OSDN Git Service
Andreas Gampe [Sun, 27 Jul 2014 21:56:42 +0000 (21:56 +0000)]
Merge "ART: Fix target test paths"
Andreas Gampe [Sat, 12 Jul 2014 04:14:35 +0000 (21:14 -0700)]
ART: Fix target test paths
Nativetest is now nativetest64 on 64b targets.
Change-Id: I4c4c6bba1a56525df2993708caaae3e6f7f5f2f6
Ian Rogers [Sun, 27 Jul 2014 18:32:40 +0000 (18:32 +0000)]
Merge "Remove the per-project FDO support. Use the common FDO support from build/core. LOCAL_FDO_SUPPORT indicates if the module should enable FDO builds."
Dehao Chen [Thu, 24 Jul 2014 23:12:59 +0000 (16:12 -0700)]
Remove the per-project FDO support. Use the common FDO support from build/core.
LOCAL_FDO_SUPPORT indicates if the module should enable FDO builds.
Change-Id: I1eab8bf418b7ceb060eea18bdd304a06597af3d3
Bill Buzbee [Sun, 27 Jul 2014 17:49:57 +0000 (17:49 +0000)]
Merge "Revert "Enable Load Store Elimination for ARM and ARM64""
Bill Buzbee [Sun, 27 Jul 2014 17:49:42 +0000 (17:49 +0000)]
Revert "Enable Load Store Elimination for ARM and ARM64"
On extended testing, I'm seeing a CHECK failure at utility_arm.cc:1201.
This reverts commit
fcc36ba2a2b8fd10e6eebd21ecb6329606443ded.
Change-Id: Icae3d49cd7c8fcab09f2f989cbcb1d7e5c6d137a
Serban Constantinescu [Tue, 15 Jul 2014 16:44:21 +0000 (17:44 +0100)]
Enable Load Store Elimination for ARM and ARM64
This patch refactors the implementation of the LoadStoreElimination
optimisation pass. Please note that this pass was disabled and not
functional for any of the backends.
The current implementation tracks aliases and handles DalvikRegs as well
as Heap memory regions. It has been tested and it is known to optimise
out the following:
* Load - Load
* Store - Load
* Store - Store
* Load Literals
Change-Id: Iefae9b696f87f833ef35c451ed4d49c5a1b6fde0
Bill Buzbee [Sun, 27 Jul 2014 02:46:58 +0000 (02:46 +0000)]
Merge "SetConstantWide should mark both SSA regs as constant"
Bill Buzbee [Sun, 27 Jul 2014 17:30:43 +0000 (17:30 +0000)]
Merge "Enable Load Store Elimination for ARM and ARM64"
Bill Buzbee [Sun, 27 Jul 2014 02:46:40 +0000 (02:46 +0000)]
Merge "ART: ReEnable check for a gap between method's arg "in" vregs and rest of its vregs"
Andreas Gampe [Sun, 27 Jul 2014 01:36:55 +0000 (01:36 +0000)]
Merge "ART: Native bridge command-line parameter"
Andreas Gampe [Fri, 25 Jul 2014 09:32:19 +0000 (02:32 -0700)]
ART: Native bridge command-line parameter
Add a command-line parameter for the native bridge library, slight
refactor/cleanup.
Add run-test 115 to test the native bridge interface. Currently the
tests are black-listed for the target, as the setup for the test
is too complicated in the current infrastructure.
Change-Id: I6ccf19485e8c30b96e9f2fd5425278cb1ebd403f
Andreas Gampe [Sun, 27 Jul 2014 01:31:16 +0000 (01:31 +0000)]
Merge "ART: Allow arrays with erroneous component type"
Andreas Gampe [Thu, 24 Jul 2014 03:18:36 +0000 (20:18 -0700)]
ART: Allow arrays with erroneous component type
Array classes must tolerate having component type classes that are
erroneous. Change CreateArrayClass to use LookupClass when FindClass
failed.
Bug:
16019155
Change-Id: Id4868c5498431c85c199aa3cbecd23566dce3601
Andreas Gampe [Sat, 26 Jul 2014 19:57:14 +0000 (19:57 +0000)]
Merge "ART: Special-case cb(n)z even for in-reg constant"
Andreas Gampe [Sat, 26 Jul 2014 08:40:39 +0000 (01:40 -0700)]
ART: Special-case cb(n)z even for in-reg constant
Call out to OpCmpImmBranch in GenCompareAndBranch if the constant
is zero and we are testing == or !=, even when zero has been loaded
to a register already. This avoids a register size mismatch on 64b
architectures when basically doing a null check, and generally
squashes a cmp + branch to a cbz or cbnz on Arm and Mips. X86 is
not degraded.
Bug:
16562601
Change-Id: I1997760f43dc186a84247ad30ae91053f71d102d
Alexei Zavjalov [Tue, 15 Jul 2014 19:17:58 +0000 (02:17 +0700)]
ART: ReEnable check for a gap between method's arg "in" vregs and rest of its vregs
The Quick backend's runtime model includes a gap between a method's
argument ("in") vregs and the rest of its vregs. Handling a range instruction
which spans the gap is somewhat complicated, and should not happen
in normal usage of dx. Punt to the interpreter if it happened.
This issue relates to https://code.google.com/p/android/issues/detail?id=66371
Which was fixed, but the problem returned after some changes in the verify
flags set for the range bytecodes.
Change-Id: I214f32e7b4227a450257f30607135930d0adff53
Signed-off-by: Katkov, Serguei I <serguei.i.katkov@intel.com>
Signed-off-by: Alexei Zavjalov <alexei.zavjalov@intel.com>
Sebastien Hertz [Thu, 24 Jul 2014 07:24:12 +0000 (07:24 +0000)]
Merge "Fix access to long/double stack values from debugger"
Stephen Hines [Thu, 24 Jul 2014 02:49:36 +0000 (02:49 +0000)]
Merge "Increase limit to 2600 on host clang builds for frame size."
Stephen Hines [Thu, 24 Jul 2014 02:47:35 +0000 (19:47 -0700)]
Increase limit to 2600 on host clang builds for frame size.
Change-Id: I573e1bcdc7a935145e88077def9c90a5573d2e90
Stephen Hines [Thu, 24 Jul 2014 02:14:06 +0000 (02:14 +0000)]
Merge "Update frame-larger-than limit to 2440 for new clang."
Stephen Hines [Thu, 17 Jul 2014 08:49:18 +0000 (01:49 -0700)]
Update frame-larger-than limit to 2440 for new clang.
Bug:
16362912
The latest version of clang is generating larger frames on elf_writer_quick.cc
and dex2oat.cc than we had previously encountered. This change unblocks the
LLVM rebase temporarily.
Change-Id: I20d9f88959b91e0509e58f01c22532720225f44d
Mathieu Chartier [Thu, 24 Jul 2014 01:30:03 +0000 (01:30 +0000)]
Merge "Reduce allocations in ParallelGC test."
Mathieu Chartier [Thu, 24 Jul 2014 00:25:41 +0000 (17:25 -0700)]
Reduce allocations in ParallelGC test.
Fixes GC stress test mode.
Bug:
16406852
Change-Id: I86c8fd7516c3f41075bca00c7d5b4dab2bb7b1d4
Nicolas Geoffray [Wed, 23 Jul 2014 23:49:26 +0000 (23:49 +0000)]
Merge "Use a (somewhat) more documented environment variable."
Nicolas Geoffray [Wed, 23 Jul 2014 23:36:48 +0000 (00:36 +0100)]
Use a (somewhat) more documented environment variable.
I believe OUT is not mandatory to define, whereas
ANDROID_PRODUCT_OUT is. Not sure our continuous tests define it.
Change-Id: I7cd23971384df265e59587b49760aba29133a59c
Andreas Gampe [Wed, 23 Jul 2014 22:07:30 +0000 (22:07 +0000)]
Merge "x86: GenSelect utility update"
Jeff Hao [Wed, 23 Jul 2014 20:41:11 +0000 (20:41 +0000)]
Merge "Fix run-test --trace option if --host is specified after."
Jeff Hao [Wed, 23 Jul 2014 18:52:52 +0000 (11:52 -0700)]
Fix run-test --trace option if --host is specified after.
Change-Id: I3911428ba762657c4433ab23e60f1771c9ddb2fe
Andreas Gampe [Fri, 25 Jul 2014 23:43:42 +0000 (23:43 +0000)]
Merge "ART: Fix run-test 114 ParallelGC to account for OOM"
Andreas Gampe [Fri, 25 Jul 2014 23:37:09 +0000 (16:37 -0700)]
ART: Fix run-test 114 ParallelGC to account for OOM
This catches any OOMs and doesn't pollute the log.
Bug:
16406852
Change-Id: I1bc95091ccae35a8cb5f2ef0a789f8c8ce5209d0
Mathieu Chartier [Fri, 25 Jul 2014 19:06:45 +0000 (19:06 +0000)]
Merge "Fix main space memory leak and add checks."
Mathieu Chartier [Fri, 25 Jul 2014 18:50:47 +0000 (11:50 -0700)]
Fix main space memory leak and add checks.
The hypothesis is that we were leaking the main space and its
bitmaps, then eventually we would run out of virtual address space,
which would cause a null bitmap (DCHECK). Finally when we tried
adding the space with a null bitmap to the heap bitmap it segfaulted.
Changed some non performance critical DCHECK -> CHECK.
Bug:
16563323
Change-Id: I08a1f873752e28ebcf63ebbd90f92d994d7ca96b
Mathieu Chartier [Fri, 25 Jul 2014 18:42:53 +0000 (18:42 +0000)]
Merge "Fix dangling pointer bug when transitioning to background."
Mathieu Chartier [Fri, 25 Jul 2014 17:13:51 +0000 (10:13 -0700)]
Fix dangling pointer bug when transitioning to background.
Dangling pointer left behind from the old rosalloc / dlmalloc
spaces. We now avoid using this pointer by using main_space_
and non_moving_space_ as well as clear the pointer when we remove
the space.
Bug:
16567203
Change-Id: Ida9ff30783e89cd4a4d86a4d0e912701692101f1
Andreas Gampe [Fri, 25 Jul 2014 22:37:09 +0000 (22:37 +0000)]
Merge "ART: Clean up API after change 102631"
Andreas Gampe [Fri, 25 Jul 2014 20:36:56 +0000 (13:36 -0700)]
ART: Clean up API after change 102631
This adds an abort_on_error parameter to GetCurrentMethod, which is
by default true. This restores all previous behavior, except for
monitor installation, where it follows 101639.
Bug:
16556938
Change-Id: I9a12b9a21ccb9f558c86236bb58d15ff2fafaec0
Vladimir Marko [Fri, 25 Jul 2014 14:55:49 +0000 (14:55 +0000)]
Merge "Fix JNI compiler frame size adjustments."
Vladimir Marko [Thu, 24 Jul 2014 16:01:58 +0000 (17:01 +0100)]
Fix JNI compiler frame size adjustments.
Bug:
16321952
Change-Id: I6f84a59c679e335e02a8e70944a5d0bc8d73f90b
Narayan Kamath [Fri, 25 Jul 2014 14:26:26 +0000 (14:26 +0000)]
Merge "Rename openDexFileNative to openDexFile."
Calin Juravle [Wed, 16 Jul 2014 16:55:55 +0000 (17:55 +0100)]
Rename openDexFileNative to openDexFile.
We no longer need two distinct methods.
Bug:
15563230
(cherry picked from commit
8edcb9c8bc62fcbd181e136c32086f2b970306a4)
Change-Id: Ib95098af0dd26733b5946d8e24ae4dfbcdbec3ba
Nicolas Geoffray [Fri, 25 Jul 2014 13:34:39 +0000 (13:34 +0000)]
Merge "Add a new stack map scheme that encodes compilation info per pc."
Andreas Gampe [Fri, 25 Jul 2014 10:37:19 +0000 (10:37 +0000)]
Merge "ART: Fix wrong CHECK in GetCurrentLocationForThrow"
Andreas Gampe [Fri, 25 Jul 2014 10:07:06 +0000 (03:07 -0700)]
ART: Fix wrong CHECK in GetCurrentLocationForThrow
Under certain circumstances kDexNoIndex is actually an encoded
dex pc. Change the setup of CurrentMethodVisitor to account for that.
Bug:
16556938
Change-Id: I4671f8c71c76f3e17e02db10f6da8ecc8331c87d
Andreas Gampe [Fri, 25 Jul 2014 09:38:04 +0000 (09:38 +0000)]
Merge "ART: Turn on ART_TEST_KEEP_GOING by default, clean up unused vars"
Andreas Gampe [Fri, 25 Jul 2014 04:41:06 +0000 (21:41 -0700)]
ART: Turn on ART_TEST_KEEP_GOING by default, clean up unused vars
Change-Id: I8db60eade0338e53e76a87ef38ad661550475e3c
Andreas Gampe [Fri, 25 Jul 2014 04:01:32 +0000 (04:01 +0000)]
Merge "Integrate ART with NativeBridge interfaces"
Yong WU [Thu, 24 Jul 2014 13:32:15 +0000 (21:32 +0800)]
Integrate ART with NativeBridge interfaces
Native-bridge will provide the following interfaces to ART:
struct NativeBridgeCallbacks {
bool (*initialize )(NativeBridgeArtCallbacks* vm_itf);
void* (*loadLibrary )(const char* libpath, int flag);
void* (*getTrampoline)(void* handle, const char* name, const char* shorty,
uint32_t len);
bool (*isSupported )(const char* libpath);
};
Native-bridge will expose a symbol NativeBridgeItf with the
type of NativeBridgeCallbacks to ART.
And ART will provide the interfaces below to native-bridge:
struct NativeBridgeArtCallbacks {
int (*logger )(int prio, const char* tag, const char* fmt, ...);
const char* (*getMethodShorty)(JNIEnv* env, jmethodID mid);
int (*getNativeMethodCount )(JNIEnv* env, jclass clazz);
int (*getNativeMethods )(JNIEnv* env, jclass clazz, JNINativeMethod* methods,
uint32_t method_count);
};
Based on the interfaces, if an ART call to dlopen fails to open a native library,
it queries the native bridge by using NativeBridgeCallbacks::isSupported(). If the
native library is supported by native-bridge, ART can load the native library
using NativeBridgeCallbacks::loadLibrary() and get a trampoline for a specific
native method using NativeBridgeCallbacks::getTrampoline(). ART can then call
the native method using the normal signature and the address of the trampoline.
On the other side, in the case of a native method calling JNI native function
CallXXXXMethodY(), native-bridge calls back to Art for the shorty of the method
using NativeBridgeArtCallbacks::getMethodShorty() so that it can prepare based
on host calling convention.
In case of JNI function RegisterNatives()/UnregisterNatives(), native bridge can
call back to NativeBridgeArtCallbacks::getNativeMethodCount() and NativeBridgeArtCallbacks
::getNativeMethods() to get all native methods of specified class so that all
corresponding trampolines can be prepared/destroyed.
Class NativeBridge is created to encapsulate the function pointers of
NativeBridgeCallbacks and provides better abstraction to ART.
Note: functionality is turned off in native_bridge.cc at the moment.
Change-Id: I652755044957a7960254648652b538cce70dd011
Mathieu Chartier [Thu, 24 Jul 2014 20:47:09 +0000 (20:47 +0000)]
Merge "Fix memory leak in RemoveRememberedSet."
Mathieu Chartier [Thu, 24 Jul 2014 18:11:05 +0000 (11:11 -0700)]
Fix memory leak in RemoveRememberedSet.
RemoveRememberedSet now deletes the remembered set.
Bug:
16532086
Change-Id: I01092931cc20cd0688dd42eed3dde9ad140889b2
Mathieu Chartier [Thu, 24 Jul 2014 17:53:11 +0000 (17:53 +0000)]
Merge "Fix stale remembered sets error."
Mathieu Chartier [Thu, 24 Jul 2014 01:45:17 +0000 (18:45 -0700)]
Fix stale remembered sets error.
We were forgetting to remove the remembered set in transition to
background. This resulted in remembered sets being added for
spaces which no longer existed. This finally caused an error when
a new space happened to have the same address as the old space,
resulting in a CHECK failure.
Also tuned the number of ParallelGC to prevent spurrious failures
and removed the ParallelGC from broken tests in the make file.
Bug:
16532086
Bug:
16406852
Change-Id: I00bbcbd7daa03c867732d165be62b72e6c43bce1
Andreas Gampe [Sat, 26 Jul 2014 09:17:40 +0000 (09:17 +0000)]
Merge "ART: Rewrite ParallelGC run-test"
Andreas Gampe [Sat, 26 Jul 2014 08:13:13 +0000 (01:13 -0700)]
ART: Rewrite ParallelGC run-test
To better test what ParallelGC is intended to do, write a version
that will more fairly allocate objects between the threads, and
will let each thread OOM only once. Use a barrier to wait for
completion of all threads, and force a final GC by allocating some
more objects.
Bug:
16406852
Change-Id: I019ddc08515b9610c18d55280cd46cbe3ae3e5ac
Andreas Gampe [Fri, 25 Jul 2014 09:23:04 +0000 (09:23 +0000)]
Merge "ART: Relax CurrentMethodVisitor requirements on GetDexPC"
Andreas Gampe [Thu, 17 Jul 2014 05:20:31 +0000 (22:20 -0700)]
ART: Relax CurrentMethodVisitor requirements on GetDexPC
In case we want to dump a Java stack after an unhandled fault, in
case we hold a thinlocked monitor, that monitor might get inflated.
That can cause an abort as we may not have enough/correct information
for the state at the bottom-most call.
Relax GetDexPc in the CurrentMethodVisitor to not abort when it cannot
find a dex pc. Instead, let the caller handle such a case. This CL
allows the locking_dex_pc_ in Monitor to be DexFile::kDexNoIndex,
which avoids the above abort.
Bug:
16352802,
16556938
Change-Id: I3adf89b2d8f018a0c3e3abdd26e542f46ee59eef
Andreas Gampe [Fri, 25 Jul 2014 08:46:44 +0000 (08:46 +0000)]
Merge "ART: Account for multidex location strings in VMClassLoader"
Andreas Gampe [Thu, 24 Jul 2014 22:35:50 +0000 (15:35 -0700)]
ART: Account for multidex location strings in VMClassLoader
To look up resources, look in the unadorned location.
Bug:
16530747
Change-Id: Ia97e39366444f6666a78ade7298d3c22b4b79f8a
Andreas Gampe [Sat, 26 Jul 2014 09:07:18 +0000 (09:07 +0000)]
Merge "Tweaks to patchoat and other related things."
Alex Light [Thu, 24 Jul 2014 18:29:14 +0000 (11:29 -0700)]
Tweaks to patchoat and other related things.
Removed some flags from patchoat that were poorly specified and fixed
some other issues with the relocation system.
Bug:
15358152
Change-Id: Ia6d47b1a008f02373307d833ba45f00ea408d76f
Serguei Katkov [Thu, 17 Jul 2014 07:39:03 +0000 (14:39 +0700)]
x86: GenSelect utility update
The is follow-up https://android-review.googlesource.com/#/c/101396/
to make x86 GenSelectConst32 implementation complete.
Change-Id: I69f318e18093f9a5b00f8f00f0f1c2e4ff7a9ab2
Signed-off-by: Serguei Katkov <serguei.i.katkov@intel.com>
Andreas Gampe [Wed, 23 Jul 2014 18:07:50 +0000 (18:07 +0000)]
Merge "Add test-art-*-run-test-###-test-name32/64 test targets."
Alex Light [Wed, 23 Jul 2014 17:31:34 +0000 (10:31 -0700)]
Add test-art-*-run-test-###-test-name32/64 test targets.
This allows one to run only a single architecture for one of the run-tests.
Change-Id: Icb06c2b36074cbec7f7808a2fcaba97ec5501578
Andreas Gampe [Wed, 23 Jul 2014 18:12:08 +0000 (18:12 +0000)]
Merge "ART: Fix x86_64 instrumentation_exit, also movsd -> movq"
Andreas Gampe [Wed, 23 Jul 2014 17:05:02 +0000 (10:05 -0700)]
ART: Fix x86_64 instrumentation_exit, also movsd -> movq
Change movd/movsd to movq.
Bug:
16386215
Change-Id: Icca71ca2aeeb2917aff46043051d6046f04395d4
Andreas Gampe [Wed, 23 Jul 2014 17:51:24 +0000 (17:51 +0000)]
Merge "ART: Fix erroneous output"
Andreas Gampe [Wed, 23 Jul 2014 17:47:54 +0000 (10:47 -0700)]
ART: Fix erroneous output
Make run-tests quiet.
Change-Id: I80da9ac04fa95a4094dee874c8703ea563acc1ee
Jeff Hao [Wed, 23 Jul 2014 01:50:09 +0000 (01:50 +0000)]
Merge "Fix x86 instrumentation exit entrypoint and trace size limit."
Jeff Hao [Wed, 23 Jul 2014 01:38:42 +0000 (18:38 -0700)]
Fix x86 instrumentation exit entrypoint and trace size limit.
The x86 instruction movd only copies a 32-bit value when used with a
memory location. movsd properly copies 64-bits.
Bug:
16386215
Change-Id: Ia59b2c2af72ce5202c7b62413687aedb48cfd25e
Mathieu Chartier [Wed, 23 Jul 2014 01:01:30 +0000 (01:01 +0000)]
Merge "Recycle mem-maps for collector transitions."
Mathieu Chartier [Wed, 16 Jul 2014 20:28:58 +0000 (13:28 -0700)]
Recycle mem-maps for collector transitions.
We now create spaces when we need them for collector transitions or
homogeneous compaction by recycling mem maps. Change the bump
pointer space size to be as large as the heap capacity instead of
1/2 heap capacity like it used to be. For GSS, bump pointer spaces
are set to 32MB currently.
Changed GSS to have main space == non moving space since we don't
need to copy from the main space.
Fixes GC stress tests 074, 096.
Fixed test 080 oom throw with -Xmx2m for GC stress test, this was
broken since it was allocating a 4 MB array before starting the
OOM process.
Bug:
14059466
Bug:
16406852
Change-Id: I62877cfa24ec944a6f34ffac30334f454a8002fd
Andreas Gampe [Wed, 23 Jul 2014 17:03:23 +0000 (17:03 +0000)]
Merge "ART: Make run tests out of oat tests"
Andreas Gampe [Wed, 23 Jul 2014 01:52:29 +0000 (18:52 -0700)]
ART: Make run tests out of oat tests
Transforms all former oat tests into run tests.
Change-Id: I190dd39456454c36e5538a2c044d993965a67533
Andreas Gampe [Wed, 23 Jul 2014 01:43:30 +0000 (01:43 +0000)]
Merge "Changed default non-full tests to be prebuild."
Alex Light [Wed, 23 Jul 2014 01:07:12 +0000 (18:07 -0700)]
Changed default non-full tests to be prebuild.
Also added ability to turn off non-prebuild manually.
Change-Id: I0315ac43703a9c7e79812658523013799fb5d5dd
Mingyao Yang [Wed, 23 Jul 2014 00:35:01 +0000 (00:35 +0000)]
Merge "Fix build, missing spaces around =/<."
Mingyao Yang [Wed, 23 Jul 2014 00:33:25 +0000 (17:33 -0700)]
Fix build, missing spaces around =/<.
Change-Id: I2e7824075626a07eccb0a5eb77ef157214fe70fb
Ian Rogers [Tue, 22 Jul 2014 22:05:57 +0000 (22:05 +0000)]
Merge "Put oat test output on stderr."
Ian Rogers [Tue, 22 Jul 2014 21:22:22 +0000 (14:22 -0700)]
Put oat test output on stderr.
Change-Id: Iad318f63263dc5d264d8a84eacedf5065a4e5248
Andreas Gampe [Tue, 22 Jul 2014 21:09:07 +0000 (21:09 +0000)]
Merge "ART: Fix checks for relocation delta"
Andreas Gampe [Tue, 22 Jul 2014 21:04:34 +0000 (14:04 -0700)]
ART: Fix checks for relocation delta
Change < to <= and > to >=.
Change-Id: I7579e3c13dc1b33801ace3a0f8d85cadaee41e6e
Andreas Gampe [Tue, 22 Jul 2014 20:57:39 +0000 (20:57 +0000)]
Merge "ART: Blacklist relocate option in oat tests"
Andreas Gampe [Tue, 22 Jul 2014 18:46:28 +0000 (11:46 -0700)]
ART: Blacklist relocate option in oat tests
To avoid overwriting issues when running the relocate option for
oat tests, blacklist them.
Change-Id: I37685a761fd0949502da134690d88ba2a399ea44
Ian Rogers [Wed, 23 Jul 2014 15:18:54 +0000 (15:18 +0000)]
Merge "Use the TMPDIR environment variable for test directory if set."
Ian Rogers [Wed, 23 Jul 2014 15:03:36 +0000 (08:03 -0700)]
Use the TMPDIR environment variable for test directory if set.
Bug:
16499668
Change-Id: I7da5559c13597d7bca4e4037a1e9335f0b6d8230
Nicolas Geoffray [Wed, 23 Jul 2014 12:26:17 +0000 (12:26 +0000)]
Merge "Fix implicit stack overflow check on optimizing/x86."
Nicolas Geoffray [Wed, 23 Jul 2014 11:57:19 +0000 (12:57 +0100)]
Fix implicit stack overflow check on optimizing/x86.
They need to happen before changing ESP, and require a suspend point.
Change-Id: Id41aa9c99714f7ab8591367ea5cb9ca105b17ce8
Vladimir Marko [Wed, 23 Jul 2014 12:12:24 +0000 (12:12 +0000)]
Merge "Rewrite topological sort order and improve GVN."
Vladimir Marko [Thu, 10 Jul 2014 11:42:52 +0000 (12:42 +0100)]
Rewrite topological sort order and improve GVN.
Rewrite the topological sort order to include a full loop
before the blocks that go after the loop. Add a new iterator
class LoopRepeatingTopologicalSortIterator that differs from
the RepeatingTopologicalSortIterator by repeating only loops
and repeating them early. It returns to the loop head if the
head needs recalculation when we reach the end of the loop.
In GVN, use the new loop-repeating topological sort iterator
and for a loop head merge only the preceding blocks' LVNs
if we're not currently recalculating this loop.
Also fix LocalValueNumbering::InPlaceIntersectMaps() which
was keeping only the last element of the intersection, avoid
some unnecessary processing during LVN merge and add some
missing braces to MIRGraph::InferTypeAndSize().
Bug:
16398693
Change-Id: I4e10d4acb626a5b8a28ec0de106a7b37f9cbca32
Nicolas Geoffray [Wed, 23 Jul 2014 12:07:12 +0000 (12:07 +0000)]
Merge "Use the correct HOST_ARCH for oat files."
Nicolas Geoffray [Wed, 23 Jul 2014 11:37:10 +0000 (12:37 +0100)]
Use the correct HOST_ARCH for oat files.
This typo explains why test-art-host-oat-optimizing-SignalTest32
would work on a 32bit host build and fail on a 64bit host build:
because dex2oat does not generate the code in the expected path
(x86_64 instead of x86), dalvikvm would re-generate it, with
the default settings (that is with quick compiler), and hence the
test would not fail.
Change-Id: I08b5ec316deb4d669fe76066d33a91fc3ed9e3f9
Vladimir Marko [Wed, 23 Jul 2014 12:05:26 +0000 (12:05 +0000)]
Merge "Fix null pointer check elimination for catch entries."
Nicolas Geoffray [Wed, 23 Jul 2014 09:12:11 +0000 (09:12 +0000)]
Merge "Rules depend on DEX2OATD, not DEX2OATD_DEPENDENCY."
Nicolas Geoffray [Wed, 23 Jul 2014 08:59:21 +0000 (09:59 +0100)]
Rules depend on DEX2OATD, not DEX2OATD_DEPENDENCY.
If dex2oatd changes, we need to re-run .oat/.art file generation.
Change-Id: I43487e024745b22b1681f4ff558bac6fd07d9881
Vladimir Marko [Fri, 11 Jul 2014 13:44:36 +0000 (14:44 +0100)]
Fix null pointer check elimination for catch entries.
Remove the special treatment of catch blocks for null
pointer check elimination and class initialization check
elimination. In both cases this can help optimizing
previously missed cases. In the null check case, this
avoids incorrect optimization as exposed by the new test.
Bug:
16230771
Change-Id: I834b7a1835d9ca8572f4f8d8516d93913c701ad1
Mingyao Yang [Tue, 22 Jul 2014 22:05:50 +0000 (22:05 +0000)]
Merge "Set vtable in class object to null after linking."
Mingyao Yang [Wed, 16 Jul 2014 17:44:41 +0000 (10:44 -0700)]
Set vtable in class object to null after linking.
This is follow-up work of embedding imt and vtable for
faster interface/virtual call dispatching.
Once vtable becomes embedded, the original vtable is nulled.
Change-Id: I307696657d1e283654169dbecb8f7815c42bbabc
Andreas Gampe [Tue, 22 Jul 2014 21:40:47 +0000 (21:40 +0000)]
Merge "Made sync-test actually wait for first thread to start."
Alex Light [Tue, 22 Jul 2014 21:04:31 +0000 (14:04 -0700)]
Made sync-test actually wait for first thread to start.
This makes sure that the threads execute in the order we expect.
Change-Id: I0c4976e72ef2eb9e5fd66184eb1cd4821ea1e42b
Andreas Gampe [Tue, 22 Jul 2014 15:25:28 +0000 (15:25 +0000)]
Merge "Runtime can now be set to require relocation"