summaryrefslogtreecommitdiff
path: root/runtime/entrypoints_order_test.cc
Commit message (Collapse)AuthorAgeFilesLines
* Revert^3 "VIXL simulator for ART (Stage1)"Ulya Trafimovich2020-07-291-3/+3
| | | | | | | | | | This reverts commit e886d68b9c40c941d8966b9c90d0e265c75fb19e. Reason for revert: simulator implemention is not ready yet. Test: lunch aosp_cf_x86_phone-userdebug && m Test: art/test.py --run-test --optimizing --host Change-Id: I03c8c09ea348205b0238d7a26caef3477cd6ae3b
* Remove deprecated debug_suspend_count TLS valueAlex Light2020-07-281-2/+1
| | | | | | | | The debug_suspend_count TLS value has been dead for a while and was accidentally left in. Remove it entirely. Test: ./test.py --host Change-Id: Ie2ead0d30e5ff3885cdd83242cad2c826c7fb732
* Revert^2 "VIXL simulator for ART (Stage1)"Ulyana Trafimovich2020-07-171-3/+3
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | This reverts commit 3060bb919cd2f37c6a97e87c1581ac5294af72b3. Reason for revert: relanding original change. The fix is setting `device_supported: false` for libart(d)-simulator module in the .bp file (`m checkbuild` attempted to build it for arm32 and failed). Original commit message: VIXL simulator for ART (Stage1) Quick User Guide: test/README.simulator.md This CL enables running ART run-tests in a simulator on host machine. Some benefits of using this simulator approach: - No need to use a target device at all. Save developers from solving the device troubles: build, flash, usb, adb, etc. - Speed up development/debug/test cycle. - Allows easy debugging/testing new instruction features without real hardware. - Allows using a smaller AOSP Android manifest master-art. The Stage1 CL provides support for running 30% of current run-tests. The rest unsupported test cases are kept in knownfailures.json. Future work will be supporting proper stack frame layout between simulator and quick entrypoints, so that stack walk, QuickArgumentVisitor, deoptimization, etc can be supported. This CL adds libart(d)-simulator-container library to the ART APEX. It has caused the following increase of the APEX size (small, about 0.13% for release APEX, measured for target aosp_arm64-userdebug): Before: 88992 com.android.art.debug.apex 51612 com.android.art.release.apex 112352 com.android.art.testing.apex After: 89124 com.android.art.debug.apex 51680 com.android.art.release.apex 112468 com.android.art.testing.apex Change-Id: I461c80aa9c4ce0673eef1c0254d2c539f2b6a8d5 Test: art/test.py --run-test --optimizing --simulate-arm64 Test: art/test.py --run-test --optimizing --host Test: m test-art-host-gtest
* Revert "VIXL simulator for ART (Stage1)"Ulyana Trafimovich2020-07-161-3/+3
| | | | | | | | | This reverts commit 48ca6a681efe1fa1cf82d8af918bf9bbfd35ae96. Reason for revert: broken build 6685551 on aosp-master on full-eng Bug: 161440641 Change-Id: I849fe53f56c4786f0f2a1605cbfd215559f11072
* VIXL simulator for ART (Stage1)Xueliang Zhong2020-07-161-3/+3
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Quick User Guide: test/README.simulator.md This CL enables running ART run-tests in a simulator on host machine. Some benefits of using this simulator approach: - No need to use a target device at all. Save developers from solving the device troubles: build, flash, usb, adb, etc. - Speed up development/debug/test cycle. - Allows easy debugging/testing new instruction features without real hardware. - Allows using a smaller AOSP Android manifest master-art. The Stage1 CL provides support for running 30% of current run-tests. The rest unsupported test cases are kept in knownfailures.json. Future work will be supporting proper stack frame layout between simulator and quick entrypoints, so that stack walk, QuickArgumentVisitor, deoptimization, etc can be supported. This CL adds libart(d)-simulator-container library to the ART APEX. It has cause the following increase of the APEX size (small, about 0.13% for release APEX, measured for target aosp_arm64-userdebug): Before: 88992 com.android.art.debug.apex 51612 com.android.art.release.apex 112352 com.android.art.testing.apex After: 89124 com.android.art.debug.apex 51680 com.android.art.release.apex 112468 com.android.art.testing.apex Test: art/test.py --run-test --optimizing --simulate-arm64 Test: art/test.py --run-test --optimizing --host Test: m test-art-host-gtest Change-Id: I078812dde9aaf7128d9f262b2102251927596b7f
* Allow late lookup for @CriticalNative methods.Vladimir Marko2020-02-191-2/+6
| | | | | | | | | | | | | | Test: Add and enable tests in 178-app-image-native-method Test: Add and enable tests in jni_compiler_test Test: Manually step through the new stub in GDB and check that backtrace works at various points. Test: m test-art-host-gtest Test: testrunner.py --host --optimizing Test: aosp_taimen-userdebug boots. Test: run-gtests.sh Test: testrunner.py --target --optimizing Bug: 112189621 Change-Id: If094e5062acbb99eefa88f2afb4815f93730cb82
* Remove old JDWP implementation from ARTAlex Light2020-01-291-3/+1
| | | | | | | | | | | | | | | | | | | | The old 'internal' JDWP implementation hasn't been used for a few releases and it's a lot of code that's barely being tested and is at risk of bit-rot. To simplify the runtime and remove potentially buggy code this removes it. We also needed to rewrite the DdmThreadNotification code since it relied on the suspension functionality from the old debugger and was generally unsafe. Test: ./test.py --host Test: atest --test-mapping cts/tests/jdwp/TEST_MAPPING Test: atest --test-mapping cts/hostsidetests/jdwptunnel/TEST_MAPPING Test: Manual ddms Bug: 119034743 Change-Id: I775f310a009141296b730e4a6c2503506a329481
* JIT baseline: trigger optimized compilation on hotness threshold.Nicolas Geoffray2019-12-031-1/+3
| | | | | | | | | | | | | - Add a new hotness count in the ProfilingInfo to not conflict with interpreter hotness which may use it for OSR. - Add a baseline flag in the OatQuickMethodHeader to identify baseline compiled methods. - Add a -Xusetieredjit flag to experiment and test. Bug: 119800099 Test: test.py with Xusetieredjit to true Change-Id: I8512853f869f1312e3edc60bf64413dee9143c52
* Baseline JIT: update inline caches in compiled code.Nicolas Geoffray2019-11-271-2/+5
| | | | | | | | | In trying to remove profiling from interpreter, to speed up interpreter performance. Bug: 119800099 Test: test.py --baseline Change-Id: Ica1fa41a889b31262d9f5691b30a31fbcec01b34
* Revert^4 "Walk internal ArtField/ArtMethod pointers"Alex Light2019-10-071-1/+4
| | | | | | | | | | | | | Several of the new tests make use of the invoke-custom opcode. This opcode is not supported by dexter/slicer causing the tests to fail. This reverts commit c34eab45161c51bf63e548e44645cbcc59d01268. Reason for revert: Added tests to redefine-stress known failures Test: ./test.py --host --redefine-stress Bug: 134162467 Change-Id: Ic1b375a0cb1e44d0252c17115af92c269fb8efc5
* Revert "Revert^2 "Walk internal ArtField/ArtMethod pointers""Nicolas Geoffray2019-10-071-4/+1
| | | | | | | | | | This reverts commit ea2a3d949354c8b054983ba629c81bc5ff7163da. Bug: 134162467 Reason for revert: Fails redefine stress Change-Id: If487c0bcacaf3a3f565ff475b6dad8321e3428b9
* Revert^2 "Walk internal ArtField/ArtMethod pointers"Alex Light2019-10-041-1/+4
| | | | | | | | | | | | | | | | | This reverts commit 712fa800b2b78e527d36c88dc369bf4b723587ea. We incorrectly didn't check if a method was obsolete before giving its class's MethodIds array. We then incorrectly used this array and the (placeholder) -1 index to try to find the previous method-id. Since -1 is not a valid array index we got check failures. To fix this we simply added a check that the method is not obsolete and if it is we go to the slow-path. Reason for revert: Fixed issue causing out-of-bounds array access Test: ./test.py --host --debuggable --ndebuggable Bug: 134162467 Change-Id: Iaffefeab6e889b4fb6554a11452d0af051001cb7
* Revert "Walk internal ArtField/ArtMethod pointers"Nicolas Geoffray2019-10-041-4/+1
| | | | | | | | | | This reverts commit c84fc3a742b160ce51cbf01c2e5f971ccc0a2c6c. Bug: 134162467 Reason for revert: Test fails on debuggable. Change-Id: I240d58fafcc7434749947330b64c67d65b9b7a1e
* Walk internal ArtField/ArtMethod pointersAlex Light2019-10-031-1/+4
| | | | | | | | | | | | | | | During structural class redefinition we sometimes need to update some of the ArtMethod/ArtField pointers held by runtime frames. This adds support for doing this through a StackReflectiveHandleScope similar to the StackHandleScope used for holding object references. This also updates various places where reflective-handles to ArtMethods and ArtFields are needed, for example the JniIdManager, field Read/Write operations and events, field resolution, and the old debugger. Test: ./test.py --host Bug: 134162467 Change-Id: I4ea73e85956a07735c6d7b125c5828a4233670bc
* ART: Optimize StringBuilder append pattern.Vladimir Marko2019-05-161-1/+3
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Recognize appending with StringBuilder and replace the entire expression with a runtime call that perfoms the append in a more efficient manner. For now, require the entire pattern to be in a single block and be very strict about the StringBuilder environment uses. Also, do not accept StringBuilder/char[]/Object/float/double arguments as they throw non-OOME exceptions and/or require a call from the entrypoint back to a helper function in Java; these shall be implemented later. Boot image size for aosp_taimen-userdebug: - before: arm/boot*.oat: 19653872 arm64/boot*.oat: 23292784 oat/arm64/services.odex: 22408664 - after: arm/boot*.oat: 19432184 (-216KiB) arm64/boot*.oat: 22992488 (-293KiB) oat/arm64/services.odex: 22376776 (-31KiB) Note that const-string in compiled boot image methods cannot throw, but for apps it can and therefore its environment can prevent the optimization for apps. We could implement either a simple carve-out for const-string or generic environment pruning to allow this pattern to be applied more often. Results for the new StringBuilderAppendBenchmark on taimen: timeAppendLongStrings: ~700ns -> ~200ns timeAppendStringAndInt: ~220ns -> ~140ns timeAppendStrings: ~200ns -> 130ns Bug: 19575890 Test: 697-checker-string-append Test: m test-art-host-gtest Test: testrunner.py --host --optimizing Test: aosp_taimen-userdebug boots. Test: run-gtests.sh Test: testrunner.py --target --optimizing Test: vogar --benchmark art/benchmark/stringbuilder-append/src/StringBuilderAppendBenchmark.java Change-Id: I51789bf299f5219f68ada4c077b6a1d3fe083964
* ART: Remove setjmp.h includesAndreas Gampe2018-11-211-2/+0
| | | | | | | | We are not using jmpbuf and co. Bug: 119869270 Test: m Change-Id: I85993e2ce506b059801d8d8da8b440e93ee9e3fd
* Remove mterp "alternate" opcode handlers.David Srbecky2018-10-161-3/+1
| | | | | | | | | | They are currently unused and I don't have plans to use them. The alternate table made it possible to enable extra mterp checks. However, it is possible to move the debug checks to the main path. Test: test.py -b -r --interpreter -t 001-HelloWorld Change-Id: I45a39ec73abaefaecf5b8c636f3f9d519a0a8bb0
* Move InterpreterCache to right below tlsPtr_Alex Light2018-10-011-2/+5
| | | | | | | | | | | | Change the position of the InterpreterCache field in Thread to be directly below the tlsPtr_ field. Since both members of the tlsPtr_ and InterpreterCache fields are used by asm_code we need their offsets in asm_support.h. The fields at the end of the Thread struct have been undergoing changes. By moving this field up we avoid the need to update asm_support.h whenever one of the fields is modified. Test: ./test.py --host Change-Id: Ic2863116ed446af155badfc3bf098add7ba0b699
* Rename type resolution entrypoints.Vladimir Marko2018-08-031-3/+3
| | | | | | | | | | | | Rename the InitializeType and InitializeTypeAndVerifyAccess entrypoints to Resolve* to better match their semantics. Keep the InitializeStaticStorage name for now as the most appropriate name InitializeType would clash with the old name of the ResolveType entrypoint. Test: m test-art-host-gtest Test: testrunner.py --host --optimizing Change-Id: Ide55b58c490d085ab37d8536f90699f7ed571d59
* Don't use StringFactory.newEmptyString in compiled codeAlex Light2018-07-101-1/+3
| | | | | | | | | | | | | | | | | | | When compiling debuggable code we would compile a new-instance String instruction into a StringFactory.newEmptyString invoke. This additional invoke could be observed using tracing and is inconsistent with the interpreter, where the string is simply allocated directly. In order to bring these two modes into alignment we added a new AllocStringObject quick entrypoint that will be used instead of the normal AllocObject<...> entrypoints when allocating a string. This entrypoint directly allocates a new string in the same manner the interpreter does. Needs next CL for test to work. Bug: 110884646 Test: ./test/testrunner/testrunner.py --host --runtime-option=-Xjitthreshold:0 --jit Test: Manual inspection of compiled code. Change-Id: I7b4b084bcf7dd9a23485c0e3cd2cd04a04b43d3d
* ART: Adds an entrypoint for invoke-customOrion Hodson2018-06-111-2/+2
| | | | | | | | | | | Add support for the compiler to call into the runtime for invoke-custom bytecodes. Bug: 35337872 Test: art/test.py --host -r -t 952 Test: art/test.py --target --64 -r -t 952 Test: art/test.py --target --32 -r -t 952 Change-Id: I821432e7e5248c91b8e1d36c3112974c34171803
* ART: Compiler support for const-method-handleOrion Hodson2018-05-111-1/+2
| | | | | | | | | | Implemented as a runtime call. Bug: 66890674 Test: art/test.py --target -r -t 979 Test: art/test.py --target --64 -r -t 979 Test: art/test.py --host -r -t 979 Change-Id: I67f461c819a7d528d7455afda8b4a59e9aed381c
* ART: Compiler support for const-method-typeOrion Hodson2018-05-101-1/+2
| | | | | | | | | | Implemented as a runtime call. Bug: 66890674 Test: art/test.py --target -r -t 979 Test: art/test.py --target --64 -r -t 979 Test: art/test.py --host -r -t 979 Change-Id: I4b3d3969d455d0198cfe122eea8abd54e0ea20ee
* ART: Add entrypoint and intrinsic for Math.pow().Vladimir Marko2018-01-221-1/+2
| | | | | | | | | | | | | | | | | | MathBenchmarks.java#timePow results on taimen's little cores fixed at frequency 1401600 with forced JIT compilation: - before: - X32: 356.33 (@FastNative), 315.39 (@CriticalNative) - X64: 357.31 (@FastNative), 315.37 (@CriticalNative) - after (LICM defeats the benchmark): - X32: 2.88 - X64: 2.87 - after but with kAllSideEffects to prevent LICM: - X32: 275.42 - X64: 275.67 Test: Rely on TreeHugger. Bug: 70727450 Change-Id: Iaa31f70acabbd57c163cfeafe02eed67c1348861
* Implement asynchronous exception supportAlex Light2017-09-281-1/+2
| | | | | | | | | | | | | | | This CL implements async exception support in the switch interpreter. It also adds support for the MTerp to detect and switch back to the switch interpreter in cases where an async exception is detected. Tests follow in next CL. Test: ./test.py --host -j50 Bug: 62821960 Bug: 34415266 Change-Id: Idb53711a40c20f962de8aa6b74662676b8bd25c6
* ART: Include cleanupAndreas Gampe2017-07-241-1/+2
| | | | | | | | | | | | | | | | | | | Let clang-format reorder the header includes. Derived with: * .clang-format: BasedOnStyle: Google IncludeIsMainRegex: '(_test|-inl)?$' * Steps: find . -name '*.cc' -o -name '*.h' | xargs sed -i.bak -e 's/^#include/ #include/' ; git commit -a -m 'ART: Include cleanup' git-clang-format -style=file HEAD^ manual inspection git commit -a --amend Test: mmma art Change-Id: Ia963a8ce3ce5f96b5e78acd587e26908c7a70d02
* Fix some gtestsMathieu Chartier2017-04-191-1/+2
| | | | | | | | Forgot to retest. Test: test-art-host-gtest Change-Id: I72b07c3872079452a3a01db4fbd2c4ee0060f294
* Catch signals that happen inside a fault handler.Josh Gao2017-03-081-2/+1
| | | | | | | | | | | | | | | | Unblock some signals (SIGABRT, SIGBUS, SIGSEGV) that could happen inside of the ART internal fault handlers, to report crashes inside of the signal handler. Because we can't use sigaction to change the handler when this happens, because it modifies global state, add a new member variable in Thread to track whether a call to the fault handler is reentrant or not. Remove the old nested signal implementation that attempted to do this. Bug: http://b/35853436 Test: changed the #if 0 to #if 1, ran a dummy process that threw a NullPointerException, inspected logcat Change-Id: I04bb4a09433c6817933d64ec681ec433b528f2a5
* Add per array size allocation entrypoints.Nicolas Geoffray2017-01-261-1/+9
| | | | | | | | | - Update architectures that have fast paths for array allocation to use it. - Will add more fast paths in follow-up CLs. Test: test-art-target test-art-host. Change-Id: I138cccd16464a85de22a8ed31c915f876e78fb04
* Remove unused array entrypoints.Nicolas Geoffray2017-01-191-6/+3
| | | | | Test: test-art-host test-art-target Change-Id: I910d1c912c7c9056ecea0e1e7da7afb2a7220dfa
* Entrypoints cleanup.Nicolas Geoffray2017-01-181-12/+3
| | | | | | | | Remove unused ones to facilitate the transition to compressed dex caches. test: test-art-host, test-art-target Change-Id: I1d1cb0daffa86dd9dda2eaa3c1ea3650a5c8d9d0
* Move some fields in art::Thread to reduce maintenance burden.Roland Levillain2017-01-171-6/+5
| | | | | | | | | | | | | Move fields `thread_local_start`, `thread_local_pos`, `thread_local_end` and `thread_local_objects` before fields `jni_entrypoints` and `quick_entrypoints` within art::Thread, to avoid repetitive art::Thread field moves in future CLs caused by the addition or deletion of entry points. Test: m test-art-host test: m test-art-target (on ARM) Change-Id: Ib67842e44a7f21a871ca4d1bb95dc6f7cfedc829
* Revert "Revert "ART: Compiler support for invoke-polymorphic.""Orion Hodson2017-01-151-3/+5
| | | | | | | | | | | | | | | | | | | This reverts commit 0fb5af1c8287b1ec85c55c306a1c43820c38a337. This takes us back to the original change and attempts to fix the issues encountered: - Adds transition record push/pop around artInvokePolymorphic. - Changes X86/X64 relocations for MacSDK. - Implements MIPS entrypoint for art_quick_invoke_polymorphic. - Corrects size of returned reference in art_quick_invoke_polymorphic on ARM. Bug: 30550796,33191393 Test: art/test/run-test 953 Test: m test-art-run-test Change-Id: Ib6b93e00b37b9d4ab743a3470ab3d77fe857cda8
* Revert "Revert "Make object allocation entrypoints only take a class.""Nicolas Geoffray2017-01-121-6/+6
| | | | | | This reverts commit f7aaacd97881c6924b8212c7f8fe4a4c8721ef53. Change-Id: I6756cd1e6110bb45231f62f5e388f16c044cb145
* Revert "Make object allocation entrypoints only take a class."Hiroshi Yamauchi2017-01-121-6/+6
| | | | | | | | 960-default-smali64 is failing. This reverts commit 2b615ba29c4dfcf54aaf44955f2eac60f5080b2e. Change-Id: Iebb8ee5a917fa84c5f01660ce432798524d078ef
* Make object allocation entrypoints only take a class.Nicolas Geoffray2017-01-111-6/+6
| | | | | | | | | | | | | | | | | | | | | | | Change motivated by: - Dex cache compression: having the allocation fast path do a dex cache lookup will be too expensive. So instead, rely on the compiler having direct access to the class (either through BSS for AOT, or JIT tables for JIT). - Inlining: the entrypoints relied on the caller of the allocation to have the same dex cache as the outer method (stored at the bottom of the stack). This meant we could not inline methods from a different dex file that do allocations. By avoiding the dex cache lookup in the entrypoint, we can now remove this restriction. Code expansion on average for Docs/Gms/FB/Framework (go/lem numbers): - Around 0.8% on arm64 - Around 1% for x64, arm - Around 1.5% on x86 Test: test-art-host, test-art-target, ART_USE_READ_BARRIER=true/false Test: test-art-host, test-art-target, ART_DEFAULT_GC_TYPE=SS ART_USE_TLAB=true Change-Id: I41f3748bb4d251996aaf6a90fae4c50176f9295f
* Use a per-thread VerifierDeps.Nicolas Geoffray2016-11-221-2/+2
| | | | | | | | | | | | | | | 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
* Change check cast entrypoint to check instance ofMathieu Chartier2016-11-081-2/+3
| | | | | | | | | | | | | | | Reduces code size since we do not need to reload class before calling slow path. TODO: Delete read barriers in the check cast code since the slow path will retry with the proper read barriers if the check fails. Bug: 12687968 Bug: 29516974 Test: test-art-host + test-art-target with CC Change-Id: Ia4eb9bbe3fe2d2016e44523cf0451210828d7b88
* jni: Support @FastNative methods that return objectsIgor Murashkin2016-10-191-2/+4
| | | | | Bug: 32088975 Change-Id: I16f8b7ec6b251812af60ab25f2153d9b72f37044
* Remove ThrowNoSuchMethod entrypoint.Vladimir Marko2016-09-141-5/+4
| | | | | | Run ART test suite on host and Nexus 6. Bug: 31464666 Change-Id: I5aa737726031adae0b132f759cf802a93d581a7f
* Fix clang-tidy warnings in art.Chih-Hung Hsieh2016-08-261-3/+4
| | | | | | | | | * Add parentheses around macro parameters, or use NOLINT to suppress warning. Bug: 28705665 Test: build with WITH_TIDY=1 Change-Id: Ifc922c2e66215772042bac372754ea70074f0053
* jni: Fast path for @FastNative annotated java methodsIgor Murashkin2016-08-161-2/+5
| | | | | | | | | | | | | | | | Adds a faster path for java methods annotated with dalvik.annotation.optimization.FastNative . Intended to replace usage of fast JNI (registering with "!(FOO)BAR" descriptors). Performance Microbenchmark Results (Angler): * Regular JNI cost in nanoseconds: 115 * Fast JNI cost in nanoseconds: 60 * @FastNative cost in nanoseconds: 36 Summary: Up to 67% faster (vs fast jni) JNI transition cost Change-Id: Ic23823ae0f232270c068ec999fd89aa993894b0e
* Remove two ReadBarrierMarkRegX entrypoints.Roland Levillain2016-07-251-5/+1
| | | | | | | | | | | | As entry points ReadBarrierMarkReg30 and ReadBarrierMarkReg31 are undefined on all architectures supporting the read barrier configuration (ARM, ARM64, x86 and x86-64), remove them from the entry point list. Test: ART host and target (ARM, ARM64) tests. Bug: 29506760 Bug: 12687968 Change-Id: I500626e54f00aebfc095b4ef5f81b49fa43f7768
* Introduce more compact ReadBarrierMark slow-paths.Roland Levillain2016-07-131-5/+67
| | | | | | | | | | | | | Replace entry point ReadBarrierMark with 32 ReadBarrierMarkRegX entry points, using register number X as input and output (instead of the standard runtime calling convention) to save two moves in Baker's read barrier mark slow-path code. Test: ART host and target (ARM, ARM64) tests. Bug: 29506760 Bug: 12687968 Change-Id: I73cfb82831cf040b8b018e984163c865cc44ed87
* Replace String.charAt() with HIR.Vladimir Marko2016-06-211-3/+4
| | | | | | | | | | | Replace String.charAt() with HArrayLength, HBoundsCheck and HArrayGet. This allows GVN on the HArrayLength and BCE on the HBoundsCheck as well as using the infrastructure for HArrayGet, i.e. better handling of constant indexes than the old intrinsic and using the HArm64IntermediateAddress. Bug: 28330359 Change-Id: I32bf1da7eeafe82537a60416abf6ac412baa80dc
* Support unlimited pending checkpointsMathieu Chartier2016-06-131-3/+5
| | | | | | | | | | | | | | | | | | | Prevents the spinning that used to happen if RunCheckpoint was called with 3 pending checkpoints. This spinning was done when holding thread_list_lock_ and thread_suspend_count_lock_ and could deadlock if any of the pending checkpoints required any of these locks. The fix is to use an overflow list instead of having a fixed limit of 3. Changed suspend stress test to have more threads and only compare last line since there may be libbacktrace spam like: "+E/libbacktrace(69891): void SignalHandler(int, siginfo_t *, void *): Timed out waiting for unwind thread to indicate it completed." Bug: 28988206 Change-Id: I2ae611506147d5199d59a08eee0395f7fa35d448
* Assembly TLAB allocation fast path for arm.Hiroshi Yamauchi2016-02-251-2/+2
| | | | | | | | | Speedup (GSS GC with TLAB on N5): BinaryTrees: 1872 -> 796 ms (-57%) MemAllocTest: 2522 -> 2219 ms (-12%) Bug: 9986565 Change-Id: Icb9d1259461f3abe83a4a82c8aff911937eaf57d
* Use 8-byte increment bracket sizes for rosalloc thread local runs.Hiroshi Yamauchi2016-02-051-1/+1
| | | | | | | | | | | Very small space savings (< 1%) after device boot and up to 10% allocation speedup. Some minor cleanup. Bug: 9986565 Change-Id: I51d791c4674d6944fe9a7ee78537ac3490c1a02c
* Fast Art interpreterbuzbee2016-01-111-1/+4
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Add a Dalvik-style fast interpreter to Art. Three primary deficiencies in the existing Art interpreter will be addressed: 1. Structural inefficiencies (primarily the bloated fetch/decode/execute overhead of the C++ interpreter implementation). 2. Stack memory wastage. Each managed-language invoke adds a full copy of the interpreter's compiler-generated locals on the shared stack. We're at the mercy of the compiler now in how much memory is wasted here. An assembly based interpreter can manage memory usage more effectively. 3. Shadow frame model, which not only spends twice the memory to store the Dalvik virtual registers, but causes vreg stores to happen twice. This CL mostly deals with #1 (but does provide some stack memory savings). Subsequent CLs will address the other issues. Current status: Passes all run-tests. Phone boots interpret-only. 2.5x faster than Clang-compiled Art goto interpreter on fetch/decode/execute microbenchmark, 5x faster than gcc-compiled goto interpreter. 1.6x faster than Clang goto on Caffeinemark overall 2.0x faster than Clang switch on Caffeinemark overall 68% of Dalvik interpreter performance on Caffeinemark (still much slower, primarily because of poor invoke performance and lack of execute-inline) Still nearly an order of magnitude slower than Dalvik on invokes (but slightly better than Art Clang goto interpreter. Importantly, saves ~200 bytes of stack memory per invoke (but still wastes ~400 relative to Dalvik). What's needed: Remove the (large quantity of) bring-up hackery in place. Integrate into the build mechanism. I'm still using the old Dalvik manual build step to generate assembly code from the stub files. Remove the suspend check hack. For bring-up purposes, I'm using an explicit suspend check (like the other Art interpreters). However, we should be doing a Dalvik style suspend check via the table base switch mechanism. This should be done during the alternative interpreter activation. General cleanup. Add CFI info. Update the new target bring-up README documentation. Add other targets. In later CLs: Consolidate mterp handlers for expensive operations (such as new-instance) with the code used by the switch interpreter. No need to duplicate the code for heavyweight operations (but will need some refactoring to align). Tuning - some fast paths needs to be moved down to the assembly handlers, rather than being dealt with in the out-of-line code. JIT profiling. Currently, the fast interpreter is used only in the fast case - no instrumentation, no transactions and no access checks. We will want to implement fast + JIT-profiling as the alternate fast interpreter. All other cases can still fall back to the reference interpreter. Improve invoke performance. We're nearly an order of magnitude slower than Dalvik here. Some of that is unavoidable, but I suspect we can do better. Add support for our other targets. Change-Id: I43e25dc3d786fb87245705ac74a87274ad34fedc
* x86 Baker's read barrier fast path implementation.Roland Levillain2015-12-151-1/+2
| | | | | | | | | | | | | | | Introduce an x86 fast path implementation in Optimizing for Baker's read barriers (for both heap reference loads and GC root loads). The marking phase of the read barrier is performed by a slow path, invoking a new runtime entry point (artReadBarrierMark). Other read barrier algorithms continue to use the original slow path based implementation, which has been renamed as GenerateReadBarrierSlow/GenerateReadBarrierForRootSlow. Bug: 12687968 Change-Id: Ie610c4befc19ff22378a8cba38b422dcacb54320
generated by cgit v1.2.3 (git 2.25.1) at 2025-12-20 17:26:16 +0000