OSDN Git Service

android-x86/art.git
10 years agoFaster TLAB allocator.
Mathieu Chartier [Fri, 11 Jul 2014 17:26:37 +0000 (10:26 -0700)]
Faster TLAB allocator.

New TLAB allocator doesn't increment bytes allocated until we allocate
a new TLAB. This increases allocation performance by avoiding a CAS.

MemAllocTest:
Before GSS TLAB: 3400ms.
After GSS TLAB: 2750ms.

Bug: 9986565

Change-Id: I1673c27555330ee90d353b98498fa0e67bd57fad

10 years agoMerge "Add missing class initialization during compilation and tests"
Sebastien Hertz [Fri, 11 Jul 2014 14:52:06 +0000 (14:52 +0000)]
Merge "Add missing class initialization during compilation and tests"

10 years agoMerge "Fix missing single-step event"
Sebastien Hertz [Fri, 11 Jul 2014 13:36:05 +0000 (13:36 +0000)]
Merge "Fix missing single-step event"

10 years agoMerge "Revert "Revert "Revert "Add intrinsic for Reference.get()""""
Sebastien Hertz [Fri, 11 Jul 2014 10:11:12 +0000 (10:11 +0000)]
Merge "Revert "Revert "Revert "Add intrinsic for Reference.get()""""

10 years agoRevert "Revert "Revert "Add intrinsic for Reference.get()"""
Sebastien Hertz [Fri, 11 Jul 2014 10:09:13 +0000 (10:09 +0000)]
Revert "Revert "Revert "Add intrinsic for Reference.get()"""

This reverts commit d4415e8bd04c4a9367744ff0149597b4f37a0e0a.

Change-Id: I34553ccbdcfea35c7742d21be2a74dc7085ab2a0

10 years agoRevert "Revert "Add intrinsic for Reference.get()""
Christopher Ferris [Fri, 11 Jul 2014 06:44:39 +0000 (06:44 +0000)]
Revert "Revert "Add intrinsic for Reference.get()""

This reverts commit a9b870b73a155ce70c867d5b3f9758fab0b45f07.

Change-Id: Ic2a9b47f2b911bef4b764d10bc33cf000e4b4211

10 years agoRevert "Add intrinsic for Reference.get()"
Christopher Ferris [Fri, 11 Jul 2014 04:18:58 +0000 (04:18 +0000)]
Revert "Add intrinsic for Reference.get()"

This reverts commit 460503b13bc894828a2d2d47d09e5534b3e91aa1.

Change-Id: Ie63f43049307e02e3b90f4e034abc9ea54ca4e24

10 years agoMerge "Revert "Revert "Revert "Add implicit null and stack checks for x86""""
Nicolas Geoffray [Fri, 11 Jul 2014 08:27:55 +0000 (08:27 +0000)]
Merge "Revert "Revert "Revert "Add implicit null and stack checks for x86""""

10 years agoRevert "Revert "Revert "Add implicit null and stack checks for x86"""
Nicolas Geoffray [Fri, 11 Jul 2014 08:26:40 +0000 (08:26 +0000)]
Revert "Revert "Revert "Add implicit null and stack checks for x86"""

Broke the build.

This reverts commit 7fb36ded9cd5b1d254b63b3091f35c1e6471b90e.

Change-Id: I9df0e7446ff0913a0e1276a558b2ccf6c8f4c949

10 years agoRevert "Revert "Add implicit null and stack checks for x86""
Dave Allison [Thu, 10 Jul 2014 02:05:10 +0000 (02:05 +0000)]
Revert "Revert "Add implicit null and stack checks for x86""

Fixes x86_64 cross compile issue.  Removes command line options
and property to set implicit checks - this is hard coded now.

This reverts commit 3d14eb620716e92c21c4d2c2d11a95be53319791.

Change-Id: I5404473b5aaf1a9c68b7181f5952cb174d93a90d

10 years agoMerge "ART: Do not dump hidden basic blocks and add a counter to file naming"
Ian Rogers [Thu, 10 Jul 2014 21:20:14 +0000 (21:20 +0000)]
Merge "ART: Do not dump hidden basic blocks and add a counter to file naming"

10 years agoMerge "x86_64: Enable fp-reg promotion"
Andreas Gampe [Thu, 10 Jul 2014 21:12:02 +0000 (21:12 +0000)]
Merge "x86_64: Enable fp-reg promotion"

10 years agox86_64: Enable fp-reg promotion
Serguei Katkov [Tue, 8 Jul 2014 10:21:53 +0000 (17:21 +0700)]
x86_64: Enable fp-reg promotion

Patch introduces 4 register XMM12-15 available for promotion of
fp virtual registers.

Change-Id: I3f89ad07fc8ae98b70f550eada09be7b693ffb67
Signed-off-by: Serguei Katkov <serguei.i.katkov@intel.com>
Signed-off-by: Chao-ying Fu <chao-ying.fu@intel.com>
10 years agoMerge "Revert "Fix mac build""
Nicolas Geoffray [Fri, 11 Jul 2014 08:26:20 +0000 (08:26 +0000)]
Merge "Revert "Fix mac build""

10 years agoRevert "Fix mac build"
Christopher Ferris [Fri, 11 Jul 2014 05:43:02 +0000 (05:43 +0000)]
Revert "Fix mac build"

This reverts commit e9343344d9bd268a05d1eae1ce80a3278ec19c89.

Change-Id: I43d1717af9c3b1237dcacec66f55a4e4b8e1f0fe

10 years agoFix mac build
Dave Allison [Thu, 10 Jul 2014 22:29:28 +0000 (15:29 -0700)]
Fix mac build

Fixes x86 fault handler, sigchain and quick_entrypoints for x86_64.

Bug: 16215218
Change-Id: I5e58660ea815042968444e6352c57a5f53314cfd

10 years agoMerge "Updates to help classes derived from X86Mir2Lir"
Ian Rogers [Thu, 10 Jul 2014 22:00:23 +0000 (22:00 +0000)]
Merge "Updates to help classes derived from X86Mir2Lir"

10 years agoUpdates to help classes derived from X86Mir2Lir
Mark Mendell [Fri, 4 Jul 2014 01:34:41 +0000 (21:34 -0400)]
Updates to help classes derived from X86Mir2Lir

Just a couple of extra changes to help me out.  These changes won't
affect anyone else.

Change-Id: I0e0985a4f16822d5cbfabbf81c9902d34ebdb5da
Signed-off-by: Mark Mendell <mark.p.mendell@intel.com>
10 years agoMerge "Revert "Revert "Add implicit null and stack checks for x86"""
Dave Allison [Thu, 10 Jul 2014 21:25:05 +0000 (21:25 +0000)]
Merge "Revert "Revert "Add implicit null and stack checks for x86"""

10 years agoMerge "Revert "Revert "Add intrinsic for Reference.get()"""
Christopher Ferris [Fri, 11 Jul 2014 07:09:25 +0000 (07:09 +0000)]
Merge "Revert "Revert "Add intrinsic for Reference.get()"""

10 years agoMerge "ART: Fix GenSelect and GenFusedLongCmpBranch for ARM64"
Andreas Gampe [Fri, 11 Jul 2014 06:12:08 +0000 (06:12 +0000)]
Merge "ART: Fix GenSelect and GenFusedLongCmpBranch for ARM64"

10 years agoART: Fix GenSelect and GenFusedLongCmpBranch for ARM64
Andreas Gampe [Thu, 10 Jul 2014 10:23:41 +0000 (03:23 -0700)]
ART: Fix GenSelect and GenFusedLongCmpBranch for ARM64

Depending on the result, we need to issue a wide csel. Also need
to handle constants, and src and dest being the same.

In GenFusedLongCmpBranch there is an ordering issue. If we swap
the inputs, we did not Load the second one.

Change-Id: Icb9876ca1288602d078b9fb89ea964ec2c910e0c

10 years agoMerge "Revert "Add intrinsic for Reference.get()""
Christopher Ferris [Fri, 11 Jul 2014 04:45:03 +0000 (04:45 +0000)]
Merge "Revert "Add intrinsic for Reference.get()""

10 years agoMerge "Fix mac build"
Dave Allison [Fri, 11 Jul 2014 03:42:17 +0000 (03:42 +0000)]
Merge "Fix mac build"

10 years agoMerge "Add intrinsic for Reference.get()"
Mathieu Chartier [Fri, 11 Jul 2014 01:54:07 +0000 (01:54 +0000)]
Merge "Add intrinsic for Reference.get()"

10 years agoMerge "Change default heap maximum size to be 256m."
Mathieu Chartier [Fri, 11 Jul 2014 01:47:39 +0000 (01:47 +0000)]
Merge "Change default heap maximum size to be 256m."

10 years agoChange default heap maximum size to be 256m.
Mathieu Chartier [Fri, 11 Jul 2014 00:50:34 +0000 (17:50 -0700)]
Change default heap maximum size to be 256m.

Useful for command line benchmarks.

Change-Id: Ie525863cd8eff93c64ce76639b1108fbdad91633

10 years agoMerge "Fix race condition in release pages."
Mathieu Chartier [Fri, 11 Jul 2014 00:31:37 +0000 (00:31 +0000)]
Merge "Fix race condition in release pages."

10 years agoFix race condition in release pages.
Mathieu Chartier [Thu, 10 Jul 2014 17:16:44 +0000 (10:16 -0700)]
Fix race condition in release pages.

There was a race condition where another thread could coalesce the
free page run before we acquired the lock. In that case
free_page_runs_.find will not find a run starting at fpr. Added a
condition to handle this case. Also added handling for free page
runs with begin with released pages but end with empty pages.

Bug: 16191993
Change-Id: Ib12fdac8c246eae29c36f6a6728eb11d85553bbb

10 years agoART: Do not dump hidden basic blocks and add a counter to file naming
Jean Christophe Beyler [Thu, 19 Jun 2014 16:34:51 +0000 (09:34 -0700)]
ART: Do not dump hidden basic blocks and add a counter to file naming

Currently, if the system dumps the CFG, it dumps the hidden BasicBlocks.
Also, the patch adds a counter that gets incremented to make the file naming
  unique, using AtomicInteger.

Change-Id: I55b489c2c2bded73b62d64a94a4a8a54d2ebed2b
Signed-off-by: Jean Christophe Beyler <jean.christophe.beyler@intel.com>
10 years agoAdd intrinsic for Reference.get()
Fred Shih [Wed, 18 Jun 2014 18:26:11 +0000 (11:26 -0700)]
Add intrinsic for Reference.get()

Added an intrinsic function for Reference.get(). Return immediately
without going through JNI if the slow path is not currently in use.
Otherwise, branch off to the the existing JNI function.

Approximately 47x speedup for cases where slow path is not enabled.

Change-Id: I13ad65a356fe4e104d8d83980694dc2740d7d039

10 years agoMerge "ART: Add Invokes to DecodedInstruction"
Ian Rogers [Thu, 10 Jul 2014 19:15:04 +0000 (19:15 +0000)]
Merge "ART: Add Invokes to DecodedInstruction"

10 years agoART: Add Invokes to DecodedInstruction
Jean Christophe Beyler [Mon, 2 Jun 2014 16:03:14 +0000 (09:03 -0700)]
ART: Add Invokes to DecodedInstruction

Add a method Invokes to test for the kInvoke flag.
Also moved IsPseudoMirOp to DecodedInstruction to use it for the various
  querry methods.

Change-Id: I59a2056b7b802b8393fa2b0d977304d252b38c89
Signed-off-by: Jean Christophe Beyler <jean.christophe.beyler@intel.com>
10 years agoMerge "ART: Handle Extended MIRs in a uniform manner"
Ian Rogers [Thu, 10 Jul 2014 18:56:00 +0000 (18:56 +0000)]
Merge "ART: Handle Extended MIRs in a uniform manner"

10 years agoMerge "ART: Detached blocks should not be processed by compiler"
Ian Rogers [Thu, 10 Jul 2014 18:40:06 +0000 (18:40 +0000)]
Merge "ART: Detached blocks should not be processed by compiler"

10 years agoMerge "Make CAS operations in Object use art::Atomic."
Ian Rogers [Thu, 10 Jul 2014 17:35:57 +0000 (17:35 +0000)]
Merge "Make CAS operations in Object use art::Atomic."

10 years agoAdd missing class initialization during compilation and tests
Sebastien Hertz [Tue, 24 Jun 2014 12:35:40 +0000 (14:35 +0200)]
Add missing class initialization during compilation and tests

Adds missing class initialization during compilation and tests, especially
java.lang.Class. Otherwise, we'd be able to execute code while the referring
class is not initialized or initializing.

Also adds mirror::Class::AssertInitializedOrInitializingInThread method to
check class initialization when entering the interpreter: the called method's
declaring class must either be initialized or be initializing by the current
thread (other threads must be waiting for the class initialization to complete
holding its lock). Note we only do this check in debug build.

Bump oat version to force compilation.

Bug: 15899971
Change-Id: I9a4edd3739a3ca4cf1c4929dcbb44cdf7a1ca1fe

10 years agoFix missing single-step event
Sebastien Hertz [Wed, 11 Jun 2014 17:45:05 +0000 (19:45 +0200)]
Fix missing single-step event

During debugging, we used to suspend too lately after a step. It occurred when
we were stepping out of an interpreted method into a compiled "caller" method.
The issue is we did not deoptimize when returning from the interpreted method
but only did it when returning from the compiled method. Therefore we were not
executing the rest of the compiled method's code with interpreter which
prevents from debugging it.

This CL fixes this issue by using instrumentation entry/exit stubs when calling
interpreted method from compiled code. Therefore, we execute instrumentation
exit stub when returning from interpreter and are able to deoptimize from this
point. This allows to execute compiled method's code with interpreter and to
debug it.

We now also prevent from reporting method entry/exit twice while instrumenting
interpreted methods. We report method entry/exit events only from interpreter.

Bug: 14422182
Bug: 11705760
Change-Id: Ia1175d36202239273083c4e9733c7e9290244090

10 years agoMake CAS operations in Object use art::Atomic.
Ian Rogers [Thu, 10 Jul 2014 09:07:54 +0000 (02:07 -0700)]
Make CAS operations in Object use art::Atomic.

Make naming consistent with art::Atomic.

Change-Id: If3abdb019ef8b53bd809e3fef3fd5248aeb27e9a

10 years agoFix GC to use art::Atomic rather than compiler intrinsics.
Ian Rogers [Thu, 10 Jul 2014 07:56:36 +0000 (00:56 -0700)]
Fix GC to use art::Atomic rather than compiler intrinsics.

Changes to SpaceBitmap::AtomicTestAndSet and Space::end_. Space::end_ is made
atomic rather than volatile to fully capture all its uses multi-threaded or not
uses.

Change-Id: I3058964b8ad90a8c253b3d7f75585f63ca2fb5e3

10 years agoART: Handle Extended MIRs in a uniform manner
Udayan Banerji [Thu, 10 Jul 2014 02:14:53 +0000 (19:14 -0700)]
ART: Handle Extended MIRs in a uniform manner

The special handling is needed since some extended MIRs can hold values in
args array, and we might want to handle the dataflow for those in a
specialized manner. Current dataflow attributes may not be able to describe
it for the extended MIRs.

Change-Id: I8b64f3142a4304282bb31f1d4686eba72284d97d
Signed-off-by: Jean Christophe Beyler <jean.christophe.beyler@intel.com>
Signed-off-by: Udayan Banerji <udayan.banerji@intel.com>
10 years agoMerge "Slow path for iget should expect return in core reg"
Andreas Gampe [Thu, 10 Jul 2014 09:29:20 +0000 (09:29 +0000)]
Merge "Slow path for iget should expect return in core reg"

10 years agoSlow path for iget should expect return in core reg
Serguei Katkov [Mon, 7 Jul 2014 17:45:45 +0000 (00:45 +0700)]
Slow path for iget should expect return in core reg

Slow path for iget invokes the C implementation.
In all cases the C function returns the result in core reg.
So implementation should expect the result in core reg
independent on whether it is fp or not.

Change-Id: I57fb0e684c38af22316398d8071f087bd4bd253c
Signed-off-by: Serguei Katkov <serguei.i.katkov@intel.com>
10 years agoMerge "Improve the OOME fragmentation message."
Hiroshi Yamauchi [Thu, 10 Jul 2014 19:56:12 +0000 (19:56 +0000)]
Merge "Improve the OOME fragmentation message."

10 years agoImprove the OOME fragmentation message.
Hiroshi Yamauchi [Wed, 9 Jul 2014 19:54:32 +0000 (12:54 -0700)]
Improve the OOME fragmentation message.

Change-Id: I390d3622f8d572ec7e34ea6dff9e1e0936e81ac1

10 years agoMerge "Move thread state to art::Atomic."
Ian Rogers [Thu, 10 Jul 2014 04:38:36 +0000 (04:38 +0000)]
Merge "Move thread state to art::Atomic."

10 years agoMove thread state to art::Atomic.
Ian Rogers [Thu, 10 Jul 2014 04:12:06 +0000 (21:12 -0700)]
Move thread state to art::Atomic.

Leaves the CAS operations as relaxed although art::Atomic treats relaxed CAS
as a strong CAS when not compiling with clang.

Change-Id: I6d37c22173540d166b624385e52e4ad05e592adc

10 years agoMerge "ART: Rename CallInlining to SpecialMethodInliner"
Vladimir Marko [Thu, 10 Jul 2014 18:14:31 +0000 (18:14 +0000)]
Merge "ART: Rename CallInlining to SpecialMethodInliner"

10 years agoART: Rename CallInlining to SpecialMethodInliner
Razvan A Lupusoru [Wed, 9 Jul 2014 23:42:19 +0000 (16:42 -0700)]
ART: Rename CallInlining to SpecialMethodInliner

The CallInlining pass is used to inline just a set of pre-categorized methods.
This set of methods includes empty, instance getters, instance setters, argument
return, and constant return. Since it inlines only "special methods", it makes
sense to name it to reflect that.

Change-Id: Iea2c1820080b0c212c99e977f6b5d34ee0774868
Signed-off-by: Razvan A Lupusoru <razvan.a.lupusoru@intel.com>
10 years agoMerge "Global Value Numbering."
Vladimir Marko [Thu, 10 Jul 2014 09:12:40 +0000 (09:12 +0000)]
Merge "Global Value Numbering."

10 years agoGlobal Value Numbering.
Vladimir Marko [Fri, 30 May 2014 09:01:32 +0000 (10:01 +0100)]
Global Value Numbering.

Implement the Global Value Numbering for optimization
purposes. Use it for the null check and range check
elimination as the LVN used to do.

The order of evaluation of basic blocks needs improving as
we currently fail to recognize some obviously identical
values in methods with more than one loop. (There are three
disabled tests that check this. This is just a missed
optimization, not a correctness issue.)

Change-Id: I0d0ce16b2495b5a3b17ad1b2b32931cd69f5a25a

10 years agoMerge "Handle potential <clinit>() correctly in LVN."
Vladimir Marko [Thu, 10 Jul 2014 08:22:12 +0000 (08:22 +0000)]
Merge "Handle potential <clinit>() correctly in LVN."

10 years agoMerge "Faster deduplication in OatWriter."
Vladimir Marko [Thu, 10 Jul 2014 08:21:49 +0000 (08:21 +0000)]
Merge "Faster deduplication in OatWriter."

10 years agoMerge "Fix GC to use art::Atomic rather than compiler intrinsics."
Ian Rogers [Thu, 10 Jul 2014 08:14:55 +0000 (08:14 +0000)]
Merge "Fix GC to use art::Atomic rather than compiler intrinsics."

10 years agoFaster deduplication in OatWriter.
Vladimir Marko [Wed, 9 Jul 2014 15:06:40 +0000 (16:06 +0100)]
Faster deduplication in OatWriter.

Use lower_bound() to look for duplicates and use it as
a hint for insertion of new entries. Add a few useful
functions to SafeMap<>.

Change-Id: If7eab3f5d153be6e0d7ae040929849f1a636ee29

10 years agoART: Detached blocks should not be processed by compiler
Niranjan Kumar [Thu, 12 Jun 2014 19:15:48 +0000 (12:15 -0700)]
ART: Detached blocks should not be processed by compiler

It is possible for blocks to be detached. This means that 'successor_block_info->block'
may evaluate to 'NullBasicBlockId'  The code should detect this case and handle it
appropriately.

Signed-off-by: vladimir.a.ivanov <vladimir.a.ivanov@intel.com
Signed-off-by: Niranjan Kumar <niranjan.kumar@intel.com
Change-Id: I410059cd2cbda342cc1380050c0972fcaa2b7a8e

10 years agoHandle potential <clinit>() correctly in LVN.
Vladimir Marko [Wed, 9 Jul 2014 13:45:36 +0000 (14:45 +0100)]
Handle potential <clinit>() correctly in LVN.

Bug: 16177324
Change-Id: I727ab6ce9aa9a608fe570cf391a6b732a12a8655

10 years agoMerge "ART: Enable some ARM64 optimizations."
Andreas Gampe [Thu, 10 Jul 2014 10:03:29 +0000 (10:03 +0000)]
Merge "ART: Enable some ARM64 optimizations."

10 years agoART: Enable some ARM64 optimizations.
Andreas Gampe [Thu, 10 Jul 2014 09:04:01 +0000 (02:04 -0700)]
ART: Enable some ARM64 optimizations.

Enables kSuppressLoads, kTrackLiveTemps, kSafeOptimizations,
kPromoteCompilerTemps, kClassInitCheckElimination,
kSuppressExceptionEdges and kMatch.

Change-Id: Id3650adce7140dde8d667cd3f1b4a1c2598f156e

10 years agoMerge "Use memory chunks for monitors on LP64"
Andreas Gampe [Thu, 10 Jul 2014 07:54:06 +0000 (07:54 +0000)]
Merge "Use memory chunks for monitors on LP64"

10 years agoUse memory chunks for monitors on LP64
Andreas Gampe [Thu, 17 Apr 2014 17:35:09 +0000 (10:35 -0700)]
Use memory chunks for monitors on LP64

Monitor IDs in lock words are only 30b. On a 32b system that works
fine, as memory is usually aligned enough that shifting works out.
On 64b systems, the virtual memory space is too large for that.
This adds memory chunks into which we allocate the monitors so that
we have base_addr + offset and can use the offset as the monitor ID.
To allow for relatively compact but growable storage, we use a list
of chunks.

Added a global lock for the monitor pool.

Change-Id: I0e290c4914a2556e0b2eef9902422d7c4dcf536d

10 years agoMerge "Missed use of android_atomic and thread state_."
Ian Rogers [Thu, 10 Jul 2014 06:49:02 +0000 (06:49 +0000)]
Merge "Missed use of android_atomic and thread state_."

10 years agoMissed use of android_atomic and thread state_.
Ian Rogers [Thu, 10 Jul 2014 05:02:36 +0000 (22:02 -0700)]
Missed use of android_atomic and thread state_.

Move to using art::Atomic, add necessary FetchAnd... operations to art::Atomic.

Change-Id: I32f1cdc4e0a2037b73f459bf4bb4d544f357f41b

10 years agoMerge "Fix tracing."
Ian Rogers [Thu, 10 Jul 2014 06:37:47 +0000 (06:37 +0000)]
Merge "Fix tracing."

10 years agoFix tracing.
Ian Rogers [Thu, 10 Jul 2014 06:16:06 +0000 (23:16 -0700)]
Fix tracing.

Change-Id: If6837270baec694c00cc1884bae0f1842d49da75

10 years agoMerge "x86_64: enable Peek and Poke intrinsics"
Ian Rogers [Wed, 9 Jul 2014 15:26:09 +0000 (15:26 +0000)]
Merge "x86_64: enable Peek and Poke intrinsics"

10 years agoMerge "Move another field away from android_atomic_cas."
Ian Rogers [Thu, 10 Jul 2014 03:36:51 +0000 (03:36 +0000)]
Merge "Move another field away from android_atomic_cas."

10 years agoMove another field away from android_atomic_cas.
Ian Rogers [Thu, 10 Jul 2014 01:00:50 +0000 (18:00 -0700)]
Move another field away from android_atomic_cas.

Change-Id: If63aa2811e06ec401a601286a3bacb62a0da96ad

10 years agoMerge "Allow method tracing for run-test"
Ian Rogers [Wed, 9 Jul 2014 18:58:43 +0000 (18:58 +0000)]
Merge "Allow method tracing for run-test"

10 years agoMerge "Fix method tracing from command-line"
Ian Rogers [Wed, 9 Jul 2014 18:52:43 +0000 (18:52 +0000)]
Merge "Fix method tracing from command-line"

10 years agoMerge "ART: Refactor frontend.cc"
Andreas Gampe [Thu, 10 Jul 2014 08:50:46 +0000 (08:50 +0000)]
Merge "ART: Refactor frontend.cc"

10 years agoART: Refactor frontend.cc
Andreas Gampe [Thu, 10 Jul 2014 08:43:08 +0000 (01:43 -0700)]
ART: Refactor frontend.cc

Refactor frontend.cc. Pull out flags, merge them in arrays keyed by
the instruction set. Simplify the checks and application of flags.

Change-Id: I12d5216df8d1f12e7fbe39d8132e4725c55bc8e7

10 years agoMerge "art: fix host dex2oat runtime args"
Colin Cross [Thu, 10 Jul 2014 06:10:48 +0000 (06:10 +0000)]
Merge "art: fix host dex2oat runtime args"

10 years agoart: fix host dex2oat runtime args
Colin Cross [Thu, 10 Jul 2014 05:14:12 +0000 (22:14 -0700)]
art: fix host dex2oat runtime args

The runtime arguments need to be prefixed with -Xmx or -Xms.
This only worked before because DEX2OAT_XMS, etc. were blank
due to bugs in build/, which resulted in --runtime-arg --runtime-arg
and anything starting with -- as the argument to --runtime-arg is
silently ignored.

Change-Id: I18607a6fd8d0004466404045f8dc4495eb6ab275

10 years agoMerge "Revert "Add implicit null and stack checks for x86""
Dave Allison [Thu, 10 Jul 2014 01:55:25 +0000 (01:55 +0000)]
Merge "Revert "Add implicit null and stack checks for x86""

10 years agoRevert "Add implicit null and stack checks for x86"
Dave Allison [Thu, 10 Jul 2014 01:54:57 +0000 (01:54 +0000)]
Revert "Add implicit null and stack checks for x86"

It breaks cross compilation with x86_64.

This reverts commit 34e826ccc80dc1cf7c4c045de6b7f8360d504ccf.

Change-Id: I34ba07821fc0a022fda33a7ae21850957bbec5e7

10 years agoAdd implicit null and stack checks for x86
Dave Allison [Thu, 29 May 2014 15:20:04 +0000 (08:20 -0700)]
Add implicit null and stack checks for x86

This adds compiler and runtime changes for x86
implicit checks.  32 bit only.

Both host and target are supported.
By default, on the host, the implicit checks are null pointer and
stack overflow.  Suspend is implemented but not switched on.

Change-Id: I88a609e98d6bf32f283eaa4e6ec8bbf8dc1df78a

10 years agoMerge "Remove legacy CAS implementations from mutex."
Ian Rogers [Thu, 10 Jul 2014 01:37:50 +0000 (01:37 +0000)]
Merge "Remove legacy CAS implementations from mutex."

10 years agoRemove legacy CAS implementations from mutex.
Ian Rogers [Wed, 9 Jul 2014 06:50:26 +0000 (23:50 -0700)]
Remove legacy CAS implementations from mutex.

Removes the use of __sync_bool_compare_and_swap and android_atomic_cas and uses
intention revealing atomic operations from art::Atomic (which will eventually
give way to std::atomic).

Change-Id: Iea44e1923f6706ec04b5459fe25427282c189a7e

10 years agoMerge "ART: Check slow_paths_.Size() every time"
Ian Rogers [Wed, 9 Jul 2014 20:31:18 +0000 (20:31 +0000)]
Merge "ART: Check slow_paths_.Size() every time"

10 years agoART: Check slow_paths_.Size() every time
Chao-ying Fu [Tue, 8 Jul 2014 00:13:52 +0000 (17:13 -0700)]
ART: Check slow_paths_.Size() every time

This patch fixes a bug, when a new slow path is created
during slowpath->Compile().

Change-Id: I4896a82781102694c25f4483112c6de3c56e072c
Signed-off-by: Chao-ying Fu <chao-ying.fu@intel.com>
10 years agoMerge "x86_64: GenInlinedCas must use wide rl_src_offset under 64-bit targets"
Ian Rogers [Wed, 9 Jul 2014 20:24:26 +0000 (20:24 +0000)]
Merge "x86_64: GenInlinedCas must use wide rl_src_offset under 64-bit targets"

10 years agox86_64: GenInlinedCas must use wide rl_src_offset under 64-bit targets
Chao-ying Fu [Wed, 9 Jul 2014 18:32:31 +0000 (11:32 -0700)]
x86_64: GenInlinedCas must use wide rl_src_offset under 64-bit targets

This patch fixes to use wide rl_src_offset for int and long types
under 64-bit targets, and fixes movzx8 and movsx8 to use r8_form
on the second register only.

Change-Id: Ib8c0756609100f9bc5c228f1eb391421416f3af6
Signed-off-by: Chao-ying Fu <chao-ying.fu@intel.com>
10 years agoMerge "Add notion of released vs empty pages to ROSAlloc."
Mathieu Chartier [Thu, 10 Jul 2014 00:32:52 +0000 (00:32 +0000)]
Merge "Add notion of released vs empty pages to ROSAlloc."

10 years agoAdd notion of released vs empty pages to ROSAlloc.
Mathieu Chartier [Tue, 24 Jun 2014 21:48:59 +0000 (14:48 -0700)]
Add notion of released vs empty pages to ROSAlloc.

A notion of released vs empty pages helps get a more accurate view of
how much memory was released during heap trimming. Otherwise we get
that the same pages possibly get madvised multiple times without
getting dirtied.

Also enabled heap trimming of rosalloc spaces even when we care about
jank. This is safe to do since the trimming process only acquires
locks for short periods of time.

Dalvik PSS reduces from ~52M to ~50M after boot on N4.

Bug: 9969166

Change-Id: I4012e0a2554f413d18efe1a0371fe18d1edabaa9

10 years agoMerge "Add implicit null and stack checks for x86"
Dave Allison [Thu, 10 Jul 2014 00:16:22 +0000 (00:16 +0000)]
Merge "Add implicit null and stack checks for x86"

10 years agoMerge "Move card table away from android_atomic_cas."
Ian Rogers [Thu, 10 Jul 2014 00:09:25 +0000 (00:09 +0000)]
Merge "Move card table away from android_atomic_cas."

10 years agoMove card table away from android_atomic_cas.
Ian Rogers [Wed, 9 Jul 2014 05:55:18 +0000 (22:55 -0700)]
Move card table away from android_atomic_cas.

For x86 use byte CAS operations for byte CAS.
Fix bug in ModifyCardsAtomic where CAS was 32-bit instead of 64-bit.

Change-Id: Ieb3fe695b4699750abf04642b0abe94103976817

10 years agox86_64: enable Peek and Poke intrinsics
Alexei Zavjalov [Tue, 8 Jul 2014 09:27:17 +0000 (16:27 +0700)]
x86_64: enable Peek and Poke intrinsics

This implements intrinsics for:
Memory.peekByte/Short/Int/Long()
Memory.pokeByte/Short/Int/Long()

Change-Id: I6da6250f262dfd7aded35c2e3ade2d0916bd73cb
Signed-off-by: Alexei Zavjalov <alexei.zavjalov@intel.com>
10 years agoFix method tracing from command-line
Sebastien Hertz [Tue, 17 Dec 2013 09:42:03 +0000 (10:42 +0100)]
Fix method tracing from command-line

Transitions current thread to the new kWaitingForMethodTracingStart thread
state when starting method tracing.

Ensures there is a current thread when method tracing is stopped due to runtime
shutdown. If the current thread has been detached, we now re-attach it.
Note: we only do this if method tracing has been activated from command-line.

Fixes instrumentation when forcing interpreter mode (-Xint) with method tracing
enabled.

Removes unused parameter from UnsafeLogFatalForThreadSuspendAllTimeout.

Bug: https://code.google.com/p/android/issues/detail?id=72094
Bug: 11683397
Change-Id: I70f000fb46ddd95d6ad51ea0a8eee77697a045e9

10 years agoAllow method tracing for run-test
Sebastien Hertz [Wed, 9 Jul 2014 13:59:05 +0000 (15:59 +0200)]
Allow method tracing for run-test

Adds "--trace" option to enable method tracing.

Bug: 11683397
Change-Id: I20a6b25c71067eafd848db26f13d62cfdb9a6159

10 years agoMerge "CW on Master: Propagate or throw exception when no class found happens in...
Brian Carlstrom [Wed, 9 Jul 2014 23:50:28 +0000 (23:50 +0000)]
Merge "CW on Master: Propagate or throw exception when no class found happens in interpreter."

10 years agoCW on Master: Propagate or throw exception when no class found happens in interpreter.
Sean Wan [Wed, 9 Jul 2014 19:08:29 +0000 (12:08 -0700)]
CW on Master: Propagate or throw exception when no class found happens in interpreter.

The old behavior is a check fail which causes zygote process crash.

This is particular a problem for CW in which webview is not used, and
this stops CW system start.

(cherry picked from commit 41a71f3db62e5bccb162a2b18ed3801e00ff6f87)

Change-Id: Iabe091ebe4bbdd86d931ac6c06abd089f1338d59

10 years agoMerge "Use the icu4c headers from the new location."
Elliott Hughes [Wed, 9 Jul 2014 22:54:14 +0000 (22:54 +0000)]
Merge "Use the icu4c headers from the new location."

10 years agoUse the icu4c headers from the new location.
Elliott Hughes [Wed, 9 Jul 2014 22:31:20 +0000 (15:31 -0700)]
Use the icu4c headers from the new location.

Change-Id: I6ca56e8f9909bf3644a5f79f2d705547ddba6d03

10 years agoMerge "ART: Add simple tests for inlining of CAS"
Andreas Gampe [Wed, 9 Jul 2014 22:18:27 +0000 (22:18 +0000)]
Merge "ART: Add simple tests for inlining of CAS"

10 years agoART: Add simple tests for inlining of CAS
Andreas Gampe [Wed, 9 Jul 2014 04:33:47 +0000 (21:33 -0700)]
ART: Add simple tests for inlining of CAS

Add simple test cases for the inlining of CAS in the quick compiler to
run-test 082. The tests are not multi-threaded and will just establish
that the baseline behavior is correct. For extensive evaluation consider
tests available in libcore.

Change-Id: I9f463599e48ab7abc725769dda84758c9c6a76c2

10 years agoMerge "Add easy way to ensure the next allocation does GC."
Mathieu Chartier [Wed, 9 Jul 2014 22:10:01 +0000 (22:10 +0000)]
Merge "Add easy way to ensure the next allocation does GC."

10 years agoAdd easy way to ensure the next allocation does GC.
Mathieu Chartier [Wed, 9 Jul 2014 00:46:19 +0000 (17:46 -0700)]
Add easy way to ensure the next allocation does GC.

Added a class called ScopedHeapFill which changes the bytes allocated
counter to be equal to the growth limit. This causes the next
allocation to do a GC and possibly generate an OOM error. This is
useful for tests which need GC to happen at specific point.

Change-Id: Ibd8f3d5928b58534c5165ba7c296980002aa2c28

10 years agoMerge "Remove incorrect check for sa_mask in signal chaining"
Dave Allison [Wed, 9 Jul 2014 21:45:13 +0000 (21:45 +0000)]
Merge "Remove incorrect check for sa_mask in signal chaining"