This reverts commit 2a71e8637f.
Reason for revert: This probably blocked the V8 roll.
https://chromium-review.googlesource.com/c/chromium/src/+/4116621?tab=checks
Original change's description:
> Reland "[heap] Fix conservative stack scanning for client isolates"
>
> This is a reland of commit 36bac1bcae
>
> Original change's description:
> > [heap] Fix conservative stack scanning for client isolates
> >
> > With this CL, the context of stacks corresponding to all client isolates
> > are saved, so that conservative stack scanning can be used correctly
> > during a shared garbage collection. This happens:
> >
> > 1) in Heap::PerformSharedGarbageCollection, for the stacks of the shared
> > isolate and the initiator;
> > 2) when an isolate's main thread is waiting in a safepoint; and
> > 3) when an isolate's main thread is parked.
> >
> > Bug: v8:13257
> > Change-Id: I9ff060f2c0c1ec12977c70d67d65d9c543e2d165
> > Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4027210
> > Reviewed-by: Michael Lippautz <mlippautz@chromium.org>
> > Reviewed-by: Omer Katz <omerkatz@chromium.org>
> > Commit-Queue: Nikolaos Papaspyrou <nikolaos@chromium.org>
> > Reviewed-by: Dominik Inführ <dinfuehr@chromium.org>
> > Cr-Commit-Position: refs/heads/main@{#84712}
>
> Bug: v8:13257
> Change-Id: I61df6eeca5a28e04eb3a858f7d601bc5f6312e49
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4092737
> Reviewed-by: Dominik Inführ <dinfuehr@chromium.org>
> Reviewed-by: Omer Katz <omerkatz@chromium.org>
> Commit-Queue: Nikolaos Papaspyrou <nikolaos@chromium.org>
> Cr-Commit-Position: refs/heads/main@{#84963}
Bug: v8:13257
Change-Id: I3a235f11e5fe55c476591a5274946aeb6cc9bf6e
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4111178
Bot-Commit: Rubber Stamper <rubber-stamper@appspot.gserviceaccount.com>
Reviewed-by: Dominik Inführ <dinfuehr@chromium.org>
Commit-Queue: Nikolaos Papaspyrou <nikolaos@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84973}
This reverts commit f51e0bb1db.
Reason for revert: Dependent on crrev.com/c/4092737 that is being reverted.
Original change's description:
> [heap] Merge mechanisms for disabling CSS
>
> EmbedderStackStateScope is used to disable conservative stack scanning
> for cppgc when the stack is known to not contain heap pointers. Also,
> DisableConservativeStackScanningScopeForTesting is used to disable CSS
> for the V8 heap in tests that assume a precise GC. Until now, these two
> have used two different mechanisms for disabling CSS. This CL merges
> the two mechanisms and implements the latter scope via the former.
>
> Bug: v8:13257
> Change-Id: Ieca082657854fe2eff9eb5d95a30d48bb8eab44f
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4111954
> Reviewed-by: Michael Lippautz <mlippautz@chromium.org>
> Commit-Queue: Nikolaos Papaspyrou <nikolaos@chromium.org>
> Cr-Commit-Position: refs/heads/main@{#84964}
Bug: v8:13257
Change-Id: Id769af6215a2ed319ec96b354734a5362b2384cf
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4111179
Commit-Queue: Nikolaos Papaspyrou <nikolaos@chromium.org>
Reviewed-by: Dominik Inführ <dinfuehr@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84972}
Maglev job destruction is suprisingly expensive, taking up roughly a
third of total finalization time. Rather than destroying jobs as part of
finalization, re-post them to the concurrent dispatcher to be destroyed
in the background
Bug: v8:7700
Change-Id: I450d8a7b49737504c2aaebbfa7754e0ae15e7f05
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4111238
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@{#84970}
Usage of scratch registers acquired via scratch register scope in
deferred code is not valid, as the scope is not active within deferred
code.
Use proper temporaries (assigned by register allocation) instead.
Bug: v8:7700
Change-Id: If4ef4c09ea6f5a551f44bda15b8eeb1055d07070
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4111170
Reviewed-by: Victor Gomes <victorgomes@chromium.org>
Commit-Queue: Patrick Thier <pthier@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84969}
MakeHeapIterable() in a shared heap isolate requires a global
safepoint. OptimizedCodeIterator uses MakeHeapIterable() but always
entered a local safepoint.
This CL changes OptimizedCodeIterator to request a global safepoint
for shared heap isolates.
Bug: v8:13267
Change-Id: I642a7731f4294cc1558de92d4daf5a1e3603f6cc
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4110854
Commit-Queue: Nikolaos Papaspyrou <nikolaos@chromium.org>
Auto-Submit: Dominik Inführ <dinfuehr@chromium.org>
Commit-Queue: Dominik Inführ <dinfuehr@chromium.org>
Reviewed-by: Nikolaos Papaspyrou <nikolaos@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84967}
EmbedderStackStateScope is used to disable conservative stack scanning
for cppgc when the stack is known to not contain heap pointers. Also,
DisableConservativeStackScanningScopeForTesting is used to disable CSS
for the V8 heap in tests that assume a precise GC. Until now, these two
have used two different mechanisms for disabling CSS. This CL merges
the two mechanisms and implements the latter scope via the former.
Bug: v8:13257
Change-Id: Ieca082657854fe2eff9eb5d95a30d48bb8eab44f
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4111954
Reviewed-by: Michael Lippautz <mlippautz@chromium.org>
Commit-Queue: Nikolaos Papaspyrou <nikolaos@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84964}
This is a reland of commit 36bac1bcae
Original change's description:
> [heap] Fix conservative stack scanning for client isolates
>
> With this CL, the context of stacks corresponding to all client isolates
> are saved, so that conservative stack scanning can be used correctly
> during a shared garbage collection. This happens:
>
> 1) in Heap::PerformSharedGarbageCollection, for the stacks of the shared
> isolate and the initiator;
> 2) when an isolate's main thread is waiting in a safepoint; and
> 3) when an isolate's main thread is parked.
>
> Bug: v8:13257
> Change-Id: I9ff060f2c0c1ec12977c70d67d65d9c543e2d165
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4027210
> Reviewed-by: Michael Lippautz <mlippautz@chromium.org>
> Reviewed-by: Omer Katz <omerkatz@chromium.org>
> Commit-Queue: Nikolaos Papaspyrou <nikolaos@chromium.org>
> Reviewed-by: Dominik Inführ <dinfuehr@chromium.org>
> Cr-Commit-Position: refs/heads/main@{#84712}
Bug: v8:13257
Change-Id: I61df6eeca5a28e04eb3a858f7d601bc5f6312e49
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4092737
Reviewed-by: Dominik Inführ <dinfuehr@chromium.org>
Reviewed-by: Omer Katz <omerkatz@chromium.org>
Commit-Queue: Nikolaos Papaspyrou <nikolaos@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84963}
The spec does not allow side effects on wrapping the exceptions
crossing the realm boundaries. We need to provide an easy way to inspect the exception-thrown cross-realms according to the last TC39 meeting
consensus.
Related spec change: https://github.com/tc39/proposal-shadowrealm/pull/382.
Bug: v8:11989
Change-Id: Ia78d94fd33cba689267aeacd028d662bd4a37fe9
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3618759
Commit-Queue: Chengzhong Wu (legendecas) <legendecas@gmail.com>
Reviewed-by: Shu-yu Guo <syg@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84961}
The write barrier for DescriptorArray was skipping slot recording
unless is_compacting_ was set. However, that flag only applies to
local objects but not shared objects in client isolates.
This CL fixes this by using the same logic as the regular write
barrier for deciding whether to record slots or not.
Bug: v8:13267
Change-Id: Ib7c5567615aa04cf79225fc7613a9db1d4ae0f0e
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4118866
Commit-Queue: Dominik Inführ <dinfuehr@chromium.org>
Reviewed-by: Michael Lippautz <mlippautz@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84958}
With this change property accesses on WasmObject and WasmArray return
undefined instead of throwing.
Furthermore retrieving the prototype of a WasmObject/WasmArray returns
null.
Bug: v8:13523
Change-Id: I12e3b9fb6ac9a7305f001a3b3835854068c3064f
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4110995
Reviewed-by: Toon Verwaest <verwaest@chromium.org>
Commit-Queue: Matthias Liedtke <mliedtke@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84957}
This way we do not lose type information while optimizing.
Bug: v8:7748
Change-Id: I5c25a3863e4400f2f69431e73b30fd7c0e626a27
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4117071
Reviewed-by: Matthias Liedtke <mliedtke@chromium.org>
Commit-Queue: Matthias Liedtke <mliedtke@chromium.org>
Auto-Submit: Manos Koukoutos <manoskouk@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84956}
We add final types to wasm-gc.
- We introduce a `kWasmSubtypeFinalCode` as an alternative to
`kWasmSubtypeCode`.
- Behind a flag, we interpret types behind this code as final, as well
as types outside a subtype definition by default.
- For final types, type checks for call_indirect and ref.test etc. are
reduced to simple type identity checks.
Bug: v8:7748
Change-Id: Iabf147b2a15f43abc4c7d1c582f460dbdc645d66
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4096478
Reviewed-by: Matthias Liedtke <mliedtke@chromium.org>
Reviewed-by: Clemens Backes <clemensb@chromium.org>
Commit-Queue: Manos Koukoutos <manoskouk@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84955}
The SaveFPRegsMode::kSave specializations of CEntry were unused.
Remove this parameter to eliminate dead code.
Bug: v8:13606
Change-Id: If3d6f5382101acd477c5d9559a84c88b02a72123
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4114283
Commit-Queue: Jakob Linke <jgruber@chromium.org>
Reviewed-by: Toon Verwaest <verwaest@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84954}
- Add timing infos to TraceAbortedJob output since aborted jobs still
take time and resources:
[aborted optimizing ... - took 0.005, 17.757, 0.061 ms]
- Add the calling code kind to 'not marking ... for optimization'
since it matters which tier we're currently stuck in:
[not marking function ... (MAGLEV) for optimization: already queued]
Bug: v8:7700
Change-Id: I9a123ddaa58ed310605cd28473f53ce8ea004fd3
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4110747
Reviewed-by: Toon Verwaest <verwaest@chromium.org>
Auto-Submit: Jakob Linke <jgruber@chromium.org>
Commit-Queue: Toon Verwaest <verwaest@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84953}
Error mode: a TypedArray backed by RAB claims it's not backed by RAB.
Drive-by: disable resizability even harder when --harmony-rab-gsab is
not on.
Bug: v8:11111, chromium:1402139
Change-Id: I937c69f6124419cc8d29da0195686bc3b9a5c281
Fixed: chromium:1402139
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4110751
Reviewed-by: Shu-yu Guo <syg@chromium.org>
Commit-Queue: Marja Hölttä <marja@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84952}
- Converts tests in EmbedderTracingTest to UnifiedHeapTest
- Move EmbedderRootsHandler tests to their own file
- Adds support for CppHeap in ManualGCScope
- Drive-by fix typo
Reland: Fix issue with attaching a CppHeap while incremental marking
is already running.
This reverts commit d90a98edc1.
Change-Id: Ifafa9145df3103578c4c7f1b3b0336b4bd9f34dd
Bug: v8:13207
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4110941
Commit-Queue: Michael Lippautz <mlippautz@chromium.org>
Reviewed-by: Dominik Inführ <dinfuehr@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84950}
This change reduce the TruncateInt64ToInt32 and BitcastWord32ToWord64 pair at machine operator reducer, since the upper 32 bits are not important.
With this change, a reduncant mov can be avoid e.g. at,
// script
function func() {
a = [];
for (j = 0; j < 13; j++) {
a[j] = j + 1;
}
return a;
}
// Assembly
// ...
movl rdi,[r8+0xb]
movl rbx,[r8+0xb] // can be removed
REX.W movq r9,rdi
sarl r9, 1
// ...
Change-Id: I6a128d9606c093df50fe9286201c1b979900864d
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4115155
Commit-Queue: Fanchen Kong <fanchen.kong@intel.com>
Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84945}
ShadowRealms have a callable boundary: there cannot be any edges between
objects in the ShadowRealm and those from other realms. V8's
prepareCallStack API breaks this invariant via getFunction() and
getThis(). This CL makes those functions throw when called inside
ShadowRealms.
See also
https://docs.google.com/document/d/1aXEy4YCC9CduxLs7MGw-UOm0P4OuG7W-cScBnLG3ALI/edit?usp=sharing
Bug: v8:11989
Change-Id: I5a2b8fa735c0f10583c8cede4062645986b2d914
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4108810
Reviewed-by: Camillo Bruni <cbruni@chromium.org>
Commit-Queue: Shu-yu Guo <syg@chromium.org>
Reviewed-by: Adam Klein <adamk@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84944}
All IFS values of resumable loops are loop phis, so values from before
the loop can safely be dropped. In fact, if we don't drop them we might
not be able to restore them and crash in a safety check. Eagerly
dropping the values allows us to keep the safety check.
Bug: v8:7700
Change-Id: I99ef80e35db5aba965dc1c5c9d78f1d085e589fb
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4110940
Commit-Queue: Victor Gomes <victorgomes@chromium.org>
Auto-Submit: Toon Verwaest <verwaest@chromium.org>
Reviewed-by: Victor Gomes <victorgomes@chromium.org>
Commit-Queue: Toon Verwaest <verwaest@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84943}
Usually sharable strings are automatically promoted to shared old space
and can be shared in-place.
There are currently two exceptions:
- When using a non-moving GC (e.g. minor MC)
- Strings in LO space
Account for these exceptions by copying strings to the respective shared
space when sharing them.
Bug: chromium:1400048
Change-Id: I20713b5f32f449c14febd848e289b5767530a257
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4110752
Reviewed-by: Dominik Inführ <dinfuehr@chromium.org>
Commit-Queue: Patrick Thier <pthier@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84942}
This is a reland of commit 833731b20b
Original change's description:
> [maglev] Fix %OptimizeOsr when called from Maglev frames
>
> This (test-only) runtime function only supported unoptimized frames as
> callers. Add support for Maglev frames as well by extracting the
> relevant BytecodeArray and bytecode offset.
>
> This reverts commit 955de73ee5.
>
> Bug: chromium:1400549,v8:7700
> Change-Id: I80f80f8736ff0400d6d47e355add2a07cdc4559e
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4111851
> Reviewed-by: Victor Gomes <victorgomes@chromium.org>
> Commit-Queue: Jakob Linke <jgruber@chromium.org>
> Cr-Commit-Position: refs/heads/main@{#84931}
Bug: chromium:1400549,v8:7700
Change-Id: I79fadaa0a82314a3fdd4970e9429d83ab06aff09
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4111950
Auto-Submit: Jakob Linke <jgruber@chromium.org>
Reviewed-by: Victor Gomes <victorgomes@chromium.org>
Commit-Queue: Victor Gomes <victorgomes@chromium.org>
Commit-Queue: Jakob Linke <jgruber@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84938}
There is no need to park the main thread while waiting for
concurrent tasks to finish. In case they allocate threads know from
the TEAR_DOWN state and the shutdown flag in CollectionBarrier that
no further GC is going to happen.
Bug: v8:13267
Change-Id: I19f8cf8a71c2fb91618ed0be5ab82cf134edb9ab
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4111446
Reviewed-by: Michael Lippautz <mlippautz@chromium.org>
Commit-Queue: Dominik Inführ <dinfuehr@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84936}
We use 0xffffffff as a sentinel for "no supertype". Therefore we
should reject it as we parse it. We implement this by rejecting
supertypes outside V8's type definition limit.
Bug: v8:7748
Change-Id: I7942d94073d8f7350528fb0e364e91f7359c8cec
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4110750
Auto-Submit: Manos Koukoutos <manoskouk@chromium.org>
Reviewed-by: Matthias Liedtke <mliedtke@chromium.org>
Commit-Queue: Matthias Liedtke <mliedtke@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84934}
Aligning struct fields to multiples of their own size can leave gaps
between them, e.g. when i8 and i32 fields alternate. This patch
introduces a simple optimization: it keeps track of the most recent
such gap, and attempts to use it for later fields that are small enough.
Bonus changes:
- Cap field alignment to 4 bytes (because we only have 4-byte object
alignment anyway).
- Don't re-compute field offsets when canonicalizing types. Instead,
re-use the original type's offsets.
Bug: v8:7748
Change-Id: Iabfc8e7cda94f16d196ed4429f3aa92d249b3b72
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4092494
Commit-Queue: Manos Koukoutos <manoskouk@chromium.org>
Reviewed-by: Manos Koukoutos <manoskouk@chromium.org>
Auto-Submit: Jakob Kummerow <jkummerow@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84933}
Most verification calls are gated behined both a build flag and a
runtime flag. Some calls were missing the runtime flag.
Bug: v8:12612
Change-Id: I482bf7cd3900e860f9db1932f9490d1af9b19df1
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4085007
Reviewed-by: Igor Sheludko <ishell@chromium.org>
Reviewed-by: Michael Lippautz <mlippautz@chromium.org>
Commit-Queue: Omer Katz <omerkatz@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84932}
This (test-only) runtime function only supported unoptimized frames as
callers. Add support for Maglev frames as well by extracting the
relevant BytecodeArray and bytecode offset.
This reverts commit 955de73ee5.
Bug: chromium:1400549,v8:7700
Change-Id: I80f80f8736ff0400d6d47e355add2a07cdc4559e
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4111851
Reviewed-by: Victor Gomes <victorgomes@chromium.org>
Commit-Queue: Jakob Linke <jgruber@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84931}
Maglev OSR uses the deopt_reason to tell the deoptimizer not to
invalidate optimized code on deopt (since we deopted not because of a
failed assumption, but just as part of the OSR procedure). This deopt
reason must always be emitted for the code to work as expected.
Bug: v8:7700
Change-Id: I96b09eef52b2b90c6c491ffec3f87538124cdc88
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4111165
Auto-Submit: Jakob Linke <jgruber@chromium.org>
Commit-Queue: Victor Gomes <victorgomes@chromium.org>
Reviewed-by: Victor Gomes <victorgomes@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84930}
This is a reland of commit 5d95bd39ca
Original change's description:
> [maglev] Prevent lazy deopts during maglev's JumpLoop (=OSR)
>
> The problem was that synchronous Maglev OSR potentially caused
> code deoptimization during compilation dependency finalization; this
> led to a lazy deopt when returning from the call to
> Runtime_CompileOptimizedOSRFromMaglev. However, a lazy deopt is
> disallowed at this point, since a) Maglev doesn't support marking an opcode as both lazy- and eager deopt, and b) the JumpLoop opcode
> is already marked as eager deopt since that's how OSR is implemented
> under the hood. See also the comment in runtime-compiler.cc.
>
> We fix this by changing synchronous Maglev-to-Turbofan OSR
> behavior s.t. actual OSR compilation is triggered from Ignition
> (and not from Maglev). In other words, when synchronous OSR is
> requested:
>
> 1. trigger an eager deopt from Maglev to Ignition by returning a
> non-null code object from Runtime_CompileOptimizedOSRFromMaglev.
> 2. Ignition handles the pending OSR compile request (through
> osr_urgency).
>
> This CL also reverts previous partial fixes:
>
> This reverts commit 21969e8e24.
> This reverts commit 6bcbcfed5c.
>
> Bug: chromium:1394279,v8:13585,v8:7700
> Change-Id: I3d64aa39575ad806ba2623102092176ca160ef0b
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4110740
> Commit-Queue: Jakob Linke <jgruber@chromium.org>
> Reviewed-by: Victor Gomes <victorgomes@chromium.org>
> Cr-Commit-Position: refs/heads/main@{#84922}
Bug: chromium:1394279,v8:13585,v8:7700
Change-Id: Id9d1a1ab2dc36e481287a1a25863b45bf281920c
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4110746
Commit-Queue: Jakob Linke <jgruber@chromium.org>
Auto-Submit: Jakob Linke <jgruber@chromium.org>
Commit-Queue: Victor Gomes <victorgomes@chromium.org>
Reviewed-by: Victor Gomes <victorgomes@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84928}
... and other nodes needed for testing it.
Bug: v8:7700
Change-Id: I2e71916112bf9ac5ab336adb819aad4c4c5c6f8b
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4111161
Commit-Queue: Patrick Thier <pthier@chromium.org>
Reviewed-by: Patrick Thier <pthier@chromium.org>
Auto-Submit: Victor Gomes <victorgomes@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84927}
Small optimization: We can save about 10% of the calls to
`FastZoneVector::Grow` during validation of large modules (e.g.,
psweb_2021_09.wasm: before 237k, after 215k calls to Grow) by
pre-allocating the `stack_` and `control_` vectors (which are unlikely
to remain empty for most functions). The runtime doesn't improve much
though (before 69.08ms validation time, after 68.45ms), but I guess it
is still worth the fewer re-allocations+copies.
I tried different amounts of pre-allocation as well (8, 16, 32, 64
elements). For both value and control stack 16 was the sweet spot: It
reduced more re-allocations than 8, but after 16 the number of Grow
calls plateaus and instead runtime increases slightly.
Change-Id: Idc57e0a31c6fa1bbdc98bfd4ffd5522d34f09e81
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4107389
Commit-Queue: Daniel Lehmann <dlehmann@chromium.org>
Reviewed-by: Clemens Backes <clemensb@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84926}
This reverts commit 5d95bd39ca.
Reason for revert: https://ci.chromium.org/ui/p/v8/builders/ci/V8%20Linux64%20-%20gc%20stress/2101/overview
Original change's description:
> [maglev] Prevent lazy deopts during maglev's JumpLoop (=OSR)
>
> The problem was that synchronous Maglev OSR potentially caused
> code deoptimization during compilation dependency finalization; this
> led to a lazy deopt when returning from the call to
> Runtime_CompileOptimizedOSRFromMaglev. However, a lazy deopt is
> disallowed at this point, since a) Maglev doesn't support marking an opcode as both lazy- and eager deopt, and b) the JumpLoop opcode
> is already marked as eager deopt since that's how OSR is implemented
> under the hood. See also the comment in runtime-compiler.cc.
>
> We fix this by changing synchronous Maglev-to-Turbofan OSR
> behavior s.t. actual OSR compilation is triggered from Ignition
> (and not from Maglev). In other words, when synchronous OSR is
> requested:
>
> 1. trigger an eager deopt from Maglev to Ignition by returning a
> non-null code object from Runtime_CompileOptimizedOSRFromMaglev.
> 2. Ignition handles the pending OSR compile request (through
> osr_urgency).
>
> This CL also reverts previous partial fixes:
>
> This reverts commit 21969e8e24.
> This reverts commit 6bcbcfed5c.
>
> Bug: chromium:1394279,v8:13585,v8:7700
> Change-Id: I3d64aa39575ad806ba2623102092176ca160ef0b
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4110740
> Commit-Queue: Jakob Linke <jgruber@chromium.org>
> Reviewed-by: Victor Gomes <victorgomes@chromium.org>
> Cr-Commit-Position: refs/heads/main@{#84922}
Bug: chromium:1394279,v8:13585,v8:7700
Change-Id: Ib82d06ab8281f0e59a2af2b631bf93b25064df1f
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4110745
Commit-Queue: Manos Koukoutos <manoskouk@chromium.org>
Bot-Commit: Rubber Stamper <rubber-stamper@appspot.gserviceaccount.com>
Owners-Override: Manos Koukoutos <manoskouk@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84925}
