Microbenchmarks show 25% improvement over C++, 11% improvement over JS
for the constructor. We don't have a microbenchmark covering the compile
method.
Locally, octane/regexp improved by 2%.
BUG=v8:5339
Review-Url: https://codereview.chromium.org/2551443002
Cr-Commit-Position: refs/heads/master@{#41490}
Calls that are potential eval calls are funneled through bytecode and
hence no longer use the {AstGraphBuilder} by now. Associated bailout
points can be removed.
R=jarin@chromium.org
Review-Url: https://codereview.chromium.org/2549113002
Cr-Commit-Position: refs/heads/master@{#41489}
This CL adds a new header src/debug/interface-types.h, moves the
definition of Location from the debug-interface.h to this new header,
and adds a new definition for the WasmDisassembly types.
This allows to use the types in other implementation files or headers
without having to include the entire debug-interface.h, reducing build
dependencies and compile time (especially for incremental builds).
The WasmDisassembly type replaces the old
std::pair<std::string, std::vector<std::tuple<...>>>, which was a bit
hard to unravel.
R=yangguo@chromium.org, kozyatinskiy@chromium.org, titzer@chromium.org
Review-Url: https://codereview.chromium.org/2529383002
Cr-Commit-Position: refs/heads/master@{#41488}
This allows us to optimise the bytecode liveness analysis to jump
directly to previously seen indices. The analysis is optimised to store
a stack of loop ends (JumpLoop bytecode indices), and iterate through
these indices directly rather than looping through the bytecode array to
find them.
Review-Url: https://codereview.chromium.org/2536653003
Cr-Commit-Position: refs/heads/master@{#41485}
Don't unconditionally assume that Tagged->Word32 changes are always
truncating independent of the type of the input. Work-around the
remaining issues with for-in by properly renaming the index if it's
not already of appropriate type (happens with generators or OSR).
R=jarin@chromium.org
Review-Url: https://codereview.chromium.org/2545393002
Cr-Commit-Position: refs/heads/master@{#41484}
1) CSA::Select() receives lambdas for generating true/false values.
The representation parameter made mandatory.
2) CSA::Select[Type]Constant() methods receive true/false nodes directly.
These methods are intended to be used for generating "selects" when
true/false values are already computed.
BUG=
Review-Url: https://codereview.chromium.org/2550683003
Cr-Commit-Position: refs/heads/master@{#41483}
The changes are:
1) Pass lambdas as const references to avoid unnecessary copying.
2) Remove CodeStubAssembler* parameter from loop bodies and let the lambdas
capture the right assembler instead. It makes the loop body code look
uniform with the surrounding code and unblocks splitting of a CSA
into different classes.
BUG=
Review-Url: https://codereview.chromium.org/2535753012
Cr-Commit-Position: refs/heads/master@{#41482}
This removes support for accessing super properties (load, store & call)
from the {FullCodeGenerator}. Consequently optimized code containing
such constructs must use the {BytecodeGraphBuilder} and can no longer
use the {AstGraphBuilder} for graph building.
R=bmeurer@chromium.org
BUG=v8:5657
Review-Url: https://codereview.chromium.org/2544123004
Cr-Commit-Position: refs/heads/master@{#41478}
When enforcing Ignition+TurboFan for a function literal during
AstNumbering, make sure to print the bailout reason with --trace-opt
to make it easier to figure out what caused the bailout.
BUG=chromium:670691
R=jarin@chromium.org
Review-Url: https://codereview.chromium.org/2550093002
Cr-Commit-Position: refs/heads/master@{#41472}
The current Array.prototype.sort implementation is pretty sensitive to
compiler changes, i.e. switching to I+TF completely, so refactor it a
bit so that it can stay with FCG+CS for now.
Middle-term the Array builtins need to be refactored to TurboFan
builtins anyways.
BUG=chromium:670691,v8:5666
R=jarin@chromium.org
Review-Url: https://codereview.chromium.org/2553523002
Cr-Commit-Position: refs/heads/master@{#41471}
Add support to lower the %_DebugIsActive intrinsic during
JSIntrinsicLowering instead of always going to the runtime
for this.
This addresses part of the Bluebird regression caused by
sending let and const to TurboFan and Ignition.
R=jarin@chromium.org
Review-Url: https://codereview.chromium.org/2550043002
Cr-Commit-Position: refs/heads/master@{#41468}
In the current implementation, WasmInstanceWrapper is allocated after the imports for the instance are processed, and before the InstanceFinalizer callback is associated with the instance. This raises the possibility of triggering a gc in the middle of the instantiate flow which is incorrect.
BUG=5707
R=titzer@chromium.org, petermarshall@chromium.org
Review-Url: https://codereview.chromium.org/2544273002
Cr-Commit-Position: refs/heads/master@{#41464}
Several cleanups to the output of the debug-only --print-scopes flag:
- Function name variable only printed once
- Only print headers for sections of variables that are non-empty
- Assume Variables stored in Scope::variables_ are never null
R=verwaest@chromium.org
Review-Url: https://codereview.chromium.org/2544063004
Cr-Commit-Position: refs/heads/master@{#41463}
Reason for revert:
Speculative revert for causing timeouts on Win Debug gpu fyi bot
Nothing else looks even remotely relevant in the list of changes.
Will reland if this doesn't fix the issues.
BUG=670396
Original issue's description:
> [typedarrays] remove invalid optimization in NAMEConstructor()
>
> Before, we were treating objects with the builtin ArrayValues iterator
> method as array-like, where the iterator would iterate through to the
> full length of the object.
>
> This optimization was not sound, because it does not ensure that the
> next method hasn't been modified. Even if it hasn't been modified,
> it's entirely possible to be modified during iteration. Thus, this
> optimization has been removed due to its observability.
>
> BUG=v8:5699
> R=littledan@chromium.org, cbruni@chromium.org
>
> Committed: https://crrev.com/77df8c67d9609ada3b7d79e8e6d33f198bbad5a1
> Cr-Commit-Position: refs/heads/master@{#41394}
TBR=cbruni@chromium.org,littledan@chromium.org,caitp@igalia.com
# Not skipping CQ checks because original CL landed more than 1 days ago.
BUG=v8:5699
Review-Url: https://codereview.chromium.org/2548583003
Cr-Commit-Position: refs/heads/master@{#41461}
Turbofan uses 64-bit load followed by 32 bit arithmetic shift when
loading higher 32 bits of 64-bit value. We simplify this
by loading higher 32 bits directly.
BUG=
Review-Url: https://codereview.chromium.org/2532333003
Cr-Commit-Position: refs/heads/master@{#41455}
This apparently gradually fell out of use after the more general
Token::IsIdentifer was introduced, and whoever left last forgot to
turn out the lights.
The only remaining use was in an assertion in DeclareLabel, but
meanwhile DeclareLabel is only called if IsIdentifier. I added re-formulated assertions, just in case.
R=verwaest@chromium.org
BUG=v8:4947
Review-Url: https://codereview.chromium.org/2549493002
Cr-Commit-Position: refs/heads/master@{#41452}
This enables {FLAG_debug_code} by default in debug builds. The advantage
is that generated code contained within the snapshot will contain such
debug code. Before we would only get coverage for these pieces with the
no-snapshot builds, which have a meager coverage. One can still pass the
inverse --no-debug-code flag to ensure generated code remains readable
within debug builds as well.
R=machenbach@chromium.org
Review-Url: https://codereview.chromium.org/2528913002
Cr-Commit-Position: refs/heads/master@{#41451}
Reason for revert:
My assumption that eval and top-level code should be handled differently was wrong
Original issue's description:
> Introduce a separate FunctionLiteral ID for Eval
>
> Top level SharedFunctionInfos will end up in a scripts SFI list, but
> eval'd SFIs shouldn't. Separate IDs will allow for adding a
> corresponding DCHECK.
>
> BUG=v8:5589
> R=marja@chromium.org
>
> Committed: https://crrev.com/c6d421ff9aee7f3cab9e48faac88f6b08d2f1cf5
> Cr-Commit-Position: refs/heads/master@{#41421}
TBR=marja@chromium.org
# Skipping CQ checks because original CL landed less than 1 days ago.
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
BUG=v8:5589
Review-Url: https://codereview.chromium.org/2544003003
Cr-Commit-Position: refs/heads/master@{#41448}
Tests were falling in qemu because of inexact computation in tests.
After correcting tests, simulator also had to be fixed.
Review-Url: https://codereview.chromium.org/2539133002
Cr-Commit-Position: refs/heads/master@{#41447}
Crankshaft has this highly unpredictable performance cliff around the
"Unsupported phi use of const or let variable", which is due to the
fact that Crankshaft refuses to do hole checks (for a couple of
reasons). So ideally we should not even try to send any lexically bound
variables to fullcodegen+Crankshaft, but instead give them to Ignition
and TurboFan.
This CL only adds the appropriate check to the AstNumbering, but doesn't
remove the functionality from fullcodegen/Crankshaft yet. This would be
step two in case this CL sticks. If you see any major performance
regressions with this CL in the range, just revert it and ping me.
See Node.js issue https://github.com/nodejs/node/issues/9729 for
additional information.
BUG=v8:5666
R=hablich@chromium.org, mstarzinger@chromium.org, rmcilroy@chromium.org
Review-Url: https://codereview.chromium.org/2525243002
Cr-Commit-Position: refs/heads/master@{#41445}
- These operations are identical for Float32x4 and Int32x4.
- Make them generic, following the naming for generic Simd128 / S128
opcodes.
- F32x4/I32x4 -> S32x4, similarly to S128
- Float32x4/Int32x4 -> Simd32x4, similarly to Simd128.
LOG=N
BUG=v8:4124
Review-Url: https://codereview.chromium.org/2543773002
Cr-Commit-Position: refs/heads/master@{#41437}
1) CSA::StoreMap() added which triggers map-specific write barrier
unlike StoreObjectField() which triggers the full write barrier.
2) CSA::StoreMapNoWriteBarrier(object, map_root_index) added.
3) StoreMapNoWriteBarrier() is used for storing immortal immovable
maps since they don't require write barriers even for objects in
old space.
BUG=
Review-Url: https://codereview.chromium.org/2544793002
Cr-Commit-Position: refs/heads/master@{#41435}
This moves the location of the bytecode-offset translation that turns
offsets of back jumps into offsets of loop headers. This translation is
now done by the {BytecodeGraphBuilder} after loop analysis has been
performed. It safes one redudant iteration over the bytecode array. Note
that this changes the semantics of the BailoutId used as an {osr_ast_id}
throughout the compiler pipeline for OSR from Ignition.
R=jarin@chromium.org
Review-Url: https://codereview.chromium.org/2465913002
Cr-Commit-Position: refs/heads/master@{#41431}
The overload simply wraps creation of an IntPtrConstant and makes things more
readable.
Review-Url: https://codereview.chromium.org/2541843006
Cr-Commit-Position: refs/heads/master@{#41430}
Before, it was a method in wasm namespace, and received a
Handle<WasmCompiledModule>. As it does not allocate on the heap, we can
just make it a non-static method on WasmCompiledModule.
R=titzer@chromium.org
Review-Url: https://codereview.chromium.org/2536373007
Cr-Commit-Position: refs/heads/master@{#41429}
The loop for non-"static" properties is no longer needed in
full-codegen since all computed property
names in object literals go through Ignition first.
BUG=v8:5657
Review-Url: https://codereview.chromium.org/2547433003
Cr-Commit-Position: refs/heads/master@{#41428}
Debugging API is still in use by Node.
Revert "[debug] remove deprecated debug command message queue."
This reverts commit abdbfc953d.
Revert "[debug] mark more unused debug API as deprecated."
This reverts commit d5ada19ce7.
BUG=v8:5530
Review-Url: https://codereview.chromium.org/2537313005
Cr-Commit-Position: refs/heads/master@{#41427}
... because only MachineRepresentation::kTagged makes sense there.
BUG=
Review-Url: https://codereview.chromium.org/2544713003
Cr-Commit-Position: refs/heads/master@{#41426}
Trampolines are generated when the value of pc_offset is greater than
next_buffer_check_ (attribute from Assembler class). This value
shouldn't be incremented in bind_to() method when internal reference
label is bound, because it is not decremented when the switch table is
generated (dd() method from Assemler class).
This patch fixes this problem. Regression test are also included for
mips and mips64 arch.
BUG=
Review-Url: https://codereview.chromium.org/2530143002
Cr-Commit-Position: refs/heads/master@{#41423}
The loop for non-"static" properties is no longer needed in
full-codegen since all computed property
names in object literals go through Ignition first.
BUG=v8:5657
Review-Url: https://codereview.chromium.org/2546473006
Cr-Commit-Position: refs/heads/master@{#41422}
Top level SharedFunctionInfos will end up in a scripts SFI list, but
eval'd SFIs shouldn't. Separate IDs will allow for adding a
corresponding DCHECK.
BUG=v8:5589
R=marja@chromium.org
Review-Url: https://codereview.chromium.org/2533303006
Cr-Commit-Position: refs/heads/master@{#41421}
This fixes the existing workaround in {BytecodeGraphBuilder} where the
number of elements in an array literal is unknown just from the bytecode
alone and needs to be deduced from the constant elements.
Note that this is just a quick fix to prevent calling the fast-clone
stub for boilerplates that are too big to fit on a regular page. In the
long run we need something more solid here.
R=mvstanton@chromium.org
TEST=mjsunit/regress/regress-crbug-669850
BUG=chromium:669850
Review-Url: https://codereview.chromium.org/2542633002
Cr-Commit-Position: refs/heads/master@{#41420}
Currently In LCodeGen::DoWrapReceiver(), the x87 jitted code's size for debug mode between label's define and bind
exceeds 128 bytes whether FLAG_deopt_every_n_times is set or not.
So always use Label:kFar as label distance in LCodeGen::DoWrapReceiver() for debug mode.
This CL also unify the label's distance value to avoid potential bugs caused by unconsistent distance value usage
for the same label when DeoptEveryNTimes() return true.
BUG=
Review-Url: https://codereview.chromium.org/2539403002
Cr-Commit-Position: refs/heads/master@{#41419}
We can skip RegExpResult construction on the fast path for several functions to
be more efficient.
BUG=v8:5330,v8:5674
Review-Url: https://codereview.chromium.org/2543483003
Cr-Commit-Position: refs/heads/master@{#41418}
Bonus: fixed a couple of places where 32-bit comparison was used.
BUG=
Review-Url: https://codereview.chromium.org/2543873003
Cr-Commit-Position: refs/heads/master@{#41417}
Unfortunately we have to split this up into two cases: those with exactly one spread argument as the final argument, and all others, due to any side-effects of evaluation being visible.
This is in preparation for a new bytecode which handles super calls.
BUG=v8:5659
Review-Url: https://codereview.chromium.org/2540593003
Cr-Commit-Position: refs/heads/master@{#41415}
This refactors portions of exec into a new function without RegExpResult
construction, which will be used in the future by test, @@match, and @@search
fast paths.
Unnecessary ToString and ToLength calls as well as repeated map checks were
removed.
BUG=v8:5339
Review-Url: https://codereview.chromium.org/2540153002
Cr-Commit-Position: refs/heads/master@{#41414}
This ensure that all inline allocations generated by {JSCreateLowering}
will fit into a regular heap page. Allocations targeting LO-space must
be done via a slower runtime call.
R=bmeurer@chromium.org
BUG=chromium:669850
Review-Url: https://codereview.chromium.org/2533353003
Cr-Commit-Position: refs/heads/master@{#41412}
The current CHECK/DCHECK implementation fails statically if a signed
value is compared against an unsigned value. The common solution is to
cast on each caller, which is tedious and error-prone (might hide bugs).
This CL implements signed vs. unsigned comparisons by executing up to
two comparisons. For example, if i is int32_t and u is uint_32_t, a
DCHECK_LE(i, u) would create the check
i <= 0 || static_cast<uint32_t>(i) <= u.
For checks against constants, at least one of the checks can be removed
by compiler optimizations.
The tradeoff we have to make is to sometimes silently execute an
additional comparison. And we increase code complexity of course, even
though the usage is just as easy (or even easier) as before.
The compile time impact seems to be minimal:
I ran 3 full compilations for Optdebug on my local machine, one time on
the current ToT, one time with this CL plus http://crrev.com/2524093002.
Before: 143.72 +- 1.21 seconds
Now: 144.18 +- 0.67 seconds
In order to check that the new comparisons are working, I refactored
some DCHECKs in wasm to use the new magic, and added unit test cases.
R=ishell@chromium.org, titzer@chromium.orgCC=ahaas@chromium.org, bmeurer@chromium.org
Committed: https://crrev.com/5925074a9dab5a8577766545b91b62f2c531d3dc
Review-Url: https://codereview.chromium.org/2526783002
Cr-Original-Commit-Position: refs/heads/master@{#41275}
Cr-Commit-Position: refs/heads/master@{#41411}
This shows around a 2.2x speedup compared to the old JS implementation (and
3.5x compared to CPP) for the fast path.
Adds ToUint32 to CodeStubAssembler.
BUG=v8:5339
Review-Url: https://codereview.chromium.org/2532403002
Cr-Commit-Position: refs/heads/master@{#41408}
This code is no longer used in full-codegen since all computed property
names in object literals go through Ignition first.
BUG=v8:5657
Review-Url: https://codereview.chromium.org/2543643002
Cr-Commit-Position: refs/heads/master@{#41405}
Incremental parsing of asm.js means we can see function tables that
are unused in the AsmWasmBuilder before they've been initialized.
BUG=669899
R=aseemgarg@chromium.org
Review-Url: https://codereview.chromium.org/2546553002
Cr-Commit-Position: refs/heads/master@{#41403}
If we just call CreateDebugInfo in GetPossibleBreakpoints then we won't call PrepareFunctionForBreakPoints and won't be able to step into this function or pause at breakpoint inside.
BUG=v8:5695
R=dgozman@chromium.org,yangguo@chromium.org
Review-Url: https://codereview.chromium.org/2540943002
Cr-Commit-Position: refs/heads/master@{#41401}
This is necessary for signal-based out of bounds handling in WebAssembly.
Adds a ProtectedStore instruction that is analogous to the previously added
ProtectedLoad instruction. Rather than using bounds checks, ProtectedStore emits
an out of line section of code that throws a JavaScript exception and provides
the necessary metadata for a signal handler to be able to find the out of line
code.
BUG= https://bugs.chromium.org/p/v8/issues/detail?id=5277
Review-Url: https://codereview.chromium.org/2516413003
Cr-Commit-Position: refs/heads/master@{#41398}
The "writable" property descriptor may legally change during the call to
AnythingToArrayLength(). This change needs to be honoured before calling
JSArray::SetLength(). The change is only honoured when the "length"
property was previously writable, so that changes during a call to
DefineOwnPropertyIgnoreAttributes() is ignored.
BUG=v8:5688
R=cbruni@chromium.org, verwaest@chromium.org, jkummerow@chromium.org
Review-Url: https://codereview.chromium.org/2543553002
Cr-Commit-Position: refs/heads/master@{#41396}
Before, we were treating objects with the builtin ArrayValues iterator
method as array-like, where the iterator would iterate through to the
full length of the object.
This optimization was not sound, because it does not ensure that the
next method hasn't been modified. Even if it hasn't been modified,
it's entirely possible to be modified during iteration. Thus, this
optimization has been removed due to its observability.
BUG=v8:5699
R=littledan@chromium.org, cbruni@chromium.org
Review-Url: https://codereview.chromium.org/2544503002
Cr-Commit-Position: refs/heads/master@{#41394}
This was causing more confusion than benefit, so we're removing it.
It's re-defined to empty for now, to avoid touching the ~100 files which
use it, we can remove it completely during a quiet period when it's less
likely to conflict with other work.
Review-Url: https://codereview.chromium.org/2535383005
Cr-Commit-Position: refs/heads/master@{#41393}
JS operators always have an implicit context input, so just use that instead.
BUG=
Review-Url: https://codereview.chromium.org/2541813002
Cr-Commit-Position: refs/heads/master@{#41392}
This is an experiment to see the impact of the limit on OOM crashes.
BUG=chromium:667388
Review-Url: https://codereview.chromium.org/2514313004
Cr-Commit-Position: refs/heads/master@{#41391}
These byte pointers (module_start and module_end) were only valid
during decoding. During instantiation or execution, they can get
invalidated by garbage collection.
This CL removes them from the WasmModule struct, and introduces a new
ModuleStorage struct as interface to the wasm wire bytes.
Since the storage is often needed together with the ModuleEnv, a new
ModuleStorageEnv struct holds both a ModuleEnv and a ModuleStorage.
The pointers in the ModuleStorage should never escape the live range of
this struct, as they might point into a SeqOneByteString or ArrayBuffer.
Therefore, the WasmInterpreter needs to create its own copy of the
whole module.
Runtime functions that previously used the raw pointers in WasmModule
(leading to memory errors) now have to use the SeqOneByteString in the
WasmCompiledModule.
R=titzer@chromium.org
BUG=chromium:669518
Review-Url: https://codereview.chromium.org/2540133002
Cr-Commit-Position: refs/heads/master@{#41388}
JSFrameSpecialization depends on the layout of the frame and doesn't work
with interpreted frames. Disable it since it is only used for OSR from asmjs code, which shouldn't go through the bytecode graph builder in many cases.
BUG=669517
Review-Url: https://codereview.chromium.org/2538823002
Cr-Commit-Position: refs/heads/master@{#41387}
port df2578d2ec (r41368)
original commit message:
Improves performance in simple, single element case by 5% and in multiple
elements cases by 2%.
BUG=
Review-Url: https://codereview.chromium.org/2540803004
Cr-Commit-Position: refs/heads/master@{#41377}
This adds consistency checks for function kind and scope type to
SharedFunctionInfoVerify.
It also fixes an inconsistency in the creation of a ScopeInfo.
R=adamk@chromium.org
BUG=
Review-Url: https://codereview.chromium.org/2537093002
Cr-Commit-Position: refs/heads/master@{#41375}
When removing a scope (see FinalizeBlockScope), remember the removal by making
the scope its own sibling. This avoid recalculating the information later on.
R=adamk@chromium.org
BUG=
Review-Url: https://codereview.chromium.org/2536993003
Cr-Commit-Position: refs/heads/master@{#41374}
Make the AsmWasmBuilder drive the process of typing and potentially parsing
function bodies. This will allow us to keep only a single asm.js function's
AST in memory as we convert to WebAssembly.
This is needed to keep our memory footprint low.
Add some additional output to a few tests that's helpful to see which stage they fail at.
BUG= https://bugs.chromium.org/p/v8/issues/detail?id=4203
LOG=N
R=marja@chromium.org,adamk@chromium.org,aseemgarg@chromium.org,titzer@chromium.org
Review-Url: https://codereview.chromium.org/2398023002
Cr-Commit-Position: refs/heads/master@{#41372}
