The approach works as follows:
Step 1: Kernel transform. Under the closed world assumption compute the
set of selectors dispatched dynamically, then mark all procedures that don't
match any of those selectors as 'not-dispatched-dynamically'.
Step 2: VM backend. When building IR for a function if this function was
marked as not-dispatched-dynamically then omit type checks for any parameter
that is not marked as generic-covariant-impl, as such arguments are guaranteed
to be checked on the caller side (by front-end).
# Performance Impact
+------------------------+------------+----------+-----------+--------------+
| benchmark | baseline | current | with opt | improved by |
+------------------------+------------+----------+-----------+--------------+
| stock_layout_iteration | 2366.3786 | 2724.3 | 2562.75 | -5.93% |
| stock_build_iteration | 3824.3 | 4914.8 | 4681 | -4.76% |
+------------------------+------------+----------+-----------+--------------+
* Flutter gallery Instructions size is reduced by 11% (8748720 bytes to 7846368 bytes).
Baseline is at 6196496 bytes.
# Alternative Implementations
Alternatively to annotating individual procedures, I considered annotating Program node
with a set of dynamically dispatched selectors. Decoding and passing this information
around proved to be quite cumbersome in the "streaming" world, so I opted for a simpler
approach where all individual procedures are annotated.
Change-Id: I363db6d5dd1138fe25917646313c16d0b213c3b4
Reviewed-on: https://dart-review.googlesource.com/37822
Commit-Queue: Vyacheslav Egorov <vegorov@google.com>
Reviewed-by: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Samir Jindel <sjindel@google.com>
When the type to test against is instantiated and has no type arguments
there is a high probability that we receive instances of that class or
subclasses at runtime.
This CL therefore extends the fast-path of AssertAssignable/InstanceOf
by checking whether the instance class id is within the cid ranges that
directly/indirectly implement/extend the type to test against.
Currently we have an almost depth-first preorder numbering of class ids
in AOT, but there are exceptions. So each class can have a number of
cid-ranges as subclasses / classes which implement it's interface.
This seems to improve performance of dart-aot-v2
* flutter stock build by 15+%
* DeltaBlueClosures by 10+%
and reduces code size on
* flutter gallery by -3%
Issue https://github.com/dart-lang/sdk/issues/31798
Change-Id: I07dd91589cc3fcd8c5952bdba339e2e2a459e08e
Reviewed-on: https://dart-review.googlesource.com/35620
Commit-Queue: Martin Kustermann <kustermann@google.com>
Reviewed-by: Régis Crelier <regis@google.com>
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
This reverts commit 12759870a4.
Reason for revert: Broke many bots.
Original change's description:
> [VM] Use IR for code in [CatchEntryInstr]s to populate captured exception/stacktrace variables
>
> This fixes an issue when a program got loaded via dill, a function
> with a try-catch got optimized and the exception/stacktrace variables
> got captured.
>
> Change-Id: Ia6b62f2a0986c78b90afe7fae25025ca4e5b09db
> Reviewed-on: https://dart-review.googlesource.com/35182
> Commit-Queue: Martin Kustermann <kustermann@google.com>
> Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
TBR=vegorov@google.com,kustermann@google.com
Change-Id: I2ec7b4bea401e6d5665f6e789eb8e241b99db7b3
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://dart-review.googlesource.com/36200
Reviewed-by: Martin Kustermann <kustermann@google.com>
Commit-Queue: Martin Kustermann <kustermann@google.com>
This fixes an issue when a program got loaded via dill, a function
with a try-catch got optimized and the exception/stacktrace variables
got captured.
Change-Id: Ia6b62f2a0986c78b90afe7fae25025ca4e5b09db
Reviewed-on: https://dart-review.googlesource.com/35182
Commit-Queue: Martin Kustermann <kustermann@google.com>
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
This CL extends support of mixed operations in strong-mode AOT
from double/smi to double/int (if underlying code generator supports
conversions from int64 to double).
* MintToDouble instruction is renamed to Int64ToDouble.
* Code generation of Int64ToDouble is implemented on x64.
Issue: https://github.com/dart-lang/sdk/issues/30480
Change-Id: I3142302e80c2785be71d7a8baff42d40f0b5b1b8
Reviewed-on: https://dart-review.googlesource.com/35382
Commit-Queue: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
Update status files.
This cl includes implementing these two features on all platforms:
1) Support calling generic functions via DartEntry::InvokeFunction().
2) Support native generic functions. These are currently allowed, but type
arguments are ignored, and therefore not accessible from the C++ side.
Change-Id: Id39e8ca46c2ba1ba3d46946c16712a8572ff64ea
Reviewed-on: https://dart-review.googlesource.com/34023
Commit-Queue: Régis Crelier <regis@google.com>
Reviewed-by: Siva Annamalai <asiva@google.com>
Currently we only enable the strengthening of AssertAssignableInstr's
when having asserts enabled (which happens in checked mode). But in
strong mode we use these instructions as well, so we need to enable the
optimization as well.
Change-Id: If889d0e14b590e77377c9828e80a00e4852fed42
Reviewed-on: https://dart-review.googlesource.com/31944
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
Commit-Queue: Vyacheslav Egorov <vegorov@google.com>
If we know that receiver is a subclass of a certain type and that this
type has accessor get:m then call o.m(...) is guaranteed to be an
invocation through a getter. Such invocations are executed most
efficiently when expanded into o.get:m().call(...).
Source based pipeline handles this case (at least for invocations on
`this`) directly in the parser, but Kernel based graph builder does not
have this sort of special case.
Instead of teaching Kernel flow graph builder to specially handle
invocations on `this` we teach AOT call specializer to specially handle
all invocations where receiver is known to be a subclass of certain
class. Such optimization is more generic and handles things that
previously were not handled by the optimization.
AOT compiler also has heuristics for injecting field dispatchers into
classes but these heuristics only handle fields of function type and
don't work for fields like `Function f` or `var f`.
Improves ParserCombinators benchmark by 4x.
This relands eea2c168f9 with a fix.
Change-Id: I13a41544c737b980efd431e31e4d15ad31da853e
Reviewed-on: https://dart-review.googlesource.com/30455
Reviewed-by: Alexander Markov <alexmarkov@google.com>
Commit-Queue: Vyacheslav Egorov <vegorov@google.com>
Strong mode types are queried from:
* Return types of interface targets of instance calls.
* Return types of static calls.
* Types of instance and static fields.
* Types of local variables (for initial types of phis).
The strong mode types are used to specialize int and double operations
and optimize 'is' checks.
https://github.com/dart-lang/sdk/issues/30480
Change-Id: I2302509f396eacfcab8ed41e3f50e8c74dd662fd
Reviewed-on: https://dart-review.googlesource.com/30386
Commit-Queue: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
* Fixes kernel flow graph builder to use instantiated type argumenst
for CreateArray
* Fixes prologue builder to mask the positon of named arguments in the
arguments descriptor array
* Add a SmiBitAndTOS instruction to DBC
* Implement missing AssertSubtype DBC instruction
* Ensure we don't use field guards in DBC mode
* Remove incorrect assertions in LoadIndexedUnsafe/StoreIndexedUnsafe
* Save argdesc_ in DBC simulator when calling the runtime to optimize
a function (which can call re-entrant to Simulator via const
evalulation)
Issue https://dart-review.googlesource.com/c/sdk/+/29581
Change-Id: Ia14b657db66f90643822b6986ec7f75f746ab0d8
Reviewed-on: https://dart-review.googlesource.com/30340
Commit-Queue: Martin Kustermann <kustermann@google.com>
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
This reverts commit eea2c168f9.
Reason for revert: standalone_2/io/test_runner_test times out
python tools/test.py -m release -c precompiler -r dart_precompiled -a simarm64 --use-blobs standalone_2/io/test_runner_test
Original change's description:
> [vm] Detect and expand calls through getters in AOT call specializer.
>
> If we know that receiver is a subclass of a certain type and that this
> type has accessor get:m then call o.m(...) is guaranteed to be an
> invocation through a getter. Such invocations are executed most
> efficiently when expanded into o.get:m().call(...).
>
> Source based pipeline handles this case (at least for invocations on
> `this`) directly in the parser, but Kernel based graph builder does not
> have this sort of special case.
>
> Instead of teaching Kernel flow graph builder to specially handle
> invocations on `this` we teach AOT call specializer to specially handle
> all invocations where receiver is known to be a subclass of certain
> class. Such optimization is more generic and handles things that
> previously were not handled by the optimization.
>
> AOT compiler also has heuristics for injecting field dispatchers into
> classes but these heuristics only handle fields of function type and
> don't work for fields like `Function f` or `var f`.
>
> Improves ParserCombinators benchmark by 4x.
>
> Bug:
> Change-Id: I365d4fb2140577d02aefbf16fcf6be1bd12413a6
> Reviewed-on: https://dart-review.googlesource.com/29840
> Commit-Queue: Vyacheslav Egorov <vegorov@google.com>
> Reviewed-by: Alexander Markov <alexmarkov@google.com>
> Reviewed-by: Martin Kustermann <kustermann@google.com>
TBR=vegorov@google.com,kustermann@google.com,alexmarkov@google.com
Change-Id: I2b891b1674dd07fa43d6414428438c41e3332391
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://dart-review.googlesource.com/30400
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
Commit-Queue: Vyacheslav Egorov <vegorov@google.com>
If we know that receiver is a subclass of a certain type and that this
type has accessor get:m then call o.m(...) is guaranteed to be an
invocation through a getter. Such invocations are executed most
efficiently when expanded into o.get:m().call(...).
Source based pipeline handles this case (at least for invocations on
`this`) directly in the parser, but Kernel based graph builder does not
have this sort of special case.
Instead of teaching Kernel flow graph builder to specially handle
invocations on `this` we teach AOT call specializer to specially handle
all invocations where receiver is known to be a subclass of certain
class. Such optimization is more generic and handles things that
previously were not handled by the optimization.
AOT compiler also has heuristics for injecting field dispatchers into
classes but these heuristics only handle fields of function type and
don't work for fields like `Function f` or `var f`.
Improves ParserCombinators benchmark by 4x.
Bug:
Change-Id: I365d4fb2140577d02aefbf16fcf6be1bd12413a6
Reviewed-on: https://dart-review.googlesource.com/29840
Commit-Queue: Vyacheslav Egorov <vegorov@google.com>
Reviewed-by: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
Apart from removing almost 1000 lines of very repetitive code, the idea here is
to change the assembler from a huge pile of arbitrary code into something that
can be used to generate tables of opcodes and their structure. Later, I'd like to
use this to make the disassembler more table driven and less arbitrary, and
perhaps build an x86 simulator in the same vein as the ARM simulator, which
would help me debug (I find the ARM simulator very useful when making low level
changes to the VM and miss its functionality on x86).
R=vegorov@google.com
Bug:
Change-Id: I1ae2c1696f88b67862843c9ac05c827a7c9b9a6e
Reviewed-on: https://dart-review.googlesource.com/25241
Commit-Queue: Erik Corry <erikcorry@google.com>
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
This CL extends RelationalOpInstr, BinaryInt64OpInstr, BinaryDoubleOpInstr,
UnaryDoubleOpInstr and unboxing instructions (UnboxInstr, UnboxIntegerInstr,
UnboxInteger32Instr, UnboxUint32Instr, UnboxInt32Instr and UnboxInt64Instr)
with non-speculative mode, when types of the inputs are not checked.
The non-speculative mode is used when generating double and int64 operations
in AOT using strong mode types.
Also, this CL disables couple of cases in --experimental-strong-mode which
do not work yet. This is a preparation for turning --experimental-strong-mode
on by default.
Issue: https://github.com/dart-lang/sdk/issues/30480
Change-Id: Ic1f89e7f6fd9592d5a92671963283c7181ce1504
Reviewed-on: https://dart-review.googlesource.com/27841
Commit-Queue: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
This allows us to avoid going into runtime even if strong mode is
enabled: type checking is handled in the []= operator itself and
_setIndexed does not need to check types and thus can be intrinsified.
This CL also removes inline type checking done in ia32 version,
not other platform had it.
# Performance Impact
Mostly noticable on Dart v2 AOT configuration where AOT compiler
did not inline []=:
Richards +26.27%
Meteor +33.63%
BuildTableInterpolation +68.98%
DeltaBlueClosures +46.43%
StarryStrings +22.08%
NavierStokes +94.81%
JsonParseCustomReviver +45.22%
DeltaBlue +56.44%
BuildTableBufferWithCodes +84.40%
JsonParseDefaultReviver +46.13%
BuildTableBuffer +125.8%
BuildTableIdiomatic +125.8%
StringBuffer +127.3%
StringIdiomatic +127.2%
JsonObjectRoundTrip +95.42%
JsonStringifyFinancialOnce +125.7%
JsonRoundTrip +96.01%
Bug:
Change-Id: Ie6d33ac55b5a0c0909ca06e0c88a60d8e0d2604e
Reviewed-on: https://dart-review.googlesource.com/26800
Commit-Queue: Vyacheslav Egorov <vegorov@google.com>
Reviewed-by: Alexander Markov <alexmarkov@google.com>
Code generation is changed to handle correctly values of
* ArgumentsDescriptor::TypeArgsLenField::mask_in_place()
* ArgumentsDescriptor::PositionalCountField::mask_in_place()
which cannot be represented as immediates on ARM/ARM64.
Closes https://github.com/dart-lang/sdk/issues/31559
Change-Id: I94bb470bf95dcdde48dca56e7128d00b95a974da
Reviewed-on: https://dart-review.googlesource.com/27081
Commit-Queue: Vyacheslav Egorov <vegorov@google.com>
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
This CL makes NativeCallInstr instructions use PushArgument()s when
calling instead of directly using a pointer to the caller-provided
arguments (for functions without optional parameters) or relying on copy
arguments prologue (for functions with optional parameters).
Issue https://github.com/dart-lang/sdk/issues/31495
Change-Id: I36d4bf1b7dd10f48c7e40b0cafe37f8698aacb71
Reviewed-on: https://dart-review.googlesource.com/24961
Commit-Queue: Martin Kustermann <kustermann@google.com>
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
This fixes some incorrect asserts that were breaking the debug bots.
The original revision is available in Patchset 1.
This reverts commit 26735519cb.
Bug:
Change-Id: Ifa599b7bff752dec4c505e10fd6db206e1abd977
Reviewed-on: https://dart-review.googlesource.com/23820
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
Commit-Queue: Samir Jindel <sjindel@google.com>
This reverts commit f13f772bb2.
Reason for revert: dartk(p)-strong debug mode is completely broken
because we convert int values non-representable as Smi to Smi, which
triggers that assert. The code that does tagging checks that
Untag(Tag(x)) == x, however for our purposes this identity does not
need to hold because we only care about lower bits and ignore the sign.
Reverting to restore green-ness of DEBUG builds.
TBR=sjindel@google.com
Bug:
Change-Id: Id436cbe000d6dec8db3469070ed531327cc82d89
Reviewed-on: https://dart-review.googlesource.com/23661
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
Several tricks here:
* When zeroing registers we can use xorl instead of xorq because the 'l'
variant will zero the top bits.
* test and 'and' instructions with immediate arguments can use 8-bit and 32 bit
variants more heavily.
* mov reg, immediate can use more compact encodings when sign-extension is not
needed.
Performance is better than +1% when measured on Dart2JS.
R=vegorov@google.com
Intel optimization manual says: "Assembly/Compiler Coding Rule 64. (H impact, M
generality) Use the 32-bit versions of instructions in 64-bit mode to reduce
code size unless the 64-bit version is necessary to access 64-bit data or
additional registers."
Bug:
Change-Id: I2a989315c45f8d8ebab719653fbfa2b18ebb77c9
Reviewed-on: https://dart-review.googlesource.com/23400
Commit-Queue: Erik Corry <erikcorry@google.com>
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
It appears like the manual flow graph building code for
MethodRecognizer::kObjectRuntimeType is wrong and was exposed by
enabling it recently.
Change-Id: Ia4be8c4bc7c3342e0cdf1cb8b270dc48b5f2fde4
Reviewed-on: https://dart-review.googlesource.com/17920
Reviewed-by: Martin Kustermann <kustermann@google.com>
Previously we would sometimes build specifically hand-crafted flow graph
code for recognized methods and sometimes let the normal flow graph
builder do it (depending on whether inlining was triggered via
[CallSiteInliner] or [CallSpecializer]).
This CL makes us use the hand-crafted flow graph building for recognized
methods in both cases.
Change-Id: I58405b16adca810fd7ccff32027b9d4a13858976
Reviewed-on: https://dart-review.googlesource.com/17445
Commit-Queue: Martin Kustermann <kustermann@google.com>
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
CloneContext is implemented as a runtime call so it is considerably slower
than AllocateContext which leads to performance issues in loops that capture
iteration variables in for loops.
Lowering CloneContext into AllocateUninitializedContext and a sequence of
explicit load&store instructions both improves the performance by avoiding
the runtime call and also makes CloneContext semantics exposed to store-load
forwarding and allocation sinking.
PutIfAbsent benchmark from issue #31113 is improved by 5x, while
ForInGeneratedLoop benchmark is improved by 8x.
Bug: https://github.com/dart-lang/sdk/issues/31113
Change-Id: I03fdede75f1a85fda92487aa8f9307ce06f23903
Reviewed-on: https://dart-review.googlesource.com/16280
Commit-Queue: Vyacheslav Egorov <vegorov@google.com>
Reviewed-by: Alexander Markov <alexmarkov@google.com>
Vyacheslav Egorov