This fixes a huge performance regression (introduced in [0]) of a
synthetic microbenchmark - MicroClosureCreateTearoffClassSecondTime -
that happens to have a gigantic function with a large number of call
sites in them.
All of those call sites stay monomorphic but all of them will have to
go to runtime for miss handler because the target code was disabled (the
unoptimized code got disabled, since optimized target code was
installed). The particular target function in question is
`_GrowableArray.add()`.
The cause for the regression is that the miss handler will lookup the
ICData based on linearly walking PcDescriptors to find deopt-id which it
uses to get ICData from function's ic-data array.
Interestingly enough it only regresses ia32 but not x64 (the only two
architectures this benchmark is run on). The reason for that is that
x64 uses exactness tracking and therefore doesn't switch to
monomorphic calls.
This CL:
If a monomorphic call site had a miss because the cached old target
was disabled, we go to runtime to find the real target. Another thread
may have transitioned the call site already from monomorphic to
polymorphic.
When resolving the target function we take advantage of a back reference
from the `ICData::entries()` back to the `ICData` to avoid the O(N)
lookups.
[0] https://dart-review.googlesource.com/c/sdk/+/206373
TEST=Existing test suite.
Change-Id: Ia066a846f5f50e5a575f0734c9453d2368f80780
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/207133
Commit-Queue: Martin Kustermann <kustermann@google.com>
Reviewed-by: Alexander Markov <alexmarkov@google.com>
Previously, the code in the TypeCheck runtime entry assumed that if a
lazily specialized type testing stub (TTS) returned a false negative in
JIT mode, that it should always be regenerated and that regeneration
would always result in different code. In AOT mode, false negatives
instead always cause the stub to go to runtime, even if that false
negative had been seen before, because the assumption is that false
negatives shouldn't happen when the whole class hierarchy is known at
compile time.
However, even in the current implementation of optimized TTSes, there
are cases where this assumption is false. For example, the code
generated by BuildOptimizedSubclassRangeCheckWithTypeArguments allows
for provided type arguments to be type parameters. When this happens,
the type parameter is instantiated at runtime using the instantiator or
function type arguments, and the instantiated type parameter must be
identical to the result retrieved from the type arguments of the
instance. That means that if the instantiated type parameter is not the
same type, but a supertype, of the result, then a false negative is
generated.
This CL changes TypeCheck's handling of false negatives from lazily
specialized TTSes as follows: in JIT, if the regenerated stub is the
same as the old stub, or in AOT, a false negative causes the same fall
back to SubtypeTestCaches as unoptimized stubs. This way, further
checks with the same false negative will be caught via the STC before
going to runtime, assuming the STC hasn't already filled up with false
negatives.
Currently, we only generate false negatives for reasons that will not
change when respecialization occurs due to additions to the hierarchy,
so we do not need to clear affected STCs during respecialization.
However, the previous approach to resetting STCs on reload (in
CallSiteResetter::ResetCaches) is insufficient, since there may be
caches containing reloaded types in non-reloaded code (like the TTS
invoker stub created by the TTS testing framework). Instead, clear
all caches on reload using the same ObjectVisitor as deoptimizing type
testing stubs.
Since we now have to check instruction equality to determine whether
to add to the STC, we now only replace the existing stub if the
instructions are different. This makes it easier to test whether a
TTS invocation on a false positive caused respecialization or not.
This CL also reworks the testing framework for type testing stubs,
1) creating a test case object that stores the particulars of a given
invocation, including expectations, and 2) moving most checks and
access to appropriate data structures into a state object that handles
setup and performing invocations given test cases.
TEST=vm/cc/TTS_Partial
Cq-Include-Trybots: luci.dart.try:vm-kernel-linux-release-x64-try,vm-kernel-precomp-linux-release-x64-try,vm-kernel-precomp-nnbd-linux-release-x64-try,vm-kernel-nnbd-linux-release-x64-try,vm-kernel-tsan-linux-release-x64-try,vm-kernel-linux-product-x64-try,vm-kernel-precomp-linux-product-x64-try,vm-kernel-linux-release-simarm-try,vm-kernel-linux-release-simarm64-try
Change-Id: I139608c5a0f2442a85a1cf39d1c04104db7a5593
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/208653
Commit-Queue: Tess Strickland <sstrickl@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
Reviewed-by: Alexander Markov <alexmarkov@google.com>
Revert the changes in 0920ac88 to use Closure::signature() and
remove the previously existing uses.
Refactor the majority of Object_haveSameRuntimeType out into a helper
function, which lifts the various Bool::Get()s out of the return
statements for easier reading. Also delay retrieving the class out of
the instance until after checking for closure, since we can check the
retrieved cids against kClosureCid instead.
Also move ASSERTs from debug printing of SubtypeTestCache entries
to UpdateTypeTestCache, to ensure they are checked when
--trace-type-checks is not enabled.
TEST=Refactoring, so existing tests. Checked manually on a separate
CL which triggered a case where Closure::signature() was
erroneously returning null due to the direct trip through ptrs.
Change-Id: Iff9b5c8e3f6426cc7deb5ce73c02606597d99d5b
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/207124
Commit-Queue: Tess Strickland <sstrickl@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
Closures created via Closure::New() set the delayed type arguments to
the null type argument vector for non-generic closure functions and to
the empty type argument vector for generic closure functions. However,
closures created via the method extractor stub would always have the
empty type argument vector for delayed type arguments. Make the two
consistent by creating two method extractor stubs, one for generic
extracted methods and one for non-generic ones.
With this change, Closure::IsGeneric simply compares the delayed type
arguments against the empty type argument vector. This change also makes
method extraction for non-generic functions slightly faster, since the
delayed type arguments field is not changed post-allocation.
The old version of Closure::IsGeneric was the only use of
Closure::NumTypeParameters, so remove the latter method.
Related cleanups:
* Use similar checks for Class::IsGeneric and FunctionType::IsGeneric,
since they only have non-null type_parameters fields when generic.
For Class::IsGeneric, this avoids having to fetch the current thread
if the class declaration has been loaded.
* Refactor methods that read packed fields in FunctionType into two
methods, a static method that takes a FunctionTypePtr and an instance
method that calls the static method. Replace the methods in Function
that also directly read FunctionType packed fields with delegations to
the static methods.
* Rework Closure_equals to avoid unneeded handle allocation in certain
cases for non-equal closures.
* Replace cases that allocates a handle for the closure function only
for retrieving the signature with calls to Closure::signature().
TEST=language_2/closure/tearoff_dynamic_test
language_2/regress/regress45890_test
Cq-Include-Trybots: luci.dart.try:vm-kernel-linux-debug-x64-try,vm-kernel-linux-debug-x64c-try,vm-kernel-nnbd-linux-debug-x64-try,vm-kernel-precomp-linux-debug-x64-try,vm-kernel-precomp-linux-debug-x64c-try,vm-kernel-reload-rollback-linux-debug-x64-try,vm-kernel-reload-linux-debug-x64-try,vm-kernel-linux-release-simarm-try,vm-kernel-linux-release-simarm64-try,vm-kernel-nnbd-linux-release-simarm-try,vm-kernel-nnbd-linux-release-simarm64-try,vm-kernel-precomp-linux-release-simarm-try,vm-kernel-precomp-linux-release-simarm64-try,vm-kernel-precomp-nnbd-linux-release-simarm64-try,vm-kernel-linux-debug-simarm64c-try,vm-kernel-precomp-linux-debug-simarm_x64-try,vm-kernel-precomp-nnbd-linux-debug-simarm_x64-try
Change-Id: I29c8859c3350ed7b3f1a8f71d82a4393496b7c2b
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/206820
Commit-Queue: Tess Strickland <sstrickl@google.com>
Reviewed-by: Alexander Markov <alexmarkov@google.com>
MSVC(unlike clang) seems to have a problem with such a use:
```
ninja -t msvc -e environment.x64 --
"C:\src\depot_tools\win_toolchain\vs_files\20d5f2553f\VC\Tools\MSVC\14.26.28801\bin\Hostx64\x64/cl.exe"
/nologo /showIncludes /FC
@obj/runtime/vm/libdart_vm_jit.runtime_entry.obj.rsp /c
../../runtime/vm/runtime_entry.cc
/Foobj/runtime/vm/libdart_vm_jit.runtime_entry.obj
/Fdobj/runtime/vm/libdart_vm_jit_cc.pdb
C:\src\dart-sdk\sdk\runtime\vm\runtime_entry.cc(3683): error C2121: '#':
invalid character: possibly the result of a macro expansion
C:\src\dart-sdk\sdk\runtime\vm\runtime_entry.cc(3683): error C2059:
syntax error: 'if'
C:\src\dart-sdk\sdk\runtime\vm\runtime_entry.cc(3693): error C2121: '#':
invalid character: possibly the result of a macro expansion
C:\src\dart-sdk\sdk\runtime\vm\runtime_entry.cc(3693): error C2059:
syntax error: 'if'
```
TEST=build sdk on Windows with msvc(rather than clang)
Change-Id: I520b32bf38bc7880f45d93223779f10cbde59af6
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/207041
Reviewed-by: Martin Kustermann <kustermann@google.com>
Commit-Queue: Alexander Aprelev <aam@google.com>
If two mutators concurrently execute a switchable call that targets a
code object which has been disabled, they both go to runtime trying to
fix that switchable call site's object pool.
This Cl ensures we use the same logic as for other miss handlers to
ensure we properly guard against concurrent accesses.
The attached regression test will trigger a segfault - though only
with low probability.
The changes to first land is: Avoid assuming the target can be called
directly. Instead always go through IC Stub when returning from miss
handler. We do that since a call site might not provide ARGS_DESC but
the target might need it.
Fixes https://github.com/dart-lang/sdk/issues/46539
Fixes https://github.com/dart-lang/sdk/issues/46553
TEST=vm/dart{,_2}/isolates/regress_46539_test
Change-Id: I18018fa286173d905c52b6e058a6c87b544ff18e
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/206373
Commit-Queue: Martin Kustermann <kustermann@google.com>
Reviewed-by: Slava Egorov <vegorov@google.com>
Reviewed-by: Ryan Macnak <rmacnak@google.com>
This relands https://dart-review.googlesource.com/c/sdk/+/205633
but without renaming TARGET_OS_IPHONE to DART_TARGET_OS_IPHONE.
It also changes uses of TARGET_OS_IOS to
DART_TARGET_OS_MACOS_IOS to be consistent with the rest of the
VM.
TargetConditionals.h for XCode 13 defines several
TARGET_OS_* preprocessor symbols that confuse the
Dart build. There is probably a more targeted fix
for this, but renaming the symbols that Dart uses
will also prevent this problem if more symbols
are added to the platform headers in the future.
See: https://github.com/dart-lang/sdk/issues/46499
TEST=It builds.
Change-Id: Ie775c19dd23cfdf5f65e5ebc6ee4ec3a561676fa
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/205860
Commit-Queue: Zach Anderson <zra@google.com>
Reviewed-by: Alexander Aprelev <aam@google.com>
If two mutators concurrently execute a switchable call that targets a
code object which has been disabled, they both go to runtime trying to
fix that switchable call site's object pool.
This Cl ensures we use the same logic as for other miss handlers to
ensure we properly guard against concurrent accesses.
The attached regression test will trigger a segfault - though only
with low probability.
Fixes https://github.com/dart-lang/sdk/issues/46539
TEST=vm/dart{,_2}/isolates/regress_46539_test
Change-Id: I91d84d25d74fb742ea992016a581b118345dd404
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/205649
Commit-Queue: Martin Kustermann <kustermann@google.com>
Reviewed-by: Slava Egorov <vegorov@google.com>
TargetConditionals.h for XCode 13 defines several
TARGET_OS_* preprocessor symbols that confuse the
Dart build. There is probably a more targeted fix
for this, but renaming the symbols that Dart uses
will also prevent this problem if more symbols
are added to the platform headers in the future.
See: https://github.com/dart-lang/sdk/issues/46499
TEST=It builds.
Change-Id: I3b33a03b4a9a14b76d55fe12f8cdefec4b3c3664
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/205633
Commit-Queue: Zach Anderson <zra@google.com>
Reviewed-by: Siva Annamalai <asiva@google.com>
With this change, allocation instructions can now be direct inputs to
other allocation instructions, so the redundancy eliminator is extended
to handle this possibility.
Methods for working with instruction input-related slots are added to
subclasses of AllocationInstr, so that the redundancy eliminator can be
written more generically in places, instead of needing to add cases for
new allocation instructions and/or instruction inputs.
Code size different in Flutter gallery (release-sizeopt):
* ARM7: Total -0.30%, instructions -0.41%
* ARM8: Total -0.31%, instructions -0.47%
TEST=Current test suite.
Cq-Include-Trybots: luci.dart.try:vm-kernel-precomp-linux-debug-simarm64c-try,vm-kernel-precomp-linux-debug-simarm_x64-try,vm-kernel-precomp-linux-debug-x64-try,vm-kernel-precomp-linux-debug-x64c-try,vm-kernel-precomp-linux-product-x64-try,vm-kernel-precomp-linux-release-x64-try,vm-kernel-precomp-linux-release-simarm64-try,vm-kernel-precomp-linux-release-simarm-try,vm-kernel-linux-debug-ia32-try,vm-kernel-linux-debug-x64-try,vm-kernel-linux-product-x64-try,vm-kernel-linux-release-simarm64-try,vm-kernel-linux-release-simarm-try,vm-kernel-linux-release-x64-try,vm-kernel-linux-debug-x64c-try
Change-Id: Idc1aa2a1cb8c0c62f0bcb64aee89a7525dd3d1e1
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/198406
Reviewed-by: Martin Kustermann <kustermann@google.com>
Reviewed-by: Slava Egorov <vegorov@google.com>
The same signature (canonical function type) may be shared between different closure functions.
Clean up asserts verifying that members are canonical when updating type test cache.
TEST=existing ones
Change-Id: Ic206409e56d34dfd02d6705af93065fd3b83b8fa
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/199561
Commit-Queue: Régis Crelier <regis@google.com>
Reviewed-by: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Ryan Macnak <rmacnak@google.com>
Now AllocateClosureInstr takes the closure function as a Value* instead
of as a const Function&, a new kFunctionReg is added to
AllocateClosureABI, and the AllocateClosure stub sets the closure
function internally instead of initializing it to null and requiring it
to be set post-allocation.
A convenience method known_function() is added to AllocateClosureInstr
which returns either the closure function, if known at compile time,
or null otherwise.
TEST=More refactorings, so just passing current tests.
Cq-Include-Trybots: luci.dart.try:vm-kernel-precomp-linux-debug-simarm64c-try,vm-kernel-precomp-linux-debug-simarm_x64-try,vm-kernel-precomp-linux-debug-x64-try,vm-kernel-precomp-linux-debug-x64c-try,vm-kernel-precomp-linux-product-x64-try,vm-kernel-precomp-linux-release-x64-try,vm-kernel-precomp-linux-release-simarm64-try,vm-kernel-precomp-linux-release-simarm-try,vm-kernel-linux-debug-ia32-try,vm-kernel-linux-debug-x64-try,vm-kernel-linux-product-x64-try,vm-kernel-linux-release-simarm64-try,vm-kernel-linux-release-simarm-try,vm-kernel-linux-release-x64-try,vm-kernel-linux-debug-x64c-try
Change-Id: If843b401e0f282b191dd60b1c6c73c01f1d5bc70
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/198285
Commit-Queue: Tess Strickland <sstrickl@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
Reviewed-by: Alexander Markov <alexmarkov@google.com>
Also create a subclass of AllocationInstr, AllocateClosureInstr,
which is used when allocating closures.
Followup CLs add inputs to this instruction and to the AllocateClosure
stub, starting with the closure function. Adding these inputs allows
the related field to be set within the stub, instead of using
StoreInstanceField manually at each closure allocation point.
Since this CL only adds the initial stub/instruction implementation,
the overhead on snapshots is minimal: just the addition of the new
stub to each isolate.
-----
This CL also adds virtual and non-virtual methods to assembler_base.h
revolving around field loads and stores and attempted inline object
allocation, to ensure all architectures have these methods.
It also adds LoadFromSlot/StoreToSlot/StoreToSlotNoBarrier, which
appropriately calls one of the other methods based on whether the slot
is compressed or not and whether it is a boxed or unboxed field.
With these additions, the AllocateClosure stub generator can be defined
in an architecture-independent way.
TEST=Basically a refactoring, so check using current test suites.
Cq-Include-Trybots: luci.dart.try:vm-kernel-precomp-linux-debug-simarm64c-try,vm-kernel-precomp-linux-debug-simarm_x64-try,vm-kernel-precomp-linux-debug-x64-try,vm-kernel-precomp-linux-debug-x64c-try,vm-kernel-precomp-linux-product-x64-try,vm-kernel-precomp-linux-release-x64-try,vm-kernel-precomp-linux-release-simarm64-try,vm-kernel-precomp-linux-release-simarm-try,vm-kernel-linux-debug-ia32-try,vm-kernel-linux-debug-x64-try,vm-kernel-linux-product-x64-try,vm-kernel-linux-release-simarm64-try,vm-kernel-linux-release-simarm-try,vm-kernel-linux-release-x64-try,vm-kernel-linux-debug-x64c-try
Change-Id: I71f5691307679f8d5e3604007699de4706f86eb8
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/198284
Commit-Queue: Tess Strickland <sstrickl@google.com>
Reviewed-by: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
The "RunWithMutatorsStopped" wrapper is used in places where the invoked
callback can cause deoptimization of code. It should therefore ensure
it's running with a "deopt safpoint operation scope" to ensure mutators
are stopped at well-defined places that allow lazy-deopt.
Issue https://github.com/dart-lang/sdk/issues/45213
TEST=Existing code base, will add "fuzzer"-like test to nightly/weekend builders.
Change-Id: Icb9a4183c13fab0f084e481c10dfc56a0308126a
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/197162
Commit-Queue: Martin Kustermann <kustermann@google.com>
Reviewed-by: Ryan Macnak <rmacnak@google.com>
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
The optimizing compiler can insert box instructions at various places
during the optimization pipeline. There is no guarantee that a box
instruction will get a proper (deopt-id, deopt-env) assigned to it - in
many places it will not.
Furthermore StoreInstanceField's into "unboxed fields" might cause
box allocations to happen. Those stores might also not have deopt
information associated with them.
In general we require stackmaps when allocation slow paths go to runtime
and cause GC. If GC throws OOM we theoretically would also need deopt
information in order to populate correct catch-entry state. Though OOM
is uncommon and the VM could decide to ignore the top-frame - if it's
missing deopt information - since box allocations aren't something
users are in control of (as opposed to user-allocated objects inside
try/catch).
=> This CL makes box allocations go to a runtime entry that doesn't
support lazy-deopt. While being in those runtime entries the mutators
will also not participate in "deopt safepoint operation" requests.
Issue https://github.com/dart-lang/sdk/issues/45213
TEST=Existing test suite.
Change-Id: I1b61f77e3166da82efad08bb49bc1756576d220c
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/196928
Commit-Queue: Martin Kustermann <kustermann@google.com>
Reviewed-by: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
This extends the existing safepoint operation mechanism by allowing to
perform two different operations:
* "gc safepoint operations": All mutators are stopped at places where
it's safe to GC. It therefore requires stackmaps to be available for
all optimized mutator frames.
* "deopt safepoint operations": All mutators are stopped at places
where it's safe to GC, but also safe to lazy-deopt mutator frames.
It therefore requires deopt-id/deopt-info to be available for all
optimized mutator frames.
Mutators can be asked to block for any of those two safepoint operations.
If a mutator is at a place where its safe to GC it will respond to "gc
safepoint operations" requests, if a mutator is additionally at a place
where it's also safe to lazy-deopt it will respond to "deopt safepoint
operation" requests.
Depending on how the runtime was entered (which is tracked via the
[Thread::runtime_call_deopt_ability_] value) - the mutator might
participate in both or only in gc safepoint operations.
During the start of a "deopt safepoint operation", the safepoint handler
will request all threads to stop at a "deopt safepoint". Some threads
might first want to initiate their own "gc safepoint operation"
(e.g. due to allocation failure) before they reach a "deopt safepoint".
We do allow this by letting the safepoint handler own a "deopt safepoint
operation" but still participate in other thread's "gc safepoint
operation" requests until all mutators are checked into places where
it's safe to lazy-deopt at which point the "deopt safepoint operation"
also owns a "gc safepoint operation".
In order to facilitate this, the Thread's safepoint_state will be
extended to consist of the following bits:
* AtSafepoint
* SafepointRequested
* AtDeoptSafepoint
* DeoptSafepointRequested
* BlockedForSafepoint
Issue https://github.com/dart-lang/sdk/issues/45213
TEST=vm/cc/SafepointOperation_*
Change-Id: Icdc2827718f6780818f99b829a5e806d6bb5b130
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/196927
Commit-Queue: Martin Kustermann <kustermann@google.com>
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
Reviewed-by: Ryan Macnak <rmacnak@google.com>
Creates HasParent() as a predicate for Function objects with a non-null
parent function and makes IsLocalFunction() specific to local closures
(i.e. explicit closures with non-null parent functions).
Creates HasSavedArgumentsDescriptor() as a predicate to determine when
saved_args_desc() can be used, to avoid the caller having to do explicit
predicate checks for each dispatcher type that contains one.
Also simplifies Function::PrintName, creates different suffixes for a
SyncGenClosureMaker and its associateed SyncGenClosure, adds suffixes
for skipping more than one generated body when printing the parent name,
and changes ArgumentsDescriptor::PrintTo to match the compact syntax
Function::PrintName used for signifying the arguments descriptor for
certain dispatchers.
TEST=pkg/vm_snapshot_analysis/test/instruction_sizes_test
Cq-Include-Trybots: luci.dart.try:app-kernel-linux-release-x64-try,vm-kernel-precomp-linux-release-x64-try,pkg-linux-release-try
Change-Id: I0ac1226d1ffe1d81743b5c6788da170d5a4010c5
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/191560
Commit-Queue: Tess Strickland <sstrickl@google.com>
Reviewed-by: Daco Harkes <dacoharkes@google.com>
CodeBreakpoint class tracks patching of the call targets for the purpose of routing the call to the debugger runtime.
With this CL CodeBreakpoint instance is managed by GroupDebugger, debugger class owned by IsolateGroup, and can be associated with multiple BreakpointLocation, one location per isolate that set a breakpoint at particular code location.
TEST=observatory/tests/service/break_on_function_child_isolate_test
Issue https://github.com/dart-lang/sdk/issues/36097
Change-Id: I3b3d0974a0fb9668742e5c1e2a2ce13b84b1d630
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/190732
Commit-Queue: Alexander Aprelev <aam@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
This CL adds code to detect situations when we hit
null error through GDT call with non-null receiver.
We achieve this by making receiver's cid part of the
GDT calling convention and checking this cid
in runtime entry responsible for throwing the null
error.
This CL is an attempt to narrow down b/179632636
and collapse various impossible crash reports
into a single native crash cluster by crashing VM
instead of throwing a null error.
TEST=tested manually
Cq-Include-Trybots: luci.dart.try:vm-kernel-precomp-linux-release-x64-try,vm-kernel-precomp-linux-release-simarm-try,vm-kernel-precomp-linux-release-simarm64-try
Change-Id: If2ed4646c4c0f403016266e4e83e296a7b234cb5
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/191141
Auto-Submit: Vyacheslav Egorov <vegorov@google.com>
Commit-Queue: Vyacheslav Egorov <vegorov@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
The deoptimization environment of AssertAssignable should not have it's
inputs in it, since those get poped before the call to the
TypeTestingStub.
A lazy deopt should therefore continue in unoptimized code after the TTS
call, without the inputs.
This was discovered by running many lightweight isolates concurrently on
the JIT.
To simplify a regression test we'll add new VM flags that can trigger
deoptimization on any runtime call (optionally filtered by name of
runtime call).
We can use this mechanism to possibly discover more bugs of the same
kind.
Closes https://github.com/dart-lang/sdk/issues/45207
Issue https://github.com/dart-lang/sdk/issues/36097
TEST=runtime/tests/vm/dart{,_2}/regress_45207_test
Change-Id: Ib713fa8be5914eaf4c7395d2e1d8dd88714f1ce2
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/189202
Commit-Queue: Martin Kustermann <kustermann@google.com>
Reviewed-by: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
To avoid a O(n) traversal of isolates in order to find pending deopts,
we'll move the pending deopt state to the Thread. It's really a property
of a thread: when the thread's stack gets unwound - via return or
exceptional flow - we remove pending deopts.
This CL also encapsulates the pending deopt logic into a [PendingDeopts]
class - thereby avoiding logic around manipulating this array being
spread around. It simplifies the already very big [Isolate] class.
TEST=Refactoring of existing code.
Issue https://github.com/dart-lang/sdk/issues/36097
Closes https://github.com/dart-lang/sdk/issues/45168
Change-Id: I5a9b18c577f21a0f2a88332b1bc26641ec39d2af
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/188521
Commit-Queue: Martin Kustermann <kustermann@google.com>
Reviewed-by: Alexander Aprelev <aam@google.com>
The background compiler no longer has bailouts due to something that
cannot be done on BG compiler and has to be done on mutator (was
originally introduced in [0])
All optimizing compilations can happen on the BG thread. Class
finalization and other things no longer cause a bailout on BG
compiler.
If e.g. invalidated field guards cause a compilation to be discarded, it
can be retried on BG compiler.
[0] https://dart-review.googlesource.com/c/sdk/+/54886
Closes https://github.com/dart-lang/sdk/issues/45136
TEST=Fixes flaky hits of an assertion.
Change-Id: I50313dbedf96b8dec205acdb8c1ed5731d00433b
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/187901
Reviewed-by: Alexander Aprelev <aam@google.com>
Commit-Queue: Martin Kustermann <kustermann@google.com>
Whether a given isolate allows reloading is really a property of the
thread not the isolate (where it was before) or isolate group (where
it was moved to), so this CL moves it to Thread.
As the thread's stack is going down and up, NoReloadScope's might be
entered or exited, just as with our many other scopes, such as
NoOOBMessageScope/...
Longer term we might need to find a better mechanism to deal with
different OOB messages. For example: Reload as well as OOB messages
are disabled during static field initialization. Though that means *all*
OOB messages cannot reach the isolate during initialization of a static
field (which can do arbitrary amount of work).
This is part of making hot-reload work with isolate groups.
Issue https://github.com/dart-lang/sdk/issues/36097
TEST=Refactoring of existing code.
Change-Id: I9e80c6d4a184c54c0373fef13af3264cc3f27ece
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/187002
Reviewed-by: Alexander Aprelev <aam@google.com>
Reviewed-by: Ryan Macnak <rmacnak@google.com>
Commit-Queue: Martin Kustermann <kustermann@google.com>
This CL moves the [ProgramReloadContext] from [Isolate] to
[IsolateGroup] - since the program is shared across isolates. Amongst
other things, it also deoptimizes stacks of all isolate mutators.
There is still future work to do for hot-reloading multiple isolates
within a group, this is only the first change.
Issue https://github.com/dart-lang/sdk/issues/36097
TEST=Refactoring of existing implementation.
Change-Id: Ic394de265e3922f84657de2da25a46179b624647
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/183693
Commit-Queue: Martin Kustermann <kustermann@google.com>
Reviewed-by: Alexander Aprelev <aam@google.com>
The logic for background compiler can be simplified due to:
* we only have one bg compiler
* it always optimizes
* it is allocated at isolate startup and freed on isolate shutdown
* an existing check in the compiler for whether the BG compiler is
disabled is racy and therefore not helpful
* when a mutator communicates with BG compiler it will use
SafepointMutexLocker instead of MutexLocker
This CL also ecapsulates disabling the BG compiler via a
NoBackgroundCompilerScope object. It allows nested disabling and guards
against races.
Issue https://github.com/dart-lang/sdk/issues/36097
TEST=Refactoring of existing code.
Change-Id: I36ed6592c9a8b91e94181698b7de662b4ee984e9
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/183400
Commit-Queue: Martin Kustermann <kustermann@google.com>
Reviewed-by: Alexander Aprelev <aam@google.com>
The background compiler can cause finalization of classes which can
cause deoptimization of code.
Any deoptimization will iterate over all mutator threads and marks
frames on the stack to be deoptimized. This requires saving old
(fp, pc) state. This state is currently saved on the current isolate's
pending deopt array (via `Thread::Current()->isolate()->AddPendingDeopt()`)
instead of the isolate whose stack got walked.
As part of making the compiler independent of `Isolate::Current()` we
pass the isolate explicitly.
Issue https://github.com/dart-lang/sdk/issues/36097
TEST=Refactoring of existing code.
Change-Id: I9327f3b76981fc16c1879a873edf37df1cbdd8bf
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/182380
Commit-Queue: Martin Kustermann <kustermann@google.com>
Reviewed-by: Alexander Aprelev <aam@google.com>
Reviewed-by: Ryan Macnak <rmacnak@google.com>
As part of making the compiler independent of a particular isolate
(since all isolates in a group share the same JITed code) we move the
boxed field list to IsolateGroup.
Any isolate that hits the lazy compile stub will then drain the fields
to be deoptimized and performs that work.
A follow-up investigation can decide whether it's beneficial (or
hurtful) for single vs multi isolate scenarios to deoptimize the field
immediately in BG compiler (and interrupt mutator). It is possibly
disadvantagues for single isolate scenarios to do so, since it causes
more mutator interruptions.
Issue https://github.com/dart-lang/sdk/issues/36097
TEST=Refactoring of existing code.
Change-Id: I64b42ad99056b43a7264bcc30280b9ba39389a2e
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/182505
Commit-Queue: Martin Kustermann <kustermann@google.com>
Reviewed-by: Ryan Macnak <rmacnak@google.com>
Martin Kustermann