OSDN Git Service

android-x86/art.git
8 years agoART: Make dex2oat timing a bit more granular
Andreas Gampe [Wed, 16 Sep 2015 03:04:54 +0000 (20:04 -0700)]
ART: Make dex2oat timing a bit more granular

Add scoped timing for runtime creation and dex file opening. Allows
comparing (JIT) compile time to specific overhead inherent to all
compilations.

Bug: 24103765
Change-Id: I1f83daa7015745bffa0cec3a3357b045c8493d6a

8 years agoMerge "Support X86 intrinsic System.arraycopy char"
Andreas Gampe [Wed, 16 Sep 2015 02:58:04 +0000 (02:58 +0000)]
Merge "Support X86 intrinsic System.arraycopy char"

8 years agoMerge "Add X86 bsf and rotate instructions"
Andreas Gampe [Wed, 16 Sep 2015 02:56:18 +0000 (02:56 +0000)]
Merge "Add X86 bsf and rotate instructions"

8 years agoAdd X86 bsf and rotate instructions
Mark Mendell [Wed, 16 Sep 2015 01:45:01 +0000 (21:45 -0400)]
Add X86 bsf and rotate instructions

These are for use in new intrinsics.  Bsf (Bit Scan Forward) is used in
{Long,Integer}NumberOfTrailingZeros and the rotates are used in
{Long,Integer}Rotate{Left,Right}.

Change-Id: Icb599d7e1eec4e4ea9e5b4f0b1654c7b8d4de678
Signed-off-by: Mark Mendell <mark.p.mendell@intel.com>
8 years agoMerge "Expect null referent in DequeuePendingReference()."
Hiroshi Yamauchi [Tue, 15 Sep 2015 17:54:44 +0000 (17:54 +0000)]
Merge "Expect null referent in DequeuePendingReference()."

8 years agoExpect null referent in DequeuePendingReference().
Hiroshi Yamauchi [Mon, 14 Sep 2015 22:10:50 +0000 (15:10 -0700)]
Expect null referent in DequeuePendingReference().

Following up on CL 170735.

It's possible that the referent may potentially be cleared which would
cause a check failure. Avoid that.

Bug: 12687968
Bug: 23896462
Change-Id: I8ccc5936b61ceacf250624681e65307f23ce0405

8 years agoMerge "Use image pointer size for profile info"
Mathieu Chartier [Tue, 15 Sep 2015 17:03:55 +0000 (17:03 +0000)]
Merge "Use image pointer size for profile info"

8 years agoUse image pointer size for profile info
Mathieu Chartier [Tue, 15 Sep 2015 01:50:08 +0000 (18:50 -0700)]
Use image pointer size for profile info

May fix some random crashes in dex2oat due to cross compilation.

Change-Id: I633652500e8c7dfec38044dffd07eb467973d82a

8 years agoMerge "Rename parameter in ArtMethod::ToNativeQuickPc"
David Brazdil [Tue, 15 Sep 2015 17:01:47 +0000 (17:01 +0000)]
Merge "Rename parameter in ArtMethod::ToNativeQuickPc"

8 years agoRename parameter in ArtMethod::ToNativeQuickPc
David Brazdil [Tue, 15 Sep 2015 16:57:31 +0000 (17:57 +0100)]
Rename parameter in ArtMethod::ToNativeQuickPc

Addressing a nit for CL Idb2b34aabf1ac7249c30a00806af7d63d7e682dd.

Change-Id: I1112312b90aa6123e7e9ca8f67132dd7cc57a62f

8 years agoMerge "Cleanup thread access in StackVisitor"
Sebastien Hertz [Tue, 15 Sep 2015 16:54:02 +0000 (16:54 +0000)]
Merge "Cleanup thread access in StackVisitor"

8 years agoMerge "ART: Fix 004-ReferenceMap run test"
David Brazdil [Tue, 15 Sep 2015 16:52:02 +0000 (16:52 +0000)]
Merge "ART: Fix 004-ReferenceMap run test"

8 years agoART: Fix 004-ReferenceMap run test
David Brazdil [Tue, 15 Sep 2015 16:00:52 +0000 (17:00 +0100)]
ART: Fix 004-ReferenceMap run test

This patch adds a new option to ArtMethod::ToNativeQuickPc to select
the order of iteration over stack maps. The method is only used by
the runtime to find native_pc of catch blocks, but also by the
004-ReferenceMap test which uses it to find native_pc of a safepoint.

Change-Id: Idb2b34aabf1ac7249c30a00806af7d63d7e682dd

8 years agoMerge "Use $ for f registers to please clang."
Nicolas Geoffray [Tue, 15 Sep 2015 15:59:53 +0000 (15:59 +0000)]
Merge "Use $ for f registers to please clang."

8 years agoUse $ for f registers to please clang.
Nicolas Geoffray [Tue, 15 Sep 2015 15:36:50 +0000 (16:36 +0100)]
Use $ for f registers to please clang.

Change-Id: I723ce53e344ee98334a556fbe231d8588fa7d1b6

8 years agoMerge "Revert "Revert "ART: Register allocation and runtime support for try/catch"""
David Brazdil [Tue, 15 Sep 2015 12:34:35 +0000 (12:34 +0000)]
Merge "Revert "Revert "ART: Register allocation and runtime support for try/catch"""

8 years agoRevert "Revert "ART: Register allocation and runtime support for try/catch""
David Brazdil [Tue, 15 Sep 2015 12:34:04 +0000 (12:34 +0000)]
Revert "Revert "ART: Register allocation and runtime support for try/catch""

The original CL triggered b/24084144 which has been fixed
by Ib72e12a018437c404e82f7ad414554c66a4c6f8c.

This reverts commit 659562aaf133c41b8d90ec9216c07646f0f14362.

Change-Id: Id8980436172457d0fcb276349c4405f7c4110a55

8 years agoMerge "ART: Fix bug in reference type propagation"
David Brazdil [Tue, 15 Sep 2015 11:11:49 +0000 (11:11 +0000)]
Merge "ART: Fix bug in reference type propagation"

8 years agoMerge "Temporarly disable SSA creation in try/catch/debuggable."
Nicolas Geoffray [Tue, 15 Sep 2015 11:01:11 +0000 (11:01 +0000)]
Merge "Temporarly disable SSA creation in try/catch/debuggable."

8 years agoART: Fix bug in reference type propagation
David Brazdil [Tue, 15 Sep 2015 10:36:54 +0000 (11:36 +0100)]
ART: Fix bug in reference type propagation

Reference type propagation assumed that type is exact if the class is
final. This does not hold for arrays which are always final and the
component type needs to be considered.

Bug: 24084144
Change-Id: Ib72e12a018437c404e82f7ad414554c66a4c6f8c

8 years agoTemporarly disable SSA creation in try/catch/debuggable.
Nicolas Geoffray [Tue, 15 Sep 2015 10:01:01 +0000 (11:01 +0100)]
Temporarly disable SSA creation in try/catch/debuggable.

This is to enable optimizations when the debuggable flag is set.

bug:24054676

Change-Id: I34eef2481ae44ad3550e04ec6f796d421d03adc0

8 years agoCleanup thread access in StackVisitor
Sebastien Hertz [Tue, 15 Sep 2015 07:52:07 +0000 (09:52 +0200)]
Cleanup thread access in StackVisitor

Adds method StackVisitor::GetThread to give access to the visited
Thread* so we no longer need to copy that pointer in subclasses.

Also adds a few missing const and DISALLOW_COPY_AND_ASSIGN.

Change-Id: I57649ee7742ef4ef1e01447ac2fbb66f977b22eb

8 years agoMerge "Revert "Revert "Re-enable Clang's -Wpessimizing-move"""
Pirama Arumuga Nainar [Mon, 14 Sep 2015 21:47:03 +0000 (21:47 +0000)]
Merge "Revert "Revert "Re-enable Clang's -Wpessimizing-move"""

8 years agoMerge "Revert "ART: Register allocation and runtime support for try/catch""
David Brazdil [Mon, 14 Sep 2015 21:27:03 +0000 (21:27 +0000)]
Merge "Revert "ART: Register allocation and runtime support for try/catch""

8 years agoRevert "ART: Register allocation and runtime support for try/catch"
David Brazdil [Mon, 14 Sep 2015 21:26:33 +0000 (21:26 +0000)]
Revert "ART: Register allocation and runtime support for try/catch"

Breaks libcore test org.apache.harmony.security.tests.java.security.KeyStorePrivateKeyEntryTest#testGetCertificateChain. Need to investigate.

This reverts commit b022fa1300e6d78639b3b910af0cf85c43df44bb.

Change-Id: Ib24d3a80064d963d273e557a93469c95f37b1f6f

8 years agoMerge "Convert ART benchmarks into caliper benchmarks"
Mathieu Chartier [Mon, 14 Sep 2015 21:02:07 +0000 (21:02 +0000)]
Merge "Convert ART benchmarks into caliper benchmarks"

8 years agoMerge "ART: Fix mac build"
Andreas Gampe [Mon, 14 Sep 2015 21:01:30 +0000 (21:01 +0000)]
Merge "ART: Fix mac build"

8 years agoART: Fix mac build
Andreas Gampe [Mon, 14 Sep 2015 20:59:51 +0000 (13:59 -0700)]
ART: Fix mac build

Parameter is unused in the mac build. We now build the test libraries
in checkbuild, which exposed the issue.

Change-Id: Ib246d87927740e2e65a63111bdf0cc4f261b4c1c

8 years agoConvert ART benchmarks into caliper benchmarks
Mathieu Chartier [Sat, 12 Sep 2015 23:55:26 +0000 (16:55 -0700)]
Convert ART benchmarks into caliper benchmarks

Move the ART benchmarks into the benchmarks directory and make them
caliper friendly.

To run you need to build libartbenchmark.so and sync it then use
vogar --benchmark.

Bug: 24055147

Change-Id: Idc9caf8dc8022190d737840ba0eb9e356f9872cb

8 years agoRevert "Revert "Re-enable Clang's -Wpessimizing-move""
Pirama Arumuga Nainar [Mon, 14 Sep 2015 20:39:53 +0000 (20:39 +0000)]
Revert "Revert "Re-enable Clang's -Wpessimizing-move""

This reverts commit 4a37953377092ade93c15c72ad49cde4adaabd94.

Change-Id: Iae3bab9b104c1391070b329ef6a6a73b836189c8

8 years agoMerge "ART: Build tests in checkbuild"
Andreas Gampe [Mon, 14 Sep 2015 20:37:48 +0000 (20:37 +0000)]
Merge "ART: Build tests in checkbuild"

8 years agoMerge "ART: Register allocation and runtime support for try/catch"
David Brazdil [Mon, 14 Sep 2015 19:44:35 +0000 (19:44 +0000)]
Merge "ART: Register allocation and runtime support for try/catch"

8 years agoART: Register allocation and runtime support for try/catch
David Brazdil [Thu, 20 Aug 2015 16:47:48 +0000 (17:47 +0100)]
ART: Register allocation and runtime support for try/catch

This patch completes a series of CLs that add support for try/catch
in the Optimizing compiler. With it, Optimizing can compile all
methods containing try/catch, provided they don't contain catch loops.
Future work will focus on improving performance of the generated code.

SsaLivenessAnalysis was updated to propagate liveness information of
instructions live at catch blocks, and to keep location information on
instructions which may be caught by catch phis.

RegisterAllocator was extended to spill values used after catch, and
to allocate spill slots for catch phis. Catch phis generated for the
same vreg share a spill slot as the raw value must be the same.

Location builders and slow paths were updated to reflect the fact that
throwing an exception may not lead to escaping the method.

Instruction code generators are forbidden from using of implicit null
checks in try blocks as live registers need to be saved before handing
over to the runtime.

CodeGenerator emits a stack map for each catch block, storing locations
of catch phis. CodeInfo and StackMapStream recognize this new type of
stack map and store them separate from other stack maps to avoid dex_pc
conflicts.

After having found the target catch block to deliver an exception to,
QuickExceptionHandler looks up the dex register maps at the throwing
instruction and the catch block and copies the values over to their
respective locations.

The runtime-support approach was selected because it allows for the
best performance in the normal control-flow path, since no propagation
of catch phi values is necessary until the exception is thrown. In
addition, it also greatly simplifies the register allocation phase.

ConstantHoisting was removed from LICMTest because it instantiated
(now abstract) HConstant and was bogus anyway (constants are always in
the entry block).

Change-Id: Ie31038ad8e3ee0c13a5bbbbaf5f0b3e532310e4e

8 years agoART: Build tests in checkbuild
Andreas Gampe [Mon, 14 Sep 2015 18:18:12 +0000 (11:18 -0700)]
ART: Build tests in checkbuild

We should build the tests when a checkbuild is run. That is the
standard configuration to test that everything compiles.

Bug: 24059936
Change-Id: I5a72d9a1b7accdd3e0dcb14edaeccde7ec126d44

8 years agoMerge "Remove unnecessary std::move from test_dex_file_builder"
Pirama Arumuga Nainar [Mon, 14 Sep 2015 18:04:34 +0000 (18:04 +0000)]
Merge "Remove unnecessary std::move from test_dex_file_builder"

8 years agoRemove unnecessary std::move from test_dex_file_builder
Pirama Arumuga Nainar [Mon, 14 Sep 2015 18:00:16 +0000 (11:00 -0700)]
Remove unnecessary std::move from test_dex_file_builder

This stops Clang from warning about -Wpessimizing-move.

Change-Id: Id40acf1c398c615faf6486ef700df6975a5f013f

8 years agoMerge "Fix the DequeuePendingReference crash."
Hiroshi Yamauchi [Mon, 14 Sep 2015 17:36:34 +0000 (17:36 +0000)]
Merge "Fix the DequeuePendingReference crash."

8 years agoFix the DequeuePendingReference crash.
Hiroshi Yamauchi [Thu, 10 Sep 2015 23:01:30 +0000 (16:01 -0700)]
Fix the DequeuePendingReference crash.

In DequeuePendingReference, acknowledge a black/white Reference object
in the queue if its referent was marked right after it's enqueued.

Bug: 12687968
Bug: 23896462
Change-Id: I33c802e04e1688a54a70ad3935628e3853c46e44

8 years agoMerge "Revert "Re-enable Clang's -Wpessimizing-move""
Pirama Arumuga Nainar [Mon, 14 Sep 2015 16:50:58 +0000 (16:50 +0000)]
Merge "Revert "Re-enable Clang's -Wpessimizing-move""

8 years agoRevert "Re-enable Clang's -Wpessimizing-move"
Pirama Arumuga Nainar [Mon, 14 Sep 2015 16:50:30 +0000 (16:50 +0000)]
Revert "Re-enable Clang's -Wpessimizing-move"

This reverts commit 8f5d2d84efb90e7220dd42c7aa104dbc19c52f4e.

Change-Id: I24fa681c0639bc4bf7af1b821f620780cd7a474d

8 years agoMerge "Re-enable Clang's -Wpessimizing-move"
Pirama Arumuga Nainar [Mon, 14 Sep 2015 16:21:09 +0000 (16:21 +0000)]
Merge "Re-enable Clang's -Wpessimizing-move"

8 years agoMerge "Be very patient with the emulator."
Nicolas Geoffray [Mon, 14 Sep 2015 13:47:18 +0000 (13:47 +0000)]
Merge "Be very patient with the emulator."

8 years agoMerge "Optimizing: Tag basic block allocations with their source."
Vladimir Marko [Mon, 14 Sep 2015 11:43:25 +0000 (11:43 +0000)]
Merge "Optimizing: Tag basic block allocations with their source."

8 years agoMerge "ART: Add some utilities for working with containers."
Vladimir Marko [Mon, 14 Sep 2015 10:42:01 +0000 (10:42 +0000)]
Merge "ART: Add some utilities for working with containers."

8 years agoRe-enable Clang's -Wpessimizing-move
Pirama Arumuga Nainar [Sun, 13 Sep 2015 05:35:30 +0000 (22:35 -0700)]
Re-enable Clang's -Wpessimizing-move

The warning on variant_map.h is fixed by
https://android-review.googlesource.com/#/c/170745/.

Change-Id: I802dd26538637ff276b7c24c1f5bca280b0bc428

8 years agoMerge "Remove unnecessary std::move from variant_map.h"
Pirama Arumuga Nainar [Sun, 13 Sep 2015 02:31:32 +0000 (02:31 +0000)]
Merge "Remove unnecessary std::move from variant_map.h"

8 years agoRemove unnecessary std::move from variant_map.h
Pirama Arumuga Nainar [Sat, 12 Sep 2015 23:04:42 +0000 (16:04 -0700)]
Remove unnecessary std::move from variant_map.h

This triggers Clang's pessimizing-move warning.

Change-Id: Icb33643f28d1d159a87a17235c4d210209199dac

8 years agoMerge "Add and use ScopedSuspendAll"
Mathieu Chartier [Fri, 11 Sep 2015 21:20:54 +0000 (21:20 +0000)]
Merge "Add and use ScopedSuspendAll"

8 years agoMerge "Add benchmark for ScopedPrimitiveArray"
Mathieu Chartier [Fri, 11 Sep 2015 20:43:10 +0000 (20:43 +0000)]
Merge "Add benchmark for ScopedPrimitiveArray"

8 years agoAdd and use ScopedSuspendAll
Mathieu Chartier [Fri, 4 Sep 2015 20:26:21 +0000 (13:26 -0700)]
Add and use ScopedSuspendAll

Usage replaces most SuspendAll and ResumeAll calls.

Change-Id: I355683a5365876242cea85a656dcb58455f7a294

8 years agoAdd benchmark for ScopedPrimitiveArray
Mathieu Chartier [Fri, 4 Sep 2015 23:38:55 +0000 (16:38 -0700)]
Add benchmark for ScopedPrimitiveArray

TODO: Make this into a caliper benchmark.

Bug: 19664826
Change-Id: If1484a68cc62603f1d216b1662ca55cceac1d619

8 years agoMerge "ART: Fix Quick-style LR vs PC core spill mask bug"
Andreas Gampe [Fri, 11 Sep 2015 17:50:01 +0000 (17:50 +0000)]
Merge "ART: Fix Quick-style LR vs PC core spill mask bug"

8 years agoMerge "Support deoptimization on exception"
Sebastien Hertz [Fri, 11 Sep 2015 15:40:48 +0000 (15:40 +0000)]
Merge "Support deoptimization on exception"

8 years agoMerge "Record profiling information before Jitting."
Nicolas Geoffray [Fri, 11 Sep 2015 14:13:08 +0000 (14:13 +0000)]
Merge "Record profiling information before Jitting."

8 years agoRecord profiling information before Jitting.
Nicolas Geoffray [Fri, 21 Aug 2015 17:38:30 +0000 (18:38 +0100)]
Record profiling information before Jitting.

- Add a new instrumentation kind to record dynamic invokes.
- Use the JNI entry point field to store the profiling data.
- Record seen receivers for every dynamic invoke.

Change-Id: I2c1738ab2a72052d45964d055dc16b44b906e54c

8 years agoMerge "ARM/ARM64: Intrinsics - numberOfTrailingZeros, rotateLeft, rotateRight"
Vladimir Marko [Fri, 11 Sep 2015 13:20:10 +0000 (13:20 +0000)]
Merge "ARM/ARM64: Intrinsics - numberOfTrailingZeros, rotateLeft, rotateRight"

8 years agoBe very patient with the emulator.
Nicolas Geoffray [Fri, 11 Sep 2015 11:08:49 +0000 (12:08 +0100)]
Be very patient with the emulator.

Change-Id: I9f1f3a5d825fea85115b3331da317eefd3dbe17b

8 years agoARM/ARM64: Intrinsics - numberOfTrailingZeros, rotateLeft, rotateRight
Scott Wakeling [Thu, 23 Jul 2015 09:44:35 +0000 (10:44 +0100)]
ARM/ARM64: Intrinsics - numberOfTrailingZeros, rotateLeft, rotateRight

Change-Id: I2a07c279756ee804fb7c129416bdc4a3962e93ed

8 years agoMerge "ART: Narrow parameter types."
Andreas Gampe [Fri, 11 Sep 2015 03:10:58 +0000 (03:10 +0000)]
Merge "ART: Narrow parameter types."

8 years agoART: Narrow parameter types.
Andreas Gampe [Fri, 11 Sep 2015 03:09:11 +0000 (20:09 -0700)]
ART: Narrow parameter types.

Follow-up to e6215c0ec4b1bb71b722fdbf7e62eaf3be8a91d5. Those should
have been uint16_t.

Change-Id: I2cc13701eb937dad488bcf405a9a7b26eabb656e

8 years agoMerge "ART: Move access flags checking to dex file verifier"
Andreas Gampe [Fri, 11 Sep 2015 01:35:46 +0000 (01:35 +0000)]
Merge "ART: Move access flags checking to dex file verifier"

8 years agoART: Move access flags checking to dex file verifier
Andreas Gampe [Tue, 1 Sep 2015 01:54:38 +0000 (18:54 -0700)]
ART: Move access flags checking to dex file verifier

Actually implement all the access flags checking in the dex file
verifier. Add tests.

Change-Id: I8b797357831b588589d56d6e2e22f7b410f33008

8 years agoART: Fix Quick-style LR vs PC core spill mask bug
Andreas Gampe [Thu, 10 Sep 2015 23:11:06 +0000 (16:11 -0700)]
ART: Fix Quick-style LR vs PC core spill mask bug

It's always been a bug that Quick marked PC as spilled instead of
LR. The root cause was a mutation of the spill mask at frame exit,
when LR is being restored into PC to return. A local should have
been used to keep the actual spill mask safe and sound.

This has only worked because nobody ever uses LR, even after long
jumps for exception dispatch. However, single-frame deoptimization
needs this to work, and I'd rather fix this than being forced to
have machine-specific fixups.

Also fix in optimizing, and bump the oat version.

Change-Id: Ib032a533408bf464097fc96dcbfc5b6a68bf59a1

8 years agoMerge "Increase the run-test timeout for the read barrier config."
Hiroshi Yamauchi [Thu, 10 Sep 2015 23:32:47 +0000 (23:32 +0000)]
Merge "Increase the run-test timeout for the read barrier config."

8 years agoIncrease the run-test timeout for the read barrier config.
Hiroshi Yamauchi [Thu, 10 Sep 2015 01:59:42 +0000 (18:59 -0700)]
Increase the run-test timeout for the read barrier config.

ThreadStress sometimes times out without hanging.

Bug: 23485091
Change-Id: Id6833c7887ece74aebafc6e1d5a8c772d7e0ae62

8 years agoMerge "Induction variable range analysis."
Aart Bik [Thu, 10 Sep 2015 17:33:44 +0000 (17:33 +0000)]
Merge "Induction variable range analysis."

8 years agoInduction variable range analysis.
Aart Bik [Tue, 8 Sep 2015 22:25:15 +0000 (15:25 -0700)]
Induction variable range analysis.

Rationale: by computing an upper bound on the trip count of each
           loop after induction var analysis has completed, a
   simple range analysis yields lower and upper bounds on
   all induced expressions in a loop; this analysis
   plugs directly into BCE (follow-up CL).

Change-Id: I46a3fe48721ca372547199b39a3498c47992597d

8 years agoSupport deoptimization on exception
Sebastien Hertz [Tue, 8 Sep 2015 15:03:36 +0000 (17:03 +0200)]
Support deoptimization on exception

Allows to deoptimize when an exception is being thrown. We only
deoptimize if an executable frame (starting from the catch handler)
needs to be executed with the interpreter.

Before executing deoptimized frames, the exception is restored. The
interpreter starts by handling this exception at the point of the
throwing instruction.

Bug: 23714835
Change-Id: I0c5f7d4b257644acf12210aae8e5b6bb0f4af1f7

8 years agoMerge "Add extra options to tools/run-jdwp-tests.sh"
Sebastien Hertz [Thu, 10 Sep 2015 10:31:05 +0000 (10:31 +0000)]
Merge "Add extra options to tools/run-jdwp-tests.sh"

8 years agoAdd extra options to tools/run-jdwp-tests.sh
Sebastien Hertz [Thu, 10 Sep 2015 10:03:51 +0000 (12:03 +0200)]
Add extra options to tools/run-jdwp-tests.sh

Adds options for local debugging:
--verbose: passes '-verbose:jdwp' to the debuggee command-line to
           enable JDWP verbose logs.
--test <name>: runs a specific test instead of the whole test
               suite.

Change-Id: Ia04dd26c16c0b6733bb1c8a1b15ec585a24872c8

8 years agoMerge "ART: Add missing GetInterfaceMethodIfProxy"
Andreas Gampe [Wed, 9 Sep 2015 17:15:23 +0000 (17:15 +0000)]
Merge "ART: Add missing GetInterfaceMethodIfProxy"

8 years agoART: Add missing GetInterfaceMethodIfProxy
Andreas Gampe [Wed, 9 Sep 2015 00:42:59 +0000 (17:42 -0700)]
ART: Add missing GetInterfaceMethodIfProxy

Add missing uses of GetInterfaceMethodIfProxy in reflection code.

Add a test case for a JNI call to a proxy method.

Bug: https://code.google.com/p/android-developer-preview/issues/detail?id=2973
Bug: 23886441
Change-Id: I5b66b64b5561fcee15d0314707d67e8abc02ce5b

8 years agoMerge "Revert "Revert "Do a second check for testing intrinsic types."""
Andreas Gampe [Wed, 9 Sep 2015 16:39:16 +0000 (16:39 +0000)]
Merge "Revert "Revert "Do a second check for testing intrinsic types."""

8 years agoMerge "Add dex_pc to all HInstructions in builder."
Nicolas Geoffray [Wed, 9 Sep 2015 13:16:59 +0000 (13:16 +0000)]
Merge "Add dex_pc to all HInstructions in builder."

8 years agoAdd dex_pc to all HInstructions in builder.
Yevgeny Rouban [Mon, 7 Sep 2015 11:57:00 +0000 (17:57 +0600)]
Add dex_pc to all HInstructions in builder.

Optimizing compiler generates minimum debug line info that
is built using the dex_pc information about suspend points.
This is not enough for performance and debugging needs.

This patch makes all HInstructions contain
dex_pc and all allocations in the builder define this value.

Change-Id: I1d14aefe075189b7b1b41b4384c3499474c19afc
Signed-off-by: Yevgeny Rouban <yevgeny.y.rouban@intel.com>
Signed-off-by: Serdjuk, Nikolay Y <nikolay.y.serdjuk@intel.com>
8 years agoMerge "[optimizing] Add support for x86 constant area"
Nicolas Geoffray [Wed, 9 Sep 2015 10:04:41 +0000 (10:04 +0000)]
Merge "[optimizing] Add support for x86 constant area"

8 years agoRevert "Revert "Do a second check for testing intrinsic types.""
Andreas Gampe [Tue, 1 Sep 2015 15:45:02 +0000 (15:45 +0000)]
Revert "Revert "Do a second check for testing intrinsic types.""

This reverts commit a14b9fef395b94fa9a32147862c198fe7c22e3d7.

When an intrinsic with invoke-type virtual is recognized, replace
the instruction with a new HInvokeStaticOrDirect.

Minimal update for dex-cache rework. Fix includes.

Change-Id: I1c8e735a2fa7cda4419f76ca0717125ef236d332

8 years agoMerge "Allow null current thread in jni weak ref decode during shutdown."
Hiroshi Yamauchi [Wed, 9 Sep 2015 01:35:11 +0000 (01:35 +0000)]
Merge "Allow null current thread in jni weak ref decode during shutdown."

8 years agoAllow null current thread in jni weak ref decode during shutdown.
Hiroshi Yamauchi [Wed, 9 Sep 2015 00:50:48 +0000 (17:50 -0700)]
Allow null current thread in jni weak ref decode during shutdown.

Fix a crash during a runtime shutdown for the read barrier config.

Bug: 23897251
Bug: 12687968
Change-Id: Iea0888cb6d052d4becddba289dc0ed121461e97d

8 years ago[optimizing] Add support for x86 constant area
Mark Mendell [Fri, 17 Apr 2015 16:49:27 +0000 (12:49 -0400)]
[optimizing] Add support for x86 constant area

Use the Quick trick of finding the address of the method by calling the
next instruction and popping the return address into a register.  This
trick is used because of the lack of PC-relative addressing in 32 bit
mode on the X86.

Add a HX86ComputeBaseMethodAddress instruction to trigger generation
of the method address, which is referenced by instructions needing
access to the constant area.

Add a HX86LoadFromConstantTable instruction that takes a
HX86ComputeBaseMethodAddress and a HConstant that will be used to load
the value when needed.

Change Add/Sub/Mul/Div to detect a HX86LoadFromConstantTable right hand
side, and generate code that directly references the constant area.
Other uses will be added later.

Change the inputs to HReturn and HInvoke(s), replacing the FP constants
with HX86LoadFromConstantTable instead.  This allows values to be
loaded from the constant area into the right location.

Port the X86_64 assembler constant area handling to the X86.

Use the new per-backend optimization framework to do this conversion.

Change-Id: I6d235a72238262e4f9ec0f3c88319a187f865932
Signed-off-by: Mark Mendell <mark.p.mendell@intel.com>
8 years agoOptimizing: Tag basic block allocations with their source.
Vladimir Marko [Thu, 3 Sep 2015 13:35:12 +0000 (13:35 +0000)]
Optimizing: Tag basic block allocations with their source.

Replace GrowableArray with ArenaVector in HBasicBlock and,
to track the source of allocations, assign one new and two
Quick's arena allocation types to these vectors. Rename
kArenaAllocSuccessor to kArenaAllocSuccessors.

Bug: 23736311
Change-Id: Ib52e51698890675bde61f007fe6039338cf1a025

8 years agoART: Add some utilities for working with containers.
Vladimir Marko [Fri, 4 Sep 2015 11:47:41 +0000 (12:47 +0100)]
ART: Add some utilities for working with containers.

Add utility functions for searching, removing and replacing
existing values in a container, to be used with std::vector
(including the ArenaVector alias) and other containers.

Also move UniqueCPtr<> and MakeUnique() to base/stl_utils.h
and clean up related includes.

Change-Id: I1e61762df91c046866591bda167d42bf8b67a692

8 years agoMerge "Simplify loop invariant operations during induction analysis."
Aart Bik [Tue, 8 Sep 2015 16:21:14 +0000 (16:21 +0000)]
Merge "Simplify loop invariant operations during induction analysis."

8 years agoMerge "ART: Clean up Thumb2Assembler's AddConstant()."
Vladimir Marko [Tue, 8 Sep 2015 13:06:07 +0000 (13:06 +0000)]
Merge "ART: Clean up Thumb2Assembler's AddConstant()."

8 years agoART: Clean up Thumb2Assembler's AddConstant().
Vladimir Marko [Tue, 8 Sep 2015 11:16:45 +0000 (12:16 +0100)]
ART: Clean up Thumb2Assembler's AddConstant().

Change-Id: I6a4c32d1bba79879e5514059df6336dc331246c1

8 years agoSimplify loop invariant operations during induction analysis.
Aart Bik [Sat, 5 Sep 2015 01:22:11 +0000 (18:22 -0700)]
Simplify loop invariant operations during induction analysis.

Rationale:
Saves some memory for nodes that are not really required, and
yields slightly more readable debugging strings.

Change-Id: I95b64b48869699137b5d49e26eb20091e264de7a

8 years agoMerge "Remove framework start from script."
Nicolas Geoffray [Mon, 7 Sep 2015 13:15:23 +0000 (13:15 +0000)]
Merge "Remove framework start from script."

8 years agoRemove framework start from script.
Nicolas Geoffray [Mon, 7 Sep 2015 12:33:36 +0000 (13:33 +0100)]
Remove framework start from script.

It's not needed anymore, as it was a one-off instruction that
needed to be executed by the devices.

Change-Id: I5eadb74aee7a9cc03be71b6e5f53e0cfd8d2352f

8 years agoMerge "Don't stop framework when testing."
Nicolas Geoffray [Mon, 7 Sep 2015 11:40:57 +0000 (11:40 +0000)]
Merge "Don't stop framework when testing."

8 years agoDon't stop framework when testing.
Nicolas Geoffray [Mon, 7 Sep 2015 11:04:35 +0000 (12:04 +0100)]
Don't stop framework when testing.

It prevents our infra team from knowing if a device is responsive.

Change-Id: Ie62df378e6fd540bcbfa2982a67b40a0dce85cb8

8 years agoMerge "Fix the read barrier config build."
Hiroshi Yamauchi [Sat, 5 Sep 2015 00:45:13 +0000 (00:45 +0000)]
Merge "Fix the read barrier config build."

8 years agoFix the read barrier config build.
Hiroshi Yamauchi [Sat, 5 Sep 2015 00:29:20 +0000 (17:29 -0700)]
Fix the read barrier config build.

Change-Id: I580d85609e8fa9b8c65466c6596d65bc3b155330

8 years agoMerge "Further development of induction variable analysis."
Aart Bik [Sat, 5 Sep 2015 00:25:41 +0000 (00:25 +0000)]
Merge "Further development of induction variable analysis."

8 years agoFurther development of induction variable analysis.
Aart Bik [Thu, 27 Aug 2015 20:46:58 +0000 (13:46 -0700)]
Further development of induction variable analysis.

Various improvements:
(1) Introduced period sequences.
(2) Extended all transfer functions to deal with all cases;
    also refactored these to read more compactly.
(3) Improved debugging output for constants for readability.
(4) Used direct pointer in mappings for clarify.
(5) Several induction info "constructors" for readability.
(6) Various other changes suggested in earlier code reviews.

Change-Id: I9d5381f1676b63d30cea6f5e304d4b7bda7acb96

8 years agoMerge "Make class loaders weak roots"
Mathieu Chartier [Fri, 4 Sep 2015 23:43:18 +0000 (23:43 +0000)]
Merge "Make class loaders weak roots"

8 years agoMake class loaders weak roots
Mathieu Chartier [Thu, 3 Sep 2015 01:51:54 +0000 (18:51 -0700)]
Make class loaders weak roots

Making the class loaders weak roots in the class linker prevents them
from keeping the classes as live. However we currently do mark them
as strong roots to make sure no accidental class unloading occurs
until the logic to free from linear alloc is complete.

Bug: 22720414

Change-Id: I57466236d9ce6fd064dda9a30ce8ab68094fb8b0

8 years agoMerge "Increase the run test failure output diff line count."
Hiroshi Yamauchi [Fri, 4 Sep 2015 20:17:31 +0000 (20:17 +0000)]
Merge "Increase the run test failure output diff line count."

8 years agoIncrease the run test failure output diff line count.
Hiroshi Yamauchi [Fri, 4 Sep 2015 19:52:03 +0000 (12:52 -0700)]
Increase the run test failure output diff line count.

If there are a lot of threads and deep stacks, we may not see the
beginning of the crash with the failure message.

Bug: 23485091
Change-Id: I854a05d9d9840f0ba10d6317396eb34e1da58c4d

8 years agoMerge "Fix art_quick_alloc_object_tlab"
Mathieu Chartier [Fri, 4 Sep 2015 18:24:43 +0000 (18:24 +0000)]
Merge "Fix art_quick_alloc_object_tlab"

8 years agoFix art_quick_alloc_object_tlab
Mathieu Chartier [Fri, 4 Sep 2015 17:40:01 +0000 (10:40 -0700)]
Fix art_quick_alloc_object_tlab

Was not updated for the new dex cache arrays.

Change-Id: I47b14fbaa071428abf87b18a045009a1c04d2376