This CL introduces typed optimizations for Turboshaft, which replaces all operations that produce a constant output (and don't have side effects) by the corresponding constant.
In addition, a new pass for eliminating dead code is introduced that cannot only remove dead operations, but also rewrite branches that are not required into GotoOps.
Drive-by: Introduce -0 as a "special value" for Float32Type and Float64Type to fix a few issues where 0 and -0 have been treated as identical.
Bug: v8:12783
Change-Id: Ia1450ad7a9abb5d58c7d753596ed08a33a73184f
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4110993
Reviewed-by: Darius Mercadier <dmercadier@chromium.org>
Commit-Queue: Nico Hartmann <nicohartmann@chromium.org>
Cr-Commit-Position: refs/heads/main@{#85143}
This reverts commit ebd933037e.
Reason for revert: Breaks a test: https://logs.chromium.org/logs/v8/buildbucket/cr-buildbucket/8792462319927467985/+/u/OptimizeForSize/CreateIsolateFromReadOnlySnapshot
Original change's description:
> [flags,testrunner] Consider readonly flags for conflict detection
>
> Flag conflict detection 1) bails out on incompatible flag values (e.g.
> --jitless and --turbofan) and 2) handles such bailouts transparently in
> the test runner by marking affected tests as OUTCOMES_FAIL.
>
> This CL adds full support for readonly flags to this system, together
> with required additional annotations in variants.py.
>
> Drive-by: assert proper use of v8_enable_slow_dchecks, and add
> support when dcheck_always_on is set.
> Drive-by: introduce has_maglev build variable detection based on
> v8_enable_maglev and use that for .status file annotations.
> Drive-by: protect against unintended overwrites of build variables
> in statusfile.py.
>
> Cq-Include-Trybots: luci.v8.try:v8_linux64_fyi_rel
> Bug: v8:13629,v8:10577
> Change-Id: I04de399139a0490806df8bfee7e75e2ec767b4b5
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4135879
> Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
> Reviewed-by: Victor Gomes <victorgomes@chromium.org>
> Commit-Queue: Jakob Linke <jgruber@chromium.org>
> Cr-Commit-Position: refs/heads/main@{#85130}
Bug: v8:13629,v8:10577
Change-Id: I0cb072c6c9f05d92894cc0af83c4d1a28df100d5
Cq-Include-Trybots: luci.v8.try:v8_linux64_fyi_rel
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4147098
Bot-Commit: Rubber Stamper <rubber-stamper@appspot.gserviceaccount.com>
Commit-Queue: Victor Gomes <victorgomes@chromium.org>
Reviewed-by: Victor Gomes <victorgomes@chromium.org>
Auto-Submit: Leszek Swirski <leszeks@chromium.org>
Reviewed-by: Michael Achenbach <machenbach@chromium.org>
Cr-Commit-Position: refs/heads/main@{#85133}
Flag conflict detection 1) bails out on incompatible flag values (e.g.
--jitless and --turbofan) and 2) handles such bailouts transparently in
the test runner by marking affected tests as OUTCOMES_FAIL.
This CL adds full support for readonly flags to this system, together
with required additional annotations in variants.py.
Drive-by: assert proper use of v8_enable_slow_dchecks, and add
support when dcheck_always_on is set.
Drive-by: introduce has_maglev build variable detection based on
v8_enable_maglev and use that for .status file annotations.
Drive-by: protect against unintended overwrites of build variables
in statusfile.py.
Cq-Include-Trybots: luci.v8.try:v8_linux64_fyi_rel
Bug: v8:13629,v8:10577
Change-Id: I04de399139a0490806df8bfee7e75e2ec767b4b5
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4135879
Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
Reviewed-by: Victor Gomes <victorgomes@chromium.org>
Commit-Queue: Jakob Linke <jgruber@chromium.org>
Cr-Commit-Position: refs/heads/main@{#85130}
This reverts commit 31ccfed461.
Reason for revert: Fails compilation on: https://ci.chromium.org/ui/p/v8/builders/ci/V8%20Linux64%20-%20cfi%20-%20builder/6527/overview
Original change's description:
> [x64] Add support for "cold calls" in hot paths
>
> This makes (specially annotated) calls to "cold functions" in hot paths
> more efficient by hiding the fact that we are actually calling a
> function here. Clang would otherwise unconditionally spill and reload
> registers that might be clobbered by the call. This would slow down the
> fast path.
>
> This CL allows to reverse priorities here: The fast path can stay fast
> (no spills and loads), but the slow path gets even slower. The inline
> assembly that implements the cold call spills and reloads *all*
> registers, because we do not know which registers are in use in the
> scope where the cold call is being emitted.
>
> I.e. this behaves like a custom calling convention with no caller-saved
> registers.
>
> The `preserve_all` attribute (experimental in clang, and incomplete for
> C++) would also solve this, but it is not production-ready yet (leads to
> crashes of clang and crashes of the generated code).
>
> R=leszeks@chromium.org
> CC=dlehmann@chromium.org
>
> Bug: v8:13565, v8:13570
> Change-Id: I2b54a480da1c689113a67c601c29d73239b0ff2b
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4116584
> Commit-Queue: Clemens Backes <clemensb@chromium.org>
> Reviewed-by: Anton Bikineev <bikineev@chromium.org>
> Reviewed-by: Leszek Swirski <leszeks@chromium.org>
> Cr-Commit-Position: refs/heads/main@{#85127}
Bug: v8:13565, v8:13570
Change-Id: I2f5b3343eb372fea13d2c4ab6354f2bc52e2c338
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4145819
Bot-Commit: Rubber Stamper <rubber-stamper@appspot.gserviceaccount.com>
Auto-Submit: Clemens Backes <clemensb@chromium.org>
Commit-Queue: Rubber Stamper <rubber-stamper@appspot.gserviceaccount.com>
Cr-Commit-Position: refs/heads/main@{#85128}
This makes (specially annotated) calls to "cold functions" in hot paths
more efficient by hiding the fact that we are actually calling a
function here. Clang would otherwise unconditionally spill and reload
registers that might be clobbered by the call. This would slow down the
fast path.
This CL allows to reverse priorities here: The fast path can stay fast
(no spills and loads), but the slow path gets even slower. The inline
assembly that implements the cold call spills and reloads *all*
registers, because we do not know which registers are in use in the
scope where the cold call is being emitted.
I.e. this behaves like a custom calling convention with no caller-saved
registers.
The `preserve_all` attribute (experimental in clang, and incomplete for
C++) would also solve this, but it is not production-ready yet (leads to
crashes of clang and crashes of the generated code).
R=leszeks@chromium.org
CC=dlehmann@chromium.org
Bug: v8:13565, v8:13570
Change-Id: I2b54a480da1c689113a67c601c29d73239b0ff2b
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4116584
Commit-Queue: Clemens Backes <clemensb@chromium.org>
Reviewed-by: Anton Bikineev <bikineev@chromium.org>
Reviewed-by: Leszek Swirski <leszeks@chromium.org>
Cr-Commit-Position: refs/heads/main@{#85127}
This will prevent building v8 arm code when all we want is the snapshot
file.
Bug: chromium:1402705
Change-Id: If8137e8ef79ff1943c24d5bd71d8374e59bdf561
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4117695
Commit-Queue: Adam Klein <adamk@chromium.org>
Auto-Submit: Andrew Grieve <agrieve@chromium.org>
Reviewed-by: Adam Klein <adamk@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84981}
This CL implements a new %CheckTurboshaftTypeOf(e, type_string)
intrinsic allowing tests to express that the expression e is supposed
to have the turboshaft type expressed by type_string eventually during
lowering.
Test that use this intrinsic are verifying implementation details and
are thus somewhat brittle and potentially platform depedent. This
intrinsic is not supposed to be used broadly, but rather to write
some tests that check the precision of turboshaft's new type system.
This intrinsic may be removed once the type system is shipped and gets
coverage in other ways.
Bug: v8:12783
Change-Id: I4cc2582273f3d668601a3203c400a8461b470cac
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4061889
Reviewed-by: Darius Mercadier <dmercadier@chromium.org>
Commit-Queue: Nico Hartmann <nicohartmann@chromium.org>
Reviewed-by: Maya Lekova <mslekova@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84894}
... and any other node needed to test it.
Bug: v8:7700
Change-Id: Ia37fdcb1db3b6fb986f026696454d443236d011c
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4111600
Reviewed-by: Patrick Thier <pthier@chromium.org>
Auto-Submit: Victor Gomes <victorgomes@chromium.org>
Commit-Queue: Patrick Thier <pthier@chromium.org>
Commit-Queue: Victor Gomes <victorgomes@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84893}
This stores the CallDescriptors used for i64 <-> BigInt conversion
builtins as process-globals on the WasmEngine, instead of creating
them as temporaries whenever they're needed.
The primary purpose of this change is to simplify the interface to
the Int64Lowering by eliminating its "special case" parameter, in
preparation for moving that lowering to a different point in the
compilation pipeline.
A minor secondary benefit is that this will save a little bit of
(repeated) Zone memory usage.
Bonus change: drop the Int64Lowering from 64-bit builds, where it
isn't used, but was compiled in up to now. This saves ~17 KiB on
Android-arm64 builds.
Change-Id: Ib35d2e0f772110652eb05abd4c42d848108164b9
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4110898
Auto-Submit: Jakob Kummerow <jkummerow@chromium.org>
Commit-Queue: Jakob Kummerow <jkummerow@chromium.org>
Commit-Queue: Maya Lekova <mslekova@chromium.org>
Reviewed-by: Maya Lekova <mslekova@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84884}
Add gen-static-roots.py to conveniently re-generate the static roots
table when it needs changing.
Additionally, ensure the first read-only page is allocated as first page
during mksnapshot, to move static roots closer to start.
Bug: v8:13466
Change-Id: Ie72b64d0ad0dd3e5fccd3b41e8ed00a4a55a0033
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4096481
Reviewed-by: Toon Verwaest <verwaest@chromium.org>
Auto-Submit: Olivier Flückiger <olivf@chromium.org>
Commit-Queue: Olivier Flückiger <olivf@chromium.org>
Reviewed-by: Igor Sheludko <ishell@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84844}
Build mode for serializing the read only heap as a memory dump in the
startup snapshot. This makes compressed pointers of root objects
statically known at mksnapshot time.
This CL also adds a feature to mksnapshot to dump the static addresses
to a C++ header file. This will allow us to use these addresses in the
future.
The mode is disabled for now since we need some build infrastructure
first to conveniently re-generate the table when the layout changes.
Bug: v8:13466
Change-Id: I975b15bd89fedf713fb7d12b4929935ece78139d
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4056181
Commit-Queue: Olivier Flückiger <olivf@chromium.org>
Reviewed-by: Michael Lippautz <mlippautz@chromium.org>
Reviewed-by: Igor Sheludko <ishell@chromium.org>
Reviewed-by: Toon Verwaest <verwaest@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84815}
Shared heap is not available without write barriers, therefore skip all
tests that require shared heap when write barriers are disabled.
Change-Id: I66a9a94941ffab792ced964f12df6930b13c92cb
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4079120
Reviewed-by: Camillo Bruni <cbruni@chromium.org>
Commit-Queue: Patrick Thier <pthier@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84682}
This reverts commit 0bd121f8e6.
MemoryAnalyzer wasn't calling LateEscapeAnalysisReducer's
ShouldSkipOperation method, but instead was calling the BaseReducer's
method (because it was using a generic Operation) for the call, which
resulted in some memory corruptions, because MemoryAnalyzer was
planning some folding which was never actually happening.
Original change's description:
> [turboshaft] Port LateEscapeAnalysis
>
> Bug: v8:12783
> Change-Id: Id5fa026d103dc67e05322b725f34186124bc5936
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4054621
> Commit-Queue: Darius Mercadier <dmercadier@chromium.org>
> Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
> Cr-Commit-Position: refs/heads/main@{#84603}
Bug: v8:12783
Change-Id: I103eb2f518943c0c57bc3e10471d1c47f5262599
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4075724
Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
Commit-Queue: Darius Mercadier <dmercadier@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84660}
This differs from the patch that landed in so far as the libm target
is only defined if v8_use_libm_trig_functions is defined. Doing this
ensures building the 'all' target only builds libm is appropriate.
You can diff between patchset 1 and 2 to see the change.
This is controlled by a gn arg, which defaults to true for clang
builds. I'm limiting to clang builds as the macros for determining
endian type are currently clang specific. My understanding is that
chrome only uses clang. I can update the endian macros if necessary
for other targets.
Bug=v8:13477
Change-Id: I59cd450facc9fcb8987fe56e8cfc1c13522e1f6d
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4070924
Commit-Queue: Hannes Payer <hpayer@chromium.org>
Auto-Submit: Scott Violet <sky@chromium.org>
Reviewed-by: Hannes Payer <hpayer@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84620}
Rename the vreg allocation to "SetValueLocationConstraints", which is a
more accurate description of what it does. Also, remove the actual vreg
state, since we don't care about unallocated operand vregs (ValueNodes
act as their own vreg).
This allows us to remove a post-processing pass which sets input
constraints for Phi nodes, which previously couldn't do so in the foward
pass due to loop Phis' back-edge inputs not having a vreg yet.
Bug: v8:7700
Change-Id: I24e75f0f600aa5fbd2169a189c0863e4153f3439
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4075529
Reviewed-by: Victor Gomes <victorgomes@chromium.org>
Auto-Submit: Leszek Swirski <leszeks@chromium.org>
Commit-Queue: Leszek Swirski <leszeks@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84615}
This is split out from
https://chromium-review.googlesource.com/c/v8/v8/+/3963708, and
implements only the change to using a dispatcher code stub for different
object types in GetOwnPropertyDescriptor, which returns a
PropertyDescriptor struct, and converts the descriptor into a JSObject
in the final step of Reflect.getOwnPropertyDescriptor or
Object.getOwnPropertyDescriptor.
This does not include the path in the dispatcher for Proxy objects, and
thus Proxy objects are still handled in the runtime.
Change-Id: I3960615aa0941d02d32717ccea291f6b63ae180e
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4024866
Commit-Queue: Caitlin Potter <caitp@igalia.com>
Reviewed-by: Toon Verwaest <verwaest@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84602}
A gn build with no specified targets builds all visible targets. The
postmortem-metadata action is visible, but its output is only needed
if v8_postmortem_support is enabled, hence we now guard the whole
action by the flag.
Depends on https://crrev.com/c/4064686
Bug: chromium:1394626
Change-Id: I8444021cfe253cedee5c1578c733a3ad973726db
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4061865
Reviewed-by: Victor Gomes <victorgomes@chromium.org>
Reviewed-by: Liviu Rau <liviurau@chromium.org>
Commit-Queue: Michael Achenbach <machenbach@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84599}
This reverts commit 4588fe544f.
Reason for revert: Fails on MSVC Windows builds: https://ci.chromium.org/ui/p/v8/builders/ci/V8%20Win64%20-%20msvc%20-%20builder/4779/overview
Original change's description:
> adds the ability for v8 to use sin/cos from libm
>
> This is controlled by a gn arg, which defaults to true for clang
> builds. I'm limiting to clang builds as the macros for determining
> endian type are currently clang specific. My understanding is that
> chrome only uses clang. I can update the endian macros if necessary
> for other targets.
>
> Bug=v8:13477
>
> Change-Id: I604f99a2464b1d57f792bb339f9240ef043251e7
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4000442
> Reviewed-by: Hannes Payer <hpayer@chromium.org>
> Commit-Queue: Scott Violet <sky@chromium.org>
> Cr-Commit-Position: refs/heads/main@{#84577}
Change-Id: I96a1571196fe658568c626e5d36559e496ba4d45
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4067303
Bot-Commit: Rubber Stamper <rubber-stamper@appspot.gserviceaccount.com>
Commit-Queue: Matthias Liedtke <mliedtke@chromium.org>
Owners-Override: Matthias Liedtke <mliedtke@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84586}
This is controlled by a gn arg, which defaults to true for clang
builds. I'm limiting to clang builds as the macros for determining
endian type are currently clang specific. My understanding is that
chrome only uses clang. I can update the endian macros if necessary
for other targets.
Bug=v8:13477
Change-Id: I604f99a2464b1d57f792bb339f9240ef043251e7
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4000442
Reviewed-by: Hannes Payer <hpayer@chromium.org>
Commit-Queue: Scott Violet <sky@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84577}
Profiles for 64/32 bits architectures are not interchangeable but
profiles collected for x64/x86 can be used for arm64/arm respectively.
Attempt to use an incompatible profile is now a fatal build error.
Note that some Float64RoundXXX instructions that are available on
arm64 are not available on x64, so we generate a bunch of code
containing branches instead. This means that we are safe to use x64
profiles for arm64 but not the other way round, otherwise we'll miss
the profile info for the subgraph that's not generated on arm64.
Bug: v8:10470
Change-Id: I6a34836866363207b5ed767e343100b406ac7990
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4055274
Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
Commit-Queue: Igor Sheludko <ishell@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84564}
As this is by now equivalent to v8_enable_sandbox.
Bug: v8:10391
Change-Id: I1eef40a86b984a1867de3513a108ece0d43ec394
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4055273
Commit-Queue: Samuel Groß <saelo@chromium.org>
Reviewed-by: Igor Sheludko <ishell@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84494}
... instead of computing them on the fly. This approach seems to
perform slightly better because it requires less code.
Bug: v8:7703, v8:11460
Change-Id: If31a06fbc748251c491c011e9e3f118665e20159
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4020456
Reviewed-by: Dominik Inführ <dinfuehr@chromium.org>
Commit-Queue: Igor Sheludko <ishell@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84413}
The CL implements an old-V8-to-young-Oilpan remembered set together with
a generational barrier.
Bug: v8:13475
Change-Id: I5f09f7c6db397f2a49cb0c47fd758a1604af4e83
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4030433
Reviewed-by: Omer Katz <omerkatz@chromium.org>
Commit-Queue: Anton Bikineev <bikineev@chromium.org>
Reviewed-by: Michael Lippautz <mlippautz@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84365}
Invariant in the normal mode:
- If the data is invalid, we'll fail gracefully (no crash, no DCHECK
failures)
Invariant in the hard fail mode:
- If the data is invalid (in a way we can detect), a CHECK fails at the
earliest location where we detect the inconsistency
Bug: chromium:1381404
Change-Id: Icae077a5c76329018fdb759122297134ae70b897
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4013142
Reviewed-by: Camillo Bruni <cbruni@chromium.org>
Commit-Queue: Marja Hölttä <marja@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84265}
This prepares work for speeding up gcmole infrastructure by adding
more checks and testing.
- Add a presubmit check that verifies gcmole directives in gn files.
- Simplify gcmole directives in gn files. Drop all but the
architecture property as the others were unused. Simplify regular
expressions.
- Fix some gcmole directives (fly by - only on untested
architectures).
- Add a python test for the changed function in gcmole.py. There's
no other testing yet and this starts somewhere. The test simulates
real BUILD.gn files via checked-in test data.
Bug: v8:12660
Change-Id: I9ca3152b2c3b806de5d4509c152283b36294950d
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4020268
Commit-Queue: Michael Achenbach <machenbach@chromium.org>
Reviewed-by: Alexander Schulze <alexschulze@chromium.org>
Reviewed-by: Leszek Swirski <leszeks@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84239}
This CL refactors the trampoline that pushes the values of callee-saved
registers on the stack, which used before stack scanning.
At the low level, it defines simpler architecture-specific functions
that save the values of these registers in a caller-supplied buffer of
the appropriate size.
The trampoline is now implemented using this mechanism. However, the
low-level functions will be used in subsequent CLs for storing the
registers without using a callback, when setting a stack marker for
conservative stack scanning.
Bug: v8:13257
Change-Id: I86dae66e8613b839c694dc004747e04d1dfad7c2
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3989143
Reviewed-by: Omer Katz <omerkatz@chromium.org>
Commit-Queue: Nikolaos Papaspyrou <nikolaos@chromium.org>
Reviewed-by: Anton Bikineev <bikineev@chromium.org>
Reviewed-by: Michael Lippautz <mlippautz@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84160}
The idea is that later maglev-ir.cc would invoke higher
level macros (AllocateString, CharCodeAt, etc)
and these inline computations can then be shared.
Bug: v8:7700
Change-Id: Icb279cc335515263e1ec29c61ba318f33143f9b8
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4000484
Auto-Submit: Victor Gomes <victorgomes@chromium.org>
Reviewed-by: Leszek Swirski <leszeks@chromium.org>
Commit-Queue: Leszek Swirski <leszeks@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84029}
