OSDN Git Service

android-x86/art.git
8 years agoMerge "Thumb2: Show the immediate in ROR (immediate) disassembly."
Vladimir Marko [Thu, 7 Apr 2016 14:38:29 +0000 (14:38 +0000)]
Merge "Thumb2: Show the immediate in ROR (immediate) disassembly."

8 years agoMerge "Add 20s timeout to 570-checker-osr."
Vladimir Marko [Thu, 7 Apr 2016 12:12:35 +0000 (12:12 +0000)]
Merge "Add 20s timeout to 570-checker-osr."

8 years agoAdd 20s timeout to 570-checker-osr.
Vladimir Marko [Thu, 7 Apr 2016 09:12:31 +0000 (10:12 +0100)]
Add 20s timeout to 570-checker-osr.

It's better to report a timeout than to hang indefinitely.

Change-Id: Ibfa8059c8b44220c7a55801b020094fc342c8047

8 years agoMerge "Optimizing: Fix codegens for MethodLoadKind::kDexCacheViaMethod."
Vladimir Marko [Thu, 7 Apr 2016 09:18:36 +0000 (09:18 +0000)]
Merge "Optimizing: Fix codegens for MethodLoadKind::kDexCacheViaMethod."

8 years agoMerge "Move optimizations out of HInliner::TryBuildAndInlineHelper."
Roland Levillain [Thu, 7 Apr 2016 09:08:06 +0000 (09:08 +0000)]
Merge "Move optimizations out of HInliner::TryBuildAndInlineHelper."

8 years agoMerge changes I775a57c9,I14c0836f,I075c1b11
Andreas Gampe [Thu, 7 Apr 2016 00:42:30 +0000 (00:42 +0000)]
Merge changes I775a57c9,I14c0836f,I075c1b11

* changes:
  ART: Add cutout for Unsafe.putObjectVolatile
  ART: Add integer & long parsing cutout to unstarted runtime
  ART: Add getDeclaredConstructor cutout for unstarted runtime

8 years agoMerge "Check if we require barrier if we did not resolve classes"
Mathieu Chartier [Wed, 6 Apr 2016 23:05:00 +0000 (23:05 +0000)]
Merge "Check if we require barrier if we did not resolve classes"

8 years agoCheck if we require barrier if we did not resolve classes
Mathieu Chartier [Wed, 6 Apr 2016 21:41:23 +0000 (14:41 -0700)]
Check if we require barrier if we did not resolve classes

Check fields instead of just always inserting the return void
barrier.

Bug: 28005874

(cherry picked from commit 03c7df9648028228d75dddfc25cbc81ab8096167)

Change-Id: I1207fecc9a4425d024a7e5e1a850c9e2a6e17c31

8 years agoMerge "Fix regression in ModUnionTable RAM usage"
Mathieu Chartier [Wed, 6 Apr 2016 22:53:45 +0000 (22:53 +0000)]
Merge "Fix regression in ModUnionTable RAM usage"

8 years agoFix regression in ModUnionTable RAM usage
Mathieu Chartier [Wed, 6 Apr 2016 22:27:19 +0000 (15:27 -0700)]
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

8 years agoART: Add cutout for Unsafe.putObjectVolatile
Andreas Gampe [Wed, 6 Apr 2016 04:12:51 +0000 (21:12 -0700)]
ART: Add cutout for Unsafe.putObjectVolatile

Add an unstarted-runtime cutout for putObjectVolatile.

Allows to compile-time initialize:
* android.content.res.Configuration
* android.icu.text.PluralRules$SimpleTokenizer
* android.icu.text.UnicodeSet
* android.icu.util.VersionInfo
* android.util.LocaleList

May allow follow-up changes to compile-time initialize
more ICU classes.

Bug: 27265238

(cherry picked from commit 1813d116cb1367f0ae449221c7ff6c6b421d269f)

Change-Id: I775a57c9f15aa8f42a4dffbd1dd40c641c30f645

8 years agoART: Add integer & long parsing cutout to unstarted runtime
Andreas Gampe [Wed, 6 Apr 2016 03:14:30 +0000 (20:14 -0700)]
ART: Add integer & long parsing cutout to unstarted runtime

Add a cutout for Integer.parseInt and Long.parseLong. Only handle
(some) successful cases, and abort the transaction in the rest.
Add tests.

Allows to compile-time initialize:
* com.android.org.bouncycastle.asn1.pkcs.PKCSObjectIdentifiers
* com.android.org.bouncycastle.asn1.x509.X509ObjectIdentifiers
* sun.security.x509.InhibitAnyPolicyExtension

Bug: 27265238

(cherry picked from commit 9c62dab16e4be0c19a2e6c6c0cc2b24c2814c6f2)

Change-Id: I14c0836fd9ec7c9bf49d8186ff14a3c6830ff711

8 years agoART: Add getDeclaredConstructor cutout for unstarted runtime
Andreas Gampe [Wed, 6 Apr 2016 01:18:43 +0000 (18:18 -0700)]
ART: Add getDeclaredConstructor cutout for unstarted runtime

In the vein of getDeclaredMethod and getDeclaredField. Refactor
code from java_lang_Class to share code.

Allows to compile-time initialize:
* sun.security.x509.AVAKeyword
* sun.security.x509.X500Name

Bug: 27265238

(cherry picked from commit f72f19fac0016499d33fd75b0463d0128a58a9f4)

Change-Id: I075c1b11f3408f5b31e4f84140a24fd8d3eaa17e

8 years agoMerge "Shard classloader classes lock"
Mathieu Chartier [Wed, 6 Apr 2016 19:25:39 +0000 (19:25 +0000)]
Merge "Shard classloader classes lock"

8 years agoShard classloader classes lock
Mathieu Chartier [Tue, 5 Apr 2016 21:36:57 +0000 (14:36 -0700)]
Shard classloader classes lock

Used to guard adding and removing classes.

Previously we used the class linker classes lock, but this had
a deadlock issue since the reference processor may need to acquire
the lock to mark the classes of a class loader. Another thread could
be blocked trying to access weak globals while also holding the
class linker classes lock.

Bug: 27946564

(cherry picked from commit 1609e3a42051769f4a8be3b6731e7bb2f828b3bb)

Change-Id: Ic5cfe573c4e6822d49ad0862ffdd9d036e439a96

8 years agoMerge "Change RequiresConstructorBarrier default to yes"
Mathieu Chartier [Wed, 6 Apr 2016 19:22:28 +0000 (19:22 +0000)]
Merge "Change RequiresConstructorBarrier default to yes"

8 years agoChange RequiresConstructorBarrier default to yes
Mathieu Chartier [Wed, 6 Apr 2016 17:47:45 +0000 (10:47 -0700)]
Change RequiresConstructorBarrier default to yes

Previously it defaulted to false, this caused incorrect dex2dex if
the resolution pass was not run.

TODO: Re-enable the no barrier for no resolution case to improve
interpreter performance slightly.

Bug: 28005874

(cherry picked from commit e5d8020b03956db6d533852b3ae940e0eb0fafc4)

Change-Id: Ibde6e18b8543f7bc9e17db6941a710756cd8b617

8 years agoMerge "Change hash set to use noexcept"
Mathieu Chartier [Wed, 6 Apr 2016 18:35:18 +0000 (18:35 +0000)]
Merge "Change hash set to use noexcept"

8 years agoChange hash set to use noexcept
Mathieu Chartier [Wed, 6 Apr 2016 02:13:37 +0000 (19:13 -0700)]
Change hash set to use noexcept

Otherwise it uses copy constructor instead of move constructor in
some cases (like std::vector).

Reduces shared dirty native by around 350k.

Bug: 27860101

(cherry picked from commit 5ef868c8332db11bb90284887a7f676f5dbef373)

Change-Id: I0311fa530fc2436630abebfdac2cad375eb4d691

8 years agoThumb2: Show the immediate in ROR (immediate) disassembly.
Vladimir Marko [Wed, 6 Apr 2016 18:05:19 +0000 (19:05 +0100)]
Thumb2: Show the immediate in ROR (immediate) disassembly.

Now disassembled as "mov ., ., ror #imm5".

Change-Id: Iad201662fd6aa3f87b95c7293fafe030c4bbdbf3

8 years agoMerge "Fix and re-enable compiler_driver_test"
Yi Kong [Wed, 6 Apr 2016 16:59:02 +0000 (16:59 +0000)]
Merge "Fix and re-enable compiler_driver_test"

8 years agoMerge "ART: Handle OOM in stack overflow"
Mathieu Chartier [Wed, 6 Apr 2016 16:51:48 +0000 (16:51 +0000)]
Merge "ART: Handle OOM in stack overflow"

8 years agoFix and re-enable compiler_driver_test
Yi Kong [Mon, 4 Apr 2016 16:44:59 +0000 (17:44 +0100)]
Fix and re-enable compiler_driver_test

This reverts commit 6f58cb85328bed2f3ea246cce4080244544d7c2b.

The move to OpenJDK introduces new dependencies to compile these
methods, however we don't actually need them to run this test. Remove
them here to enable the test again.

Bug: 25836016
Change-Id: I75b12c48941fca6646142004accae8364e8c6dbc

8 years agoMove optimizations out of HInliner::TryBuildAndInlineHelper.
Roland Levillain [Wed, 6 Apr 2016 16:45:58 +0000 (17:45 +0100)]
Move optimizations out of HInliner::TryBuildAndInlineHelper.

This is a followup CL to
https://android-review.googlesource.com/#/c/212651.

Change-Id: I37d1c3fc06d8e36c335dd62887936486f7deeba4

8 years agoOptimizing: Fix codegens for MethodLoadKind::kDexCacheViaMethod.
Vladimir Marko [Wed, 6 Apr 2016 16:33:41 +0000 (17:33 +0100)]
Optimizing: Fix codegens for MethodLoadKind::kDexCacheViaMethod.

Use the original method index instead of the target method
index because the target method may point to a different dex
file.

No regression test: this currently happens only if the
codegen uses the kDexCacheViaMethod as a fallback for
another load kind and we aim to avoid that fallback, so it
would be difficult to write a reliable regression test. We
could try and exploit current fallbacks for irreducible
loops on x86 and arm but those fallbacks will eventually
disappear anyway.

Bug: 28036230
Change-Id: I4cc9e046480d3d60a7fb521f0ca6a98914625cdc

8 years agoART: Handle OOM in stack overflow
Mathieu Chartier [Mon, 4 Apr 2016 23:49:44 +0000 (16:49 -0700)]
ART: Handle OOM in stack overflow

When creating a stack overflow error, we must not allocate a
new out-of-memory error. Running its constructor will lead to
a nested exception.

Bug: 27663199

Change-Id: I2a7b5b937730eeade22dce654cfc4ad903c5f040

8 years agoMerge "ART: Add arraycopy(byte) to unstarted runtime"
Andreas Gampe [Wed, 6 Apr 2016 01:18:15 +0000 (01:18 +0000)]
Merge "ART: Add arraycopy(byte) to unstarted runtime"

8 years agoART: Add arraycopy(byte) to unstarted runtime
Andreas Gampe [Tue, 5 Apr 2016 21:16:10 +0000 (14:16 -0700)]
ART: Add arraycopy(byte) to unstarted runtime

The byte-array version is now in use in libcore.

Allows to compile-time initialize:
* java.security.cert.X509CertSelector
* sun.security.pkcs.PKCS9Attribute
* sun.security.provider.certpath.AdaptableX509CertSelector
* sun.security.x509.AccessDescription
* sun.security.x509.AlgorithmId
* sun.security.x509.ExtendedKeyUsageExtension
* sun.security.x509.NetscapeCertTypeExtension
* sun.security.x509.OIDMap
* sun.security.x509.PKIXExtensions

Bug: 27265238

(cherry picked from commit cc44581beacd61a633b02a1223d1bf6fea278f94)

Change-Id: If6b680a455809152c9bd7a679a8ab430936c46a1

8 years agoMerge "ART: Don't attempt write flock if not zygote"
Andreas Gampe [Tue, 5 Apr 2016 23:02:42 +0000 (23:02 +0000)]
Merge "ART: Don't attempt write flock if not zygote"

8 years agoART: Don't attempt write flock if not zygote
Andreas Gampe [Tue, 5 Apr 2016 17:26:42 +0000 (10:26 -0700)]
ART: Don't attempt write flock if not zygote

When loading an image, don't attempt a RDWR flock if the process
is not the zygote. The lock will be rejected as the process does
not have permission to write the file.

Bug: 28011897

(cherry picked from commit fff377b13469b8c84ce55ee20b10ca80a5a0813d)

Change-Id: I567934c4e9a9a13e9131711240e9ecfdd58b69f7

8 years agoMerge "Fix issued with non-public interface methods."
Alex Light [Tue, 5 Apr 2016 17:49:22 +0000 (17:49 +0000)]
Merge "Fix issued with non-public interface methods."

8 years agoMerge "Remove unnecessary TEMP_FAILURE_RETRY."
Vladimir Marko [Tue, 5 Apr 2016 15:43:16 +0000 (15:43 +0000)]
Merge "Remove unnecessary TEMP_FAILURE_RETRY."

8 years agoMerge "Implement ARM64 support for SystemArrayCopy()"
Roland Levillain [Tue, 5 Apr 2016 13:25:12 +0000 (13:25 +0000)]
Merge "Implement ARM64 support for SystemArrayCopy()"

8 years agoRemove unnecessary TEMP_FAILURE_RETRY.
Vladimir Marko [Tue, 5 Apr 2016 13:19:08 +0000 (14:19 +0100)]
Remove unnecessary TEMP_FAILURE_RETRY.

unlink, rmdir, rename and closedir do not generate EINTR.

Change-Id: Ia5daab3e19f7373d7c27cdb6a800351c86b5a4e5

8 years agoMerge "Remove unused 'status' variable"
Vladimir Marko [Tue, 5 Apr 2016 12:36:16 +0000 (12:36 +0000)]
Merge "Remove unused 'status' variable"

8 years agoMerge "Revert "Remove many `instrumentation` variables from goto interpreter.""
Vladimir Marko [Tue, 5 Apr 2016 11:52:23 +0000 (11:52 +0000)]
Merge "Revert "Remove many `instrumentation` variables from goto interpreter.""

8 years agoRevert "Remove many `instrumentation` variables from goto interpreter."
Vladimir Marko [Tue, 5 Apr 2016 11:48:02 +0000 (11:48 +0000)]
Revert "Remove many `instrumentation` variables from goto interpreter."

This reverts commit 4cb22b4f6d0afbdf424369df99c6eb11999f2473.

Change-Id: I1d84965c2cde2ffc2ebdb46fb5251691ea12569b

8 years agoRemove unused 'status' variable
Goran Jakovljevic [Tue, 5 Apr 2016 11:08:21 +0000 (13:08 +0200)]
Remove unused 'status' variable

GCC (mips32 Buildbot) is complaining about unused variable.

This is follow up change for
    https://android-review.googlesource.com/#/c/212523/.

Change-Id: Id6ac2736d30f324f2cad68d8dabe409092f000ac

8 years agoMerge "Clean up OatQuickMethodHeader after Quick removal."
Vladimir Marko [Tue, 5 Apr 2016 09:23:52 +0000 (09:23 +0000)]
Merge "Clean up OatQuickMethodHeader after Quick removal."

8 years agoMerge "Move DCHECK into loop."
Aart Bik [Mon, 4 Apr 2016 22:08:30 +0000 (22:08 +0000)]
Merge "Move DCHECK into loop."

8 years agoFix issued with non-public interface methods.
Alex Light [Thu, 31 Mar 2016 17:03:07 +0000 (10:03 -0700)]
Fix issued with non-public interface methods.

Some APK's ship with dex files containing non-public interface
methods. These would cause crashes on newer versions of ART due to
stricter verification of dex files. Make ART emit a warning but allow
this behavior.

Bug: 27928832

Change-Id: Ia3689ee091aa154fb5954b1f6dd50c489128acce

8 years agoMerge "Fix issue with verification check failure on erroneous classes."
Alex Light [Mon, 4 Apr 2016 21:46:30 +0000 (21:46 +0000)]
Merge "Fix issue with verification check failure on erroneous classes."

8 years agoMerge "ART: Dump less maps"
Andreas Gampe [Mon, 4 Apr 2016 21:24:59 +0000 (21:24 +0000)]
Merge "ART: Dump less maps"

8 years agoART: Dump less maps
Andreas Gampe [Sat, 2 Apr 2016 00:20:49 +0000 (17:20 -0700)]
ART: Dump less maps

Security has been tightened, so there are situations where we can't
load our files, and produce "log spam."

Bug: 27925454

(cherry picked from commit dfd7b70bbab8157fb385e4a6d6e4b55ae2e4e892)

Change-Id: Ia8c76184ea5087d625cf52d4459f8a557eebfafe

8 years agoMove DCHECK into loop.
Aart Bik [Mon, 4 Apr 2016 21:19:01 +0000 (14:19 -0700)]
Move DCHECK into loop.

Rationale:
Since pointer is non-null on loop entry, moving the check
into the while loop (rather than after) gives a more friendly
error message in debug mode when something goes wrong.

BUG=27825731
BUG=27627004

Change-Id: I2ea67b3ab6c4edc6815fea38a436e5c3c194e540

8 years agoFix issue with verification check failure on erroneous classes.
Alex Light [Mon, 4 Apr 2016 20:23:14 +0000 (13:23 -0700)]
Fix issue with verification check failure on erroneous classes.

Bug: 27924355

Change-Id: I7d680e7196aeb9e2e6ef54139634564e058174d7

8 years agoMerge "Mark array classes as verification attempted"
Mathieu Chartier [Mon, 4 Apr 2016 18:40:51 +0000 (18:40 +0000)]
Merge "Mark array classes as verification attempted"

8 years agoMark array classes as verification attempted
Mathieu Chartier [Sat, 2 Apr 2016 00:33:31 +0000 (17:33 -0700)]
Mark array classes as verification attempted

We now initialize all array classes in the compiler driver. This in
turn will ensure they are marked verified in the image. This
prevents dirty pages in the image since we would otherwise set the
flags in the zygote.

On BusinessCard:
Reduces shared dirty from 636k -> 432k for boot.art, and dirty pages
from 183 to 132.

(cherry picked from commit 3b674098bef6eaf7d3fb731878293ef715fe1080)

Change-Id: If3093e4e3242e4ee3ea120abe5be7db028290260

8 years agoMerge "Revert "Refactor HGraphBuilder and SsaBuilder to remove HLocals""
David Brazdil [Mon, 4 Apr 2016 17:48:02 +0000 (17:48 +0000)]
Merge "Revert "Refactor HGraphBuilder and SsaBuilder to remove HLocals""

8 years agoRevert "Refactor HGraphBuilder and SsaBuilder to remove HLocals"
David Brazdil [Mon, 4 Apr 2016 17:47:42 +0000 (17:47 +0000)]
Revert "Refactor HGraphBuilder and SsaBuilder to remove HLocals"

Bug: 27995065
This reverts commit e3ff7b293be2a6791fe9d135d660c0cffe4bd73f.

Change-Id: I5363c7ce18f47fd422c15eed5423a345a57249d8

8 years agoMerge "Have Checker print lines of unmatched test cases."
Roland Levillain [Mon, 4 Apr 2016 16:58:06 +0000 (16:58 +0000)]
Merge "Have Checker print lines of unmatched test cases."

8 years agoClean up OatQuickMethodHeader after Quick removal.
Vladimir Marko [Thu, 31 Mar 2016 11:02:28 +0000 (12:02 +0100)]
Clean up OatQuickMethodHeader after Quick removal.

This reduces the size of the pre-header by 8 bytes, reducing
oat file size and mmapped .text section size. The memory
needed to store a CompiledMethod by dex2oat is also reduced,
for 32-bit dex2oat by 8B and for 64-bit dex2oat by 16B. The
aosp_flounder-userdebug 32-bit and 64-bit boot.oat are each
about 1.1MiB smaller.

Disable the broken StubTest.IMT, b/27991555 .

Change-Id: I05fe45c28c8ffb7a0fa8b1117b969786748b1039

8 years agoMerge "Please Clang wrt stack frames in HInliner::TryBuildAndInlineHelper."
Roland Levillain [Mon, 4 Apr 2016 14:36:12 +0000 (14:36 +0000)]
Merge "Please Clang wrt stack frames in HInliner::TryBuildAndInlineHelper."

8 years agoPlease Clang wrt stack frames in HInliner::TryBuildAndInlineHelper.
Roland Levillain [Mon, 4 Apr 2016 14:34:31 +0000 (15:34 +0100)]
Please Clang wrt stack frames in HInliner::TryBuildAndInlineHelper.

This change enables Clang to compile ART on MIPS64 without
complaining about stack frames larger than what
`-Wframe-larger-than` allows.

Change-Id: I72f424922d9cb3d01f7b8ba4e5cb9170b82870d9

8 years agoHave Checker print lines of unmatched test cases.
Roland Levillain [Mon, 4 Apr 2016 13:56:36 +0000 (14:56 +0100)]
Have Checker print lines of unmatched test cases.

Change-Id: I8ea7eee4c29f23ad10fb4d49ca927fdde39c1ef9

8 years agoMerge "Use bit-level packing for InlineInfo in stackmaps as well."
Calin Juravle [Mon, 4 Apr 2016 13:35:53 +0000 (13:35 +0000)]
Merge "Use bit-level packing for InlineInfo in stackmaps as well."

8 years agoMerge "Fix BitCount intrinsics assertions."
Roland Levillain [Mon, 4 Apr 2016 12:58:23 +0000 (12:58 +0000)]
Merge "Fix BitCount intrinsics assertions."

8 years agoMerge "Improve debugging in art/compiler/optimizing/common_arm64.h."
Roland Levillain [Mon, 4 Apr 2016 11:37:58 +0000 (11:37 +0000)]
Merge "Improve debugging in art/compiler/optimizing/common_arm64.h."

8 years agoMerge "Refactor HGraphBuilder and SsaBuilder to remove HLocals"
David Brazdil [Mon, 4 Apr 2016 11:12:14 +0000 (11:12 +0000)]
Merge "Refactor HGraphBuilder and SsaBuilder to remove HLocals"

8 years agoMerge "Build dominator tree before generating HInstructions"
David Brazdil [Mon, 4 Apr 2016 11:11:34 +0000 (11:11 +0000)]
Merge "Build dominator tree before generating HInstructions"

8 years agoRefactor HGraphBuilder and SsaBuilder to remove HLocals
David Brazdil [Wed, 2 Mar 2016 16:48:20 +0000 (16:48 +0000)]
Refactor HGraphBuilder and SsaBuilder to remove HLocals

This patch merges the instruction-building phases from HGraphBuilder
and SsaBuilder into a single HInstructionBuilder class. As a result,
it is not necessary to generate HLocal, HLoadLocal and HStoreLocal
instructions any more, as the builder produces SSA form directly.

Saves 5-15% of arena-allocated memory (see bug for more data):
  GMS      20.46MB  =>  19.26MB  (-5.86%)
  Maps     24.12MB  =>  21.47MB  (-10.98%)
  YouTube  28.60MB  =>  26.01MB  (-9.05%)

Bug: 27894376
Change-Id: Iefe28d40600c169c5d306fd2c77034ae19476d90

8 years agoBuild dominator tree before generating HInstructions
David Brazdil [Tue, 16 Feb 2016 09:26:07 +0000 (09:26 +0000)]
Build dominator tree before generating HInstructions

Second CL in the series of merging HGraphBuilder and SsaBuilder. This
patch refactors the builders so that dominator tree can be built
before any HInstructions are generated. This puts the SsaBuilder
removal of HLoadLocals/HStoreLocals straight after HGraphBuilder's
HInstruction generation phase. Next CL will therefore be able to
merge them.

This patch also adds util classes for iterating bytecode and switch
tables which allowed to simplify the code.

Bug: 27894376
Change-Id: Ic425d298b2e6e7980481ed697230b1a0b7904526

8 years agoMerge "Dump different fields in imgdiag"
Mathieu Chartier [Fri, 1 Apr 2016 23:36:19 +0000 (23:36 +0000)]
Merge "Dump different fields in imgdiag"

8 years agoDump different fields in imgdiag
Mathieu Chartier [Fri, 1 Apr 2016 20:56:41 +0000 (13:56 -0700)]
Dump different fields in imgdiag

Dump which fields are different, also print field values.

(cherry picked from commit ec1533880f3606546baceafe2db513d1a63c20da)

Change-Id: Ie91ea07fb8ff7822fb7a03227cd7139e5e54db0e

8 years agoMerge changes I297e2f20,Ia2e475e4,I9f1b4139
Hiroshi Yamauchi [Fri, 1 Apr 2016 20:53:55 +0000 (20:53 +0000)]
Merge changes I297e2f20,Ia2e475e4,I9f1b4139

* changes:
  Revert "Revert "Revert "ART: Improve JitProfile perf in arm/arm64 mterp"""
  Revert "ART: Improve JitProfile perf in x86 mterp"
  Revert "ART: Improve JitProfile perf in x86_64 mterp"

8 years agoRevert "Revert "Revert "ART: Improve JitProfile perf in arm/arm64 mterp"""
Hiroshi Yamauchi [Fri, 1 Apr 2016 19:02:58 +0000 (12:02 -0700)]
Revert "Revert "Revert "ART: Improve JitProfile perf in arm/arm64 mterp"""

This reverts commit 4a8ac9cee4312ac910fabf31c64d28d4c8362836.

570-checker-osr intermittently failing.

Bug: 27939339

8 years agoRevert "ART: Improve JitProfile perf in x86 mterp"
Hiroshi Yamauchi [Fri, 1 Apr 2016 19:01:48 +0000 (12:01 -0700)]
Revert "ART: Improve JitProfile perf in x86 mterp"

This reverts commit 3e9edd1c63c1760f1bcffdbeaf721ebe3320f386.

570-checker-osr intermittently failing.

Bug: 27939339

8 years agoRevert "ART: Improve JitProfile perf in x86_64 mterp"
Hiroshi Yamauchi [Fri, 1 Apr 2016 19:00:10 +0000 (12:00 -0700)]
Revert "ART: Improve JitProfile perf in x86_64 mterp"

This reverts commit 099a611a418df6f0695e3bcd32fe896043ca1398.

570-checker-osr intermittently failing.

Bug: 27939339
Change-Id: I9f1b4139118b1d803ea9c21319c3147d2f40fec9

8 years agoMerge "Make test 146 run with secondary dex file in classpath."
Alex Light [Fri, 1 Apr 2016 18:42:25 +0000 (18:42 +0000)]
Merge "Make test 146 run with secondary dex file in classpath."

8 years agoMake test 146 run with secondary dex file in classpath.
Alex Light [Fri, 1 Apr 2016 18:40:16 +0000 (11:40 -0700)]
Make test 146 run with secondary dex file in classpath.

Bug: 27931085
Change-Id: Ie0b15d5fc94c2bb6d3707ecf985f6f72110e4d68

8 years agoImprove debugging in art/compiler/optimizing/common_arm64.h.
Roland Levillain [Fri, 1 Apr 2016 17:37:46 +0000 (18:37 +0100)]
Improve debugging in art/compiler/optimizing/common_arm64.h.

Change-Id: I44ff2cb64c1fd45390ed4a6517af2488fdbdaf41

8 years agoMerge "Make InvokeInterfaceTrampoline check methods"
Alex Light [Fri, 1 Apr 2016 17:30:41 +0000 (17:30 +0000)]
Merge "Make InvokeInterfaceTrampoline check methods"

8 years agoFix BitCount intrinsics assertions.
Roland Levillain [Fri, 1 Apr 2016 17:21:55 +0000 (18:21 +0100)]
Fix BitCount intrinsics assertions.

Bug: 27852035
Change-Id: Iba43039aadd9ba288b476d53cc2306a58356465f

8 years agoMerge "ART: Better support for arraycopy in unstarted runtime"
Andreas Gampe [Fri, 1 Apr 2016 16:51:47 +0000 (16:51 +0000)]
Merge "ART: Better support for arraycopy in unstarted runtime"

8 years agoMake InvokeInterfaceTrampoline check methods
Alex Light [Thu, 31 Mar 2016 21:34:33 +0000 (14:34 -0700)]
Make InvokeInterfaceTrampoline check methods

InvokeInterfaceTrampoline was causing problems by looking into the
ImtConflictTable of non-imt-conflict-methods. This makes it check for
that before doing so.

Bug: 27931085

Change-Id: I993178a371f8f46535a752e5c4d46d74777cefaf

8 years agoMerge "Fix thumb bit in the oatdump symbolizer."
David Srbecky [Fri, 1 Apr 2016 16:46:32 +0000 (16:46 +0000)]
Merge "Fix thumb bit in the oatdump symbolizer."

8 years agoART: Better support for arraycopy in unstarted runtime
Andreas Gampe [Thu, 31 Mar 2016 20:30:53 +0000 (13:30 -0700)]
ART: Better support for arraycopy in unstarted runtime

Extend the System.arraycopy() cutout in the unstarted runtime
to support arrays with differing component types.

Add tests.

Bug: 27805718
Change-Id: Iaacd95a372e9bfa26e9055a06b0d8f0335b8d6d1

8 years agoUse bit-level packing for InlineInfo in stackmaps as well.
David Srbecky [Thu, 25 Feb 2016 21:55:03 +0000 (21:55 +0000)]
Use bit-level packing for InlineInfo in stackmaps as well.

Use the same approach as we do for stackmaps to reduce the size.

It saves 4.0 MB from non-debuggable boot.oat (AOSP).
It does not affect debuggable boot.oat.

It saves 3.6 MB (of 96.6 MB) from /system/framework/arm/ (GOOG).
It saves 0.6 MB (of 26.7 MB) from /system/framework/oat/arm/ (GOOG).

Field loads from inline-info get around 5% slower.
(based on the time it takes to load all inline-infos from boot.oat)

Change-Id: I67b0fa5eef74c1fdb013680d0231fd44ea696176

8 years agoMerge "Fix a test info referencing AMD64 instead of ARM64."
Vladimir Marko [Fri, 1 Apr 2016 14:40:08 +0000 (14:40 +0000)]
Merge "Fix a test info referencing AMD64 instead of ARM64."

8 years agoFix a test info referencing AMD64 instead of ARM64.
Vladimir Marko [Fri, 1 Apr 2016 13:35:01 +0000 (14:35 +0100)]
Fix a test info referencing AMD64 instead of ARM64.

Change-Id: I195551b984c9ad1dce2bbf978c78831c5b98c803

8 years agoMerge "Pack stack map entries on bit level to save space."
Calin Juravle [Fri, 1 Apr 2016 12:28:51 +0000 (12:28 +0000)]
Merge "Pack stack map entries on bit level to save space."

8 years agoMerge "unstarted_runtime: Update references to decimal format classes."
Narayan Kamath [Fri, 1 Apr 2016 11:26:45 +0000 (11:26 +0000)]
Merge "unstarted_runtime: Update references to decimal format classes."

8 years agounstarted_runtime: Update references to decimal format classes.
Narayan Kamath [Wed, 30 Mar 2016 14:41:54 +0000 (15:41 +0100)]
unstarted_runtime: Update references to decimal format classes.

We had a few special cases for thread local buffers and instances
created by our formatters / parsers. Update them to point to the
OpenJdk equivalents of the same classes.

bug: 27265238
Change-Id: Ida36a89bf414aeabea64b25004e193c770f906c2

8 years agoMerge "unstarted_runtime: Add additional cutouts"
Narayan Kamath [Fri, 1 Apr 2016 11:25:58 +0000 (11:25 +0000)]
Merge "unstarted_runtime: Add additional cutouts"

8 years agounstarted_runtime: Add additional cutouts
Narayan Kamath [Wed, 30 Mar 2016 12:11:18 +0000 (13:11 +0100)]
unstarted_runtime: Add additional cutouts

These allow us to initialize a further 90 classes. getIntVolatile
was used by the static initializer for ThreadLocal, which is used
quite frequently. This class was compile time initializable for M.

java.lang.System.getSecurityManager : 5 classes
sun.misc.Unsafe.getIntVolatile : 85 classes

bug: 27265238
Change-Id: I7e9820112bc87aec47c9b1b40ec6ba4f56172916

8 years agoImplement ARM64 support for SystemArrayCopy()
donghui.bai [Thu, 10 Mar 2016 06:02:55 +0000 (14:02 +0800)]
Implement ARM64 support for SystemArrayCopy()

Change-Id: Iaaf6a9e8c1bea17fa9fde768869bcd4442ebc7d9

8 years agoMerge "ART: Actually run tests cross arch for oatdump"
Andreas Gampe [Fri, 1 Apr 2016 01:13:17 +0000 (01:13 +0000)]
Merge "ART: Actually run tests cross arch for oatdump"

8 years agoMerge changes from topic 'image_verify'
Mathieu Chartier [Thu, 31 Mar 2016 23:03:07 +0000 (23:03 +0000)]
Merge changes from topic 'image_verify'

* changes:
  Disable checks for b/27493510
  Add verification for image ArtMethod declaring classes

8 years agoMerge "ART: Do not run verification under lock"
Andreas Gampe [Thu, 31 Mar 2016 22:55:31 +0000 (22:55 +0000)]
Merge "ART: Do not run verification under lock"

8 years agoDisable checks for b/27493510
Mathieu Chartier [Thu, 31 Mar 2016 22:05:45 +0000 (15:05 -0700)]
Disable checks for b/27493510

Disabled for performance now that the bug is fixed. Still enabled
for debug builds.

Bug: 27493510

(cherry picked from commit 8790c7f9f7d684bff40d5731b2d8e3462ef8ed90)

Change-Id: I049d4319bca1114bd11a35bb13fc9d80c993260c

8 years agoART: Do not run verification under lock
Andreas Gampe [Thu, 31 Mar 2016 02:52:58 +0000 (19:52 -0700)]
ART: Do not run verification under lock

Do not hold the object lock for the duration of the verification.
Instead, use the kStatusVerifying indicator to wait, similar to
resolution.

Bug: 27924355
Change-Id: Ie831f47dd830756a1b7002ca9c792f8ff67570dc

8 years agoAdd verification for image ArtMethod declaring classes
Mathieu Chartier [Tue, 8 Mar 2016 00:13:54 +0000 (16:13 -0800)]
Add verification for image ArtMethod declaring classes

Try and see if any ArtMethods in the image have declaring classes
that are not marked in the heap live bitmap. Trying to mark these
would cause a SIGSEGV since the bitmap is read only.

TODO: Disable for performance.

(cherry-picked from commit 03c1dd9858962071bf3f0c988abc39e0e6ed7a1e)

Bug: 27493510
Change-Id: I231b07b84d98dcd7ee1557c18956c86264058ee1

8 years agoMerge "ART: Add support for DexFile compiler filter function"
Andreas Gampe [Thu, 31 Mar 2016 22:52:51 +0000 (22:52 +0000)]
Merge "ART: Add support for DexFile compiler filter function"

8 years agoMerge "Remove obsolete ANDROID_COMPILE_WITH_JACK"
Ying Wang [Thu, 31 Mar 2016 22:05:04 +0000 (22:05 +0000)]
Merge "Remove obsolete ANDROID_COMPILE_WITH_JACK"

8 years agoMerge "Add way to disable resolving for stack walk"
Mathieu Chartier [Thu, 31 Mar 2016 21:59:35 +0000 (21:59 +0000)]
Merge "Add way to disable resolving for stack walk"

8 years agoAdd way to disable resolving for stack walk
Mathieu Chartier [Thu, 31 Mar 2016 18:07:09 +0000 (11:07 -0700)]
Add way to disable resolving for stack walk

Only occurs for walking the inlined frames case.

Bug: 27857910

(cherry picked from commit be2892bf771435eb451c74297b553f7dc9fea4ca)

Change-Id: Ied92be96f2f74ba02c02168f704443b95e7a4b04

8 years agoPack stack map entries on bit level to save space.
David Srbecky [Fri, 12 Feb 2016 21:58:43 +0000 (21:58 +0000)]
Pack stack map entries on bit level to save space.

Use only the minimum number of bits required to store stack map data.
For example, if native_pc needs 5 bits and dex_pc needs 3 bits, they
will share the first byte of the stack map entry.

The header is changed to store bit offsets of the fields rather than
byte sizes. Offsets also make it easier to access later fields without
calculating sum of all previous sizes.

All of the header fields are byte sized or encoded as ULEB128 instead
of the previous fixed size encoding. This shrinks it by about half.

It saves 3.6 MB from non-debuggable boot.oat (AOSP).
It saves 3.1 MB from debuggable boot.oat (AOSP).

It saves 2.8 MB (of 99.4 MB) from /system/framework/arm/ (GOOG).
It saves 1.0 MB (of 27.8 MB) from /system/framework/oat/arm/ (GOOG).

Field loads from stackmaps seem to get around 10% faster.
(based on the time it takes to load all stackmap entries from boot.oat)

Bug: 27640410
Change-Id: I8bf0996b4eb24300c1b0dfc6e9d99fe85d04a1b7

8 years agoMerge "Obtain stack trace outside of critical section"
Mathieu Chartier [Thu, 31 Mar 2016 21:05:30 +0000 (21:05 +0000)]
Merge "Obtain stack trace outside of critical section"

8 years agoObtain stack trace outside of critical section
Mathieu Chartier [Tue, 29 Mar 2016 21:02:55 +0000 (14:02 -0700)]
Obtain stack trace outside of critical section

Fixes deadlock if the stack walk does allocations, changed stack
trace format to prevent slowdown.

Added missing GetInterfaceMethodIfProxy to fix a crash in maps.

Bug: 27857910

(cherry picked from commit 23428587d32361736d4c5e0ba7270c7602695a43)

Change-Id: I64373bcd87a68fdd1b58fb855db2b16c9f6ed36b

8 years agoMerge "Refine OatFileAssistant.MakeUpToDate failure result."
Richard Uhler [Thu, 31 Mar 2016 19:27:54 +0000 (19:27 +0000)]
Merge "Refine OatFileAssistant.MakeUpToDate failure result."