This is similar to the previous SIMD spilling fixes, but this time at
block merges. The logic is similar to the existing cases, but not quite
the same. I did not find a nice way to unify the different locations
where we check for SIMD register overlap.
R=thibaudm@chromium.org
Bug: chromium:1283395, v8:12330
Change-Id: I5ab9b6831368cbce40b8368e4ec7954e985bff96
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3404780
Reviewed-by: Thibaud Michaud <thibaudm@chromium.org>
Commit-Queue: Clemens Backes <clemensb@chromium.org>
Cr-Commit-Position: refs/heads/main@{#78720}
Save the PC in the jump buffer and implement the suspend builtin.
R=ahaas@chromium.org
CC=fgm@chromium.org
Bug: v8:12191
Change-Id: I1a6d965d7864dce0a572f6c8d7102046dad190fd
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3345006
Reviewed-by: Jakob Kummerow <jkummerow@chromium.org>
Reviewed-by: Andreas Haas <ahaas@chromium.org>
Commit-Queue: Thibaud Michaud <thibaudm@chromium.org>
Cr-Commit-Position: refs/heads/main@{#78715}
Allow Wasm to generate calls directly to Fast API C functions.
This massively reduces the overhead of these calls (~300%).
Currently options parameter is not supported.
This is a reland of
https://chromium-review.googlesource.com/c/v8/v8/+/3364356
with a fix to a data race.
Bug: chromium:1052746
Change-Id: I8c1c255419496d03a94ec2b443329842469586d5
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3398394
Reviewed-by: Maya Lekova <mslekova@chromium.org>
Reviewed-by: Manos Koukoutos <manoskouk@chromium.org>
Reviewed-by: Camillo Bruni <cbruni@chromium.org>
Commit-Queue: Paolo Severini <paolosev@microsoft.com>
Cr-Commit-Position: refs/heads/main@{#78714}
Use the FatalProcessOutOfMemory function such that tooling recognizes
these crashes as OOM's.
Drive-by: Skip one more test that leads to such stack overflows.
Fixed: v8:12555, chromium:1288456
Bug: v8:12472
Change-Id: Ib9203a4aa0487744f7cea9a212aeeffda579ae23
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3401861
Auto-Submit: Jakob Gruber <jgruber@chromium.org>
Reviewed-by: Clemens Backes <clemensb@chromium.org>
Commit-Queue: Clemens Backes <clemensb@chromium.org>
Cr-Commit-Position: refs/heads/main@{#78692}
super.property accesses in heritage positions like `class C extends
super.property` should resolve super in the current scope, not C's
class scope.
Bug: chromium:1282096
Change-Id: I7ef815bc02cfff35a2898ef9f39b133d1114046c
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3400150
Reviewed-by: Marja Hölttä <marja@chromium.org>
Commit-Queue: Shu-yu Guo <syg@chromium.org>
Cr-Commit-Position: refs/heads/main@{#78687}
Apply case-insensitive comparisons not only for the initial character,
but for the entire prefix. This avoids degenerate behavior for patterns
like /aaaa|AAAA|AAAA/i (i.e. generate a single 4-char prefix instead of
four 1-char prefixes).
Bug: v8:12472
Change-Id: Ib2b49fe73ca846a1b7ec90056cc64bdf5cf33026
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3398114
Reviewed-by: Patrick Thier <pthier@chromium.org>
Commit-Queue: Jakob Gruber <jgruber@chromium.org>
Cr-Commit-Position: refs/heads/main@{#78668}
Recursive ToNode node generation may overflow the stack for large
graphs. As a quick fix, insert periodic stack overflow checks in
selected ToNode methods.
As a more permanent fix, in the future we could abort gracefully
(instead of crashing on a CHECK), and/or refactor into iterative node
generation.
Bug: v8:12472
Change-Id: Ie5fbe838c5f6a5192d7d9b44bfe6f6c76a8d26e7
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3398112
Reviewed-by: Leszek Swirski <leszeks@chromium.org>
Commit-Queue: Jakob Gruber <jgruber@chromium.org>
Cr-Commit-Position: refs/heads/main@{#78667}
This reverts commit bd72152e7d.
Reason for revert: TSAN reports a data race, please see https://ci.chromium.org/ui/p/v8/builders/ci/V8%20Linux64%20TSAN%20-%20isolates/18124/overview
Original change's description:
> [fastcall] Add Wasm entry for Fast API calls
>
> Allow Wasm to generate calls directly to Fast API C functions.
> This massively reduces the overhead of these calls (~300%).
> Currently options parameter is not supported.
>
> This is a rebase of the work originally done by devsnek in:
> https://chromium-review.googlesource.com/c/v8/v8/+/2718666.
>
> Bug: chromium:1052746
> Change-Id: I1bb1de68b440044cc8a4e528adf9d8e0e6692a07
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3364356
> Reviewed-by: Clemens Backes <clemensb@chromium.org>
> Reviewed-by: Manos Koukoutos <manoskouk@chromium.org>
> Reviewed-by: Maya Lekova <mslekova@chromium.org>
> Commit-Queue: Paolo Severini <paolosev@microsoft.com>
> Cr-Commit-Position: refs/heads/main@{#78664}
Bug: chromium:1052746
Change-Id: I957708cf1cff6ee8f90678ee48428f5c12f75a53
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3398121
Auto-Submit: Maya Lekova <mslekova@chromium.org>
Owners-Override: Maya Lekova <mslekova@chromium.org>
Commit-Queue: Rubber Stamper <rubber-stamper@appspot.gserviceaccount.com>
Bot-Commit: Rubber Stamper <rubber-stamper@appspot.gserviceaccount.com>
Cr-Commit-Position: refs/heads/main@{#78665}
Allow Wasm to generate calls directly to Fast API C functions.
This massively reduces the overhead of these calls (~300%).
Currently options parameter is not supported.
This is a rebase of the work originally done by devsnek in:
https://chromium-review.googlesource.com/c/v8/v8/+/2718666.
Bug: chromium:1052746
Change-Id: I1bb1de68b440044cc8a4e528adf9d8e0e6692a07
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3364356
Reviewed-by: Clemens Backes <clemensb@chromium.org>
Reviewed-by: Manos Koukoutos <manoskouk@chromium.org>
Reviewed-by: Maya Lekova <mslekova@chromium.org>
Commit-Queue: Paolo Severini <paolosev@microsoft.com>
Cr-Commit-Position: refs/heads/main@{#78664}
Previously, StoreOwnIC incorrectly reuses the [[Set]] semantics
when initializing public literal class fields and object literals in
certain cases (e.g. when there's no feedback).
This was less of an issue for object literals, but with public class
fields it's possible to define property attributes while the
instance is still being initialized, or to encounter existing static
"name" or "length" properties that should be readonly. This patch
fixes it by
1) Emitting code that calls into the slow stub when
handling StoreOwnIC with existing read-only properties.
2) Adding extra steps in StoreIC::Store to handle such stores
properly with [[DefineOwnProperty]] semantics.
Bug: v8:12421, v8:9888
Change-Id: I6547320a1caba58c66ee1043cd3183a2de7cefef
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3300092
Reviewed-by: Shu-yu Guo <syg@chromium.org>
Reviewed-by: Toon Verwaest <verwaest@chromium.org>
Commit-Queue: Joyee Cheung <joyee@igalia.com>
Cr-Commit-Position: refs/heads/main@{#78659}
Now that concurrent inlining is shipping on stable, remove support
--no-concurrent-inlining.
Note that it's still possible to run Turbofan exclusively on the
main thread by passing --no-concurrent-recompilation.
Bug: v8:7790, v8:12142, chromium:1240585
Change-Id: I1943bbbcad7dea7e3a3c337c239f14f7d96c23cd
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3308798
Reviewed-by: Liviu Rau <liviurau@chromium.org>
Reviewed-by: Maya Lekova <mslekova@chromium.org>
Commit-Queue: Jakob Gruber <jgruber@chromium.org>
Cr-Commit-Position: refs/heads/main@{#78644}
The fast path has an early return if the two inputs are the same
object. However, this was missing the check that the receiver
is not undefined required by the spec.
This fixes it by first checking that the receiver is a string and
only afterwards checking for reference equality.
Bug: v8:12495
Change-Id: I4c5fc80e09060b013c94b05bbc9da504ddbb5206
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3386602
Auto-Submit: Tobias Tebbi <tebbi@chromium.org>
Reviewed-by: Jakob Gruber <jgruber@chromium.org>
Commit-Queue: Jakob Gruber <jgruber@chromium.org>
Cr-Commit-Position: refs/heads/main@{#78637}
Shell::FetchModuleTree assumes that the module at file_name wasn't
already fetched. Shell::ExecuteModule is calling into
FetchModuleTree without checking if the module is already in the module
map, violating this assumption.
This change fixes this by having Shell::ExecuteModule check for the
existence of the module before calling into Shell::ExecuteModule, the
same way that Shell::DoHostImportModuleDynamically does.
Bug: v8:12530
Change-Id: Ia038cbd1715e85c9c92c4554fd486c657ef952e8
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3388130
Reviewed-by: Marja Hölttä <marja@chromium.org>
Commit-Queue: Marja Hölttä <marja@chromium.org>
Cr-Commit-Position: refs/heads/main@{#78636}
Similar to the case of fixed registers, we need to consider both cases:
A SIMD register might collide with either the low or high FP register,
or the FP register might collide with a previously allocated SIMD
register. We did only consider the first case so far.
R=thibaudm@chromium.org
Bug: chromium:1286253, v8:12330
Change-Id: Id4c995586cc8b97a2e131ee9d3417525e409bcef
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3380597
Reviewed-by: Thibaud Michaud <thibaudm@chromium.org>
Commit-Queue: Clemens Backes <clemensb@chromium.org>
Cr-Commit-Position: refs/heads/main@{#78633}
When FLAG_shared_string_table is true, postMessaging strings will share
instead of copy.
Note that not all operations on shared strings are supported, and shared
strings may be slower than non-shared strings for some operations.
Bug: v8:12007
Change-Id: I3462128e15410d2568868143571571b3025722c1
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3277250
Reviewed-by: Toon Verwaest <verwaest@chromium.org>
Commit-Queue: Shu-yu Guo <syg@chromium.org>
Cr-Commit-Position: refs/heads/main@{#78614}
It is possible for KeyedDefineOwnICKind to go into
ElementsTransitionAndStoreIC_Miss when a computed field key
is a valid index and the lazy feedback allocation is disabled.
Bug: chromium:1277863
Change-Id: If8a81384257647426607495b6e3d8f235913e8f3
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3322634
Reviewed-by: Igor Sheludko <ishell@chromium.org>
Commit-Queue: Joyee Cheung <joyee@igalia.com>
Cr-Commit-Position: refs/heads/main@{#78573}
We introduce a type arrayref, which is a supertype of all array types
and a subtype of dataref. We change array.len to accept values of type
(ref null array).
Drive-by: Fix kEq/kData case in TypecheckJSObject.
Bug: v8:7748
Change-Id: I47c6a4487ddf5e7280c1427f43abe87a97c896bd
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3368105
Reviewed-by: Simon Zünd <szuend@chromium.org>
Reviewed-by: Jakob Kummerow <jkummerow@chromium.org>
Commit-Queue: Manos Koukoutos <manoskouk@chromium.org>
Cr-Commit-Position: refs/heads/main@{#78565}
The original CL introduced a test that does not work when it is executed
concurrently on multiple isolates. This CL skips this test
configuration.
Original change's description:
> [wasm] Lazy compilation after deserialization
>
> The serialization format contains one boolean flag per function which
> specifies whether the function code exists in the serialized module or
> not. With this CL, this boolean flag is extended to a three-value flag
> which indicates whether the function exists, and if not, whether the
> function was executed before serialization. This information can then be
> used upon deserialization to compile only those functions that were
> executed before serialization.
>
> Design doc: https://docs.google.com/document/d/1U3uqq4njqLqFhr1G2sU_bmpQxY-3bvfG55udSb-DvA4/edit?usp=sharing
>
> Bug: v8:12281
Change-Id: I36ce90b37736172aa01c47ab04e154ec8ea2d8aa
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3380590
Reviewed-by: Clemens Backes <clemensb@chromium.org>
Commit-Queue: Andreas Haas <ahaas@chromium.org>
Cr-Commit-Position: refs/heads/main@{#78564}
- Add Suspender.suspendOnReturnedPromise method
- Extend the WasmApiFunctionRef data with the suspender
- Detect wrapped WasmJSFunctions when we resolve the import
For now the generated wrapper is still a regular wasm-to-js wrapper, but
this sets the ground for generating specific wrappers for functions
wrapped by suspendOnReturnedPromise, and to access the suspender from
the wrapper code.
R=ahaas@chromium.org
CC=fgm@chromium.org
Bug: v8:12191
Change-Id: I81cbec6b023507e47e6e1463b5f9b912f807da6a
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3345000
Reviewed-by: Andreas Haas <ahaas@chromium.org>
Reviewed-by: Dominik Inführ <dinfuehr@chromium.org>
Commit-Queue: Thibaud Michaud <thibaudm@chromium.org>
Cr-Commit-Position: refs/heads/main@{#78560}
This reverts commit fbcdb28178.
Reason for revert: New test fails for multiple (concurrent) isolates: https://ci.chromium.org/ui/p/v8/builders/ci/V8%20Linux/45152/overview
Original change's description:
> [wasm] Lazy compilation after deserialization
>
> The serialization format contains one boolean flag per function which
> specifies whether the function code exists in the serialized module or
> not. With this CL, this boolean flag is extended to a three-value flag
> which indicates whether the function exists, and if not, whether the
> function was executed before serialization. This information can then be
> used upon deserialization to compile only those functions that were
> executed before serialization.
>
> Design doc: https://docs.google.com/document/d/1U3uqq4njqLqFhr1G2sU_bmpQxY-3bvfG55udSb-DvA4/edit?usp=sharing
>
> Bug: v8:12281
> Change-Id: I465e31e5422fa45163256be0e6594045865f0174
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3364089
> Reviewed-by: Clemens Backes <clemensb@chromium.org>
> Commit-Queue: Andreas Haas <ahaas@chromium.org>
> Cr-Commit-Position: refs/heads/main@{#78545}
Bug: v8:12281
Change-Id: If0e327d02e8257a4d1cfcf8b82381af11f28e91c
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3377126
Auto-Submit: Clemens Backes <clemensb@chromium.org>
Commit-Queue: Rubber Stamper <rubber-stamper@appspot.gserviceaccount.com>
Bot-Commit: Rubber Stamper <rubber-stamper@appspot.gserviceaccount.com>
Cr-Commit-Position: refs/heads/main@{#78546}
The serialization format contains one boolean flag per function which
specifies whether the function code exists in the serialized module or
not. With this CL, this boolean flag is extended to a three-value flag
which indicates whether the function exists, and if not, whether the
function was executed before serialization. This information can then be
used upon deserialization to compile only those functions that were
executed before serialization.
Design doc: https://docs.google.com/document/d/1U3uqq4njqLqFhr1G2sU_bmpQxY-3bvfG55udSb-DvA4/edit?usp=sharing
Bug: v8:12281
Change-Id: I465e31e5422fa45163256be0e6594045865f0174
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3364089
Reviewed-by: Clemens Backes <clemensb@chromium.org>
Commit-Queue: Andreas Haas <ahaas@chromium.org>
Cr-Commit-Position: refs/heads/main@{#78545}
We should only check the "SIMD sibling" register if we are handling a
SIMD register. This avoids unneeded spills, and in this particular case
ran into a DCHECK because there are only 29 registers, but we tried
checking #29.
R=thibaudm@chromium.org
Bug: v8:12330, v8:1285007
Change-Id: Ife8b295ac958990611ca8816bbfbfb5124a4297d
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3372916
Reviewed-by: Thibaud Michaud <thibaudm@chromium.org>
Commit-Queue: Clemens Backes <clemensb@chromium.org>
Cr-Commit-Position: refs/heads/main@{#78532}
The test was added in https://crrev.com/c/3372910, but needs to be
skipped on non-SIMD hardware because it contains SIMD instructions.
R=thibaudm@chromium.org
Bug: v8:12330, chromium:1284980
Change-Id: Ifaede466b24aea4f9ef6b062414a31698bcca864
No-Try: true
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3372917
Auto-Submit: Clemens Backes <clemensb@chromium.org>
Reviewed-by: Deepti Gandluri <gdeepti@chromium.org>
Commit-Queue: Deepti Gandluri <gdeepti@chromium.org>
Cr-Commit-Position: refs/heads/main@{#78517}
Spilling was already fixed if a fixed SIMD register overlaps with an
allocated FP register, but the other way around was missing: If an odd
FP register (in this case d1) is used as a fixed output register, but
this register is in use as the upper half of a SIMD register (in this
case q0), we did not detect this and would just use overwrite the SIMD
half.
This CL also fixes this case.
R=thibaudm@chromium.org
Bug: v8:12330, chromium:1284980
Change-Id: Id3f98b7accd77e38ab4cd5ff8910aaf5ad96a1ed
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3372910
Reviewed-by: Thibaud Michaud <thibaudm@chromium.org>
Commit-Queue: Clemens Backes <clemensb@chromium.org>
Cr-Commit-Position: refs/heads/main@{#78514}
We implement loop peeling for wasm, currently available behind a flag.
Loops are peeled regardless of size.
Bug: v8:11510
Change-Id: Ia4c883abdee83df632b2611584d608c44e3295c8
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3367615
Reviewed-by: Andreas Haas <ahaas@chromium.org>
Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
Commit-Queue: Manos Koukoutos <manoskouk@chromium.org>
Cr-Commit-Position: refs/heads/main@{#78496}
If a fixed register is defined for an input, we did only spill the
sibling SIMD register if the other sibling was allocated. This is not
correct. If only the sibling is in use (e.g. s1 colliding with q0) we
also have to spill that sibling.
R=mslekova@chromium.org
Bug: chromium:1283042, v8:12330
Change-Id: I6a22eaf461774a0b4603ec3ff17062134a528161
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3359615
Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
Reviewed-by: Maya Lekova <mslekova@chromium.org>
Commit-Queue: Clemens Backes <clemensb@chromium.org>
Cr-Commit-Position: refs/heads/main@{#78483}
The mid-tier register allocator did not handle block merges correctly
where a SIMD register was partially overlapping with a non-SIMD
register. This CL fixes that, and reorders the code to allow for early
exits.
R=mslekova@chromium.org
Bug: chromium:1282224, v8:12330
Change-Id: I2e9275d5c1aaa764ecb63fbf8fa197b68d6b6c3c
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3358294
Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
Reviewed-by: Maya Lekova <mslekova@chromium.org>
Commit-Queue: Clemens Backes <clemensb@chromium.org>
Cr-Commit-Position: refs/heads/main@{#78482}
We unify the implementation of element segment expression entries with
other initializer expressions: we represent them with a {WireBytesRef}
and decode them with {InitExprInterface}. Except for reducing code
duplication, this also fixes a bug where {global.get} entries in element
segments could reference invalid globals.
Changes:
- Change {WasmElemSegment::Entry} to a union of a {WireBytesRef}
initializer expression and a {uint32_t} function index.
- In module-decoder, change parsing of expression entries to use
{consume_init_expr}. Add type checking to
{consume_element_func_index}, to complement type checking happening in
{consume_init_expr}.
- In module-instantiate.cc:
- Move instantiation of indirect tables before loading of element
segments. This way, when we call {UpdateDispatchTables} in
{SetTableEntry}, the indirect table for the current table will also
be updated.
- Consolidate table entry instantiation into {SetTableEntry}, which
handles lazily instantiated functions, or dispatches to
{WasmTableObject::Set}.
- Rename {InitializeIndirectFunctionTables} to
{InitializeNonDefaultableTables}.
- Change {InitializeNonDefaultableTables} and {LoadElemSegmentImpl}
to use {EvaluateInitExpression}.
- Add a test to exclude mutable/non-imported globals from the element
section.
- Update tests as needed.
- Update .js module emission in wasm-fuzzer-common.
Change-Id: I29c541bbca8531e8d0312ed95869c8e78a5a0c57
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3364082
Reviewed-by: Andreas Haas <ahaas@chromium.org>
Commit-Queue: Manos Koukoutos <manoskouk@chromium.org>
Cr-Commit-Position: refs/heads/main@{#78476}
This CL splits the TF type for JSFunction into CallableFunction and
ClassConstructor. This differentiation allows us to lower calls to the
CallFunction Builtin only for functions that we can actually call.
Class Constructors are special, as they are callable but should raise
an exception if called.
By not lowering class constructors to calls to CallFunction (but the
more generall Call) builtin, we can remove the checks for class
constructors from CallFunction (in a follow-up CL).
Bug: chromium:1262750
Change-Id: I399967eb03b2f20d2dcb67aef2243b32c9d3174e
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3350457
Reviewed-by: Nico Hartmann <nicohartmann@chromium.org>
Commit-Queue: Patrick Thier <pthier@chromium.org>
Cr-Commit-Position: refs/heads/main@{#78445}
Private method loads are compiled to a named load of a private brand,
which always loads a BlockContext. This BlockContext holds the private
methods common to all instances of a class. TurboFan currently considers
JSLoadNamed to be of Type::NonInternal(). Private methods break this
assumption, since BlockContext is of Type::OtherInternal().
This CL changes the typing of JSLoadNamed of private brands to be
Type::OtherInternal().
Bug: v8:12500
Change-Id: I91f39747bf9422bd419d299f44152f567d8be8db
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3351167
Reviewed-by: Maya Lekova <mslekova@chromium.org>
Commit-Queue: Shu-yu Guo <syg@chromium.org>
Cr-Commit-Position: refs/heads/main@{#78431}
Without simple FP aliasing, a SIMD register will overlap with two
floating-point registers. If we spill an FP register to use it for a
SIMD operation, we need to make sure to also spill the "sibling" FP
register.
R=leszeks@chromium.org
Bug: v8:12330, chromium:1271244
Change-Id: I7fdc6cb8da35d66b4862a8a913ba4ff906cf05aa
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3347576
Reviewed-by: Maya Lekova <mslekova@chromium.org>
Commit-Queue: Clemens Backes <clemensb@chromium.org>
Cr-Commit-Position: refs/heads/main@{#78414}
The mid-tier register allocator could not handle the case that the same
virtual register was used for
- the input corresponding to the 'same-as-input' output, and
- another 'unique register' input.
In this case, it cannot choose the already assigned register for the
'unique' register. Instead, it needs to allocate a new register and
introduce a gap move to duplicate the input value in two different
registers.
FYI, the instruction where the current logic failed was:
(v5(0), v6(R)) = IA32AddPair v7(R) v7(*) v8(R) v7(R)
(where the last input was marked 'unique').
R=leszeks@chromium.orgCC=thibaudm@chromium.org
Bug: v8:12330, chromium:1272204
Change-Id: Ie4843aa9f5e027afe503e0481a4acdfa325dfe0e
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3347821
Reviewed-by: Leszek Swirski <leszeks@chromium.org>
Reviewed-by: Maya Lekova <mslekova@chromium.org>
Commit-Queue: Clemens Backes <clemensb@chromium.org>
Cr-Commit-Position: refs/heads/main@{#78411}
A SIMD register can "block" more than one FP register. In that case, no
virtual register will be assigned for one of the FP registers. This is
fine, we just need to detect and handle that case correctly.
R=thibaudm@chromium.org
CC=leszeks@chromium.org
Bug: chromium:1271538, v8:12330
Change-Id: I7ec19229445c5ace0782f63945acb89322816540
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3293082
Reviewed-by: Thibaud Michaud <thibaudm@chromium.org>
Reviewed-by: Maya Lekova <mslekova@chromium.org>
Commit-Queue: Clemens Backes <clemensb@chromium.org>
Cr-Commit-Position: refs/heads/main@{#78400}
