| d9e4d73b20d68aa387f5837e1535b6fc26b2859a |
|
05-Feb-2018 |
Gupta Kumar, Sanjiv <sanjiv.kumar.gupta@intel.com> |
Fix iCache misses for GetKind on x86,x86_64
GetKind() takes about 2.6% of total compilation time on x86_64.
The primary reason is that the target call GetKindInternal() is often
beyond the page boundary causing frequent i-cache misses.
This patch removes the virtual call to GetKindInternal () and instead
keeps the InstructionKind into each constructed instruction.
Since we have about 121 instructions in total as of now,
it takes about 7 extra bits in each instruction.
dex2oat runs about 12% faster with --compiler-filter=everything on an
APK of 25MB.
Test: Tested the patch by running host art tests.
Rebased.
Change-Id: Ia7bbcd67180151e4565507164a718acbb6284885
Signed-off-by: Gupta Kumar, Sanjiv <sanjiv.kumar.gupta@intel.com>
|
| bff7a52e2c6c9e988c3ed1f12a2da0fa5fd37cfb |
|
25-Jan-2018 |
Nicolas Geoffray <ngeoffray@google.com> |
Revert "Compiler changes for bitstring based type checks."
Bug: 64692057
Bug: 71853552
Bug: 26687569
This reverts commit eb0ebed72432b3c6b8c7b38f8937d7ba736f4567.
Change-Id: I7daeaa077960ba41b2ed42bc47f17501621be4be
|
| eb0ebed72432b3c6b8c7b38f8937d7ba736f4567 |
|
10-Jan-2018 |
Vladimir Marko <vmarko@google.com> |
Compiler changes for bitstring based type checks.
We guard the use of this feature with a compile-time flag,
set to true in this CL.
Boot image size for aosp_taimen-userdebug in AOSP master:
- before:
arm boot*.oat: 63604740
arm64 boot*.oat: 74237864
- after:
arm boot*.oat: 63531172 (-72KiB, -0.1%)
arm64 boot*.oat: 74135008 (-100KiB, -0.1%)
The new TypeCheckBenchmark yields the following changes
using the little cores of taimen fixed at 1.4016GHz:
32-bit 64-bit
timeCheckCastLevel1ToLevel1 11.48->15.80 11.47->15.78
timeCheckCastLevel2ToLevel1 15.08->15.79 15.08->15.79
timeCheckCastLevel3ToLevel1 19.01->15.82 17.94->15.81
timeCheckCastLevel9ToLevel1 42.55->15.79 42.63->15.81
timeCheckCastLevel9ToLevel2 39.70->14.36 39.70->14.35
timeInstanceOfLevel1ToLevel1 13.74->17.93 13.76->17.95
timeInstanceOfLevel2ToLevel1 17.02->17.95 16.99->17.93
timeInstanceOfLevel3ToLevel1 24.03->17.95 24.45->17.95
timeInstanceOfLevel9ToLevel1 47.13->17.95 47.14->18.00
timeInstanceOfLevel9ToLevel2 44.19->16.52 44.27->16.51
This suggests that the bitstring typecheck should not be
used for exact type checks which would be equivalent to the
"Level1ToLevel1" benchmark. Whether the implementation is
a beneficial replacement for the kClassHierarchyCheck and
kAbstractClassCheck on average depends on how many levels
from the target class (or Object for a negative result) is
a typical object's class.
Test: m test-art-host-gtest
Test: testrunner.py --host --optimizing --jit
Test: testrunner.py --host -t 670-bitstring-type-check
Test: Pixel 2 XL boots.
Test: testrunner.py --target --optimizing --jit
Test: testrunner.py --target -t 670-bitstring-type-check
Bug: 64692057
Bug: 71853552
Bug: 26687569
Change-Id: I538d7e036b5a8ae2cc3fe77662a5903d74854562
|
| 7f4aff6705f46f411874b5ca8c4856b8ed5bfb13 |
|
21-Jun-2017 |
Artem Serov <artem.serov@linaro.org> |
ART: Implement SuperblockCloner.
SuperblockCloner provides a feature of cloning subgraphs in a
smart, high level way without fine grain manipulation with IR;
data flow and graph properties are resolved/adjusted automatically.
The clone transformation is defined by specifying a set of basic
blocks to copy and a set of rules how to treat edges, remap their
successors. By using this approach such optimizations as Branch
Target Expansion, Loop Peeling, Loop Unrolling can be implemented.
Test: superblock_cloner_test.cc.
Change-Id: Ibeede38195376ca35f44ba9015491e50b3a5b87e
|
| 04366f382239f4bcf1f9c67bb1ff6975607cd8e4 |
|
14-Dec-2017 |
Nicolas Geoffray <ngeoffray@google.com> |
Revert "ART: Try to statically evaluate some conditions."
CL has an unwanted 10-15% compile-time impact.
This reverts commit 1de1e11ac90db9fad8916ac43d43714ccb8d978f.
Change-Id: I76b45aa95bbd24dd025d2ee6cf37d77fe17b8497
|
| 09faaea17b75269805b4857ed3c9cd04c7273959 |
|
07-Dec-2017 |
Artem Serov <artem.serov@linaro.org> |
ART: Fix single-preheader transformation.
Original implementation of "Make sure the loop has only one
pre-header" had an assumption that the header had no phi
functions since loops with multiple preheaders now only may exist
during graph building before ssa construction; all of the
optimizations preserve the single-preheader invariant. This code is
used by DCE; DCE was called multiple times but after graph building
preheader transformation was never executed. However if someone
introduces a optimization which might not keep the invariant
(e.g. loop peeling) the data flow adjustments must be performed.
Test: loop_optimization_test.cc
Test: test-art-target, test-art-host
Change-Id: I88bb0aad2dd5241addef7fe9cda474a6868bf532
|
| 1de1e11ac90db9fad8916ac43d43714ccb8d978f |
|
20-Jul-2017 |
Artem Serov <artem.serov@linaro.org> |
ART: Try to statically evaluate some conditions.
If a condition 'cond' is evaluated in an HIf instruction then in
the successors of the this HIF_BLOCK we statically know the value
of the condition (TRUE in TRUE_SUCC, FALSE in FALSE_SUCC). Using
that we could replace another evaluation (use) EVAL of the same
'cond' with TRUE value (FALSE value) if every path from the
ENTRY_BLOCK to EVAL_BLOCK contains the edge HIF_BLOCK->TRUE_SUCC
(HIF_BLOCK->FALSE_SUCC).
if (cond) {
...
if (cond) {
...
}
...
int a = cond ? 5 : 105;
...
}
The patch is a prerequisite step for "Loop peeling to eliminate
invariant exits" however it brings some value on its own with
a tiny code size reduction in boot-framework.oat (-8Kb).
Test: 458-checker-instruct-simplification
Test: test-art-target, test-art-host.
Change-Id: Ifbe45097dc2b5f098176fa1a1d023ea90b76d396
|
| 28e012a4af2d710e5e5f824709ffd6432e4f549f |
|
07-Dec-2017 |
Vladimir Marko <vmarko@google.com> |
Determine HLoadClass/String load kind early.
This helps save memory by avoiding the allocation of
HEnvironment and related objects for AOT references to
boot image strings and classes (kBootImage* load kinds)
and also for JIT references (kJitTableAddress).
Compiling aosp_taimen-userdebug boot image, the most memory
hungry method BatteryStats.dumpLocked() needs
- before:
Used 55105384 bytes of arena memory...
...
UseListNode 10009704
Environment 423248
EnvVRegs 20676560
...
- after:
Used 50559176 bytes of arena memory...
...
UseListNode 8568936
Environment 365680
EnvVRegs 17628704
...
Test: m test-art-host-gtest
Test: testrunner.py --host --optimizing --jit
Bug: 34053922
Change-Id: I68e73a438e6ac8e8908e6fccf53bbeea8a64a077
|
| dbb9aef046301940d0b253c918a5c78b277330ba |
|
23-Nov-2017 |
Nicolas Geoffray <ngeoffray@google.com> |
Log at places we fail to compile.
Useful when diagnosing some compiler issues / limitations.
Test: test.py
Change-Id: I8759d0e78b0682b300ddcadfe02793432cab2036
|
| 75bb2f3c85330b2aeba9e0a4a25f7eb059bcd754 |
|
30-Nov-2017 |
Mingyao Yang <mingyao@google.com> |
Type conversion elimination of constants
A better way of eliminating type conversion for constants.
Test: run-test on host. 711-checker-type-conversion.
Change-Id: I457bc091542a5ac4cc4e77cadb012ee7cb040ce8
|
| 46721ef33e8f5cd405c291d72e3f259e3085fb5f |
|
05-Oct-2017 |
Mingyao Yang <mingyao@google.com> |
Don't merge values for exit block in LSE.
This enables some additional optimizations since exit block doesn't
really merge values.
Test: run-test on host.
Change-Id: I21ed7e0e43a3bc5d9ed2dabfad8462129b904eb7
|
| cced8ba4245a061ab047a0a6882468d75d619dd9 |
|
19-Jul-2017 |
Artem Serov <artem.serov@linaro.org> |
ART: Introduce individual HInstruction cloning.
Introduce API for HInstruction cloning, support it for a few
instructions. add a gtest.
Test: cloner_test.cc, test-art-target, test-art-host.
Change-Id: I8b6299be5d04a26390d9ef13a20ce82ee5ae4afe
|
| 69d310e0317e2fce97bf8c9c133c5c2c0332e61d |
|
09-Oct-2017 |
Vladimir Marko <vmarko@google.com> |
Use ScopedArenaAllocator for building HGraph.
Memory needed to compile the two most expensive methods for
aosp_angler-userdebug boot image:
BatteryStats.dumpCheckinLocked() : 21.1MiB -> 20.2MiB
BatteryStats.dumpLocked(): 42.0MiB -> 40.3MiB
This is because all the memory previously used by the graph
builder is reused by later passes.
And finish the "arena"->"allocator" renaming; make renamed
allocator pointers that are members of classes const when
appropriate (and make a few more members around them const).
Test: m test-art-host-gtest
Test: testrunner.py --host
Bug: 64312607
Change-Id: Ia50aafc80c05941ae5b96984ba4f31ed4c78255e
|
| ca6fff898afcb62491458ae8bcd428bfb3043da1 |
|
03-Oct-2017 |
Vladimir Marko <vmarko@google.com> |
ART: Use ScopedArenaAllocator for pass-local data.
Passes using local ArenaAllocator were hiding their memory
usage from the allocation counting, making it difficult to
track down where memory was used. Using ScopedArenaAllocator
reveals the memory usage.
This changes the HGraph constructor which requires a lot of
changes in tests. Refactor these tests to limit the amount
of work needed the next time we change that constructor.
Test: m test-art-host-gtest
Test: testrunner.py --host
Test: Build with kArenaAllocatorCountAllocations = true.
Bug: 64312607
Change-Id: I34939e4086b500d6e827ff3ef2211d1a421ac91a
|
| d5d2f2ce627aa0f6920d7ae05197abd1a396e035 |
|
26-Sep-2017 |
Vladimir Marko <vmarko@google.com> |
ART: Introduce Uint8 compiler data type.
This CL adds all the necessary codegen for the Uint8 type
but does not add code transformations that use that code.
Vectorization codegens are modified to use Uint8 as the
packed type when appropriate. The side effects are now
disconnected from the instruction's type after the graph has
been built to allow changing HArrayGet/H*FieldGet/HVecLoad
to use a type different from the underlying field or array.
Note: HArrayGet for String.charAt() is modified to have
no side effects whatsoever; Strings are immutable.
Test: m test-art-host-gtest
Test: testrunner.py --host --optimizing --jit
Test: testrunner.py --target --optimizing on Nexus 6P
Test: Nexus 6P boots.
Bug: 23964345
Change-Id: If2dfffedcfb1f50db24570a1e9bd517b3f17bfd0
|
| 0ebe0d83138bba1996e9c8007969b5381d972b32 |
|
21-Sep-2017 |
Vladimir Marko <vmarko@google.com> |
ART: Introduce compiler data type.
Replace most uses of the runtime's Primitive in compiler
with a new class DataType. This prepares for introducing
new types, such as Uint8, that the runtime does not need
to know about.
Test: m test-art-host-gtest
Test: testrunner.py --host
Bug: 23964345
Change-Id: Iec2ad82454eec678fffcd8279a9746b90feb9b0c
|
| 94ec2db21332ee1dcdbbf254b99a9a999a304fe0 |
|
06-Sep-2017 |
Vladimir Marko <vmarko@google.com> |
Use mmapped boot image class table for PIC app HLoadClass.
Implement new HLoadClass load kind for boot image classes
referenced by PIC-compiled apps (i.e. prebuilts) that uses
PC-relative load from a boot image ClassTable mmapped into
the apps .bss. This reduces the size of the PIC prebuilts
that reference boot image classes compared to the kBssEntry
as we can completely avoid the slow path and stack map
unless we need to do the class initialization check.
Prebuilt services.odex for aosp_angler-userdebug (arm64):
- before: 20312800
- after: 19775352 (-525KiB)
Test: m test-art-host-gtest
Test: testrunner.py --host
Test: testrunner.py --host --pictest
Test: testrunner.py --target on Nexus 6P.
Test: testrunner.py --target --pictest on Nexus 6P.
Test: Nexus 6P boots.
Bug: 31951624
Change-Id: I13adb19a1fa7d095a72a41f09daa6101876e77a8
|
| dd018df8a00e841fe38fabe38520b7d297a885c1 |
|
09-Aug-2017 |
Igor Murashkin <iam@google.com> |
optimizing: add block-scoped constructor fence merging pass
Introduce a new "Constructor Fence Redundancy Elimination" pass.
The pass currently performs local optimization only, i.e. within instructions
in the same basic block.
All constructor fences preceding a publish (e.g. store, invoke) get
merged into one instruction.
==============
OptStat#ConstructorFenceGeneratedNew: 43825
OptStat#ConstructorFenceGeneratedFinal: 17631 <+++
OptStat#ConstructorFenceRemovedLSE: 164
OptStat#ConstructorFenceRemovedPFRA: 9391
OptStat#ConstructorFenceRemovedCFRE: 16133 <---
Removes ~91.5% of the 'final' constructor fences in RitzBenchmark:
(We do not distinguish the exact reason that a fence was created, so
it's possible some "new" fences were also removed.)
==============
Test: art/test/run-test --host --optimizing 476-checker-ctor-fence-redun-elim
Bug: 36656456
Change-Id: I8020217b448ad96ce9b7640aa312ae784690ad99
|
| 6cfbdbc359ec5414d3e49f70d28f8c0e65b98d63 |
|
25-Jul-2017 |
Vladimir Marko <vmarko@google.com> |
Use mmapped boot image intern table for PIC app HLoadString.
Implement new HLoadString load kind for boot image strings
referenced by PIC-compiled apps (i.e. prebuilts) that uses
PC-relative load from a boot image InternTable mmapped into
the apps .bss. This reduces the size of the PIC prebuilts
that reference boot image strings compared to the kBssEntry
as we can completely avoid the slow path and stack map.
We separate the InternedStrings and ClassTable sections of
the boot image (.art) file from the rest, aligning the
start of the InternedStrings section to a page boundary.
This may actually increase the size of the boot image file
by a page but it also allows mprotecting() these tables as
read-only. The ClassTable section is included in
anticipation of a similar load kind for HLoadClass.
Prebuilt services.odex for aosp_angler-userdebug (arm64):
- before: 20862776
- after: 20308512 (-541KiB)
Note that 92KiB savings could have been achieved by simply
avoiding the read barrier, similar to the HLoadClass flag
IsInBootImage(). Such flag is now unnecessary.
Test: m test-art-host-gtest
Test: testrunner.py --host
Test: testrunner.py --host --pictest
Test: testrunner.py --target on Nexus 6P.
Test: testrunner.py --target --pictest on Nexus 6P.
Test: Nexus 6P boots.
Bug: 31951624
Change-Id: I5f2bf1fc0bb36a8483244317cfdfa69e192ef6c5
|
| 6ef45677305048c2bf0600f1c4b98a11b2cfaffb |
|
08-Aug-2017 |
Igor Murashkin <iam@google.com> |
optimizing: Add statistics for # of constructor fences added/removed
Statistics are attributed as follows:
Added because:
* HNewInstances requires a HConstructorFence following it.
* HReturn requires a HConstructorFence (for final fields) preceding it.
Removed because:
* Optimized in Load-Store-Elimination.
* Optimized in Prepare-For-Register-Allocation.
Test: art/test.py
Bug: 36656456
Change-Id: Ic119441c5151a5a840fc6532b411340e2d68e5eb
|
| 16e528957869c7debb1f6758c9a364819e15ee1a |
|
14-Jul-2017 |
Mads Ager <ager@google.com> |
RFC: Generate select instruction for conditional returns.
The select generator currently only inserts select instructions
if there is a diamond shape with a phi.
This change extends the select generator to also deal with the
pattern:
if (condition) {
movable instruction 0
return value0
} else {
movable instruction 1
return value1
}
which it turns into:
moveable instruction 0
moveable instruction 1
return select (value0, value1, condition)
Test: 592-checker-regression-bool-input
Change-Id: Iac50fb181dc2c9b7619f28977298662bc09fc0e1
|
| c73ee37b76494253862ee17933acfe2b88de1a01 |
|
31-Jul-2017 |
Artem Serov <artem.serov@linaro.org> |
ART: Fix loop header's predecessors reordering in SimplifyLoops.
Fix the issue when after loop header's predecessors reordering in
SimplifyLoops phi inputs are not reordered correspondingly.
Test: loop_optimization_test.cc, test-art-host, test-art-target.
Change-Id: I8a251a0a953d751f9bb67da58181e47d225d90e6
|
| 21c7e6fbcabef2f22b467e1e89f4abe1aa43e459 |
|
27-Jul-2017 |
Artem Serov <artem.serov@linaro.org> |
ART: Fix SimplifyInduction for an instruction with HEnvironment.
After an instruction is removed during RemoveFromCycle its
environment isn't properly cleaned: it still has input instructions
present and registered (those instructions still hold records for
that).
Test: test-art-target, test-art-host.
Change-Id: Iea315bdf735d75fe477f43671f05b40dfecc63a8
|
| 8cf9cb386cd9286d67e879f1ee501ec00d72a4e1 |
|
19-Jul-2017 |
Andreas Gampe <agampe@google.com> |
ART: Include cleanup
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
|
| a4b58ed1ef8dceafbffcfa88bf2c11144e302d18 |
|
23-Jun-2017 |
George Burgess IV <gbiv@google.com> |
Fix static analyzer warning
nodes.cc: warning: Access to field 'next_' results in a dereference of a
null pointer (loaded from field 'last_instruction_')
This was split from
https://android-review.googlesource.com/#/c/416101/. Please see the
discussion nodes.cc (patch set 1) for why this warning is triggered.
Bug: 32619234
Test: test-art-host. Rebuilt ART with the analyzer to verify that these
issues are gone.
Change-Id: Id5da00ceee0667441233153a7971d238ea8c8650
|
| 0eb882bfc5d260e8014c26adfda11602065aa5d8 |
|
15-May-2017 |
Vladimir Marko <vmarko@google.com> |
Use ArtMethod* .bss entries for HInvokeStaticOrDirect.
Test: m test-art-host-gtest
Test: testrunner.py --host
Test: testrunner.py --target
Test: Nexus 6P boots.
Test: Build aosp_mips64-userdebug.
Bug: 30627598
Change-Id: I0e54fdd2e91e983d475b7a04d40815ba89ae3d4f
|
| e7197bf7d58c705a048e13e241d7ca320502cd40 |
|
02-Jun-2017 |
Vladimir Marko <vmarko@google.com> |
Replace invoke kind kDexCacheViaMethod with kRuntimeCall.
In preparation for replacing the dex cache method array
with a hash-based array, get rid of one unnecessary use.
This method load kind is currently used only on mips for
irreducible loops and OSR, so this should have no impact
on x86/x86-64/arm/arm64.
Test: m test-art-host-gtest
Test: testrunner.py --host
Test: Repeat the above tests with manually changing
kDexCachePcRelative to kRuntimeCall in sharpening.cc.
(Ignore failures in 552-checker-sharpening.)
Bug: 30627598
Change-Id: Ifce42645f2dcc350bbb88c2f4642e88fc5f98152
|
| 847e6ce98b4b822fd94c631975763845978ebaa3 |
|
02-Jun-2017 |
Vladimir Marko <vmarko@google.com> |
Rename kDexCacheViaMethod to kRuntimeCall for HLoadClass/String.
The old name does not reflect the actual code anymore.
Test: testrunner.py --host
Change-Id: I2e13cf727bba9d901c4d3fc821bb526d38a775b8
|
| 19d7d50c0ae23f2041b1aee4aa8a5b789e3d2020 |
|
24-May-2017 |
Vladimir Marko <vmarko@google.com> |
ARM64: Fix IsAdrpPatch().
kMethodRelative patch can be an ADRP patch. Note that this
minor bug had no real consequences as the patch would later
be filtered out in NeedsErratum843419Thunk() because we
generate the instruction sequence ADRP+ADD in this case.
Also change the output for kDirectAddress from "Direct" to
"DirectAddress". This kind is currently unused in checker
tests.
Test: Rely on TreeHugger.
Bug: 30627598
Change-Id: I4e7c033aa010bbff53a554bfadd2ba0f03f69528
|
| 6597946d29be9108e2cc51223553d3db9290a3d9 |
|
19-May-2017 |
Vladimir Marko <vmarko@google.com> |
Use PC-relative pointer to boot image methods.
In preparation for adding ArtMethod entries to the .bss
section, add direct PC-relative pointers to methods so that
the number of needed .bss entries for boot image is small.
Test: m test-art-host-gtest
Test: testrunner.py --host
Test: testrunner.py --target on Nexus 6P
Test: Nexus 6P boots.
Test: Build aosp_mips64-userdebug
Bug: 30627598
Change-Id: Ia89f5f9975b741ddac2816e1570077ba4b4c020f
|
| 79d8fa7c52c1810d4618c9bd1d43994be5abb53d |
|
18-Apr-2017 |
Igor Murashkin <iam@google.com> |
optimizing: Build HConstructorFence for HNewArray/HNewInstance nodes
Also fixes:
* LSE, code_sinking to keep optimizing new-instance if it did so before
* Various tests to expect constructor fences after new-instance
Sidenote: new-instance String does not get a ConstructorFence; the
special StringFactory calls are assumed to be self-fencing.
Metric changes on go/lem:
* CodeSize -0.262% in ART-Compile (ARMv8)
* RunTime -0.747% for all (linux-armv8)
(No changes expected to x86, constructor fences are no-op).
The RunTime regression is temporary until art_quick_alloc_* entrypoints have their
DMBs removed in a follow up CL.
Test: art/test.py
Bug: 36656456
Change-Id: I6a936a6e51c623e1c6b5b22eee5c3c72bebbed35
|
| 764d454d1d51448deb81f6e8d2d7d317c7f4d1b4 |
|
16-May-2017 |
Vladimir Marko <vmarko@google.com> |
Remove LoadString/Class kind kBootImageLinkTimeAddress.
We no longer support non-PIC boot image compilation.
Also clean up some obsolete code for method patches
and make JIT correctly report itself as non-PIC.
Test: testrunner.py --host
Test: testrunner.py --target
Bug: 33192586
Change-Id: I593289c5c1b0e88b82b86a933038be97bbb15ad2
|
| 79efadfdd861584f1c47654ade975eae6c43c360 |
|
08-May-2017 |
Nicolas Geoffray <ngeoffray@google.com> |
Add runtime reasons for deopt.
Currently to help investigate. Also:
1) Log when deoptimization happens (which method and what reason)
2) Trace when deoptimization happens (to make it visible in systrace)
bug:37655083
Test: test-art-host test-art-target
(cherry picked from commit 4e92c3ce7ef354620a785553bbada554fca83a67)
Change-Id: I992398a1038ab61ea0e5106af6b6ad0a3305312e
|
| 4e92c3ce7ef354620a785553bbada554fca83a67 |
|
08-May-2017 |
Nicolas Geoffray <ngeoffray@google.com> |
Add runtime reasons for deopt.
Currently to help investigate. Also:
1) Log when deoptimization happens (which method and what reason)
2) Trace when deoptimization happens (to make it visible in systrace)
bug:37655083
Test: test-art-host test-art-target
Change-Id: I0c2d87b40db09e8e475cf97a7c784a034c585e97
|
| d01745ef88bfd25df574a885d90a1a7785db5f5b |
|
06-Apr-2017 |
Igor Murashkin <iam@google.com> |
optimizing: constructor fence redundancy elimination - remove dmb after LSE
Part one of a few upcoming CLs to optimize constructor fences.
This improves load-store-elimination; all singleton objects that are not
returned will have their associated constructor fence removed.
If the allocation is removed, so is the fence. Even if allocation is not
removed, fences can sometimes be removed.
This change is enabled by tracking the "this" object associated with the
constructor fence as an input. Fence inputs are considered weak; they do not keep
the "this" object alive; if the instructions for "this" are all deleted,
the fence can also be deleted.
Bug: 36656456
Test: art/test.py --host && art/test.py --target
Change-Id: I05659ab07e20d6e2ecd4be051b722726776f4ab1
|
| c6ea7d00ad069a2736f603daa3d8eaa9a1f8ea11 |
|
02-Feb-2017 |
Andreas Gampe <agampe@google.com> |
ART: Clean up art_method.h
Clean up the header. Fix up other headers including the -inl file,
in an effort to prune the include graph. Fix broken transitive
includes by making includes explicit. Introduce new -inl files
for method handles and reference visiting.
Test: source build/envsetup.sh && lunch aosp_angler-userdebug && mmma art
Test: source build/envsetup.sh && lunch aosp_mips64-userdebug && mmma art
Change-Id: I8f60f1160c2a702fdf3598149dae38f6fa6bc851
|
| b07d1bcf055d3eb8c6b4c45b359ad8ef30909af7 |
|
05-Apr-2017 |
Aart Bik <ajcbik@google.com> |
Ensure environment is ready when populating loop.
Rationale:
OSR requires the suspend check to already have an environment,
albeit just for testing irreducible loops. This CL fixes the
omission. Note, the error is spurious on OSR and writing a
unit or regression test for this is hard.
Test: test-art-host
Bug: 36950873
Change-Id: Ica89e18e10deb438dead79e2cc40dd00a60b529f
|
| f8f5a16ed7bad1e18179e38453e59c96a944de10 |
|
07-Feb-2017 |
Aart Bik <ajcbik@google.com> |
ART vectorizer.
Rationale:
Make SIMD great again with a retargetable and easily extendable vectorizer.
Provides a full x86/x86_64 and a proof-of-concept ARM implementation. Sample
improvement (without any perf tuning yet) for Linpack on x86 is about 20% to 50%.
Test: test-art-host, test-art-target (angler)
Bug: 34083438, 30933338
Change-Id: Ifb77a0f25f690a87cd65bf3d5e9f6be7ea71d6c1
|
| dd0fc0481cdc4e04abb7b7300e76edbd2f07f011 |
|
28-Mar-2017 |
Nicolas Geoffray <ngeoffray@google.com> |
Make data dependency around HDeoptimize correct.
We use HDeoptimize in a few places, but when it comes to data
dependency we either:
- don't have any (BCE, CHA), in which case we should make sure no
code that the deoptimzation guards moves before the HDeoptimize
- have one on the receiver (inline cache), in which case we can
update the dominated users with the HDeoptimize to get the data
dependency correct.
bug:35661819
bug:36371709
test: 644-checker-deopt
Change-Id: I30b750f97b656dede9e10e7e43ac02c8604d7b7a
(cherry picked from commit 6f8e2c9913b24f746a154dda700f609cee3095f9)
|
| 6f8e2c9913b24f746a154dda700f609cee3095f9 |
|
23-Mar-2017 |
Nicolas Geoffray <ngeoffray@google.com> |
Make data dependency around HDeoptimize correct.
We use HDeoptimize in a few places, but when it comes to data
dependency we either:
- don't have any (BCE, CHA), in which case we should make sure no
code that the deoptimzation guards moves before the HDeoptimize
- have one on the receiver (inline cache), in which case we can
update the dominated users with the HDeoptimize to get the data
dependency correct.
bug:35661819
bug:36371709
test: 644-checker-deopt
Change-Id: I4820c6710b06939e7f5a59606971693e995fb958
|
| b13c65bb46544821a84ff2106d0710d77b0fb463 |
|
22-Mar-2017 |
Aart Bik <ajcbik@google.com> |
Saves full XMM state along suspend check's slow path.
Rationale:
Break-out CL of ART Vectorizer. We need to save 128-bit
of data (default ABI of ART runtime only saves 64-bit)
Note that this is *only* done for xmm registers that
are live, so overhead is not too big.
Bug: 34083438
Test: test-art-host
Change-Id: Ic89988b0acb0c104634271d0c6c3e29b6596d59b
|
| 1eede6ae9b08d305d0c1123284ff958373916474 |
|
02-Mar-2017 |
Nicolas Geoffray <ngeoffray@google.com> |
Don't inline methods that throw in graph with irreducible loops.
Re-computing the loop information is not supported in graphs
with irreducible loops, as it is not deterministic, and the
loop header of a loop could change. That would lead to having the
suspend check in the wrong block.
Test: test-art-host
Test: 641-irreducible-inline
bug: 35757766
Change-Id: I6a435885461fbeca035e4f5d94f055fc3262adca
|
| 69d75ffac23fe1e655b7e81f0454c2841280dc1f |
|
07-Feb-2017 |
Mingyao Yang <mingyao@google.com> |
Skip loop optimization if there is no loop in the graph.
LinearizeGraph() does quite some allocations.
Also add some comments on the possible false positives of
some flags.
Test: m test-art-host
Change-Id: I80ef89a2dc031d601e7621d0b22060cd8c17fae3
|
| fdb7d638c17dab47984e1d325d04796bb426d9b3 |
|
08-Feb-2017 |
Nicolas Geoffray <ngeoffray@google.com> |
Inline methods that throw.
Forked from https://android-review.googlesource.com/214292
test: 637-checker-throw-inline
bug: 30933338
Change-Id: I184be82dfab0710af3f3796e9e486c7817fa9c60
|
| 83c8e27a292e6e002fb3b3def75cf6d8653378e8 |
|
31-Jan-2017 |
Nicolas Geoffray <ngeoffray@google.com> |
Code refactoring around sharpening HLoadClass.
Even if the class is not accessible through the dex cache, we
can access it by other means (eg boot class, jit table). So rewrite
static field access instruction builder to not bail out if a class
cannot be accessed through the dex cache.
bug:34966607
test: test-art-host test-art-target
Change-Id: I88e4e09951a002b480eb8f271726b56f981291bd
|
| 22aa54bf8469689c7c6c33f15ff4df2ffba8fa15 |
|
18-Oct-2016 |
Alexandre Rames <alexandre.rames@linaro.org> |
AArch64: Add HInstruction scheduling support.
This commit adds a new `HInstructionScheduling` pass that performs
basic scheduling on the `HGraph`.
Currently, scheduling is performed at the block level, so no
`HInstruction` ever leaves its block in this pass.
The scheduling process iterates through blocks in the graph. For
blocks that we can and want to schedule:
1) Build a dependency graph for instructions. It includes data
dependencies (inputs/uses), but also environment dependencies and
side-effect dependencies.
2) Schedule the dependency graph. This is a topological sort of the
dependency graph, using heuristics to decide what node to schedule
first when there are multiple candidates. Currently the heuristics
only consider instruction latencies and schedule first the
instructions that are on the critical path.
Test: m test-art-host
Test: m test-art-target
Change-Id: Iec103177d4f059666d7c9626e5770531fbc5ccdc
|
| 1ea9efcbd22127b75865f9a7c2949e20f5553744 |
|
16-Jan-2017 |
Nicolas Geoffray <ngeoffray@google.com> |
Acquire the mutator lock before comparing classes/strings.
Scratch my initial thought we woudldn't need it because the
handlescope is visited during the pause: as the compiler thread
is in state native, the GC can concurrently update the handlescope,
leading to false negatives when doing class/string equality.
bug:34240874
test: test-art-host gcstress
Change-Id: Icda0722fb49300a7de57e1c5d1efaa9e8dbda83f
|
| 5247c08fb186a5a2ac02226827cf6b994f41a681 |
|
13-Jan-2017 |
Nicolas Geoffray <ngeoffray@google.com> |
Put the resolved class in HLoadClass.
To avoid repeated lookups in sharpening/rtp/inlining.
Test: test-art-host test-art-target
Change-Id: I08d0da36a4bb061cdaa490ea2af3a3217a875bbe
|
| 5d37c152f21a0807459c6f53bc25e2d84f56d259 |
|
12-Jan-2017 |
Nicolas Geoffray <ngeoffray@google.com> |
Put inlined ArtMethod pointer in stack maps.
Currently done for JIT. Can be extended for AOT and inlined boot
image methods.
Also refactor the lookup of a inlined method at runtime to not
rely on the dex cache, but look at the class loader tables.
bug: 30933338
test: test-art-host, test-art-target
Change-Id: I58bd4d763b82ab8ca3023742835ac388671d1794
|
| 6bec91c7d4670905cd67440991ec76fd54d0f000 |
|
09-Jan-2017 |
Vladimir Marko <vmarko@google.com> |
Store resolved types for AOT code in .bss.
Test: m test-art-host
Test: m test-art-target on Nexus 9.
Test: Nexus 9 boots.
Test: Build aosp_mips64-eng.
Bug: 30627598
Bug: 34193123
Change-Id: I8ec60a98eb488cb46ae3ea56341f5709dad4f623
|
| 48886c2ee655a16224870fee52dc8721a52babcf |
|
06-Jan-2017 |
Vladimir Marko <vmarko@google.com> |
Remove HLoadClass::LoadKind::kDexCachePcRelative.
Test: m test-art-host
Test: m test-art-target-run-test-552-checker-sharpening
Bug: 30627598
Change-Id: Ic809b0f3a8ed0bd4dc7ab67aa64866f9cdff9bdb
|
| 6b69e0acb0e4c506ce2587e362c38e36e41e34ab |
|
11-Jan-2017 |
Aart Bik <ajcbik@google.com> |
Complete unrolling of loops with small body and trip count one.
Rationale:
Avoids the unnecessary loop control overhead, suspend check,
and exposes more opportunities for constant folding in the
resulting loop body. Fully unrolls loop in execute() of
the Dhrystone benchmark (3% to 8% improvements).
Test: test-art-host
Change-Id: If30f38caea9e9f87a929df041dfb7ed1c227aba3
|
| f0acfe7a812a332122011832074142718c278dae |
|
09-Jan-2017 |
Nicolas Geoffray <ngeoffray@google.com> |
Keep resolved String in HLoadString.
For the following reasons:
- Avoids needing to do a lookup again in CodeGenerator::EmitJitRoots.
- Fixes races where we the string was GC'ed before CodeGenerator::EmitJitRoots.
- Makes it possible to do GVN on the same string but defined in different
dex files.
Test: test-art-host, test-art-target
Change-Id: If2b5d3079f7555427b1b96ab04546b3373fcf921
|
| c1a42cf3873be202c8c0ca3c4e67500b470ab075 |
|
18-Dec-2016 |
Nicolas Geoffray <ngeoffray@google.com> |
Remove soon to be obsolete call kinds for direct calls.
And remove CompilerDriver::GetCodeAndMethodForDirectCall in
preparation of removing non-PIC prebuild and non-PIC on-device
boot image compilation.
Test: test-art-host test-art-target
bug:33192586
Change-Id: Ic48e3e8b9d7605dd0e66f31d458a182198ba9578
|
| b0b051ad6c9fab511346882650d5d689f805a980 |
|
17-Nov-2016 |
Mingyao Yang <mingyao@google.com> |
CHA guard optimization (elimination/hoisting).
Test: manual by checking the dump-cfg output.
Change-Id: I254e168b9a85d2d3d23e02eea7e129c1bc9ab920
|
| a9dbe8333d4df5447157fe575a805003172af047 |
|
15-Dec-2016 |
Mingyao Yang <mingyao@google.com> |
Add HVariableInputSizeInstruction.
Make HPhi and HInvoke subclasses of the new instruction.
Test: m test-art-host-run-test
Change-Id: I303c725876f1f4407b98702d92370be25193fc53
|
| 9b1583e7d799a3bb3c0036abb8a0b9fcbfad360a |
|
13-Dec-2016 |
Nicolas Geoffray <ngeoffray@google.com> |
Support GVN for HLoadClass::LoadKind::kJitTableAddress.
Fixes performance regressions seen in eg Dhrystone.
Also add comment on why a class may not be found when sharpening.
Test: manual Dhrystone run, performance recovers
Test: ART_TEST_JIT=true test-art-host-run-test-jit
Change-Id: I8e879f1c390f83e8bc930f343beb7b4a41c2f190
|
| 22384aeab988df7fa5ccdc48a668589c5f602c39 |
|
12-Dec-2016 |
Nicolas Geoffray <ngeoffray@google.com> |
Revert "Revert "Add kJitTableAddress for HLoadClass.""
This reverts commit d2d5262c8370309e1f2a009f00aafc24f1cf00a0.
Change-Id: I6149d5c7d5df0b0fc5cb646a802a2eea8d01ac08
|
| d2d5262c8370309e1f2a009f00aafc24f1cf00a0 |
|
12-Dec-2016 |
Nicolas Geoffray <ngeoffray@google.com> |
Revert "Add kJitTableAddress for HLoadClass."
One test failure after merge.
This reverts commit 5b12f7973636bfea29da3956a9baa7a6bbe2b666.
Change-Id: I120c49e53274471fc1c82a10d52e99c83f5f85cc
|
| 5b12f7973636bfea29da3956a9baa7a6bbe2b666 |
|
09-Dec-2016 |
Nicolas Geoffray <ngeoffray@google.com> |
Add kJitTableAddress for HLoadClass.
This new kind loads classes from the root table associated with
JIT compiled code.
Also remove kDexCacheAddress, which is replaced by kJitTableAddress.
test: ART_TEST_JIT=true test-art-host-jit test-art-target-jit
Change-Id: Ia23029688d1a60c178bf2ffa7463927c5d5de4d0
|
| be44dcf6b2338782c276483c6c79a4ab17ca98a9 |
|
30-Nov-2016 |
Mingyao Yang <mingyao@google.com> |
Add LoadString kind of kJitTableAddress for dump-cfg.
Test: manual
Change-Id: Ifcae7b26f666930766635d8b3ed7a495494cddf7
|
| c757decb04d8535fd806b9bce1c2fe5e52c228dc |
|
04-Nov-2016 |
David Sehr <sehr@google.com> |
Do not inline loops without exit edges
Fixes an issue with LinearOrder after inlining a function containing a
loop that has no exit edge. The failure is due to incorrect loop
information being computed for blocks that are not on a path to the
inlined function's return. They should not be considered part of the
caller's enclosing loop, but are today.
Bug: 32547653
Test: run-test --host 478-checker-inline-noreturn
Change-Id: I9694a1cb861430051c801d07f7ce29752332cba5
|
| 661b69bd7b19f90b0c0cba7f354e61f7ff9cc583 |
|
09-Nov-2016 |
Vladimir Marko <vmarko@google.com> |
Reduce arena memory usage when changing graph structure.
Use swap() to reuse previously allocated memory. Use
range insertion instead of individual push_back()s.
Test: m test-art-host.
Change-Id: I7174f8cc66937f34b452ff7268a8806ca94e7573
|
| 54d6a207341ad45cb5eceed71a344073ed6d4e31 |
|
09-Nov-2016 |
Vladimir Marko <vmarko@google.com> |
Fix 552-checker-sharpening for PIC test.
And remove obsolete HLoadString::LoadKind::kDexCacheAddress.
Test: m ART_TEST_PIC_TEST=true test-art-host
Change-Id: I3e7a1a98c2c7eba5ea10954d7efcf743a807c300
|
| 2c45bc9137c29f886e69923535aff31a74d90829 |
|
25-Oct-2016 |
Vladimir Marko <vmarko@google.com> |
Remove H[Reverse]PostOrderIterator and HInsertionOrderIterator.
Use range-based loops instead, introducing helper functions
ReverseRange() for iteration in reverse order in containers.
When the contents of the underlying container change inside
the loop, use an index-based loop that better exposes the
container data modifications, compared to the old iterator
interface that's hiding it which may lead to subtle bugs.
Test: m test-art-host
Change-Id: I2a4e6c508b854c37a697fc4b1e8423a8c92c5ea0
|
| 709b070044354d9f47641f273edacaeeb0240ab7 |
|
13-Oct-2016 |
David Sehr <sehr@google.com> |
Remove mirror:: and ArtMethod deps in utils.{h,cc}
The latest chapter in the ongoing saga of attempting to dump a DEX
file without having to start a whole runtime instance. This episode
finds us removing references to ArtMethod/ArtField/mirror.
One aspect of this change that I would like to call out specfically
is that the utils versions of the "Pretty*" functions all were written
to accept nullptr as an argument. I have split these functions up as
follows:
1) an instance method, such as PrettyClass that obviously requires
this != nullptr.
2) a static method, that behaves the same way as the util method, but
calls the instance method if p != nullptr.
This requires using a full class qualifier for the static methods,
which isn't exactly beautiful. I have tried to remove as many cases
as possible where it was clear p != nullptr.
Bug: 22322814
Test: test-art-host
Change-Id: I21adee3614aa697aa580cd1b86b72d9206e1cb24
|
| e8a3c576301fd531d5f73a65fc8b84a63619d580 |
|
12-Oct-2016 |
Mathieu Chartier <mathieuc@google.com> |
Replace StackHandleScopeCollection with VariableSizedHandleScope
VariableSizedHandleScope's internal handle scopes are not pushed
directly on the thread. This means that it is safe to intermix with
other types of handle scopes.
Added test.
Test: clean-oat-host && test-art-host
Change-Id: Id2fd1155788428f394d49615d337d9134824c8f0
|
| 9620230700d4b451097c2163faa70627c9d8088a |
|
05-Oct-2016 |
Aart Bik <ajcbik@google.com> |
Refactoring of graph linearization and linear order.
Rationale:
Ownership of graph's linear order and iterators was
a bit unclear now that other phases are using it.
New approach allows phases to compute their own
order, while ssa_liveness is sole owner for graph
(since it is not mutated afterwards).
Also shortens lifetime of loop's arena.
Test: test-art-host
Change-Id: Ib7137d1203a1e0a12db49868f4117d48a4277f30
|
| aad75c6d5bfab2dc8e30fc99fafe8cd2dc8b74d8 |
|
03-Oct-2016 |
Vladimir Marko <vmarko@google.com> |
Revert "Revert "Store resolved Strings for AOT code in .bss.""
Fixed oat_test to keep dex files alive. Fixed mips build.
Rewritten the .bss GC root visiting and added write barrier
to the artResolveStringFromCode().
Test: build aosp_mips-eng
Test: m ART_DEFAULT_GC_TYPE=SS test-art-target-host-gtest-oat_test
Test: Run ART test suite on host and Nexus 9.
Bug: 20323084
Bug: 30627598
This reverts commit 5f926055cb88089d8ca27243f35a9dfd89d981f0.
Change-Id: I07fa2278d82b8eb64964c9a4b66cb93726ccda6b
|
| 281c681a0852c10f5ca99b351650b244e878aea3 |
|
26-Aug-2016 |
Aart Bik <ajcbik@google.com> |
A first implementation of a loop optimization framework.
Rationale:
We are planning to add more and more loop related optimizations
and this framework provides the basis to do so. For starters,
the framework optimizes dead induction, induction that can be
replaced with a simpler closed-form, and eliminates dead loops
completely (either pre-existing or as a result of induction
removal).
Speedup on e.g. Benchpress Loop is 73x (17.5us. -> 0.24us.)
[with the potential for more exploiting outer loop too]
Test: 618-checker-induction et al.
Change-Id: If80a809acf943539bf6726b0030dcabd50c9babc
|
| 5f926055cb88089d8ca27243f35a9dfd89d981f0 |
|
30-Sep-2016 |
Vladimir Marko <vmarko@google.com> |
Revert "Store resolved Strings for AOT code in .bss."
There are some issues with oat_test64 on host and aosp_mips-eng.
Also reverts "compiler_driver: Fix build."
Bug: 20323084
Bug: 30627598
This reverts commit 63dccbbefef3014c99c22748d18befcc7bcb3b41.
This reverts commit 04a44135ace10123f059373691594ae0f270a8a4.
Change-Id: I568ba3e58cf103987fdd63c8a21521010a9f27c4
|
| 0795f23920ee9aabf28e45c63cd592dcccf00216 |
|
28-Sep-2016 |
Mathieu Chartier <mathieuc@google.com> |
Clean up ScopedThreadStateChange to use ObjPtr
Also fixed inclusion of -inl.h files in .h files by adding
scoped_object_access-inl.h and scoped_fast_natvie_object_access-inl.h
Changed AddLocalReference / Decode to use ObjPtr.
Changed libartbenchmark to be debug to avoid linkage errors.
Bug: 31113334
Test: test-art-host
Change-Id: I4d2e160483a29d21e1e0e440585ed328b9811483
|
| 63dccbbefef3014c99c22748d18befcc7bcb3b41 |
|
21-Sep-2016 |
Vladimir Marko <vmarko@google.com> |
Store resolved Strings for AOT code in .bss.
And do some related refactorings.
Bug: 20323084
Bug: 30627598
Test: Run ART test suite including gcstress on host and Nexus 9.
Test: Run ART test suite including gcstress with baker CC on host and Nexus 9.
Test: Build aosp_mips64-eng.
Change-Id: I1b12c1570fee8e5da490b47f231050142afcbd1e
|
| 20e9db6db787e007e7032878c9899b28ec43e93f |
|
14-Sep-2016 |
Aart Bik <ajcbik@google.com> |
Make LinearizeGraph() public (and move it to nodes files)
Rationale:
It is strange that HLinearOrderIterator is defined (and visible)
in nodes.h, but clients have no way to build this order. This CL
makes the building available at the usual place.
Change-Id: Ib66f2edf6dfc8edd6b429bd4bea3ac7e37440b28
Tests: m test-art
|
| bdf7f1c3ab65ccb70f62db5ab31dba060632d458 |
|
31-Aug-2016 |
Andreas Gampe <agampe@google.com> |
ART: SHARED_REQUIRES to REQUIRES_SHARED
This coincides with the actual attribute name and upstream usage.
Preparation for deferring to libbase.
Test: m
Test: m test-art-host
Change-Id: Ia8986b5dfd926ba772bf00b0a35eaf83596d8518
|
| 75d2df275abedf90be8c8e31202389788c3b079b |
|
28-Jul-2016 |
Andreas Gampe <agampe@google.com> |
ART: Use old operator<< for MemBarrierKind
To fix assumptions in the tests.
Test: m test-art-host
Change-Id: Ie2e738524ee8f75a3c41730f6a1449dab490e535
|
| 26de38bb7f2122417388809f4ff88a7cb5c4af5e |
|
28-Jul-2016 |
Andreas Gampe <agampe@google.com> |
ART: Delete old compiler_enums.h
Holdover from the Quick days. Move the two enums that are still
used closer to the actual users (and prune no longer used cases).
Test: m test-art-host
Change-Id: I88aa49961a54635788cafac570ddc3125aa38262
|
| e90049140fdfb89080e5cc9b000b0c9be8c18bcd |
|
16-Jun-2016 |
Vladimir Marko <vmarko@google.com> |
Create a typedef for HInstruction::GetInputs() return type.
And some other cleanup after
https://android-review.googlesource.com/230742
Test: No new tests. ART test suite passed (tested on host).
Change-Id: I4743bf17544d0234c6ccb46dd0c1b9aae5c93e17
|
| dbb7f5bef10138ade0fb202da1d61f562b2df649 |
|
30-Mar-2016 |
Vladimir Marko <vmarko@google.com> |
Improve HLoadClass code generation.
For classes in the boot image, use either direct pointers
or PC-relative addresses. For other classes, use PC-relative
access to the dex cache arrays for AOT and direct address of
the type's dex cache slot for JIT.
For aosp_flounder-userdebug:
- 32-bit boot.oat: -252KiB (-0.3%)
- 64-bit boot.oat: -412KiB (-0.4%)
- 32-bit dalvik cache total: -392KiB (-0.4%)
- 64-bit dalvik-cache total: -2312KiB (-1.0%)
(contains more files than the 32-bit dalvik cache)
For aosp_flounder-userdebug forced to compile PIC:
- 32-bit boot.oat: -124KiB (-0.2%)
- 64-bit boot.oat: -420KiB (-0.5%)
- 32-bit dalvik cache total: -136KiB (-0.1%)
- 64-bit dalvik-cache total: -1136KiB (-0.5%)
(contains more files than the 32-bit dalvik cache)
Bug: 27950288
Change-Id: I4da991a4b7e53c63c92558b97923d18092acf139
|
| 206fbf51ccd2dbbcf106e480cc6c5bb96fc7befb |
|
10-Jun-2016 |
Nicolas Geoffray <ngeoffray@google.com> |
Remove too aggressive DCHECKs.
A class can move from a state greater or equal than resolved
to erroneous concurrently to the verifier or the compiler.
bug:29239283
Change-Id: I89f3fe1c1d9556c6c99b8e005b3ec02de7f01b85
(cherry picked from commit f7d994622aabcc689f62253a9253e0c67d9e787e)
|
| f7d994622aabcc689f62253a9253e0c67d9e787e |
|
10-Jun-2016 |
Nicolas Geoffray <ngeoffray@google.com> |
Remove too aggressive DCHECKs.
A class can move from a state greater or equal than resolved
to erroneous concurrently to the verifier or the compiler.
bug:29239283
Change-Id: I89f3fe1c1d9556c6c99b8e005b3ec02de7f01b85
|
| d6c205eb9b04bcfa072cd5ffdd93deef167ec340 |
|
07-Jun-2016 |
David Brazdil <dbrazdil@google.com> |
ART: Remove redundant MoveInstructionBefore method
Change-Id: If53d7011197cc6b9c1702a3d98ef11b59eb76f0c
|
| f479d7758dbe7e8740386fbf1d73e05b0277c5e3 |
|
16-May-2016 |
Anton Shamin <anton.shamin@intel.com> |
ART: ArrayGet hoisting restriction added.
Currently if we hoist ArrayGet from loop there is no guarantee
that insn will be executed at runtime. Because of that we could
face issues like crashes in generated code.
This patch introduces restriction for ArrayGet hoisting. We say
that ArrayGet execution is guaranteed at least one time if its bb
dominates all exit blocks.
Signed-off-by: Anton Shamin <anton.shamin@intel.com>
(cherry picked from commit f89381fed12faf96c45a83a989ae2fff82c05f3b)
BUG=29145171
Change-Id: Ia5664dedb1543d78a7b4038801b8372572f069f6
|
| f89381fed12faf96c45a83a989ae2fff82c05f3b |
|
16-May-2016 |
Anton Shamin <anton.shamin@intel.com> |
ART: ArrayGet hoisting restriction added.
Currently if we hoist ArrayGet from loop there is no guarantee
that insn will be executed at runtime. Because of that we could
face issues like crashes in generated code.
This patch introduces restriction for ArrayGet hoisting. We say
that ArrayGet execution is guaranteed at least one time if its bb
dominates all exit blocks.
Change-Id: I9f72c0f4c33b358341109238bea46cb5a82f490f
Signed-off-by: Anton Shamin <anton.shamin@intel.com>
|
| 372f10e5b0b34e2bb6e2b79aeba6c441e14afd1f |
|
17-May-2016 |
Vladimir Marko <vmarko@google.com> |
Refactor handling of input records.
Introduce HInstruction::GetInputRecords(), a new virtual
function that returns an ArrayRef<> to all input records.
Implement all other functions dealing with input records as
wrappers around GetInputRecords(). Rewrite functions that
previously used multiple virtual calls to deal with input
records, especially in loops, to prefetch the ArrayRef<>
only once for each instruction. Besides avoiding all the
extra calls, this also allows the compiler (clang++) to
perform additional optimizations.
This speeds up the Nexus 5 boot image compilation by ~0.5s
(4% of "Compile Dex File", 2% of dex2oat time) on AOSP ToT.
Change-Id: Id8ebe0fb9405e38d918972a11bd724146e4ca578
|
| d7c2fdc939bb7efb3e7204d62e54c6a3f7d77f9b |
|
10-May-2016 |
Nicolas Geoffray <ngeoffray@google.com> |
Fix another case of live_in at irreducible loop entry.
GVN was implicitly extending the liveness of an instruction across
an irreducible loop.
Fix this problem by clearing the value set at loop entries that contain
an irreducible loop.
bug:28252896
(cherry picked from commit 77ce6430af2709432b22344ed656edd8ec80581b)
Change-Id: Ie0121e83b2dfe47bcd184b90a69c0194d13fce54
|
| 77ce6430af2709432b22344ed656edd8ec80581b |
|
10-May-2016 |
Nicolas Geoffray <ngeoffray@google.com> |
Fix another case of live_in at irreducible loop entry.
GVN was implicitly extending the liveness of an instruction across
an irreducible loop.
Fix this problem by clearing the value set at loop entries that contain
an irreducible loop.
bug:28252896
Change-Id: I68823cb88dceb4c2b4545286ba54fd0c958a48b0
|
| a26b3c51bfd97be1100d267f20c46535913e6bb7 |
|
09-May-2016 |
Vladimir Marko <vmarko@google.com> |
Attribute arena allocations previously marked as STL.
Bug: 28603175
Bug: 28684584
(cherry picked from commit 3ea5a97d27468cec846d958c38d0d706ef7ec67e)
Change-Id: I7f1bd22e7710cca74f4b10fd13cb8fa2c3b1b318
|
| 7e589feab1b35203fbb8c431213f1d2b2a4ad530 |
|
06-May-2016 |
David Brazdil <dbrazdil@google.com> |
ART: Fix dominance for irreducible loops
Computation of dominance was broken in the presence of irreducible
loops because the algorithm assumed that back edges are always
dominated by their respective headers and a fix-point iteration is
therefore unnecessary.
This is not true for irreducible loops, forcing us to revisit their
loop headers and all dependent blocks. This patch adds a fix-point
iteration if a back edge not dominated by its header is found.
Bug: 28611485
Change-Id: If84044e49d5b9c682949648033d2861628d7fe05
(cherry picked from commit 3f4a522cc39f5c651e7c718196e989bc81d8c6ef)
|
| bf12e4d4209ac4e8fb98b4fd5193208adc7fe3ff |
|
05-Apr-2016 |
Vladimir Marko <vmarko@google.com> |
Optimizing: LoadString may not have any side effects.
LoadString does not have any side effects if the string is
known to be in the dex cache or it's a boot image string
referenced directly, as specified by the string load kind.
We need to clear the side effects for these cases to avoid
a DCHECK() failure when such LoadString instruction ends up
between a ClinitCheck and an instruction to which we want to
merge that ClinitCheck. This may happen as a consequence of
inlining, LICM and DCE as shown by a regression test.
Bug: 27929914
(cherry picked from commit ace7a000a433ce4ecf94f30adea39c01a76fa936)
Change-Id: Iaf9c63b6e58aae1e246b43ca52eea0b47a6ad565
|
| 3ea5a97d27468cec846d958c38d0d706ef7ec67e |
|
09-May-2016 |
Vladimir Marko <vmarko@google.com> |
Attribute arena allocations previously marked as STL.
Bug: 28603175
Change-Id: I488e39b23afb86f3ff5a2df16d2df59eb3adaa0f
|
| e4a97f926882ffb12707655a755a1c8f96620f91 |
|
05-May-2016 |
David Brazdil <dbrazdil@google.com> |
Stop populating irreducible loop at header
Recent CL modified the (previously exponential) algorithm for
populating irreducible loops so as to not visit blocks multiple times.
However, the new condition for an early exit did not take into account
that it is not necessary to visit predecessors of the loop header.
The algorithm remains correct but does unnecessary work.
Bug: 27856014
Change-Id: If72f8be82e79838f0dd9678082a3bbaabb51e43b
(cherry picked from commit 5a620590e5cf6d6817693edffd661371555de88b)
|
| 3f4a522cc39f5c651e7c718196e989bc81d8c6ef |
|
06-May-2016 |
David Brazdil <dbrazdil@google.com> |
ART: Fix dominance for irreducible loops
Computation of dominance was broken in the presence of irreducible
loops because the algorithm assumed that back edges are always
dominated by their respective headers and a fix-point iteration is
therefore unnecessary.
This is not true for irreducible loops, forcing us to revisit their
loop headers and all dependent blocks. This patch adds a fix-point
iteration if a back edge not dominated by its header is found.
Bug: 28611485
Change-Id: If84044e49d5b9c682949648033d2861628d7fe05
|
| 5a620590e5cf6d6817693edffd661371555de88b |
|
05-May-2016 |
David Brazdil <dbrazdil@google.com> |
Stop populating irreducible loop at header
Recent CL modified the (previously exponential) algorithm for
populating irreducible loops so as to not visit blocks multiple times.
However, the new condition for an early exit did not take into account
that it is not necessary to visit predecessors of the loop header.
The algorithm remains correct but does unnecessary work.
Bug: 27856014
Change-Id: If72f8be82e79838f0dd9678082a3bbaabb51e43b
|
| d1d7c40c8004303d1131ebb1956fd0ade54f8404 |
|
25-Apr-2016 |
Aart Bik <ajcbik@google.com> |
Test component type for errors too.
With regression test.
Rationale:
Moved erroneous check in convenience method, so we
put all the same logic in one place. When testing
for erroneous T[], check both the array type
as well at the component type T for errors
(it is possible T[] is not marked erroneous
even though T is eventually).
BUG=28358598
(cherry picked from commit f417ff44d1eb111854d7a213f106912b3dd9e3d4)
Change-Id: Ieba66aa4b55d8e7ebddf200239c7e4095dfd4678
|
| f417ff44d1eb111854d7a213f106912b3dd9e3d4 |
|
25-Apr-2016 |
Aart Bik <ajcbik@google.com> |
Test component type for errors too.
With regression test.
Rationale:
Moved erroneous check in convenience method, so we
put all the same logic in one place. When testing
for erroneous T[], check both the array type
as well at the component type T for errors
(it is possible T[] is not marked erroneous
even though T is eventually).
BUG=28358598
Change-Id: I11339a976dc83e0493a99e6bb97f3a058ca3f796
|
| 9ef26544e6d9c438b150dfda39be61a7b4fd8b20 |
|
20-Apr-2016 |
Vladimir Marko <vmarko@google.com> |
Fix HInstruction::ReplaceInput(), allow no-op.
Allow HInstruction::ReplaceInput() to be called with
a `replacement` being the same as the old input and
do nothing in that case.
This is a follow-up to
https://android-review.googlesource.com/216923
where I erroneously assumed that it never happens.
Also adhere to the standard C++ std::forward_list<>
semantics in the single-element overload of
`IntrusiveForwardList<>::splice_after()`.
Bug: 28173563
(cherry picked from commit c6b5627c25ff5653e97ccff8c5ccf6ac967b6f83)
Change-Id: I66b7d4b48f629284d0bcb1d31100519e02a872e5
|
| c6b5627c25ff5653e97ccff8c5ccf6ac967b6f83 |
|
20-Apr-2016 |
Vladimir Marko <vmarko@google.com> |
Fix HInstruction::ReplaceInput(), allow no-op.
Allow HInstruction::ReplaceInput() to be called with
a `replacement` being the same as the old input and
do nothing in that case.
This is a follow-up to
https://android-review.googlesource.com/216923
where I erroneously assumed that it never happens.
Also adhere to the standard C++ std::forward_list<>
semantics in the single-element overload of
`IntrusiveForwardList<>::splice_after()`.
Bug: 28173563
Change-Id: I5cea14c212b1083f90ffe6b5b53324ad663d57d8
|
| fa7f58935491fad5cb8e5af9ee38d1a4f73e872d |
|
19-Apr-2016 |
Vladimir Marko <vmarko@google.com> |
Reuse HUseListNode<>s when replacing instruction or input.
Compiling the Nexus 5 boot image with the 64-bit dex2oat
on host this CL reduces the memory used for compiling the
most hungry method, BatteryStats.dumpLocked(), by ~5.6MiB:
Before:
MEM: used: 44393040, allocated: 45361248, lost: 968208
Number of arenas allocated: 319,
Number of allocations: 815492, avg size: 54
...
UseListNode 10308480
...
After:
MEM: used: 38554536, allocated: 39463008, lost: 908472
Number of arenas allocated: 274,
Number of allocations: 572221, avg size: 67
...
UseListNode 4469976
...
With 32-bit dex2oat, the UseListNode would be 2/3 of the
values for 64-bit dex2oat (both before and after).
Bug: 28173563
Bug: 27856014
(cherry picked from commit 3c19d3e029a9fcc123d2c6fd1e5e13867d2cfe1f)
Change-Id: Iddd42d7545e4f97a13da63590b33711a5bbdd43b
|
| d59f3b1b7f5c1ab9f0731ff9dc60611e8d9a6ede |
|
29-Mar-2016 |
Vladimir Marko <vmarko@google.com> |
Use iterators "before" the use node in HUserRecord<>.
Create a new template class IntrusiveForwardList<> that
mimicks std::forward_list<> except that all allocations
are handled externally. This is essentially the same as
boost::intrusive::slist<> but since we're not using Boost
we have to reinvent the wheel.
Use the new container to replace the HUseList and use the
iterators to "before" use nodes in HUserRecord<> to avoid
the extra pointer to the previous node which was used
exclusively for removing nodes from the list. This reduces
the size of the HUseListNode by 25%, 32B to 24B in 64-bit
compiler, 16B to 12B in 32-bit compiler. This translates
directly to overall memory savings for the 64-bit compiler
but due to rounding up of the arena allocations to 8B, we
do not get any improvement in the 32-bit compiler.
Compiling the Nexus 5 boot image with the 64-bit dex2oat
on host this CL reduces the memory used for compiling the
most hungry method, BatteryStats.dumpLocked(), by ~3.3MiB:
Before:
MEM: used: 47829200, allocated: 48769120, lost: 939920
Number of arenas allocated: 345,
Number of allocations: 815492, avg size: 58
...
UseListNode 13744640
...
After:
MEM: used: 44393040, allocated: 45361248, lost: 968208
Number of arenas allocated: 319,
Number of allocations: 815492, avg size: 54
...
UseListNode 10308480
...
Note that while we do not ship the 64-bit dex2oat to the
device, the JIT compilation for 64-bit processes is using
the 64-bit libart-compiler.
Bug: 28173563
Bug: 27856014
(cherry picked from commit 46817b876ab00d6b78905b80ed12b4344c522b6c)
Change-Id: Ifb2d7b357064b003244e92c0d601d81a05e56a7b
|
| 7de2439d357710aaf8bfe02b8cf9c196d3c77705 |
|
06-Apr-2016 |
Aart Bik <ajcbik@google.com> |
Avoid constructing types with errors.
BUG=27626735
Rationale:
Do not construct classes with a link error. Without this,
the error type thought it was Object (mirror's method
IsObjectClass() returns true if there is no superclass).
(cherry picked from commit 8b3f9b246d5bdbf67faeb2b872b75b8d72777bc0)
(also contains follow-up commit 31244b4cde9156632a08103a8bf1cbff4cbae3cc)
Change-Id: I4443779dda47c320115975c1c71b22e118bd8252
|
| 3c19d3e029a9fcc123d2c6fd1e5e13867d2cfe1f |
|
19-Apr-2016 |
Vladimir Marko <vmarko@google.com> |
Reuse HUseListNode<>s when replacing instruction or input.
Compiling the Nexus 5 boot image with the 64-bit dex2oat
on host this CL reduces the memory used for compiling the
most hungry method, BatteryStats.dumpLocked(), by ~5.6MiB:
Before:
MEM: used: 44393040, allocated: 45361248, lost: 968208
Number of arenas allocated: 319,
Number of allocations: 815492, avg size: 54
...
UseListNode 10308480
...
After:
MEM: used: 38554536, allocated: 39463008, lost: 908472
Number of arenas allocated: 274,
Number of allocations: 572221, avg size: 67
...
UseListNode 4469976
...
With 32-bit dex2oat, the UseListNode would be 2/3 of the
values for 64-bit dex2oat (both before and after).
Bug: 28173563
Change-Id: Ia4fabe03568f0e0dbf2cdf2b031863602aea3530
|
| 46817b876ab00d6b78905b80ed12b4344c522b6c |
|
29-Mar-2016 |
Vladimir Marko <vmarko@google.com> |
Use iterators "before" the use node in HUserRecord<>.
Create a new template class IntrusiveForwardList<> that
mimicks std::forward_list<> except that all allocations
are handled externally. This is essentially the same as
boost::intrusive::slist<> but since we're not using Boost
we have to reinvent the wheel.
Use the new container to replace the HUseList and use the
iterators to "before" use nodes in HUserRecord<> to avoid
the extra pointer to the previous node which was used
exclusively for removing nodes from the list. This reduces
the size of the HUseListNode by 25%, 32B to 24B in 64-bit
compiler, 16B to 12B in 32-bit compiler. This translates
directly to overall memory savings for the 64-bit compiler
but due to rounding up of the arena allocations to 8B, we
do not get any improvement in the 32-bit compiler.
Compiling the Nexus 5 boot image with the 64-bit dex2oat
on host this CL reduces the memory used for compiling the
most hungry method, BatteryStats.dumpLocked(), by ~3.3MiB:
Before:
MEM: used: 47829200, allocated: 48769120, lost: 939920
Number of arenas allocated: 345,
Number of allocations: 815492, avg size: 58
...
UseListNode 13744640
...
After:
MEM: used: 44393040, allocated: 45361248, lost: 968208
Number of arenas allocated: 319,
Number of allocations: 815492, avg size: 54
...
UseListNode 10308480
...
Note that while we do not ship the 64-bit dex2oat to the
device, the JIT compilation for 64-bit processes is using
the 64-bit libart-compiler.
Bug: 28173563
Change-Id: I985eabd4816f845372d8aaa825a1489cf9569208
|
| 3563c44464ca55b2106373b35110e5ecaae38abf |
|
18-Apr-2016 |
Vladimir Marko <vmarko@google.com> |
Fix inlining loops in OSR mode.
When compiling a method in OSR mode and the method does not
contain a loop (arguably, a very odd case) but we inline
another method with a loop and then the final DCE re-runs
the loop identification, the inlined loop would previously
be marked as irreducible. However, the SSA liveness analysis
expects irreducible loop to have extra loop Phis which were
already eliminated from the loop before the inner graph was
inlined to the outer graph, so we would fail a DCHECK().
We fix this by not marking inlined loops as irreducible when
compiling in OSR mode.
Bug: 28210356
(cherry picked from commit fd66c50d64c38e40bafde83b4872e27bbff7546d)
Change-Id: I149273b766d1c713c571baad6033c5f70e6dd960
|
| fd66c50d64c38e40bafde83b4872e27bbff7546d |
|
18-Apr-2016 |
Vladimir Marko <vmarko@google.com> |
Fix inlining loops in OSR mode.
When compiling a method in OSR mode and the method does not
contain a loop (arguably, a very odd case) but we inline
another method with a loop and then the final DCE re-runs
the loop identification, the inlined loop would previously
be marked as irreducible. However, the SSA liveness analysis
expects irreducible loop to have extra loop Phis which were
already eliminated from the loop before the inner graph was
inlined to the outer graph, so we would fail a DCHECK().
We fix this by not marking inlined loops as irreducible when
compiling in OSR mode.
Bug: 28210356
Change-Id: If10057ed883333c62a878ed2ae3fe01bb5280e33
|
| 8b3f9b246d5bdbf67faeb2b872b75b8d72777bc0 |
|
06-Apr-2016 |
Aart Bik <ajcbik@google.com> |
Avoid constructing types with errors.
BUG=27626735
Rationale:
Do not construct classes with a link error. Without this,
the error type thought it was Object (mirror's method
IsObjectClass() returns true if there is no superclass).
Change-Id: I55ca8cc8cfc042210edf748aab10da4c6e345980
|
| c2e8af9659db7e456b26febb1b971900057ad427 |
|
05-Apr-2016 |
David Brazdil <dbrazdil@google.com> |
ART: Speed up HGraph::PopulateIrreducibleRecursive
Populating an irreducible loop can potentially traverse all possible
paths through the HGraph, leading to an exponential algorithm.
This patch adds a bit vector of nodes whose membership in the loop
has been decided and need not be revisited again.
Bug: 27856014
Change-Id: I3696f08c846e6f40e5de44cb771811bac7e3e08a
|
| dee58d6bb6d567fcd0c4f39d8d690c3acaf0e432 |
|
07-Apr-2016 |
David Brazdil <dbrazdil@google.com> |
Revert "Revert "Refactor HGraphBuilder and SsaBuilder to remove HLocals""
This patch merges the instruction-building phases from HGraphBuilder
and SsaBuilder into a single HInstructionBuilder class. As a result,
it is not necessary to generate HLocal, HLoadLocal and HStoreLocal
instructions any more, as the builder produces SSA form directly.
Saves 5-15% of arena-allocated memory (see bug for more data):
GMS 20.46MB => 19.26MB (-5.86%)
Maps 24.12MB => 21.47MB (-10.98%)
YouTube 28.60MB => 26.01MB (-9.05%)
This CL fixed an issue with parsing quickened instructions.
Bug: 27894376
Bug: 27998571
Bug: 27995065
Change-Id: I20dbe1bf2d0fe296377478db98cb86cba695e694
|
| ace7a000a433ce4ecf94f30adea39c01a76fa936 |
|
05-Apr-2016 |
Vladimir Marko <vmarko@google.com> |
Optimizing: LoadString may not have any side effects.
LoadString does not have any side effects if the string is
known to be in the dex cache or it's a boot image string
referenced directly, as specified by the string load kind.
We need to clear the side effects for these cases to avoid
a DCHECK() failure when such LoadString instruction ends up
between a ClinitCheck and an instruction to which we want to
merge that ClinitCheck. This may happen as a consequence of
inlining, LICM and DCE as shown by a regression test.
Bug: 27929914
Change-Id: I7b3bddf7d8c79ce1828a4a751f1270cf2e3d61f0
|
| 60328910cad396589474f8513391ba733d19390b |
|
04-Apr-2016 |
David Brazdil <dbrazdil@google.com> |
Revert "Refactor HGraphBuilder and SsaBuilder to remove HLocals"
Bug: 27995065
This reverts commit e3ff7b293be2a6791fe9d135d660c0cffe4bd73f.
Change-Id: I5363c7ce18f47fd422c15eed5423a345a57249d8
|
| e3ff7b293be2a6791fe9d135d660c0cffe4bd73f |
|
02-Mar-2016 |
David Brazdil <dbrazdil@google.com> |
Refactor HGraphBuilder and SsaBuilder to remove HLocals
This patch merges the instruction-building phases from HGraphBuilder
and SsaBuilder into a single HInstructionBuilder class. As a result,
it is not necessary to generate HLocal, HLoadLocal and HStoreLocal
instructions any more, as the builder produces SSA form directly.
Saves 5-15% of arena-allocated memory (see bug for more data):
GMS 20.46MB => 19.26MB (-5.86%)
Maps 24.12MB => 21.47MB (-10.98%)
YouTube 28.60MB => 26.01MB (-9.05%)
Bug: 27894376
Change-Id: Iefe28d40600c169c5d306fd2c77034ae19476d90
|
| 86ea7eeabe30c98bbe1651a51d03cb89776724e7 |
|
16-Feb-2016 |
David Brazdil <dbrazdil@google.com> |
Build dominator tree before generating HInstructions
Second CL in the series of merging HGraphBuilder and SsaBuilder. This
patch refactors the builders so that dominator tree can be built
before any HInstructions are generated. This puts the SsaBuilder
removal of HLoadLocals/HStoreLocals straight after HGraphBuilder's
HInstruction generation phase. Next CL will therefore be able to
merge them.
This patch also adds util classes for iterating bytecode and switch
tables which allowed to simplify the code.
Bug: 27894376
Change-Id: Ic425d298b2e6e7980481ed697230b1a0b7904526
|
| cac5a7e871f1f346b317894359ad06fa7bd67fba |
|
22-Feb-2016 |
Vladimir Marko <vmarko@google.com> |
Optimizing: Improve const-string code generation.
For strings in the boot image, use either direct pointers
or pc-relative addresses. For other strings, use PC-relative
access to the dex cache arrays for AOT and direct address of
the string's dex cache slot for JIT.
For aosp_flounder-userdebug:
- 32-bit boot.oat: -692KiB (-0.9%)
- 64-bit boot.oat: -948KiB (-1.1%)
- 32-bit dalvik cache total: -900KiB (-0.9%)
- 64-bit dalvik cache total: -3672KiB (-1.5%)
(contains more files than the 32-bit dalvik cache)
For aosp_flounder-userdebug forced to compile PIC:
- 32-bit boot.oat: -380KiB (-0.5%)
- 64-bit boot.oat: -928KiB (-1.0%)
- 32-bit dalvik cache total: -468KiB (-0.4%)
- 64-bit dalvik cache total: -1928KiB (-0.8%)
(contains more files than the 32-bit dalvik cache)
Bug: 26884697
Change-Id: Iec7266ce67e6fedc107be78fab2e742a8dab2696
|
| 9eeebf646b118269b3f5bd6fc6c1b0a58c26c6d4 |
|
24-Mar-2016 |
David Brazdil <dbrazdil@google.com> |
ART: Fix order of operations in HBasicBlock::DisconnectAndDelete
The method would remove predecessors before successors. As a result,
instructions used by dead loop phis would see dangling uses, causing
a DCHECK to fail.
Steps were reordered to remove dependencies in post order.
Bug: 27683071
Change-Id: I8e0e976443fb410908321a065276f1340b757c41
|
| c9b21f83c2eab8bdd16241992193c3049dc68e43 |
|
23-Mar-2016 |
Roland Levillain <rpl@google.com> |
Fix some typos in art/compiler/optimizing/nodes.cc.
Change-Id: I11be5a9b73da207c9eb497bcaffc49d614c1ca89
|
| 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
|
| 1a65388f1d86bb232c2e44fecb44cebe13105d2e |
|
18-Mar-2016 |
Roland Levillain <rpl@google.com> |
Clean up art::HConstant predicates.
- Make the difference between arithmetic zero and zero-bit
pattern non ambiguous.
- Introduce Boolean predicates in art::HIntConstant for when
they are used as Booleans.
- Introduce aritmetic positive and negative zero predicates
for floating-point constants.
Bug: 27639313
Change-Id: Ia04ecc6f6aa7450136028c5362ed429760c883bd
|
| 18401b748a3180f52e42547ede22d1b184fe8c43 |
|
11-Mar-2016 |
Nicolas Geoffray <ngeoffray@google.com> |
Fix invariant in reference type propagation.
Also some cleanups.
Change-Id: I7f0ec7d06b4bab10dbfa230c757447d311658f93
|
| 7ba9966b76bbf818513018fa1da72c89330fe384 |
|
02-Mar-2016 |
Serguei Katkov <serguei.i.katkov@intel.com> |
ART: cleanup exit_block_ in graph if exit block is removed
If we remove the exit block from the graph (for example method
contains infinite loop) we should also clean the field exit_block_
in graph. At least inliner expects it.
Change-Id: Icda668da2233cdd6cd673635a1949f5ed34cf270
Signed-off-by: Serguei Katkov <serguei.i.katkov@intel.com>
|
| 3f52306b259caed1c654c4b3fd5b594d5ec8d46c |
|
29-Feb-2016 |
David Brazdil <dbrazdil@google.com> |
ART: Fix overlapping instruction IDs in inliner
Inliner creates the inner graph so that it generates instruction IDs
higher than the outer graph. This was broken because the inliner
would create instructions in the outer graph before the inner graph
is inlined.
The bug cannot be triggered because the offending instruction would
share the same ID as the first inner HLocal, which is removed before
the inner graph is inlined. The added DCHECKs reveal the hidden problem
and make it safe for HLocals to be removed in the future.
Change-Id: I486eb0f3987e20c50cbec0fb06332229e07fbae9
|
| d65925108fc99607f3b447d8505fe6713acda55c |
|
29-Feb-2016 |
Nicolas Geoffray <ngeoffray@google.com> |
Bug fix for polymorphic inlining.
The code used to wrongly propagate try/catch information to
new blocks. Since it has the same logic as Hraph::InlineInto,
extract the code that updates loop and try/catch information
to blocks to a shared method.
bug:27330865
bug:27372101
bug:27360329
(cherry picked from commit a1d8ddfaf09545f99bc326dff97ab604d4574eb6)
Change-Id: Ice0373ec0a1c24d78121634a377f6f502e814cfb
|
| a1d8ddfaf09545f99bc326dff97ab604d4574eb6 |
|
29-Feb-2016 |
Nicolas Geoffray <ngeoffray@google.com> |
Bug fix for polymorphic inlining.
The code used to wrongly propagate try/catch information to
new blocks. Since it has the same logic as Hraph::InlineInto,
extract the code that updates loop and try/catch information
to blocks to a shared method.
bug:27330865
bug:27372101
bug:27360329
Change-Id: I4386f724d8d412bde5bcc04fda6955bc3bacf5a9
|
| a1de9188a05afdecca8cd04ecc4fefbac8b9880f |
|
25-Feb-2016 |
Vladimir Marko <vmarko@google.com> |
Optimizing: Reduce memory usage of HInstructions.
Pack narrow fields and flags into a single 32-bit field.
Change-Id: Ib2f7abf987caee0339018d21f0d498f8db63542d
|
| e53bd8160ad2892f33849108d3b1099992a311fd |
|
24-Feb-2016 |
Roland Levillain <rpl@google.com> |
Remove unreachable code paths in constant folding.
Change-Id: I7ffb361711c87f6b1b98d172d2cfdf9b2ba65607
|
| 916cc1d504f10a24f43b384e035fdecbe6a74b4c |
|
18-Feb-2016 |
Nicolas Geoffray <ngeoffray@google.com> |
Implement polymorphic inlining.
For example, before:
HInvokeVirtual
After:
If (receiver == Foo) {
// inlined code.
} else if (receiver == Bar) {
// inlined code
} else {
// HInvokeVirtual or HDeoptimize(receiver != Baz)
}
Change-Id: I5ce305aef8f39f8294bf2b2bcfe60e0dddcfdbec
|
| 31dd3d60491148d345c1edae1ccd090a1b67dd2b |
|
16-Feb-2016 |
Roland Levillain <rpl@google.com> |
Extend constant folding to float and double operations.
Change-Id: I2837064b2ceea587bc171fc520507f13355292c6
|
| 55bd749991f9a0a73f612696e1a93e739380546b |
|
16-Feb-2016 |
Nicolas Geoffray <ngeoffray@google.com> |
Refactor the inliner.
In preparation for more polymorphic inlining, refactor the inliner
a bit.
Change-Id: Ie3fd6c1ef205f1089989c67a527e6f57ff3c8b5d
|
| badd826664896d4a9628a5a89b78016894aa414b |
|
02-Feb-2016 |
David Brazdil <dbrazdil@google.com> |
ART: Run SsaBuilder from HGraphBuilder
First step towards merging the two passes, which will later result in
HGraphBuilder directly producing SSA form. This CL mostly just updates
tests broken by not being able to inspect the pre-SSA form.
Using HLocals outside the HGraphBuilder is now deprecated.
Bug: 27150508
Change-Id: I00fb6050580f409dcc5aa5b5aa3a536d6e8d759e
|
| 86503785cd6414b8692e5c83cadaa2972b6a099b |
|
11-Feb-2016 |
Roland Levillain <rpl@google.com> |
Fix x86-64 Baker's read barrier fast path for CheckCast.
Use an art::x86_64::Label instead of an
art::x86_64::NearLabel as end label when emitting code for a
HCheckCast instruction, as the range of the latter may
sometimes be too short when Baker's read barriers are
enabled.
Bug: 12687968
Change-Id: Ia9742dce65be7d4fb104688f3c4717b65df1fb54
|
| b331febbab8e916680faba722cc84b66b84218a3 |
|
05-Feb-2016 |
Nicolas Geoffray <ngeoffray@google.com> |
Revert "Revert "Implement on-stack replacement for arm/arm64/x86/x86_64.""
This reverts commit bd89a5c556324062b7d841843b039392e84cfaf4.
Change-Id: I08d190431520baa7fcec8fbdb444519f25ac8d44
|
| bd89a5c556324062b7d841843b039392e84cfaf4 |
|
05-Feb-2016 |
David Brazdil <dbrazdil@google.com> |
Revert "Implement on-stack replacement for arm/arm64/x86/x86_64."
DCHECK whether loop headers are covered fails.
This reverts commit 891bc286963892ed96134ca1adb7822737af9710.
Change-Id: I0f9a90630b014b16d20ba1dfba31ce63e6648021
|
| 891bc286963892ed96134ca1adb7822737af9710 |
|
29-Jan-2016 |
Nicolas Geoffray <ngeoffray@google.com> |
Implement on-stack replacement for arm/arm64/x86/x86_64.
High-level overview:
- osr_method_threshold is used to know when to compile a method
in osr mode (-> treat all loops as irreducible).
- branch instructions in the compiler query whether they can
jump to an osr method.
- An osr entry point is found through the stack maps: if a stack
map is duplicated in the CodeInfo, it is an osr entry point.
Change-Id: Ifb39338cd281e2c7eccce67f4e18d46428be71e4
|
| 3f1a8be7c9511afbc1ea0ce2e76a018269382336 |
|
03-Feb-2016 |
Aart Bik <ajcbik@google.com> |
Fixed bug on premature DCHECK.
With fail-before/pass-after test
bug=26947011
Rationale:
During BCE, the phi structure is under construction,
to be fixed by InsertPhiNodes() and carefully checked
with the SSA checker. So utilities should not overly
DCHECK on SSA consistency during the modifications.
Change-Id: Ia9df9ee5aac0c1dd2c3e3a447c730246d5e48bbb
|
| 74eb1b264691c4eb399d0858015a7fc13c476ac6 |
|
14-Dec-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Implement HSelect
This patch adds a new HIR instruction to Optimizing. HSelect returns
one of two inputs based on the outcome of a condition.
This is only initial implementation which:
- defines the new instruction,
- repurposes BooleanSimplifier to emit it,
- extends InstructionSimplifier to statically resolve it,
- updates existing code and tests accordingly.
Code generators currently emit fallback if/then/else code and will be
updated in follow-up CLs to use platform-specific conditional moves
when possible.
Change-Id: Ib61b17146487ebe6b55350c2b589f0b971dcaaee
|
| 09e8d5ffe52c738c6a74984b1cbc7ad4bc8f5e2c |
|
23-Jan-2016 |
Aart Bik <ajcbik@google.com> |
Some minor simplifications in code and tests.
Background:
This is actually a resubmit of an earlier cl that was
reverted because was test was less robust against
inlining changes (it assumed a virtual call would
never be inlined).
original cl: If8ada79dfd70bea991c11d2b18661b951b6c4cd4
revert cl: I739aaaccd0509d02a62ef01e797a6d45bfe941df
Change-Id: I952680d60ff488874907f066bfdf156a45b409ba
|
| 788f2f05c3e5b0e5bda247b00e34f0094585546f |
|
22-Jan-2016 |
Nicolas Geoffray <ngeoffray@google.com> |
Revert "Revert "Inline methods with loops.""
Bug: 26689526
This reverts commit 451ad8d1be9a1949ea3c3e3a713a9e76198a8b2d.
Change-Id: If484fe4c0744254dd7568fd5006e574d621a1855
|
| 69fd1b56021ac62c17e188bd0d4dd22fc911558e |
|
22-Jan-2016 |
Nicolas Geoffray <ngeoffray@google.com> |
Revert "Some minor simplifications in code and tests."
Fails 530-checker-loops on arm
This reverts commit bf03fcd10a3ffa15468d335f26697b0473e45b36.
Change-Id: I739aaaccd0509d02a62ef01e797a6d45bfe941df
|
| bf03fcd10a3ffa15468d335f26697b0473e45b36 |
|
04-Jan-2016 |
Aart Bik <ajcbik@google.com> |
Some minor simplifications in code and tests.
Rationale: fell through the cracks of previous "intrinsics" CL.
Change-Id: If8ada79dfd70bea991c11d2b18661b951b6c4cd4
|
| a0ee77157ee6ceb72292c20bc299c1d24fe95a39 |
|
20-Jan-2016 |
Andreas Gampe <agampe@google.com> |
Revert "Inline methods with loops."
This reverts commit 82fc9bb45dbf8ff728122fb7ab72d1eb7b2f4869.
Loop inlining exposes issues with BCE.
Bug: 26689526
(cherry picked from commit 451ad8d1be9a1949ea3c3e3a713a9e76198a8b2d)
Change-Id: Ie6f260e6a224aeb7f5ed93df378b6cefba10d35f
|
| 451ad8d1be9a1949ea3c3e3a713a9e76198a8b2d |
|
20-Jan-2016 |
Andreas Gampe <agampe@google.com> |
Revert "Inline methods with loops."
This reverts commit 82fc9bb45dbf8ff728122fb7ab72d1eb7b2f4869.
Loop inlining exposes issues with BCE.
Bug: 26689526
Change-Id: Id9983d7f9d3c5579d91e56e4699d4d939517b2dc
|
| 82fc9bb45dbf8ff728122fb7ab72d1eb7b2f4869 |
|
19-Jan-2016 |
Nicolas Geoffray <ngeoffray@google.com> |
Inline methods with loops.
Compiling Gms/Fb/Framework/Docs:
- Overall compilation-time increase: 2.2%
- Overall code size increase: 1.1%
Performance improvements:
- Richards with jit: +6%
- Takl: +11%
Change-Id: I0a6fcf2a360e5ad193cd95b5c4fe92227ac6bd96
|
| 09aa147f0891ef28a95d89e8ad61c429f82ddd5b |
|
19-Jan-2016 |
Nicolas Geoffray <ngeoffray@google.com> |
Disable DCE when there are irreducible loops.
Also ensure an instruction that requires an environment
does have one.
Change-Id: I41a8460e05ef320f872197d3be7847e7ffaa6ee8
|
| 6de1938e562b0d06e462512dd806166e754035ea |
|
08-Jan-2016 |
David Brazdil <dbrazdil@google.com> |
ART: Remove incorrect HFakeString optimization
Simplification of HFakeString assumes that it cannot be used until
String.<init> is called which is not true and causes different
behaviour between the compiler and the interpreter. This patch
removes the optimization together with the HFakeString instruction.
Instead, HNewInstance is generated and an empty String allocated
until it is replaced with the result of the StringFactory call. This
is consistent with the behaviour of the interpreter but is too
conservative. A follow-up CL will attempt to optimize out the initial
allocation when possible.
Bug: 26457745
Bug: 26486014
Change-Id: I7139e37ed00a880715bfc234896a930fde670c44
|
| 15bd22849ee6a1ffb3fb3630f686c2870bdf1bbc |
|
05-Jan-2016 |
Nicolas Geoffray <ngeoffray@google.com> |
Implement irreducible loop support in optimizing.
So we don't fallback to the interpreter in the presence of
irreducible loops.
Implications:
- A loop pre-header does not necessarily dominate a loop header.
- Non-constant redundant phis will be kept in loop headers, to
satisfy our linear scan register allocation algorithm.
- while-graph optimizations, such as gvn, licm, lse, and dce
need to know when they are dealing with irreducible loops.
Change-Id: I2cea8934ce0b40162d215353497c7f77d6c9137e
|
| a3eca2d7300f35c66cf4b696d788a8b7ba74eb99 |
|
12-Jan-2016 |
Nicolas Geoffray <ngeoffray@google.com> |
Do not leave intermediate addresses across Java calls.
bug:26472446
Change-Id: Ie4a9b5fe6f1d61a76c71eceaa2299fe55512c612
|
| c928591f5b2c544751bb3fb26dc614d3c2e67bef |
|
18-Dec-2015 |
Roland Levillain <rpl@google.com> |
ARM Baker's read barrier fast path implementation.
Introduce an ARM 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 the 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: Ie7ee85b1b4c0564148270cebdd3cbd4c3da51b3a
|
| f555258861aea7df8af9c2241ab761227fd2f66a |
|
27-Dec-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Create BoundType for CheckCast early
ReferenceTypePropagation creates a BoundType for each CheckCast and
replaces all dominated uses of the casted object with it. This does
not include Phi uses on the boundary of the dominated scope, reducing
typing precision. This patch creates the BoundType in Builder, causing
SsaBuilder to replace uses of the object automatically.
Bug: 26081304
Change-Id: I083979155cccb348071ff58cb9060a896ed7d2ac
|
| 4833f5a1990c76bc2be89504225fb13cca22bedf |
|
16-Dec-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Refactor SsaBuilder for more precise typing info
This reverts commit 68289a531484d26214e09f1eadd9833531a3bc3c.
Now uses Primitive::Is64BitType instead of Primitive::ComponentSize
because it was incorrectly optimized by GCC.
Bug: 26208284
Bug: 24252151
Bug: 24252100
Bug: 22538329
Bug: 25786318
Change-Id: Ib39f3da2b92bc5be5d76f4240a77567d82c6bebe
|
| 5d75afe333f57546786686d9bee16b52f1bbe971 |
|
14-Dec-2015 |
Aart Bik <ajcbik@google.com> |
Improved side-effects/can-throw information on intrinsics.
Rationale: improved side effect and exception analysis gives
many more opportunities for GVN/LICM/BCE.
Change-Id: I8aa9b757d77c7bd9d58271204a657c2c525195b5
|
| 5f7b58ea1adfc0639dd605b65f59198d3763f801 |
|
23-Nov-2015 |
Vladimir Marko <vmarko@google.com> |
Rewrite HInstruction::Is/As<type>().
Make Is<type>() and As<type>() non-virtual for concrete
instruction types, relying on GetKind(), and mark GetKind()
as PURE to improve optimization opportunities. This reduces
the number of relocations in libart-compiler.so's .rel.dyn
section by ~4K, or ~44%, and in .data.rel.ro by ~18K, or
~65%. The file is 96KiB smaller for Nexus 5, including 8KiB
reduction of the .text section.
Unfortunately, the g++/clang++ __attribute__((pure)) is not
strong enough to avoid duplicated virtual calls and we would
need the C++ [[pure]] attribute proposed in n3744 instead.
To work around this deficiency, we introduce an extra
non-virtual indirection for GetKind(), so that the compiler
can optimize common expressions such as
instruction->IsAdd() || instruction->IsSub()
or
instruction->IsAdd() && instruction->AsAdd()->...
which contain two virtual calls to GetKind() after inlining.
Change-Id: I83787de0671a5cb9f5b0a5f4a536cef239d5b401
|
| 68289a531484d26214e09f1eadd9833531a3bc3c |
|
16-Dec-2015 |
Alex Light <allight@google.com> |
Revert "ART: Refactor SsaBuilder for more precise typing info"
This reverts commit d9510dfc32349eeb4f2145c801f7ba1d5bccfb12.
Bug: 26208284
Bug: 24252151
Bug: 24252100
Bug: 22538329
Bug: 25786318
Change-Id: I5f491becdf076ff51d437d490405ec4e1586c010
|
| d9510dfc32349eeb4f2145c801f7ba1d5bccfb12 |
|
05-Nov-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Refactor SsaBuilder for more precise typing info
This patch refactors the SsaBuilder to do the following:
1) All phis are constructed live and marked dead if not used or proved
to be conflicting.
2) Primitive type propagation, now not a separate pass, identifies
conflicting types and marks corresponding phis dead.
3) When compiling --debuggable, DeadPhiHandling used to revive phis
which had only environmental uses but did not attempt to resolve
conflicts. This pass was removed as obsolete and is now superseded
by primitive type propagation (identifying conflicting phis) and
SsaDeadPhiEliminiation (keeping phis live if debuggable + env use).
4) Resolving conflicts requires correct primitive type information
on all instructions. This was not the case for ArrayGet instructions
which can have ambiguous types in the bytecode. To this end,
SsaBuilder now runs reference type propagation and types ArrayGets
from the type of the input array.
5) With RTP being run inside the SsaBuilder, it is not necessary to
run it as a separate optimization pass. Optimizations can now assume
that all instructions of type kPrimNot have reference type info after
SsaBuilder (with the exception of NullConstant).
6) Graph now contains a reference type to be assigned to NullConstant.
All reference type instructions therefore have RTI, as now enforced
by the SsaChecker.
Bug: 24252151
Bug: 24252100
Bug: 22538329
Bug: 25786318
Change-Id: I7a3aee1ff66c82d64b4846611c547af17e91d260
|
| 04ff4e8463ac68752638d305eeb84b457fd8289c |
|
10-Dec-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Fix bug in DCE not removing phis from catch phi uses
Due to the missing edges between throwing instructions and catch phis
DCE needs to manually remove dead instructions from catch phi users,
being overly conservative if the inputs are not in the dead blocks.
DCE used to do this for normal instructions, but it needs to do the
same for phis.
Change-Id: I7edfcb84ec6ff7303945d5d5cd436b1d1e95df2a
|
| 4b467ed97bc5886fb800209c0ee94df10163b88d |
|
20-Nov-2015 |
Mingyao Yang <mingyao@google.com> |
Simplify and rename IsLoopInvariant() test.
Simplify IsLoopInvariant() test. Also rename it to IsDefinedOutOfTheLoop()
so there is no ambiguity for example whether a instruction after the loop counts
as a loop invariant. It's up to the caller to make the interpretation.
Change-Id: I999139032b0e4d815dd1e2276f2bd428cf558686
|
| f64242a30c6e05a8e4302a64eab4bcc28297dc9e |
|
01-Dec-2015 |
Vladimir Marko <vmarko@google.com> |
Optimizing: Add checker tests for sharpening.
This is a follow-up to
https://android-review.googlesource.com/184116 .
Change-Id: Ib03c424fb673afc5ccce15d7d072b7572b47799a
|
| fb337ea53d1e6fe68b24217a9ea95e9f544ef697 |
|
25-Nov-2015 |
Vladimir Marko <vmarko@google.com> |
Move PC-relative addressing bases to a better position.
Move the platform-specific HX86ComputeBaseMethodAddress and
HArmDexCacheArraysBase to the latest dominator of their uses
outside any loop. This brings the base closer to the first
use (previously, it was in the entry block) and relieves
some pressure on the register allocator while avoiding
recalculation of the base in a loop.
Change-Id: I231aa81eb5b4de9af2d0167054d06b65eb18a636
|
| 5c0048565e78ff53fd2b3a2e446c72ea2fffe239 |
|
23-Nov-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Fix uninitialized variable
Change-Id: I906de334b3c3cb1e36eff4944457f4598b7c174f
|
| 3fc7f357170311689c4c31007a5e168ddea321d5 |
|
21-Nov-2015 |
Aart Bik <ajcbik@google.com> |
Accept synthetic phi nodes and general names for blocks.
Rationale: these changes were already approved as part of the dynamic
bce changes, but I am now sending them out separately.
Change-Id: I3564bac9f6a0b6a89466457836ff54ad09164faf
|
| d59c70627cc42878cc30b46bd29ff497b4483b22 |
|
21-Nov-2015 |
Aart Bik <ajcbik@google.com> |
Revert "Dynamic BCE (based on induction range analysis)"
This reverts commit 0b5849be045c5683d4a6b6b6c306abadba5f0fcc.
Change-Id: Id33f5da42bbdfb1aff7e2281417c8a7aa492df05
Rationale: so close :-( but bullhead-userdebug (linux) build in git_mnc-dr-dev-plus-aosp reported a breakage with a type inconsistency (long vs int in probably the codegen of dynamic bce); no time to investigate and fix this fully before my trip, so rolling back for now
|
| 0b5849be045c5683d4a6b6b6c306abadba5f0fcc |
|
19-Oct-2015 |
Aart Bik <ajcbik@google.com> |
Dynamic BCE (based on induction range analysis)
Rationale: A rewritten dynamic BCE that uses induction variable analysis
to generate the run-time tests before a loop in order to
eliminate bounds-checks from its body. This CL removes now
obsoleted induction related code inside the BCE module.
Also, the dynamic test generation is placed more strategically,
since we missed a few cases where static analysis does better.
Most significant performance improvements (after filtering noise) is about:
Linpack +20%
LU > +10%
Change-Id: I4e7b8bab0288beff6f98a14856e3536103d32742
|
| f652917de5634b30c974c81d35a72871915b352a |
|
17-Nov-2015 |
Mark Mendell <mark.p.mendell@intel.com> |
Simplify boolean condition compared to 0
CaffeineMarkRR Logic has some boolean flipping which can be helped by
some simplification.
Simplify non-FP (A COND_OP B) != 0 to A OPPOSITE_COND_OP B.
This is better than the original code, which would use a HBooleanNot
after the condition.
Also simplify non-FP (A COND_OP B) == 1 to A OPPOSITE_COND_OP B.
Move GetOppositeCondition to nodes.h/nodes.cc to share with Boolean
Simplification, renaming it to InsertOppositeCondition, as it inserts
the new HInstruction (unless it is a constant).
Change-Id: I34ded7758836e375de0d6fdba9239d2d451928d0
Signed-off-by: Mark Mendell <mark.p.mendell@intel.com>
|
| fbb184a1c6df22d9302b32b55206396c8278edcf |
|
13-Nov-2015 |
Vladimir Marko <vmarko@google.com> |
Fix ClinitCheck pruning.
Make sure we merge the ClinitCheck only with LoadClass and
HInvokeStaticOrDirect that is a part of the very same dex
instruction. This fixes incorrect stack traces from class
initializers (wrong dex pcs).
Rewrite the pruning to do all the ClinitCheck merging when
we see the ClinitCheck, instead of merging ClinitCheck into
LoadClass and then LoadClass into HInvokeStaticOrDirect.
When we later see an HInvokeStaticOrDirect with an explicit
check (i.e. not merged), we know that some other instruction
is doing the check and the invoke doesn't need to, so we
mark it as not requiring the check at all. (Previously it
would have been marked as having an implicit check.)
Remove the restriction on merging with inlined invoke static
as this is not necessary anymore. This was a workaround for
X.test():
invoke-static C.foo() [1]
C.foo():
invoke-static C.bar() [2]
After inlining and GVN we have
X.test():
LoadClass C (from [1])
ClinitCheck C (from [1], to be merged to LoadClass)
InvokeStaticOrDirect C.bar() (from [2])
and the LoadClass must not be merged into the invoke as this
would cause the resolution trampoline to see an inlined
frame from the not-yet-loaded class C during the stack walk
and try to load the class. However, we're not allowed to
load new classes at that point, so an attempt to do so leads
to an assertion failure. With this CL, LoadClass is not
merged when it comes from a different instruction, so we can
guarantee that all inlined frames seen by the stack walk in
the resolution trampoline belong to already loaded classes.
Change-Id: I2b8da8d4f295355dce17141f0fab2dace126684d
|
| 0f7dca4ca0be8d2f8776794d35edf8b51b5bc997 |
|
02-Nov-2015 |
Vladimir Marko <vmarko@google.com> |
Optimizing/X86: PC-relative dex cache array addressing.
Add PC-relative dex cache array addressing for X86 and use
it for better invoke-static/-direct dispatch. Also delay
the initialization to the PC-relative base until needed.
Change-Id: Ib8634d5edce4920cd70172fd13211809cf6948d1
|
| d26a411adee1e71b3f09dd604ab9b23018037138 |
|
10-Nov-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Refactor iteration over normal/exceptional successors
Add helper methods on HBasicBlock which return ArrayRef with the
suitable sub-array of the `successors_` list.
Change-Id: I66c83bb56f2984d7550bf77c48110af4087515a8
|
| 9e23df5c21bed53ead79e3131b67105abc8871e4 |
|
10-Nov-2015 |
Vladimir Marko <vmarko@google.com> |
Optimizing: Improve constant folding + DCE for inlining.
Run constant folding before DCE in inliner to eliminate more
code that can prevent inlining. Improve the constant folding
to evaluate Equals and NotEquals for null inputs.
Change-Id: I876ffb903ef39484370b6c8793f0f8467a977362
|
| 59a850ee63a637660599f215901058b059f3a4b4 |
|
10-Nov-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Improve comment about inlining into try/catch
Change-Id: I66a4fd3206847c8d5bb57b1678d9d3dc94331294
|
| dc0d1eb8a4b033db4acf136cd563de865542098a |
|
10-Nov-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Add clarifying comment
Change-Id: I189ec4cae0aa1a5245a79e86d1ec0592e38eac4a
|
| 8a7c0fe837bb00b02dfcfc678910d81d07fb2136 |
|
02-Nov-2015 |
David Brazdil <dbrazdil@google.com> |
Revert "Revert "ART: Update DCE to work with try/catch""
The previous CL failed because it did not update inputs of catch phis.
Since phi input indices cannot be easily mapped back to throwing
instructions, this new implementation at least removes catch phi uses
of values defined in the removed blocks to preserve graph consistency.
This reverts commit fb552d7061746f7a90fdd5002696e255e2e15c35.
Change-Id: I63d95915d1ef50e71d3bcf0cd10aaded554035b4
|
| 391d01f3e8dedf3af3727bdf5d5b76ab35d52795 |
|
06-Nov-2015 |
Vladimir Marko <vmarko@google.com> |
Optimizing: Rewrite search for common dominators.
Provide a utility class that can be used to quickly search
for common dominators of two or more blocks. Change the
algorithm to avoid memory allocations.
Change-Id: Id72c975fc42377cb7622902f87c4262ea7b3cc38
|
| db51efb3617d15f1cd9e5ff0cc2d934777014e9a |
|
06-Nov-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Fix critical edge splitting under try/catch
A critical edge would not be split if the predecessor ends with
TryBoundary. This would eventually trip liveness analysis because
a back edge block would have smaller liveness position than a nested
loop.
Another implication of this change is that an edge between a loop's
pre-header ending with TryBoundary and the header will be split,
guaranteeing that a pre-header always has just one successor.
Bug: 25493695
Bug: 25454012
Change-Id: I5a13b8bb74509b48f5d628906f7158af007f99ae
|
| b554b5a5ae3cdc66969d61be20783a8af816206e |
|
06-Nov-2015 |
Vladimir Marko <vmarko@google.com> |
Optimizing: Remove unused ArtMethod* input from HInvokeStaticOrDirect.
Change-Id: Iea99fa683440673ff517e246f35fade96600f229
|
| 9bc436160b4af99067973affb0b1008de9a2b04c |
|
05-Nov-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Fix simplification of catch blocks in the presence of dead code
Simplification of catch blocks transforms the code so that catch
blocks have only exceptional predecessors. However, it is invoked
before trivially dead code is eliminated which breaks simple
assumptions such as the fact that a catch block cannot start with
move-exception if it has non-exceptional predecessors. This patch
fixes the algorithm to work under these relaxed conditions.
Bug: 25494450
Bug: 25492628
Change-Id: Idc8d010102a4b8b9a6cd918b98d6e11d1838db0c
|
| fb552d7061746f7a90fdd5002696e255e2e15c35 |
|
02-Nov-2015 |
David Brazdil <dbrazdil@google.com> |
Revert "ART: Update DCE to work with try/catch"
This reverts commit ce52901e2c8377fc1c331ae0faf7fbcb46b9da97.
Change-Id: I6b3a1f2a3dc036030b089b0df2005ecefa66b949
|
| ce52901e2c8377fc1c331ae0faf7fbcb46b9da97 |
|
29-Oct-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Update DCE to work with try/catch
Dead block elimination was previously disabled because it needed
to be updated. With this patch, try/catch blocks can be removed
as a result of a dead if/switch branch.
Change-Id: I3261060bf24fd5fe7bb0f989247f0ef62ec5fd7b
|
| 951779839f0d35ed5336f399c8f521fd9a6b7c27 |
|
30-Oct-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Enable inlining under try/catch
This patch updates the inliner to set try/catch information
when inlining into a method with try/catch. It does not yet
allow inlining of methods with try/catch because that will
require generating catch stack maps with inline info.
Change-Id: I7d57e1454e7da537d75c5c7eda60b22f3a30fa60
|
| 771e5cc519665ce0cc76985bb4803f0dd50c3b40 |
|
29-Oct-2015 |
David Brazdil <dbrazdil@google.com> |
Revert "ART: Enable more passes under try/catch"
BCE does not set TryCatchInformation when creating new blocks.
Will be fixed with DynamicBCE CL.
This reverts commit 39fabd6bb6fcf7a712b370d3b6fd0ada83e2e5d8.
Change-Id: I76ae707ac132bb1a4a9f64f4916ffcd786ef730c
|
| 73f1f3be46652d3f6df61b4234c366ebbf81274a |
|
28-Oct-2015 |
Aart Bik <ajcbik@google.com> |
Move loop invariant utility to more general place.
Change-Id: I15ebfbf9684f0fcce9e63d078ff8dc1381fd1ca3
|
| 39fabd6bb6fcf7a712b370d3b6fd0ada83e2e5d8 |
|
26-Oct-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Enable more passes under try/catch
LICM, BCE, LSE are all safe under try/catch. Inliner and DCE
need updating and will be enabled in follow-up CLs.
Change-Id: I86db5f811257d5e765fea91666a2a2af0fb24ec3
|
| dc151b2346bb8a4fdeed0c06e54c2fca21d59b5d |
|
15-Oct-2015 |
Vladimir Marko <vmarko@google.com> |
Optimizing: Determine invoke-static/-direct dispatch early.
Determine the dispatch type of invoke-static/-direct in a
special pass right after the type inference. This allows the
inliner to pass the "needs dex cache" check and inline more.
It also allows the code generator to avoid requesting a
register location for the ArtMethod* for kDexCachePcRelative
and direct methods.
The supported dispatch check handles also situations that
the CompilerDriver currently doesn't allow. The cleanup of
the CompilerDriver and required changes to Quick will come
in a separate change.
Change-Id: I3f8e903a119949e95871d8ab0a995f4731a13a07
|
| 214bbcd1d7454197427c13cc082860619357d847 |
|
20-Oct-2015 |
Calin Juravle <calin@google.com> |
Inliner: make sure the returned value is in the outer graph.
The returned value may be a constant or a parameter value. If so, it
will be in the inlined entry_block and (before this CL) we would not
update its block or graph. This CL fixes this and makes sure that the
returned value belongs to the outer graph.
Change-Id: Ie296f0d5a320c33f39eb187df6d328371ccf6500
|
| 805b3b56c6eb542298db33e0181f135dc9fed3d9 |
|
18-Sep-2015 |
Mark Mendell <mark.p.mendell@intel.com> |
X86 jump tables for PackedSwitch
Implement X86PackedSwitch using a jump table of offsets to blocks. The
X86PackedSwitch version just adds an input to address the constant area.
Change-Id: Id2752a1ee79222493040c6fd0e59aee9a544b76a
Bug: 21119474
Signed-off-by: Mark Mendell <mark.p.mendell@intel.com>
|
| ec7802a102d49ab5c17495118d4fe0bcc7287beb |
|
01-Oct-2015 |
Vladimir Marko <vmarko@google.com> |
Add DCHECKs to ArenaVector and ScopedArenaVector.
Implement dchecked_vector<> template that DCHECK()s element
access and insert()/emplace()/erase() positions. Change the
ArenaVector<> and ScopedArenaVector<> aliases to use the new
template instead of std::vector<>. Remove DCHECK()s that
have now become unnecessary from the Optimizing compiler.
Change-Id: Ib8506bd30d223f68f52bd4476c76d9991acacadc
|
| a83a54d7f2322060f08480f8aabac5eb07268912 |
|
02-Oct-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Add support for intrinsic optimizations.
Change-Id: Ib5a4224022f9360e60c09a19ac8642270a7f3b64
|
| 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
|
| d7558daaa86decf5a38f4f9bcd82267ab6e3e17f |
|
22-Sep-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Preserve loop headers with try/catch
Algorithm for inserting HTryBoundary instructions would generate a
non-natural loop when a loop header block was covered by a TryItem.
This patch changes the approach to fix the issue.
Bug: 23895756
Change-Id: I0e1ee6cf135cea326a96c97954907d202c9793cc
|
| 1f8695ca0c0d443a3d2754637ea5c9459147af55 |
|
24-Sep-2015 |
Vladimir Marko <vmarko@google.com> |
Optimizing: Rewrite HGraph::FindBackEdges().
Replace a recursive implementation with a loop using a work
list to avoid stack overflow for 702-LargeBranchOffset in
host debug build with -O0, 512KiB thread pool worker stack.
Change-Id: Iaa91f006fa1099913aeffc9c764879bd004d56de
|
| d76d1390b04a4db9ca1f74eb4873d926643d979b |
|
23-Sep-2015 |
Vladimir Marko <vmarko@google.com> |
Optimizing: Rewrite HGraph::ComputeDominanceInformation().
Replace a recursive implementation with a loop using a work
list to avoid stack overflow for 702-LargeBranchOffset in
host debug build with -O0.
Bug: 24133462
Change-Id: I444cc85733a9212403a071ea98b9ddfb52bfc402
|
| fe57faa2e0349418dda38e77ef1c0ac29db75f4d |
|
18-Sep-2015 |
Mark Mendell <mark.p.mendell@intel.com> |
[optimizing] Add basic PackedSwitch support
Add HPackedSwitch, and generate it from the builder. Code generators
convert this to a series of compare/branch tests. Better implementation
in the code generators as a real jump table will follow as separate CLs.
Change-Id: If14736fa4d62809b6ae95280148c55682e856911
Signed-off-by: Mark Mendell <mark.p.mendell@intel.com>
|
| b7d8e8cf7063fdec1cce6ebd33e33804976bd978 |
|
17-Sep-2015 |
Vladimir Marko <vmarko@google.com> |
Optimizing: Do not use range-based loop when inserting elements.
When we iterate over the elements of a container and we may
insert new elements into that container, it's wrong to use
the range-based loop.
Bug: 24133462
Change-Id: Iee35fbcf88ed3bcd6155cbeba09bd256032a16be
|
| 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
|
| 77a48ae01bbc5b05ca009cf09e2fcb53e4c8ff23 |
|
15-Sep-2015 |
David Brazdil <dbrazdil@google.com> |
Revert "Revert "ART: Register allocation and runtime support for try/catch""
The original CL triggered b/24084144 which has been fixed
by Ib72e12a018437c404e82f7ad414554c66a4c6f8c.
This reverts commit 659562aaf133c41b8d90ec9216c07646f0f14362.
Change-Id: Id8980436172457d0fcb276349c4405f7c4110a55
|
| baf89b8f2591df93686df545bc6ccc6eb3b128cc |
|
15-Sep-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Fix bug in reference type propagation
Reference type propagation assumed that type is exact if the class is
final. This does not hold for arrays which are always final and the
component type needs to be considered.
Bug: 24084144
Change-Id: Ib72e12a018437c404e82f7ad414554c66a4c6f8c
|
| 659562aaf133c41b8d90ec9216c07646f0f14362 |
|
14-Sep-2015 |
David Brazdil <dbrazdil@google.com> |
Revert "ART: Register allocation and runtime support for try/catch"
Breaks libcore test org.apache.harmony.security.tests.java.security.KeyStorePrivateKeyEntryTest#testGetCertificateChain. Need to investigate.
This reverts commit b022fa1300e6d78639b3b910af0cf85c43df44bb.
Change-Id: Ib24d3a80064d963d273e557a93469c95f37b1f6f
|
| b022fa1300e6d78639b3b910af0cf85c43df44bb |
|
20-Aug-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Register allocation and runtime support for try/catch
This patch completes a series of CLs that add support for try/catch
in the Optimizing compiler. With it, Optimizing can compile all
methods containing try/catch, provided they don't contain catch loops.
Future work will focus on improving performance of the generated code.
SsaLivenessAnalysis was updated to propagate liveness information of
instructions live at catch blocks, and to keep location information on
instructions which may be caught by catch phis.
RegisterAllocator was extended to spill values used after catch, and
to allocate spill slots for catch phis. Catch phis generated for the
same vreg share a spill slot as the raw value must be the same.
Location builders and slow paths were updated to reflect the fact that
throwing an exception may not lead to escaping the method.
Instruction code generators are forbidden from using of implicit null
checks in try blocks as live registers need to be saved before handing
over to the runtime.
CodeGenerator emits a stack map for each catch block, storing locations
of catch phis. CodeInfo and StackMapStream recognize this new type of
stack map and store them separate from other stack maps to avoid dex_pc
conflicts.
After having found the target catch block to deliver an exception to,
QuickExceptionHandler looks up the dex register maps at the throwing
instruction and the catch block and copies the values over to their
respective locations.
The runtime-support approach was selected because it allows for the
best performance in the normal control-flow path, since no propagation
of catch phi values is necessary until the exception is thrown. In
addition, it also greatly simplifies the register allocation phase.
ConstantHoisting was removed from LICMTest because it instantiated
(now abstract) HConstant and was bogus anyway (constants are always in
the entry block).
Change-Id: Ie31038ad8e3ee0c13a5bbbbaf5f0b3e532310e4e
|
| 3ecfd65143d95bd7c6cbe4f58c33af517d3761e0 |
|
07-Sep-2015 |
Yevgeny Rouban <yevgeny.y.rouban@intel.com> |
Add dex_pc to all HInstructions in builder.
Optimizing compiler generates minimum debug line info that
is built using the dex_pc information about suspend points.
This is not enough for performance and debugging needs.
This patch makes all HInstructions contain
dex_pc and all allocations in the builder define this value.
Change-Id: I1d14aefe075189b7b1b41b4384c3499474c19afc
Signed-off-by: Yevgeny Rouban <yevgeny.y.rouban@intel.com>
Signed-off-by: Serdjuk, Nikolay Y <nikolay.y.serdjuk@intel.com>
|
| 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
|
| bbd733e4ef277eff19bf9a6601032da081e9b68f |
|
18-Aug-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Enable basic optimizations for try/catch
Generating code for try/catch methods requires having run at least the
instruction simplifier to remove redundant suspend checks. This patch
enables the first group of optimizations when try/catch is present.
Enabled optimizations:
1) IntrinsicsRecognizer
Does not modify the graph, only sets HInvoke::intrinsic_.
2) ConstantFolding
Does not deal with throwing instructions.
3) InstructionSimplifier
May remove a throwing instruction (e.g. LoadClass in VisitCheckCast),
or may turn a throwing instruction into a non-throwing one (ArraySet).
Their corresponding catch phi inputs are not removed but correctness
is preserved.
4) ReferenceTypePropagation
Does not modify the graph, only sets type properties. Typing of
LoadException from catch handler information was added.
5) DeadCodeElimination
Removing individual instructions is fine (same as 3). Removal of dead
blocks was disabled for try/catch.
Change-Id: I2722c3229eb8aaf326391e07f522dbf5186774b8
|
| ec16f79a4d0aeff319bf52139a0c82de3080d73c |
|
19-Aug-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Refactor try/catch block info, store exception type
This patch replaces HBasicBlock fields storing try/catch info with a
single TryCatchInformation data structure, saving memory for the
majority of non-try/catch blocks. It also changes builder to store
the exception type for catch blocks.
Change-Id: Ib3e43f7db247e6915d67c267fc62410420e230c9
|
| 29fc008c9689e9036a3f5e3bd186bbfb5de3cb82 |
|
18-Aug-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Revert storing of exceptional predecessors
After change of the approach for try/catch register allocation, it is
no longer necessary to record instructions which might throw into a
catch block.
Change-Id: I7ef12ed06c49a35280029810975fa2a50fe4a424
|
| 9867bc722f7c41e07a95397bc08b790cd21dc758 |
|
05-Aug-2015 |
Roland Levillain <rpl@google.com> |
Have constant folding be more flexible.
- Have Evaluate methods take as argument(s) and return value
instances of HConstant (instead of built-in 32- or 64-bit
integer values), to let the evaluated instruction choose
the type of the statically evaluated node; for instance,
art::HEqual::Evaluate shall return a HIntConstant
node (as implementation of a Boolean constant) whatever
the type of its inputs (a pair of HIntConstant or a pair
of HLongConstant).
- Split the evaluation job from the operation logic: the
former is addressed by Evaluate methods, while the latter
is done by a generic Compute method.
- Adress valid BinOp(int, long) and BinOp(long, int) cases.
- Add a constructor to art::HIntConstant to build an integer
constant from a `bool` value.
Change-Id: If84b6fe8406bb94ddb1aa8b02e36628dff526db3
|
| c90bc7c07f9bd24b5424cfb1e3f064fbae5334d6 |
|
11-Dec-2014 |
Roland Levillain <rpl@google.com> |
Add constant folding for long unary operations in opt. compiler.
Add tests to exercise the constant folding of these
instructions.
Also, prevent Java methods from run-tests exercising the
code generation of these instruction from being inlined,
so that they continue to check the generated code (and
not the code produced by the constant folding pass).
Change-Id: I28efca7cdb5142ac2b6d158ba296fb9136d62481
|
| b618adebbc19e50d7b1aa2f11b84341beb3c64dc |
|
29-Jul-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Store and check exceptional predecessors
Future CL on register allocation for try/catch will require the
knowledge of instructions which throw into a catch block. This patch
stores that information with the basic block and verifies it in the
graph checker.
More checks on try catch also added to the graph checker and an order
of exception handlers is enforced in TryBoundary successors.
Change-Id: I3034c610791ea51d96724bcca97f49ec6ecf2af3
|
| 2e76830f0b3f23825677436c0633714402715099 |
|
28-Jul-2015 |
Calin Juravle <calin@google.com> |
Revert "Revert "Revert "Revert "Use the object class as top in reference type propagation""""
This reverts commit b734808d0c93af98ec4e3539fdb0a8c0787263b0.
Change-Id: Ifd925f166761bcb9be2268ff0fc9fa3a72f00c6f
|
| b734808d0c93af98ec4e3539fdb0a8c0787263b0 |
|
28-Jul-2015 |
Calin Juravle <calin@google.com> |
Revert "Revert "Revert "Use the object class as top in reference type propagation"""
This reverts commit 80caa1478cf3df4eac1214d8a63a4da6f4fe622b.
Change-Id: I63b51ca418b19b2bfb5ede3f8444f8fbeb8a339d
|
| 80caa1478cf3df4eac1214d8a63a4da6f4fe622b |
|
16-Jul-2015 |
Calin Juravle <calin@google.com> |
Revert "Revert "Use the object class as top in reference type propagation""
This reverts commit 7733bd644ac71f86d4b30a319624b23343882e53.
Change-Id: I7d393a808c01c084c18d632a54e0554b4b455f2c
|
| 7733bd644ac71f86d4b30a319624b23343882e53 |
|
22-Jul-2015 |
Calin Juravle <calin@google.com> |
Revert "Use the object class as top in reference type propagation"
This reverts commit 3fabec7a25d151b26ba7de13615bbead0dd615a6.
Change-Id: Id8614f6b6e3e0e4c9caeb9f771e4c145d9fec64f
|
| 3fabec7a25d151b26ba7de13615bbead0dd615a6 |
|
16-Jul-2015 |
Calin Juravle <calin@google.com> |
Use the object class as top in reference type propagation
This properly types all instructions, making it safe to query the type
at any time.
This also moves a few functions from class.h to class-inl.h to please
gcc linker when compiling for target.
Change-Id: I6b7ce965c10834c994b95529ab65a548515b4406
|
| ffee3d33f3ea39aa6031c3d2ff29c4806c8dcc51 |
|
06-Jul-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Build SSA form when try/catch is present
This patch implements support for try/catch in the SsaBuilder.
Values of locals are propagated from throwing sites inside try
blocks to their respective catch blocks and phis ("catch phis")
are created when necessary.
Change-Id: I0736565c2c4ff3f9f0924b6e3a785a50023f875a
|
| beba9302bec33d72beb582970bf23d056f62641f |
|
08-Jul-2015 |
Calin Juravle <calin@google.com> |
Revert "Use the object class as top in reference type propagation"
failing on the build bot on some targets but not locally. needs more investigation.
This reverts commit 20e6071362b84a9782b633a893c29ebde458205e.
Change-Id: I6965483f569fb862f9bdb66d459b747ded54de71
|
| 20e6071362b84a9782b633a893c29ebde458205e |
|
01-Jul-2015 |
Calin Juravle <calin@google.com> |
Use the object class as top in reference type propagation
This properly types all instructions, making it safe to query the type
at any time.
Change-Id: I3ee2f0f79253cdf45b10ddab37ecb473345ca53a
|
| c470193cfc522fc818eb2eaab896aef9caf0c75a |
|
10-Apr-2015 |
Mark Mendell <mark.p.mendell@intel.com> |
Fuse long and FP compare & condition on x86/x86-64 in Optimizing.
This is a preliminary implementation of fusing long/float/double
compares with conditions to avoid materializing the result from the
compare and condition.
The information from a HCompare is transferred to the HCondition if it
is legal. There must be only a single use of the HCompare, the HCompare
and HCondition must be in the same block, the HCondition must not need
materialization.
Added GetOppositeCondition() to HCondition to return the flipped
condition.
Bug: 21120453
Change-Id: I1f1db206e6dc336270cd71070ed3232dedc754d6
Signed-off-by: Mark Mendell <mark.p.mendell@intel.com>
|
| 56e1accf3966ae92e151567abf4561ef3f6466f4 |
|
30-Jun-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Changes to try-catch in GraphBuilder
This patch adds an additional case into the insertion algorithm for
HTryBoundary inside HGraphBuilder in order to better handle catch
blocks covered by a TryItem.
Building SSA form also required to stop combining HTryBoundaries for
neighbouring TryItems because it was not clear which exception
handlers belong to which try block.
Change-Id: Ic68bd6ef98fee784609fa593cb08dca1f00a15e0
|
| fc6a86ab2b70781e72b807c1798b83829ca7f931 |
|
26-Jun-2015 |
David Brazdil <dbrazdil@google.com> |
Revert "Revert "ART: Implement try/catch blocks in Builder""
This patch enables the GraphBuilder to generate blocks and edges which
represent the exceptional control flow when try/catch blocks are
present in the code. Actual compilation is still delegated to Quick
and Baseline ignores the additional code.
To represent the relationship between try and catch blocks, Builder
splits the edges which enter/exit a try block and links the newly
created blocks to the corresponding exception handlers. This layout
will later enable the SsaBuilder to correctly infer the dominators of
the catch blocks and to produce the appropriate reverse post ordering.
It will not, however, allow for building the complete SSA form of the
catch blocks and consequently optimizing such blocks.
To this end, a new TryBoundary control-flow instruction is introduced.
Codegen treats it the same as a Goto but it allows for additional
successors (the handlers).
This reverts commit 3e18738bd338e9f8363b26bc895f38c0ec682824.
Change-Id: I4f5ea961848a0b83d8db3673763861633e9bfcfb
|
| 3e18738bd338e9f8363b26bc895f38c0ec682824 |
|
26-Jun-2015 |
David Brazdil <dbrazdil@google.com> |
Revert "ART: Implement try/catch blocks in Builder"
Causes OutOfMemory issues, need to investigate.
This reverts commit 0b5c7d1994b76090afcc825e737f2b8c546da2f8.
Change-Id: I263e6cc4df5f9a56ad2ce44e18932ca51d7e349f
|
| 0b5c7d1994b76090afcc825e737f2b8c546da2f8 |
|
11-Jun-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Implement try/catch blocks in Builder
This patch enables the GraphBuilder to generate blocks and edges which
represent the exceptional control flow when try/catch blocks are
present in the code. Actual compilation is still delegated to Quick
and Baseline ignores the additional code.
To represent the relationship between try and catch blocks, Builder
splits the edges which enter/exit a try block and links the newly
created blocks to the corresponding exception handlers. This layout
will later enable the SsaBuilder to correctly infer the dominators of
the catch blocks and to produce the appropriate reverse post ordering.
It will not, however, allow for building the complete SSA form of the
catch blocks and consequently optimizing such blocks.
To this end, a new TryBoundary control-flow instruction is introduced.
Codegen treats it the same as a Goto but it allows for additional
successors (the handlers).
Change-Id: I415b985596d5bebb7b1bb358a46e08b7b04bb53a
|
| 18b236e5261d2b1f312e632a4d3bb2273c8bf641 |
|
24-Jun-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Recompute dominator tree after DCE.
bug:22031382
(cherry picked from commit 1f82ecc6a0c9f88d03d6d1a6d95eeb8707bd06c1)
Change-Id: I9a74edb185cb806045903dfe9695d9cc1a02e86b
|
| fe659462e7d58bb2585b1bd029f9e08fd9dd32ae |
|
24-Jun-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Stop creating a fallthrough block for Goto
Optimizing's Builder used to create a basic block after a Goto under
the assumption that control flow can fall through.
Bug: 19084197
Change-Id: Id85f31df98a4177466750d3cd0bc8bb74782ca2d
|
| 1f82ecc6a0c9f88d03d6d1a6d95eeb8707bd06c1 |
|
24-Jun-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Recompute dominator tree after DCE.
bug:22031382
Change-Id: Ifebe169897b76872015e3ce0ed7d0a9662f80cef
|
| 1e256bf257e8d97df9b2178ae8658b731ca2d662 |
|
19-Jun-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Be careful with predecessor/successor index.
When we simplify the CFG, we must preserve things that were already
simplified. For example, the index in the predecessor list or
successor list of a block must be preserved for ensuring the
first block is a loop pre header.
bug:21867463
(cherry picked from commit 8b20f88b0a8d1b374dd5eaae289d19734c77b8f8)
Change-Id: I2581b5a50942290da96cd9ec876f6f2573e0a6c4
|
| 25fde612b0df01a086cd4c801b7bd3a10e93a0e9 |
|
18-Jun-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Fix a bug in optimizing when the null constant has been DCE.
If it has been DCE, we should create a new one, instead of
using the old one.
Also move the first DCE to a place where it could actually
be useful.
bug:21870788
(cherry picked from commit 18e6873c469b48aaed22148451523479eece98e3)
Change-Id: I3b3ab2dafe8ce5fb60868fd1a6ef0eeefe666e0c
|
| bca381a12965a98e3727e93986dd0a195db500a0 |
|
20-May-2015 |
Mingyao Yang <mingyao@google.com> |
Fix premature deoptimization if the loop body isn't entered.
Add a test between initial_ and end_ to see if the loop body is entered.
If the loop body isn't entered at all, we jump to the loop header. Loop header is
still executed and is going to test the condition again and loop body won't be
entered. This makes sure no deoptimization is triggered if the loop body isn't
even entered.
Bug: 21034044
(cherry picked from commit 3584bce5b1f45e5741d3a6ca24884a36320ecb6b)
Change-Id: I2b6de1f22fbc4568ca419f76382ebd87806d9694
|
| 8b20f88b0a8d1b374dd5eaae289d19734c77b8f8 |
|
19-Jun-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Be careful with predecessor/successor index.
When we simplify the CFG, we must preserve things that were already
simplified. For example, the index in the predecessor list or
successor list of a block must be preserved for ensuring the
first block is a loop pre header.
bug:21867463
Change-Id: Ic3fcb3eb2c3fb109d8a57ee2a6b6d4d65fdb9410
|
| 18e6873c469b48aaed22148451523479eece98e3 |
|
18-Jun-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Fix a bug in optimizing when the null constant has been DCE.
If it has been DCE, we should create a new one, instead of
using the old one.
Also move the first DCE to a place where it could actually
be useful.
bug:21870788
Change-Id: I28fc52ae481ef92cba45fc1b5abcf07c995f524c
|
| f78848f2ced8466b5fb2d7148d608288ee88757b |
|
17-Jun-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Don't special case HCurrentMethod in DCE.
Instead, re-create the HCurrentMethod if it is needed
after it has been removed.
Change-Id: Id3bf15ae87b00a1d7eb35bf36d58fe96f788fba4
|
| 3584bce5b1f45e5741d3a6ca24884a36320ecb6b |
|
20-May-2015 |
Mingyao Yang <mingyao@google.com> |
Fix premature deoptimization if the loop body isn't entered.
Add a test between initial_ and end_ to see if the loop body is entered.
If the loop body isn't entered at all, we jump to the loop header. Loop header is
still executed and is going to test the condition again and loop body won't be
entered. This makes sure no deoptimization is triggered if the loop body isn't
even entered.
Bug: 21034044
Change-Id: I2b6de1f22fbc4568ca419f76382ebd87806d9694
|
| 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
|
| d23eeef3492b53102eb8093524cf37e2b4c296db |
|
18-May-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Support for inlining methods that call/throw.
Mostly fixes here and there to make it working.
Change-Id: I1b535e895105d78b65634636d675b818551f783e
|
| 76b1e1799a713a19218de26b171b0aef48a59e98 |
|
27-May-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Add a HCurrentMethod node.
This enables register allocation for the current method, so
that users of it don't always load it from the stack.
Currently only used by HLoadClass. Will make follow-up
CLs for the other users.
Change-Id: If73324d85643102faba47fabbbd2755eb258c59c
|
| 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
|
| c7af85dad0dc392cfc0b373b0c1cb4b4197c89f4 |
|
26-May-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Update graph's exit block field if removed
Running DCE on an infinite loop will delete the exit block but the
corresponding field is currently not cleared in the parent graph.
This does not cause any problems at the moment as that information is
only used in codegens to DCHECK that a block is not the exit block.
However, it will be necessary to update the inliner once we start to
inline methods with loops.
With this patch, DCE will update the HGraph::exit_block_ field. DCHECK
was also added to HGraph::InlineInto to make sure that the inlined
graph does have an exit block.
Change-Id: Ia8ddca375bbc6830cd919af6059a52cc9b73a023
|
| e82549b14c7def0a45461183964f7e6a34cbb70c |
|
06-May-2015 |
Mark Mendell <mark.p.mendell@intel.com> |
[optimizing] Fold HTypeConversion of constants
While looking into optimizing long shifts on x86, I found that the
compiler wasn't folding HTypeConversion of constants. Add simple
conversions of constants, taking care of float/double values
with NaNs and small/large values, ensuring Java conversion semantics.
Add checker cases to see that constant folding of HTypeConversion is
done.
Ensure 422-type-conversion type conversion routiness don't get
inlined to avoid compile time folding.
Change-Id: I5a4eb376b64bc4e41bf908af5875bed312efb228
Signed-off-by: Mark Mendell <mark.p.mendell@intel.com>
|
| e8ff50df01c89e1b5264a5a900cfebdde87a9b44 |
|
07-May-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Rediscover loops after deleting blocks in DCE
The way DCE currently updates loop information does not cover all
cases. This patch removes the logic, resets loop information of live
blocks to pre-SSA state and reanalyzes the affected loops.
Change-Id: I0b996a70235b95a8db0de9a23a03f71db57a21b8
(cherry picked from commit a4b8c21dae70ae34aee13628632c39a675c06022)
|
| a4b8c21dae70ae34aee13628632c39a675c06022 |
|
07-May-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Rediscover loops after deleting blocks in DCE
The way DCE currently updates loop information does not cover all
cases. This patch removes the logic, resets loop information of live
blocks to pre-SSA state and reanalyzes the affected loops.
Change-Id: I0b996a70235b95a8db0de9a23a03f71db57a21b8
|
| 0a23d74dc2751440822960eab218be4cb8843647 |
|
07-May-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Add a parent environment to HEnvironment.
This code has no functionality change. It adds a placeholder
for chaining inlined frames.
Change-Id: I5ec57335af76ee406052345b947aad98a6a4423a
|
| 8c0c91a845568624815df026cfdac8c42ecccdf6 |
|
07-May-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Use a growable array instead of an environment during SSA.
Using an environment was convenient because it contains
a growable array. But there's no need for the environment
abstraction when being used as a temporary holder for values
of locals.
Change-Id: Idf2883fe4b8f97a31ee70b3627c1bdd23ebfff0e
|
| db216f4d49ea1561a74261c29f1264952232728a |
|
05-May-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Relax the only one back-edge restriction.
The rule is in the way for better register allocation, as
it creates an artificial join point between multiple paths.
Change-Id: Ia4392890f95bcea56d143138f28ddce6c572ad58
|
| 6db49a74e8402d3b6c66536ea7ec988144c05d24 |
|
28-Apr-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Update the remaining input index of phis after deleting an input.
bug:20715803
bug:20690906
(cherry picked from commit 5d7b7f81ed5455893f984752c00571ef27cc97c5)
Change-Id: Ie55739601b8d6fedc830d6e19d8a053392047d34
|
| 5d7b7f81ed5455893f984752c00571ef27cc97c5 |
|
28-Apr-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Update the remaining input index of phis after deleting an input.
bug:20715803
bug:20690906
Change-Id: Iaf08f0c30d629e766be2b04815dc3e38b6e7ff35
|
| 395086f0a9e0658a2d33eeade7121db55c1f5dc8 |
|
29-Apr-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Fix loop information after dead code elimination
Compilation failed when only some blocks of a loop were removed during
dead code elimination.
Bug: 20680703
(cherry picked from commit 69a2804c3bb48cf4fd00a66080f613a4fd96c422)
Change-Id: If9988381236e4d8d8c3b508dfce1376b27c20d75
|
| 69a2804c3bb48cf4fd00a66080f613a4fd96c422 |
|
29-Apr-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Fix loop information after dead code elimination
Compilation failed when only some blocks of a loop were removed during
dead code elimination.
Bug: 20680703
Change-Id: If31025169ca493f0d7f7f2788576e98d05f03394
|
| 2b1c622d5db941fe06b3ea9c1a5366358fa298c6 |
|
27-Apr-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Fix removing a Phi with RemoveInstruction
Boolean simplifier might attempt to remove a Phi from the Instruction
list.
(cherry picked from commit c7508e93fa3df3a3890f6b62550cbd5e35bdd8df)
Change-Id: Ic8ad31967aa3e47c1fb1c67553d08681b6063a16
|
| f213e05cef6d38166cfe0cce8f3b0a53225a1b39 |
|
27-Apr-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Add support for caching float and double constants.
Change-Id: Ib5205bad1006bc5e3c9cc86bc82a6b4b1ce9bef9
|
| c7508e93fa3df3a3890f6b62550cbd5e35bdd8df |
|
27-Apr-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Fix removing a Phi with RemoveInstruction
Boolean simplifier might attempt to remove a Phi from the Instruction
list.
Change-Id: I698cc616549bd88dac96395cb2e5d09b5433d157
|
| 2967ec6c3dad1c1dc15fc827188bd5ecfa75493b |
|
24-Apr-2015 |
Guillaume "Vermeille" Sanchez <guillaumesa@google.com> |
Add InsertInstructionAfter in HBasicBlock.
Change-Id: I56e4e6edb39d1aab747877b7e517e94f0393f296
|
| 206d6fd6cae5ba8c4d5f0e230111fe77b9d5c0a5 |
|
14-Apr-2015 |
Mingyao Yang <mingyao@google.com> |
Deoptimization-based BCE for unknown loop bounds.
For loop like:
for (int i = start; i < end; i++) {
array[i] = 1;
}
We add the following to the loop pre-header:
if (start < 0) deoptimize();
if (end > array.length) deoptimize();
Then we can eliminate bounds-check of array[i] inside the loop.
We also take care of indexing with induction variable plus some offsets,
like array[i - 1]/array[i + 1] inside the loop, and adjust the condition
for deoptimization accordingly.
Change-Id: I9e24c6b5e134ff95eff5b5605ff8f95d6546616f
|
| 067cae2c86627d2edcf01b918ee601774bc76aeb |
|
26-Apr-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Revert "[optimizing] Replace FP divide by power of 2"
Fails compiling docs.
This reverts commit b0bd8915cb257cdaf46ba663c450a6543bca75af.
Change-Id: I47d32525c83a73118e2163eb58c68bbb7a28bb38
|
| 4c0eb42259d790fddcd9978b66328dbb3ab65615 |
|
24-Apr-2015 |
Roland Levillain <rpl@google.com> |
Ensure inlined static calls perform clinit checks in Optimizing.
Calls to static methods have implicit class initialization
(clinit) checks of the method's declaring class in
Optimizing. However, when such a static call is inlined,
the implicit clinit check vanishes, possibly leading to an
incorrect behavior.
To ensure that inlining static methods does not change the
behavior of a program, add explicit class initialization
checks (art::HClinitCheck) as well as load class
instructions (art::HLoadClass) as last input of static
calls (art::HInvokeStaticOrDirect) in Optimizing' control
flow graphs, when the declaring class is reachable and not
known to be already initialized. Then when considering the
inlining of a static method call, proceed only if the method
has no implicit clinit check requirement.
The added explicit clinit checks are already removed by the
art::PrepareForRegisterAllocation visitor. This CL also
extends this visitor to turn explicit clinit checks from
static invokes into implicit ones after the inlining step,
by removing the added art::HLoadClass nodes mentioned
hereinbefore.
Change-Id: I9ba452b8bd09ae1fdd9a3797ef556e3e7e19c651
|
| 2d7352ba5311b8f57427b91b7a891e61497373c1 |
|
20-Apr-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Dead block removal
Adds a new pass which finds all unreachable blocks, typically due to
simplifying an if-condition to a constant, and removes them from the
graph. The patch also slightly generalizes the graph-transforming
operations.
Change-Id: Iff7c97f1d10b52886f3cd7401689ebe1bfdbf456
|
| 2cebb24bfc3247d3e9be138a3350106737455918 |
|
22-Apr-2015 |
Mathieu Chartier <mathieuc@google.com> |
Replace NULL with nullptr
Also fixed some lines that were too long, and a few other minor
details.
Change-Id: I6efba5fb6e03eb5d0a300fddb2a75bf8e2f175cb
|
| c3d743fa2a26effcb35627d8a1338029c86e582a |
|
22-Apr-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Update last_instruction when adding Phis
HBasicBlock::InsertPhiAfter would not update the last_instruction
pointer when adding at the end of the list. This could cause problems
when iterating over phis backwards. Fortunately, we don't do that
anywhere in the existing code.
Change-Id: I4487265bf2cf3d3819623fafd7ce7c359bac190e
|
| 7d275379bf490a87805852129e3fe2e8afe961e7 |
|
21-Apr-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Update loop info of all nested loops when inlining
When inlining into a nested loop, the inliner would only add the new
blocks into the innermost loop info object. This patch fixes that and
modifies SsaChecker to verify the property.
Change-Id: I21d343a6f7d972f5b7420701f816c65ab3f20566
|
| b0bd8915cb257cdaf46ba663c450a6543bca75af |
|
16-Apr-2015 |
Mark Mendell <mark.p.mendell@intel.com> |
[optimizing] Replace FP divide by power of 2
Replace a floating point division by a power of two by a multiplication
of the reciprocal. This is guarenteed to have the exact same result as
it is exactly representable.
Add routines to allow generation of float and double constants after the
SSA Builder. I was unsure if float and double caches should be
implemented. Under the assumption that there is probably not a lot of
repetition of FP values. Please let me know.
Change-Id: I3a6c3847b49b4e747a7e7e8843ca32bb174b1584
Signed-off-by: Mark Mendell <mark.p.mendell@intel.com>
|
| 05144f4322eed049f4878015bf1f0381d419b785 |
|
16-Apr-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Hot fix for an inliner issue
Change 147130 caused a problem with duplicit instruction ids when
inlining methods with constants. This is a hot fix to unblock build.
Change-Id: Ieddadcd94135930a1f29ad64ad57349a384da07f
|
| 4a3faecbe4157225a3fe83a9ef7f4992dfc9c19d |
|
16-Apr-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Don't duplicate null/int/long constants when inlining
Change-Id: I7e6a3393fcbbcf76b4ba2000915ba6bbbfb7c70e
|
| f776b92a0d52bb522043812dacb9c21ac11858e2 |
|
15-Apr-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Remove dead blocks for the blocks_ array.
This prevents crashing because of structurally incorrect
blocks. Also we now don't need to remove its instructions.
Test case courtesy of Serguei I Katkov.
Change-Id: Ia3ef9580549fc3546e8cd5f346079b1f0ceb2a61
|
| d43b3ac88cd46b8815890188c9c2b9a3f1564648 |
|
01-Apr-2015 |
Mingyao Yang <mingyao@google.com> |
Revert "Revert "Deoptimization-based bce.""
This reverts commit 0ba627337274ccfb8c9cb9bf23fffb1e1b9d1430.
Change-Id: I1ca10d15bbb49897a0cf541ab160431ec180a006
|
| 8d5b8b295930aaa43255c4f0b74ece3ee8b43a47 |
|
24-Mar-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Force constants into the entry block
Optimizations such as GVN and BCE make the assumption that all
constants are located in the entry block of the CFG, but not all
passes adhere to this rule.
This patch makes constructors of constants private and only accessible
to friend classes - HGraph for int/long constants and SsaBuilder for
float/double - which ensure that they are placed correctly and not
duplicated.
Note that the ArenaAllocatorAdapter was modified to not increment
the ArenaAllocator's internal reference counter in order to allow
for use of ArenaSafeMap inside an arena-allocated objects. Because
their destructor is not called, the counter does not get decremented.
Change-Id: I36a4fa29ae34fb905cdefd482ccbf386cff14166
|
| b2bd1c5f9171f35fa5b71ada42d1a9e11189428d |
|
25-Mar-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Formatting and comments in BooleanSimplifier
Change-Id: I9a5aa3f2aa8b0a29d7b0f1e5e247397cf8e9e379
|
| 46e2a3915aa68c77426b71e95b9f3658250646b7 |
|
16-Mar-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Boolean simplifier
The optimization recognizes the negation pattern generated by 'javac'
and replaces it with a single condition. To this end, boolean values
are now consistently assumed to be represented by an integer.
This is a first optimization which deletes blocks from the HGraph and
does so by replacing the corresponding entries with null. Hence,
existing code can continue indexing the list of blocks with the block
ID, but must check for null when iterating over the list.
Change-Id: I7779da69cfa925c6521938ad0bcc11bc52335583
|
| 0ba627337274ccfb8c9cb9bf23fffb1e1b9d1430 |
|
24-Mar-2015 |
Andreas Gampe <agampe@google.com> |
Revert "Deoptimization-based bce."
This breaks compiling the core image:
Error after BCE: art::SSAChecker: Instruction 219 in block 1 does not dominate use 221 in block 1.
This reverts commit e295e6ec5beaea31be5d7d3c996cd8cfa2053129.
Change-Id: Ieeb48797d451836ed506ccb940872f1443942e4e
|
| e295e6ec5beaea31be5d7d3c996cd8cfa2053129 |
|
07-Mar-2015 |
Mingyao Yang <mingyao@google.com> |
Deoptimization-based bce.
A mechanism is introduced that a runtime method can be called
from code compiled with optimizing compiler to deoptimize into
interpreter. This can be used to establish invariants in the managed code
If the invariant does not hold at runtime, we will deoptimize and continue
execution in the interpreter. This allows to optimize the managed code as
if the invariant was proven during compile time. However, the exception
will be thrown according to the semantics demanded by the spec.
The invariant and optimization included in this patch are based on the
length of an array. Given a set of array accesses with constant indices
{c1, ..., cn}, we can optimize away all bounds checks iff all 0 <= min(ci) and
max(ci) < array-length. The first can be proven statically. The second can be
established with a deoptimization-based invariant. This replaces n bounds
checks with one invariant check (plus slow-path code).
Change-Id: I8c6e34b56c85d25b91074832d13dba1db0a81569
|
| 4f1a384762bf49fe8f3ecae8dd2bcb0e19d044a9 |
|
12-Mar-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Give an expected type to phis created for multiple returns.
When inlining, we used to take the type of the inlined method
for the phi in case of multiple returns. I recently changed the
logic of phi types to only be of int/float/double/ref, so we
need to call ToPhiType when creating the phi.
Change-Id: I960067ca8a8814509c2a7c52c08387d892ebf4a3
|
| b2fd7bca70b580921eebf7c45769c39d2dfd8a5a |
|
11-Mar-2015 |
Alexandre Rames <alexandre.rames@arm.com> |
Opt compiler: Basic simplification for arithmetic operations.
The optimisations in this patch do not look further than the
inputs of each operation.
Change-Id: Iddd0ab6b360b9e7bb042db22086d51a31be85530
|
| 817bce7658918b7a70c17b70aa5e6a46b1ae8b3d |
|
24-Feb-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Fix inlining in the presence of multiple returns.
One return could actually return a phi, so doing a phi check for
knowing if the result phi was already created was bogus.
Bug: 19454010
Change-Id: Iee703a2d1071ae263092354465eda368e5d6770d
|
| 1abb4191a2e56d8dbf518efcaeefb266c1acdf2b |
|
17-Feb-2015 |
David Brazdil <dbrazdil@google.com> |
Optimizing: Speed up HInstruction use removal
Similarly to a previous commit on HEnvironment use removal, this patch
adds links from instructions to their respective inputs' use lists for
contant-time removal at the cost of doubling the size of input lists
(from one pointer per entry to two). Manual testing shows that this
significantly reduces the time required to transform HGraph to SSA
form for some huge methods.
Change-Id: I8dc3e4b0c48a50ac1481eb55c31093b99f4dc29f
|
| acf735c13998ad2a175f5a17e7bfce220073279d |
|
12-Feb-2015 |
Calin Juravle <calin@google.com> |
Reference type propagation
- propagate reference types between instructions
- remove checked casts when possible
- add StackHandleScopeCollection to manage an arbitrary number of stack
handles (see comments)
Change-Id: I31200067c5e7375a5ea8e2f873c4374ebdb5ee60
|
| d6138ef1ea13d07ae555542f8898b30d89e9ac9a |
|
18-Feb-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Ensure the graph is correctly typed.
We used to be forgiving because of HIntConstant(0) also being
used for null. We now create a special HNullConstant for such uses.
Also, we need to run the dead phi elimination twice during ssa
building to ensure the correctness.
Change-Id: If479efa3680d3358800aebb1cca692fa2d94f6e5
|
| be31ff94d66a0037c445eb57dc82f2a51bb46d9e |
|
04-Feb-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Fix a bug in the inliner.
Code did not work in the presence of multiple returns.
Spotted by Mark P. Mendell.
Change-Id: I237050a0d79c0cfaa479e9b886f7450879e84713
|
| 276d9daaedfbff716339f94d55e6eff98b7434c6 |
|
02-Feb-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Inline methods with multiple blocks.
Change-Id: I3431af60e97fae230e0b6e98bcf0acc0ee9abf8c
|
| 82091dad38f3e5bfaf3b6984c9ab73069fb68310 |
|
26-Jan-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Implement LICM in optimizing compiler.
Change-Id: I9c8afb0a58ef45e568576015473cbfd5f011c242
|
| ed59619b370ef23ffbb25d1d01f615e60a9262b6 |
|
23-Jan-2015 |
David Brazdil <dbrazdil@google.com> |
Optimizing: Speed up HEnvironment use removal
Removal of use records from HEnvironment vregs involved iterating over
potentially large linked lists which made compilation of huge methods
very slow. This patch turns use lists into doubly-linked lists, stores
pointers to the relevant nodes inside HEnvironment and subsequently
turns the removals into constant-time operations.
Change-Id: I0e1d4d782fd624e7b8075af75d4adf0a0634a1ee
|
| 0ada95d8de4b04b5f201b4b7e9c3c2fd2cc321ae |
|
04-Dec-2014 |
Jean Christophe Beyler <jean.christophe.beyler@intel.com> |
ART: Replace NULL to nullptr in the optimizing compiler
Replace macro NULL to the nullptr variation for C++.
Change-Id: Ib6e48dd4bb3c254343383011b67372622578ca76
Signed-off-by: Jean Christophe Beyler <jean.christophe.beyler@intel.com>
|
| 6c2dff8ff8e1440fa4d9e1b2ba2a44d036882801 |
|
21-Jan-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Revert "Revert "Fully support pairs in the register allocator.""
This reverts commit c399fdc442db82dfda66e6c25518872ab0f1d24f.
Change-Id: I19f8215c4b98f2f0827e04bf7806c3ca439794e5
|
| 77520bca97ec44e3758510cebd0f20e3bb4584ea |
|
12-Jan-2015 |
Calin Juravle <calin@google.com> |
Record implicit null checks at the actual invoke time.
ImplicitNullChecks are recorded only for instructions directly (see NB
below) preceeded by NullChecks in the graph. This way we avoid recording
redundant safepoints and minimize the code size increase.
NB: ParallalelMoves might be inserted by the register allocator between
the NullChecks and their uses. These modify the environment and the
correct action would be to reverse their modification. This will be
addressed in a follow-up CL.
Change-Id: Ie50006e5a4bd22932dcf11348f5a655d253cd898
|
| c399fdc442db82dfda66e6c25518872ab0f1d24f |
|
21-Jan-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Revert "Fully support pairs in the register allocator."
Libcore tests fail.
This reverts commit 41aedbb684ccef76ff8373f39aba606ce4cb3194.
Change-Id: I2572f120d4bbaeb7a4d4cbfd47ab00c9ea39ac6c
|
| 41aedbb684ccef76ff8373f39aba606ce4cb3194 |
|
14-Jan-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Fully support pairs in the register allocator.
Enabled on ARM for longs and doubles.
Change-Id: Id8792d08bd7ca9fb049c5db8a40ae694bafc2d8b
|
| 9ee66183d8e046ea661f642ba884626f16b46e06 |
|
16-Jan-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Constant fold after inlining.
- Inlining opens up new opportunities for constant folding.
- Fix a bug in constant folder where the result type was not
correctly set for the folding of a HCompare.
- Improve graph checker's coverage.
Change-Id: I0943bf8ff65505c4addc4a555a526b55e00b5268
|
| 7c5367badfe61b96c5836d495d286cee64861579 |
|
17-Dec-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Fix ids and remove invoke when inlining.
Bugs found by Razvan Lupusoru.
Change-Id: I3a5a9af280d8700d18f52abb4a2cff0e3a9aac74
|
| 53d9da8507a1b68f036ce8669ad3f2ae9fc3d225 |
|
04-Dec-2014 |
Jean Christophe Beyler <jean.christophe.beyler@intel.com> |
ART: Create a RemoveBlock method
The RemoveDeadBlocks should be separated into a utility function to remove
a single block so that it can be used as a future utility method.
Change-Id: I4c67113fff24e92a66a81bc0e8edf9fbdda08cdf
Signed-off-by: Jean Christophe Beyler <jean.christophe.beyler@intel.com>
|
| e53798a7e3267305f696bf658e418c92e63e0834 |
|
01-Dec-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Inlining support in optimizing.
Currently only inlines simple things that don't require an
environment, such as:
- Returning a constant.
- Returning a parameter.
- Returning an arithmetic operation.
Change-Id: Ie844950cb44f69e104774a3cf7a8dea66bc85661
|
| fc600dccd7797a9a10cdd457034ea8e148ccd631 |
|
02-Dec-2014 |
Roland Levillain <rpl@google.com> |
Fix a compiler bug related to a catch-less try-finally statement.
Ensure a dead basic block produced in this case is properly
removed.
Change-Id: I7c88e26aaa6c6378892f7c7c299494fa42312db2
|
| f537012ceb6cba8a78b36a5065beb9588451a250 |
|
02-Dec-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Treat SSA transformation special, as we may have to bailout.
We forgot to bailout when we found a non-natural loop (on which
our optimizations don't work).
Change-Id: I11976b5af4c98f4f29267a74c74d34b5ad81e20c
|
| 6a3c1fcb4ba42ad4d5d142c17a3712a6ddd3866f |
|
31-Oct-2014 |
Ian Rogers <irogers@google.com> |
Remove -Wno-unused-parameter and -Wno-sign-promo from base cflags.
Fix associated errors about unused paramenters and implict sign conversions.
For sign conversion this was largely in the area of enums, so add ostream
operators for the effected enums and fix tools/generate-operator-out.py.
Tidy arena allocation code and arena allocated data types, rather than fixing
new and delete operators.
Remove dead code.
Change-Id: I5b433e722d2f75baacfacae4d32aef4a828bfe1b
|
| b762d2ebf9dc604561d9915c96b377235c94960c |
|
22-Oct-2014 |
Roland Levillain <rpl@google.com> |
Various fixes related to integer negate operations.
- Emit an RSB instruction for HNeg nodes in the ARM code
generator instead of RSBS, as we do not need to update the
condition code flags in this case.
- Simply punt when trying to statically evaluate a long
unary operation, instead of aborting.
- Move a test case to the right place.
Change-Id: I35eb8dea58ed35258d4d8df77181159c3ab07b6f
|
| 102cbed1e52b7c5f09458b44903fe97bb3e14d5f |
|
15-Oct-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Implement register allocator for floating point registers.
Also:
- Fix misuses of emitting the rex prefix in the x86_64 assembler.
- Fix movaps code generation in the x86_64 assembler.
Change-Id: Ib6dcf6e7c4a9c43368cfc46b02ba50f69ae69cbe
|
| 9240d6a2baa9ed1e18ee08744b461fe49a1ee269 |
|
20-Oct-2014 |
Roland Levillain <rpl@google.com> |
Constant folding on unary operations in the optimizing compiler.
Change-Id: I4b77afa2a89f5ad2eedd4d6c0c6c382585419349
|
| 6c82d40eb142771086f5531998de2273ba5cc08c |
|
13-Oct-2014 |
Roland Levillain <rpl@google.com> |
Have HInstruction::StrictlyDominates compute strict dominance.
Change-Id: I3a4fa133268615fb4ce54a0bcb43e0c2458cc865
|
| 633021e6ff6b9a57a374a994e74cfd69275ce100 |
|
01-Oct-2014 |
Roland Levillain <rpl@google.com> |
Implement default traversals in CFG & SSA graph checkers.
- Check CFG graphs using an insertion order traversal.
- Check SSA form graphs using a reverse post-order traversal.
Change-Id: Ib9062599bdbf3c17b9f213b743274b2d71a9fa90
|
| 476df557fed5f0b3f32f8d11a654674bb403a8f8 |
|
09-Oct-2014 |
Roland Levillain <rpl@google.com> |
Use Is*() helpers to shorten code in the optimizing compiler.
Change-Id: I79f31833bc9a0aa2918381aa3fb0b05d45f75689
|
| 360231a056e796c36ffe62348507e904dc9efb9b |
|
08-Oct-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Fix code generation of materialized conditions.
Move the logic for knowing if a condition needs to be materialized
in an optimization pass (so that the information does not change
as a side effect of another optimization).
Also clean-up arm and x86_64 codegen:
- arm: ldr and str are for power-users when a constant is
in play. We should use LoadFromOffset and StoreToOffset.
- x86_64: fix misuses of movq instead of movl.
Change-Id: I01a03b91803624be2281a344a13ad5efbf4f3ef3
|
| 191c4b1372aef7c0272f8fa3985b55513029e728 |
|
07-Oct-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Inserting a node must also update its inputs users.
Change-Id: I55357564b81efcc0cf52fffdf23289696fe27dd1
|
| 3c04974a90b0e03f4b509010bff49f0b2a3da57f |
|
24-Sep-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Optimize suspend checks in optimizing compiler.
- Remove the ones added during graph build (they were added
for the baseline code generator).
- Emit them at loop back edges after phi moves, so that the test
can directly jump to the loop header.
- Fix x86 and x86_64 suspend check by using cmpw instead of cmpl.
Change-Id: I6fad5795a55705d86c9e1cb85bf5d63dadfafa2a
|
| 6b46923ff0197c95f1e7ea0bc730961df6725cc9 |
|
25-Sep-2014 |
Roland Levillain <rpl@google.com> |
Optimizing compiler: check inputs & uses definitions in CFG.
Ensure each input and each use of an instruction is defined
in a block of the control-flow graph.
Change-Id: If4a83b02825230329b0b4fd84255dcb7c3219684
|
| 18efde5017369e005f1e8bcd3bbfb04e85053640 |
|
22-Sep-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Fix code generation with materialized conditions.
Change-Id: I8630af3c13fc1950d3fa718d7488407b00898796
|
| 724c96326dea6ec33287a0076279c136abb0208a |
|
22-Sep-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Also remove environment links to removed instructions.
Change-Id: I505163fb8683269c7d3fe21b34df92337d244552
|
| d31cf3d55a0847c018c4eaa2b349b8eea509de64 |
|
08-Sep-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
First optimization in new compiler: simple GVN.
Change-Id: Ibe0efa4e84fd020a53ded310a92e0b4363f91b12
|
| 556c3d193134f6461f3e1fe17c032b087c5931a0 |
|
18-Sep-2014 |
Roland Levillain <rpl@google.com> |
Initiate a constant propagation pass in the optimizing compiler.
- Perform constant folding on int and long additions and subtractions in
the optimizing compiler.
- Apply constant folding to conditions and comparisons.
Change-Id: Ic88783a3c975fda777c74c531e257fa777be42eb
|
| ccc07a9579c554443cd03a306ca9b4f943fd2a93 |
|
16-Sep-2014 |
Roland Levillain <rpl@google.com> |
Add CFG and SSA form checkers in the optimizing compiler.
Checks performed on control-flow graphs:
- Ensure that the predecessors and successors of a basic block are
consistent within a control-flow graph.
- Ensure basic blocks end with a branch instruction.
- Detect phi functions listed in non-phi instruction lists and vice
versa.
- Ensure a block's instructions (and phi functions) are associated
with this very block.
Checks performed on SSA form graphs:
- Ensure an instruction dominates all its uses.
- Ensure there are no critical edges.
Change-Id: I1c12b4a61ecf608682152c897980ababa7eca847
|
| 604c6e4764edb2fd244e9f47626868cda5644a7a |
|
17-Sep-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Ensure the first predecessor of a loop is the pre header.
Note that the check in ssa_phi_elimination.cc was very defensive:
it does not affect the outcome of the algorithm whether the
loop phi takes itself as the first input.
It makes things consistent to always have the pre header as first
input.
Change-Id: Ic86248c1f38af67f7432782f6deefae1f4bf1ab6
|
| 065bf77b43c39da315b974ea08a5ed25e9049681 |
|
03-Sep-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Add (simple) side effects flags and equality methods on nodes.
This is in preparation of doing GVN and LICM.
Change-Id: I43050ff846755f9387a62b893d548ecdb54e7e95
|
| 3946844c34ad965515f677084b07d663d70ad1b8 |
|
02-Sep-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Runtime support for the new stack maps for the opt compiler.
Now most of the methods supported by the compiler can be optimized,
instead of using the baseline.
Change-Id: I80ab36a34913fa4e7dd576c7bf55af63594dc1fa
|
| 20dfc797dc631bf8d655dcf123f46f13332d3074 |
|
17-Jun-2014 |
Dave Allison <dallison@google.com> |
Add some more instruction support to optimizing compiler.
This adds a few more DEX instructions to the optimizing compiler's
builder (constants, moves, if_xx, etc).
Also:
* Changes the codegen for IF_XX instructions to use a condition
rather than comparing a value against 0.
* Fixes some instructions in the ARM disassembler.
* Fixes PushList and PopList in the thumb2 assembler.
* Switches the assembler for the optimizing compiler to thumb2
rather than ARM.
Change-Id: Iaafcd02243ccc5b03a054ef7a15285b84c06740f
|
| ec7e4727e99aa1416398ac5a684f5024817a25c7 |
|
06-Jun-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Fix some bugs in graph construction/simplification methods.
Also fix a brano during SSA construction. The code should
not have been commented out. Added a test to cover what the code
intends.
Change-Id: Ia00ae79dcf75eb0d412f07649d73e7f94dbfb6f0
|
| ffddfdf6fec0b9d98a692e27242eecb15af5ead2 |
|
03-Jun-2014 |
Tim Murray <timmurray@google.com> |
DO NOT MERGE
Merge ART from AOSP to lmp-preview-dev.
Change-Id: I0f578733a4b8756fd780d4a052ad69b746f687a9
|
| a7062e05e6048c7f817d784a5b94e3122e25b1ec |
|
22-May-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Add a linear scan register allocator to the optimizing compiler.
This is a "by-the-book" implementation. It currently only deals
with allocating registers, with no hint optimizations.
The changes remaining to make it functional are:
- Allocate spill slots.
- Resolution and placements of Move instructions.
- Connect it to the code generator.
Change-Id: Ie0b2f6ba1b98da85425be721ce4afecd6b4012a4
|
| 4e3d23aa1523718ea1fdf3a32516d2f9d81e84fe |
|
22-May-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Import Dart's parallel move resolver.
And write a few tests while at it.
A parallel move resolver will be needed for performing multiple moves
that are conceptually parallel, for example moves at a block
exit that branches to a block with phi nodes.
Change-Id: Ib95b247b4fc3f2c2fcab3b8c8d032abbd6104cd7
|
| f635e63318447ca04731b265a86a573c9ed1737c |
|
14-May-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Add a compilation tracing mechanism to the new compiler.
Code mostly imported from: https://android-review.googlesource.com/#/c/81653/.
Change-Id: I150fe942be0fb270e03fabb19032180f7a065d13
|
| 622d9c31febd950255b36a48b47e1f630197c5fe |
|
12-May-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Add loop recognition and CFG simplifications in new compiler.
We do three simplifications:
- Split critical edges, for code generation from SSA (new).
- Ensure one back edge per loop, to simplify loop recognition (new).
- Ensure only one pre header for a loop, to simplify SSA creation (existing).
Change-Id: I9bfccd4b236a00486a261078627b091c8a68be33
|
| 804d09372cc3d80d537da1489da4a45e0e19aa5d |
|
02-May-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Build live-in, live-out and kill sets for each block.
This information will be used when computing live ranges of
instructions.
Change-Id: I345ee833c1ccb4a8e725c7976453f6d58d350d74
|
| c32e770f21540e4e9eda6dc7f770e745d33f1b9f |
|
24-Apr-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Add a Transform to SSA phase to the optimizing compiler.
Change-Id: Ia9700756a0396d797a00b529896487d52c989329
|
| 43c86422e210a3883729ab90997711e79f26bccc |
|
18-Mar-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Fix lint error, and Makefile that could be confused with local files.
Change-Id: I780cc0d6593eadd6f82e1126d7ad445894af666c
|
| 787c3076635cf117eb646c5a89a9014b2072fb44 |
|
17-Mar-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Plug new optimizing compiler in compilation pipeline.
Also rename accessors to ART's conventions.
Change-Id: I344807055b98aa4b27215704ec362191464acecc
|
| bab4ed7057799a4fadc6283108ab56f389d117d4 |
|
11-Mar-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
More code generation for the optimizing compiler.
- Add HReturn instruction
- Generate code for locals/if/return
- Setup infrastructure for register allocation. Currently
emulate a stack.
Change-Id: Ib28c2dba80f6c526177ed9a7b09c0689ac8122fb
|
| 3ff386aafefd5282bb76c8a50506a70a4321e698 |
|
04-Mar-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Add register support to the optimizing compiler.
Also make if take an input and build the use list for instructions.
Change-Id: I1938cee7dce5bd4c66b259fa2b431d2c79b3cf82
|
| d4dd255db1d110ceb5551f6d95ff31fb57420994 |
|
28-Feb-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Add codegen support to the optimizing compiler.
Change-Id: I9aae76908ff1d6e64fb71a6718fc1426b67a5c28
|
| be9a92aa804c0d210f80966b74ef8ed3987f335a |
|
25-Feb-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Add conditional branches, and build dominator tree.
Change-Id: I4b151a07b72692961235a1419b54b6b45cf54e63
|
| 818f2107e6d2d9e80faac8ae8c92faffa83cbd11 |
|
18-Feb-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Re-apply: Initial check-in of an optimizing compiler.
The classes and the names are very much inspired by V8/Dart.
It currently only supports the RETURN_VOID dex instruction,
and there is a pretty printer to check if the building of the
graph is correct.
Change-Id: I28e125dfee86ae6ec9b3fec6aa1859523b92a893
|
| 1af0c0b88a956813eb0ad282664cedc391e2938f |
|
19-Feb-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Revert "Initial check-in of an optimizing compiler."
g++ warnings turned into errors.
This reverts commit 68a5fefa90f03fdf5a238ac85c9439c6b03eae96.
Change-Id: I09bb95d9cc13764ca8a266c41af04801a34b9fd0
|
| 68a5fefa90f03fdf5a238ac85c9439c6b03eae96 |
|
18-Feb-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Initial check-in of an optimizing compiler.
The classes and the names are very much inspired by V8/Dart.
It currently only supports the RETURN_VOID dex instruction,
and there is a pretty printer to check if the building of the
graph is correct.
Change-Id: Id5ef1b317ab997010d4e3888e456c26bef1ab9c0
|