OSDN Git Service
Treehugger Robot [Mon, 28 Nov 2016 12:05:22 +0000 (12:05 +0000)]
Merge "Cache flush/invalidate needs RWX permission"
Artem Udovichenko [Thu, 17 Nov 2016 07:51:58 +0000 (10:51 +0300)]
Cache flush/invalidate needs RWX permission
When generating JIT code, perform cache maintenance operations before
removing page write permissions. Errata on some cores require data
flush operations to be followed by data invalidate operations
requiring write permission.
Test: ART_TEST_JIT=true test-art-target on an arm64 device.
bug:
27265969
Change-Id: I70cbb0dc2ea37e42f2011d23333e386ae3f639c7
Orion Hodson [Mon, 28 Nov 2016 09:39:44 +0000 (09:39 +0000)]
Merge "ART: Throw InternalError rather than VirtualMachineError"
Orion Hodson [Mon, 28 Nov 2016 09:10:32 +0000 (09:10 +0000)]
Merge "Revert "Revert "Method Handles: Add VarargsCollector."""
Orion Hodson [Fri, 25 Nov 2016 15:46:49 +0000 (15:46 +0000)]
Revert "Revert "Method Handles: Add VarargsCollector.""
This reverts commit
12af7a41c224923fb55699305008214636a1a3da.
Test: m test-art-host-run-test-956-methodhandles
Bug:
30550796
Change-Id: Ia820093bab919667729daba5b14330de51a63f79
Orion Hodson [Fri, 25 Nov 2016 16:07:11 +0000 (16:07 +0000)]
ART: Throw InternalError rather than VirtualMachineError
Instantiate InternalError instead of VirtualMachineError as the latter
is abstract and not instantiable.
Change-Id: I3e90b28a34bef02dd1c200c02cb832aa8dce10b3
Orion Hodson [Fri, 25 Nov 2016 14:31:43 +0000 (14:31 +0000)]
Merge "Revert "Method Handles: Add VarargsCollector.""
Orion Hodson [Fri, 25 Nov 2016 14:27:07 +0000 (14:27 +0000)]
Revert "Method Handles: Add VarargsCollector."
This reverts commit
ab52ce1480d9c4701bfe3ff3530596ec0c5760ee.
Change-Id: Idc6dd28b58a61cbd2c9ee72e8a0eeb07b2ca0edf
Nicolas Geoffray [Fri, 25 Nov 2016 11:25:43 +0000 (11:25 +0000)]
Merge "ARM: VIXL32: Improve codegen on CBZ/CBNZ"
Nicolas Geoffray [Fri, 25 Nov 2016 08:32:37 +0000 (08:32 +0000)]
Merge "Use input-vdex-fd, or input-vdex in dex2oat."
Richard Uhler [Fri, 25 Nov 2016 07:53:25 +0000 (07:53 +0000)]
Merge changes Ia3853e46,Ic32af2ec,I79c6cc12,Idfd59472,I68720661
* changes:
Refine OatStatus enum.
Factor out code for picking the best oat file.
Move dumping of oat file status inside of the OatFileAssistant.
Remove OatFileAssistant IsOutOfDate, IsUpToDate, and NeedsRelocation.
Simplify decision of which is the best oat file.
Richard Uhler [Tue, 22 Nov 2016 09:42:04 +0000 (09:42 +0000)]
Refine OatStatus enum.
Replace Exists() with a separate OatStatus code, and distinguish
between the oat file being out of date due to the dex being out of
date or the image being out of date.
Bug:
30937355
Test: oat_file_assistant_test
Change-Id: Ia3853e461beb07f0b417276b9277b9c562b20865
Nicolas Geoffray [Sat, 19 Nov 2016 10:42:37 +0000 (10:42 +0000)]
Use input-vdex-fd, or input-vdex in dex2oat.
input-vdex-fd is used by installd
input-vdex is used by run-tests, and (will be used by) go/lem
This change copies the contents of the passed vdex to the new one,
unquicken the new vdex, and run the fast verification on the new
vdex.
bug:
30937355
Test: device boots, apps get updated faster with vdex
Test: set TEST_VDEX to true in run-test-jar, run all tests
Test: 628-vdex
Change-Id: Idfbac4de411cebcf8ea7a6af7a417d7c7908dd72
Vladimir Marko [Thu, 24 Nov 2016 13:05:12 +0000 (13:05 +0000)]
Merge "Revert "Revert "Revert "Make sure that const-class linkage is preserved.""""
Vladimir Marko [Thu, 24 Nov 2016 13:03:57 +0000 (13:03 +0000)]
Revert "Revert "Revert "Make sure that const-class linkage is preserved."""
626-const-class-linking is failing (spurious wakeups?)
970-iface-super-resolution-gen is failing:
dex2oatd F 11-24 10:57:16 6410 6410 image_writer.cc:1144]
Check failed: !IsBootClassLoaderClass(as_klass)
java.lang.Class<java.lang.NoSuchMethodError>
Bug:
30627598
This reverts commit
25dcbad4462ea7279ee2bbe0884abc25bdfac77a.
Change-Id: Ie010169bdde45e6ccf2e04a521da4682bd817114
Orion Hodson [Thu, 24 Nov 2016 11:57:29 +0000 (11:57 +0000)]
Merge "Method Handles: Add VarargsCollector."
xueliang.zhong [Fri, 4 Nov 2016 09:23:32 +0000 (09:23 +0000)]
ARM: VIXL32: Improve codegen on CBZ/CBNZ
This patch improves codegen on CBZ/CBNZ.
CompareAndBranchIfZero/CompareAndBranchIfNonZero are introduced.
These two functions can generate CMP+Bcc or Cbz/Cbnz.
CMP+Bcc are generated by default.
If a hint is given (is_far_target = false) and rn and label
can all fit into Cbz/Cbnz, then Cbz/Cbnz is generated.
Prefer these two interfaces to using vixl32::MacroAssembler::Cbz/Cbnz.
In T32, Cbz/Cbnz instructions have following limitations:
- Far targets, which are over 126 bytes away, are not supported.
- Only low registers can be encoded.
- Backward branches are not supported.
Test: ART_USE_VIXL_ARM_BACKEND=true m test-art-host
Test: ART_USE_VIXL_ARM_BACKEND=true m test-art-target
Change-Id: I5d2ada19ea4f83dab78baf0cf78c72e99e58d946
Vladimir Marko [Thu, 24 Nov 2016 10:07:53 +0000 (10:07 +0000)]
Merge "Revert "Revert "Make sure that const-class linkage is preserved."""
Richard Uhler [Mon, 14 Nov 2016 14:38:03 +0000 (14:38 +0000)]
Factor out code for picking the best oat file.
Factor out common code for determining the best oat file.
Bug:
30937355
Test: oat_file_assistant_test
Change-Id: Ic32af2ec493675be1a22e81c2de46a848c398f6b
Richard Uhler [Mon, 14 Nov 2016 14:53:55 +0000 (14:53 +0000)]
Move dumping of oat file status inside of the OatFileAssistant.
Rather than exposing what could otherwise be internal to the
OatFileAssistant, move the logic for dumping a human readable
representation of the status of the oat files inside the
OatFileAssistant.
Bug:
30937355
Test: oat_file_assistant_test
Change-Id: I79c6cc1286a822f1dbe0035be934a2be4792563c
Richard Uhler [Tue, 8 Nov 2016 16:23:07 +0000 (16:23 +0000)]
Remove OatFileAssistant IsOutOfDate, IsUpToDate, and NeedsRelocation.
Relatively recent design changes made these functions equivalent to
the Status function. Prefer to use the Status function instead to
simplify the OatFileAssistant API, especially in preparation for
expanding the number of possible Status results.
Bug:
30937355
Test: oat_file_assistant_test
Change-Id: Idfd594729210257f693f03ca823a8d54e74034c5
Richard Uhler [Tue, 8 Nov 2016 16:51:51 +0000 (16:51 +0000)]
Simplify decision of which is the best oat file.
Previously the best oat file was determined by looking at all
combination of states of the oat file in the odex location and the oat
location. This change simplifies the decision by using the oat
location if that oat file is not out of date and using the odex
location otherwise. For practical cases, both decisions lead to the
same result.
Bug:
30937355
Test: oat_file_assistant_test
Change-Id: I687206612914cdbd43950fcdd7dd818a54255a6b
Mathieu Chartier [Wed, 23 Nov 2016 19:19:22 +0000 (19:19 +0000)]
Merge "Speed up verified methods"
Nicolas Geoffray [Wed, 23 Nov 2016 17:26:20 +0000 (17:26 +0000)]
Merge "libopenjdkjvmti is actually not a test library."
Vladimir Marko [Wed, 23 Nov 2016 13:35:26 +0000 (13:35 +0000)]
Revert "Revert "Make sure that const-class linkage is preserved.""
Fix jdwp getting multiple instances of the same class.
Fix counting "zygote"/"non-zygote" classes in class table.
Fix FindClass() to EnsureResolved() classes added by
a racing thread to the class table.
Test: m test-art-host
Test: art/tools/run-jdwp-tests.sh --mode=host --variant=X64 --debug
Bug:
30627598
This reverts commit
ecffc67068a28d55d131553bf915fdb9fafbbc03.
Change-Id: I15807949da4f163c2693fac2fbfc274f17685f8a
Orion Hodson [Fri, 11 Nov 2016 11:11:31 +0000 (11:11 +0000)]
Method Handles: Add VarargsCollector.
Enable support for methods with variable arity arguments.
Test: m test-art-host-run-test-956-methodhandles
Bug:
30550796
Change-Id: I205724ddacc6c9840e18093a3984c1ad6e54394a
Nicolas Geoffray [Wed, 23 Nov 2016 16:25:15 +0000 (16:25 +0000)]
libopenjdkjvmti is actually not a test library.
Use the right target for the libopenjdkjvmti library.
Test: test-art-target-sync
Change-Id: I8c23c87911bc890482e5e9be3d64ce0a362b6924
Nicolas Geoffray [Wed, 23 Nov 2016 14:23:05 +0000 (14:23 +0000)]
Merge "Update expectations in SetVerifierDeps/SetStackSample."
Calin Juravle [Wed, 23 Nov 2016 12:53:24 +0000 (12:53 +0000)]
Merge "Revert "Use interpret-only instead of verify-at-runtime when testing JIT""
Calin Juravle [Wed, 23 Nov 2016 12:52:53 +0000 (12:52 +0000)]
Revert "Use interpret-only instead of verify-at-runtime when testing JIT"
This reverts commit
8b2cf761df20e60a9e0e57564db3ab4eae5d64a7.
Change-Id: I3a15a20eb9d3c71902f42cd5e559d8ab4e15f07d
Nicolas Geoffray [Wed, 23 Nov 2016 12:52:01 +0000 (12:52 +0000)]
Update expectations in SetVerifierDeps/SetStackSample.
SetStackSample might update an existing sample. So instead
check that SetStackSample is only called in non-AOT mode, and
SetVerifierDeps is only called in AOT mode.
Test: 099-vmdebug
Change-Id: I80dc3d61da651fb311364418802b0bb7e1ce9351
Nicolas Geoffray [Wed, 23 Nov 2016 12:35:42 +0000 (12:35 +0000)]
Merge "Update run-test dependencies on device."
Nicolas Geoffray [Wed, 23 Nov 2016 10:47:02 +0000 (10:47 +0000)]
Update run-test dependencies on device.
Test: test-art-target
Change-Id: I5f58bcbf9eb41571d9511eba8b3f11765baf24a2
Roland Levillain [Wed, 23 Nov 2016 10:33:03 +0000 (10:33 +0000)]
Merge "Fix build rules for statically linked oatdump on host."
Mathieu Chartier [Mon, 21 Nov 2016 23:05:23 +0000 (15:05 -0800)]
Speed up verified methods
Switch to an array of atomic pointers instead of a map. Removes lock
and map lookup. Also address comments from previous CL.
GetVerifiedMethod: 1.59% -> 0.18% of compilation time.
Install time seems to goes down by around 1%.
Also has significant RAM savings (FB host compile):
dex2oat native alloc:
84695472B ->
71268736B
For the JIT case, the old method is used to prevent any increase in
RAM usage.
Bug:
32641252
Test: test-art-host
Change-Id: I47b4b8a4a3cb3f8ef23e36a888b8885e12168787
Jeff Hao [Tue, 22 Nov 2016 23:49:19 +0000 (23:49 +0000)]
Merge changes from topic 'dex2oat_dexlayout'
* changes:
Add layout scheme to dexlayout for dex file reordering.
Add new --layout-profile compiler-filter for dex2oat.
Dexlayout cleanup and refactoring.
Mathieu Chartier [Tue, 22 Nov 2016 23:09:48 +0000 (23:09 +0000)]
Merge "Always fill in root table length"
Jeff Hao [Wed, 19 Oct 2016 18:17:11 +0000 (11:17 -0700)]
Add layout scheme to dexlayout for dex file reordering.
Currently takes startup classes from profile and groups them together
with the lowest class indexes. Moves the ClassDef, ClassData, and
CodeItem sections.
Results gathered from 5 second systrace of app startup (no other input):
Maps
base.vdex - 24.70MB -> 19.35MB (30.50MB file size)
base.odex - 13.22MB -> 12.82MB (13.22MB file size)
Running time of dex2oat: 27.5s -> 37.1s
Photos
base.vdex - 12.98MB -> 9.77MB (15.10MB file size)
base.odex - 6.84MB -> 6.24MB ( 6.84MB file size)
Running time of dex2oat: 15.6s -> 19.9s
Plus
base.vdex - 14.09MB -> 13.29MB (16.72MB file size)
base.odex - 2.47MB -> 2.47MB ( 2.47MB file size)
Running time of dex2oat: 14.1s -> 18.5s
Performance and memory tracking bug: b/
33017139
Test: mm test-art-host-gtest-dexlayout_test
Bug:
29921113
Change-Id: Ib500ed353d71a606e2db0dd80378750c5f7b8e33
Mathieu Chartier [Tue, 22 Nov 2016 21:24:40 +0000 (13:24 -0800)]
Always fill in root table length
Otherwise, there may be DCHECK failure in JitCodeCache::ClearData.
Fixes gcstress JIT tests.
Test: test/run-test --host --relocate --prebuild --64 --jit --gcstress --dev 021-string2
Change-Id: I757ccd1e5a015c5e686254417a2a3f68b650da1c
Treehugger Robot [Tue, 22 Nov 2016 20:30:12 +0000 (20:30 +0000)]
Merge "Verify for type_id out of range errors"
Mathieu Chartier [Tue, 22 Nov 2016 19:26:09 +0000 (19:26 +0000)]
Merge "Correctly handle EINTR return from futex"
David Sehr [Mon, 21 Nov 2016 20:52:12 +0000 (12:52 -0800)]
Verify for type_id out of range errors
The dalvik documentation states that several fields used as type ids are
uint, which is 32 bits. Other places use uint16, which is obviously 16.
This adds assertions that there are fewer than 65536
class_def_items/class_data_items and that the fields in question are
either valid (0 <= id <= 0xffff) or id == kDexNoIndex (0xffffffff).
Test: m test-art-host
Change-Id: I584615192086e6d8bb30aa11217d569d6dbe1280
Vladimir Marko [Tue, 22 Nov 2016 17:49:05 +0000 (17:49 +0000)]
Merge "Revert "Make sure that const-class linkage is preserved.""
Vladimir Marko [Tue, 22 Nov 2016 17:37:21 +0000 (17:37 +0000)]
Revert "Make sure that const-class linkage is preserved."
Bug:
30627598
This reverts commit
19886dbeced9430579a27979beb92b09c95197b4.
Squashed reverts of subsequent partial fixes.
This reverts commit
bad475a63b7c938b7ff7903ea1dae27369c3f629.
This reverts commit
0ba0c7d9b0328af39a385014652eb9af4e42d8b5.
This reverts commit
ddb8afec79492750f16f98e5bd07d44f3dc485d4.
Change-Id: Id94af5ae56a0335f9641a79c885cc75bba971713
Roland Levillain [Tue, 22 Nov 2016 17:20:46 +0000 (17:20 +0000)]
Fix build rules for statically linked oatdump on host.
Test: m build-art-host
Test: m test-art-host-gtest-oatdump_test
Bug:
29530992
Change-Id: Iaa1ff13f0d42e5bc9670fc87da5b6e291482f626
Jeff Hao [Wed, 19 Oct 2016 18:17:11 +0000 (11:17 -0700)]
Add new --layout-profile compiler-filter for dex2oat.
New compiler filter makes dex2oat call into dexlayout.
Added basic test for --layout-profile filter to make sure dex2oat runs
to completion and file is valid. Contests of file are not checked.
Test: mm test-art-host-gtest-dexlayout_test
Bug:
29921113
Change-Id: I4bd0dea3d3f1284c155d1d9dea80a48062e67770
Jeff Hao [Tue, 15 Nov 2016 02:10:16 +0000 (18:10 -0800)]
Dexlayout cleanup and refactoring.
Created option to output to a mem map in preparation of hooking
dexlayout into dex2oat.
Test: mm test-art-host-gtest-dexlayout_test
Bug:
29921113
Change-Id: Id42ef15cb8f83cc8d05b025b7647a4338e9b96b0
Nicolas Geoffray [Tue, 22 Nov 2016 17:07:46 +0000 (17:07 +0000)]
Merge "Use a per-thread VerifierDeps."
Nicolas Geoffray [Tue, 22 Nov 2016 16:40:20 +0000 (16:40 +0000)]
Merge "Clear the right data pointer."
Vladimir Marko [Tue, 22 Nov 2016 14:58:42 +0000 (14:58 +0000)]
Merge changes If9d1da45,I6daecb52
* changes:
Do not test class unloading in 626-const-class-linking
Fix 626-const-class-linking for no-dex2oat configuration.
Vladimir Marko [Tue, 22 Nov 2016 14:38:31 +0000 (14:38 +0000)]
Do not test class unloading in 626-const-class-linking
Keep tests that classes were not unloaded but do not test
that they are actually unloaded. Both JIT and tracing can
hold onto those classes.
Test: m ART_USE_JIT=true test-art-host-run-test-626-const-class-linking
Test: m ART_TEST_TRACE=true test-art-host-run-test-626-const-class-linking
Bug:
30627598
Change-Id: If9d1da45d8691c5ee05b6cf500d7d132bcab611f
Nicolas Geoffray [Tue, 22 Nov 2016 13:45:36 +0000 (13:45 +0000)]
Clear the right data pointer.
Internally, stack_map_data is an offset of the roots_data allocation.
Pass both to the ClearData method, which will use the right pointer
to deallocate.
Test: test-art-host in debug mode
Change-Id: Ibfe38e7f1853870076048427f2efe7121c11c136
Vladimir Marko [Tue, 22 Nov 2016 13:18:00 +0000 (13:18 +0000)]
Fix 626-const-class-linking for no-dex2oat configuration.
For this configuration, the MisbehavingLoader.loadClass()
is called twice, so we cannot rely on printing the message
exactly once.
Test: m ART_TEST_RUN_TEST_NO_DEX2OAT=true \
test-art-host-run-test-626-const-class-linking
Bug:
30627598
Change-Id: I6daecb52efc1d25fc5cf796f8aef726666e97c8b
Nicolas Geoffray [Tue, 22 Nov 2016 13:18:55 +0000 (13:18 +0000)]
Merge "ARM: VIXL32: Fix the assembler usage in the intrinsics."
Nicolas Geoffray [Fri, 18 Nov 2016 16:03:10 +0000 (16:03 +0000)]
Use a per-thread VerifierDeps.
Avoid lock contention on a singleton VerifierDeps by allocating
temporary per-thread VerifierDeps that get merged after verification.
This saves around ~35% compile-times on interpret-only.
Only the creation of extra strings is guarded by a lock, for simplicity.
Test: test-art-host, test-art-target
bug:
32641252
bug:
30937355
Change-Id: I11a2367da882b58e39afa7b42cba2e74a209b75d
Vladimir Marko [Tue, 22 Nov 2016 12:12:12 +0000 (12:12 +0000)]
Merge "Fix build: update test to use dex::TypeIndex."
Vladimir Marko [Tue, 22 Nov 2016 11:56:48 +0000 (11:56 +0000)]
Fix build: update test to use dex::TypeIndex.
Test: m test-art-host
Bug:
30627598
Change-Id: I1aad84c6367b7b60b60391488aeb9777d3a627e4
Vladimir Marko [Tue, 22 Nov 2016 11:23:16 +0000 (11:23 +0000)]
Merge "Make sure that const-class linkage is preserved."
Daniel Colascione [Mon, 21 Nov 2016 22:35:42 +0000 (14:35 -0800)]
Correctly handle EINTR return from futex
We should handle EINTR the same way we do EAGAIN: they're both
transient failures. Either way, higher-level code (see
ReaderWriterMutex::SharedLock) will do the right thing, loop around,
and try again.
Test: code inspection
Change-Id: I1eb7fb6851cc1a65acc9a0cf7cbeef787b6b28b6
Mathieu Chartier [Mon, 21 Nov 2016 23:32:45 +0000 (23:32 +0000)]
Merge "Check same dex file for finding class descriptor string id"
Treehugger Robot [Mon, 21 Nov 2016 22:01:14 +0000 (22:01 +0000)]
Merge changes I8e51415a,I79a255c1
* changes:
ART: Add JNI local details for FollowReferences
ART: Fix generic_ptr handling in GetMethodName
Mathieu Chartier [Thu, 17 Nov 2016 21:08:35 +0000 (13:08 -0800)]
Check same dex file for finding class descriptor string id
Use ClassDef for finding same dex file class descriptor string id
for vdex. Removes some calls to FindStringId, the remaining calls to
FindStringId are due to multidex or referring to types in the boot
class path.
Also check method index slot to see if we get lucky.
Host perf:
DexFile::FindStringId: 2.57% -> 0.94%
Not done for fields since these do not seem to reach the FindStringId
case accordingto perf. The remaining calls to FindStringId come from:
100.00% verifier::VerifierDeps::GetClassDescriptorStringId
-> 66.62% verifier::VerifierDeps::AddAssignability
-> 33.38% verifier::VerifierDeps::AddMethodResolution
FB install time: 35.54s -> 34.41s (avg of 40 samples at 960MHZ N6P).
Bug:
32641252
Test: test-art-host
Change-Id: If50408cf1be5c2d2127140780b8d5af66306939c
Treehugger Robot [Mon, 21 Nov 2016 21:17:32 +0000 (21:17 +0000)]
Merge "ART: Add dex::TypeIndex"
Treehugger Robot [Mon, 21 Nov 2016 20:43:12 +0000 (20:43 +0000)]
Merge "ART: Add boot classpath check for initialization"
Mathieu Chartier [Mon, 21 Nov 2016 20:19:58 +0000 (20:19 +0000)]
Merge "Use type lookup tables in compiler"
Anton Kirilov [Mon, 14 Nov 2016 14:37:00 +0000 (14:37 +0000)]
ARM: VIXL32: Fix the assembler usage in the intrinsics.
The assembler must not be used outside of any scopes;
the macroassembler methods should be called instead.
Test: ART_USE_VIXL_ARM_BACKEND=true m test-art-host
Test: ART_USE_VIXL_ARM_BACKEND=true m test-art-target
Change-Id: I1358bea5330547db9b7a8446932d85ba946e5683
Vladimir Marko [Fri, 11 Nov 2016 11:40:40 +0000 (11:40 +0000)]
Make sure that const-class linkage is preserved.
After calling loadClass() on the initiating loader, insert
the Class to the ClassTable. Check that the returned class
has the correct name and that the ClassTable does not yet
contain a different Class for the same name. If either check
fails, follow the RI behavior for parallel capable class
loaders, i.e. throw NoClassDefFoundError for name mismatch
(with more descriptive message than RI) or return the Class
already present in the ClassTable. This ensures that the
const-class instruction for a given name always gives the
same result as the first time it succeeded for the same
initiating loader, even for non-well-behaved loader.
Inserting the Class to the ClassTable is also a necessary
step for changing to hash-based dex cache type array when
the DexCache will not necessarily keep the Class and
therefore the defining loader alive while the initiating
loader stays alive, so the ClassTable entry is necessary
to keep the dependecy.
Add tests that a const-class instruction always returns the
same Class<?> as it returned the first time, i.e. when it
was "linked", even for class loaders that are not
well-behaved, even if the dex cache type array is cleared.
Also test the edge cases, i.e. NoClassDefFoundError and
returning old entry from ClassTable when already inserted
by another thread.
Bug:
30627598
Test: 626-const-class-linking
Test: m test-art-host
Change-Id: Ie9b721464b4e9a5dcce8df8095548e983bba1fe8
Andreas Gampe [Thu, 17 Nov 2016 23:21:22 +0000 (15:21 -0800)]
ART: Add dex::TypeIndex
Add abstraction for uint16_t type index.
Test: m test-art-host
Change-Id: I47708741c7c579cbbe59ab723c1e31c5fe71f83a
Andreas Gampe [Mon, 21 Nov 2016 18:10:08 +0000 (10:10 -0800)]
ART: Add boot classpath check for initialization
Boot classpath classes should not fail initialization.
Test: m test-art-host
Change-Id: I48af3ab6ab3d5c2589b804df8f4585f5a81ea169
Mathieu Chartier [Thu, 17 Nov 2016 00:22:37 +0000 (16:22 -0800)]
Use type lookup tables in compiler
This recently regressed and stopped happening, the fix creates fake
OatDexFiles so that the compiler uses the type lookup tables instead
of slow FindClassDef and FindTypeId.
Perf on host compile Facebook:
Before:
2.49%: art::DexFile::FindClassDef(unsigned short) const
1.59%: art::DexFile::FindTypeId(char const*) const
After:
0.42%: art::OatDexFile::FindClassDef(art::DexFile const&, char const*, unsigned long)
0%: art::DexFile::FindTypeId(char const*) const
0%: art::DexFile::FindClassDef(unsigned short) const
Average install (N6P 960 mhz average of 40 samples): 38.2s -> 35.64s
Bug:
32641252
Test: test-art-host, adb install.
Change-Id: I34df21dc2c155bc2579c5cafdd91f9cb0fead1a9
Treehugger Robot [Mon, 21 Nov 2016 17:38:14 +0000 (17:38 +0000)]
Merge "ART: Fix line break in Throwable::Dump"
Nicolas Geoffray [Mon, 21 Nov 2016 10:14:06 +0000 (10:14 +0000)]
Merge "Revert "Revert "Revert "Revert "JIT root tables."""""
Treehugger Robot [Sun, 20 Nov 2016 00:37:56 +0000 (00:37 +0000)]
Merge changes I240f8bb7,I27defaa7,Ib82d64cc
* changes:
ART: Do not DCHECK in verifier when already aborting
ART: Add debug exception check to JIT
ART: Extend DumpRecursiveAbort
Andreas Gampe [Sat, 19 Nov 2016 01:43:05 +0000 (17:43 -0800)]
ART: Fix line break in Throwable::Dump
Add line breaks for missing stack traces.
Test: m test-art-host
Change-Id: I0624ed19d6d025b7c00d6e6fb55827577bffc7b6
Andreas Gampe [Sat, 19 Nov 2016 22:51:52 +0000 (14:51 -0800)]
ART: Do not DCHECK in verifier when already aborting
Only log an error in FailOrAbort in the debug build when the
runtime is already aborting.
Test: m test-art-host
Change-Id: I240f8bb7695e42d60b5805ebae2ab9dc72e5fdb1
Andreas Gampe [Sat, 19 Nov 2016 01:39:45 +0000 (17:39 -0800)]
ART: Add debug exception check to JIT
Check for a pending exception after JIT compilation finishes.
Test: m ART_TEST_JIT=true test-art-host
Change-Id: I27defaa733f360cb1dbabede62d4908777bc5d2b
Andreas Gampe [Sat, 19 Nov 2016 01:21:12 +0000 (17:21 -0800)]
ART: Extend DumpRecursiveAbort
Accept recursive dump request for deeper recursions. Otherwise,
two or more threads aborting concurrently would ensure that there
are no dumps.
Test: m test-art-host
Change-Id: Ib82d64cceba0de89e352d9b15bcd5708db82498c
Treehugger Robot [Sat, 19 Nov 2016 01:07:35 +0000 (01:07 +0000)]
Merge "Fix valgrind issue with thread_pool_test."
Andreas Gampe [Fri, 18 Nov 2016 23:06:21 +0000 (15:06 -0800)]
Fix valgrind issue with thread_pool_test.
Make sure the task list is empty at the end of the test. Follow-up
to commit
e2fc98e88106b9cfa44eeacc6765c291feda04a8.
Test: m test-art-host-gtest-thread_pool_test
Test: m valgrind-test-art-host-gtest-thread_pool_test
Change-Id: Ib41a1f7dde9d45f1195e18ff6db4241fb7dc7428
Andreas Gampe [Fri, 18 Nov 2016 21:30:30 +0000 (13:30 -0800)]
ART: Add JNI local details for FollowReferences
Give details for the thread.
Stack depth and method are currently approximated (as
bottom of the stack), as the RootInfo does not have that
information.
Bug:
31385354
Test: m test-art-host-run-test-913-heaps
Change-Id: I8e51415a02678c55b687144b4cb749c4105bd3c1
Andreas Gampe [Fri, 18 Nov 2016 21:31:13 +0000 (13:31 -0800)]
ART: Fix generic_ptr handling in GetMethodName
Do not write null into a null pointer.
Bug:
31684812
Test: m test-art-host-run-test-910-methods
Change-Id: I79a255c1403ca9621390c9cdc09d2381c26fffb6
Treehugger Robot [Fri, 18 Nov 2016 21:04:59 +0000 (21:04 +0000)]
Merge "Revert "Revert "Use interpret-only instead of verify-at-runtime when testing JIT"""
Andreas Gampe [Fri, 18 Nov 2016 21:03:13 +0000 (21:03 +0000)]
Merge "ART: Dump native stack on recursive abort"
Hiroshi Yamauchi [Fri, 18 Nov 2016 20:16:30 +0000 (20:16 +0000)]
Merge "Run 103-string-append with libartd."
Treehugger Robot [Fri, 18 Nov 2016 19:33:25 +0000 (19:33 +0000)]
Merge "ART: Use cerr in Runtime::Abort on the host"
Calin Juravle [Fri, 18 Nov 2016 17:39:26 +0000 (17:39 +0000)]
Revert "Revert "Use interpret-only instead of verify-at-runtime when testing JIT""
This reverts commit
0b4d05e096d31e9c2d5ae522dbcc7b190b60a6f6.
Test: m test-art-host-run-test
Change-Id: Ie946ebd899002bdc7502276b3d53000c551be768
Andreas Gampe [Fri, 18 Nov 2016 17:51:06 +0000 (09:51 -0800)]
ART: Dump native stack on recursive abort
Dump the native stack of the current thread on recursive abort.
That is safe to do, as it requires no runtime interaction.
Test: m test-art-host
Change-Id: I4e052916f1036c74dc9fa82b049b4574626a70bb
Andreas Gampe [Fri, 18 Nov 2016 17:30:08 +0000 (09:30 -0800)]
ART: Use cerr in Runtime::Abort on the host
Use cerr to immediately write out abort info on the host, in an
effort to make recursive aborts diagnosable. This is similar
(but slightly inferior) to the previous INTERNAL_FATAL fastpathing.
Test: m test-art-host
Change-Id: I7c49874951acd47e58faceeb84a9ba93159ca3b0
Calin Juravle [Fri, 18 Nov 2016 16:09:11 +0000 (16:09 +0000)]
Merge "Revert "Revert Revert "Use interpret-only instead of verify-at-runtime when testing JIT"""
Calin Juravle [Fri, 18 Nov 2016 16:07:58 +0000 (16:07 +0000)]
Revert "Revert Revert "Use interpret-only instead of verify-at-runtime when testing JIT""
test-art-host-run-test-debug-prebuild-jit-relocate-ntrace-gcstress-checkjni-picimage-npictest-ndebuggable-974-verify-interface-super64 fails on the build servers.
It's unclear if it's because of this CL but revert again to check if it fixes the problem.
Locally everything is green...
This reverts commit
0b64a5195b38773084ff94618ac22284bc855b92.
Change-Id: I4668b8d8f8c9672fd942a693d98a20113fbb6b9e
Nicolas Geoffray [Wed, 16 Nov 2016 09:19:42 +0000 (09:19 +0000)]
Revert "Revert "Revert "Revert "JIT root tables.""""
Test: 626-set-resolved-string, test-art-host, test-art-target
Test: run-libcore-tests.sh
Test: phone boots and runs
This reverts commit
3395fbc20bcd20948bec8958db91b304c17cacd8.
Change-Id: I104b73d093e3eb6a271d564cfdb9ab09c1c8cf24
Nicolas Geoffray [Fri, 18 Nov 2016 13:14:10 +0000 (13:14 +0000)]
Merge "Add dequickening support."
Nicolas Geoffray [Fri, 18 Nov 2016 12:54:00 +0000 (12:54 +0000)]
Merge "ARM: VIXL32: Fix PackedSwitch."
Nicolas Geoffray [Fri, 18 Nov 2016 12:02:15 +0000 (12:02 +0000)]
Merge "Fix valgrind issue with thread_pool_test."
Calin Juravle [Fri, 18 Nov 2016 12:01:36 +0000 (12:01 +0000)]
Merge "Revert Revert "Use interpret-only instead of verify-at-runtime when testing JIT""
Treehugger Robot [Fri, 18 Nov 2016 11:24:20 +0000 (11:24 +0000)]
Merge "Make ImageWriter and PatchOat helper classes nested."
Nicolas Geoffray [Thu, 17 Nov 2016 10:58:36 +0000 (10:58 +0000)]
Add dequickening support.
Both vdex and BCI require full dequickening support. This
change implements dequickening by using the existing encoded
quickening, and adds the checkcast quickening to that encoding.
bug:
30937355
bug:
32369913
Test: dex_to_dex_decompiler_test.cc test-art-host
Change-Id: Ie95f46946d59b28157d6e47dcf4a859be032d1c3
Nicolas Geoffray [Fri, 18 Nov 2016 10:12:51 +0000 (10:12 +0000)]
Fix valgrind issue with thread_pool_test.
Make sure the task list is empty at the end of the test.
Test: thread_pool_test
Change-Id: I155a8eb90dc83a0dd9128f0c0bc64deeba7a92e9
Treehugger Robot [Fri, 18 Nov 2016 08:29:12 +0000 (08:29 +0000)]
Merge "ART: Make libopenjdkjvmti a run-test dependency"
Andreas Gampe [Fri, 18 Nov 2016 06:50:33 +0000 (22:50 -0800)]
ART: Make libopenjdkjvmti a run-test dependency
Add libopenjdkjvmti as a host dependency for running tests, as
required by the tests 900-913.
Bug:
31455788
Test: m test-art-host
Change-Id: I4910f587e35615a10168a39ef2c60d3d80981fc4