| 8016bdee7ca1a066221a5d2fe5e60890de950a5b |
|
16-Jun-2016 |
Mathieu Chartier <mathieuc@google.com> |
Use collector specific helper classes
Changed to use inner classes. Also changed some visitors to lambdas.
Bug: 29413717
Bug: 19534862
(cherry picked from commit a07f55913824ab4215a9a4f827fa9c043c0d44d9)
Change-Id: I631c8bfe5f795eda4623c5bb4f357f2dd12358e2
|
| 56f9a895b938af96ffa1f5725904afd76fc47c73 |
|
07-Apr-2016 |
Mathieu Chartier <mathieuc@google.com> |
Fix regression in ModUnionTable RAM usage
We forgot to check if a reference changed before assigning it back,
this caused extra dirty pages in the image.
BusinessCard .art PSS
Before: 1678 KB
After: 991 KB
System wide .art PSS goes down ~20MB on my shamu.
Bug: 27906566
(cherry picked from commit 014885a8561f2415cb19ebde1e4084b88d63b746)
Change-Id: I64d149f10f3ef1ed1cb61810282559b47823014c
|
| 014885a8561f2415cb19ebde1e4084b88d63b746 |
|
07-Apr-2016 |
Mathieu Chartier <mathieuc@google.com> |
Fix regression in ModUnionTable RAM usage
We forgot to check if a reference changed before assigning it back,
this caused extra dirty pages in the image.
BusinessCard .art PSS
Before: 1678 KB
After: 991 KB
System wide .art PSS goes down ~20MB on my shamu.
Bug: 27906566
Change-Id: I6e398da79f0838960e4cb6650b24b1e6f60784ae
|
| a9d82fe8bc6960b565245b920e99107a824ca515 |
|
26-Jan-2016 |
Mathieu Chartier <mathieuc@google.com> |
Use mutator lock to guard adding and removing heap spaces
Too hard to add a new lock since dlmalloc ArtMoreCore requires
looping through the spaces while holding the allocator lock.
Bug: 22858531
Change-Id: Ieac2136da02c766b6795cd604a58798bee37ef2a
|
| 763a31ed7a2bfad22a9cb07f5301a71c0f97ca49 |
|
17-Nov-2015 |
Mathieu Chartier <mathieuc@google.com> |
Add immune spaces abstraction
ImmuneSpaces is a set of spaces which are not reclaimable by the GC in
the current collection. This set of spaces does not have requirements
about space adjacency like the old ImmuneRegion. ImmuneSpaces generates
the largest immune region for the GC. Since there is no requirement on
adjacency, it is possible to have multiple non-adjacent applicaton
image files.
For image spaces, we also look at the oat code which is normally after
the application image. In this case, we add the code as part of the
immune region. This is required to have both the boot image and the
zygote space be in the same immune region (for performance reasons).
Bug: 22858531
Change-Id: I5103b31c0e39ad63c594f5557fc848a3b288b43e
|
| 1ac1c2bba3ff0835cbae4753621f38dc935ebd91 |
|
22-Sep-2015 |
Mathieu Chartier <mathieuc@google.com> |
Some clean up for mark sweep files
Bug: 19534862
Change-Id: Ib619e8c4b9d902269d123baa18be56c94114e070
|
| 52a7f5caebdf359ab877f1928aad59f1e9ad29fa |
|
19-Aug-2015 |
Mathieu Chartier <mathieuc@google.com> |
Add class flags to class to help GC scanning
Reduces GC time and pauses by reducing the number of loads required
to scan an object.
Average total GC time before on EvaluateAndApplyChanges (EAAC): 7.452s
After: 7.144s
Average GC pause times before on EAAC: 860.67us
After: 722.75us
Adding the class flags field cause a memory increase of ~24k system
wide on low memory devices.
Change-Id: I3f04212d5787bfbf5e55026584d149f55476105e
|
| 3887c468d731420e929e6ad3acf190d5431e94fc |
|
12-Aug-2015 |
Roland Levillain <rpl@google.com> |
Remove unnecessary `explicit` qualifiers on constructors.
Change-Id: Id12e392ad50f66a6e2251a68662b7959315dc567
|
| da7c650022a974be10e2f00fa07d5109e3d8826f |
|
24-Jul-2015 |
Mathieu Chartier <mathieuc@google.com> |
Visit class native roots from VisitReferences
Visit class roots when we call Class::VisitReferences instead of in
the class linker. This makes it easier to implement class unloading
since unmarked classes won't have their roots visited by the class
linker.
Bug: 22181835
Change-Id: I63f31e5ebef7b2a0b764b3ba3cb038b3f561b379
|
| 90443477f9a0061581c420775ce3b7eeae7468bc |
|
17-Jul-2015 |
Mathieu Chartier <mathieuc@google.com> |
Move to newer clang annotations
Also enable -Wthread-safety-negative.
Changes:
Switch to capabilities and negative capabilities.
Future work:
Use capabilities to implement uninterruptible annotations to work
with AssertNoThreadSuspension.
Bug: 20072211
Change-Id: I42fcbe0300d98a831c89d1eff3ecd5a7e99ebf33
|
| e48a169605d8dc2b9af5ebe7c9c6af25f3b65dfe |
|
16-Jul-2015 |
Mathieu Chartier <mathieuc@google.com> |
Move MarkObject to mark_sweep.cc
Maybe fixes mac build.
Change-Id: I7c018922be8e159001498d30682435643bed1dfd
|
| 8118781ebc9659f806716c451bdb3fe9b77ae32b |
|
15-Jul-2015 |
Mathieu Chartier <mathieuc@google.com> |
Address some GC comments
Follow-up from:
https://android-review.googlesource.com/#/c/159650/
Change-Id: Id14f29b4ce5b70b63fcb3e74f8503ae60a3ae444
|
| 97509954404d031594b2ecbda607314d169d512e |
|
13-Jul-2015 |
Mathieu Chartier <mathieuc@google.com> |
Clean up GC callbacks to be virtual methods
Change-Id: Ia08034a4e5931c4fcb329c3bd3c4b1f301135735
|
| 375a29a824207fb2187bc3d31bfaae2d8283b3a2 |
|
14-May-2015 |
Hiroshi Yamauchi <yamauchi@google.com> |
Print field info on mark sweep invalid ref crash.
(cherry pick commit eb2baaf20d9059c0fc38141780ec05bea0486c40)
Bug: 20557050
Change-Id: I3abb7a44e3b7ed3256653824fd1705aac5f7d72c
|
| eb2baaf20d9059c0fc38141780ec05bea0486c40 |
|
14-May-2015 |
Hiroshi Yamauchi <yamauchi@google.com> |
Print field info on mark sweep invalid ref crash.
Bug: 20557050
Change-Id: I3abb7a44e3b7ed3256653824fd1705aac5f7d72c
|
| 414369a2e3f23e1408fc1cbf4f623014bd95cb8f |
|
04-May-2015 |
Mathieu Chartier <mathieuc@google.com> |
Add some more DISALLOW_COPY_AND_ASSIGN
May help prevent bugs maybe.
(cherry picked from commit 3130cdf29eb203be0c38d1107a65d920ec39c106)
Change-Id: Ie73d469dfcd078492ecb3aa28682b42707221202
|
| 3130cdf29eb203be0c38d1107a65d920ec39c106 |
|
04-May-2015 |
Mathieu Chartier <mathieuc@google.com> |
Add some more DISALLOW_COPY_AND_ASSIGN
May help prevent bugs maybe.
Change-Id: Ie73d469dfcd078492ecb3aa28682b42707221202
|
| 9086b65b2ad35dd39a8afc62d535be8217208d08 |
|
14-Apr-2015 |
Mathieu Chartier <mathieuc@google.com> |
Fix valgrind tests
Delete large objects in space destructor. Also some cleanup.
Change-Id: I4c4e90149841a156b7a3236201b37683e14890fb
|
| bb87e0f1a52de656bc77cb01cb887e51a0e5198b |
|
03-Apr-2015 |
Mathieu Chartier <mathieuc@google.com> |
Refactor and improve GC root handling
Changed GcRoot to use compressed references. Changed root visiting to
use virtual functions instead of function pointers. Changed root visting
interface to be an array of roots instead of a single root at a time.
Added buffered root marking helper to avoid dispatch overhead.
Root marking seems a bit faster on EvaluateAndApplyChanges due to batch
marking. Pause times unaffected.
Mips64 is untested but might work, maybe.
Before:
MarkConcurrentRoots: Sum: 67.678ms 99% C.I. 2us-664.999us Avg: 161.138us Max: 671us
After:
MarkConcurrentRoots: Sum: 54.806ms 99% C.I. 2us-499.986us Avg: 136.333us Max: 602us
Bug: 19264997
Change-Id: I0a71ebb5928f205b9b3f7945b25db6489d5657ca
|
| cb535da36915f9d10bec3880b46f1de1f7a69f22 |
|
23-Jan-2015 |
Mathieu Chartier <mathieuc@google.com> |
Change AtomicStack to use StackReference
Previously used Object*, using StackReference saves memory on 64 bit
devices.
Bug: 12935052
Bug: 17643507
Change-Id: I035878690054eeeb24d655a900b8f26c837703ff
|
| e34fa1df67fbe0173b4ea9abddcc3ae3d0537037 |
|
14-Jan-2015 |
Mathieu Chartier <mathieuc@google.com> |
Print more info in MarkSweep::VerifyRoot
Refactored old root callback to use a new class called RootInfo.
RootInfo contains all the relevant info related to the root
associated with the callback. The MarkSweep::VerifyRoot function
now uses this info to print the StackVisitor's described location
if the GC root is of the type kRootJavaFrame.
Some other cleanup.
Example output:
E/art (12167): Tried to mark 0x123 not contained by any spaces
E/art (12167): Attempting see if it's a bad root
E/art (12167): Found invalid root: 0x123 with type RootJavaFrame
E/art (12167): Location=Visiting method
'void java.lang.Runtime.gc()' at dex PC 0xffffffff (native PC 0x0)
vreg=0
(cherry picked from commit 12f7423a2bb4bfab76700d84eb6d4338d211983a)
Bug: 18588862
Change-Id: Ic5a2781f704e931265ffb3621c2eab4b2e25f60f
|
| 12f7423a2bb4bfab76700d84eb6d4338d211983a |
|
14-Jan-2015 |
Mathieu Chartier <mathieuc@google.com> |
Print more info in MarkSweep::VerifyRoot
Refactored old root callback to use a new class called RootInfo.
RootInfo contains all the relevant info related to the root
associated with the callback. The MarkSweep::VerifyRoot function
now uses this info to print the StackVisitor's described location
if the GC root is of the type kRootJavaFrame.
Some other cleanup.
Example output:
E/art (12167): Tried to mark 0x123 not contained by any spaces
E/art (12167): Attempting see if it's a bad root
E/art (12167): Found invalid root: 0x123 with type RootJavaFrame
E/art (12167): Location=Visiting method
'void java.lang.Runtime.gc()' at dex PC 0xffffffff (native PC 0x0)
vreg=0
Bug: 18588862
Change-Id: Ic5a2781f704e931265ffb3621c2eab4b2e25f60f
|
| 13735955f39b3b304c37d2b2840663c131262c18 |
|
08-Oct-2014 |
Ian Rogers <irogers@google.com> |
stdint types all the way!
Change-Id: I4e4ef3a2002fc59ebd9097087f150eaf3f2a7e08
|
| 52e4b43d62896b56f8c2bd041e528472bb4a0d8d |
|
10-Jun-2014 |
Mathieu Chartier <mathieuc@google.com> |
Add mark compact collector.
The mark compact collector is a 4 phase collection, doing a normal
full mark_sweep, calculating forwarding addresses of objects in the
from space, updating references of objects in the from space, and
moving the objects in the from space.
Support is diabled by default since it needs to have non movable
classes and field arrays. Performance numbers is around 50% as fast.
The main advantage that this has over semispace is that the worst
case memory usage is 50% since we only need one space isntead of two.
TODO: Make field arrays and classes movable. This causes complication
since Object::VisitReferences relies on these, so if we update the
fields of an object but another future object uses this object to
figure out what fields are reference fields it doesn't work.
Bug: 14059466
Change-Id: I661ed3b71ad4dde124ef80312c95696b4a5665a1
|
| 308351ada0008b0cbe1a5afc31c302c975554ee4 |
|
15-Jun-2014 |
Mathieu Chartier <mathieuc@google.com> |
Change reference processing to use heap references.
Removes several SetReferents for updating moved referents. Cleaned
up other aspects of the code.
Change-Id: Ibcb4d713fadea617efee7e936352ddf77ff4c370
|
| ffddfdf6fec0b9d98a692e27242eecb15af5ead2 |
|
03-Jun-2014 |
Tim Murray <timmurray@google.com> |
DO NOT MERGE
Merge ART from AOSP to lmp-preview-dev.
Change-Id: I0f578733a4b8756fd780d4a052ad69b746f687a9
|
| bbdc5bc5fd5141711879a6c85d80ac45b7aad5d0 |
|
28-May-2014 |
Hiroshi Yamauchi <yamauchi@google.com> |
Use a heap allocated free buffer in MarkSweep::SweepArray().
This is for avoiding a large stack frame size.
Bug: 15278350
Change-Id: I63e92c2dd0ad7078fd74ee2392717dc44b431e9a
|
| 3e5cf305db800b2989ad57b7cde8fb3cc9fa1b9e |
|
21-May-2014 |
Ian Rogers <irogers@google.com> |
Begin migration of art::Atomic to std::atomic.
Change-Id: I4858d9cbed95e5ca560956b9dabd976cebe68333
|
| 700a402244a1a423da4f3ba8032459f4b65fa18f |
|
20-May-2014 |
Ian Rogers <irogers@google.com> |
Now we have a proper C++ library, use std::unique_ptr.
Also remove the Android.libcxx.mk and other bits of stlport compatibility
mechanics.
Change-Id: Icdf7188ba3c79cdf5617672c1cfd0a68ae596a61
|
| 507dfdd147c97bfbadebfd63584d094b6a4e7b47 |
|
16-May-2014 |
Ian Rogers <irogers@google.com> |
Compatibility layer to transition from UniquePtr to std::unique_ptr.
Use ART_WITH_STLPORT (enabled for the target) to cause the use of UniquePtr,
for the host switch to std::unique_ptr. For now the type remains called
UniquePtr.
Make dalvik compile with clang on the host, move its build to C++11.
Change-Id: I5ba8d2757904bc089ed62047ea03de3c0853fb12
|
| eb8167a4f4d27fce0530f6724ab8032610cd146b |
|
08-May-2014 |
Mathieu Chartier <mathieuc@google.com> |
Add Handle/HandleScope and delete SirtRef.
Delete SirtRef and replaced it with Handle. Handles are value types
which wrap around StackReference*.
Renamed StackIndirectReferenceTable to HandleScope.
Added a scoped handle wrapper which wraps around an Object** and
restores it in its destructor.
Renamed Handle::get -> Get.
Bug: 8473721
Change-Id: Idbfebd4f35af629f0f43931b7c5184b334822c7a
|
| 78f7b4c04ab6e8b5581921bc95b67a9beee1c246 |
|
06-May-2014 |
Mathieu Chartier <mathieuc@google.com> |
Add concurrent reference processing.
Concurrent reference processing currently works by going into native
code from java.lang.ref.Reference.get(). From there, we have a fast
path if the references aren't being processed which returns the
referent without needing to access any locks. In the slow path we
block until reference processing is complete. It may be possible to
improve the slow path if the referent is blackened.
TODO: Investigate doing the fast path in java code by using racy reads
of a static volatile boolean. This will work as long as there are no
suspend points inbetween the boolean read and referent read.
Bug: 14381653
Change-Id: I1546b55be4691fe4ff4aa6d857b234cce7187d87
|
| b0fa5dc7769c1e054032f39de0a3f6d6dd06f8cf |
|
29-Apr-2014 |
Ian Rogers <irogers@google.com> |
Force inlining on trivial accessors.
Make volatility for GetFieldObject a template parameter.
Move some trivial mirror::String routines to a -inl.h.
Bug: 14285442
Change-Id: Ie23b11d4f18cb15a62c3bbb42837a8aaf6b68f92
|
| 6f365cc033654a5a3b45eaa1379d4b5f156b0cee |
|
23-Apr-2014 |
Mathieu Chartier <mathieuc@google.com> |
Enable concurrent sweeping for non-concurrent GC.
Refactored the GarbageCollector to let all of the phases be run by
the collector's RunPhases virtual method. This lets the GC decide
which phases should be concurrent and reduces how much baked in GC
logic resides in GarbageCollector.
Enabled concurrent sweeping in the semi space and non concurrent
mark sweep GCs. Changed the semi-space collector to have a swap semi
spaces boolean which can be changed with a setter.
Fixed tests to pass with GSS collector, there was an error related to
the large object space limit.
Before (EvaluateAndApplyChanges):
GSS paused GC time 7.81s/7.81s, score: 3920
After (EvaluateAndApplyChanges):
GSS paused GC time 6.94s/7.71s, score: 3900
Benchmark score doesn't go up since the GC happens in the allocating
thread. There is a slight reduction in pause times experienced by
other threads (0.8s total).
Added options for pre sweeping GC heap verification and pre sweeping
rosalloc verification.
Bug: 14226004
Bug: 14250892
Bug: 14386356
Change-Id: Ib557d0590c1ed82a639d0f0281ba67cf8cae938c
|
| bbd695c71e0bf518f582e84524e1cdeb3de3896c |
|
16-Apr-2014 |
Mathieu Chartier <mathieuc@google.com> |
Replace ObjectSet with LargeObjectBitmap.
Speeds up large object marking since large objects no longer required
a lock. Changed the GCs to use the heap bitmap for marking objects
which aren't in the fast path. This eliminates the need for a
MarkLargeObject function.
Maps before (10 GC iterations):
Mean partial time: 180ms
Mean sticky time: 151ms
Maps after:
Mean partial time: 161ms
Mean sticky time: 101ms
Note: the GC durations are long due to recent ergonomic changes and
because the fast bulk free hasn't yet been enabled. Over 50% of the
GC time is spent in RosAllocSpace::FreeList.
Bug: 13571028
Change-Id: Id8f94718aeaa13052672ccbae1e8edf77d653f62
|
| a8e8f9c0a8e259a807d7b99a148d14104c24209d |
|
09-Apr-2014 |
Mathieu Chartier <mathieuc@google.com> |
Refactor space bitmap to support different alignments.
Required for:
Using space bitmaps instead of std::set in mod union table +
remembered set.
Using a bitmap instead of set for large object marking.
Bug: 13571028
Change-Id: Id024e9563d4ca4278f79607cdb2f81895121b113
|
| 7bf9f190cd33a7e2f8584299eb889e9df66e0323 |
|
04-Apr-2014 |
Mathieu Chartier <mathieuc@google.com> |
Improve invalid root dumping.
The invalid root dumping now attempts to print the root type.
Change-Id: Ie821296d569f34909ba6e2705f5c347cd2143a3a
|
| 0f7bf6a3ad1798fde328a2bff48a4bf2d750a36b |
|
28-Mar-2014 |
Mathieu Chartier <mathieuc@google.com> |
Swap allocation stacks in pause.
This enables us to collect objects allocated during the GC for both
sticky, partial, and full GC. This also significantly simplifies GC
code. No measured performance impact on benchmarks, but this should
slightly increase sticky GC throughput.
Changed RevokeRosAllocThreadLocalBuffers to happen at most once per
GC. Previously it occured twice if pre-cleaning was enabled.
Renamed HandleDirtyObjectsPhase to PausePhase and enabled it for
non-concurrent GC. This helps reduce duplicated code which was in
both HandleDirtyObjectsPhase for concurrent GC and ReclaimPhase for
non-concurrent GC.
Change-Id: I533414b5c2cd2800f00724418e0ff90e7fdb0252
|
| 4aeec176eaf11fe03f342aadcbb79142230270ed |
|
28-Mar-2014 |
Mathieu Chartier <mathieuc@google.com> |
Refactor some GC code.
Reduced amount of code in mark sweep / semi space by moving
common logic to garbage_collector.cc. Cleaned up mod union tables
and deleted an unused implementation.
Change-Id: I4bcc6ba41afd96d230cfbaf4d6636f37c52e37ea
|
| 407f702da4f867c074fc3c8c688b8f8c32279eff |
|
18-Feb-2014 |
Mathieu Chartier <mathieuc@google.com> |
Refactor object reference visiting logic.
Refactored the reference visiting logic to be in mirror::Object
instead of MarkSweep.
Change-Id: I773249478dc463d83b465e85c2402320488577c0
|
| 0e54cd0d8fc635d3dc8bf88a465fdade151a098f |
|
20-Mar-2014 |
Mathieu Chartier <mathieuc@google.com> |
Refactor and optimize GC code.
Fixed the reference cache mod union table, and re-enabled it by
default. Added a boolean flag to count how many null objects,
immune, fast path, slow path objects we marked.
Slight speedup in mark stack processing, large speedup in image mod
union table scanning.
EvaluateAndApplyChanges Before:
Process mark stack time for full GC only:
12.464089s, 12.357870s, 12.538028s
Time spent marking mod image union table ~240ms.
After:
Process mark stack time: 12.299375s, 12.217142s, 12.187076s
Time spent marking mod image union table ~40ms.
TODO: Refactor reference visiting logic into mirror::Object.
Change-Id: I91889ded9d3f2bf127bc0051c1b1ff77e792e94f
|
| c93c530efc175954160c3834c93961a1a946a35a |
|
21-Mar-2014 |
Hiroshi Yamauchi <yamauchi@google.com> |
Revoke rosalloc thread-local buffers at the checkpoint.
In the mark sweep collector, rosalloc thread-local buffers were
revoked during the pause. Now, they are revoked at the thread
checkpoint, as opposed to during the pause, which appears to help
reduce the pause time.
In Ritz MemAllocTest, the average sticky pause time went down ~20%
(925 us -> 724 us).
Bug: 13394464
Bug: 9986565
Change-Id: I104992a11b46d59264c0b9aa2db82b1ccf2826bc
|
| 601276abdb746b03675ff945745aa936694d3439 |
|
20-Mar-2014 |
Mathieu Chartier <mathieuc@google.com> |
Fix RecordFree to take signed parameters.
RecordFree can get negative bytes allocated when background
compaction foreground transitions occur. This caused a DCHECK to
fail on debug builds. Also did some refactoring in
PreProcessReferences.
Bug: 13568814
Change-Id: I57543f1c78544a94f1d241459698b736dba8cfa8
|
| 3e41780cb3bcade3b724908e00443a9caf6977ef |
|
20-Mar-2014 |
Hiroshi Yamauchi <yamauchi@google.com> |
Refactor the garbage collector driver (GarbageCollector::Run).
Bug: 12687968
Change-Id: Ifc9ee86249f7938f51495ea1498cf0f7853a27e8
|
| 1ad2784ad9f311ebf9fe0677d33818648f423f9c |
|
20-Mar-2014 |
Mathieu Chartier <mathieuc@google.com> |
Add soft reference pre processing.
Soft reference pre-processing does soft reference preservation with
mutators running. After this is done, it does another pass with
mutators paused in the ProcessReference code. This helps lower pauses
since most preserved soft references have their referents recursive
marked outside the pause.
Changed ergonomics to have non sticky collectors always clear the
soft references.
Maps pauses ~10ms -> ~3ms on Nexus 4.
Bug: 13421927
Change-Id: I1370f7bb6934034869aa5afca0c377876267aa8e
|
| 8d562103c3a3452fb15ef4b1c64df767b70507a4 |
|
13-Mar-2014 |
Mathieu Chartier <mathieuc@google.com> |
Refactor immune region logic into its own file.
Added immune_region.cc/.h in the collector directory. Changed the
functionality to no longer assume spaces are added to immune region
in ascending order.
Change-Id: Id1d643b3849ad2695e8a151dbbb74a5035644472
|
| 719d1a33f6569864f529e5a3fff59e7bca97aad0 |
|
06-Mar-2014 |
Ian Rogers <irogers@google.com> |
Enable annotalysis on clang ART builds.
Fix clang build errors aswell as restructure locking/mutex code for correct
thread safety analysis support.
Reorder make dependencies so that host builds build first as they should
provide better compilation errors than target.
Remove host's use of -fno-omit-frame-pointer as it has no value with correct
use of CFI, which we should have.
Change-Id: I72cea8da9a3757b1a0b3acb4081feccb7c6cef90
|
| 893263b7d5bc2ca43a91ecb8071867f5134fc60a |
|
04-Mar-2014 |
Mathieu Chartier <mathieuc@google.com> |
Avoid marking old class linker and intern table roots during pause.
The new root visiting logic has a concept of a root log which holds
new roots which were added since the start of the GC. This is an
optimization since it lets us only mark these newly added roots
during the pause (or pre-cleaning) since the other roots intern table
and class linker roots were marked concurrently at the start of the
GC.
Before (EvaluateAndApplyChanges):
MarkConcurrentRoots: Sum: 605.193ms
After:
MarkConcurrentRoots: Sum: 271.858ms
This should also reduce pathological GC pauses which used to be able
to happen when the intern table or class linker became "dirty"
during the concurrent GC.
Change-Id: I433fab021f2c339d50c35aaae7161a50a0901dec
|
| 6fac447555dc94a935b78198479cce645c837b89 |
|
26-Feb-2014 |
Ian Rogers <irogers@google.com> |
Make allocations report usable size.
Work-in-progress to allow arrays to fill usable size. Bug: 13028925.
Use C++11's override keyword on GCC >= 2.7 to ensure that we override GC and
allocator methods.
Move initial mirror::Class set up into a Functor so that all allocated objects
have non-zero sizes. Use this property to assert that all objects are never
larger than their usable size.
Other bits of GC related clean-up, missing initialization, missing use of
const, hot methods in .cc files, "unimplemented" functions that fail at
runtime in header files, reducing header file includes, move valgrind's space
into its own files, reduce number of array allocation routines.
Change-Id: Id5760041a2d7f94dcaf17ec760f6095ec75dadaa
|
| c22c59ef8513b4cbbfd25073d1afbf58196b522a |
|
25-Feb-2014 |
Mathieu Chartier <mathieuc@google.com> |
Remove started runtime check in RevokeAllThreadLocalAllocationStacks
This check occasionally caused some thread local allocation stacks
to incorrectly not get revoked when multiple threads were allocating
without a started runtime. This showed up in image_test with
compaction enabled when we were initializing classes in the compiler
driver.
Change-Id: I7f28d072feea333c2503e35265ba25c51a6308fe
|
| dda54f59271464b5a72bf4cde6d9010e8dc1f337 |
|
24-Feb-2014 |
Mathieu Chartier <mathieuc@google.com> |
Card pre-cleaning.
We now pre-clean cards before the pause in the concurrent mark sweep
collectors. This provides substantial a pause time reduction for GC
iterations which have a lot of dirty cards. The only downside is a
slight GC time increase for large heaps.
Benchmark FormulaEvaluationActions.EvaluateAndApplyChanges:
Before:
Partial average pause: 5.47ms
Sticky average pause: 2.91ms
Total GC time: 25.8s
After:
Partial average pause: 1.98ms
Sticky average pause: 1.66ms
Total GC time: 27.0s
Benchmark score difference in the noise.
Change-Id: If9f01f8c1501f122e19432438108d48e723b332e
|
| 3bb57c7b41bf5419fe895e7aa664d8d430205ba8 |
|
18-Feb-2014 |
Mathieu Chartier <mathieuc@google.com> |
Change ProcessReferences to not use RecursiveMarkObject.
Calling ProcessMarkStack in RecursiveMarkObject caused a lot of
overhead due to timing logger splits. Changed the logic to be the
same as prior to the reference queue refactoring which involves
calling process mark stack after preserving soft references and
enqueueing finalizer references.
FinalizingGC longest pause is reduced by around 1/2 down to ~300ms.
Benchmark score ~400000 -> ~600000.
Also changed the timing logger splits in the GC to have (Paused) if
the split is a paused part of the GC.
Bug: 12129382
Change-Id: I7476d4f23670b19d70738e2fd48e37ec2f57e9f4
|
| 815873ecc312b1d231acce71e1a16f42cdaf09f2 |
|
14-Feb-2014 |
Mathieu Chartier <mathieuc@google.com> |
Change root visitor to use Object**.
Simplifies code and improves the performance of root visiting since
we usually don't need to check to see if the object moved.
Change-Id: Iba998f5a15ae1fa1b53ca5226dd2168a411196cf
|
| 83c8ee000d525017ead8753fce6bc1020249b96a |
|
28-Jan-2014 |
Mathieu Chartier <mathieuc@google.com> |
Add root types and thread id to root visiting.
Enables us to pass the root type and thread id to hprof.
Bug: 12680863
Change-Id: I6a0f1f9e3aa8f9b4033d695818ae7ca3460d67cb
|
| ef7d42fca18c16fbaf103822ad16f23246e2905d |
|
06-Jan-2014 |
Ian Rogers <irogers@google.com> |
Object model changes to support 64bit.
Modify mirror objects so that references between them use an ObjectReference
value type rather than an Object* so that functionality to compress larger
references can be captured in the ObjectRefererence implementation.
ObjectReferences are 32bit and all other aspects of object layout remain as
they are currently.
Expand fields in objects holding pointers so they can hold 64bit pointers. Its
expected the size of these will come down by improving where we hold compiler
meta-data.
Stub out x86_64 architecture specific runtime implementation.
Modify OutputStream so that reads and writes are of unsigned quantities.
Make the use of portable or quick code more explicit.
Templatize AtomicInteger to support more than just int32_t as a type.
Add missing, and fix issues relating to, missing annotalysis information on the
mutator lock.
Refactor and share implementations for array copy between System and uses
elsewhere in the runtime.
Fix numerous 64bit build issues.
Change-Id: I1a5694c251a42c9eff71084dfdd4b51fff716822
|
| a1602f28c0e3127ad511712d4b08db89737ae901 |
|
14-Jan-2014 |
Mathieu Chartier <mathieuc@google.com> |
Add zygote space as its own space type.
Helps prevent errors caused from doing invalid operations on the
old alloc space.
Removed some duplicated code in mark_sweep.cc and semi_space.cc.
Change-Id: I67a772cab30d698744c918aad581053f282a4a99
|
| ec05007f8619f8b0cc868d06731e07f84bb74c5b |
|
08-Jan-2014 |
Mathieu Chartier <mathieuc@google.com> |
Refactor sweeping logic into malloc space.
Removes duplicated code in MarkSweep/SemiSpace.
Deleted VerifyImageRoots since it had race conditions and is tested
by pre/post GC heap verification.
Change-Id: I9636359ff6adb3e93d56ce77a3e15299ed23dfd5
|
| c528dba35b5faece51ca658fc008b688f8b690ad |
|
26-Nov-2013 |
Mathieu Chartier <mathieuc@google.com> |
Enable moving classes.
Slight reduction in Zygote size, memory savings are in the noise.
Before: Zygote size: 8739224
After: Zygote size: 8733568
Fixed a bug where we didn't set the concurrent start bytes after
switching the allocator from bump pointer to ROSAlloc in the
zygote. This caused excessive memory usage.
Added the method verifiers as roots to fix an issue caused by
RegTypes holding a Class*.
Added logic to clear card table in the SemiSpace collector, this
reduces DalvikOther from ~2400k -> ~1760k when using the SemiSpace
collector.
Added a missing lock to the timing loggers which caused a rare
one time crash in std::set.
Bug: 11771255
Bug: 8499494
Bug: 10802951
Change-Id: I99d2b528cd51c1c5ed7012e3220b3aefded680ae
|
| 39e3261168e7761fea6d873494d7c5d191285791 |
|
13-Nov-2013 |
Mathieu Chartier <mathieuc@google.com> |
Refactor reference queues.
Refactored the reference queue processing to reside in the heap code.
This removes significant code duplication in the semispace and
marksweep garbage collectors.
Changed the soft reference behaviour to preserve all soft references
unless the GC requires them to be cleared to avoid an out of memory
error. It may be worth investigating a better heuristic in the
future to preserve soft references by LRU order.
Change-Id: I1f3ff5bd4b3c5149271f4bb4fc94ba199e2f9bc2
|
| 590fee9e8972f872301c2d16a575d579ee564bee |
|
13-Sep-2013 |
Mathieu Chartier <mathieuc@google.com> |
Compacting collector.
The compacting collector is currently similar to semispace. It works by
copying objects back and forth between two bump pointer spaces. There
are types of objects which are "non-movable" due to current runtime
limitations. These are Classes, Methods, and Fields.
Bump pointer spaces are a new type of continuous alloc space which have
no lock in the allocation code path. When you allocate from these it uses
atomic operations to increase an index. Traversing the objects in the bump
pointer space relies on Object::SizeOf matching the allocated size exactly.
Runtime changes:
JNI::GetArrayElements returns copies objects if you attempt to get the
backing data of a movable array. For GetArrayElementsCritical, we return
direct backing storage for any types of arrays, but temporarily disable
the GC until the critical region is completed.
Added a new runtime call called VisitObjects, this is used in place of
the old pattern which was flushing the allocation stack and walking
the bitmaps.
Changed image writer to be compaction safe and use object monitor word
for forwarding addresses.
Added a bunch of added SIRTs to ClassLinker, MethodLinker, etc..
TODO: Enable switching allocators, compacting on background, etc..
Bug: 8981901
Change-Id: I3c886fd322a6eef2b99388d19a765042ec26ab99
|
| 0941b0423537a6a5d7c1df6dd23e9864ea8f319c |
|
05-Nov-2013 |
Hiroshi Yamauchi <yamauchi@google.com> |
Fix a DCHECK failure due to unmatching numbers of cards scanned.
- See the bug for details of the failure.
- After a discussion, we decided to get rid of the DCHECK as a simple
solution would not detect corner failure cases and a full solution
would add undesired complexity, and left a comment that explains
what situation had caused a DCHECK failure.
- Fix a potential error of failing to scan the last card that the end
of the image space falls on as a result of the image end being not
necessarily aligned by the card size.
- Remove dead/unused MarkSweep::ScanRoot().
- Add AlignUp and AlignDown for aligning pointers.
Bug: 11465268
Change-Id: Iee3018a42c48a159feb0e9cf77b1a6b303f5d245
|
| ad2541a59c00c2c69e8973088891a2b5257c9780 |
|
25-Oct-2013 |
Mathieu Chartier <mathieuc@google.com> |
Fix object identity hash.
The object identity hash is now stored in the monitor word after
being computed. Hashes are computed by a pseudo random number
generator.
When we write the image, we eagerly compute object hashes to
prevent pages getting dirtied.
Bug: 8981901
Change-Id: Ic8edacbacb0afc7055fd740a52444929f88ed564
|
| 11409ae81a3eaf84d7fd2b3c85b8b06d2bae27f0 |
|
23-Sep-2013 |
Mathieu Chartier <mathieuc@google.com> |
Refactor and improve mod-union tables.
Allow support for adding more mod union tables, reduces the amount
of baked in logic. Adds support for updating mod union table references
from compaction (not for ReferenceCache table yet).
Change-Id: I1beeda00839ed86ef0e853beff5ce10d0ab2b9d1
|
| c11d9b8870de5f860b13c84003ade7b3f3125a52 |
|
19-Sep-2013 |
Mathieu Chartier <mathieuc@google.com> |
Re-enable concurrent system weak sweeping.
Enabled by disallowing new system weaks during the pause and
re-allowing it after the system weaks have been swept. Reduces
GC pause by ~1ms.
Fixes pause regression caused by fix for
Bug: 10626133
Change-Id: If49d33e7ef19cb728ed3cef5187acfa53b9b05d8
|
| 0f72e4136aecaf6976fdb55916bbd7b6d5c9c77b |
|
07-Sep-2013 |
Mathieu Chartier <mathieuc@google.com> |
Improve heap verification.
Re-enabled checking the allocation stack for heap verification.
Added tracking of recent frees in DlMallocSpace if debug spaces is
enabled. This is useful when you have heap corruption caused by a
live object referencing a recently freed object.
Added various other sanity checks in the GC.
Bug: 10626133
Change-Id: I5ada11966336ae9a06615b16f4b933f05b5d0c32
|
| 6aa3df965395566ed6a4fec4af37c2b7577992e9 |
|
18-Sep-2013 |
Mathieu Chartier <mathieuc@google.com> |
Refactor system weak sweeping, add support for modification.
Required for moving collectors.
Change-Id: Ib97ba4a05af1139f8d388077a15e62bcb9534855
|
| 423d2a3dcbb260b020efb5da59f784c9f02accbf |
|
13-Sep-2013 |
Mathieu Chartier <mathieuc@google.com> |
Add support for changing roots through the root visitor callback.
Needed for copying collectors.
Change-Id: Icc4a342a57e0cfb79587edb02ef8c85e08808877
|
| 958291c7afe723d846a39539fd00410c102485f3 |
|
28-Aug-2013 |
Mathieu Chartier <mathieuc@google.com> |
Fix build.
Build broken due to a rare lock checking violation happening during
oat generation. Added a new lock level for the mark stack lock since
it may be acquired with the globals/pins lock.
Change-Id: I944ae0700c66ad050fe3d23aba2f2c377c5df799
|
| ba311b4385fa3f382f01312a8cc97b52011232e3 |
|
27-Aug-2013 |
Mathieu Chartier <mathieuc@google.com> |
Fix mark stack expand race.
We now guard parallel mark stack pushing with a lock. This is
only used by checkpoint root marking. I did not observe a
significant slowdown by looking at ritzperf and maps, but it may
be worth reinvestigating in the future.
Also a bit of refactoring.
Bug: 10113123
Change-Id: Ifcb12d14df437e2aea9a1165a9568054f80d91b3
|
| 2775ee4f82dff260663ca16adddc0b15327aaa42 |
|
21-Aug-2013 |
Mathieu Chartier <mathieuc@google.com> |
Add more runtime options.
Changed HeapGCThreads to be split into two different options:
-XX:ParallelGCThreads: Which specifies how many threads the GC may
use when the mutators are suspended.
-XX:ConcGCThreads: Which specifies how many threads the GC may use
when the mutators are running.
Added runtime options to specify long pause / long GC thresholds:
-XX:LongPauseThreshold (default 5ms)
-XX:LongGCThreshold (default 100ms)
These thresholds were previously constants, but are now runtime
options. If we exceed either of the thresholds, we print the GC
message.
Added a new runtime option: -XX:IgnoreMaxFootprint which makes it
that the GC only does GC when the number of bytes allocated hits
the growth limit. This causes GC to occur much less frequently and
can be useful to measure how much of an impact GC has on performance.
Changed the GC behaviour to use only one thread when we do not care
about pauses to prevent jank that can be caused by 2 simultaneous GC
on different processes fighting for CPU time.
Added thread pool functionality for changing the maximum number of
active workers.
Fixed an accounting error where we didn't count large objects in the
total freed.
Bug: 9986416
Change-Id: I86afa358d93dcd3780e18ac5d85bdb1a130cb7e7
|
| e53225c7b8c98f8fc3855fc70f718e7f8abab307 |
|
19-Aug-2013 |
Mathieu Chartier <mathieuc@google.com> |
Print information about large objects freed and AllocSpace objects.
When we print information about the GC, we now print how many
AllocSpace and LOS objects were freed along with their corresponding
total sizes.
Fixed a disabled timing longer which I had accidentally done in
another CL.
Change-Id: I6cf958ec553bdbb365f242b3844461acec801078
|
| 94c32c5f01c7d44781317bf23933ed0a5bc4b796 |
|
09-Aug-2013 |
Mathieu Chartier <mathieuc@google.com> |
More parallel GC, rewritten parallel mark stack processing.
Card scanning may now be done in parallel. This speeds up sticky and
reduces pause times for all GC types.
Speedup on my mako (ritz perf):
Average pause time for sticky GC (~250 samples):
Without parallel cards scanning enabled: 2.524904215ms
Parallel card scanning (num_gc_threads_): 1.552123552ms
Throughput (~250 samples):
Sticky GC throughput with parallel card scanning: 69MB/s
Sticky GC throughput without parallel card scanning: 51MB/s
Rewrote the mark stack processing to be LIFO and use a prefetch queue
like the non parallel version.
Cleaned up some of the logcat printing for the activity manager
process state listening.
Added unlikely hints to object scanning since arrays and classes are
scanned much less often than normal objects.
Fixed a bug where the number of GC threads was clamped to 1 due to a
bool instead of a size_t.
Fixed a race condition when we added references to the reference
queues. Sharded the reference queue lock into one lock for each reference
type (weak, soft, phatom, finalizer).
Changed timing splits to be different for processing gray objects with
and without mutators paused since sticky GC does both.
Mask out the class bit when visiting fields as an optimization, this is
valid since classes are held live by the class linker.
Partially completed: Parallel recursive mark + finger.
Bug: 10245302
Bug: 9969166
Bug: 9986532
Bug: 9961698
Change-Id: I142d09718c4609b7c2387cb28f517a6983c73288
|
| 02e25119b15a6f619f17db99f5d05124a5807ff3 |
|
15-Aug-2013 |
Mathieu Chartier <mathieuc@google.com> |
Fix up TODO: c++0x, update cpplint.
Needed to update cpplint to handle const auto.
Fixed a few cpplint errors that were being missed before.
Replaced most of the TODO c++0x with ranged based loops. Loops which
do not have a descriptive container name have a concrete type instead
of auto.
Change-Id: Id7cc0f27030f56057c544e94277300b3f298c9c5
|
| 9642c96bd5a1ccc4e221de9c0af4a545af8182d2 |
|
06-Aug-2013 |
Mathieu Chartier <mathieuc@google.com> |
Move allocation stack unmarking outside of pause.
Reduces pause time by moving the allocation stack unmarking outside
of the pause. This is especially helpful for devices which have longer
GC times since these times result in having more things to be
unmarked in the allocation stack.
Bug: 9969166
Change-Id: I570f2213cbdda9d90545b64538e2cbeb0dc32d16
|
| 184e322fe8ddd75c844a1eb2eb1ca32bc02f2d45 |
|
03-Aug-2013 |
Mathieu Chartier <mathieuc@google.com> |
Disable and remove finger.
Finger is useless for multithreaded GC, removing it should provide a
slight speed up due to avoided comparison.
Change-Id: I7eb7abcbab8d3307807b1086c06d68b2d4bcd2e9
|
| 7934ac288acfb2552bb0b06ec1f61e5820d924a4 |
|
26-Jul-2013 |
Brian Carlstrom <bdc@google.com> |
Fix cpplint whitespace/comments issues
Change-Id: Iae286862c85fb8fd8901eae1204cd6d271d69496
|
| fc0e3219edc9a5bf81b166e82fd5db2796eb6a0d |
|
17-Jul-2013 |
Brian Carlstrom <bdc@google.com> |
Fix multiple inclusion guards to match new pathnames
Change-Id: Id7735be1d75bc315733b1773fba45c1deb8ace43
|
| 7940e44f4517de5e2634a7e07d58d0fb26160513 |
|
12-Jul-2013 |
Brian Carlstrom <bdc@google.com> |
Create separate Android.mk for main build targets
The runtime, compiler, dex2oat, and oatdump now are in seperate trees
to prevent dependency creep. They can now be individually built
without rebuilding the rest of the art projects. dalvikvm and jdwpspy
were already this way. Builds in the art directory should behave as
before, building everything including tests.
Change-Id: Ic6b1151e5ed0f823c3dd301afd2b13eb2d8feb81
|