Use the TFA direct-call metadata to directly call a closure in function
invocations.
Closes#55231.
Tested: existing tests cover the new code paths, but I also added a new
test.
Change-Id: Ib5f26b10efd77570e256196b4bfb07e6bef800c0
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/397260
Reviewed-by: Martin Kustermann <kustermann@google.com>
Commit-Queue: Ömer Ağacan <omersa@google.com>
This reverts commit 63179aa49d.
See commit 1ab083ae48 for the original
description of the change.
The original change was reverted because it did not take
into account that in Dart it is legal to make some
parameters optional when you override a method. Such parameters
should not be passed in registers - which means for each
possible interface target we need to compute the minimum
number of direct parameters accepted by all overrides.
I have also added two comprehensive tests for different
combinations of direct, optional and named parameters.
TEST=vm/dart/calling_conventions_test.dart,vm/dart/calling_conventions2_test.dart
Cq-Include-Trybots: luci.dart.try:vm-aot-linux-debug-simarm_x64-try,vm-aot-linux-debug-simriscv64-try,vm-aot-linux-debug-x64-try,vm-aot-linux-debug-x64c-try,vm-aot-linux-product-x64-try,vm-aot-linux-release-arm64-try,vm-aot-linux-release-simarm_x64-try,vm-aot-linux-release-x64-try,vm-aot-optimization-level-linux-release-x64-try
Change-Id: Ib76e2512d3e4077a40265d315483ebe3b0f5551c
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/357862
Reviewed-by: Martin Kustermann <kustermann@google.com>
Commit-Queue: Slava Egorov <vegorov@google.com>
TFA now produces direct call metadata for function calls based on the
results of the analysis.
VM reads direct call metadata for function calls, but only uses it
if target is a tear-off, as it cannot handle inlining of an arbitrary
closure function yet (due to scope building dependencies between
compilation of a closure and its parent function).
TEST=pkg/vm/testcases/transformations/type_flow/transformer/closures.dart
Issue: https://github.com/dart-lang/sdk/issues/39692
Change-Id: Idf0b6988534bfca1a64c6661a82df8d453272abf
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/357180
Reviewed-by: Martin Kustermann <kustermann@google.com>
Reviewed-by: Slava Egorov <vegorov@google.com>
Commit-Queue: Alexander Markov <alexmarkov@google.com>
Following the lead of unboxed calling conventions augment
out backend to support passing arguments in registers on
ARM, ARM64 and X64.
Similar to unboxing the calling convention change only applies
to functions which are invoked statically or through
global dispatch table. Functions which are invoked from native
code or directly invoked by dynamic call sites are not allowed
to use register calling convention.
This CL avoid messing too much with call instructions and
instead leans on adjusting existing machinery: MoveArgument
instructions corresponding to register arguments are simply
detached from the graph and are processed as input constraints
by the register allocator.
The computation of the calling convention is centralized in
`ComputeCallingConvention` in `flow_graph.cc` and is used
by callees and callers.
While working on this CL I have also discovered an existing bug
in catch move handling (see added test runtime/tests/vm/dart/try_catch_parameter_sync_test.dart).
On a large Flutter app this decreases the size of generated code
by 1.8% (380Kb) and there a mixed bag of improvements in runing time, e.g.
```
build_bench (Pixel 6a)
-3.536% (0.2 noise)423.16 438.68
animation_bench_close_first_frame (Pixel 6a)
-3.717% (0.0 noise)253.64 263.43
rrect_contains_bench (Pixel 6a)
-3.796% (0.0 noise)0.088700 0.092200
sync_star_sync_star (Pixel 6a)
-5.039% (0.0 noise)2.4310 2.5600
sync_star_list (Pixel 6a)
-5.329% (0.0 noise)1.0660 1.1260
velocity_tracker_bench (Pixel 6a)
-5.731% (0.1 noise)62.039 65.811
sync_star_generated (Pixel 6a)
-6.438% (0.0 noise)1.090 01.1650
matrix_utils_transform_point_perspective (Pixel 6a)
-6.844% (0.2 noise)0.012520 0.013439
matrix_utils_transform_rect_perspective (Pixel 6a)
-7.818% (0.2 noise)0.028117 0.030502
matrix_utils_transform_rect_affine
-27.05% (0.7 noise) 0.017494 0.023981
```
TEST=ci
Issue https://github.com/dart-lang/sdk/issues/54955
Cq-Include-Trybots: luci.dart.try:vm-aot-linux-debug-simarm_x64-try,vm-aot-linux-debug-simriscv64-try,vm-aot-linux-debug-x64-try,vm-aot-linux-debug-x64c-try,vm-aot-linux-product-x64-try,vm-aot-linux-release-arm64-try,vm-aot-linux-release-simarm_x64-try,vm-aot-linux-release-x64-try,vm-aot-optimization-level-linux-release-x64-try
Change-Id: I9a134a764666a541d4693e6ca9f6813025ecd63e
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/353404
Reviewed-by: Alexander Markov <alexmarkov@google.com>
Commit-Queue: Slava Egorov <vegorov@google.com>
The modular parts in package:vm/modular/** will only depend on
* package:kernel
* package:front_end
=> Only for auto-generated error code messages
=> Those error codes can be their own dart_library() in g3
Based on https://dart-review.googlesource.com/c/sdk/+/353980, with
changes:
* Exclude package:kernel from "no exports" CFE presubmit.
* Add package:vm/modular to CFE "allowed deps".
* Fix import in `front_end/tool/perf_common.dart` and various others.
Tested: no code changes, only moves.
Change-Id: I9a44ac6ee05478812a9c8af31f6c4bbcf44b7c42
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/354221
Reviewed-by: Martin Kustermann <kustermann@google.com>
Commit-Queue: Morgan :) <davidmorgan@google.com>
Reviewed-by: Johnni Winther <johnniwinther@google.com>
Sometimes inferred closure value may reference a closure
which was eliminated by the tree shaker (as tree-shaker can prune
deeper AST). That should not cause a crash and inferred closure value
should be disregarded.
TEST=pkg/vm/testcases/transformations/type_flow/transformer/eliminated_closure.dart
Fixes b/305730721
Change-Id: Ia26e620ebc06f949be1e05c734be65ce1d708b64
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/330661
Commit-Queue: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
This change adds propagation of closure values to TFA.
For now, inferred closure values are used only when they are
constant (tear-off of a static method or a constructor).
Arbitrary closures can now be referenced from unrelated
members via closure-id metadata.
In addition, this change fixes an incorrect stack trace when
an implicit closure (tear-off) was propagated and its call was
inlined. Inlining interval was not recorded because of the missing
source position of a call to a target method within implicit closure.
TEST=pkg/vm/testcases/transformations/type_flow/transformer/closures.dart
Issue: https://github.com/dart-lang/sdk/issues/39692
Change-Id: I3590da91b6057e0b55a8614382dba1bbcc267b39
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/325447
Commit-Queue: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
Reviewed-by: Slava Egorov <vegorov@google.com>
This change moves inferred types of incoming arguments to a separate
metadata. Incoming argument type may be different from inferred
type of a parameter variable as parameter variable could be
modified in the method body.
This is a preparation for adding inferred types of captured
variables - parameter variable can be captured and can have both
inferred type of incoming argument and inferred type of a captured
variable.
TEST=ci
Issue: https://github.com/dart-lang/sdk/issues/51102
Change-Id: I9f77b8f5ed87477424f4fd3d339fdfb8e55abf82
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/320121
Commit-Queue: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
Reviewed-by: Slava Egorov <vegorov@google.com>
Unboxing of records in return values is switched to use inferred record
shape instead of a static type. This is more accurate, as
static return type 'Object', 'dynamic' or a type parameter would
not prevent record unboxing.
TEST=runtime/tests/vm/dart/records_return_value_unboxing_il_test.dart
TEST=pkg/vm/testcases/transformations/type_flow/transformer/unboxed_records.dart
Issue: https://github.com/dart-lang/sdk/issues/49719
Cq-Include-Trybots: luci.dart.try:vm-kernel-precomp-nnbd-linux-debug-x64-try,vm-kernel-precomp-nnbd-linux-release-x64-try
Change-Id: I56528e114bc2de94c4a1ec09c48eb5b9ed3c3d3f
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/288824
Commit-Queue: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
Functions returning non-nullable records of 2 fields can now return
them as separate values on 2 registers. This postpones and sometimes
eliminates creation of record instances. As other unboxing
optimizations of return values, this optimization is only
performed in AOT.
Pair of values for an unboxed record are represented using
kPairOfTagged representation, which can be now used for the input of
Return and for the output of StaticCall/InstanceCall
PolymorphicInstanceCall/DispatchTableCall instructions.
In order to combine separate values for Return, a new MakePair
instruction is added. Extracting separate values from the result of
a call is implemented using existing ExtractNthOutput instruction.
Benchmarks (AOT mode):
MultipleReturns.Forwarded.Record +45-57%
MultipleReturns.Forwarded.RecordNamed +43-57%
MultipleReturns.NotInlined.Record +58-79%
MultipleReturns.NotInlined.RecordNamed +53-76%
TEST=runtime/tests/vm/dart/records_return_value_unboxing_il_test.dart
TEST=benchmarks/MultipleReturns/dart/MultipleReturns.dart
Issue: https://github.com/dart-lang/sdk/issues/49719
Change-Id: I7117b19a134c1db0ba5117a1ef093867f9ba20e2
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/269100
Commit-Queue: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
This enables the serialization to handle references to libraries that
are not part of the serialized libraries. This is a step towards
removing the need to call computeCanonicalNames from outside the
serialization and to always use library filters for serialization
of a subset of the libraries in a component.
In response to b/235393918
TEST=existing
Change-Id: I48cd8b027795850ac2411a80e4b55efd517bb2cc
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/248588
Reviewed-by: Jens Johansen <jensj@google.com>
Reviewed-by: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Srujan Gaddam <srujzs@google.com>
Commit-Queue: Johnni Winther <johnniwinther@google.com>
Since the VM package isn't opted in, the individual libraries have
been annotated with `@dart=2.12` to opt in to null safety.
TEST=existing
Change-Id: I0bfbcf69cb80d32bb6b80a171f7bdb62fde7ca65
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/195277
Commit-Queue: Johnni Winther <johnniwinther@google.com>
Reviewed-by: Aske Simon Christensen <askesc@google.com>
The CL is a step towards have a more restricted and wellstructured
handled of references and canonical names.
The CL moves Reference to canonical_name.dart and makes
CanonicalName.reference private, and replaces CanonicalName.getReference
with a 'reference' getter.
It also removes NamedNode.canonicalName, Field.getterCanonicalName and
Field.setterCanonicalName so that these can only be accessed through the
corresponding reference. This is to reduce the reliance on the
canonical names which, ideally, should only be part of serialization and
deserialization.
TEST=existing
Change-Id: I955fb7d52d4e112d8741f7c12dcf38b74ae0c91a
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/190442
Commit-Queue: Johnni Winther <johnniwinther@google.com>
Reviewed-by: Jens Johansen <jensj@google.com>
The partitioning has the property that for every reference from loading unit S to unit T, either
- the reference is through a deferred prefix, or
- T is guarenteed to be loaded before S
The partitioning is constructed by cutting the dominator tree of the library import graph at deferred imports. It is suboptimal if there are deferred imports to multiple libraries in a connected component, but the expected use case involves loading units with a single deferred entry.
Bug: https://github.com/dart-lang/sdk/issues/41974
Change-Id: I3c49044ae19112525f197b077b36e4ce874b81ac
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/151470
Commit-Queue: Ryan Macnak <rmacnak@google.com>
Reviewed-by: Alexander Markov <alexmarkov@google.com>
This CL adds a Node.toText method together with an AstPrinter. These
facility and better toString implementation on AST nodes while allowing
for toString independent printing of AST to use in testing. This also
add support for an integrated toString of custom/internal nodes.
Some work is still needed in bringing the toString implementation on
all nodes to the old quality, and not all internal nodes have
customized textual representations yet. This work is left for future
CLs.
Change-Id: Ib0bf8a0bc02f489dfacdc8aa5f96da9c52f26058
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/150923
Commit-Queue: Johnni Winther <johnniwinther@google.com>
Reviewed-by: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Jens Johansen <jensj@google.com>
Adds another interface target to instance calls, pointing to the torn
off method for tearoff getter calls. This is then used to direct the
call to the method extractor for the method.
Increases the fraction of dynamically dispatched instance calls in
Flutter Gallery that are handled by the dispatch table from 95.8% to
98.4%. The remaining 92 IC calls are all on static receiver type
dynamic or Function (34 function calls, 22 method calls, 21 operator
calls, 14 getter calls and 1 setter call).
Change-Id: I0bde1a8271c8563f4631534bd39d1254cd779380
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/142155
Commit-Queue: Aske Simon Christensen <askesc@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
In order to have instance calls with unboxed parameters, all
the possible targets need to be able to those parameters unboxed.
Then, the methods were partitioned according to the class hierarchy,
and the inferred types from TFA are used to determine the parameters
that could be unboxed.
This is done for non-nullable integer and double parameters, and
return value.
dart-aot linux x64:
Bench2D 8.028%
DeltaBlue 4.688%
ParticleSystemPaint 27.79%
Flutter release android-armv8:
velocity_tracker_bench -2.026% (less is better)
flutter_gallery_total_size -0.1460%
flutter_gallery_instructions_size -0.2715%
flutter_gallery_readonlydata_size 0.0539%
Issue https://github.com/dart-lang/sdk/issues/33549
Change-Id: Iedd9b4321c3d78b406ea1708e15c747754433027
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/138501
Commit-Queue: Victor Agnez Lima <victoragnez@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
Reviewed-by: Samir Jindel <sjindel@google.com>
NullCheck pc descriptors are added to all dispatch table calls where the
receiver may be null (and the selector is not one implemented by null).
All null entries in the table go to the NullError runtime entry, which
reads the NullCheck pc descriptor to get the name of the called member
for the error message.
Change-Id: I9d2847d0ccdfdb735b06e879916920ec299f39bc
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/134294
Commit-Queue: Aske Simon Christensen <askesc@google.com>
Reviewed-by: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
For each selector, count the number of calls to the selector that are
both reachable and polymorphic according to the TFA. Only include
selectors in the table with non-zero counts.
This reduces the dispatch table size by 30% in dart2js and by 49% in
Flutter Gallery (312k memory use reduction on ARM64).
Call counts are transferred in a dedicated metadata block with
per-selector information. This mechanism also prepares for transferring
other per-selector information in the future.
Change-Id: Iba15aa4d6c50e67e53c3fd8e542123d3fc98bd07
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/132603
Reviewed-by: Alexander Markov <alexmarkov@google.com>
This is a Dart-tailored implementation of the "row-displacement
dispatch table" technique for closed-world instance calls:
All interface targets in the program are grouped into selectors such
that all targets that could potentially be called from the same call
site have the same selector (currently just grouped by name).
Each selector is assigned a selector offset such that offset + classid
is unique for all selector/classid combinations where the class
implements the selector.
At every instance call site that has an interface target (i.e. where
the static type of the receiver is not dynamic), the selector offset +
receiver classid is computed and used as index into a global table of
entry points.
If the receiver can be null (as determined by the front-end TFA and the
VM type propagation), a null check is inserted before the call.
An arguments descriptor is provided (only) for selectors that need it
(those which have type parameters or optional/named parameters).
The dispatch table calls don't need the monomorphic entry code, so for
functions that are only called via dispatch table calls (i.e. never
called dynamically), the monomorphic entry code is left out.
Some future improvements to the table dispatch implementation are
mentioned in https://github.com/dart-lang/sdk/issues/40188
The table dispatch flag is disabled by default in this commit. A
separate commit enables the flag.
Change-Id: Ic2911742b4a2c9a8d3bc7df60605454cbe4c0714
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/126648
Commit-Queue: Aske Simon Christensen <askesc@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
Reviewed-by: Alexander Markov <alexmarkov@google.com>
This reverts commit 5f81efb4a3.
Reason for revert: 3xH bot is green again.
Original change's description:
> Revert "[vm/aot] Only patch call-sites to monomorphic entry-points when the receiver is proven to be a heap object."
>
> This reverts commit ab2026af45.
>
> Reason for revert: Flutter 3xH is broken on Golem.
>
> Original change's description:
> > [vm/aot] Only patch call-sites to monomorphic entry-points when the receiver is proven to be a heap object.
> >
> > Flutter Gallery ARMv8: -0.8% total, -1.3% instructions
> > Flutter Gallery ARMv7: -0.7% total, -1.2% instructions
> >
> > There are some performance regressions, in four categories:
> >
> > 1. Arithmetic on num types (e.g. MinLib, MaxLib):
> > We believe arithmetic on num types is rare in practice, and can be optimized later
> > by specializing the call-sites.
> >
> > 2. Lack of types (e.g. MeshDecompression, ImagingGaussianBlur, Crypto*):
> > These (and similar benchmarks) are written in a very untyped style, which is not
> > representative of Dart 2 style. We've confirmed that the regressions disappear if
> > the benchmarks are annotated with appropriate types.
> >
> > 3. Megamorphic calls which are specialized in the benchmark (e.g. DeltaBlue):
> > DeltaBlue uses string interpolation in one place, so the toString() method in the
> > string interpolation helper is monomorphic for _Smis. However, in a realistic programs,
> > string interpolation is used much more frequently and with different types, so it's very
> > unlikely that this call would be monomorphic. When string interpolation is removed, there's
> > a remaining 1.3% regression on DeltaBlue which is not yet accounted for.
> >
> > 4. Noisy benchmarks
> > The remaining regressions are on benchmarks which show frequent jitter historically.
> >
> > There are also 1-4% improvements on many benchmarks (Golem's "noise" analysis is hiding
> > many of them, look at the graphs).
> >
> > For the reasons above, we believe the regressions justified by the code size improvement.
> >
> > Change-Id: Ic9b281d4383a6111de9d6f44347976ffa61a6ca6
> > Cq-Include-Trybots:luci.dart.try:vm-kernel-precomp-linux-debug-x64-try,vm-kernel-precomp-linux-release-simarm-try,vm-kernel-precomp-linux-release-simarm64-try,vm-kernel-precomp-bare-linux-release-simarm64-try
> > Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/128667
> > Commit-Queue: Samir Jindel <sjindel@google.com>
> > Reviewed-by: Ryan Macnak <rmacnak@google.com>
> > Reviewed-by: Martin Kustermann <kustermann@google.com>
>
> TBR=kustermann@google.com,rmacnak@google.com,alexmarkov@google.com,sjindel@google.com
>
> Change-Id: I411aae4b230f2a08146ad3bf3e7a10aa6592db0e
> No-Presubmit: true
> No-Tree-Checks: true
> No-Try: true
> Cq-Include-Trybots: luci.dart.try:vm-kernel-precomp-linux-debug-x64-try, vm-kernel-precomp-linux-release-simarm-try, vm-kernel-precomp-linux-release-simarm64-try, vm-kernel-precomp-bare-linux-release-simarm64-try
> Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/131660
> Reviewed-by: Samir Jindel <sjindel@google.com>
> Commit-Queue: Samir Jindel <sjindel@google.com>
TBR=kustermann@google.com,rmacnak@google.com,alexmarkov@google.com,sjindel@google.com
# Not skipping CQ checks because original CL landed > 1 day ago.
Change-Id: I36e4e9d598033fcc934c2cdf5b061aa11255d3b2
Cq-Include-Trybots: luci.dart.try:vm-kernel-precomp-linux-debug-x64-try, vm-kernel-precomp-linux-release-simarm-try, vm-kernel-precomp-linux-release-simarm64-try, vm-kernel-precomp-bare-linux-release-simarm64-try
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/131837
Reviewed-by: Samir Jindel <sjindel@google.com>
Commit-Queue: Samir Jindel <sjindel@google.com>
This reverts commit ab2026af45.
Reason for revert: Flutter 3xH is broken on Golem.
Original change's description:
> [vm/aot] Only patch call-sites to monomorphic entry-points when the receiver is proven to be a heap object.
>
> Flutter Gallery ARMv8: -0.8% total, -1.3% instructions
> Flutter Gallery ARMv7: -0.7% total, -1.2% instructions
>
> There are some performance regressions, in four categories:
>
> 1. Arithmetic on num types (e.g. MinLib, MaxLib):
> We believe arithmetic on num types is rare in practice, and can be optimized later
> by specializing the call-sites.
>
> 2. Lack of types (e.g. MeshDecompression, ImagingGaussianBlur, Crypto*):
> These (and similar benchmarks) are written in a very untyped style, which is not
> representative of Dart 2 style. We've confirmed that the regressions disappear if
> the benchmarks are annotated with appropriate types.
>
> 3. Megamorphic calls which are specialized in the benchmark (e.g. DeltaBlue):
> DeltaBlue uses string interpolation in one place, so the toString() method in the
> string interpolation helper is monomorphic for _Smis. However, in a realistic programs,
> string interpolation is used much more frequently and with different types, so it's very
> unlikely that this call would be monomorphic. When string interpolation is removed, there's
> a remaining 1.3% regression on DeltaBlue which is not yet accounted for.
>
> 4. Noisy benchmarks
> The remaining regressions are on benchmarks which show frequent jitter historically.
>
> There are also 1-4% improvements on many benchmarks (Golem's "noise" analysis is hiding
> many of them, look at the graphs).
>
> For the reasons above, we believe the regressions justified by the code size improvement.
>
> Change-Id: Ic9b281d4383a6111de9d6f44347976ffa61a6ca6
> Cq-Include-Trybots:luci.dart.try:vm-kernel-precomp-linux-debug-x64-try,vm-kernel-precomp-linux-release-simarm-try,vm-kernel-precomp-linux-release-simarm64-try,vm-kernel-precomp-bare-linux-release-simarm64-try
> Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/128667
> Commit-Queue: Samir Jindel <sjindel@google.com>
> Reviewed-by: Ryan Macnak <rmacnak@google.com>
> Reviewed-by: Martin Kustermann <kustermann@google.com>
TBR=kustermann@google.com,rmacnak@google.com,alexmarkov@google.com,sjindel@google.com
Change-Id: I411aae4b230f2a08146ad3bf3e7a10aa6592db0e
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Cq-Include-Trybots: luci.dart.try:vm-kernel-precomp-linux-debug-x64-try, vm-kernel-precomp-linux-release-simarm-try, vm-kernel-precomp-linux-release-simarm64-try, vm-kernel-precomp-bare-linux-release-simarm64-try
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/131660
Reviewed-by: Samir Jindel <sjindel@google.com>
Commit-Queue: Samir Jindel <sjindel@google.com>
Flutter Gallery ARMv8: -0.8% total, -1.3% instructions
Flutter Gallery ARMv7: -0.7% total, -1.2% instructions
There are some performance regressions, in four categories:
1. Arithmetic on num types (e.g. MinLib, MaxLib):
We believe arithmetic on num types is rare in practice, and can be optimized later
by specializing the call-sites.
2. Lack of types (e.g. MeshDecompression, ImagingGaussianBlur, Crypto*):
These (and similar benchmarks) are written in a very untyped style, which is not
representative of Dart 2 style. We've confirmed that the regressions disappear if
the benchmarks are annotated with appropriate types.
3. Megamorphic calls which are specialized in the benchmark (e.g. DeltaBlue):
DeltaBlue uses string interpolation in one place, so the toString() method in the
string interpolation helper is monomorphic for _Smis. However, in a realistic programs,
string interpolation is used much more frequently and with different types, so it's very
unlikely that this call would be monomorphic. When string interpolation is removed, there's
a remaining 1.3% regression on DeltaBlue which is not yet accounted for.
4. Noisy benchmarks
The remaining regressions are on benchmarks which show frequent jitter historically.
There are also 1-4% improvements on many benchmarks (Golem's "noise" analysis is hiding
many of them, look at the graphs).
For the reasons above, we believe the regressions justified by the code size improvement.
Change-Id: Ic9b281d4383a6111de9d6f44347976ffa61a6ca6
Cq-Include-Trybots:luci.dart.try:vm-kernel-precomp-linux-debug-x64-try,vm-kernel-precomp-linux-release-simarm-try,vm-kernel-precomp-linux-release-simarm64-try,vm-kernel-precomp-bare-linux-release-simarm64-try
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/128667
Commit-Queue: Samir Jindel <sjindel@google.com>
Reviewed-by: Ryan Macnak <rmacnak@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
Previously, generated bytecode was cached using metadata in kernel AST.
When serializaing kernel AST (for subsequent compilations) incremental
serialization was not used, so AST serialization was considerably
slower in bytecode mode compared to non-bytecode mode.
Apparently, metadata is not compatible with incremental serialization
(https://github.com/dart-lang/sdk/issues/39302).
This change reimplements caching of generated bytecode using a separate
'foo.dill.cache' file near the generated 'foo.dill' and 'foo.dill.ast'.
AST is now written using incremental serialization both in bytecode
and non-bytecode modes.
flutter_test_performance benchmark:
AST mode:
"without_change_elapsed_time_ms": 2615-2652,
"implementation_change_elapsed_time_ms": 6292-6394,
"interface_change_elapsed_time_ms": 6183-6484,
"with_coverage_time_ms": 2723-2834
Bytecode mode, before this change:
"without_change_elapsed_time_ms": 3246-3295,
"implementation_change_elapsed_time_ms": 7998-8068,
"interface_change_elapsed_time_ms": 7899-8029,
"with_coverage_time_ms": 3316-3378
Bytecode mode, after this change:
"without_change_elapsed_time_ms": 2689-2737,
"implementation_change_elapsed_time_ms": 7630-7677,
"interface_change_elapsed_time_ms": 7724-7917,
"with_coverage_time_ms": 2841-2949
Change-Id: I603d33f49949146cf6620226d810f2f3cb9853a9
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/124592
Reviewed-by: Ryan Macnak <rmacnak@google.com>
Commit-Queue: Alexander Markov <alexmarkov@google.com>
Previously, before switching to compact encoding of bytecode
instructions, we had to align bytecode instructions by 4.
Since that time sections of bytecode binary were aligned.
However, this alignment doesn't work well in AOT mode if dill file
contains both bytecode and AST, as other metadata (such as obfuscation
prohibitions) may precede bytecode metadata and may make the whole
bytecode binary misaligned.
This results in incorrect calculation of section offsets,
failed assertion
assert(writer.offset - start == section.offset);
and VM may fail to load bytecode binary with error
bytecode_reader.cc: 1368: error: unreachable code
In order to fix these problems, extra alignment of bytecode
sections is removed.
Change-Id: Iaa3b7c46999615aaa446781a7018db409261ead8
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/123686
Reviewed-by: Régis Crelier <regis@google.com>
Reviewed-by: Ryan Macnak <rmacnak@google.com>
Commit-Queue: Alexander Markov <alexmarkov@google.com>
After d65a60db00, we don't need AST with the bytecode to initialize a kernel compiler since the kernel compiler will initialize from a separate file.
Remove duplicated BinaryCacheMetadataRepository.
Fix generating bytecode in non-incremental mode.
Change-Id: Ica7bfb5a2d86dcc224849a9934a7a4676b01f075
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/119702
Commit-Queue: Ryan Macnak <rmacnak@google.com>
Reviewed-by: Alexander Markov <alexmarkov@google.com>
Daco Harkes