OSDN Git Service

android-x86/art.git
9 years agoMerge "[optimizing compiler] Support x86 hard float ABI"
Nicolas Geoffray [Thu, 29 Jan 2015 11:05:48 +0000 (11:05 +0000)]
Merge "[optimizing compiler] Support x86 hard float ABI"

9 years agoMerge "Small optimization for recursive calls: avoid dex cache."
Nicolas Geoffray [Thu, 29 Jan 2015 10:55:11 +0000 (10:55 +0000)]
Merge "Small optimization for recursive calls: avoid dex cache."

9 years agoSmall optimization for recursive calls: avoid dex cache.
Nicolas Geoffray [Fri, 12 Dec 2014 19:22:03 +0000 (19:22 +0000)]
Small optimization for recursive calls: avoid dex cache.

Change-Id: I044757a2f06e535cdc1480c4fc8182b89635baf6

9 years agoMerge "Clean up pass driver"
Mathieu Chartier [Thu, 29 Jan 2015 01:17:10 +0000 (01:17 +0000)]
Merge "Clean up pass driver"

9 years agoClean up pass driver
Mathieu Chartier [Tue, 27 Jan 2015 02:30:19 +0000 (18:30 -0800)]
Clean up pass driver

Added pass manager to hold the state which used to be in global
variables.

Static variables caused issues with Runtime.exit since they are
destroyed by the global destructors while threads are still
executing.

Bug: 17950037
Change-Id: Ie0e4546dc9e48909c8df996a5c135be682d50044

9 years agoMerge "ART: Arm64 optimizing compiler intrinsics"
Andreas Gampe [Wed, 28 Jan 2015 23:36:25 +0000 (23:36 +0000)]
Merge "ART: Arm64 optimizing compiler intrinsics"

9 years agoART: Arm64 optimizing compiler intrinsics
Andreas Gampe [Fri, 16 Jan 2015 07:24:00 +0000 (23:24 -0800)]
ART: Arm64 optimizing compiler intrinsics

Implement most intrinsics for the optimizing compiler for Arm64.

Change-Id: Idb459be09f0524cb9aeab7a5c7fccb1c6b65a707

9 years agoMerge "Force set resolved method for static invokes"
Mathieu Chartier [Wed, 28 Jan 2015 22:34:48 +0000 (22:34 +0000)]
Merge "Force set resolved method for static invokes"

9 years agoForce set resolved method for static invokes
Mathieu Chartier [Wed, 28 Jan 2015 21:11:44 +0000 (13:11 -0800)]
Force set resolved method for static invokes

For static invokes, we may dispatch to the static method in the
superclass but resolve using the subclass. To prevent getting slow
paths on each invoke, we force set the resolved method for the
super class dex method index if we are in the same dex file.

Added test.

Bug: 19175856
Change-Id: I26f8644a7f725f5c2dc2a94a8e9578f573792507

9 years agoMerge "Print lock level in ExclusiveUnlock"
Mathieu Chartier [Wed, 28 Jan 2015 02:21:29 +0000 (02:21 +0000)]
Merge "Print lock level in ExclusiveUnlock"

9 years agoPrint lock level in ExclusiveUnlock
Mathieu Chartier [Wed, 28 Jan 2015 01:14:16 +0000 (17:14 -0800)]
Print lock level in ExclusiveUnlock

Bug: 18713034
Change-Id: I83e1bf1e2023b595d8ddf20008935c75b2b0aa0c

9 years ago[optimizing compiler] Support x86 hard float ABI
Mark Mendell [Mon, 26 Jan 2015 16:21:33 +0000 (11:21 -0500)]
[optimizing compiler] Support x86 hard float ABI

Add support for the new ABI passing FP parameters in XMM0-XMM3.  This
allows us to optimize for x86 methods that don't use 'long'.

Change-Id: Ic79a24767173451e7d7095ccc2a00b307593a868
Signed-off-by: Mark Mendell <mark.p.mendell@intel.com>
9 years agoMerge "ART: Even more Quick cleanup"
Andreas Gampe [Wed, 28 Jan 2015 00:18:07 +0000 (00:18 +0000)]
Merge "ART: Even more Quick cleanup"

9 years agoMerge "Revert "Revert "ART: Implement X86 hard float (Quick/JNI/Baseline)"""
Andreas Gampe [Tue, 27 Jan 2015 23:57:47 +0000 (23:57 +0000)]
Merge "Revert "Revert "ART: Implement X86 hard float (Quick/JNI/Baseline)"""

9 years agoMerge "ART: Clean up some build flags"
Andreas Gampe [Tue, 27 Jan 2015 23:29:53 +0000 (23:29 +0000)]
Merge "ART: Clean up some build flags"

9 years agoART: Even more Quick cleanup
Andreas Gampe [Tue, 27 Jan 2015 22:31:40 +0000 (14:31 -0800)]
ART: Even more Quick cleanup

Remove Backend.

Change-Id: I247cc65ccda6a362ba1a8f5e73e7f12ecd980a87

9 years agoART: Clean up some build flags
Andreas Gampe [Tue, 27 Jan 2015 22:36:38 +0000 (14:36 -0800)]
ART: Clean up some build flags

Remove some no-longer necessary build flags.

Change-Id: Iea85a21689c20931b88d0e84b84410652b66b38d

9 years agoRevert "Revert "ART: Implement X86 hard float (Quick/JNI/Baseline)""
Mark P Mendell [Tue, 27 Jan 2015 15:45:27 +0000 (15:45 +0000)]
Revert "Revert "ART: Implement X86 hard float (Quick/JNI/Baseline)""

This reverts commit 949c91fb91f40a4a80b2b492913cf8541008975e.

This time, don't clobber EBX before saving it.

Redo some of the macros to make register usage explicit.

Change-Id: I8db8662877cd006816e16a28f42444ab7c36bfef

9 years agoMerge "ART: reduce UpdateModUnionTable time for partial GC"
Hiroshi Yamauchi [Tue, 27 Jan 2015 20:20:55 +0000 (20:20 +0000)]
Merge "ART: reduce UpdateModUnionTable time for partial GC"

9 years agoMerge "ART: clear dirty cards of alloc space for MS/CMS partial and full GCs"
Hiroshi Yamauchi [Tue, 27 Jan 2015 18:54:18 +0000 (18:54 +0000)]
Merge "ART: clear dirty cards of alloc space for MS/CMS partial and full GCs"

9 years agoMerge "Checker: Allow don't-care output on a line"
David Brazdil [Tue, 27 Jan 2015 18:21:06 +0000 (18:21 +0000)]
Merge "Checker: Allow don't-care output on a line"

9 years agoChecker: Allow don't-care output on a line
David Brazdil [Tue, 27 Jan 2015 15:54:30 +0000 (15:54 +0000)]
Checker: Allow don't-care output on a line

This patch changes the behaviour of whitespace characters in CHECK
lines, allowing for additional content between verified parts of the
matched output line. Tests therefore won't need to explicitly match
attributes which are not tested.

The way attributes are printed ensures that the right part of the
line is matched against.

Example:
  - output line:   i32 Div [ i4 i8 ] ( loop_header:null )
  - CHECK before:  Div [ {{i\d+}} {{i\d+}} ] ( loop_header:null )
  - CHECK now:     Div ( loop_header:null )

Change-Id: Icf6bacfb285ae288bea21640e860a871a94cc386

9 years agoMerge "ART: Further refactor use lists"
David Brazdil [Tue, 27 Jan 2015 17:32:30 +0000 (17:32 +0000)]
Merge "ART: Further refactor use lists"

9 years agoART: Further refactor use lists
David Brazdil [Tue, 27 Jan 2015 17:12:29 +0000 (17:12 +0000)]
ART: Further refactor use lists

Change-Id: I9e3219575a508ca5141d851bfcaf848302480c32

9 years agoMerge "run-test: enable checker in dev mode"
Calin Juravle [Tue, 27 Jan 2015 16:14:43 +0000 (16:14 +0000)]
Merge "run-test: enable checker in dev mode"

9 years agoMerge "ART: More Quick cleanup"
Andreas Gampe [Tue, 27 Jan 2015 16:14:26 +0000 (16:14 +0000)]
Merge "ART: More Quick cleanup"

9 years agorun-test: enable checker in dev mode
Calin Juravle [Mon, 26 Jan 2015 16:47:33 +0000 (16:47 +0000)]
run-test: enable checker in dev mode

Also, use explicit flush (instead of passing flush parameter to print)
to enable better compatibility across python versions.

Change-Id: Iaf294d88e932b778d5dce7f3c2b8eca775849973

9 years agoMerge "Revert "ART: Implement X86 hard float (Quick/JNI/Baseline)""
Vladimir Marko [Tue, 27 Jan 2015 15:10:10 +0000 (15:10 +0000)]
Merge "Revert "ART: Implement X86 hard float (Quick/JNI/Baseline)""

9 years agoRevert "ART: Implement X86 hard float (Quick/JNI/Baseline)"
Vladimir Marko [Tue, 27 Jan 2015 10:48:44 +0000 (10:48 +0000)]
Revert "ART: Implement X86 hard float (Quick/JNI/Baseline)"

And the 3 Mac build fixes. Fix conflicts in context_x86.* .

This reverts commits
  3d2c8e74c27efee58e24ec31441124f3f21384b9 ,
  34eda1dd66b92a361797c63d57fa19e83c08a1b4 ,
  f601d1954348b71186fa160a0ae6a1f4f1c5aee6 ,
  bc503348a1da573488503cc2819c9e30807bea31 .

Bug: 19150481
Change-Id: I6650ee30a7d261159380fe2119e14379e4dc9970

9 years agoMerge "Refactor register access from StackVisitor"
Sebastien Hertz [Tue, 27 Jan 2015 07:37:38 +0000 (07:37 +0000)]
Merge "Refactor register access from StackVisitor"

9 years agoMerge "JDWP: allocate single step control only when active"
Sebastien Hertz [Tue, 27 Jan 2015 07:32:23 +0000 (07:32 +0000)]
Merge "JDWP: allocate single step control only when active"

9 years agoART: More Quick cleanup
Andreas Gampe [Tue, 27 Jan 2015 01:37:27 +0000 (17:37 -0800)]
ART: More Quick cleanup

Remove frontend, move code to QuickCompiler.

Fix some includes.

Change-Id: Iedd8f339ec76635e48b166e285f7fc571fc948ac

9 years agoMerge "ART: Fix inlining of Mir2Lir"
Andreas Gampe [Tue, 27 Jan 2015 00:41:53 +0000 (00:41 +0000)]
Merge "ART: Fix inlining of Mir2Lir"

9 years agoART: Fix inlining of Mir2Lir
Andreas Gampe [Tue, 27 Jan 2015 00:39:37 +0000 (16:39 -0800)]
ART: Fix inlining of Mir2Lir

Missed -inl includes.

Change-Id: I39e6d603c7f5d36693aca3816653594488bff63f

9 years agoMerge "Fix 114-ParallelGC."
Hiroshi Yamauchi [Tue, 27 Jan 2015 00:27:04 +0000 (00:27 +0000)]
Merge "Fix 114-ParallelGC."

9 years agoFix 114-ParallelGC.
Hiroshi Yamauchi [Tue, 27 Jan 2015 00:05:11 +0000 (16:05 -0800)]
Fix 114-ParallelGC.

In main(), the outer ArrayList may still be reachable after the
try-catch under the interpreter or a compiler without a liveness
analysis and the last ArrayList allocation may fail due to an OOME.

This fixes 114-ParallelGC under the CC and the GSS collectors.

Bug: 12687968
Change-Id: Ie1082d38b2a677ec70fdc23b0187ae8ce0612808

9 years agoMerge "ART: Some Quick cleanup"
Andreas Gampe [Mon, 26 Jan 2015 23:46:42 +0000 (23:46 +0000)]
Merge "ART: Some Quick cleanup"

9 years agoART: Some Quick cleanup
Andreas Gampe [Fri, 23 Jan 2015 04:39:27 +0000 (20:39 -0800)]
ART: Some Quick cleanup

Make several fields const in CompilationUnit. May benefit some Mir2Lir
code that repeats tests, and in general immutability is good.

Remove compiler_internals.h and refactor some other headers to reduce
overly broad imports (and thus forced recompiles on changes).

Change-Id: I898405907c68923581373b5981d8a85d2e5d185a

9 years agoMerge "Add missing EndTiming in CompileMethod"
Mathieu Chartier [Mon, 26 Jan 2015 21:38:05 +0000 (21:38 +0000)]
Merge "Add missing EndTiming in CompileMethod"

9 years agoAdd missing EndTiming in CompileMethod
Mathieu Chartier [Mon, 26 Jan 2015 21:29:31 +0000 (13:29 -0800)]
Add missing EndTiming in CompileMethod

DCHECK failure when running tests with JIT.

Bug: 17950037
Change-Id: I4f886758b2a9322c2d7a8e8449e84c9a1709c813

9 years agoMerge "ART: Take apart stream output"
Andreas Gampe [Mon, 26 Jan 2015 20:28:47 +0000 (20:28 +0000)]
Merge "ART: Take apart stream output"

9 years agoMerge "Change AtomicStack to use StackReference"
Mathieu Chartier [Mon, 26 Jan 2015 19:19:26 +0000 (19:19 +0000)]
Merge "Change AtomicStack to use StackReference"

9 years agoJDWP: allocate single step control only when active
Sebastien Hertz [Mon, 26 Jan 2015 16:37:14 +0000 (17:37 +0100)]
JDWP: allocate single step control only when active

Only allocates SingleStepControl when a single step is requested in
a thread.

Bug: 19142632
Change-Id: I830b7eaea86b516d3fac80ce14d41c66b4bd5415

9 years agoART: Take apart stream output
Andreas Gampe [Mon, 26 Jan 2015 18:38:34 +0000 (10:38 -0800)]
ART: Take apart stream output

To investigate bug.

Bug: 19028453
Change-Id: Ib109caf2747f9adeef15dfc0f4708498e395dacf

9 years agoMerge "ART: Fix assembler labels' usage in macros for Mac."
Vladimir Marko [Mon, 26 Jan 2015 18:33:45 +0000 (18:33 +0000)]
Merge "ART: Fix assembler labels' usage in macros for Mac."

9 years agoART: Fix assembler labels' usage in macros for Mac.
Vladimir Marko [Mon, 26 Jan 2015 18:07:14 +0000 (18:07 +0000)]
ART: Fix assembler labels' usage in macros for Mac.

Change-Id: I99552528329a7f2cd5656eef0210c831cfc5fe09

9 years agoMerge "Optimizing: Speed up HEnvironment use removal"
David Brazdil [Mon, 26 Jan 2015 16:33:48 +0000 (16:33 +0000)]
Merge "Optimizing: Speed up HEnvironment use removal"

9 years agoOptimizing: Speed up HEnvironment use removal
David Brazdil [Fri, 23 Jan 2015 10:39:45 +0000 (10:39 +0000)]
Optimizing: Speed up HEnvironment use removal

Removal of use records from HEnvironment vregs involved iterating over
potentially large linked lists which made compilation of huge methods
very slow. This patch turns use lists into doubly-linked lists, stores
pointers to the relevant nodes inside HEnvironment and subsequently
turns the removals into constant-time operations.

Change-Id: I0e1d4d782fd624e7b8075af75d4adf0a0634a1ee

9 years agoMerge "Move code around and address growable_array comment."
Nicolas Geoffray [Mon, 26 Jan 2015 16:11:04 +0000 (16:11 +0000)]
Merge "Move code around and address growable_array comment."

9 years agoMerge "Quick: Fix range check for intrinsic String.charAt() on x86."
Vladimir Marko [Mon, 26 Jan 2015 16:08:22 +0000 (16:08 +0000)]
Merge "Quick: Fix range check for intrinsic String.charAt() on x86."

9 years agoMove code around and address growable_array comment.
Nicolas Geoffray [Mon, 26 Jan 2015 15:18:36 +0000 (15:18 +0000)]
Move code around and address growable_array comment.

- Move SideEffectsAnalysis to its own file.
- Move most of gvn.h to gvn.cc.
- Don't call Resize in GrowableArray constructor, but just set num_used
  directly.

Change-Id: I1f1291207945d678d3c99cc0ec1ec155bcae82f6

9 years agoMerge "Fix Mac build."
Vladimir Marko [Mon, 26 Jan 2015 15:46:40 +0000 (15:46 +0000)]
Merge "Fix Mac build."

9 years agoRefactor register access from StackVisitor
Sebastien Hertz [Thu, 22 Jan 2015 17:57:14 +0000 (18:57 +0100)]
Refactor register access from StackVisitor

Moves register access checking up to StackVisitor by adding methods
IsAccessibleGPR and IsAccessibleFPR in Context class. It allows to
simplify GetGPR/FPR and SetGPR/FPR methods in the Context class (and
its subclasses).

Also simplifies code in StackVisitor by adding IsAccessibleRegister,
GetRegister and SetRegister methods which then call either GPR or FPR
specific methods in Context depending on the nature of the accessed
register.

Bug: 18547544
Bug: 19106446
Change-Id: I6e707608d935a71571d0e975a6e766053de3763a

9 years agoFix Mac build.
Vladimir Marko [Mon, 26 Jan 2015 15:33:03 +0000 (15:33 +0000)]
Fix Mac build.

Change-Id: I88c0b3c7b799df8dd0aab86b6801c37093045129

9 years agoMerge "Fix build breakage after GVN change."
Nicolas Geoffray [Mon, 26 Jan 2015 15:17:44 +0000 (15:17 +0000)]
Merge "Fix build breakage after GVN change."

9 years agoFix build breakage after GVN change.
Nicolas Geoffray [Mon, 26 Jan 2015 15:13:47 +0000 (15:13 +0000)]
Fix build breakage after GVN change.

Change-Id: I120717f8cae452cad9967559c466c1605a3e7b14

9 years agoMerge "ART: Fix asm literals for Mac builds."
Vladimir Marko [Mon, 26 Jan 2015 14:46:12 +0000 (14:46 +0000)]
Merge "ART: Fix asm literals for Mac builds."

9 years agoMerge "Introduce a SideEffectsAnalysis class."
Nicolas Geoffray [Mon, 26 Jan 2015 14:45:59 +0000 (14:45 +0000)]
Merge "Introduce a SideEffectsAnalysis class."

9 years agoIntroduce a SideEffectsAnalysis class.
Nicolas Geoffray [Mon, 26 Jan 2015 12:54:33 +0000 (12:54 +0000)]
Introduce a SideEffectsAnalysis class.

LICM also needs the side effects information of loops, so move
the GVN::ComputeSideEffects method into its own analysis class.

Change-Id: I810c8230a0eb6b9b536e8f808e17a3a4ad72f7db

9 years agoART: Fix asm literals for Mac builds.
Vladimir Marko [Mon, 26 Jan 2015 14:30:46 +0000 (14:30 +0000)]
ART: Fix asm literals for Mac builds.

Change-Id: Ief922e1e06811b8ec25771b9854376fae3e07929

9 years agoQuick: Fix range check for intrinsic String.charAt() on x86.
Vladimir Marko [Mon, 26 Jan 2015 14:06:46 +0000 (14:06 +0000)]
Quick: Fix range check for intrinsic String.charAt() on x86.

Bug: 19125146
Change-Id: I274190a7a60cd2e29a854738ed1ec99a9e611969

9 years agoMerge "Fix a bug in the liveness analysis."
Nicolas Geoffray [Mon, 26 Jan 2015 14:26:44 +0000 (14:26 +0000)]
Merge "Fix a bug in the liveness analysis."

9 years agoFix a bug in the liveness analysis.
Nicolas Geoffray [Fri, 23 Jan 2015 18:01:51 +0000 (18:01 +0000)]
Fix a bug in the liveness analysis.

A range after a loop might also be after a lifetime hole.
In this situation we must preserve the hole, and not merge
it with the loop start.

Change-Id: I82eddef059592102a25362cdaa4273200574c2ae

9 years agoMerge "Fix lint error and typo."
Nicolas Geoffray [Mon, 26 Jan 2015 13:58:04 +0000 (13:58 +0000)]
Merge "Fix lint error and typo."

9 years agoFix lint error and typo.
Nicolas Geoffray [Mon, 26 Jan 2015 13:57:30 +0000 (13:57 +0000)]
Fix lint error and typo.

Change-Id: Ie2787a4da5b2d76975664e64f443c081127b7cd9

9 years agoMerge "Fix codegen_test."
Nicolas Geoffray [Mon, 26 Jan 2015 13:33:33 +0000 (13:33 +0000)]
Merge "Fix codegen_test."

9 years agoFix codegen_test.
Nicolas Geoffray [Mon, 26 Jan 2015 12:49:35 +0000 (12:49 +0000)]
Fix codegen_test.

Native and ART do not have the same calling convention for ART,
so we need to adjust blocked and allocated registers.

Change-Id: I606b2620c0e5a54bd60d6100a137c06616ad40b4

9 years agoMerge "ART: Implement X86 hard float (Quick/JNI/Baseline)"
Vladimir Marko [Mon, 26 Jan 2015 12:44:44 +0000 (12:44 +0000)]
Merge "ART: Implement X86 hard float (Quick/JNI/Baseline)"

9 years agoMerge "Support callee-save registers on ARM."
Nicolas Geoffray [Mon, 26 Jan 2015 10:10:46 +0000 (10:10 +0000)]
Merge "Support callee-save registers on ARM."

9 years agoART: clear dirty cards of alloc space for MS/CMS partial and full GCs
Lei Li [Thu, 15 Jan 2015 03:55:26 +0000 (11:55 +0800)]
ART: clear dirty cards of alloc space for MS/CMS partial and full GCs

For MS/CMS partial and full GCs, we could clear the dirty cards of alloc
space when we process cards as we care about the dirty cards after GC
starts.

Change-Id: I1f9b32b20d75979387bc5d26b0cf9a256dcf20b6
Signed-off-by: Lei Li <lei.l.li@intel.com>
9 years agoChange AtomicStack to use StackReference
Mathieu Chartier [Fri, 23 Jan 2015 21:50:03 +0000 (13:50 -0800)]
Change AtomicStack to use StackReference

Previously used Object*, using StackReference saves memory on 64 bit
devices.

Bug: 12935052
Bug: 17643507
Change-Id: I035878690054eeeb24d655a900b8f26c837703ff

9 years agoSupport callee-save registers on ARM.
Nicolas Geoffray [Fri, 23 Jan 2015 18:23:14 +0000 (18:23 +0000)]
Support callee-save registers on ARM.

Change-Id: I7c519b7a828c9891b1141a8e51e12d6a8bc84118

9 years agoMerge "ART: Fix new[] / delete mismatch"
Andreas Gampe [Sat, 24 Jan 2015 05:18:25 +0000 (05:18 +0000)]
Merge "ART: Fix new[] / delete mismatch"

9 years agoART: Fix new[] / delete mismatch
Andreas Gampe [Sat, 24 Jan 2015 03:51:15 +0000 (19:51 -0800)]
ART: Fix new[] / delete mismatch

The type of the unique_ptr in MarkQueue should be an array type,
as it is holding an array, actually.

Change-Id: If1d05a1d52cd58a373f240f7156fc69b70324298

9 years agoMerge "Make it possible to enable read barriers at build time."
Hiroshi Yamauchi [Sat, 24 Jan 2015 01:26:52 +0000 (01:26 +0000)]
Merge "Make it possible to enable read barriers at build time."

9 years agoMake it possible to enable read barriers at build time.
Hiroshi Yamauchi [Sat, 24 Jan 2015 01:23:42 +0000 (17:23 -0800)]
Make it possible to enable read barriers at build time.

Bug: 12687968
Change-Id: I26ec55063c2d5ff0181e628c1002bfa609815b29

9 years agoMerge "ART: On shutdown, only warn on mutex contention"
Andreas Gampe [Fri, 23 Jan 2015 23:17:16 +0000 (23:17 +0000)]
Merge "ART: On shutdown, only warn on mutex contention"

9 years agoART: On shutdown, only warn on mutex contention
Andreas Gampe [Fri, 23 Jan 2015 03:48:51 +0000 (19:48 -0800)]
ART: On shutdown, only warn on mutex contention

Do not abort, as daemon threads may still be active.

Bug: 17894429

(cherry picked from commit c0440f69ebf051ff2ffdc00de51005a040014462)

Change-Id: I7c1d50ff8d4a5e150279e703a69c8f2f1d423e6b

9 years agoMerge "More of the concurrent copying collector."
Hiroshi Yamauchi [Fri, 23 Jan 2015 23:13:11 +0000 (23:13 +0000)]
Merge "More of the concurrent copying collector."

9 years agoMore of the concurrent copying collector.
Hiroshi Yamauchi [Fri, 9 Jan 2015 22:03:35 +0000 (14:03 -0800)]
More of the concurrent copying collector.

Bug: 12687968
Change-Id: I62f70274d47df6d6cab714df95c518b750ce3105

9 years agoMerge "Fix valgrind error."
Nicolas Geoffray [Fri, 23 Jan 2015 22:01:13 +0000 (22:01 +0000)]
Merge "Fix valgrind error."

9 years agoFix valgrind error.
Nicolas Geoffray [Fri, 23 Jan 2015 21:51:55 +0000 (21:51 +0000)]
Fix valgrind error.

Also introduce kLastCpuRegister to define kFakeReturnRegister.

Change-Id: I58cef6186c0452d45b5d2dcba9298cbe07f3552d

9 years agoMerge "Fix compaction bugs related to IdentityHashCode"
Mathieu Chartier [Fri, 23 Jan 2015 19:10:41 +0000 (19:10 +0000)]
Merge "Fix compaction bugs related to IdentityHashCode"

9 years agoFix compaction bugs related to IdentityHashCode
Mathieu Chartier [Fri, 23 Jan 2015 01:02:27 +0000 (17:02 -0800)]
Fix compaction bugs related to IdentityHashCode

IdentityHashCode is a suspend point if monitor inflation occurs.

Change-Id: I114021aed8b3f3437109ef622298de05e13b4e34

9 years agoART: Implement X86 hard float (Quick/JNI/Baseline)
Mark Mendell [Tue, 13 Jan 2015 22:32:55 +0000 (17:32 -0500)]
ART: Implement X86 hard float (Quick/JNI/Baseline)

Use XMM0-XMM3 as parameter registers for float/double on X86.  X86_64
already uses XMM0-XMM7 for parameters.

Change the 'hidden' argument register from XMM0 to XMM7 to avoid a
conflict.

Add support for FPR save/restore in runtime/arch/x86.

Minimal support for Optimizing baseline compiler.

Bump the version in runtime/oat.h because this is an ABI change.

Change-Id: Ia6fe150e8488b9e582b0178c0dda65fc81d5a8ba
Signed-off-by: Mark Mendell <mark.p.mendell@intel.com>
9 years agoMerge "Fix makefile error."
Nicolas Geoffray [Fri, 23 Jan 2015 17:48:57 +0000 (17:48 +0000)]
Merge "Fix makefile error."

9 years agoFix makefile error.
Nicolas Geoffray [Fri, 23 Jan 2015 17:48:28 +0000 (17:48 +0000)]
Fix makefile error.

Change-Id: I797de0bed7b9c51cd7889de5144b8dbfd641335d

9 years agoMerge "Mark test as failing for optimizing."
Nicolas Geoffray [Fri, 23 Jan 2015 17:31:36 +0000 (17:31 +0000)]
Merge "Mark test as failing for optimizing."

9 years agoMark test as failing for optimizing.
Nicolas Geoffray [Fri, 23 Jan 2015 17:30:54 +0000 (17:30 +0000)]
Mark test as failing for optimizing.

Change-Id: I928ff56d232a647e8227f2b97a8f4a51791a5f95

9 years agoMerge "Fix exception handling during deoptimization"
Sebastien Hertz [Fri, 23 Jan 2015 16:51:17 +0000 (16:51 +0000)]
Merge "Fix exception handling during deoptimization"

9 years agoFix exception handling during deoptimization
Sebastien Hertz [Fri, 16 Jan 2015 18:49:09 +0000 (19:49 +0100)]
Fix exception handling during deoptimization

When interpreting a deoptimized shadow frame, we may start with a
pending exception thrown by a previous deoptimized shadow frame (from
a previous invoke). Therefore, we need to handle it before executing
any instruction, otherwise we execute incorrect code.

Because we need the DEX pc of the throwing instruction to find a
matching catch handler, we initialize deoptimized shadow frames with
the current DEX pc at the time the stack is deoptimized.
When we are about to interpret a deoptimized shadow frame, we need to
update the shadow frame with the DEX pc of the next instruction to
interpret. There are three cases:
- if there is no pending exception, this is the instruction following
the current one.
- if there is a pending exception and we found a matching catch
handler, this is the first instruction of this handler.
- if there is a pending exception but there is no matching catch
handler, we do not execute the deoptimized shadow frame and continue
to its caller.

The verifier now fails when a method starts with a move-exception
instruction. Indeed we cannot start executing a method with a pending
exception.

Bug: 19057915
Bug: 19041195
Bug: 18607595
Change-Id: I355ac81e6ac098edc7e3cc8c13dbfa24a2969ab2

9 years agoMerge "Fix BitVector::IndexIterator::operator*() to return uint32_t."
Vladimir Marko [Fri, 23 Jan 2015 11:49:09 +0000 (11:49 +0000)]
Merge "Fix BitVector::IndexIterator::operator*() to return uint32_t."

9 years agoFix BitVector::IndexIterator::operator*() to return uint32_t.
Vladimir Marko [Fri, 2 Jan 2015 17:00:44 +0000 (17:00 +0000)]
Fix BitVector::IndexIterator::operator*() to return uint32_t.

Change-Id: I3cfc028b1c3744ec85ea00eadcbccfdde6fd51d3

9 years agoMerge "Implicit stack overflow checks and null checks are the norm now."
Nicolas Geoffray [Fri, 23 Jan 2015 09:38:05 +0000 (09:38 +0000)]
Merge "Implicit stack overflow checks and null checks are the norm now."

9 years agoImplicit stack overflow checks and null checks are the norm now.
Nicolas Geoffray [Fri, 23 Jan 2015 09:36:52 +0000 (09:36 +0000)]
Implicit stack overflow checks and null checks are the norm now.

Make the default to true so that unit tests don't need to
use the full CompilerOptions constructor.

Change-Id: Ieb52235af73dce2fa03f12932e22ebe2294c805b

9 years agoMerge "Support callee save floating point registers on x64."
Nicolas Geoffray [Fri, 23 Jan 2015 09:23:12 +0000 (09:23 +0000)]
Merge "Support callee save floating point registers on x64."

9 years agoSupport callee save floating point registers on x64.
Nicolas Geoffray [Thu, 22 Jan 2015 13:50:01 +0000 (13:50 +0000)]
Support callee save floating point registers on x64.

- Share the computation of core_spill_mask and fpu_spill_mask
  between backends.
- Remove explicit stack overflow check support: we need to adjust
  them and since they are not tested, they will easily bitrot.

Change-Id: I0b619b8de4e1bdb169ea1ae7c6ede8df0d65837a

9 years agoART: reduce UpdateModUnionTable time for partial GC
Lei Li [Fri, 23 Jan 2015 08:37:59 +0000 (16:37 +0800)]
ART: reduce UpdateModUnionTable time for partial GC

UpdateModUnionTable is to scan objects with dirty cards in the immune
region. The immune region of partial GC are image space and zygote
space. As image space is always immuned for any kind of GC, there is no
need to scan dirty cards of zygote space pointing to image space for
partial GC.

Change-Id: I2e1f61c2b361985c211076a441418797abf324c2

9 years agoMerge "ART: clear dirty cards of alloc space in pause phase"
Mathieu Chartier [Fri, 23 Jan 2015 03:51:17 +0000 (03:51 +0000)]
Merge "ART: clear dirty cards of alloc space in pause phase"

9 years agoMerge "Fix 116-nodex2oat to use arguments in custom run script"
Brian Carlstrom [Fri, 23 Jan 2015 01:22:20 +0000 (01:22 +0000)]
Merge "Fix 116-nodex2oat to use arguments in custom run script"