Currently we count optimizations to decide to disable optimization, and
count deopts to detect this decision and allow re-enabling optimizations
after a while.
However, throwing out TurboFan OSR code and GC optimized code evictions
do not count as deopts, which means that the optimization count
increases without increasing the deopt count. This increased optimization
count disables further optimization -- which is bad, because these are
not "true" deopts -- and can stop the optimization from being re-enabled,
because the deopt count can't go high enough.
Instead, we now only ever look at deopts to disable/re-enable
optimization, and opt counts are only used for naming log files and in
tests.
Change-Id: I0c7d6be497545449a38cf952cd2f007ee51982ba
Reviewed-on: https://chromium-review.googlesource.com/468811
Commit-Queue: Leszek Swirski <leszeks@chromium.org>
Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
Reviewed-by: Jaroslav Sevcik <jarin@chromium.org>
Reviewed-by: Benedikt Meurer <bmeurer@chromium.org>
Reviewed-by: Michael Starzinger <mstarzinger@chromium.org>
Cr-Commit-Position: refs/heads/master@{#44647}
If the new.target passed to the Array constructor cannot be a Proxy,
then we know that the Array constructor cannot trigger any observable
side-effects.
R=jarin@chromium.org
BUG=v8:6262
Review-Url: https://codereview.chromium.org/2821453002
Cr-Commit-Position: refs/heads/master@{#44645}
The JSArgumentsObject::length representation is initially Smi, so we can
record that on the initial map and use it to optimize the accesses in
TurboFan based on that. Similar for JSSloppyArgumentsObject::caller.
BUG=v8:6262
R=yangguo@chromium.org
Review-Url: https://codereview.chromium.org/2810333004
Cr-Commit-Position: refs/heads/master@{#44644}
Previously, we would only consider it if it pointed to a full-code JS function.
Thus we could miss both optimized functions and bytecode handlers if they
called frame-less code.
Review-Url: https://codereview.chromium.org/2822433002
Cr-Commit-Position: refs/heads/master@{#44640}
This fixes the validation of "|0" annotations of call sites that are
used to indicate a "signed" return type of functions. We use lookahead
during call validation and request deferred validation as part of the
actual OR-expression. Special care has to be taken to get precedence
levels of all involved operators right.
R=clemensh@chromium.org
TEST=mjsunit/asm/call-annotation
BUG=v8:6183
Change-Id: If0586f669e7cee26a13425b0fd9f41098e852d68
Reviewed-on: https://chromium-review.googlesource.com/475871
Commit-Queue: Michael Starzinger <mstarzinger@chromium.org>
Reviewed-by: Clemens Hammacher <clemensh@chromium.org>
Cr-Commit-Position: refs/heads/master@{#44639}
The condition only applies in unicode mode, where any lone surrogates
are desugared into a character class (and will not be considered in this
optimization). Non-unicode mode treats lone surrogates exactly like
any other codepoint.
BUG=chromium:711092
Review-Url: https://codereview.chromium.org/2808403006
Cr-Commit-Position: refs/heads/master@{#44638}
In eval scripts, the source code position can be 0 rather
than greater than 0.
Add regression test.
Drive-by fix: unrelated typo.
Bug: 707223
Change-Id: If52c0736daac3ad42ac6d324eb8ec5f1798f6f5a
Reviewed-on: https://chromium-review.googlesource.com/476630
Reviewed-by: Michael Starzinger <mstarzinger@chromium.org>
Commit-Queue: Franziska Hinkelmann <franzih@chromium.org>
Cr-Commit-Position: refs/heads/master@{#44637}
When optimizing calls to get Object.prototype.__proto__ in
JSCallReducer, we can also consume unreliable receiver map
information, as long as the receiver maps are stable. In
that case we also need to install proper stability dependencies.
BUG=chromium:711195,v8:5267,v8:6241
R=jarin@chromium.org
Review-Url: https://codereview.chromium.org/2816993002
Cr-Commit-Position: refs/heads/master@{#44634}
This CL implements the proposed change to show information about
WebAssembly values and call frames via the inspector interface.
Each interpreted WebAssembly frame will have two scopes: A global scope
showing information about the memory (to be extended for globals), and
a local scope showing information about parameters, local variables, and
stack values.
Names of local variables will be added later.
R=ahaas@chromium.org, yangguo@chromium.org
BUG=v8:6245,v8:5822
Change-Id: I0a35fddd0a353933c86adf62083233b08098a2c7
Reviewed-on: https://chromium-review.googlesource.com/474865
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Reviewed-by: Aleksey Kozyatinskiy <kozyatinskiy@chromium.org>
Reviewed-by: Andreas Haas <ahaas@chromium.org>
Cr-Commit-Position: refs/heads/master@{#44633}
This fixes the existing lowering of {JSGetSuperConstructor} nodes to
unconditional throws. The above operator is marked as {kNoWrite} but
runtime calls are not marked as such. Any deoptimizing operation after
the throw would not be able to find a valid {Checkpoint}. We remove the
lowering case in question.
R=bmeurer@chromium.org
TEST=mjsunit/regress/regress-6248
BUG=v8:6248
Change-Id: I22c922947336254502f698b02f944cf35dd8688a
Reviewed-on: https://chromium-review.googlesource.com/476570
Reviewed-by: Benedikt Meurer <bmeurer@chromium.org>
Commit-Queue: Michael Starzinger <mstarzinger@chromium.org>
Cr-Commit-Position: refs/heads/master@{#44632}
When the incoming value to ChangeFloat64ToTagged is in Smi range, we
represent it as Smi instead of a HeapNumber. This addresses a range of
problems where TurboFan unnecessarily deoptimizes because an operation
learned Smi feedback in Ignition, but was then confronted with a tagged
HeapNumber in TurboFan, just because the value was also represented as
unboxed double somewhere in the meantime.
BUG=v8:6256
R=yangguo@chromium.org
Review-Url: https://codereview.chromium.org/2815283002
Cr-Commit-Position: refs/heads/master@{#44631}
This patch delays the DisposeModuleEmbedderData call for
the interative shell case until we exit the RunShell function.
Bug: v8:5785
Change-Id: I01ff76000882cd1d6801fefc9ea3770c3f38c83b
Reviewed-on: https://chromium-review.googlesource.com/476024
Commit-Queue: Sathya Gunasekaran <gsathya@chromium.org>
Reviewed-by: Adam Klein <adamk@chromium.org>
Cr-Commit-Position: refs/heads/master@{#44630}
Adds some benchmarks copied from v8:4762.
BUG=v8:4762, v8:5460
R=adamk@chromium.org
Change-Id: I0b96080042781c2c46c0c8a3896a921bde97c1e5
Reviewed-on: https://chromium-review.googlesource.com/475934
Reviewed-by: Adam Klein <adamk@chromium.org>
Commit-Queue: Caitlin Potter <caitp@igalia.com>
Cr-Commit-Position: refs/heads/master@{#44629}
It's required by the spec -- and observable -- that the index be validated
before the conversion of the value(s) via ToInteger.
The previous implementation also had an old test for validating the atomic
index, which has now been switched to ToIndex.
This also exposed an issue in the ia32 code generator: cmpxchg_b requires a
byte register, but the ia32 instruction selector was ensuring that the
new_value was a byte register, not the TempRegister. This change forces the
temp register to use edx, which always can be used as a byte register (dl).
This is the same behavior as currently used in UseByteRegister.
BUG=v8:4614
R=jarin@chromium.org,jkummerow@chromium.org
Review-Url: https://codereview.chromium.org/2814753003
Cr-Commit-Position: refs/heads/master@{#44626}
When doing VS 2017 official builds of some targets, including
mksnapshot.exe, a warning about a buffer overrun is shown. After
analysis and discussion with Microsoft it was decided that this warning
is spurious. The warning is:
warning C4789: buffer 'key' of size 16 bytes will be overrun; 4
bytes will be written starting at offset 16
Despite the certain language "4 bytes *will* be written..." it is in
fact a heuristic based warning. Suppressing it at this point in the
inlining stack appears to avoid the issues.
R=hablich@chromium.org
BUG=v8:6068
Review-Url: https://codereview.chromium.org/2804033005
Cr-Commit-Position: refs/heads/master@{#44625}
The arm64 MacroAssembler expects buffer_size to be an unsigned, not a
size_t.
BUG=chromium:710913
Review-Url: https://codereview.chromium.org/2818513002
Cr-Commit-Position: refs/heads/master@{#44623}
Optimizations are supposed to be disabled in our stack-trace code when
building with VC++. However the check used #if defined(COMPILER_MSVC)
when that is never defined in v8. The correct define in v8 is
V8_CC_MSVC.
R=hablich@chromium.org
Review-Url: https://codereview.chromium.org/2800043003
Cr-Commit-Position: refs/heads/master@{#44621}
gdb_index is not in declare_args() and has no effect.
NOTRY=true
Change-Id: I88a9558937aa8fea30ab246899bea4a123947f82
Reviewed-on: https://chromium-review.googlesource.com/475772
Commit-Queue: Jakob Kummerow <jkummerow@chromium.org>
Reviewed-by: Jakob Kummerow <jkummerow@chromium.org>
Cr-Commit-Position: refs/heads/master@{#44620}
The BytecodePipeline is no longer used by any optimizers, so remove it and
connect the BytecodeArrayBuilder directly to the BytecodeWriter.
Also remove some functions from BytecodeNode which are no longer used.
BUG=v8:6194
Change-Id: Id2ec94ff1d4db41b108a778100459283fbb2256c
Reviewed-on: https://chromium-review.googlesource.com/471528
Reviewed-by: Leszek Swirski <leszeks@chromium.org>
Commit-Queue: Ross McIlroy <rmcilroy@chromium.org>
Cr-Commit-Position: refs/heads/master@{#44619}
This relands commit a79f903155.
Original change's description:
> [Interpreter] Unify approach to building interpreter handler and Turbofan stubs.
>
> Moves interpreter-generator.cc to a similar model of building handlers as
> Turbofan stubs elsewhere, to simplify moving code between stubs / builtins and
> bytecode handlers. This removes the "__" hack from the Interpreter generator
> code.
>
> Also make SetBytecodeOffset private to InterpreterAssembler and make
> LdaImmutable[Current]ContextSlot and Lda[Current]ContextSlot share
> handlers since they are identical.
>
> Change-Id: I9e91e7d37c2ea75513e4dcc3b95b4bb6517f83da
> Reviewed-on: https://chromium-review.googlesource.com/471987
> Commit-Queue: Ross McIlroy <rmcilroy@chromium.org>
> Reviewed-by: Jakob Kummerow <jkummerow@chromium.org>
> Reviewed-by: Igor Sheludko <ishell@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#44534}
>
TBR=rmcilroy@chromium.org,jkummerow@chromium.org,machenbach@chromium.org,cbruni@chromium.org,leszeks@chromium.org,v8-reviews@googlegroups.com,ishell@chromium.org
Change-Id: I282fe5582f681ccb0642537a70f89185558ee195
Reviewed-on: https://chromium-review.googlesource.com/474755
Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
Reviewed-by: Camillo Bruni <cbruni@chromium.org>
Commit-Queue: Ross McIlroy <rmcilroy@chromium.org>
Cr-Commit-Position: refs/heads/master@{#44617}
Unfortunately, this test cannot test that a function was really skipped (i.e.,
not parsed).
BUG=v8:5516
Change-Id: I8db5027d2216a95cc012ceae8e17554095cc1d4f
Reviewed-on: https://chromium-review.googlesource.com/457037
Reviewed-by: Daniel Vogelheim <vogelheim@chromium.org>
Commit-Queue: Marja Hölttä <marja@chromium.org>
Cr-Commit-Position: refs/heads/master@{#44615}
Reason for revert:
Roll blocker: https://bugs.chromium.org/p/chromium/issues/detail?id=710824
Original issue's description:
> [wasm] instantiate expressed in terms of compile
>
> Today, the semantics of:
>
> WebAssembly.instantiate
>
> and
>
> WebAssembly.compile().then(new WebAssemblyInstance)
>
> are subtly different, to the point where attempting the proposed
> change uncovered bugs.
>
> In the future, it's possible that .instantiate actually have different
> semantics - if we pre-specialized to the provided ffi, for example.
> Right now that's not the case.
>
> This CL:
> - gets our implementation closer to what developers may write using
> the compile -> new Instance alternative, in particular wrt promise
> creation. By reusing code paths, we uncover more bugs, and keep
> maintenance cost lower.
>
> - it gives us the response-based WebAssembly.instantiate implicitly.
> Otherwise, we'd need that same implementation on the blink side. The
> negative is maintenance: imagine if the bugs I mentioned could only be
> found when running in Blink.
>
> BUG=chromium:697028
>
> Review-Url: https://codereview.chromium.org/2806073002
> Cr-Commit-Position: refs/heads/master@{#44592}
> Committed: 7829af3275TBR=bradnelson@chromium.org,ahaas@chromium.org,adamk@chromium.org,mtrofin@chromium.org
# Skipping CQ checks because original CL landed less than 1 days ago.
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
BUG=chromium:697028
Review-Url: https://codereview.chromium.org/2810203002
Cr-Commit-Position: refs/heads/master@{#44614}
The biggest problem is isolate.h (this CL doesn't solve that yet).
BUG=v8:5294
Change-Id: I56b32109f501c48facd99cd12ca6c8f427e188a9
Reviewed-on: https://chromium-review.googlesource.com/471487
Reviewed-by: Yang Guo <yangguo@chromium.org>
Reviewed-by: Michael Starzinger <mstarzinger@chromium.org>
Commit-Queue: Marja Hölttä <marja@chromium.org>
Cr-Commit-Position: refs/heads/master@{#44613}
BUG=v8:4742
R=machenbach@chromium.org,jkummerow@chromium.org
Change-Id: I03e87db1536f33a67593437f8c72c33486ecdbd1
Reviewed-on: https://chromium-review.googlesource.com/474787
Commit-Queue: Loo Rong Jie <loorongjie@gmail.com>
Reviewed-by: Jakob Kummerow <jkummerow@chromium.org>
Reviewed-by: Michael Achenbach <machenbach@chromium.org>
Cr-Commit-Position: refs/heads/master@{#44610}
This adds a fast path to skip runtime calls to GetSubstitution when
the replacer string does not contain a '$' char.
Extended background:
String.prototype.replace is (roughly) structured as follows:
* Check if {searchValue} has a @@replace Symbol, and delegate to that if
so. We currently implement efficient fast paths when {searchValue} is
a String or a fast RegExp.
* A specialized fast path for single-char {searchValue}, "long" subject
string, and String {replaceValue} that do not contain '$' chars (yes,
this fast path is very specialized).
* Check for the location of the first match using StringIndexOf, and
exit early if no match is found.
* Finally build the return value, which is 'prefix + replacement +
suffix', where replacement is either the result of calling {replaceValue}
(if it is callable), or GetSubstitution(ToString({replaceValue}))
otherwise.
There's several spots that could be improved.
StringIndexOf currently calls into C++ runtime for all but the simple
1-byte, 1-char {searchValue} case. We need to finally add support for
remaining cases.
The runtime call to GetSubstitution can be skipped if the replacer
string does not contain any '$' syntax. This CL handles that case.
BUG=
Review-Url: https://codereview.chromium.org/2813843002
Cr-Commit-Position: refs/heads/master@{#44606}
This is necessary to appease "gn check" if gtest_prod.h becomes a part
of the Chromium checkout, instead of a third-party repository brought
over by Chromium's DEPS. The file is already listed in v8's DEPS, but gn
does not use DEPS as an input.
BUG=chromium:630705
Review-Url: https://codereview.chromium.org/2807353002
Cr-Commit-Position: refs/heads/master@{#44604}
The hole NaN should also have proper Type::Hole, and not silently hide
in the Type::Number. This way we can remove all the special casing for
the hole NaN, and we also finally get the CheckNumber right.
This also allows us to remove some ducktape from the Deoptimizer, as for
escape analyzed FixedDoubleArrays we always pass the hole value now to
represent the actual holes.
Also-By: jarin@chromium.org
BUG=chromium:684208,chromium:709753,v8:5267
R=jarin@chromium.org
Review-Url: https://codereview.chromium.org/2814013003
Cr-Commit-Position: refs/heads/master@{#44603}
The local variables were parsed two times, which in fact doubled the
amount of local variables allocated for each called function.
This was costing memory and performance. As the additional local
variables were never used, we did not recognize this before.
Add a test case for locals and stack values of interpreted frames.
R=ahaas@chromium.org
BUG=v8:5822
Change-Id: Ie5cb8d8f5441edee6abb46aa6bebef4a033d582b
Reviewed-on: https://chromium-review.googlesource.com/474749
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Reviewed-by: Andreas Haas <ahaas@chromium.org>
Cr-Commit-Position: refs/heads/master@{#44602}
Similar to WasmInterpreter::Thread, we now also use the pimpl idiom for
InterpretedFrame, hiding the implementation completely in the .cc file.
This allows us to store just two things per InterpretedFrameImpl: The
corresponding thread, and the frame index.
The external interface changes to always return a std::unique_ptr,
because the object layout is not known via the public interface, hence
objects cannot be stack allocated. They also cannot be copied or passed
by value.
The frame inspection interface will be tested after another fix in
https://chromium-review.googlesource.com/474749.
R=ahaas@chromium.org
BUG=v8:5822
Change-Id: I7b109da73df745fac97ec72cb0cf4f0ad71e5da9
Reviewed-on: https://chromium-review.googlesource.com/472887
Reviewed-by: Andreas Haas <ahaas@chromium.org>
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Cr-Commit-Position: refs/heads/master@{#44600}
RationalizeConsecutiveAtoms optimizes ab|ac|az to a(?:b|c|d).
Ensure that this optimization does not split surrogate pairs in unicode
mode.
BUG=chromium:641091
Review-Url: https://codereview.chromium.org/2813893002
Cr-Commit-Position: refs/heads/master@{#44599}
As of crrev.com/2760213003, the CheckBounds operator passes a truncation
that identfies zero and minus zero. However that was not reflected in
the typing rule, and as such the type of CheckBounds(-0,length) was
always Type::None. That confused the typed alias analysis in the
LoadElimination and led to ignoring StoreElement nodes.
BUG=chromium:708050
R=jarin@chromium.org
Review-Url: https://codereview.chromium.org/2812013006
Cr-Commit-Position: refs/heads/master@{#44598}
