green on mozilla-inbound
| jonco
| 3
| 5/9/2013
| in progress
|-
| 5
| GGC no-jit shell only build on TBPL (x64)
| green on mozilla-inbound
| terrence
| 6
| 2/21/2013
| 4/7/2013
|-
| 6
| GGC no-jit shell only build on TBPL (arm)
| green on mozilla-inbound
| sfink, terrence
| 2
|
| pending
|-
| 7
| JIT support (x64) (Bug 706885)
| TBPL build using JITs and Ion+Baseline+GGC displayed on AWFY
| bhacket
| 8
| 3/15/2013
| 4/7/2013
|-
| 8
| JIT support (arm)
| TBPL build using JITs and Ion+Baseline+GGC displayed on AWFY
| bhacket
| 2
|
| pending
|-
| 9
| Optimized GGC in shell (x64 + arm)
| Ion+GGC faster than Ion-GGC on AWFY
| bhacket
| 12
| 6/1/2013   
| not started
|-
| 10
| Static rooting analysis - browser
| available via mrgiggles and URL
| sfink, bhacket
| 12
| 6/1/2013
| done
|-
| 10.1
| Static analysis on TBPL - browser
| visible on tbpl
| sfink
|
|
| not started           
|-
| 10.2
| Static rooting analysis - b2g
| available via mrgiggles and URL
| sfink
|
|
| pending
|-
| 10.3
| Static analysis on TBPL - b2g
| visible on tbpl
| sfink
|
|
| not started           
|-
| 11
| Exactly rooted browser (x64 + arm)
| green on mozilla-inbound
| sfink, jonco, evilpie, Ms2ger, terrence
| 16
| 7/1/2013   
| in progress
|-
| 12
| GGC in the browser (x64 + arm)
| green on AWFY
| terrence
| 16
| 10/1/2013
| not started
|}


We may need to rewrite the maps in xpconnect and the browser in terms of HashTable: this could be a bunch of work.
== Steps ==
Make JS_IsAboutToBeFinalized indirect [Bug 765432] - 10 weeks


OTHER
=== Exact Rooting done in JS Shell {{bug|753203}} ===


Should we establish a new  benchmark specifically for GGC. (sfink votes yes)
<p>12 weeks started
What would it measure? there are multiple goals, e.g. throughput/MMU/pause time
</p>
Should we make it a public benchmark?
<ul><li> Dynamic rooting analysis on tbpl (sfink) - done
What workloads should we consider?
<ul><li> Unhidden on inbound</li></ul>
compartmental GC very important to us, not necessarily applicable to other implementations
</li><li> Get the root analysis build green - done
allocation rate (broken down into live vs garbage), steady behavior vs swapping between allocation + computation modes, etc.
<ul><li> Get jit-tests green <span class="fck_mw_template"><span class="fck_mw_template">{{bug|745742}}</span></span>
How can we keep our advantage on Splay benchmark
</li><li> Get js ref tests green
possible to do with TI --- look at types of objects promoted from nursery to major heap, eventually start allocating them directly in the major heap
</li><li> Get jsapi-tests green <span class="fck_mw_template"><span class="fck_mw_template">{{bug|831376}}</span></span>
Compacting GC - Revisit in Febuary/March - 1 Month
</li></ul>
Delayed until exact rooting is fully done. We can add this in if we have time in the schedule that we cannot parallelize.
</li><li> Rooting analysis fuzz bugs <span class="fck_mw_template"><span class="fck_mw_template">{{bug|773746}}</span></span>
<ul><li> Rely on static analysis to make this not be a whack-a-mole game
</li></ul>
</li><li> Remove E4X - done
</li><li> Do something about JSD
</li><li> Add exact roots to stack structures - 2 weeks
</li><li> AddRoot/RemoveRoot for Heap structures - 4 weeks
</li><li> Static code analysis <span class="fck_mw_template"><span class="fck_mw_template">{{bug|831409}}</span></span>
<ul><li> Fix all discovered rooting hazards (~800) (sfink,jonco) - 4 weeks
</li><li> Optimize all discovered over-rooting (~100)
</li><li> Automate static analysis <span class="fck_mw_template"><span class="fck_mw_template">{{bug|834912}}</span></span> - done (sfink)
</li><li> http://people.mozilla.org/~sfink/analysis/browser/rootingHazards.txt
</li><li> http://people.mozilla.org/~sfink/analysis/browser/gcFunctions.txt
</li><li> http://people.mozilla.org/~sfink/analysis/browser/allFunctions.txt
</li></ul>
</li></ul>


Risks
=== JIT Integration with post barriers - 4 weeks, parallelized ===
Team is responsible for many top crashers
 
External rooting API has not been designed  
<ul><li> IonMonkey <span class="fck_mw_template"><span class="fck_mw_template">{{bug|831506}}</span></span> - 1 week (bhackett)
GGC algorithm has not been decided on - it may not be faster
</li><li> JaegerMonkey <span class="fck_mw_template"><span class="fck_mw_template">{{bug|764876}}</span></span> - 2 days (bhackett)
JSD1 exact rooting is lurking
</li><li> Baseline JIT <span class="fck_mw_template"><span class="fck_mw_template">{{bug|831507}}</span></span> - &lt;ask jandem/djvj for an estimate&gt;
</li></ul>
 
=== Generational Garbage Collection in the Shell ===
 
<ul><li> Implement prototype algorithm (terrence) - 1 week
</li><li> This is a potential milestone: we would need to build it and test it on TBPL similar to how |r| works now -- |GGCJS| - done.
</li><li> Re-implement HashTable rekeying
<ul><li> We undid this code because it was a perf regression and was still a bit buggy. (terrence) - 3 weeks
</li><li> at least part of this is <span class="fck_mw_template"><span class="fck_mw_template">{{bug|726687}}</span></span>
</li></ul>
</li></ul>
 
=== Exactly Root the Browser <span class="fck_mw_template"><span class="fck_mw_template">{{bug|831379}}</span></span> - 10 weeks ===
 
<ul><li> This is a potential milestone: we would turn on exact rooting for release FF at this point. We would not get a performance boost from this (necessarily), but it would lock in our work to this point.
</li></ul>
 
=== Performance Tuning - 4 weeks (This should probably start asap) ===
 
<ul><li> Implement a Nursery <span class="fck_mw_template"><span class="fck_mw_template">{{bug|706885}}</span></span> - 1 week
</li><li> Test against V8 Earley-Boyer benchmark.
<ul><li> Also v8 deltablue, raytrace
</li></ul>
</li><li> We may need to Implement Bill’s Pools/Zones idea to get the nursery to the requisite perf <span class="fck_mw_template"><span class="fck_mw_template">{{bug|759585}}</span></span> - 6 weeks - done
</li><li> Refactor code to avoid rooting on hot paths and keep rooter overhead acceptable <span class="fck_mw_template"><span class="fck_mw_template">{{bug|831886}}</span></span> (bhackett) - done
</li></ul>
 
=== Make the Post Barrier Verifier Green in the browser <span class="fck_mw_template"><span class="fck_mw_template">{{bug|764882}}</span></span> ===
 
<ul><li> Investigate how long it will take to do generational barriers - 1 week
</li><li> We may need to rewrite the maps in xpconnect and the browser in terms of HashTable: this could be a bunch of work.
</li><li> Make JS_IsAboutToBeFinalized indirect <span class="fck_mw_template"><span class="fck_mw_template">{{bug|765432}}</span></span> - 10 weeks
</li></ul>
 
== Other ==
 
<ul><li> Should we establish a new benchmark specifically for GGC. (sfink votes yes)
<ul><li> What would it measure? there are multiple goals, e.g. throughput/MMU/pause time
</li><li> Should we make it a public benchmark?
</li><li> What workloads should we consider?
</li><li> compartmental GC very important to us, not necessarily applicable to other implementations
</li><li> allocation rate (broken down into live vs garbage), steady behavior vs swapping between allocation + computation modes, etc.
</li></ul>
</li></ul>
<ul><li> How can we keep our advantage on Splay benchmark
<ul><li> possible to do with TI --- look at types of objects promoted from nursery to major heap, eventually start allocating them directly in the major heap
</li></ul>
</li></ul>
<ul><li> Compacting GC - Revisit in Febuary/March - 1 Month
<ul><li> Delayed until exact rooting is fully done. We can add this in if we have time in the schedule that we cannot parallelize.
</li></ul>
</li></ul>
 
== Risks ==
 
<ul>
<li> Team is responsible for many top crashers</li>
<li> External rooting API has not been designed</li>
<li> GGC algorithm has not been decided on - it may not be faster</li>
<li> JSD1 exact rooting is lurking</li>
</ul>
52

edits