OSDN Git Service
Andreas Gampe [Fri, 25 Sep 2015 15:24:13 +0000 (08:24 -0700)]
ART: Move code from RA to T9 in Mips64 longjump
Do not use the return address. It should not be clobbered as we
may need it, e.g., for single-frame deopt.
Change-Id: I401d7b0034bc4e210bda231ec78ee31242053b36
Vladimir Marko [Thu, 24 Sep 2015 17:13:16 +0000 (17:13 +0000)]
Merge "Optimizing: Rewrite HGraph::FindBackEdges()."
Vladimir Marko [Thu, 24 Sep 2015 17:04:38 +0000 (17:04 +0000)]
Merge "Optimizing: Rewrite HGraph::ComputeDominanceInformation()."
Vladimir Marko [Thu, 24 Sep 2015 12:11:31 +0000 (13:11 +0100)]
Optimizing: Rewrite HGraph::FindBackEdges().
Replace a recursive implementation with a loop using a work
list to avoid stack overflow for 702-LargeBranchOffset in
host debug build with -O0, 512KiB thread pool worker stack.
Change-Id: Iaa91f006fa1099913aeffc9c764879bd004d56de
Vladimir Marko [Wed, 23 Sep 2015 15:07:14 +0000 (16:07 +0100)]
Optimizing: Rewrite HGraph::ComputeDominanceInformation().
Replace a recursive implementation with a loop using a work
list to avoid stack overflow for 702-LargeBranchOffset in
host debug build with -O0.
Bug:
24133462
Change-Id: I444cc85733a9212403a071ea98b9ddfb52bfc402
Vladimir Marko [Thu, 24 Sep 2015 12:19:32 +0000 (12:19 +0000)]
Merge "ART: mprotect() bottom page of ThreadPoolWorker stacks."
David Brazdil [Thu, 24 Sep 2015 11:16:17 +0000 (11:16 +0000)]
Merge "ART: Refactor run-test 482-loop-back-edge-use"
Vladimir Marko [Thu, 24 Sep 2015 09:41:33 +0000 (10:41 +0100)]
ART: mprotect() bottom page of ThreadPoolWorker stacks.
This will catch stack overflows instead of creating hard
to find issues later on.
Bug:
24133462
Change-Id: I3ae5d5da70b8167867936b2561830f3ff47f14fc
Aart Bik [Thu, 24 Sep 2015 01:44:04 +0000 (01:44 +0000)]
Merge "Minor cleanup in range analysis."
Aart Bik [Thu, 24 Sep 2015 00:50:50 +0000 (17:50 -0700)]
Minor cleanup in range analysis.
(1) replaced min/max macro as previously required.
(2) removed some redundant code by merging min/max into one.
Change-Id: I610879a06d550346bfac7e6e12ec0299ba226a37
Aart Bik [Wed, 23 Sep 2015 23:34:54 +0000 (23:34 +0000)]
Merge "Various improvements in finding induction variables."
Aart Bik [Wed, 16 Sep 2015 19:50:25 +0000 (12:50 -0700)]
Various improvements in finding induction variables.
Rationale:
(1) Analyze multi-way phis (requested by Nicolas, Igor, and Mingyao).
(2) Analyze trip count for restricted != loops
(3) Added unit test for public API of range analysis (static methods
were already well-tested).
Change-Id: I9285d22d3bb927f141204cc4697ea6fe5120994d
Aart Bik [Wed, 23 Sep 2015 18:43:56 +0000 (18:43 +0000)]
Merge "Replaced INT_MIN/MAX with modern-day limits."
Mathieu Chartier [Wed, 23 Sep 2015 18:35:57 +0000 (18:35 +0000)]
Merge "Add benchmark for jobject functions"
Aart Bik [Wed, 23 Sep 2015 18:18:57 +0000 (11:18 -0700)]
Replaced INT_MIN/MAX with modern-day limits.
Change-Id: Ia6a0df1e8c6a543c338db0acd75437e1d19701e3
Alex Light [Wed, 23 Sep 2015 16:41:24 +0000 (16:41 +0000)]
Merge "Fix falkiness in 117-nopatchoat"
Alex Light [Sat, 19 Sep 2015 00:09:43 +0000 (17:09 -0700)]
Fix falkiness in 117-nopatchoat
Previously if the image chose a relocation delta of 0 this test would
fail. Now we check for this state directly.
Bug:
24192015
Change-Id: Ie818701edc5605fed590547f0e8b1e97d1d994e3
David Brazdil [Wed, 23 Sep 2015 10:24:56 +0000 (11:24 +0100)]
ART: Refactor run-test 482-loop-back-edge-use
The test used to rely on absolute liveness positions of instructions.
This was painful but worked on all architectures because all passes
were platform agnostic. This does not apply any more and the test
needed to be rewritten.
Change-Id: I423e61ba64a9522361f8e20c27b1975656419f67
Nicolas Geoffray [Wed, 23 Sep 2015 11:04:28 +0000 (11:04 +0000)]
Merge "Don't use ninja to build targets."
Nicolas Geoffray [Wed, 23 Sep 2015 11:02:56 +0000 (12:02 +0100)]
Don't use ninja to build targets.
Ninja does not preserve the -e option given to make.
Change-Id: I3b540aa420a57369b67565477a31c6ff2c17afaf
Mathieu Chartier [Wed, 23 Sep 2015 01:14:29 +0000 (18:14 -0700)]
Add benchmark for jobject functions
Measures performance of:
Add/RemoveLocalRef
Add/RemoveGlobalRef
Add/RemoveWeakGlobalRef
Decoding local, weak, global, handle scope jobjects.
N5 results:
benchmark ns linear runtime
AddRemoveGlobal 476.7 ===========================
AddRemoveLocal 42.0 ==
AddRemoveWeakGlobal 519.7 ==============================
DecodeGlobal 69.9 ====
DecodeHandleScopeRef 38.9 ==
DecodeLocal 70.7 ====
DecodeWeakGlobal 119.7 ======
Change-Id: Iba014a9993909ee45fd52ae48d6dea8428a86bf2
Andreas Gampe [Tue, 22 Sep 2015 23:19:35 +0000 (23:19 +0000)]
Merge "ART: PathClassLoader test"
Andreas Gampe [Mon, 21 Sep 2015 18:25:10 +0000 (11:25 -0700)]
ART: PathClassLoader test
Add a test checking that subclassing PathClassLoader works as
expected.
Change-Id: Ic6bb733b748df5e29b21df021f302ed4237b0f99
Mathieu Chartier [Tue, 22 Sep 2015 23:05:14 +0000 (23:05 +0000)]
Merge "Some clean up for mark sweep files"
Andreas Gampe [Tue, 22 Sep 2015 23:05:00 +0000 (23:05 +0000)]
Merge "[optimizing] Add basic PackedSwitch support"
Mathieu Chartier [Tue, 22 Sep 2015 23:04:55 +0000 (23:04 +0000)]
Merge "Clear allocator in image writer"
Mathieu Chartier [Tue, 22 Sep 2015 22:18:04 +0000 (15:18 -0700)]
Clear allocator in image writer
Doesn't fix any active bugs, just precautionary.
Change-Id: I32a89e00b1d5964ad77fb53eded55263fea5df00
Mathieu Chartier [Tue, 22 Sep 2015 21:53:32 +0000 (14:53 -0700)]
Some clean up for mark sweep files
Bug:
19534862
Change-Id: Ib619e8c4b9d902269d123baa18be56c94114e070
Yabin Cui [Tue, 22 Sep 2015 21:18:55 +0000 (21:18 +0000)]
Merge "Fix static build when libbacktrace links llvm libraries."
Mathieu Chartier [Tue, 22 Sep 2015 18:23:19 +0000 (18:23 +0000)]
Merge "Fix CreateLinearAlloc bug"
Aart Bik [Tue, 22 Sep 2015 18:05:39 +0000 (18:05 +0000)]
Merge "Various improvements in range analysis."
Mathieu Chartier [Tue, 22 Sep 2015 18:04:26 +0000 (11:04 -0700)]
Fix CreateLinearAlloc bug
Should have been using IsAotCompiler since the JIT has IsCompiler
as true but no low 4 GB arena pool.
Fixes 64 bit JIT tests.
Change-Id: Idc4d3e66c573eacd31ae85c66f8493b418d221d7
Mathieu Chartier [Tue, 22 Sep 2015 16:28:12 +0000 (16:28 +0000)]
Merge "Revert "Revert "Add one LinearAlloc per ClassLoader"""
Mathieu Chartier [Tue, 22 Sep 2015 15:50:05 +0000 (08:50 -0700)]
Revert "Revert "Add one LinearAlloc per ClassLoader""
Issue was fixed by:
https://android-review.googlesource.com/#/c/171945/
Bug:
22720414
This reverts commit
7de5dfe37f3cf24e1166412b589f6f67dcd1f1c0.
Aart Bik [Mon, 21 Sep 2015 21:45:05 +0000 (14:45 -0700)]
Various improvements in range analysis.
Rationale:
Using min/max values for "unknowns" is a bit wasteful,
since it eliminates two useful values. Replaced this
with additional boolean to make cases more accurate.
Added few cases to handle examples found in real-life.
Change-Id: I211f8d9a28b1ae79abdb55fb4569716f21d8043b
Vladimir Marko [Tue, 22 Sep 2015 14:23:04 +0000 (14:23 +0000)]
Merge "ART: Fix destruction order in Runtime."
Vladimir Marko [Tue, 22 Sep 2015 10:43:46 +0000 (11:43 +0100)]
ART: Fix destruction order in Runtime.
Destroy arena allocators as late as possible.
With https://android-review.googlesource.com/172057 , the
ArenaPool was indirectly used by ClassLinker's destructor
(via ~LinearAlloc) after it has already been destroyed.
These allocators were being destroyed way too early anyway,
they should be orderly destroyed when the unique_ptr<>
members of Runtime have their destructors executed. However,
due to the potential dependence on MemMap, we destroy them
just before the MemMap::ShutDown().
Change-Id: Ifc3e5a24a29536d5767c82353901e2d22b13d643
Nicolas Geoffray [Tue, 22 Sep 2015 09:53:20 +0000 (09:53 +0000)]
Merge "Further refinements to checkcast/instanceof."
Nicolas Geoffray [Tue, 22 Sep 2015 09:48:28 +0000 (09:48 +0000)]
Merge "Revert "Add one LinearAlloc per ClassLoader""
Nicolas Geoffray [Tue, 22 Sep 2015 09:48:02 +0000 (09:48 +0000)]
Revert "Add one LinearAlloc per ClassLoader"
Times out on 32bit target/host for 132-daemon-locks-shutdown test.
Bug:
22720414
This reverts commit
356412e2b7ba3fde164bc08a44fee0ddc19c54e1.
Change-Id: I5ab3a09e88a5ad8c306a27d2606ecbecc80b9326
Nicolas Geoffray [Mon, 21 Sep 2015 17:41:21 +0000 (18:41 +0100)]
Further refinements to checkcast/instanceof.
- Use setcc when possible.
- Do an exact check in the Object[] case before checking the
component type.
Change-Id: Ic11c60643af9b41fe4ef2beb59dfe7769bef388f
Yabin Cui [Tue, 22 Sep 2015 03:28:36 +0000 (20:28 -0700)]
Fix static build when libbacktrace links llvm libraries.
Bug:
22229391
Change-Id: I8a2f3471cc1915e1e4e1eea0898219614e289ef3
Mathieu Chartier [Mon, 21 Sep 2015 23:29:32 +0000 (23:29 +0000)]
Merge "Add one LinearAlloc per ClassLoader"
Andreas Gampe [Mon, 21 Sep 2015 23:07:57 +0000 (23:07 +0000)]
Merge "ART: Use JIT image for JIT run tests"
Andreas Gampe [Mon, 21 Sep 2015 22:14:38 +0000 (15:14 -0700)]
ART: Use JIT image for JIT run tests
Don't use interpreter image anymore. That was a stop-gap measure.
Change-Id: If4b43bc20026b1ed89e59690fe35771746e354e6
Andreas Gampe [Mon, 21 Sep 2015 20:07:50 +0000 (20:07 +0000)]
Merge "Revert "ART: Skip image visitor in OatWriter for VerifyAtRuntime""
Mathieu Chartier [Mon, 21 Sep 2015 16:07:37 +0000 (09:07 -0700)]
Add one LinearAlloc per ClassLoader
Also added freeing linear alloc and class table when the
corresponding class loader is no longer reachable.
Bug:
22720414
Change-Id: Icb32c3a4c865f240e147bc87ed080a6b1d8a5795
Andreas Gampe [Mon, 21 Sep 2015 18:46:52 +0000 (11:46 -0700)]
Revert "ART: Skip image visitor in OatWriter for VerifyAtRuntime"
This reverts commit
d5ce6a80d6795ada9f47600287706135481cbb4b.
This has an unexpected side effect of not correctly including
reference classes in the boot image. Needs more investigation.
Bug:
24103765
Bug:
24213543
Change-Id: I2a7649f07ec73c59a8cb543f12b6496ac74f7a11
Vladimir Marko [Mon, 21 Sep 2015 18:52:46 +0000 (18:52 +0000)]
Merge "Fix locking on string init map (again)."
David Brazdil [Mon, 21 Sep 2015 17:43:03 +0000 (17:43 +0000)]
Merge "MIPS64: No call if ArraySet does not need a type check"
David Brazdil [Mon, 21 Sep 2015 17:39:16 +0000 (18:39 +0100)]
MIPS64: No call if ArraySet does not need a type check
The MIPS64 back-end did not use the ArraySet::NeedsTypeCheck query to
decide whether to invoke runtime or not like the other back-ends do.
Change-Id: Ie4ef9ee3ca97947256cd59983a2c1c3e3af58c83
David Brazdil [Mon, 21 Sep 2015 17:34:58 +0000 (17:34 +0000)]
Merge "ART: Print short register names with MIPS64"
David Brazdil [Mon, 21 Sep 2015 17:20:26 +0000 (18:20 +0100)]
ART: Print short register names with MIPS64
MIPS64 back-end would print register names as "GpuRegister: X" or
"FpuRegister: X". Space separates Checker attributes and therefore
caused a test to fail. This patch changes it to print "aX" and "fX"
respectively.
Change-Id: Iffae989f25563ef074dd7ef0b09cdfe438119cee
David Brazdil [Mon, 21 Sep 2015 15:44:52 +0000 (15:44 +0000)]
Merge "ART: Fix bug in DeadPhiHandling"
David Brazdil [Thu, 17 Sep 2015 15:47:21 +0000 (16:47 +0100)]
ART: Fix bug in DeadPhiHandling
When reviving dead phis for --debuggable, the DeadPhiHandling algorithm
could produce two phis of the same type, which causes the SSAChecker
graph verifier to fail. This patch fixes the bug.
Note that the code is currently not exercised because compilation is
delegated to Quick for --debuggable.
Bug:
24129675
Change-Id: I26b6dcf3071b325cc7871b989a36c505279ae681
David Brazdil [Mon, 21 Sep 2015 14:30:38 +0000 (14:30 +0000)]
Merge "ART: Optimize catch phi creation to save memory."
Vladimir Marko [Mon, 21 Sep 2015 11:00:16 +0000 (12:00 +0100)]
Fix locking on string init map (again).
Follow-up to
https://android-review.googlesource.com/172036 ,
https://android-review.googlesource.com/171621 .
Don't overwrite existing values, only insert new ones.
(Also improve performance by using move semantics.)
This prevents the following race: Thread 1 looks for string
init map for a method but doesn't find it, so it starts to
construct a new map. Thread 2 is doing the same but it's
faster and actually inserts the new map and keeps a pointer
to it. After Thread 2 releases the lock, Thread 1 acquires
it and starts to Overwrite() the element that the Thread 2
is currently using, so Thread 2 ends up looking at a map
that's being actively modified.
Change-Id: I135571af644363ea7bb282969a1bc7287b34f9b2
Mark Mendell [Fri, 18 Sep 2015 13:26:15 +0000 (09:26 -0400)]
[optimizing] Add basic PackedSwitch support
Add HPackedSwitch, and generate it from the builder. Code generators
convert this to a series of compare/branch tests. Better implementation
in the code generators as a real jump table will follow as separate CLs.
Change-Id: If14736fa4d62809b6ae95280148c55682e856911
Signed-off-by: Mark Mendell <mark.p.mendell@intel.com>
Nicolas Geoffray [Mon, 21 Sep 2015 10:35:24 +0000 (10:35 +0000)]
Merge "Fix braino in new build targets."
Nicolas Geoffray [Mon, 21 Sep 2015 10:20:20 +0000 (11:20 +0100)]
Fix braino in new build targets.
adb should be added to target build, not host build.
Change-Id: I91308f20cee66603441c08ba3dd57c107bf6317c
Nicolas Geoffray [Mon, 21 Sep 2015 10:13:40 +0000 (10:13 +0000)]
Merge "Remove adb from host build in buildbot."
Nicolas Geoffray [Mon, 21 Sep 2015 10:11:40 +0000 (11:11 +0100)]
Remove adb from host build in buildbot.
HOST_PREFER_32_BIT is for ART use only.
Change-Id: I080de289d22ed9ad9ab6befbd0605ef4043e8b15
Jeff Hao [Fri, 18 Sep 2015 22:58:22 +0000 (22:58 +0000)]
Merge "Fix locking on string init map."
Jeff Hao [Fri, 18 Sep 2015 21:44:32 +0000 (14:44 -0700)]
Fix locking on string init map.
Related to: https://android-review.googlesource.com/#/c/171621
Change-Id: Ib6dcf9914f490f1d744c0abe1f520e5b307c7acd
Mathieu Chartier [Fri, 18 Sep 2015 19:13:26 +0000 (19:13 +0000)]
Merge "Handle OOME during threadstress finishing message"
Andreas Gampe [Fri, 18 Sep 2015 18:07:49 +0000 (18:07 +0000)]
Merge "ART: Decrease dex2oat watchdog timeout"
Mathieu Chartier [Fri, 18 Sep 2015 03:46:56 +0000 (20:46 -0700)]
Handle OOME during threadstress finishing message
If there is an OOME that occurs after the "Finishing workers" it
caused the test to fail.
Also guard worker thread creation by try catch for OOME.
Bug:
18577101
Change-Id: I69367be0aad3f60093c02c7f63ae3c20757fb89b
Andreas Gampe [Fri, 18 Sep 2015 17:55:28 +0000 (17:55 +0000)]
Merge "ART: Clean up less in dex2oat"
Andreas Gampe [Thu, 17 Sep 2015 21:03:21 +0000 (14:03 -0700)]
ART: Clean up less in dex2oat
In non-debug builds, clean up even less. We already did not shut
down the runtime. Also skipping the compiler driver and the
verification results removes all major points of destructor
performance.
Tested with a common large app on Nexus 9. Time between dex2oat timing
message and executable exit (log from immediately-after log echo)
[w/o swap, w/ swap].
Before: 2.409s / 48.774s
After: 0.132s / 0.188s
Bug:
24199200
Change-Id: I5d8c17f8e28796545cfbb3887c07c92905f9b48d
Alex Light [Fri, 18 Sep 2015 17:37:17 +0000 (17:37 +0000)]
Merge "Make run-{libcore,jdwp}-tests work with OUT_DIR set."
Nicolas Geoffray [Fri, 18 Sep 2015 14:20:39 +0000 (14:20 +0000)]
Merge "Revert "Revert "Optimize code generation of check-cast and instance-of."""
Nicolas Geoffray [Fri, 18 Sep 2015 13:40:46 +0000 (13:40 +0000)]
Revert "Revert "Optimize code generation of check-cast and instance-of.""
This reverts commit
7537437c6a2f89249a48e30effcc27d4e7c5a04f.
Change-Id: If759cb08646e47b62829bebc3c5b1e2f2969cf84
David Brazdil [Fri, 18 Sep 2015 13:58:57 +0000 (14:58 +0100)]
ART: Optimize catch phi creation to save memory.
Catch phis are eagerly created for all vregs and then removed as
throwing instructions with undefined locals are encountered. This can
be optimized since rougly one third of vregs is already undefined with
the first throwing instruction.
Change-Id: Ia4e66f19ea03f9a129b4b64d4193d262fff45224
Nicolas Geoffray [Fri, 18 Sep 2015 13:51:30 +0000 (13:51 +0000)]
Merge "Fix x64's cmpw."
Nicolas Geoffray [Fri, 18 Sep 2015 13:36:49 +0000 (14:36 +0100)]
Fix x64's cmpw.
Change-Id: If700f2994990864c8b34aa52eb7a767153a1f917
David Brazdil [Fri, 18 Sep 2015 13:16:03 +0000 (13:16 +0000)]
Merge "ART: Delete catch phis for undefined vregs."
David Brazdil [Fri, 18 Sep 2015 09:58:32 +0000 (10:58 +0100)]
ART: Delete catch phis for undefined vregs.
SSA builder assumed that (eagerly created) catch phis for undefined
vregs must be dead and will be deleted by dead phi elimination. This,
however, does not hold for --debuggable because the catch phis might
be revived for their environment uses. This patch modifies the builder
to delete the phis itself.
Bug:
24054676
Change-Id: Iaa5f2487ff1c38353b44cb89c709bbff1ecd73cc
Vladimir Marko [Fri, 18 Sep 2015 08:31:16 +0000 (08:31 +0000)]
Merge "ART: Fix two allocation/deallocation mismatches."
Sebastien Hertz [Fri, 18 Sep 2015 07:07:01 +0000 (07:07 +0000)]
Merge "Deoptimization support in optimizing compiler for setting local values"
Andreas Gampe [Mon, 14 Sep 2015 22:34:38 +0000 (15:34 -0700)]
ART: Decrease dex2oat watchdog timeout
Keep the dex2oat watchdog timeout lower than the package manager
timeout, so that dex2oat kills itself before the system server
watchdog kills the system because of the long installation.
Bug:
23629410
Change-Id: I5faa23a1715736a1c2ba3f302a6ee41130d3fbae
Jeff Hao [Fri, 18 Sep 2015 01:58:23 +0000 (01:58 +0000)]
Merge "Add lock around interpreter string init reg map."
Alex Light [Thu, 17 Sep 2015 23:59:09 +0000 (16:59 -0700)]
Make run-{libcore,jdwp}-tests work with OUT_DIR set.
Change the two scripts to use the value of the OUT_DIR environment
variable if it is set to find the build directory. If OUT_DIR is not
set just fall back to the default 'out/'.
Change-Id: I19824d792d5dcbed32846912fb36654894fcc0ac
Jeff Hao [Fri, 18 Sep 2015 00:29:01 +0000 (17:29 -0700)]
Add lock around interpreter string init reg map.
Fixes intermittent test failures. Note that locks can't be held
while the verifier is generating the map.
Change-Id: Ie38d6bf0d9c5e81d66b8c167fde0e75a0565cea5
Andreas Gampe [Fri, 18 Sep 2015 00:10:09 +0000 (00:10 +0000)]
Merge "Implement StringGetCharsNoCheck intrinsic for X86"
Andreas Gampe [Thu, 17 Sep 2015 23:26:15 +0000 (23:26 +0000)]
Merge "ART: Skip image visitor in OatWriter for VerifyAtRuntime"
Hiroshi Yamauchi [Thu, 17 Sep 2015 21:55:46 +0000 (21:55 +0000)]
Merge "Allow null self only in DecodeWeakGlobalDuringShutdown()."
Andreas Gampe [Thu, 17 Sep 2015 21:43:46 +0000 (21:43 +0000)]
Merge "ART: Refactor intrinsics slow-paths"
Andreas Gampe [Wed, 9 Sep 2015 20:15:38 +0000 (13:15 -0700)]
ART: Refactor intrinsics slow-paths
Refactor slow paths so that there is a default implementation for
common cases (only arm64 with vixl is special). Write a generic
intrinsic slow-path that can be reused for the specific architectures.
Move helper functions into CodeGenerator so that they are accessible.
Change-Id: Ibd788dce432601c6a9f7e6f13eab31f28dcb8550
Andreas Gampe [Thu, 17 Sep 2015 00:12:55 +0000 (17:12 -0700)]
ART: Skip image visitor in OatWriter for VerifyAtRuntime
There's no use running the visitor, and it will force-resolve all
methods. Avoid the overhead. Reduces boot image compile time in
verify-at-runtime mode from 1.25s to 1s (on the host, in -j4 mode).
Bug:
24103765
Change-Id: Icdc3be628fed0c9bda01cf5f24fe377f23c571a9
Evgenii Stepanov [Thu, 17 Sep 2015 20:57:34 +0000 (20:57 +0000)]
Merge "asan: update condition to work with multiple SANITIZE_TARGET values."
Ivan Krasin [Thu, 17 Sep 2015 20:40:19 +0000 (13:40 -0700)]
asan: update condition to work with multiple SANITIZE_TARGET values.
The goal is to enable SANITIZE_TARGET='address coverage', which
will be used by LLVMFuzzer.
Bug:
22850550
Bug:
22233158
Change-Id: I909c6268e5c3d2da0361f2a2216cac080afbee64
Vladimir Marko [Thu, 17 Sep 2015 19:22:02 +0000 (20:22 +0100)]
ART: Fix two allocation/deallocation mismatches.
Change-Id: I5088126cbd5b5e4b461a449eecd5b3574883f413
Hiroshi Yamauchi [Thu, 17 Sep 2015 04:11:44 +0000 (21:11 -0700)]
Allow null self only in DecodeWeakGlobalDuringShutdown().
To follow up CL 169855, allow a null current thread only in
DecodeWeakGlobalDuringShutdown() as a special case rather than
DecodeWeakGlobal(). This is to prevent a bug where null is accidentally
passed to DecodeWeakGlobal().
Bug:
23897251
Change-Id: I5e7bb78ec739b8bfcf77284ed321d507737ee33e
Nicolas Geoffray [Thu, 17 Sep 2015 17:12:37 +0000 (17:12 +0000)]
Merge "Revert "Optimize code generation of check-cast and instance-of.""
Nicolas Geoffray [Thu, 17 Sep 2015 17:12:19 +0000 (17:12 +0000)]
Revert "Optimize code generation of check-cast and instance-of."
Failures with libcore tests.
This reverts commit
64acf303eaa2f32c0b1d8cfcbf044a822c5eec08.
Change-Id: Ie6f323fcf5d86bae5c334c1352bb21f1bad60a88
Josh Gao [Thu, 17 Sep 2015 16:46:14 +0000 (16:46 +0000)]
Merge "Use arc4random_buf instead of getauxval(AT_RANDOM)."
Vladimir Marko [Thu, 17 Sep 2015 15:53:10 +0000 (15:53 +0000)]
Merge "Optimizing: Allow storing value objects in containers."
Vladimir Marko [Thu, 17 Sep 2015 15:32:33 +0000 (15:32 +0000)]
Merge "Optimizing: Do not use range-based loop when inserting elements."
Mingyao Yang [Mon, 6 Jul 2015 18:10:37 +0000 (11:10 -0700)]
Deoptimization support in optimizing compiler for setting local values
Due to compiler optimizations, we may not always be able to update
the value of a local variable in a compiled frame (like a variable
seen as constant by the compiler). To avoid that situation, we simply
deoptimize compiled frames updated by the debugger so they are
executed by the interpreter with the updated value.
When the debugger attempts to set a local variable (actually a DEX
register or a pair of registers) in a compiled frame, we allocate a
ShadowFrame associated to that frame (using its frame id) and set the
new value in that ShadowFrame. When we know we are about to continue
the execution of the compiled frame, we deoptimize the stack using
the preallocated ShadowFrame (instead of creating a new one). We
initialize it with the current value of all DEX registers except
the ones that have been set by the debugger. Therefore, the
ShadowFrame represent the runtime context modified by the debugger.
Bumps oat version to force recompilation.
Bug:
19944235
Change-Id: I0ebe6241264f7a3be0f14ee4516c1f7436e04da6
Vladimir Marko [Thu, 17 Sep 2015 14:47:05 +0000 (15:47 +0100)]
Optimizing: Do not use range-based loop when inserting elements.
When we iterate over the elements of a container and we may
insert new elements into that container, it's wrong to use
the range-based loop.
Bug:
24133462
Change-Id: Iee35fbcf88ed3bcd6155cbeba09bd256032a16be