Introduces NativeRepresentation and NativeLocation for the compilation of FFI.
NativeRepresentations are able to express all representations (or types) of the native ABIs we bind to with FFI, this is more representations that than that are used in Dart itself.
NativeLocations are able to express all locations of the native ABIs we bind to with FFI, this is more types of locations than that are used for the Dart calling convention.
See the documentation in the respective files.
These NativeLocations and NativeRepresentations are computed by the NativeCallingConvention and consumed by the Marshaller and Assemblers.
This reenginering is required for go/dart-ffi-by-value, hardfp (Arm S and D fpu registers), and iOS 64 bit (non-word-aligned stack arguments).
In addition, by using the NativeRepresentations we also get slightly reduced code size:
* The tracking of sizes is improved, so less sign/zero-extension operations are required.
* UnboxedWidthExtenderInstr is fully removed, the size extension is done inside the native moves, coalescing moves and size extension when possible.
* BitCastInstr is only used when really needed. This reduces code-size on arm32 softfp.
This fixes the iOS arm64 calling convention, manually tested with https://github.com/flutter/flutter/pull/46078 and https://dart-review.googlesource.com/c/sdk/+/131074.
Fixes: https://github.com/dart-lang/sdk/issues/39637
Issue: https://github.com/dart-lang/sdk/issues/36309
Issue: https://github.com/dart-lang/sdk/issues/36730
Change-Id: I8878bc0f314277bab4ca22f417c6295ecc017720
Cq-Include-Trybots: luci.dart.try:vm-ffi-android-debug-arm-try,vm-ffi-android-debug-arm64-try,app-kernel-linux-debug-x64-try,vm-kernel-linux-debug-ia32-try,vm-kernel-win-debug-x64-try,vm-kernel-win-debug-ia32-try,vm-kernel-precomp-linux-debug-x64-try,vm-dartkb-linux-release-x64-abi-try,vm-kernel-precomp-android-release-arm64-try,vm-kernel-asan-linux-release-x64-try,vm-kernel-linux-release-simarm-try,vm-kernel-linux-release-simarm64-try,vm-kernel-precomp-android-release-arm_x64-try,vm-kernel-precomp-obfuscate-linux-release-x64-try,dart-sdk-linux-try,analyzer-analysis-server-linux-try,analyzer-linux-release-try,front-end-linux-release-x64-try,vm-kernel-precomp-win-release-x64-try
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/129081
Commit-Queue: Daco Harkes <dacoharkes@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
Reviewed-by: Alexander Markov <alexmarkov@google.com>
This reverts commit 7475c637c3.
Reason for revert: Breakage on simarm_x64 (and thus arm_x64)
Original change's description:
> [vm/aot] Remove object wrapping of bare instructions for AOT snapshots.
>
> Now, when writing an AOT snapshot in bare instructions mode, only
> the actual instructions in the RawInstructions payload are serialized
> instead of the entire RawInstructions object.
>
> Since there are no longer RawInstructions objects in these AOT
> snapshots, we also change how Code objects are serialized. Instead
> of just containing a reference to the RawInstructions object, we
> serialize two pieces of information: where the instructions
> payload for this Code object begins and whether there was a single
> entry for the instructions payload. (To save space, the single
> entry bit is serialized as the low order bit of the unchecked
> offset, which was already being serialized).
>
> While we also need the length of the instructions payload, we
> approximate it for all but the last Code object by subtracting
> the next Code object's payload start from this Code object's
> payload start. For the last Code object, we assume it extends
> to the end of the instructions image.
>
> Changes on flutter gallery in release mode:
> armv7: instructions size -2.66%, total size -1.68%
> armv8: instructions size -5.81%, total size -3.49%
>
> Fixes https://github.com/dart-lang/sdk/issues/38451.
>
> Change-Id: Ia458e8d99bae18f5c3b6e849df2519027f06f574
> 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,vm-kernel-precomp-linux-release-simarm_x64-try,vm-kernel-precomp-android-release-arm64-try,vm-kernel-precomp-android-release-arm_x64-try,vm-kernel-precomp-mac-release-simarm64-try,vm-kernel-precomp-win-release-x64-try
> Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/131067
> Commit-Queue: Teagan Strickland <sstrickl@google.com>
> Reviewed-by: Martin Kustermann <kustermann@google.com>
TBR=kustermann@google.com,rmacnak@google.com,sstrickl@google.com
Change-Id: If2c8ca6b0993211b2509b275af7ff11a8fa2baa3
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
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, vm-kernel-precomp-linux-release-simarm_x64-try, vm-kernel-precomp-android-release-arm64-try, vm-kernel-precomp-android-release-arm_x64-try, vm-kernel-precomp-mac-release-simarm64-try, vm-kernel-precomp-win-release-x64-try
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/134322
Reviewed-by: Teagan Strickland <sstrickl@google.com>
Commit-Queue: Teagan Strickland <sstrickl@google.com>
Now, when writing an AOT snapshot in bare instructions mode, only
the actual instructions in the RawInstructions payload are serialized
instead of the entire RawInstructions object.
Since there are no longer RawInstructions objects in these AOT
snapshots, we also change how Code objects are serialized. Instead
of just containing a reference to the RawInstructions object, we
serialize two pieces of information: where the instructions
payload for this Code object begins and whether there was a single
entry for the instructions payload. (To save space, the single
entry bit is serialized as the low order bit of the unchecked
offset, which was already being serialized).
While we also need the length of the instructions payload, we
approximate it for all but the last Code object by subtracting
the next Code object's payload start from this Code object's
payload start. For the last Code object, we assume it extends
to the end of the instructions image.
Changes on flutter gallery in release mode:
armv7: instructions size -2.66%, total size -1.68%
armv8: instructions size -5.81%, total size -3.49%
Fixes https://github.com/dart-lang/sdk/issues/38451.
Change-Id: Ia458e8d99bae18f5c3b6e849df2519027f06f574
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,vm-kernel-precomp-linux-release-simarm_x64-try,vm-kernel-precomp-android-release-arm64-try,vm-kernel-precomp-android-release-arm_x64-try,vm-kernel-precomp-mac-release-simarm64-try,vm-kernel-precomp-win-release-x64-try
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/131067
Commit-Queue: Teagan Strickland <sstrickl@google.com>
Reviewed-by: Martin Kustermann <kustermann@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>
The pc-relative calls were recorded with pc-offsets pointing to the start of the
call instruction, where as code-based calls were recorded with
pc-offsets pointing to the next instruction (i.e. return address).
This is inconsistent and caused us to hit an assert in
`Code::set_static_calls_target_table`. The assert was benign, but it is
good to maintain the uniqueness guarantee in the static calls table, so
we'll unify the encoding to use offsets to the instruction after the
call in both cases.
Closes https://github.com/dart-lang/sdk/issues/39811
Change-Id: Id0305befd78f09ed0b0e100f39641bca9e764442
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/129717
Reviewed-by: Ben Konyi <bkonyi@google.com>
Commit-Queue: Ben Konyi <bkonyi@google.com>
This CL is responsible for inlining Float32x4 and Float64x2 arithmetic
operations in order to reduce their run-time cost.
Null-checks were required in order to allow the unbox instructions to
be non-speculative.
CheckNullInstr was modified to throw ArgumentError when checking
arguments for those operations.
Change-Id: I18b36d96a36011647480f301eb6dc6bb478f59a1
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/128589
Commit-Queue: Victor Agnez Lima <victoragnez@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
Reviewed-by: Samir Jindel <sjindel@google.com>
There's just one case where we can't get the offset for Instructions
from somewhere else, and that's when we reset active_instructions_ to
instructions_ in Code::Enable(). Thus, we add a caching field for the
uncached offset to RawCode, but only when the active_instructions_
field is added (when not running in the precompiled runtime).
We also output the unchecked offset in snapshots for Code objects so
we can recalculate the entry point cache on loading. (For full JIT
snapshots, we output both the cached offset and the offset for
active_instructions_.)
Bug: https://github.com/dart-lang/sdk/issues/38451
Change-Id: I78572ec83d1574b00820df148bccb57e5c79eaaa
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/125400
Reviewed-by: Martin Kustermann <kustermann@google.com>
Commit-Queue: Teagan Strickland <sstrickl@google.com>
We eagerly add objects into the global object pool while generating
code. However code generation can be aborted and retried at any point
due to various reasons (branch distance or speculative inlining). Such
bailouts and retries would leave dead global pool entries behind.
This CL avoids this by switching to a accumulate&commit scheme for
populating global object pool.
This solution is not entirely clean because we sometimes perform
nested code generation (i.e. to create allocation stubs) which
means that we can't always commit accumulated entries into the
global pool at the same indices as we expected them to be - meaning
that commit can fail and we have to retry compilation.
We try to minimize amount of retries by pregenerating some stubs
ahead of the main code generation.
Change-Id: I5bd3f70272cf49189b754f3a9a2c31d2e6165353
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/126343
Commit-Queue: Vyacheslav Egorov <vegorov@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
Take full advantage of PushPair (stp) in code generation
visitors; use PushRegisterPair for ThrowErrorSlowPathCode.
This change improves Flutter Gallery code size by:
Instr: -0.24%
Total: -0.15%
Change-Id: Icba42cb956b07966861b52d3a8653d21eafab3b9
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/125971
Reviewed-by: Martin Kustermann <kustermann@google.com>
Commit-Queue: Martin Kustermann <kustermann@google.com>
Block R22 to hold a cached version of NullObject();
refresh it in the same way as for BARRIER_MASK register.
Alter assembler to avoid emitting load and to use NR
register directly instead.
This change improves Flutter Gallery code size by:
Instr: -1.81%.
Total: -1.16%.
Change-Id: Ifec654e799737527eec1d8a0e87b4a197ad0298a
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/125662
Commit-Queue: Vyacheslav Egorov <vegorov@google.com>
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
--code-comments should be passed both at compile/training-time and run-time to get
comments in the disassembled instructions.
The required --disassemble-* flags need only be passed at run-time. Flow-graph printing
is not yet supported because we don't save flow-graph like assembly comments.
This functionality is excluded from product builds.
Change-Id: If35fbb08cff6eae40069f916ecec9c6f589c0cad
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/124989
Commit-Queue: Samir Jindel <sjindel@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
Fixes an assumption that CompressedStackMapsIterator::Find() is never
passed a PC offset of 0. Adds back an ASSERT that was dropped which checks
for specific cases where a given PC offset does not have a stack map entry.
Original commit message:
Lifting the PC offset in a2bb730 was a small, lightweight change that
gave us big gains, at least on 32-bit architectures. Here, we make
much more invasive changes that will improve the amount of memory used
by the information previously stored in StackMap objects.
Instead of allocating separate objects for StackMaps, we instead compress
all StackMap information for a given Code object into a single object
(CompressedStackMaps, or CSM for short). This replaces the Array used to
store PC offsets (as Smis) and the individual StackMap objects.
While we lose all canonicalization for individual StackMap entries, the
drop in space required to store stack map information more than offsets that.
-----
The impact on AOT snapshot size when compiling the Flutter Gallery
in release mode:
armv7: Total size -2.58% (Isolate RO: +14.46%, Isolate snapshot: -22.93%)
armv8: Total size -1.85% (Isolate RO: +15.69%, Isolate snapshot: -22.97%)
The impact on in-memory, not on-disk, size for the Flutter Gallery as seen
in the Observatory while running a profile (not release) build:
armv7: Drops from 7.1 MB to 6.2MB (-0.9 MB)
armv8: Drops from 13.5MB to 11.7MB (-1.8 MB)
-----
Bug: https://github.com/dart-lang/sdk/issues/35274, https://github.com/dart-lang/sdk/issues/38873
Cq-Include-Trybots: luci.dart.try:vm-kernel-precomp-linux-debug-x64-try,vm-kernel-precomp-linux-debug-simarm_x64-try,vm-kernel-mac-debug-simdbc64-try
Change-Id: I111b129b0ed64f03184370bceb7cda69d5d4b3c9
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/121700
Commit-Queue: Teagan Strickland <sstrickl@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
Support unaligned memory accesses for regex compiler for arm64
target. Also fix a bug in LoadCodeUnitsInstr IL visitor where
Address::OperandSizeFor was used to get access size (which
returned size of one element independently of actual element
count).
Change-Id: I77c449c8597f8cfd500b8d701f72c39c46a443b7
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/121424
Commit-Queue: Vyacheslav Egorov <vegorov@google.com>
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
This reverts commit d77f4913a1.
Reason for revert: large number of flaky crashes on flakiness dashboard (https://github.com/dart-lang/sdk/issues/38873); crashes on vm-kernel-mac-debug-simdbc64 bot.
Original change's description:
> [vm/compiler] Further compress the information previously in StackMaps.
>
> Lifting the PC offset in a2bb730 was a small, lightweight change that
> gave us big gains, at least on 32-bit architectures. Here, we make
> much more invasive changes that will improve the amount of memory used
> by the information previously stored in StackMap objects.
>
> Instead of allocating separate objects for StackMaps, we instead compress
> all StackMap information for a given Code object into a single object
> (CompressedStackMaps, or CSM for short). This replaces the Array used to
> store PC offsets (as Smis) and the individual StackMap objects.
>
> While we lose all canonicalization for individual StackMap entries, the
> drop in space required to store stack map information more than offsets that.
>
> -----
>
> The impact on AOT snapshot size when compiling the Flutter Gallery
> in release mode:
>
> armv7: Total size -2.58% (Isolate RO: +14.46%, Isolate snapshot: -22.93%)
> armv8: Total size -1.85% (Isolate RO: +15.69%, Isolate snapshot: -22.97%)
>
> The impact on in-memory, not on-disk, size for the Flutter Gallery as seen
> in the Observatory while running a profile (not release) build:
>
> armv7: Drops from 7.1 MB to 6.2MB (-0.9 MB)
> armv8: Drops from 13.5MB to 11.7MB (-1.8 MB)
>
> -----
>
> Bug: https://github.com/dart-lang/sdk/issues/35274
> Cq-Include-Trybots: luci.dart.try:vm-kernel-precomp-linux-debug-x64-try,vm-kernel-precomp-linux-debug-simarm_x64-try
> Change-Id: Ie3bb898d557215146260a560423f5fa27bdff512
> Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/119640
> Commit-Queue: Teagan Strickland <sstrickl@google.com>
> Reviewed-by: Martin Kustermann <kustermann@google.com>
> Reviewed-by: Ryan Macnak <rmacnak@google.com>
TBR=kustermann@google.com,rmacnak@google.com,sstrickl@google.com
# Not skipping CQ checks because original CL landed > 1 day ago.
Bug: https://github.com/dart-lang/sdk/issues/35274
Change-Id: Ib844b4d52e3aa4aca62a896cfe349c680bda7461
Cq-Include-Trybots: luci.dart.try:vm-kernel-precomp-linux-debug-x64-try, vm-kernel-precomp-linux-debug-simarm_x64-try
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/121642
Reviewed-by: Alexander Markov <alexmarkov@google.com>
Commit-Queue: Alexander Markov <alexmarkov@google.com>
Lifting the PC offset in a2bb730 was a small, lightweight change that
gave us big gains, at least on 32-bit architectures. Here, we make
much more invasive changes that will improve the amount of memory used
by the information previously stored in StackMap objects.
Instead of allocating separate objects for StackMaps, we instead compress
all StackMap information for a given Code object into a single object
(CompressedStackMaps, or CSM for short). This replaces the Array used to
store PC offsets (as Smis) and the individual StackMap objects.
While we lose all canonicalization for individual StackMap entries, the
drop in space required to store stack map information more than offsets that.
-----
The impact on AOT snapshot size when compiling the Flutter Gallery
in release mode:
armv7: Total size -2.58% (Isolate RO: +14.46%, Isolate snapshot: -22.93%)
armv8: Total size -1.85% (Isolate RO: +15.69%, Isolate snapshot: -22.97%)
The impact on in-memory, not on-disk, size for the Flutter Gallery as seen
in the Observatory while running a profile (not release) build:
armv7: Drops from 7.1 MB to 6.2MB (-0.9 MB)
armv8: Drops from 13.5MB to 11.7MB (-1.8 MB)
-----
Bug: https://github.com/dart-lang/sdk/issues/35274
Cq-Include-Trybots: luci.dart.try:vm-kernel-precomp-linux-debug-x64-try,vm-kernel-precomp-linux-debug-simarm_x64-try
Change-Id: Ie3bb898d557215146260a560423f5fa27bdff512
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/119640
Commit-Queue: Teagan Strickland <sstrickl@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
Reviewed-by: Ryan Macnak <rmacnak@google.com>
Instead of storing the PC offset inside the StackMap object, store it instead
in the `Array` containing `StackMap`s in `Code` objects. That is, the `Array`
provided by `Code::stackmaps()` no longer contains just `StackMap` objects, but
instead contains `Smi`s and `StackMap` objects in alternating fashion. Each
`Smi` is the PC offset for the `StackMap` object that follows.
This ends up changing very little code outside of `StackMap`,
`Code::GetStackMap`, and `StackMapTableBuilder`, as there are only two types of
`StackMap` users:
* Users that call `Code::GetStackMap` already have the PC offset.
* Users that call Code::stackmaps() can just fetch the PC offset from the
returned `Array` instead.
On 64-bit architectures, we will use more space to represent the PC offset
as a Smi in the Array than the old uint32_t field. However, the drop in total
number of StackMap objects due to an increased ability to canonicalize
them should offset this. On 32-bit architectures, we can only represent
30 bit PC offsets, not 32 bit PC offsets, but that shouldn't be a problem
in practice except for pathological cases.
_Numbers from building the Flutter gallery in android_release mode_
Since PC offsets are no longer in the `StackMap` objects, this enables
a lot more canonicalization than before. Previously, we generated
49379 `StackMap`s, but now we only generate 16139 `StackMap`s, just under
a third of the original number. This is because there were a lot of
`StackMap`s that differed only in their PC offset, and now they can
be canonicalized into the same `StackMap` object.
When building the Flutter gallery with android_release, the app.so size
drops from 11276896 bytes to 10908256 bytes, a difference of 368640
bytes, or 3.27%.
Using the AOT snapshot profiling support, we see the following drops:
Heap snapshot size drops from 10.7 MB to 10.4 MB.
`Code` | Before | After | Difference
---------------------------------------------------------
Shallow size | 352363 | 352363 | 0
Retained size | 9114801 | 8796064 | -318737
Percent of snapshot | 81.2% | 80.7% | -0.5%
`(RO)StackMap` | Before | After | Difference
---------------------------------------------------------
Shallow size | 49381 | 16141 | -33240
Retained size | 893965 | 286613 | -607352
Percent of snapshot | 7.97% | 2.63% | -5.34%
`StackMap` | Before | After | Difference
---------------------------------------------------------
Shallow size | 844584 | 270472 | -574112
Retained size | 844584 | 270472 | -574112
Percent of snapshot | 7.53% | 2.48% | -5.05%
As we'd expect from the `StackMap` numbers above, we end up using a little
under a third of the space for `StackMap`s. We actually use even less space
(32.0% of the original) than the drop in `StackMap` numbers (32.7%), because
each `RawStackMap` instance is 32 bits smaller due to dropping the `pc_offset_`
field. Note that even though these PC offsets now show up in the
`Code::stackmaps()` `Array`, we can see this is still a net drop in space used
by looking at the retained size of `Code` objects.
Bug: https://github.com/dart-lang/sdk/issues/35274
Change-Id: I0910a43e7a5a7e2e721676209196be1884c5a71c
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/119147
Commit-Queue: Teagan Strickland <sstrickl@google.com>
Reviewed-by: Ryan Macnak <rmacnak@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>
Rationale:
Some processors may support just popcnt and, in theory,
some just lzcnt. This change makes the test more
specific. The bsf/bsr tests are never skipped anymore.
Change-Id: I0e7b8772789473ce238f0c5c93378c06c39c695b
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/117940
Reviewed-by: Ryan Macnak <rmacnak@google.com>
Commit-Queue: Aart Bik <ajcbik@google.com>
Fixes the changes to assembly snapshot writer in SIMARM_X64 mode. This unfortunately can't be tested until the ELF loader is available.
Take 2 is in patchset 1.
Change-Id: Ib8b067dd4f09dcba3b142705e9cb4dfbf4e3eb53
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/117726
Commit-Queue: Samir Jindel <sjindel@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
Daco Harkes