| 46ea0147b49e3539492be160e1631e73f58d2c3c |
|
22-Apr-2016 |
Vladimir Marko <vmarko@google.com> |
Reduce memory lost by ArenaAllocator for large allocations.
When allocating from a new arena, check if the old arena has
more remaining space than the new one after the current
allocation. If so, keep using the old arena to reduce the
amount of "lost" arena memory. This can happen when we try
to allocate more than half the default arena size. If the
allocation exceeds the default arena size, it's very likely
to happen even though the ArenaPool could still provide some
much larger previously allocated arena.
Also avoid artithmetic overflow when checking if the
request can be satisfied from the current arena.
And abort immediately if calloc() fails.
Bug: 28173563
Bug: 28256882
In addition to the initial CL
(cherry picked from commit 3e0e7173c0cdfc57dba39fe781e30d187d50fa9c)
this contains a squashed subsequent fix
Fix valgrind tests: mark allocated space as defined.
(cherry picked from commit 3f84f2cb3cadc25d75e1e3e2c1bc26c1a671f336)
Change-Id: Id80d5601874e8e28d930c0dd47a51c73c4810094
|
| d1ee80948144526b985afb44a0574248cf7da58a |
|
13-Apr-2016 |
Vladimir Marko <vmarko@google.com> |
Move Assemblers to the Arena.
And clean up some APIs to return std::unique_ptr<> instead
of raw pointers that don't communicate ownership.
(cherry picked from commit 93205e395f777c1dd81d3f164cf9a4aec4bde45f)
Bug: 27505766
Change-Id: I3017302307a0253d661240750298802fb0d9585e
|
| 3a40bf2fc1b9823e3bc8fcf96d5242668b5c088b |
|
22-Mar-2016 |
Vladimir Marko <vmarko@google.com> |
ART: Clean up arena allocation memory tracking.
Remove unused allocation types, mostly from removed Quick.
Move logging one level up to capture memory used by stack
maps during AOT compilation. Raise the reporting threshold
to 8MiB to limit the output to the worst offenders.
Change-Id: I8c7a01bfa90bc8ec5eab66187eb6850a022f3543
|
| f6a35de9eeefb20f6446f1b4815b4dcb0161d09c |
|
21-Mar-2016 |
Vladimir Marko <vmarko@google.com> |
Optimizing: Fix register allocator validation memory usage.
Also attribute ArenaBitVector allocations to appropriate
passes. This was used to track down the source of the
excessive memory alloactions.
Bug: 27690481
Change-Id: Ib895984cb7c04e24cbc7abbd8322079bab8ab100
|
| 25e0456b6ea13eba290b63ea88b6b7120ed89413 |
|
01-Mar-2016 |
Nicolas Geoffray <ngeoffray@google.com> |
Give the JIT its own arena pool to avoid lock contentions.
Sharing it with the verifier and the class loader is not ideal,
especially at startup time.
bug:27398183
bug:23128949
Change-Id: I1b91663a13f6c5b33ad3b4be780d93eb7fe445b4
|
| ca76a1a0c1737b3e04961ba382c113102fdc00bf |
|
02-Feb-2016 |
Jean-Philippe Halimi <jean-philippe.halimi@intel.com> |
Dex-wide ArenaPool scoping for AOT compilation
The Arena Pool is a structure that holds memory allocated by the Arena
Allocator, preventing system allocations and deallocations to occur too often
during AOT compilation. Currently, the Arena Pool holds memory it allocates
during the whole AOT compilation process. Unfortunately, such a behavior
generates memory usage overhead, which is exacerbated by bigger applications
such as Facebook.
In this particular app, method size imbalance provokes unnecessary memory
pressure, as one method might require a lot of arena allocations that won't be
used in the remaining compilation. Because the compiler memory footprint keeps
increasing during AOT compilation, the memory pressure becomes very high.
The proposed patch is an attempt to find a tradeoff between allocations /
deallocations time overhead, and the aforementioned memory pressure overhead
resulting of the allocations being held by the Arena Pool. The patch adds a
feature freeing up all memory allocated in the Arena Pool after each dex file
is done compiling.
We have measured no significant AOT compile-time overhead with the patch
(<0.3%), because the selected dex-file granularity is coarse enough. In the
meantime, it provides significant memory footprint improvements. The impact is
especially big with Facebook, because this app has a few methods generating
huge memory footprint peaks (peak native heap footprint goes down 34%).
Change-Id: I27e867e6a20b8a6c28a82cb83140941a8c2b5847
Signed-off-by: Jean-Philippe Halimi <jean-philippe.halimi@intel.com>
|
| fda043241caada72fbf1215ac8c6eb5ad1ad44e7 |
|
11-Nov-2015 |
Vladimir Marko <vmarko@google.com> |
ART: Add comments to Scoped-/ArenaAllocator.
Change-Id: Iba7350114fb998ac9ee1466ef12f4a5024a18e8e
|
| de40d478930d0889a2aea5cbf58aa63da24e5dfa |
|
16-Oct-2015 |
Mathieu Chartier <mathieuc@google.com> |
Use arenas for the verifier
Improvements are from using arenas for verifier. The things that
were moved into arenas are register lines, reg types, and reg type
descriptors.
Also some minor cleanup. Fixed double space formatting error in
string piece.
========================================
Before:
=======================================
compile GmsCore 50x on host -j1:
real 3m5.510s
user 2m48.139s
sys 0m11.753s
50 random APKs:
real 1m18.157s
user 1m8.167s
sys 0m8.071s
RAM (GmsCore):
(threads: 1) arena alloc=0B java alloc=27MB native alloc=64MB free=792KB
(threads: 4) arena alloc=0B java alloc=27MB native alloc=64MB free=397KB
Perf:
=============================
After:
=============================
compile GmsCore 50x on host -j1:
real 2m31.455s
user 2m14.784s
sys 0m10.888s
50 random APKs:
Speed up from arena allocating:
real 1m12.094s
user 1m2.325s
sys 0m7.872s
Speed up from areans + CreateFindClassDefIndex + unordered_map for DexFileVerifier.
real 1m8.654s
user 0m58.955s
sys 0m7.777s
RAM (GmsCore):
(threads: 1) arena alloc=1049KB java alloc=27MB native alloc=65MB free=538KB
(threads: 4) arena alloc=4MB java alloc=27MB native alloc=69MB free=650KB
Perf on host key results:
_int_malloc: 3.60% -> 1.39%
malloc: 1.48% -> 0.65%
_int_free: 1.59% -> 0.61%
RegTypeCache::FromClass: 1.47% -> 0.86%
malloc_consolidate: 0.89% -> 0.63%
Real speed up is ~22% on GmsCore, ~14% on the set of 60 APKs.
Bug: 10921004
Change-Id: If13d4ab4284a176f93a26a412549b6b1149dfb16
|
| 8df69d42a9e3ccd9456ff72fac8dbd1999f98755 |
|
23-Oct-2015 |
Mingyao Yang <mingyao@google.com> |
Revert "Revert "load store elimination.""
This reverts commit 8030c4100d2586fac39ed4007c61ee91d4ea4f25.
Change-Id: I79558d85484be5f5d04e4a44bea7201fece440f0
|
| 4b8f1ecd3aa5a29ec1463ff88fee9db365f257dc |
|
26-Aug-2015 |
Roland Levillain <rpl@google.com> |
Use ATTRIBUTE_UNUSED more.
Use it in lieu of UNUSED(), which had some incorrect uses.
Change-Id: If247dce58b72056f6eea84968e7196f0b5bef4da
|
| 8030c4100d2586fac39ed4007c61ee91d4ea4f25 |
|
15-Oct-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Revert "load store elimination."
Breaks libcore tests:
libcore.java.lang.ref.FinalizeTest#testWatchdogDoesNotFailForObjectsThatAreNearTheDeadline
libcore.java.util.ResourceLeakageDetectorTest#testDetectsUnclosedCloseGuard
org.apache.harmony.tests.java.lang.ref.ReferenceTest#test_finalizeReferenceInteraction
This reverts commit 589dac7f0ce078d19aad7e35bb0195c47ddf01d2.
Change-Id: I55115765c10762d5bc152d3425e4622560d8b9f4
|
| 589dac7f0ce078d19aad7e35bb0195c47ddf01d2 |
|
24-Aug-2015 |
Mingyao Yang <mingyao@google.com> |
load store elimination.
This adds a pass to eliminate some unnecessary heap loads/stores. It
first collects heap locations and then tracks values stored to those heap
locations. Alias analysis is done based on offset, type, singleton,
pre-existence, etc.
Change-Id: I11a9d8ef20d1b2f245607eb25118e9aff9be472a
|
| 655e585073ac271cc9afa7c9d6ff5ab4dbe4b72e |
|
12-Oct-2015 |
Vladimir Marko <vmarko@google.com> |
Optimizing: Move GraphChecker memory allocations to arena.
Bug: 18120045
Change-Id: I3934158e6ea4868d9baa1dfcc53b603ca6c521e2
|
| 7bda3b600d74c9a5746840ebb8534443e486615a |
|
07-Oct-2015 |
Vladimir Marko <vmarko@google.com> |
Revert "Quick: Work around large frame sizes for x86_64."
Instead, move the MEMORY_TOOL_MAKE_* calls out-of-line.
Bug: 24729377
This reverts commit d4cf1e4fe0624b99df22ed5556dc1d042b32a7e0.
Change-Id: I9bccc8bd4a691a2d018b84de8b95bc68fafad4e1
|
| 2a408a3bef330551818f9cec9a7c5aa7a3f1129e |
|
18-Sep-2015 |
Vladimir Marko <vmarko@google.com> |
ART: Mark deallocated arena memory as inaccessible.
Mark arena and scoped arena memory freed by allocator
adapters as inaccessible. This can help catch accesses to
old storage of a container, for example the old data of an
ArenaVector<> that's been resized.
Together with debug-mode enforcement of destruction of all
scoped arena containers, this provides strong verification
of their memory usage. However, this does not apply to the
normal (non-scoped) arena memory held by arena containers
as they are typically not destroyed if they are themselves
located in the arena. ArenaBitVector memory, whether in
normal or scoped arena, isn't marked either.
Change-Id: I4d2a80fedf7ceb7d4ce24ee8e7bcd53513171388
|
| 5233f93ee336b3581ccdb993ff6342c52fec34b0 |
|
29-Sep-2015 |
Vladimir Marko <vmarko@google.com> |
Optimizing: Tag even more arena allocations.
Tag previously "Misc" arena allocations with more specific
allocation types. Move some native heap allocations to the
arena in BCE.
Bug: 23736311
Change-Id: If8ef15a8b614dc3314bdfb35caa23862c9d4d25c
|
| 225b6464a58ebe11c156144653f11a1c6607f4eb |
|
28-Sep-2015 |
Vladimir Marko <vmarko@google.com> |
Optimizing: Tag arena allocations in code generators.
And completely remove the deprecated GrowableArray.
Replace GrowableArray with ArenaVector in code generators
and related classes and tag arena allocations.
Label arrays use direct allocations from ArenaAllocator
because Label is non-copyable and non-movable and as such
cannot be really held in a container. The GrowableArray
never actually constructed them, instead relying on the
zero-initialized storage from the arena allocator to be
correct. We now actually construct the labels.
Also avoid StackMapStream::ComputeDexRegisterMapSize() being
passed null references, even though unused.
Change-Id: I26a46fdd406b23a3969300a67739d55528df8bf4
|
| 2aaa4b5532d30c4e65d8892b556400bb61f9dc8c |
|
17-Sep-2015 |
Vladimir Marko <vmarko@google.com> |
Optimizing: Tag more arena allocations.
Replace GrowableArray with ArenaVector and tag arena
allocations with new allocation types.
As part of this, make the register allocator a bit more
efficient, doing bulk insert/erase. Some loops are now
O(n) instead of O(n^2).
Change-Id: Ifac0871ffb34b121cc0447801a2d07eefd308c14
|
| 71bf8090663d02869cafafdd530976f7f2a9db7f |
|
15-Sep-2015 |
Vladimir Marko <vmarko@google.com> |
Optimizing: Tag arena allocations in SsaBuilder.
Replace GrowableArray with ArenaVector in SsaBuilder and
tag allocations with a new arena allocation type.
Change-Id: I27312c51d7be9d2ad02a974cce93b365c65c5fc4
|
| fa6b93c4b69e6d7ddfa2a4ed0aff01b0608c5a3a |
|
15-Sep-2015 |
Vladimir Marko <vmarko@google.com> |
Optimizing: Tag arena allocations in HGraph.
Replace GrowableArray with ArenaVector in HGraph and related
classes HEnvironment, HLoopInformation, HInvoke and HPhi,
and tag allocations with new arena allocation types.
Change-Id: I3d79897af405b9a1a5b98bfc372e70fe0b3bc40d
|
| 6058455d486219994921b63a2d774dc9908415a2 |
|
03-Sep-2015 |
Vladimir Marko <vmarko@google.com> |
Optimizing: Tag basic block allocations with their source.
Replace GrowableArray with ArenaVector in HBasicBlock and,
to track the source of allocations, assign one new and two
Quick's arena allocation types to these vectors. Rename
kArenaAllocSuccessor to kArenaAllocSuccessors.
Bug: 23736311
Change-Id: Ib52e51698890675bde61f007fe6039338cf1a025
|
| 145acc5361deb769eed998f057bc23abaef6e116 |
|
03-Sep-2015 |
Vladimir Marko <vmarko@google.com> |
Revert "Optimizing: Tag basic block allocations with their source."
Reverting so that we can have more discussion about the STL API.
This reverts commit 91e11c0c840193c6822e66846020b6647de243d5.
Change-Id: I187fe52f2c16b6e7c5c9d49c42921eb6c7063dba
|
| 91e11c0c840193c6822e66846020b6647de243d5 |
|
02-Sep-2015 |
Vladimir Marko <vmarko@google.com> |
Optimizing: Tag basic block allocations with their source.
Replace GrowableArray with ArenaVector in HBasicBlock and,
to track the source of allocations, assign one new and two
Quick's arena allocation types to these vectors. Rename
kArenaAllocSuccessor to kArenaAllocSuccessors.
Bug: 23736311
Change-Id: I984aef6e615ae2380a532f5c6726af21015f43f5
|
| f9f6441c665b5ff9004d3ed55014f46d416fb1bb |
|
02-Sep-2015 |
Vladimir Marko <vmarko@google.com> |
Optimizing: Tag Arena allocations with their source.
This adds the ability to track where we allocate memory
when the kArenaAllocatorCountAllocations flag is turned on.
Also move some allocations from native heap to the Arena
and remove some unnecessary utilities.
Bug: 23736311
Change-Id: I1aaef3fd405d1de444fe9e618b1ce7ecef07ade3
|
| 3887c468d731420e929e6ad3acf190d5431e94fc |
|
12-Aug-2015 |
Roland Levillain <rpl@google.com> |
Remove unnecessary `explicit` qualifiers on constructors.
Change-Id: Id12e392ad50f66a6e2251a68662b7959315dc567
|
| 90443477f9a0061581c420775ce3b7eeae7468bc |
|
17-Jul-2015 |
Mathieu Chartier <mathieuc@google.com> |
Move to newer clang annotations
Also enable -Wthread-safety-negative.
Changes:
Switch to capabilities and negative capabilities.
Future work:
Use capabilities to implement uninterruptible annotations to work
with AssertNoThreadSuspension.
Bug: 20072211
Change-Id: I42fcbe0300d98a831c89d1eff3ecd5a7e99ebf33
|
| 1e13374baf7dfaf442ffbf9809c37c131d681eaf |
|
20-May-2015 |
Evgenii Stepanov <eugenis@google.com> |
Generalize Valgrind annotations in ART to support ASan.
Also add redzones around non-fixed mem_map(s).
Also extend -Wframe-larger-than limit to enable arm64 ASan build.
Change-Id: Ie572481a25fead59fc8978d2c317a33ac418516c
|
| 3d21bdf8894e780d349c481e5c9e29fe1556051c |
|
22-Apr-2015 |
Mathieu Chartier <mathieuc@google.com> |
Move mirror::ArtMethod to native
Optimizing + quick tests are passing, devices boot.
TODO: Test and fix bugs in mips64.
Saves 16 bytes per most ArtMethod, 7.5MB reduction in system PSS.
Some of the savings are from removal of virtual methods and direct
methods object arrays.
Bug: 19264997
(cherry picked from commit e401d146407d61eeb99f8d6176b2ac13c4df1e33)
Change-Id: I622469a0cfa0e7082a2119f3d6a9491eb61e3f3d
Fix some ArtMethod related bugs
Added root visiting for runtime methods, not currently required
since the GcRoots in these methods are null.
Added missing GetInterfaceMethodIfProxy in GetMethodLine, fixes
--trace run-tests 005, 044.
Fixed optimizing compiler bug where we used a normal stack location
instead of double on ARM64, this fixes the debuggable tests.
TODO: Fix JDWP tests.
Bug: 19264997
Change-Id: I7c55f69c61d1b45351fd0dc7185ffe5efad82bd3
ART: Fix casts for 64-bit pointers on 32-bit compiler.
Bug: 19264997
Change-Id: Ief45cdd4bae5a43fc8bfdfa7cf744e2c57529457
Fix JDWP tests after ArtMethod change
Fixes Throwable::GetStackDepth for exception event detection after
internal stack trace representation change.
Adds missing ArtMethod::GetInterfaceMethodIfProxy call in case of
proxy method.
Bug: 19264997
Change-Id: I363e293796848c3ec491c963813f62d868da44d2
Fix accidental IMT and root marking regression
Was always using the conflict trampoline. Also included fix for
regression in GC time caused by extra roots. Most of the regression
was IMT.
Fixed bug in DumpGcPerformanceInfo where we would get SIGABRT due to
detached thread.
EvaluateAndApplyChanges:
From ~2500 -> ~1980
GC time: 8.2s -> 7.2s due to 1s less of MarkConcurrentRoots
Bug: 19264997
Change-Id: I4333e80a8268c2ed1284f87f25b9f113d4f2c7e0
Fix bogus image test assert
Previously we were comparing the size of the non moving space to
size of the image file.
Now we properly compare the size of the image space against the size
of the image file.
Bug: 19264997
Change-Id: I7359f1f73ae3df60c5147245935a24431c04808a
[MIPS64] Fix art_quick_invoke_stub argument offsets.
ArtMethod reference's size got bigger, so we need to move other args
and leave enough space for ArtMethod* and 'this' pointer.
This fixes mips64 boot.
Bug: 19264997
Change-Id: I47198d5f39a4caab30b3b77479d5eedaad5006ab
|
| e401d146407d61eeb99f8d6176b2ac13c4df1e33 |
|
22-Apr-2015 |
Mathieu Chartier <mathieuc@google.com> |
Move mirror::ArtMethod to native
Optimizing + quick tests are passing, devices boot.
TODO: Test and fix bugs in mips64.
Saves 16 bytes per most ArtMethod, 7.5MB reduction in system PSS.
Some of the savings are from removal of virtual methods and direct
methods object arrays.
Bug: 19264997
Change-Id: I622469a0cfa0e7082a2119f3d6a9491eb61e3f3d
|
| 41b175aba41c9365a1c53b8a1afbd17129c87c14 |
|
19-May-2015 |
Vladimir Marko <vmarko@google.com> |
ART: Clean up arm64 kNumberOfXRegisters usage.
Avoid undefined behavior for arm64 stemming from 1u << 32 in
loops with upper bound kNumberOfXRegisters.
Create iterators for enumerating bits in an integer either
from high to low or from low to high and use them for
<arch>Context::FillCalleeSaves() on all architectures.
Refactor runtime/utils.{h,cc} by moving all bit-fiddling
functions to runtime/base/bit_utils.{h,cc} (together with
the new bit iterators) and all time-related functions to
runtime/base/time_utils.{h,cc}. Improve test coverage and
fix some corner cases for the bit-fiddling functions.
Bug: 13925192
(cherry picked from commit 80afd02024d20e60b197d3adfbb43cc303cf29e0)
Change-Id: I905257a21de90b5860ebe1e39563758f721eab82
|
| 80afd02024d20e60b197d3adfbb43cc303cf29e0 |
|
19-May-2015 |
Vladimir Marko <vmarko@google.com> |
ART: Clean up arm64 kNumberOfXRegisters usage.
Avoid undefined behavior for arm64 stemming from 1u << 32 in
loops with upper bound kNumberOfXRegisters.
Create iterators for enumerating bits in an integer either
from high to low or from low to high and use them for
<arch>Context::FillCalleeSaves() on all architectures.
Refactor runtime/utils.{h,cc} by moving all bit-fiddling
functions to runtime/base/bit_utils.{h,cc} (together with
the new bit iterators) and all time-related functions to
runtime/base/time_utils.{h,cc}. Improve test coverage and
fix some corner cases for the bit-fiddling functions.
Bug: 13925192
Change-Id: I704884dab15b41ecf7a1c47d397ab1c3fc7ee0f7
|
| 3481ba2c4e4f3aa80d8c6d50a9f85dacb56b508b |
|
13-Apr-2015 |
Vladimir Marko <vmarko@google.com> |
ART: Clean up includes.
Reduce dependencies to improve incremental build times.
Break up circular dependency involving class_linker-inl.h.
Change-Id: I4be742c5c2b5cd9855beea86630fd68aab76b0db
|
| c785344b87221f5e4e6473e5b762e4e61fe65dcf |
|
27-Mar-2015 |
Mathieu Chartier <mathieuc@google.com> |
Move ArtField to native
Add linear alloc. Moved ArtField to be native object. Changed image
writer to put ArtFields after the mirror section.
Savings:
2MB on low ram devices
4MB on normal devices
Total PSS measurements before (normal N5, 95s after shell start):
Image size: 7729152 bytes
23112 kB: .NonMoving
23212 kB: .NonMoving
22868 kB: .NonMoving
23072 kB: .NonMoving
22836 kB: .NonMoving
19618 kB: .Zygote
19850 kB: .Zygote
19623 kB: .Zygote
19924 kB: .Zygote
19612 kB: .Zygote
Avg: 42745.4 kB
After:
Image size: 7462912 bytes
17440 kB: .NonMoving
16776 kB: .NonMoving
16804 kB: .NonMoving
17812 kB: .NonMoving
16820 kB: .NonMoving
18788 kB: .Zygote
18856 kB: .Zygote
19064 kB: .Zygote
18841 kB: .Zygote
18629 kB: .Zygote
3499 kB: .LinearAlloc
3408 kB: .LinearAlloc
3424 kB: .LinearAlloc
3600 kB: .LinearAlloc
3436 kB: .LinearAlloc
Avg: 39439.4 kB
No reflection performance changes.
Bug: 19264997
Bug: 17643507
Change-Id: I10c73a37913332080aeb978c7c94713bdfe4fe1c
|
| c6201fa2ec66a218c4d0320fbcddd2fbb65cfa4a |
|
12-Mar-2015 |
Mathieu Chartier <mathieuc@google.com> |
Add way to select arena type at runtime
We now use MemMap for JIT, and malloc for everything else. This
should help fix the allegedly regressed compile times.
Change-Id: I6a6552738933f9d7ee3bd23f45e310818b19b70d
|
| 9b34b244ecddd8a35c922ed87bc3df0ca4db0282 |
|
09-Mar-2015 |
Mathieu Chartier <mathieuc@google.com> |
Trim arenas for JIT
Moved arena pool into the runtime. Added arena trimming to arena
pool. When called, this madvises the used memory.
Changed the JIT compiler to trim arenas after compilation. Changed
the arena mmap name to dalvik-LinearAlloc.
Native PSS before:
80353 kB: Native
80775 kB: Native
78116 kB: Native
After:
73357 kB: Native
70181 kB: Native
70306 kB: Native
Bug: 17950037
Bug: 17643507
Bug: 19264997
Change-Id: I63e7a898fd6e909c2c677fa57b5917a7b1398930
|
| b666f4805c8ae707ea6fd7f6c7f375e0b000dba8 |
|
18-Feb-2015 |
Mathieu Chartier <mathieuc@google.com> |
Move arenas into runtime
Moved arena pool into the runtime.
Motivation:
Allow GC to use arena allocators, recycle arena pool for linear alloc.
Bug: 19264997
Change-Id: I8ddbb6d55ee923a980b28fb656c758c5d7697c2f
|