This CL validates the hypothesis that the only misaligned reads/writes
which are not supported are double and float, and only on arm32.
Running these on the CI and Golem will validate this hypothesis.
Bug: https://github.com/dart-lang/sdk/issues/45009
Change-Id: I0a77fd1f47a388d1f454c1ded50cd7ecaeadb0f0
Cq-Include-Trybots: luci.dart.try:dart-sdk-linux-try,dart-sdk-mac-try,dart-sdk-win-try,vm-ffi-android-debug-arm-try,vm-ffi-android-debug-arm64-try,vm-kernel-asan-linux-release-x64-try,vm-kernel-mac-debug-x64-try,vm-kernel-linux-debug-ia32-try,vm-kernel-linux-debug-x64-try,vm-kernel-nnbd-linux-debug-x64-try,vm-kernel-nnbd-linux-debug-ia32-try,vm-kernel-nnbd-mac-release-x64-try,vm-kernel-nnbd-win-debug-x64-try,vm-kernel-precomp-linux-debug-x64-try,vm-kernel-precomp-linux-debug-simarm_x64-try,vm-kernel-precomp-nnbd-linux-debug-x64-try,vm-kernel-precomp-win-release-x64-try,vm-kernel-reload-linux-debug-x64-try,vm-kernel-reload-rollback-linux-debug-x64-try,vm-kernel-win-debug-x64-try,vm-kernel-win-debug-ia32-try,vm-precomp-ffi-qemu-linux-release-arm-try,vm-kernel-precomp-obfuscate-linux-release-x64-try,vm-kernel-msan-linux-release-x64-try,vm-kernel-precomp-msan-linux-release-x64-try,vm-kernel-precomp-android-release-arm_x64-try,analyzer-analysis-server-linux-try
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/190523
Commit-Queue: Daco Harkes <dacoharkes@google.com>
Reviewed-by: Aske Simon Christensen <askesc@google.com>
Inlines in the test classes:
- Native function lookup.
- The main benchmark loop.
This CL contains the refactoring necessary to update the
benchmarks to include leaf calls, as seen in:
https://dart-review.googlesource.com/c/sdk/+/179768
TEST=tools/build.py -m release dart && out/ReleaseX64/dart benchmarks/FfiCall/dart{,2}/FfiCall.dart
Change-Id: I1e85f991d895862c27ae99a71f5d4785ccdf5676
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/188465
Commit-Queue: Clement Skau <cskau@google.com>
Commit-Queue: Daco Harkes <dacoharkes@google.com>
Reviewed-by: Daco Harkes <dacoharkes@google.com>
Adds benchmarks using `asTypedList` on Pointers to investigate from
where it pays of to do construct a TypedData to do memory access
instead of on the Pointer directly.
x64 JIT results:
- Memory access on TypedData (int8) is ~2,5 faster than on Pointer.
FfiMemory.PointerInt8(RunTime): 72.23885718413639 us.
FfiMemory.PointerInt8TypedDataReuse(RunTime): 28.239710263614928 us.
- Constructing the TypedData for 1000 reads + 1000 writes takes ~20%
of the total runtime, 400 reads or writes on TypedData, 160 reads
or writes on Pointer.
FfiMemory.PointerInt8TypedDataNew(RunTime): 34.345480148372026 us.
FfiMemory.PointerInt8TypedDataReuse(RunTime): 28.239710263614928 us.
x64 AOT results:
FfiMemory.PointerInt8(RunTime): 13.862134213116345 us.
FfiMemory.PointerInt8TypedDataNew(RunTime): 26.149628021913365 us.
FfiMemory.PointerInt8TypedDataReuse(RunTime): 24.73322780505299 us.
- Using asTypedList is slower. (Some optimizations fail to kick in.)
Change-Id: I401edb88baf3a3c5094ecae808f122adf258da28
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/186289
Reviewed-by: Clement Skau <cskau@google.com>
Commit-Queue: Daco Harkes <dacoharkes@google.com>
- Fixes expect logic in Float.. and Double.. benchmarks.
- Fixes expected values in the same benchmarks.
- Fixes typos in a few FFI function lookups.
- Adds expected value to 'Unexpected result' message.
Change-Id: I16eff1cac451412b376f56c070c5b8127ff744e6
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/186284
Auto-Submit: Clement Skau <cskau@google.com>
Commit-Queue: Clement Skau <cskau@google.com>
Reviewed-by: Daco Harkes <dacoharkes@google.com>
This CL does not yet roll `package:ffi` to use `Allocator`, because that
breaks the checked in Dart in Flutter in g3. Instead, this copies
`_CallocAllocator` from `package:ffi` into the benchmarks. (We need a
copy per benchmark such that file-copying before running benchmarks
works properly.) The copies can be deleted when we can update
`package:ffi` in the DEPS file to contain `_CallocAllocator`.
New API landed in: https://dart-review.googlesource.com/c/sdk/+/177705
Issue: https://github.com/dart-lang/sdk/issues/44621
Issue: https://github.com/dart-lang/sdk/issues/38721
Change-Id: I546de7ec65ceb6f05644a5f269b83f64656892e5
Cq-Include-Trybots: luci.dart.try:benchmark-linux-try
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/178995
Commit-Queue: Daco Harkes <dacoharkes@google.com>
Reviewed-by: Aske Simon Christensen <askesc@google.com>
- Generate strings on one place rather than work hard to generate them
the same with different number types.
- Update dart2 version
Change-Id: I69e6bde8b870598e062b5009578e5f8c7b2a6add
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/176585
Auto-Submit: Stephen Adams <sra@google.com>
Reviewed-by: Stephen Adams <sra@google.com>
Commit-Queue: Stephen Adams <sra@google.com>
This reverts commit 7ebbc89e78.
The benchmark became broken because it checks for round-trip fidelity
and using a shared list of strings there is rounding on the int.*.64
case.
I'll fix it and recommit.
TBR: whesse@google.com
Change-Id: I9ba25716992e59929726745b9414eb8dc17a55fa
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/176580
Reviewed-by: Stephen Adams <sra@google.com>
Commit-Queue: Stephen Adams <sra@google.com>
- Generate strings on one place rather than work hard to generate them
the same with different number types.
- Update dart2 version
Change-Id: I449d08a5ead13504ec604b11bd63b729590d6ffe
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/175861
Reviewed-by: William Hesse <whesse@google.com>
Commit-Queue: Stephen Adams <sra@google.com>
A benchmark that is a collection of other benchmarks.
This benchmark has two purposes:
- As a medium sized compilation benchmark
- As a tool for running benchmarks as part of a larger program
This benchmark is not intended to be run for the purpose of generating
benchmark timing results on Golem.
It should be extended over time to include most public benchmarks that
are meaningful on all Dart platforms.
Change-Id: I17f034cfd68d01ff5b4842b89b216fcea0d5bd3c
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/170060
Commit-Queue: Stephen Adams <sra@google.com>
Reviewed-by: William Hesse <whesse@google.com>
The existing EventLoopLatencyJson benchmark exercises the *worst case scenario*:
A background isolate is stressing the GC where the working memory set
does not fit into new space. This causes scavenges to copy the entire
newspace either into the to-space or promote it to old space.
The benchmark in this CL is similar to EventLoopLatencyJson, but it's working
memory set is significantly lower than new space, therefore not
exercising this worst case scenario (this represents the most common
scenario, which we would expect for compute() function in flutter)
(Due to difficulty sharing code between benchmarks, this CL adds
similar code as EventLoopLatencyJson)
Issue https://github.com/dart-lang/sdk/issues/36097
Change-Id: I385a620ba8900082f8ca5fb9080fc23d95112e83
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/169100
Reviewed-by: Alexander Aprelev <aam@google.com>
Commit-Queue: Martin Kustermann <kustermann@google.com>
This CL adds Utf8 encode benchmarks to complement the existing Utf8
decode benchmarks. These benchmarks measure the time to process a
given input rune.
The motivation for adding this was after noticing in external
benchmarks that Dart utf8 encoding can sometimes be much slower
than the utf8 encoding in other languages. After adding benchmarks,
I plan to send CLs for some small tweaks that may improve performance.
Change-Id: I8751d7df22937e9a85b85794e3488c664c359bbe
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/158820
Auto-Submit: Benjamin Prosnitz <bprosnitz@google.com>
Reviewed-by: Aske Simon Christensen <askesc@google.com>
Commit-Queue: Aske Simon Christensen <askesc@google.com>
Dart 2 version of this benchmark doesn't have this call.
Dart 3 (NNBD) is more strict and allows implicit casts only if casting
from dynamic, so it highlights places where types are incompatible.
However, Iterable.cast<int>() has a high overhead and it doesn't look
like a right way to suppress static type check in this case, as it
creates a new Iterable which would additionally cast each element.
Simple 'input as dynamic' cast is enough in this case and reflects
what was happening in Dart 2 case when types are incompatible.
Fixes https://github.com/dart-lang/sdk/issues/42506
Change-Id: I8d83796a53b53e6f567f533f3d5369cd29f6ad97
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/152700
Commit-Queue: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Régis Crelier <regis@google.com>
Reviewed-by: William Hesse <whesse@google.com>
The literal_only_boolean_expressions analyzer hint flags this code:
while (true) {
...
if (cond) break;
...
}
=> This is very common code.
The unnecessary_await_in_return analyzer hint flags this code:
Future foo() async {
...
return await foo();
}
=> Addressing this hint by removing the `await` changes semantics,
performance characteristics and async stack traces.
Change-Id: I3c126123bfba7386b01033084e954267af21ba13
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/152001
Reviewed-by: Alexander Markov <alexmarkov@google.com>
Commit-Queue: Martin Kustermann <kustermann@google.com>
Hoist 'one' in all versions of benchmark for uniformity.
Change-Id: Id2aaa9b18903ff33e3057c01e13ff332fc8979ab
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/151180
Commit-Queue: Stephen Adams <sra@google.com>
Reviewed-by: Jonas Termansen <sortie@google.com>
This hint flags code such as
int time = watch.elapsedMillisecond;
and encourages using `var` everywhere:
var time = watch.elapsedMillisecond;
Though if the type is not obvious to a developer, we should be allowed
to write it.
Furthermore it copies the rules from
`package:pedantic/analysis_options-1.9.0.yaml` to avoid being
automatically opted-in to new rules when a new package:pedantic
gets rolled into the SDK.
Change-Id: I45c72584885c608a56745685e9068ba83dfbed47
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/151526
Commit-Queue: Martin Kustermann <kustermann@google.com>
Reviewed-by: Alexander Markov <alexmarkov@google.com>
On x64 desktop in JIT the trampolines with Handles are a bit slower
than trampolines passing integers or Pointers.
FfiCall.Uint64x01(RunTime): 200.8482627033541 us.
FfiCall.PointerUint8x01(RunTime): 261.3910088865656 us.
FfiCall.Handlex01(RunTime): 355.4978670458585 us.
FfiCall.Handlex02(RunTime): 384.86376755820663 us.
FfiCall.Handlex04(RunTime): 492.896007885658 us.
FfiCall.Handlex10(RunTime): 846.1214043993232 us.
FfiCall.Handlex20(RunTime): 1193.412291169451 us.
Issue: https://github.com/dart-lang/sdk/issues/36858
Issue: https://github.com/dart-lang/sdk/issues/41319
New issue for optimizing multiple handles:
https://github.com/dart-lang/sdk/issues/42341
Also cleans up dart2/native which was erroneously left over from the
benchmark duplication for NNBD.
Change-Id: I81223fefc47129d00984492efb9502d5449f0d4a
Cq-Include-Trybots: luci.dart.try:benchmark-linux-try
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/145593
Commit-Queue: Daco Harkes <dacoharkes@google.com>
Reviewed-by: Jonas Termansen <sortie@google.com>
Also:
* Clean up lints in both unmigrated and migrated versions of benchmarks.
* Cache lists in local variables in Uint8ListCopyViaLoopBenchmark and
Float64ListCopyViaLoopBenchmark to make the same number of field
accesses as in other benchmarks, in order to make comparison between
different ways of copying typed data more fair.
Change-Id: I1e7e726ebf1f3cf6c199fce515c2a56d94576ba6
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/150280
Reviewed-by: Régis Crelier <regis@google.com>
Commit-Queue: Alexander Markov <alexmarkov@google.com>
The shared libraries for the FfiBoringssl benchmark had been moved up a
level but unfortunately the same change wasn't made for the FfiCall
benchmark.
The IsolateJson benchmark had a typo in the path and broke.
Change-Id: Iec2fec1d5508022d9cfbfb040d80c74b078a0640
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/150024
Reviewed-by: William Hesse <whesse@google.com>
This change actually migrates vm/dart/* tests to NNBD by fixing
compile-time errors and adjusting tests where needed.
vm/dart/null_float32x4_simd_ops_test and vm/dart/null_float64x2_simd_ops_test
are deleted as they are superseded by static type checks.
vm/dart/regress_40462_test.dart is a huge source code auto-generated
by fuzzer and migrating and maintaining this source doesn't have much
value, so it is deleted.
There are still failures in strong mode due to dependencies on
the packages which are not migrated yet (https://github.com/dart-lang/sdk/issues/42146).
Migrated vm/dart/null_checks_with_dwarf_stack_traces_test fails both in
weak and strong mode due to https://github.com/dart-lang/sdk/issues/42149.
Issue: https://github.com/dart-lang/sdk/issues/41314
Change-Id: I5561f1c8705ec16def0c4e0efa495d15f4ea7259
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/149493
Commit-Queue: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Ryan Macnak <rmacnak@google.com>
Reviewed-by: Régis Crelier <regis@google.com>
Reviewed-by: Siva Annamalai <asiva@google.com>
This change forks the benchmarks for the null safety feature where there
now is a null safe version at benchmarks/Foo/dart and a legacy Dart 2
version at benchmarks/Foo/dart2. This change allows benchmarks to be
migrated per go/dart-nnbd-benchmark-migration.
Additionally this change pins the internal benchmarks which have been
moved into their own repository and is now versioned along with the Dart
SDK.
The dart2 implementation will be benchmarked instead of the null safe
dart implementation as of this change. The null safe dart implementation
will be used when null safe benchmarking is turned on.
Change-Id: If039fd7100c960169f4161c1d98167aca0af2ded
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/148440
Reviewed-by: Alexander Thomas <athom@google.com>