OSDN Git Service
Nicolas Geoffray [Wed, 24 Feb 2016 17:18:25 +0000 (17:18 +0000)]
Forgot to initialize JIT memory use fields.
Change-Id: I82ae97adffd8e633de048f044cdbdb953b2f9763
Roland Levillain [Wed, 24 Feb 2016 16:05:37 +0000 (16:05 +0000)]
Merge "MIPS32: Implement intrinsics from java.lang.Math:"
Roland Levillain [Wed, 24 Feb 2016 15:46:01 +0000 (15:46 +0000)]
Merge "Remove unreachable code paths in constant folding."
Roland Levillain [Wed, 24 Feb 2016 15:45:01 +0000 (15:45 +0000)]
Merge "Fix indentation of copyright headers in some ART run-tests."
Roland Levillain [Wed, 24 Feb 2016 15:41:20 +0000 (15:41 +0000)]
Fix indentation of copyright headers in some ART run-tests.
Reported by Kevin Brodsky <kevin.brodsky@linaro.org>.
Change-Id: Iec3d69aa33d3b6a31f1298607e0ee6bc4cb6d9c0
Bill Buzbee [Wed, 24 Feb 2016 14:59:07 +0000 (14:59 +0000)]
Merge "ART: Enable JitProfiling for x86 Mterp"
Roland Levillain [Wed, 24 Feb 2016 14:54:18 +0000 (14:54 +0000)]
Remove unreachable code paths in constant folding.
Change-Id: I7ffb361711c87f6b1b98d172d2cfdf9b2ba65607
Nicolas Geoffray [Wed, 24 Feb 2016 12:14:35 +0000 (12:14 +0000)]
Merge "Add a script to symbolize buildbot crashes."
Nicolas Geoffray [Wed, 24 Feb 2016 12:04:23 +0000 (12:04 +0000)]
Add a script to symbolize buildbot crashes.
Change-Id: I6e85a2f37db84a8d06c7258de1c60eaffc78e44b
Vladimir Marko [Wed, 24 Feb 2016 11:52:14 +0000 (11:52 +0000)]
Merge "ART: Allow method references across oat files for multi-image, 2nd attempt."
Vladimir Marko [Fri, 19 Feb 2016 12:27:55 +0000 (12:27 +0000)]
ART: Allow method references across oat files for multi-image, 2nd attempt.
These were disabled because we didn't have sufficient
information about the multi-image layout when processing
link-time patches in OatWriter. This CL refactors the
ELF file creation so that the information is available.
Also clean up ImageWriter to use oat file indexes instead
of filenames and avoid reopening the oat file to retrieve
the checksum.
Change-Id: Icc7b528deca29da1e473c8f079521a36d6c4892f
Roland Levillain [Wed, 24 Feb 2016 10:53:24 +0000 (10:53 +0000)]
Merge "Please Clang with respect to stack frame limits in dex2oat."
Nicolas Geoffray [Wed, 24 Feb 2016 10:35:01 +0000 (10:35 +0000)]
Merge "Introduce partial code cache collection."
David Srbecky [Wed, 24 Feb 2016 10:30:04 +0000 (10:30 +0000)]
Merge "Associate slow paths with the instruction that they belong to."
David Srbecky [Tue, 9 Feb 2016 15:24:47 +0000 (15:24 +0000)]
Associate slow paths with the instruction that they belong to.
Almost all slow paths already know the instruction they belong to,
this CL just moves the knowledge to the base class as well.
This is needed to be be able to get the corresponding dex pc for
slow path, which allows us generate better native line numbers,
which in turn fixes some native debugging stepping issues.
Change-Id: I568dbe78a7cea6a43a4a71a014b3ad135782c270
David Srbecky [Wed, 24 Feb 2016 10:23:50 +0000 (10:23 +0000)]
Merge "Remove HNativeDebugInfo from start of basic blocks."
David Srbecky [Tue, 9 Feb 2016 14:30:11 +0000 (14:30 +0000)]
Remove HNativeDebugInfo from start of basic blocks.
We do not require full environment at the start of basic block.
The dex pc contained in basic block is sufficient for line mapping.
Change-Id: I5ba9e5f5acbc4a783ad544769f9a73bb33e2bafa
Nicolas Geoffray [Wed, 24 Feb 2016 10:12:30 +0000 (10:12 +0000)]
Merge "Show battery level fo devices on buildbot."
buzbee [Tue, 23 Feb 2016 21:25:00 +0000 (13:25 -0800)]
ART: Enable JitProfiling for x86 Mterp
Adds branch profiling and enables for x86.
Change-Id: I875034d5bc6b639df08a0236e415195521994238
Nicolas Geoffray [Wed, 24 Feb 2016 09:53:09 +0000 (09:53 +0000)]
Show battery level fo devices on buildbot.
Change-Id: Ic8279d1905c8a6efe08ca190f1bb7332842e1904
Alex Light [Tue, 23 Feb 2016 22:33:55 +0000 (22:33 +0000)]
Merge "Make JNI work correctly with default methods."
Alex Light [Mon, 22 Feb 2016 21:43:29 +0000 (13:43 -0800)]
Make JNI work correctly with default methods.
Also adds some tests for JNI and DefaultMethods.
Bug:
27259142
Bug:
24618811
Change-Id: I31222e3e41059d803be1dbb0f40e1144ac4bf457
Aart Bik [Tue, 23 Feb 2016 19:42:27 +0000 (19:42 +0000)]
Merge "Improved instruction + offset hunting."
Aart Bik [Tue, 23 Feb 2016 00:22:33 +0000 (16:22 -0800)]
Improved instruction + offset hunting.
Rationale:
This is generally useful for anything using this method
but in particular for deopting something like
bs[ off] = (byte)(n >>> 24);
bs[++off] = (byte)(n >>> 16);
bs[++off] = (byte)(n >>> 8);
bs[++off] = (byte)(n );
where the base + offset is hidden in the increments.
Occurs quite often in real-life code.
Change-Id: I3fa7d285a7368a179a26e590e8eee37f3b64c25d
Mathieu Chartier [Tue, 23 Feb 2016 18:49:08 +0000 (18:49 +0000)]
Merge "Fix allocation tracking race"
Mathieu Chartier [Tue, 23 Feb 2016 18:37:32 +0000 (10:37 -0800)]
Fix allocation tracking race
Need to instrument the stubs before calling
Heap::SetAlloctrackingEnabled or else we can get this DCHECK failing
DCHECK(!IsAllocTrackingEnabled())
Change-Id: I1fb5df217a2785baf22b4f57887cd323f13f0973
Nicolas Geoffray [Tue, 23 Feb 2016 18:17:59 +0000 (18:17 +0000)]
Merge "Revert "Compile checker run-tests with Jack""
Nicolas Geoffray [Tue, 23 Feb 2016 18:17:46 +0000 (18:17 +0000)]
Revert "Compile checker run-tests with Jack"
Tests still faill.
Bug:
25635944
Bug:
19467889
This reverts commit
fbeeb47fc63ec85891198b51b2309bb15f26527e.
Change-Id: I8f49dc6e8d3eb5f0de10636ef2685b7a2e08f297
Roland Levillain [Tue, 23 Feb 2016 18:08:53 +0000 (18:08 +0000)]
Please Clang with respect to stack frame limits in dex2oat.
This change enables Clang to compile dex2oat on MIPS64
without complaining about stack frames larger than what
`-Wframe-larger-than` allows.
Bug:
27310199
Change-Id: I441a4be499959a089d3c2eae1135eb0273c1e80b
Sebastien Hertz [Tue, 23 Feb 2016 16:56:50 +0000 (16:56 +0000)]
Merge "Compile checker run-tests with Jack"
Calin Juravle [Tue, 23 Feb 2016 16:54:48 +0000 (16:54 +0000)]
Merge "Fix profman path for the gtest."
Calin Juravle [Tue, 23 Feb 2016 16:53:30 +0000 (16:53 +0000)]
Fix profman path for the gtest.
Change-Id: Id332c50658b7dc1eb1edf6a58fcbdde4c2887d27
Sebastien Hertz [Fri, 19 Feb 2016 15:54:05 +0000 (16:54 +0100)]
Compile checker run-tests with Jack
Bug:
25635944
Bug:
19467889
Change-Id: I3c3fcbcbd2f6c142dbea279a9ecd9d372e79b3b9
Nicolas Geoffray [Tue, 23 Feb 2016 13:56:43 +0000 (13:56 +0000)]
Introduce partial code cache collection.
It will collect code that is known unused (because it deoptimized),
and osr code.
bug:
26846185
Change-Id: Ic27dfeb944efb2ca464039007ba365c1e0d4a040
Calin Juravle [Tue, 23 Feb 2016 14:07:40 +0000 (14:07 +0000)]
Merge "Revert "Revert "Add profman tool: responsible to process profiles"""
Roland Levillain [Tue, 23 Feb 2016 13:57:10 +0000 (13:57 +0000)]
Merge "Re-enable tests that were causing TimeoutExceptions on x86/x86-64."
Roland Levillain [Tue, 23 Feb 2016 13:49:35 +0000 (13:49 +0000)]
Re-enable tests that were causing TimeoutExceptions on x86/x86-64.
Some libcore tests have been failing with a
java.util.concurrent.TimeoutException on ART Builbot's
x86/x86-64 concurrent collector configurations. No longer
ignore these failures to see whether recent changes (namely
https://android-review.googlesource.com/#/c/204125) fixed
the issue.
Bug:
26507762
Change-Id: Ib188e194195fc11b539bd53a03db78ab80626f33
Nicolas Geoffray [Tue, 23 Feb 2016 13:53:11 +0000 (13:53 +0000)]
Merge "Don't allocate mspaces of less than a page."
Nicolas Geoffray [Tue, 23 Feb 2016 13:52:01 +0000 (13:52 +0000)]
Don't allocate mspaces of less than a page.
Fixes jit tests in debug mode.
Change-Id: I34565e03683cee063e26975a461d8e75ad4a205f
Calin Juravle [Tue, 23 Feb 2016 12:00:03 +0000 (12:00 +0000)]
Revert "Revert "Add profman tool: responsible to process profiles""
This reverts commit
3da74687e42de7d33a8e75df9bd64374e650f75e.
Change-Id: Id005096bd8063c6c602744d4476d5eb7e0d34e90
Vladimir Marko [Tue, 23 Feb 2016 10:52:23 +0000 (10:52 +0000)]
Merge "Dex-wide ArenaPool scoping for AOT compilation"
Nicolas Geoffray [Tue, 23 Feb 2016 10:52:11 +0000 (10:52 +0000)]
Merge "Add a new kReservedCapacity to decide when to start GC code."
Calin Juravle [Tue, 23 Feb 2016 10:41:25 +0000 (10:41 +0000)]
Merge "Fix static dex2oat build"
Delphine Martin [Tue, 23 Feb 2016 08:19:30 +0000 (08:19 +0000)]
Merge "Revert "Force Jack to consider multidex annotations""
Mathieu Chartier [Tue, 23 Feb 2016 01:21:13 +0000 (01:21 +0000)]
Merge "Add regression test for alloc stack trace and unloading"
Mathieu Chartier [Tue, 23 Feb 2016 00:56:54 +0000 (16:56 -0800)]
Add regression test for alloc stack trace and unloading
Test creates allocations in a different loader, then unloads it.
This would free the ArtMethod except that we now visit them as
roots.
Bug:
26849503
Change-Id: Ied9f5942ff9022b1cd39052363f119a630ffbaef
Mathieu Chartier [Tue, 23 Feb 2016 00:18:30 +0000 (00:18 +0000)]
Merge "Visit stack trace roots"
Mathieu Chartier [Mon, 22 Feb 2016 22:49:04 +0000 (14:49 -0800)]
Visit stack trace roots
We need to visit the declaring classes of all the methods in the
allocation stack traces to prevent class unloading for these
methods. If the class gets unloaded, it will free the linear alloc
resulting in hprof crashing during dumping.
Also a bit of clean up.
Bug:
26849503
Change-Id: I5cc57436294e4399388cb497819148b8a3b973d9
Nicolas Geoffray [Mon, 22 Feb 2016 22:38:27 +0000 (22:38 +0000)]
Merge "Revert "Add profman tool: responsible to process profiles""
Nicolas Geoffray [Mon, 22 Feb 2016 22:37:52 +0000 (22:37 +0000)]
Revert "Add profman tool: responsible to process profiles"
Needs a profile_assistant_test fix.
Bug:
26719109
Bug:
26563023
This reverts commit
6caefd983a800a063b219f1d3ed71b1416cecd70.
Change-Id: Ibdeb7385737dd7846ed861e0a95f083abb9aa974
Hiroshi Yamauchi [Mon, 22 Feb 2016 21:45:16 +0000 (21:45 +0000)]
Merge "Fix class_linker_test for valgrind."
Hiroshi Yamauchi [Mon, 22 Feb 2016 21:30:35 +0000 (13:30 -0800)]
Fix class_linker_test for valgrind.
Fix a memory leak.
Bug:
27156726
Change-Id: I4d9a2dc9cff447b6f2fd02432e018ae49766563a
Hiroshi Yamauchi [Mon, 22 Feb 2016 21:41:06 +0000 (21:41 +0000)]
Merge "Fix CC collector thread flip and JNI critical section deadlocks."
Andreas Gampe [Mon, 22 Feb 2016 20:50:35 +0000 (20:50 +0000)]
Merge "ART: Add Class.getEnclosingClass() to UnstartedRuntime"
Chris Larsen [Fri, 20 Nov 2015 17:07:09 +0000 (09:07 -0800)]
MIPS32: Implement intrinsics from java.lang.Math:
- abs(double) - abs(float) - abs(int)
- abs(long) - max(double, double) - max(float, float)
- max(int, int) - max(long, long) - min(double, double)
- min(float, float) - min(int, int) - min(long, long)
- sqrt(double)
The math intrinsics:
- ceil(double) - floor(double) - rint(double)
- round(double) - round(float)
aren't implemented because they require instructions which only exist
for MIPS64, or for MIPS32r6.
Change-Id: I943be3592b52a423fcb7ac40f46f38a5e2a58c50
Hiroshi Yamauchi [Fri, 19 Feb 2016 23:44:06 +0000 (15:44 -0800)]
Fix CC collector thread flip and JNI critical section deadlocks.
There's a bug in the synchronization between the CC collector's thread
flip and JNI critical sections where it incorrectly attempts to make
the synchronization scheme to be writer (GC) preference to avoid
starvation of GC in the presence of frequent JNI critical section
enter/exit. This could cause a deadlock between them if a thread
enters a nested JNI critical section after a thread flip occurs. This
is reproduced in the added test.
The fix is to use a thread local disable counter in addition to the
global counter to detect a nested enter by the same thread and avoid
waiting if nested.
Bug:
19235243
Bug:
12687968
Change-Id: Idf7720a6906c9ea508219935af3727f76680d2d8
Andreas Gampe [Fri, 19 Feb 2016 18:49:50 +0000 (10:49 -0800)]
ART: Add Class.getEnclosingClass() to UnstartedRuntime
Required to initialize classes calling getSimpleName().
Bug:
27265238
(cherry picked from commit
fb5c22e28030cc9ba8424b7f6f70677ec3ba3afc)
Change-Id: Id63dac5b75b3bbcdfe80f73f2d9d76dea3996823
Bill Buzbee [Mon, 22 Feb 2016 19:14:13 +0000 (19:14 +0000)]
Merge "ART: Enable JitProfiling for Arm Mterp"
Vladimir Marko [Mon, 22 Feb 2016 18:37:47 +0000 (18:37 +0000)]
Merge "Optimizing/x86-64: Use MOVL in Load64BitValue() if IsUint<32>()."
Calin Juravle [Mon, 22 Feb 2016 18:20:49 +0000 (18:20 +0000)]
Fix static dex2oat build
Change-Id: I1a5fb726fa72d9ccd083f90a818e245f075fc89c
Vladimir Marko [Mon, 22 Feb 2016 16:54:39 +0000 (16:54 +0000)]
Optimizing/x86-64: Use MOVL in Load64BitValue() if IsUint<32>().
Change-Id: Ie8bfb1861a384d0906f2aff9e8a94be0925c65b6
Hans Boehm [Mon, 22 Feb 2016 15:52:53 +0000 (15:52 +0000)]
Merge "Add Math.round tests for large integers."
Sebastien Hertz [Mon, 22 Feb 2016 15:43:30 +0000 (15:43 +0000)]
Merge "Revert "Workaround JDWP failures""
Delphine Martin [Mon, 22 Feb 2016 15:16:25 +0000 (15:16 +0000)]
Revert "Force Jack to consider multidex annotations"
This reverts commit
8cc049c2354af3964d2c15223a9c8d4f57a6691b.
The bug that necessitated that workaround was fixed.
Bug:
26441710
Change-Id: Idf238e497052cb983fd56f50a36f0f7998240fa8
Calin Juravle [Mon, 22 Feb 2016 15:05:03 +0000 (15:05 +0000)]
Merge "Add profman tool: responsible to process profiles"
Calin Juravle [Mon, 1 Feb 2016 12:06:18 +0000 (12:06 +0000)]
Add profman tool: responsible to process profiles
This pulls out profile parsing from dex2oat into a separate tool.
Some additional refactoring:
- better return codes
- dex2oat now accepts only one profile file
This is the first step towards support secondary dex files and
extracting profiles out of the code_cache directory.
Bug:
26719109
Bug:
26563023
(cherry picked from commit
d81c289e58f4ced8ec7674fd377c356669f97227)
Change-Id: I34d408faa318e866e1a23d4d04c369131ee5012b
Narayan Kamath [Mon, 22 Feb 2016 13:50:02 +0000 (13:50 +0000)]
Narayan Kamath [Mon, 22 Feb 2016 13:34:07 +0000 (13:34 +0000)]
Track libcore change
b5d045f7e4023c0ba9a42fb95efad4c30dfc085f
Change-Id: I40bfd7f85c3972cb21c1ddde07ae9b5727df8909
Nicolas Geoffray [Mon, 22 Feb 2016 13:14:04 +0000 (13:14 +0000)]
Add a new kReservedCapacity to decide when to start GC code.
Change-Id: Ib9d0ad9bc86c66ac29975f7f907f79f842342500
Sebastien Hertz [Mon, 22 Feb 2016 10:55:24 +0000 (10:55 +0000)]
Revert "Workaround JDWP failures"
This reverts commit
8af4d1c84682a3108b1e5b66fd7442a69eb715c0.
Bug:
27262745
Change-Id: I4254fad87cfdd2f9c044722c2f5766c8b14aaa9c
Roland Levillain [Mon, 22 Feb 2016 10:45:59 +0000 (10:45 +0000)]
Merge "Ignore GZIPInputStreamTest#testLongMessage failures with CC."
Roland Levillain [Mon, 22 Feb 2016 10:20:57 +0000 (10:20 +0000)]
Ignore GZIPInputStreamTest#testLongMessage failures with CC.
This test sometimes fails with a
java.util.concurrent.TimeoutException on ART Builbot's
x86-64 concurrent collector configuration:
* libcore.java.util.zip.GZIPInputStreamTest#testLongMessage
Disable it while we investigate.
Bug:
26507762
Change-Id: I319a812ff7e6e95857879b2e653c886020cb462b
Nicolas Geoffray [Mon, 22 Feb 2016 09:06:00 +0000 (09:06 +0000)]
Merge "Track memory usage of code cache."
Nicolas Geoffray [Mon, 22 Feb 2016 06:56:01 +0000 (06:56 +0000)]
Merge "Use adds instead of add to set the N flag."
Nicolas Geoffray [Sun, 21 Feb 2016 17:19:18 +0000 (17:19 +0000)]
Use adds instead of add to set the N flag.
That flag will be used by the following bmi instruction.
Fixes 109-suspend-check on arm32.
Change-Id: I07ee069294e7a0d9709718375bce1e21e3db6529
Bill Buzbee [Sun, 21 Feb 2016 14:53:05 +0000 (14:53 +0000)]
Merge "Enable bytecode tracing in ART FI"
Bill Buzbee [Sun, 21 Feb 2016 14:52:27 +0000 (14:52 +0000)]
Merge "x86 ART FI clean-up"
Serguei Katkov [Sat, 20 Feb 2016 06:55:24 +0000 (12:55 +0600)]
Enable bytecode tracing in ART FI
Trace bytecode execution in Fast Interpreter similar to
other interpreters.
Update TraceExecutionEnabled function to switch on tracing.
Change-Id: Icabc17871c8198b11cd4c3dbfaa901e4fbf67946
Signed-off-by: Serguei Katkov <serguei.i.katkov@intel.com>
Mathieu Chartier [Fri, 19 Feb 2016 23:06:16 +0000 (23:06 +0000)]
Merge "Support image classes filtering in image writer"
Mathieu Chartier [Fri, 19 Feb 2016 21:42:48 +0000 (13:42 -0800)]
Support image classes filtering in image writer
New logic prunes classes that have any dependency on a non-image
class. This enables creating smaller images with class profiling.
Code is from class profiling CL.
Added a CHECK for guarding against creating an app boot image with
existing boot image(s).
Bug:
22858531
(cherry picked from commit
7fccc2de5a7ea1bb8241d50869a2a4e44b35767f)
Change-Id: Idcc1de4367b5368bbbd2881cbd63975a646d0831
Mathieu Chartier [Fri, 19 Feb 2016 19:16:20 +0000 (19:16 +0000)]
Merge "Add VLOG(collector) to reduce log spam"
Mathieu Chartier [Fri, 19 Feb 2016 18:25:39 +0000 (10:25 -0800)]
Add VLOG(collector) to reduce log spam
VLOG(gc) is enabled from AndroidRuntime, using it too much produces
log spam.
Bug:
22858531
(cherry picked from commit
a6e1c126299586932ecd3c1133a55a6f8e1107fc)
Change-Id: If60902a8d7855cdabd80208b9b385ce4ba4d9e5f
Andreas Gampe [Fri, 19 Feb 2016 18:38:22 +0000 (18:38 +0000)]
Merge "Revert "Do not build art with clang for mips64""
buzbee [Fri, 19 Feb 2016 18:10:20 +0000 (10:10 -0800)]
ART: Enable JitProfiling for Arm Mterp
Also, fix missing shadow frame clear operation for the 64-bit
shift operations.
Change-Id: Icea95b3aeb1d6d36ea92336fb738cf56edd92da4
Nicolas Geoffray [Fri, 19 Feb 2016 17:39:37 +0000 (17:39 +0000)]
Merge "Workaround JDWP failures"
Andreas Gampe [Fri, 19 Feb 2016 17:33:08 +0000 (09:33 -0800)]
Revert "Do not build art with clang for mips64"
This reverts commit
610a1420fe400d3f638fc0dadb1a012d2ac3ad19.
Bug:
25928358
Change-Id: I3fe8777ead2bd067c071277a7c15b2442f15567f
Sebastien Hertz [Fri, 19 Feb 2016 17:31:06 +0000 (18:31 +0100)]
Workaround JDWP failures
Compile JDWP tests in 1.7 to workaround debug info issues.
Bug:
27262745
Change-Id: I4630a0346d5a99f94895557d705dad310735b45f
Bill Buzbee [Fri, 19 Feb 2016 16:58:41 +0000 (16:58 +0000)]
Merge "Revert "Revert "Revert "Revert "ART: Enable Jit Profiling in Mterp for arm/arm64"""""
Nicolas Geoffray [Fri, 19 Feb 2016 16:25:57 +0000 (16:25 +0000)]
Track memory usage of code cache.
Needed once we start having a strategy based on how full the code
cache is.
Change-Id: I181ad05e1c216afabf106f91ac54bbf144d7406a
Nicolas Geoffray [Fri, 19 Feb 2016 16:01:49 +0000 (16:01 +0000)]
Merge "The JIT does not need VerifedMethod nor CompilerCallbacks."
Bill Buzbee [Thu, 11 Feb 2016 22:37:42 +0000 (22:37 +0000)]
Revert "Revert "Revert "Revert "ART: Enable Jit Profiling in Mterp for arm/arm64""""
This reverts commit
5d03317a834efdf3b5240c401f1bc2ceac7a2f25.
We need to catch all possible cases in which new instrumentation appears
or the debugger is attached, and then switch to the reference interpreter
if necessary. We may, in a future CL, use the alt-mterp mechanism to accompish
this (as did Dalvik).
Only enables Arm64 for now. Once it survives extended testing, will enable
arm and update x86.
Updated OSR handling to match other interpreters.
Change-Id: I076f1d752d6f59899876bab26b18e2221cd92f69
Nicolas Geoffray [Fri, 19 Feb 2016 13:41:08 +0000 (13:41 +0000)]
Merge "Fix braino in single frame deopt."
Jean-Philippe Halimi [Tue, 2 Feb 2016 18:48:52 +0000 (19:48 +0100)]
Dex-wide ArenaPool scoping for AOT compilation
The Arena Pool is a structure that holds memory allocated by the Arena
Allocator, preventing system allocations and deallocations to occur too often
during AOT compilation. Currently, the Arena Pool holds memory it allocates
during the whole AOT compilation process. Unfortunately, such a behavior
generates memory usage overhead, which is exacerbated by bigger applications
such as Facebook.
In this particular app, method size imbalance provokes unnecessary memory
pressure, as one method might require a lot of arena allocations that won't be
used in the remaining compilation. Because the compiler memory footprint keeps
increasing during AOT compilation, the memory pressure becomes very high.
The proposed patch is an attempt to find a tradeoff between allocations /
deallocations time overhead, and the aforementioned memory pressure overhead
resulting of the allocations being held by the Arena Pool. The patch adds a
feature freeing up all memory allocated in the Arena Pool after each dex file
is done compiling.
We have measured no significant AOT compile-time overhead with the patch
(<0.3%), because the selected dex-file granularity is coarse enough. In the
meantime, it provides significant memory footprint improvements. The impact is
especially big with Facebook, because this app has a few methods generating
huge memory footprint peaks (peak native heap footprint goes down 34%).
Change-Id: I27e867e6a20b8a6c28a82cb83140941a8c2b5847
Signed-off-by: Jean-Philippe Halimi <jean-philippe.halimi@intel.com>
Nicolas Geoffray [Fri, 19 Feb 2016 12:43:12 +0000 (12:43 +0000)]
Fix braino in single frame deopt.
handler_method_header_ is actually the OatQuickMethodHeader
of the caller of the method that is deoptimizing (as we will
resume at its call site). We should use the OatQuickMethodHeader
of the method that is deoptimizing and pass it to the code cache
to invalidate the method.
Change-Id: I2b61a96b8415e9477e3ac902c54a314aeb0811a0
Mathieu Chartier [Fri, 19 Feb 2016 02:19:24 +0000 (02:19 +0000)]
Merge "Use image oat file instead of image header for immune spaces"
Mathieu Chartier [Thu, 18 Feb 2016 00:19:53 +0000 (16:19 -0800)]
Use image oat file instead of image header for immune spaces
The old immune spaces logic used the oat file information in the
image header instead of the actual oat file pointer. This was
incorrect for the app image case since the app image oat file is
not necessarily at the address specified in the header. This bug
could cause an incorrect immune region that caused large objects
to get freed if they were within this immune region.
Added test.
Bug:
22858531
Change-Id: I243253e61ea1afd42c3bb2414c02ce6dd36d0f9c
Hiroshi Yamauchi [Fri, 19 Feb 2016 01:52:38 +0000 (01:52 +0000)]
Merge "Fix rosalloc issues with valgrind."
Mathieu Chartier [Fri, 19 Feb 2016 01:16:50 +0000 (01:16 +0000)]
Merge "Don't pass in oat file name to NativeLocationInImage"
Mathieu Chartier [Thu, 18 Feb 2016 02:02:40 +0000 (18:02 -0800)]
Don't pass in oat file name to NativeLocationInImage
We already have the oat file name stored in the relocation, using
the one passed in for the image info can cause a mismatch.
I suspect this fixes a bug repated to copied methods where there
was a oat file name mismatch. When we assign the native object
relocation for the methods of a class, we use the dex cache from
that class.
However, we use method's declaring class in NativeLocationVisitor
by using ArtMethod::GetDexCache. This could result in using the
wrong ImageInfo if the declaring class is in another image.
Change-Id: I753207f7b773fac182f84ba57c640c09f9ce911f
Hiroshi Yamauchi [Fri, 19 Feb 2016 00:17:37 +0000 (00:17 +0000)]
Merge "Disable read barriers in GC VisitReferences calls."