We do not generate explicit control flow and phis for exceptional
edges. Instead CatchBlockEntry has a Parameter instruction for
each local variable. These Parameter-s represent that state
of local variables on entry into the catch block. This means
that we should mark all variables assigned on entry to the
catch block.
Previous version of the code was incorrectly marking only
direct parameters as stored.
Fixes https://github.com/dart-lang/sdk/issues/36953
Bug: 36953
Change-Id: I2e8040d75ba23d2483e451300f44bbdcef458655
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/103802
Reviewed-by: Daco Harkes <dacoharkes@google.com>
Commit-Queue: Vyacheslav Egorov <vegorov@google.com>
In DEBUG mode during AOT compilation MetadataHelper::SetMetadataMappings
may spend considerable time verifying that node offsets are sorted
in kernel metadata. The problem is that there are a lot of MetadataHelpers
created (several per each compiled function).
This change moves verification of kernel metadata mappings to
the kernel loader, so it is performed only once per kernel program.
AOT gen_snapshot time on a small test:
Before: 11.680s
After: 9.041s
Change-Id: I1c0a676b0ed28fc5cfa756ca60159f914190fac0
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/103486
Commit-Queue: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Aart Bik <ajcbik@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
In addition, if a _ByteDataView came from a use of the ByteData
constructor, its offset is always 0.
Also add appropriate canonicalization for the following instructions for
typed data views coming from a known factory call:
* LoadField getting type arguments
* GuardFieldLength
This closes https://github.com/dart-lang/sdk/issues/36570.
Change-Id: I1c045edb2d928c3bb3340d9d12009c2afa66febb
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/102362
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
Commit-Queue: Teagan Strickland <sstrickl@google.com>
This change adds merging of cid ranges into Cids::Create if there are
gaps with the same target function. Previously, this optimization was
performed only when calculating targets via CallTargets::CreateAndExpand.
However, Cids::Create is widely used to check receiver type if there is
only one targets (e.g. in CallSpecializer::AddReceiverCheck /
CallSpecializer::AddChecksForArgNr).
This optimization is needed to shrink huge code for CheckClass instructions
in Flutter RenderObject.layout() method in JIT mode.
Issue: https://github.com/dart-lang/sdk/issues/36428
Change-Id: I97051b4213066b252ac3238df55ab4ed5c98f412
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/103402
Commit-Queue: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
Reviewed-by: Ryan Macnak <rmacnak@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
This reverts commit fde6a5917e.
Reason for revert: This commit looks to be causing new test failures on dart2js. While there are some odd results on some builders that look like infra failures, some builders (for example, dart2js-minified-strong-linux-x64-d8) show the new failing tests clearly.
Original change's description:
> [vm, compiler] Unoptimized megamorphic calls.
>
> When an instance call in unoptimized code creates more than FLAG_max_polymorphic_checks cases, switch the call to use a MegamorphicCache instead of ICData. The prevents unbounded collection of type feedback, and gives improvements on microbenchmarks in the 3-8% range for unoptimized code.
>
> It also leads to a loss of target frequency information for the optimizer, leading to different ordering for range checks in polymorphic inlining. This leads to changes on megamorphic microbenchmarks from -31% to +60%, weighted toward the negative end.
>
> In practice the frequency information seems unimportant, as dart2js has 4.01% geomean improvement.
>
> This is a step toward direct monomorphic calls in unoptimized code, which will also make use of the patching and type feedback extraction added here.
>
> Bug: https://github.com/dart-lang/sdk/issues/26780
> Bug: https://github.com/dart-lang/sdk/issues/36409
> Bug: https://github.com/dart-lang/sdk/issues/36731
> Change-Id: I29f53f23b6794c5f5f0db8b8184788cee16fd9c5
> Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/99270
> Reviewed-by: Alexander Markov <alexmarkov@google.com>
TBR=rmacnak@google.com,alexmarkov@google.com,ajcbik@google.com
Change-Id: Icad46b93cdf8541a00563f49da6b4ac0a4df1ba1
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug: https://github.com/dart-lang/sdk/issues/26780, https://github.com/dart-lang/sdk/issues/36409, https://github.com/dart-lang/sdk/issues/36731
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/103440
Reviewed-by: Teagan Strickland <sstrickl@google.com>
Commit-Queue: Teagan Strickland <sstrickl@google.com>
Until now we often leaked kernel buffers (e.g. hot reload buffers) because various
objects were referencing ExternalTypedData objects pointing into the middle of
c-allocated memory. This made it impossible for the GC to determine when the last
reference is gone.
This CL ensures that the actual buffers are *always* made available via
ExternalTypedData and any inner pointers into it are created via TypedDataViews.
The embedder guarantees to the free kernel buffers it has provided to:
- Dart_CreateIsolateFromKernel
- Dart_LoadScriptFromKernel
- Dart_LoadLibraryFromKernel
- Dart_SetDartLibrarySourcesKernel
on isolate shutdown.
All other kernel buffers will get a finalizer attached, which ensures the
kernel buffers get freed by the GC once they are no longer referenced:
- Kernel blobs for expression evaluation
- Kernel blobs for Hot-Reload
- Kernel blobs for cc tests
Fixes https://github.com/dart-lang/sdk/issues/33973
Fixes https://github.com/dart-lang/sdk/issues/36857
Issue https://github.com/dart-lang/sdk/issues/37030
Change-Id: I1cc410c94c0f4b229413e793728a261afcb10aaf
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/103130
Reviewed-by: Ryan Macnak <rmacnak@google.com>
Commit-Queue: Martin Kustermann <kustermann@google.com>
When an instance call in unoptimized code creates more than FLAG_max_polymorphic_checks cases, switch the call to use a MegamorphicCache instead of ICData. The prevents unbounded collection of type feedback, and gives improvements on microbenchmarks in the 3-8% range for unoptimized code.
It also leads to a loss of target frequency information for the optimizer, leading to different ordering for range checks in polymorphic inlining. This leads to changes on megamorphic microbenchmarks from -31% to +60%, weighted toward the negative end.
In practice the frequency information seems unimportant, as dart2js has 4.01% geomean improvement.
This is a step toward direct monomorphic calls in unoptimized code, which will also make use of the patching and type feedback extraction added here.
Bug: https://github.com/dart-lang/sdk/issues/26780
Bug: https://github.com/dart-lang/sdk/issues/36409
Bug: https://github.com/dart-lang/sdk/issues/36731
Change-Id: I29f53f23b6794c5f5f0db8b8184788cee16fd9c5
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/99270
Reviewed-by: Alexander Markov <alexmarkov@google.com>
Rationale:
A basic peephole optimizer with window size one that avoids
redundant push-pop sequences already results is substantial
savings in code size and runtime, without any noticeable impact
on compile time (the peephole is very, very fast).
Performance:
Golem unoptimized code (which typically runs -90% compared to
optimized code, sees 5-60% improvements
https://github.com/dart-lang/sdk/issues/36409
Change-Id: I08db4b3dbc92377d89340a4969db6e664e54bceb
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/102980
Commit-Queue: Aart Bik <ajcbik@google.com>
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
Reviewed-by: Ryan Macnak <rmacnak@google.com>
Type propagator adds Redefinition instructions to provide more accurate
type information only for dominated uses (for example, to mark receiver
as non-nullable after instance calls). However, adding Redefinition
instructions may interfere with branch simplification, which looks only
at empty join blocks. This change avoids adding unused Redefinition
instructions (if there are no dominated uses).
This particular situation happens in Flutter in RenderObject.layout() where
constraints.isTight is inlined with multiple polymorphic targets, including
SliverConstraints.isTight which returns false.
Branch simplification was not happening due to Redefinition instruction
for 'constraints'. That Redefinition instruction is not used.
Change-Id: I15082bacd7cc6ea1445a4dadc8ef97853142ba74
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/103023
Commit-Queue: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
When a mixin class is created, there might be anonymous closures copied over to the other script.
Previously the script of closure function is the script of its parent function, which is not true in that case.
The solution is to create a patch class as the owner if parent function has a different script object. Return script of owner when query for script of closure function.
Bug: https://github.com/dart-lang/sdk/issues/34841
Change-Id: I53065cbf13f2d0dc8da320993fd3cd425e5c9714
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/102226
Commit-Queue: Zichang Guo <zichangguo@google.com>
Reviewed-by: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Siva Annamalai <asiva@google.com>
Previously we would generate the prologue in FunctionEntryInstr or OsrEntryInstr
on X64/ARM, and in CompileGraph on ARM64/IA32/DBC. This caused a confusing asymmetry
in the code.
Now we generate it in FunctionEntryInstr/OsrEntryInstr on all backends.
Partially addresses dartbug.com/34162.
This also fixes a regression on a protobuf benchmark due to ICDatas not being saved.
Change-Id: Ie270d759234b197141025dde07a27546d221f6c8
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/101297
Commit-Queue: Samir Jindel <sjindel@google.com>
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
In many cases, the Mutexes and Monitors have to be marked "mutable"
because they're used to synchronize const accessor methods.
Small text segment improvement for Product builds:
$ size dart.{arm,x64}.{before,after}
text data bss dec hex filename
19726069 409960 392332 20528361 1393ce9 dart.arm.before
19725525 409960 392332 20527817 1393ac9 dart.arm.after
22576021 600376 1782824 24959221 17cd8f5 dart.x64.before
22574821 600376 1782824 24958021 17cd445 dart.x64.after
Change-Id: I68f5cd5ad452044df8bfebd160910496036a3e6b
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/101745
Commit-Queue: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
There were rare crashes
../../runtime/vm/compiler/call_specializer.cc: 1393: error: expected: call->MatchesCoreName(Symbols::_simpleInstanceOf())
version=2.3.0-edge.796ebc6069bde3a59475a9b45075f49e50b0cc34 (Thu May 9 09:38:28 2019 -0700) on "linux_x64"
thread=82540, isolate=vm-service(0x558748c5d200)
pc 0x000055874635dd6c fp 0x00007fa94dd3bf30 dart::Profiler::DumpStackTrace(void*)
pc 0x0000558745f90332 fp 0x00007fa94dd3c010 dart::Assert::Fail(char const*, ...)
pc 0x000055874653909b fp 0x00007fa94dd3c0a0 dart::CallSpecializer::ReplaceWithInstanceOf(dart::InstanceCallInstr*)
pc 0x0000558746496ed6 fp 0x00007fa94dd3c0f0 dart::FlowGraphVisitor::VisitBlocks()
when running
tools/test.py --repeat 5000 -n dartkb-mixed-linux-debug-x64 language_2/null_test
at the rate ~1-3 crashes per 30,000 test cases.
The problem is that in function
bool InstanceCallInstr::MatchesCoreName(const String& name) {
return function_name().raw() == Library::PrivateCoreLibName(name).raw();
}
'function_name().raw()' is evaluated before PrivateCoreLibName(name) is called
and saved in a temporary (register). PrivateCoreLibName may trigger GC and
relocate objects. In such case, 'PrivateCoreLibName(name).raw()' results in a
moved object, which is compared to a stale object address.
This CL fixes InstanceCallInstr::MatchesCoreName and other similar places
by introducing Library::IsPrivateCoreLibName, which is also a little bit more
efficient as it avoids extra symbol table lookup.
Change-Id: I4dc91c586b0c595a3e85d6da13b98fc2248fb8fd
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/102120
Commit-Queue: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Aart Bik <ajcbik@google.com>
While adding support for new compact bytecode instructions,
VM also keeps support for old bytecode instructions to preserve
backwards compatibility and allow soft transition.
This change is separate from bytecode generator changes in order
to test VM with old bytecode generator.
Corresponding bytecode generator changes:
https://dart-review.googlesource.com/c/sdk/+/99400
Change-Id: Icf5ceee7d51f27ffe3f79d0eae81e0ddc0a7e855
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/101062
Reviewed-by: Ryan Macnak <rmacnak@google.com>
It already is a static method in implementation, so this way
we don't duplicate the work of looking up the library. In addition,
it checks has_pragma, so we don't need to check before calling it.
Change-Id: I88281cf22615e30157c842c570bd0f6b499a6a3d
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/101821
Commit-Queue: Teagan Strickland <sstrickl@google.com>
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
Forward jumps in bytecode usually go to the 'else' part (skipping 'then')
or exit loops. In order to generate Branch instructions in flow graph
which would represent ifs and loop conditions more naturally the first ('true')
successor of a BranchInstr should correspond to fallthrough in bytecode, and not the
target of a Jump bytecode instructions.
This results in a slightly better register allocation.
Golem shows overall improvement +0.4620% on x64 and +0.8395% on arm64.
Issue: https://github.com/dart-lang/sdk/issues/36429
Change-Id: I1ea581ef9f5998eb960914680d0e867c57639f1f
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/101361
Reviewed-by: Martin Kustermann <kustermann@google.com>
Commit-Queue: Alexander Markov <alexmarkov@google.com>
'is' checks which are represented as _simpleInstanceOf() calls
are sometimes recognized by Token::kIS and not by a target of a call.
Before: TypeChecks.IsCheck(RunTime): 279.63057234726693 us.
After: TypeChecks.IsCheck(RunTime): 200.77956524794217 us.
(best of 5 runs)
Issue: https://github.com/dart-lang/sdk/issues/36429
Change-Id: I5a22222c59c33d2fa8262383a597c2c013ab61b6
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/101322
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
Commit-Queue: Alexander Markov <alexmarkov@google.com>
When provided, the opcode is implicitly inserted before each provided
MatchCode, which simplifies test cases that contain kMoveGlob or
kMoveParallelMoves before all other types of entries.
Change-Id: I51119fb66176cb54e1588687aa53b8d4ce80d9b5
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/101290
Reviewed-by: Martin Kustermann <kustermann@google.com>
Commit-Queue: Teagan Strickland <sstrickl@google.com>
Previously when computing aliasing during load propagation we only
propagated aliasing from X[C'|S'] to *[C''|S''] for S' < S''.
The code incorrectly assumed that this propagation is symmetric (which
it is not) and expected that *[C|S] would be cross aliased with X[C'|S']
when *[C|S] is visited.
Now we split symmetric and non-symmetric parts of the aliasing
computation: X[C'|S'] propagates aliasing to X[C''|S''] for S' < S'' -
which is symmetric. Separately if X is an aliased instance we propagate
aliasing from X[C'|S'] to *[C|S] for all sizes S.
Change-Id: Iccd4c73a18ffd3f60fd179df43a6b117e0844b64
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/101281
Commit-Queue: Vyacheslav Egorov <vegorov@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
Daco Harkes