GitLab

We would run into issues if we tried to compile an obsolete method
since some of the information needed to do so is lost as we redefine
the declaring class. Since compiling these methods is unlikely to have
a major effect on the performance of the program we simply tell the
JIT not to bother with them.

Also update all places to use the new SetDontCompile helper.

Test: mma -j40 test-art-host
Test: stress --cpu 60 &; \
    while ./test/run-test --host --jit --gcstress --64 942; do ; done

Change-Id: Ie70fee6f503818e7589571facf28ccc5e7ca79f8

We were not adding some DexFiles to the ClassPath if there were
multiple classes from the same classloader being redefined at the same
time. We fixed this issue and made a test for it.

Test: mma -j40 test-art-host

Change-Id: I6e8961c8602367ebec5d5a948d71e58f3be2f6d7

This requires two native methods, isAnonymousClass and
getDeclaringClass.

Add tests.

Bug: 34890992
Test: m
Test: m test-art-host
Test: Device boots
Change-Id: Ib3fca2a6bb6e367ef202ff864719cab979d6c793

Make BitMemoryRegion constructor inlined, remove read barrier
for IsProxyMethod.

Around 15% speedup for pmd benchmark, maybe more for CC.

Test: test-art-host

Change-Id: Ib4392649e041406e538cc944c26c69f68d388fb4

This patch implements min/max intrinsics for:
  * Long
  * Float
  * Double

Test: m test-art-host
Test: m test-art-target
Test: 082-inline-execute

Change-Id: I2dfab8ab606f3d01fba712f9014d2e0617449d74

Test: test-art-host

Change-Id: I16e3a0a17ca6193a68932c78f67e0d788684e7ef

Fix infinite loop for unrecognized options.

Test: art/tools/buildbot-build.sh --host
Test: art/tools/buildbot-build.sh --mode=host
Change-Id: I9f3b53021d83ddef2dca89b414065b9ea0bfafbf

Test: booted MIPS32R2 in QEMU
Test: booted MIPS64 (with 2nd arch MIPS32R6) in QEMU

Change-Id: Iff61353bc9e787e65f3f17d8938d4b5561d2a603

Asan reserves 0x04000000 - 0x20000000 in 32-bit.

Bug: 34606909
Test: marlin-userdebug asan coverage build boots.
Test: test-art-host
Change-Id: I3ee9cd290cd02de02e7d6c79b854479a985549c3

Using the dex cache is faster than going through the class linker.
This reverts to behavior from before aog/321573.

Speeds up pmd benchmark by ~50%.

Test: test-art-host

Change-Id: I58403aec03e2b7e7a3d3e108319cfb4c75a680cb

Cache stack frames from the counting visitor to avoid needing to
walk the stack twice in the common case.

Speeds up pmd benchark by 20%.

Test: test-art-host

Change-Id: I81e4e55280d9c1ccf1937a7ea12abff75e9abb94

The rest of utils.cc does not depend on art::Runtime. This separates
the part dependent on that class, so that including utils.cc in the
build does not require the entire Runtime. Another preparatory cleanup
to getting tools to build on Windows.

Bug: 22322814
Test: test-art
Change-Id: I194ff363fc2ab87e5311ecea6973a2d0fad2621d

We were using recursive loading of the current art method which was
making us miss obsolete methods in some cases.

We could also end up checking the wrong method when walking the stack.

We also add tests for recursive obsolete methods in general.

Bug: 34815470

Test: mma -j40 test-art-host
Test: ART_TEST_JIT=true mma -j40 test-art-host
Change-Id: I522fd4cac4e3f9d35d03b128bad6d6971cfe6c4a

Data is commonly shared between different stack maps. The register
masks are stored after the stack masks.

Oat size for a large app:
96722288 -> 94485872 (-2.31%)

Average oat size reduction according to golem -3.193%.

Bug: 34621054

Test: test-art-host

Change-Id: I5eacf668992e866d11ddba0c01675038a16cdfb4

We were letting the jit and the main thread race on class definition.
If the 'leader' changes then we would end up with the test class
having an unexpected definition, causing the test to fail. We fixed
this by making sure all threads get the same definition regardless of
speed.

Bug: 34799243
Test: mma -j40 test-art-host
Test: stress --cpu 60 &; \
      while ./test/run-test --host --jit --gcstress --64 934 ; do ; done
Test: stress --cpu 60 &; \
      while ./test/run-test --host --jit --gcstress --64 935 ; do ; done
Test: stress --cpu 60 &; \
      while ./test/run-test --host --jit --gcstress --64 938 ; do ; done

Change-Id: I0d6785b4c3c41db1d655cf1ff22be60809923a7a

Once mterp's branch profiling identifies a method hot enough
to try on-stack replacement, a request is made to the JIT to compile
the method for OSR.  Mterp then enters a mode in which it checks
for completion of the compilation in order to initiate the OSR.
Currently, this completion check happens on every branch. In
situations in which the JIT is backlogged and it takes awhile for the
compilation to complete, the overhead of doing these checks is noticable.

This change moves from a "check on every branch" model to a "check
on every Nth branch" model.  We start with N=100, which should still
yield responsive OSR but dramatically reduce the checking overhead.

Bug: 32090348
Test: m test-art-host
Test: m test-art-target (Nexus 5x)
Change-Id: I97442723397bb242163dc18bd4444977bcd469fa

Reverting to work around some programs crashing with
    Check failed: handle_scope_iface.Get() != nullptr.
though the reason for the failure not yet understood.

Test: m test-art-host
Bug: 34839984
Bug: 30627598
Bug: 34659969
Bug: 30419309

This reverts commit ec786228.

Change-Id: Ifded663633082f1e59e5b6ff2e026dc559bd6b82