This change introduces tracking of variable values with respect to
control flow (e.g. uses consume only reaching definitions).
flutter_gallery in release mode:
instructions size: arm -0.42%, arm64 -0.48%
total size: arm -0.43%, arm64 -0.45%
velocity_tracker_bench in release mode:
instructions size: arm -1.22%, arm64 -1.53%
total size: arm -1.23%, arm64 -1.37%
NavierStokes in AOT mode +135% (x64)
Closes https://github.com/dart-lang/sdk/issues/37636
Change-Id: I54d4887d57b2722b77ca98df3637f56d0f452144
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/131740
Commit-Queue: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
Reviewed-by: Samir Jindel <sjindel@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>
This is done by collecting all frames on the current stack until an async/async* frame is hit
which has yielded before (i.e. is not in sync-async case).
From there on it finds the closure of the async/async* frame and starts traversing the listeners:
while (closure != null) {
yield_index = closure.context[Context::kAsyncJumpVarIndex]
pc = closure.function.code.pc_descriptors.LookupPcFromYieldIndex(yield_index);
<emit pc in frame>
closure = closure.context[Context::kAsyncCompleterVarIndex]._future._resultOrListeners.callback;
}
Bug: https://github.com/dart-lang/sdk/issues/37668
Change-Id: I97030d22e529bf8a74ecd30a8a2589cfe57ad330
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/122644
Commit-Queue: Clement Skau <cskau@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
New class representation is used in ConcreteType and ConeType instead of
DartType. Uses of pre-NNBD legacy raw types are eliminated in most cases.
This also simplifies subtype checks.
TypesBuilder class is created to translate static types into
TFA types. It holds CoreTypes which is used to recognize Null class
and supersedes Type.cone and Type.fromStatic factory constructors.
Change-Id: Ia1cd48cf8c738173a06dd04864f6cec8ae6def83
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/124286
Commit-Queue: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Ryan Macnak <rmacnak@google.com>
This is a reland of 56991ad94b
Original change's description:
> [vm,aot] Simple unreachable code elimination before type-flow analysis, take 2
>
> This is the relanding of https://dart-review.googlesource.com/c/sdk/+/121500
> with fixes.
>
> Original CL description:
>
> This change adds transformation for very early cleanup of unreachable
> code such as code guarded by if statements with constant conditions or
> code used in assert statements when assertions are disabled.
>
> The advantage of cleaning such code early is that type-flow analysis
> won't be looking at it and TFA-based tree shaker is able to remove
> more code.
>
> flutter_gallery_total_size -0.5663% (arm), -0.5409% (arm64)
> build_bench_total_size -2.533% (arm), -2.449% (arm64)
> gesture_detector_total_size -4.183% (arm), -4.072% (arm64)
>
> Change-Id: Ief8ebd0cd828c0e7a847a824f44d2d97a3595b87
> Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/121901
> Reviewed-by: Ryan Macnak <rmacnak@google.com>
> Reviewed-by: Martin Kustermann <kustermann@google.com>
> Commit-Queue: Alexander Markov <alexmarkov@google.com>
Change-Id: I82ea0f70845851f85ec159c366f408aa7f9e5788
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/122166
Reviewed-by: Martin Kustermann <kustermann@google.com>
Reviewed-by: Ryan Macnak <rmacnak@google.com>
Commit-Queue: Alexander Markov <alexmarkov@google.com>
This reverts commit 56991ad94b.
Reason for revert: broke building Flutter apps in release mode
Original change's description:
> [vm,aot] Simple unreachable code elimination before type-flow analysis, take 2
>
> This is the relanding of https://dart-review.googlesource.com/c/sdk/+/121500
> with fixes.
>
> Original CL description:
>
> This change adds transformation for very early cleanup of unreachable
> code such as code guarded by if statements with constant conditions or
> code used in assert statements when assertions are disabled.
>
> The advantage of cleaning such code early is that type-flow analysis
> won't be looking at it and TFA-based tree shaker is able to remove
> more code.
>
> flutter_gallery_total_size -0.5663% (arm), -0.5409% (arm64)
> build_bench_total_size -2.533% (arm), -2.449% (arm64)
> gesture_detector_total_size -4.183% (arm), -4.072% (arm64)
>
> Change-Id: Ief8ebd0cd828c0e7a847a824f44d2d97a3595b87
> Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/121901
> Reviewed-by: Ryan Macnak <rmacnak@google.com>
> Reviewed-by: Martin Kustermann <kustermann@google.com>
> Commit-Queue: Alexander Markov <alexmarkov@google.com>
TBR=vegorov@google.com,kustermann@google.com,rmacnak@google.com,alexmarkov@google.com
Change-Id: I25fe5d4457bfcc5ee25506810aa7eee01ca1b28a
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/122023
Reviewed-by: Alexander Markov <alexmarkov@google.com>
Commit-Queue: Alexander Markov <alexmarkov@google.com>
This is the relanding of https://dart-review.googlesource.com/c/sdk/+/121500
with fixes.
Original CL description:
This change adds transformation for very early cleanup of unreachable
code such as code guarded by if statements with constant conditions or
code used in assert statements when assertions are disabled.
The advantage of cleaning such code early is that type-flow analysis
won't be looking at it and TFA-based tree shaker is able to remove
more code.
flutter_gallery_total_size -0.5663% (arm), -0.5409% (arm64)
build_bench_total_size -2.533% (arm), -2.449% (arm64)
gesture_detector_total_size -4.183% (arm), -4.072% (arm64)
Change-Id: Ief8ebd0cd828c0e7a847a824f44d2d97a3595b87
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/121901
Reviewed-by: Ryan Macnak <rmacnak@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
Commit-Queue: Alexander Markov <alexmarkov@google.com>
This is needed to perform lowering during inference.
Some missing expectation files were added as part of the update.
Change-Id: Ifc834617621e8d60518e840dc61f0386fb8d3b92
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/121423
Reviewed-by: Aske Simon Christensen <askesc@google.com>
This reverts commit 41b65b27c2.
Reason for revert: broke protobuf-aot benchmarks.
Original change's description:
> [vm,aot] Simple unreachable code elimination before type-flow analysis
>
> This change adds transformation for very early cleanup of unreachable
> code such as code guarded by if statements with constant conditions or
> code used in assert statements when assertions are disabled.
>
> The advantage of cleaning such code early is that type-flow analysis
> won't be looking at it and TFA-based tree shaker is able to remove
> more code.
>
> flutter_gallery_total_size -0.5663% (arm), -0.5409% (arm64)
> build_bench_total_size -2.533% (arm), -2.449% (arm64)
> gesture_detector_total_size -4.183% (arm), -4.072% (arm64)
>
> Change-Id: Ic764c056b6594164dc32eed4fd5679a7077ea2f9
> Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/121500
> Commit-Queue: Alexander Markov <alexmarkov@google.com>
> Reviewed-by: Ryan Macnak <rmacnak@google.com>
> Reviewed-by: Martin Kustermann <kustermann@google.com>
TBR=vegorov@google.com,kustermann@google.com,rmacnak@google.com,alexmarkov@google.com
Change-Id: I3e5668176a7a90d2891126bf79139ac5859bc630
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/121824
Reviewed-by: Alexander Markov <alexmarkov@google.com>
Commit-Queue: Alexander Markov <alexmarkov@google.com>
This change adds transformation for very early cleanup of unreachable
code such as code guarded by if statements with constant conditions or
code used in assert statements when assertions are disabled.
The advantage of cleaning such code early is that type-flow analysis
won't be looking at it and TFA-based tree shaker is able to remove
more code.
flutter_gallery_total_size -0.5663% (arm), -0.5409% (arm64)
build_bench_total_size -2.533% (arm), -2.449% (arm64)
gesture_detector_total_size -4.183% (arm), -4.072% (arm64)
Change-Id: Ic764c056b6594164dc32eed4fd5679a7077ea2f9
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/121500
Commit-Queue: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Ryan Macnak <rmacnak@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
This is the follow-up for 46cef9bfdd.
After that change, VM's optimizer (canonicalization) removes
initializing stores which store null, even if they're overwriting the
result of another initializing store. This causes incorrect behavior
in cases like this:
class A {
int x = 3;
A() : x = null {}
}
To accomodate for this change in the VM's optimizer, bytecode generator
should elide duplicate stores, similarly to AST-based flow graph
builder.
Fixes language_2/field_parameter_test in AOT mode with bytecode.
Change-Id: I9b1e25840f639d88bf62dc990f183878ecf1b59b
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/121649
Reviewed-by: Ryan Macnak <rmacnak@google.com>
Commit-Queue: Alexander Markov <alexmarkov@google.com>
For a switch statement
switch(foo) {
case c_1: ...
...
case c_N: ...
}
Bytecode was calling foo.operator==(c_i). This is less efficient than
c_i.operator==(foo), as c_i is a constant with a known actual type
and operator== can be always devirtualized and inlined.
Change-Id: I786ea193de1fa1830854602318e436f601296f5e
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/121405
Reviewed-by: Ryan Macnak <rmacnak@google.com>
Reviewed-by: Régis Crelier <regis@google.com>
Commit-Queue: Alexander Markov <alexmarkov@google.com>
In order to support causal-async-stacks, the following synthetic calls
are inserted: _asyncStackTraceHelper, _setAsyncThreadStackTrace and
_clearAsyncThreadStackTrace.
_asyncStackTraceHelper is always generated in the async transformation.
_setAsyncThreadStackTrace and _clearAsyncThreadStackTrace are inserted
during bytecode generation (or in the flow graph builder in case of AST
pipeline). If bytecode generation options are set inconsistently with
--causal-async-stacks VM option, then it was possible that
_asyncStackTraceHelper was generated but other calls were not generated,
causing incorrect (full, non-truncated) async stack traces. This also
causes performance regression as collecting full stack traces takes
much more time.
This change makes generation of all these calls more consistent: if
causal async stacks are disabled during bytecode generation, then all
three calls are omitted. Also, in case these calls were generated and
present in bytecode, but VM option --causal-async-stacks is disabled,
then bytecode flow graph builder turns all these three calls into no-ops.
Change-Id: I93eb1d83c675ee093799bb8e37ca3d60a3c5c19d
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/120927
Commit-Queue: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Ryan Macnak <rmacnak@google.com>
Reviewed-by: Régis Crelier <regis@google.com>
Previously, calls through field/getter were generated as dynamic calls
because InterfaceCall with field/getter target doesn't correctly
represent what such call should do. This results in a quite inefficient
code in AOT mode.
With this change, calls through fields/getters are decomposed into two
calls, while keeping the same order of evaluation.
Change-Id: Ieb33927c1b39d1b4d8eafb7bd542675820474068
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/120628
Commit-Queue: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Régis Crelier <regis@google.com>
AOT transformations add vm.direct-call.metadata with devirtualization
information to AST nodes.
Bytecode generator should take this information into account when
generating bytecode in case of AOT. Bytecode format is extended with
CheckReceiverForNull and UncheckedDirectCall instructions, and
DirectCallViaDynamicForwarder constant pool entry in order to represent
devirtualized calls.
Change-Id: I697432ddd0b58d2d0413715132ba5e90eb606ec1
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/119201
Reviewed-by: Régis Crelier <regis@google.com>
Reviewed-by: Ryan Macnak <rmacnak@google.com>
Commit-Queue: Alexander Markov <alexmarkov@google.com>
When generating deopt info in SIMDBC mode (DeoptInfoBuilder::AddCopy,
DeoptInfoBuilder::ToCpuRegisterSource) kSpecialDbcRegister source
locations are not supported.
In order to avoid such deopt info, this change moves CheckStack bytecode
instruction after function type arguments handling is finished (function
type arguments handling uses arguments descriptor which resides in
kArgsDescriptorsReg special DBC register). This placement of CheckStack is
also more aligned with what AST-based flow graph builder generates.
Fixes lib_2/math/min_max_test with bytecode in dartk-reload-mac-release-simdbc64
configuration.
Change-Id: I7d9863b8290a78a6e24e5806cc5053d55c93ce5c
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/117221
Reviewed-by: Régis Crelier <regis@google.com>
Commit-Queue: Alexander Markov <alexmarkov@google.com>
VM no longer relies on distinct ICData entries in constant pool, and
it no longer creates actual ICData objects.
This change introduces new DynamicCall constant pool entry which is
* de-duplicated;
* stores arguments descriptor inline instead of allocating a separate
constant pool entry and referencing it by index;
* reserves extra slot in constant pool (like other call constants), so
we can eventually remove creation of extra UnlinkedCall objects.
Change-Id: I8d21954803ca5be5b6fc8335b700e8a9086245cf
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/116120
Reviewed-by: Ryan Macnak <rmacnak@google.com>
Reviewed-by: Régis Crelier <regis@google.com>
Commit-Queue: Alexander Markov <alexmarkov@google.com>
PushStatic instruction expects field object to be pushed onto the stack.
This is useless, as PushStatic also references field with its D operand.
This change replaces PushStatic instruction with LoadStatic which doesn't
take field object on the stack, so PushConstant/PushStatic pair is replaced
with a single LoadStatic instruction.
This change also enables constant propagation of values of injected CID
fields, which are not known at bytecode generation time.
Change-Id: Ifbdd3aea2aab338f6c9ec3e5728948c4d8541a4b
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/112489
Commit-Queue: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Ryan Macnak <rmacnak@google.com>
Reviewed-by: Régis Crelier <regis@google.com>
Alexander Markov