OSDN Git Service

android-x86/art.git
7 years agoRevert "Use try lock to fix class resolution race"
Mathieu Chartier [Fri, 29 Jul 2016 20:06:45 +0000 (20:06 +0000)]
Revert "Use try lock to fix class resolution race"

This reverts commit a704eda0078989a73cac111ed309aca50d2e289b.

Bug: 27417671
Bug: 30500547
Change-Id: Ieea05236b9e61c722660cd9497c9d55d13ccd010

7 years agoUse try lock to fix class resolution race
Mathieu Chartier [Wed, 13 Jul 2016 16:53:35 +0000 (09:53 -0700)]
Use try lock to fix class resolution race

There was some possible deadlocks related to EnsureResolved caused by
acquiring an object lock.

Scenario:
Thread 1 acquires lock on obj1
Thread 1 begins to resolve / initialize class1
Thread 1 blocks since it sees that class1 is already being resolved and
gets preempted before it can acquire the object lock on class1
Thread 2 finishes resolving and initializing class1 and locks class1
Thread 2 blocks attempting to lock obj1
Thread 1 blocks attempting to lock class1
Deadlock

Fixed the deadlock by changing EnsureResolved to use a try lock for the
unresolved case.

Added a test.

Test: Device boot, test-art-host, monitor_test

Bug: 27417671
Change-Id: Ic6e1c3ca6f45490cf8a7bf8e137dee71ac83ff64

7 years agoMerge "Fix profiling output to properly display class info."
Treehugger Robot [Wed, 27 Jul 2016 18:38:39 +0000 (18:38 +0000)]
Merge "Fix profiling output to properly display class info."

7 years agoFix profiling output to properly display class info.
Jeff Hao [Wed, 27 Jul 2016 17:35:03 +0000 (10:35 -0700)]
Fix profiling output to properly display class info.

The profiler stores class def indexes, not type ids.

Change-Id: Ieec9374a2ba381387ed444f376922478183145f8

7 years agoMerge "Add missing filter cards to zygote mod union table"
Mathieu Chartier [Tue, 26 Jul 2016 19:54:28 +0000 (19:54 +0000)]
Merge "Add missing filter cards to zygote mod union table"

7 years agoAdd missing filter cards to zygote mod union table
Mathieu Chartier [Tue, 26 Jul 2016 00:48:52 +0000 (17:48 -0700)]
Add missing filter cards to zygote mod union table

Without filtering the cards, every object in the zygote is grayed
during the GC. This was deleted in a PS of previous CL.

GrayAllDirtyImmuneObjects goes from 1.974ms to 376.464us for CC on
N9 maps. This happens during the pause. This CL also fixes regression
in zygote PSS.

Bug: 29516968
Bug: 12687968

Change-Id: I42014e78b1de3ce9b2eefa3bd32f0d24e2ff71c6

7 years agoMerge "Improve CC handling for immune objects"
Treehugger Robot [Mon, 25 Jul 2016 23:27:42 +0000 (23:27 +0000)]
Merge "Improve CC handling for immune objects"

7 years agoMerge "Remove redundant 601 test, it's all in 600."
Aart Bik [Mon, 25 Jul 2016 23:13:18 +0000 (23:13 +0000)]
Merge "Remove redundant 601 test, it's all in 600."

7 years agoMerge "runtime: Refactor asm_support to be auto-generated (mostly)"
Treehugger Robot [Mon, 25 Jul 2016 23:06:39 +0000 (23:06 +0000)]
Merge "runtime: Refactor asm_support to be auto-generated (mostly)"

7 years agoImprove CC handling for immune objects
Mathieu Chartier [Fri, 22 Jul 2016 17:47:45 +0000 (10:47 -0700)]
Improve CC handling for immune objects

Currently we reduce ram for immune objects by racing agianst the
mutators to try and finish processing them before the mutators
change many objects to gray. However there is still a window of time
where the mutator can dirty immune pages by changing the lock words
to gray. These pages remain dirty for the lifetime of the app.

This CL changes uses the FlipCallback pause to gray all of the
immune objects that have a dirty card. Once these objects are all
gray we don't to gray any more objects in the immune spaces since
these objects are the only ones that may reference non immune
objects.

Also only scan objects that are gray when scanning immune spaces to
reduce scanning time.

System wide PSS after boot on N9, before:
61668 kB: .art mmap
11249 kB: .Zygote

After:
36013 kB: .art mmap
12251 kB: .Zygote

Results are better than demonstrated since there are more apps
running after.

Maps PSS / Private Dirty, before:
.art mmap     3703     3116
  .Zygote      577      480

After:
.art mmap     1655     1092
  .Zygote      476      392

System server before:
.art mmap     4453     3956
  .Zygote      849      780

After:
.art mmap     2326     1748
  .Zygote      640      564

EAAC:
Before:
ScanImmuneSpaces takes 669.434ms GC time
Scores: 718, 761, 753 average 744
GC time: 4.2s, 4.35s, 4.3s average 4.28s

After:
ScanImmuneSpaces takes 138.328ms GC time
Scores: 731, 730, 704 average 722
GC time: 3.92s, 3.83s, 3.85s average 3.87s

Additional GC pause time is 285us on Maps on N9.
TODO: Reduce this pause time.

Test: N9 booting, test-art-host, EAAC all run with CC

Bug: 29516968
Bug: 12687968

Change-Id: I584b10d017547b321f33eb23fb5d64372af6f69c

7 years agoMerge "ART: Add pointer-size template to some reflection functions"
Treehugger Robot [Mon, 25 Jul 2016 22:24:42 +0000 (22:24 +0000)]
Merge "ART: Add pointer-size template to some reflection functions"

7 years agoruntime: Refactor asm_support to be auto-generated (mostly)
Igor Murashkin [Fri, 22 Jul 2016 22:59:16 +0000 (15:59 -0700)]
runtime: Refactor asm_support to be auto-generated (mostly)

Usage:

* If the defines are no longer up to date, re-run generate-asm-support

* To add a new field:
--- Edit one of the offset_ or constant def files.
--- Rebuild cpp-define-generator and re-run generate-asm-support

Change-Id: I772430fcf1ad9af40898ebb453848f8412612386
TODO: Integrate this into the build somehow
TODO: Account for 32 vs 64-bit and read barrier vs no read barrier

7 years agoRemove redundant 601 test, it's all in 600.
Aart Bik [Mon, 25 Jul 2016 22:12:24 +0000 (15:12 -0700)]
Remove redundant 601 test, it's all in 600.

Rationale:
Nicolas, I don't understand why you duplicated this test.
Since all (and now more) is in 600, removing this clone.

Change-Id: Ibc8b47549f054972f07b53b1554e7ab4dc0cb33e

7 years agoMerge "Make java 8 run-tests runnable on the RI again."
Treehugger Robot [Mon, 25 Jul 2016 21:31:19 +0000 (21:31 +0000)]
Merge "Make java 8 run-tests runnable on the RI again."

7 years agoMerge "Launch ahat server before processing the heap dump."
Treehugger Robot [Mon, 25 Jul 2016 20:18:25 +0000 (20:18 +0000)]
Merge "Launch ahat server before processing the heap dump."

7 years agoART: Add pointer-size template to some reflection functions
Andreas Gampe [Mon, 25 Jul 2016 20:06:04 +0000 (13:06 -0700)]
ART: Add pointer-size template to some reflection functions

The unstarted runtime may run code for a different pointer size,
even when no transaction is active (e.g., during startup). To
retain performance when the runtime is up and executing under
normal conditions, add a template parameter and use sizeof(void*)
in places where it is adequate.

For maintainability, it is necessary to drop the default for
the transaction template parameter. Implicit conversions from
bool to size_t may lead to incorrect code and hard to diagnose
problems. So instead ensure that all callers must give all
template parameter values.

Test: m test-art-host
Change-Id: I3076883422c8553ede4de5642409c5684a5a9aa8

7 years agoMerge "Fix stream-tracing with default methods"
Treehugger Robot [Mon, 25 Jul 2016 18:44:11 +0000 (18:44 +0000)]
Merge "Fix stream-tracing with default methods"

7 years agoMerge "Make static helper methods member functions of OptimizingCompiler class"
Aart Bik [Mon, 25 Jul 2016 17:53:51 +0000 (17:53 +0000)]
Merge "Make static helper methods member functions of OptimizingCompiler class"

7 years agoFix stream-tracing with default methods
Alex Light [Tue, 19 Jul 2016 18:21:32 +0000 (11:21 -0700)]
Fix stream-tracing with default methods

Test: Ran mma ART_TEST_TRACE=true ART_TEST_TRACE_STREAM=true -j40 test-art-host

Bug: 30102284
Bug: 30286605

Change-Id: I12f2534bc2d42a7504e6c1972c4cfdda2f77286f

7 years agoMerge "Revert "Revert "ART: Add dex2oat swap usage test"""
Treehugger Robot [Mon, 25 Jul 2016 17:27:55 +0000 (17:27 +0000)]
Merge "Revert "Revert "ART: Add dex2oat swap usage test"""

7 years agoMerge "Add fast path to arm64 READ_BARRIER macro"
Mathieu Chartier [Mon, 25 Jul 2016 16:44:33 +0000 (16:44 +0000)]
Merge "Add fast path to arm64 READ_BARRIER macro"

7 years agoRevert "Revert "ART: Add dex2oat swap usage test""
Andreas Gampe [Mon, 25 Jul 2016 15:27:35 +0000 (08:27 -0700)]
Revert "Revert "ART: Add dex2oat swap usage test""

Now correctly ignores the target (but with enough infrastructure
to introduce this some day).

This reverts commit ec743ffd47ac9d3c10a449926d78c2eb51e5208e.

Bug: 29259363
Test: m test-art-host-gtest-dex2oat_test
Test: m test-art-target-gtest-dex2oat_test (shamu)
Change-Id: I11b4e755bc8cb1e2eea29cd006e8df67df632c00

7 years agoMerge "Do not emit stack maps for runtime calls to ReadBarrierMarkRegX."
Roland Levillain [Mon, 25 Jul 2016 14:49:51 +0000 (14:49 +0000)]
Merge "Do not emit stack maps for runtime calls to ReadBarrierMarkRegX."

7 years agoMerge "Revert "ART: Add dex2oat swap usage test""
Treehugger Robot [Sat, 23 Jul 2016 06:04:13 +0000 (06:04 +0000)]
Merge "Revert "ART: Add dex2oat swap usage test""

7 years agoRevert "ART: Add dex2oat swap usage test"
Andreas Gampe [Sat, 23 Jul 2016 05:17:18 +0000 (05:17 +0000)]
Revert "ART: Add dex2oat swap usage test"

The test doesn't correctly exclude the target, as it will only work
on the host.

This reverts commit ab2a54093386c85756fe78daedd11ff4408a5988.

Bug: 29259363
Change-Id: Ie50df2e6f0c63cb10359c3862ad44ee3c03d4e3b

7 years agoAdd fast path to arm64 READ_BARRIER macro
Mathieu Chartier [Thu, 21 Jul 2016 21:59:04 +0000 (14:59 -0700)]
Add fast path to arm64 READ_BARRIER macro

EAAC benchmark time from 978.7857143ms to 969.5714286ms on N9 based
on 42 samples. Reduces artReadBarrierSlow calls from 9M to 1M.

Not a huge improvement since we were already checking the lock word in
ReadBarrier::Barrier.

Test: N9 boots, test-art-host, EEAC runs. (All with CC enabled).

Bug: 30162165
Bug: 12687968

Change-Id: Ifb97b52ea84e21c7df83addfb91c5f05f41db32d

7 years agoMake static helper methods member functions of OptimizingCompiler class
Wojciech Staszkiewicz [Fri, 22 Jul 2016 20:33:11 +0000 (13:33 -0700)]
Make static helper methods member functions of OptimizingCompiler class

Make RunOptimizations, MaybeRunInliner and RunArchOptimizations member
functions of OptimizingCompiler class.

Both versions of RunOptimizations are protected in preparation for
bisection bug search CL.

Change-Id: I596efa9ed3fccd1ed3798c6427cc166e2a5d28bd

7 years agoMerge "ART: Change run-*-tests to ANDROID_{HOST|PRODUCT}_OUT"
Treehugger Robot [Fri, 22 Jul 2016 22:32:08 +0000 (22:32 +0000)]
Merge "ART: Change run-*-tests to ANDROID_{HOST|PRODUCT}_OUT"

7 years agoMerge "ART: Add dex2oat swap usage test"
Treehugger Robot [Fri, 22 Jul 2016 22:31:37 +0000 (22:31 +0000)]
Merge "ART: Add dex2oat swap usage test"

7 years agoMerge "Combine offsets in loop-based dynamic BCE."
Treehugger Robot [Fri, 22 Jul 2016 18:43:01 +0000 (18:43 +0000)]
Merge "Combine offsets in loop-based dynamic BCE."

7 years agoDo not emit stack maps for runtime calls to ReadBarrierMarkRegX.
Roland Levillain [Fri, 22 Jul 2016 16:10:06 +0000 (17:10 +0100)]
Do not emit stack maps for runtime calls to ReadBarrierMarkRegX.

* Boot image code size variation on Nexus 5X
  (aosp_bullhead-userdebug build):
  - total ARM64 framework Oat files size change:
    115584120 bytes -> 109124728 bytes (-5.59%)
  - total ARM framework Oat files size change:
    97387728 bytes -> 92517584 (-5.00%)

Test: ART host and target (ARM, ARM64) tests.
Bug: 29506760
Bug: 12687968
Change-Id: I979d9fb2b4e09f4c0c7bf33af2cd91750a67f989

7 years agoCombine offsets in loop-based dynamic BCE.
Aart Bik [Fri, 15 Jul 2016 00:19:43 +0000 (17:19 -0700)]
Combine offsets in loop-based dynamic BCE.

Rationale:
Similar to what I did recently for dom-based dynamic BCE, this
CL combines offsets for the tests generated for loop-based
dynamic BCE. For a set of n references, this reduces the
number of generated tests from 2*n+1 down to at most 4
(in some cases even less).

TEST: 530-checker-loops3

BUG=27430379

Change-Id: Ic80c2563eaae23f514c1fd52965dd83bccb9d190

7 years agoMerge "ARM64/x86-64: Fix mterp fill-array-data-payload pointer calculation."
Treehugger Robot [Fri, 22 Jul 2016 16:53:29 +0000 (16:53 +0000)]
Merge "ARM64/x86-64: Fix mterp fill-array-data-payload pointer calculation."

7 years agoARM64/x86-64: Fix mterp fill-array-data-payload pointer calculation.
Vladimir Marko [Fri, 22 Jul 2016 09:52:24 +0000 (10:52 +0100)]
ARM64/x86-64: Fix mterp fill-array-data-payload pointer calculation.

Fix the pointer calculation to sign-extend the offset
instead of zero-extending it, just like we do for the switch
table pointer calculation. Clean up comments for the switch.

Test: Additional test in 412-new-array.
Change-Id: Ibb1d2d3fcb109f59280aca08de21e42edc4ce66b

7 years agoMerge "ARM64: Improve mterp cmpl/cmpg."
Vladimir Marko [Fri, 22 Jul 2016 09:24:41 +0000 (09:24 +0000)]
Merge "ARM64: Improve mterp cmpl/cmpg."

7 years agoMerge changes I295c7876,Ib4b84b7b
Treehugger Robot [Thu, 21 Jul 2016 23:58:00 +0000 (23:58 +0000)]
Merge changes I295c7876,Ib4b84b7b

* changes:
  ART: Remove PACKED from ArtMethod's ptr_sized_fields_
  ART: Rename ArtMethod JNI field

7 years agoMerge "Use non-CAS thread flip root visitor."
Hiroshi Yamauchi [Thu, 21 Jul 2016 23:24:44 +0000 (23:24 +0000)]
Merge "Use non-CAS thread flip root visitor."

7 years agoART: Remove PACKED from ArtMethod's ptr_sized_fields_
Andreas Gampe [Wed, 20 Jul 2016 01:27:17 +0000 (18:27 -0700)]
ART: Remove PACKED from ArtMethod's ptr_sized_fields_

Remove the PACKED(4) hack, as it's highly annoying when debugging
a 64-bit process. Instead, fix the actual offset and size computation
for cross-size accesses.

Test: m test-art-host
Change-Id: I295c78760b74b6a62946e76856f218b4eb159cdc

7 years agoART: Add dex2oat swap usage test
Andreas Gampe [Thu, 21 Jul 2016 22:36:22 +0000 (15:36 -0700)]
ART: Add dex2oat swap usage test

Add a test that checks that native alloc size goes down when using
swap, as an indication that we're actually effectively using swap.

Bug: 29259363
Test: m test-art-host-gtest-dex2oat_test
Change-Id: Ifa5aef1b97696309283de78be08699a6399a4d1d

7 years agoLaunch ahat server before processing the heap dump.
Richard Uhler [Thu, 21 Jul 2016 20:52:48 +0000 (13:52 -0700)]
Launch ahat server before processing the heap dump.

Change-Id: Ic8f18ad3010cb0937f5ef68228359258ec4793fd
Test: Manually launch ahat on a large heap dump with ahat already
      running. Verify that ahat fails immediately.
Test: Manually launch ahat on a large heap dump without ahat already
      running. Verify that visiting localhost:7100 immediately
      eventually resolves to the overview page rather than say the
      site can't be reached.
Bug: 28611294

7 years agoUse non-CAS thread flip root visitor.
Hiroshi Yamauchi [Thu, 21 Jul 2016 03:25:27 +0000 (20:25 -0700)]
Use non-CAS thread flip root visitor.

We don't need to use CAS to update the thread-local GC roots for the
thread flip.

Bug: 12687968
Bug: 29517059
Test: libartd.so boot. ART tests. Ritzperf EAAC.
Change-Id: Ia2acab824f756bd7d2ad501b2040233e0d394356

7 years agoART: Rename ArtMethod JNI field
Andreas Gampe [Tue, 19 Jul 2016 15:06:07 +0000 (08:06 -0700)]
ART: Rename ArtMethod JNI field

The field is multi-purpose, rename it to data and clean up
accessors in preparation of more checks.

Test: m test-art-host
Change-Id: Ib4b84b7b1a51ca201544bc488ce8770aa858c7fd

7 years agoMerge "ART: Make run-test temp dir consistent"
Treehugger Robot [Thu, 21 Jul 2016 18:26:27 +0000 (18:26 +0000)]
Merge "ART: Make run-test temp dir consistent"

7 years agoART: Make run-test temp dir consistent
Andreas Gampe [Thu, 21 Jul 2016 04:09:29 +0000 (21:09 -0700)]
ART: Make run-test temp dir consistent

We use the username as a directory component in run-test. Use the
same when driven through the Makefile.

Drop the username in run-test when TMP_DIR is set.

Test: m test-art-host-run-test
Test: art/test/run-test --host 001-HelloWorld
Change-Id: I060997ffbd80cd4da30dd6ac8d3954641de3292b

7 years agoMerge "Add a way to measure read barrier slow paths"
Mathieu Chartier [Thu, 21 Jul 2016 16:52:44 +0000 (16:52 +0000)]
Merge "Add a way to measure read barrier slow paths"

7 years agoARM64: Improve mterp cmpl/cmpg.
Vladimir Marko [Thu, 21 Jul 2016 11:59:46 +0000 (12:59 +0100)]
ARM64: Improve mterp cmpl/cmpg.

Use CSET+CNEG instead of MOV+CNEG+CSEL. Prefer the
CNEG/CSET alias over the CSNEG/CSINC for readability.

Test: Run ART test suite on Nexus 9 with the interpreter.
Change-Id: I5c4fb0cf2c053904253e8e82f3e7e05c774b0583

7 years agoMerge "Change return type of artIsAssignableFromCode for MIPS64"
Treehugger Robot [Thu, 21 Jul 2016 14:19:33 +0000 (14:19 +0000)]
Merge "Change return type of artIsAssignableFromCode for MIPS64"

7 years agoMerge "Fix the definition of MACRO_LITERAL for OS X on x86-64."
Roland Levillain [Thu, 21 Jul 2016 14:17:40 +0000 (14:17 +0000)]
Merge "Fix the definition of MACRO_LITERAL for OS X on x86-64."

7 years agoFix the definition of MACRO_LITERAL for OS X on x86-64.
Roland Levillain [Tue, 5 Jul 2016 17:55:32 +0000 (18:55 +0100)]
Fix the definition of MACRO_LITERAL for OS X on x86-64.

Test: "ART_USE_READ_BARRIER=true mmma art" on OS X.
Change-Id: Ia2d4c7a3eb7fec346ddfa4c7b0f7b700f1137344

7 years agoChange return type of artIsAssignableFromCode for MIPS64
Goran Jakovljevic [Thu, 21 Jul 2016 12:21:46 +0000 (14:21 +0200)]
Change return type of artIsAssignableFromCode for MIPS64

This has been missed by Ic505befc6c94e2dccbc8abf2b13d4c2d662e68d1.
This fixes MIPS64 build.

Bug: 30232671
Test: make -j 32 out/target/product/generic_mips64/obj/SHARED_LIBRARIES/libart_intermediates/arch/mips64/entrypoints_init_mips64.o
Change-Id: Iec89d25e2d38c6efc0d1025767d0ac2a8bdb7dcd

7 years agoMerge "Move caller-saves saving/restoring to ReadBarrierMarkRegX."
Roland Levillain [Thu, 21 Jul 2016 12:11:15 +0000 (12:11 +0000)]
Merge "Move caller-saves saving/restoring to ReadBarrierMarkRegX."

7 years agoMerge "Clean up Class::FindStaticField()."
Vladimir Marko [Thu, 21 Jul 2016 10:33:13 +0000 (10:33 +0000)]
Merge "Clean up Class::FindStaticField()."

7 years agoMove caller-saves saving/restoring to ReadBarrierMarkRegX.
Roland Levillain [Wed, 20 Jul 2016 10:32:19 +0000 (11:32 +0100)]
Move caller-saves saving/restoring to ReadBarrierMarkRegX.

Instead of saving/restoring live caller-save registers
before/after the call to read barrier mark entry points
ReadBarrierMarkRegX, have these entry points save/restore
all the caller-save registers themselves (except register
rX, which contains the return value).

Also refactor the assembly code of these entry points
using macros.

* Boot image code size variation on Nexus 5X
  (aosp_bullhead-userdebug build):
  - total ARM64 framework Oat files size change:
    119196792 bytes -> 115575920 bytes (-3.04%)
  - total ARM framework Oat files size change:
    100435212 bytes -> 97621188 bytes (-2.80%)

* Benchmarks (ARM64) score variations on Nexus 5X
  (aosp_bullhead-userdebug build):
  - RitzPerf (lower is better)
    - average score difference: -2.71%
  - CaffeineMark (higher is better)
    - no real difference for most tests
      (absolute variation lower than 1%)
    - better score on the "Method" benchmark:
      score variation 41253 -> 44891 (+8.82%)

Test: ART host and target (ARM, ARM64) tests.
Bug: 29506760
Bug: 12687968
Change-Id: I881bf73139a3f1c2bee9ffc6fc8c00f9a392afa6

7 years agoMerge "ARM: Port instr simplification of array accesses."
Vladimir Marko [Thu, 21 Jul 2016 09:17:15 +0000 (09:17 +0000)]
Merge "ARM: Port instr simplification of array accesses."

7 years agoARM: Port instr simplification of array accesses.
Artem Serov [Wed, 6 Jul 2016 15:23:04 +0000 (16:23 +0100)]
ARM: Port instr simplification of array accesses.

After changing the addressing mode for array accesses (in
https://android-review.googlesource.com/248406) the 'add'
instruction that calculates the base address for the array can be
shared across accesses to the same array.

Before https://android-review.googlesource.com/248406:
    add IP, r[Array], r[Index0], LSL #2
    ldr r0, [IP, #12]
    add IP, r[Array], r[Index1], LSL #2
    ldr r0, [IP, #12]

Before this CL:
    add IP. r[Array], #12
    ldr r0, [IP, r[Index0], LSL #2]
    add IP. r[Array], #12
    ldr r0, [IP, r[Index1], LSL #2]

After this CL:
    add IP. r[Array], #12
    ldr r0, [IP, r[Index0], LSL #2]
    ldr r0, [IP, r[Index1], LSL #2]

Link to the original optimization:
    https://android-review.googlesource.com/#/c/127310/

Test: Run ART test suite on Nexus 6.
Change-Id: Iee26f9a0a7ca46abb90e3f60d19d22dc8dee4d8f

7 years agoMerge changes Ibcc11ce7,I9867dc11
Vladimir Marko [Thu, 21 Jul 2016 09:15:23 +0000 (09:15 +0000)]
Merge changes Ibcc11ce7,I9867dc11

* changes:
  ARM64: Improve Mterp.
  ARM64: Fix mterp switch table pointer calculation.

7 years agoAdd a way to measure read barrier slow paths
Mathieu Chartier [Thu, 14 Jul 2016 20:30:03 +0000 (13:30 -0700)]
Add a way to measure read barrier slow paths

If enabled, this option counts number of slow paths, measures the
total slow path time per GC and records the info into a histogram.
Also added support for systrace to see which threads are performing
slow paths.

Added runtime option -Xgc:measure to enable. The info is dumped
for SIGQUIT.

Test: Volantis boot with CC, test-art-host with CC, run EEAC with CC
and -Xgc:measure

Bug: 30162165

Change-Id: I3c2bdb4156065249c45695f13c77c0579bc8e57a

7 years agoMerge "Revert "Revert "Refactor GetIMTIndex"""
Treehugger Robot [Thu, 21 Jul 2016 00:04:52 +0000 (00:04 +0000)]
Merge "Revert "Revert "Refactor GetIMTIndex"""

7 years agoMerge "Make stream tracing have a higher thread count on host"
Treehugger Robot [Wed, 20 Jul 2016 22:08:47 +0000 (22:08 +0000)]
Merge "Make stream tracing have a higher thread count on host"

7 years agoMake stream tracing have a higher thread count on host
Alex Light [Wed, 20 Jul 2016 17:43:39 +0000 (10:43 -0700)]
Make stream tracing have a higher thread count on host

Test: mma ART_TEST_TRACE_STREAM=true -j40 test-art-host-run-test-debug-prebuild-optimizing-relocate-stream-cms-checkjni-image-npictest-ndebuggable-001-HelloWorld32
Bug: 30229615
Change-Id: Id396f569b9e21ff764562005624aabc964d4e95a

7 years agoMerge "Refactor register allocation to be pluggable"
Treehugger Robot [Wed, 20 Jul 2016 20:38:30 +0000 (20:38 +0000)]
Merge "Refactor register allocation to be pluggable"

7 years agoMerge "Fix accidental pass-by-value"
Treehugger Robot [Wed, 20 Jul 2016 19:42:28 +0000 (19:42 +0000)]
Merge "Fix accidental pass-by-value"

7 years agoMerge changes I328ea51d,I577c5d02
Treehugger Robot [Wed, 20 Jul 2016 18:22:10 +0000 (18:22 +0000)]
Merge changes I328ea51d,I577c5d02

* changes:
  Compute oat and odex filenames eagerly.
  Make a static OatFileAssistant::DexLocationToOatFileName function.

7 years agoARM64: Improve Mterp.
Vladimir Marko [Wed, 20 Jul 2016 13:25:30 +0000 (14:25 +0100)]
ARM64: Improve Mterp.

Several straight-forward optimizations:
  - use ubfx instead of SHR+AND,
  - do not mask shifting distance,
  - use 64-bit LDRSH to avoid subsequent sign extension,
  - use CBNZ instead of CMP+BNE for null checks,
  - style: use SXTW alias instead of explicit SBFM.

Test: Run ART test suite on Nexus 9 with the interpreter.
Change-Id: Ibcc11ce7f455432ecb789f727da21f269f8370f0

7 years agoARM64: Fix mterp switch table pointer calculation.
Vladimir Marko [Wed, 20 Jul 2016 16:52:51 +0000 (17:52 +0100)]
ARM64: Fix mterp switch table pointer calculation.

Do not mix 32-bit and 64-bit registers with
    add     x0, xPC, w0, lsl #1
that ends up compiled as
    add     x0, xPC, w0, uxtx #1
instead of the required sxtx. Just sing-extend the offset
correctly in previous instructions.

Test: Additional test in 501-regression-packed-switch.
Change-Id: I9867dc1180743e98f9707a312241d2f5b726ca8c

7 years agoFix accidental pass-by-value
Matthew Gharrity [Wed, 20 Jul 2016 17:13:45 +0000 (10:13 -0700)]
Fix accidental pass-by-value

Change-Id: I245111eabb43368875c1215ca4f3a1f1918492fe

7 years agoRefactor register allocation to be pluggable
Matthew Gharrity [Thu, 14 Jul 2016 20:24:00 +0000 (13:24 -0700)]
Refactor register allocation to be pluggable

Allow alternate register allocation strategies to be implemented
in subclasses of a common register allocation base class.

Test: m test-art-host

Change-Id: I7c5866aa9ddff8f53fcaf721bad47654ab221b4f

7 years agoMerge "ART: Change return types of field access entrypoints"
Vladimir Marko [Wed, 20 Jul 2016 12:01:15 +0000 (12:01 +0000)]
Merge "ART: Change return types of field access entrypoints"

7 years agoMerge "Fix test after rename."
Nicolas Geoffray [Wed, 20 Jul 2016 10:53:11 +0000 (10:53 +0000)]
Merge "Fix test after rename."

7 years agoFix test after rename.
Nicolas Geoffray [Wed, 20 Jul 2016 10:49:47 +0000 (11:49 +0100)]
Fix test after rename.

bug:29964720
Change-Id: I37cebc40ca83597b159eefbc492e8cc105996306

7 years agoART: Change return types of field access entrypoints
Andreas Gampe [Wed, 20 Jul 2016 05:34:53 +0000 (22:34 -0700)]
ART: Change return types of field access entrypoints

Ensure that return types guarantee full-width data as the compiled
code and mterp expect by using size_t and ssize_t.

This fixes Clang no longer sign-/zero-extending small return types.

Bug: 30232671
Test: m ART_TEST_RUN_TEST_NDEBUG=true ART_TEST_INTERPRETER=true test-art-host-run-test
Change-Id: Ic505befc6c94e2dccbc8abf2b13d4c2d662e68d1

7 years agoMerge "JIT: Don't update the dex cache of another class loader."
Nicolas Geoffray [Wed, 20 Jul 2016 10:30:09 +0000 (10:30 +0000)]
Merge "JIT: Don't update the dex cache of another class loader."

7 years agoMerge "ARM: Change mem address mode for array accesses."
Vladimir Marko [Wed, 20 Jul 2016 09:20:55 +0000 (09:20 +0000)]
Merge "ARM: Change mem address mode for array accesses."

7 years agoARM: Change mem address mode for array accesses.
Artem Serov [Mon, 11 Jul 2016 13:02:34 +0000 (14:02 +0100)]
ARM: Change mem address mode for array accesses.

Switch from:
  add IP, r[Array], r[Index], LSL #2
  ldr r0, [IP, #12]
To:
  add IP. r[Array], #12
  ldr r0, [IP, r[Index], LSL #2]

These is a base for the future TryExtractArrayAccessAddress
optimization port to arm.

Test: aosp_shamu-userdebug boots and passes "m test-art-target".
Change-Id: I6ab01ba3271a8f79599ddd91a6b63cd1b37d2d67

7 years agoMerge "Refactor SSA deconstruction into its own class"
Treehugger Robot [Wed, 20 Jul 2016 00:02:38 +0000 (00:02 +0000)]
Merge "Refactor SSA deconstruction into its own class"

7 years agoMerge "Tune the GC ergnomics for the read barrier config."
Treehugger Robot [Tue, 19 Jul 2016 22:07:57 +0000 (22:07 +0000)]
Merge "Tune the GC ergnomics for the read barrier config."

7 years agoMerge "Disable warnings triggered in Clang r271374"
Stephen Hines [Tue, 19 Jul 2016 22:02:31 +0000 (22:02 +0000)]
Merge "Disable warnings triggered in Clang r271374"

7 years agoRevert "Revert "Refactor GetIMTIndex""
Matthew Gharrity [Tue, 19 Jul 2016 21:32:52 +0000 (21:32 +0000)]
Revert "Revert "Refactor GetIMTIndex""

Originally reverted in order to revert
https://android-review.googlesource.com/#/c/244190/
but can now be merged again.

This reverts commit d4ceecc85a5aab2ec23ea1bd010692ba8c8aaa0c.

Test: m test-art-host

Change-Id: Id9205f2b77a378fc0f06088e78c66e81a49f712d

7 years agoRefactor SSA deconstruction into its own class
Matthew Gharrity [Mon, 18 Jul 2016 20:38:44 +0000 (13:38 -0700)]
Refactor SSA deconstruction into its own class

Test: m test-art-host

Change-Id: Ie82c2802f76f27512ef922ba583caeccf5675063

7 years agoDisable warnings triggered in Clang r271374
Pirama Arumuga Nainar [Tue, 28 Jun 2016 17:51:10 +0000 (10:51 -0700)]
Disable warnings triggered in Clang r271374

http://b/28149048
http://b/29823425

Disable -Wconstant-conversion and -Wundefined-var-template.  The second
bug above tracks that these warnings get reenabled.

Test: Tested build, boot and common usage for Arm, Arm64, x86, x86_64,
Mips images in AOSP and internal branch.

Change-Id: Iea20cf6b5dbec3247b55cf8130f88202e786e367

7 years agoTune the GC ergnomics for the read barrier config.
Hiroshi Yamauchi [Tue, 19 Jul 2016 00:07:26 +0000 (17:07 -0700)]
Tune the GC ergnomics for the read barrier config.

Add 1.0 to the heap growth multiplier for the read barrier config, which
reduces the GC frequency down to roughly the same as CMS in one of the
jank tests.

Bug: 29517059
Bug: 12687968
Test: ART run-tests.
Change-Id: I1302a2f17e862f152d2f92bc06a65c9e6defcba0

7 years agoJIT: Don't update the dex cache of another class loader.
Nicolas Geoffray [Tue, 19 Jul 2016 16:06:23 +0000 (17:06 +0100)]
JIT: Don't update the dex cache of another class loader.

This only works for properly delegating class loaders. But Java allows
non-delegating ones :(

bug:29964720

Change-Id: I8b785e6cdfe9a2b77322521a02b8e59ec332ad83
test:612-jit-dex-cache

7 years agoCompute oat and odex filenames eagerly.
Richard Uhler [Tue, 19 Apr 2016 20:24:41 +0000 (13:24 -0700)]
Compute oat and odex filenames eagerly.

Because we almost always need both of them anyway, and they aren't
expensive to compute.

Test: oat file assistant tests.
Change-Id: I328ea51da6eb8700329f829a0458b02e12d1ee9e

7 years agoMake a static OatFileAssistant::DexLocationToOatFileName function.
Richard Uhler [Tue, 19 Apr 2016 20:08:04 +0000 (13:08 -0700)]
Make a static OatFileAssistant::DexLocationToOatFileName function.

So that you can figure out an oat file name without instantiating an
OatFileAssistant object.

Test: oat file assistant tests.
Change-Id: I577c5d02225f926086e9833d45b88d8a92db52fa

7 years agoMerge "ARM: Fix shifted register offset mem address mode for load signed."
Vladimir Marko [Tue, 19 Jul 2016 12:37:33 +0000 (12:37 +0000)]
Merge "ARM: Fix shifted register offset mem address mode for load signed."

7 years agoARM: Fix shifted register offset mem address mode for load signed.
Artem Serov [Mon, 11 Jul 2016 13:00:46 +0000 (14:00 +0100)]
ARM: Fix shifted register offset mem address mode for load signed.

For example 'ldrsh r0, [sp, r1, LSL #2]' previously
was assembled as 'ldrh'.

Test: New test in assembler_thumb2_test.cc .
Change-Id: I1d30724f0c2745b131876bffefdc0a780d76f6a1

7 years agoMerge "Do allocation fence before pushing on allocation stack"
Treehugger Robot [Mon, 18 Jul 2016 22:53:44 +0000 (22:53 +0000)]
Merge "Do allocation fence before pushing on allocation stack"

7 years agoART: Change run-*-tests to ANDROID_{HOST|PRODUCT}_OUT
Andreas Gampe [Mon, 18 Jul 2016 21:56:56 +0000 (14:56 -0700)]
ART: Change run-*-tests to ANDROID_{HOST|PRODUCT}_OUT

Use ANDROID_{HOST|PRODUCT}_OUT to allow OUT_DIR and OUT_DIR_COMMON_BASE
usage.

Change-Id: Ia091b9be5153518ef58fc7686cb77669e7c1ca54
Test: art/tools/run-jdwp-tests
Test: art/tools/run-libcore-tests --mode=host

7 years agoDo allocation fence before pushing on allocation stack
Mathieu Chartier [Mon, 18 Jul 2016 18:11:45 +0000 (11:11 -0700)]
Do allocation fence before pushing on allocation stack

Heap::VisitObjects relies on having valid classes for objects in
the allocation stack. If the writes reorder, the thread calling
VisitObjects could see the free list pointer instead of the class
of the object. I believe this is causing crashes in VisitObjects.

Bug: 28790624

Test: Volantis booted

Change-Id: I0f2d4097de1ef3f5caf670ecc977d4d6837872ca

7 years agoMerge "Rename current register allocator implementation"
Treehugger Robot [Mon, 18 Jul 2016 19:35:11 +0000 (19:35 +0000)]
Merge "Rename current register allocator implementation"

7 years agoMerge "ART: Fix run-test script"
Treehugger Robot [Mon, 18 Jul 2016 18:49:56 +0000 (18:49 +0000)]
Merge "ART: Fix run-test script"

7 years agoART: Fix run-test script
Andreas Gampe [Sat, 25 Jun 2016 05:30:29 +0000 (22:30 -0700)]
ART: Fix run-test script

Don't assume out is under BUILD_TOP.

Test: run-test with OUT_DIR_COMMON_BASE set
Change-Id: Iba408e807e6a15ff60de54d6d4d653814d4b11d0

7 years agoMerge "MIPS64: Highest/Lowest Bit Intrinsic Support"
Roland Levillain [Mon, 18 Jul 2016 15:40:34 +0000 (15:40 +0000)]
Merge "MIPS64: Highest/Lowest Bit Intrinsic Support"

7 years agoMerge "Fix the build with respect to new VIXL."
Roland Levillain [Mon, 18 Jul 2016 15:07:19 +0000 (15:07 +0000)]
Merge "Fix the build with respect to new VIXL."

7 years agoFix the build with respect to new VIXL.
Roland Levillain [Mon, 18 Jul 2016 15:03:05 +0000 (16:03 +0100)]
Fix the build with respect to new VIXL.

Test: Build ART for ARM64.
Change-Id: I2a9ebf145c61db9f8ceec6616963bac3ad5a7eb9

7 years agoMerge "Fixes to build against new VIXL interface."
Roland Levillain [Mon, 18 Jul 2016 14:28:05 +0000 (14:28 +0000)]
Merge "Fixes to build against new VIXL interface."

7 years agoMerge "Add a new control flow simplifier."
Nicolas Geoffray [Mon, 18 Jul 2016 12:16:46 +0000 (12:16 +0000)]
Merge "Add a new control flow simplifier."

7 years agoMerge "ARM64: Improve code generated to spill/restore for slow paths."
Treehugger Robot [Mon, 18 Jul 2016 11:19:14 +0000 (11:19 +0000)]
Merge "ARM64: Improve code generated to spill/restore for slow paths."

7 years agoARM64: Improve code generated to spill/restore for slow paths.
Alexandre Rames [Fri, 15 Jul 2016 16:41:13 +0000 (17:41 +0100)]
ARM64: Improve code generated to spill/restore for slow paths.

Aligning the accesses allows generating better code.

Before:

    add x16, sp, #0x44 (68)
    stp x0, x1, [x16, #-16]

After:

    stp x0, x1, [sp, #56]

Change-Id: I3e20ad3fa59d00aee4b4d14ea9d59c7cd546509e

7 years agoAdd a new control flow simplifier.
Nicolas Geoffray [Fri, 15 Jul 2016 09:46:17 +0000 (10:46 +0100)]
Add a new control flow simplifier.

Run it in the dead code elimination phase, as it relates to
creating dead branches.

From 0.04 to 0.07% less code size framework/gms/docs/fb (70K saved on fb)
3%-5% runtime performance improvements on Richards/DeltaBlue/Ritz.
Compile-time is mixed, so in the noise (from 2% slower to 1% faster).

test:611-checker-simplify-if

Change-Id: Ife8b7882d57b5481f5ca9dc163beba655d7e78bf