This CL extends IterableToListWithSymbolLookup with fast paths
for spreading keys/values iterators of JSMap, and values iterator
of JSSet (which is also the iterator of Set.prototype.keys() and
Set.prototype[Symbol.iterator]()). The fast paths are only taken
if the target still has original iteration behavior.
For iterators it is also required that the iterator is not
partially consumed. After spreading, to be spec-compliant, the
iterator is exhausted. Tests are added.
Bug: v8:7980
Change-Id: Ida74e5ecbbc5ba5488d13a40f2c4bda14c781cbf
Reviewed-on: https://chromium-review.googlesource.com/c/1276632
Reviewed-by: Benedikt Meurer <bmeurer@chromium.org>
Reviewed-by: Georg Neis <neis@chromium.org>
Commit-Queue: Hai Dang <dhai@google.com>
Cr-Commit-Position: refs/heads/master@{#56716}
This skips the slowest tests in stress and noopt variants.
TBR=sigurds@chromium.org
NOTRY=true
Bug: v8:7783
Cq-Include-Trybots: luci.chromium.try:linux_chromium_headless_rel;master.tryserver.blink:linux_trusty_blink_rel
Change-Id: Ic471a2ab3e6806c4c60b81c0cdddfb44b199dd26
Reviewed-on: https://chromium-review.googlesource.com/c/1286334
Commit-Queue: Michael Achenbach <machenbach@chromium.org>
Reviewed-by: Michael Achenbach <machenbach@chromium.org>
Cr-Commit-Position: refs/heads/master@{#56715}
This also includes ports of Array.p.toString and Array.p.toLocaleString.
Many parts of the old JS implementation are preserved, because
TypedArray.p.join still relies on it. These will be removed once
TypedArray.p.join is ported to Torque.
To simplify implementation, special handling of extremely sparse arrays
has been removed.
Performance improvements vary by array size, elements, and sparse-ness.
Some quick numbers and graphs are here:
https://docs.google.com/spreadsheets/d/125VLmRMudk8XaomLCsZQ1ewc94WCqht-8GQwU3s9BW8/edit#gid=2087673710
Cq-Include-Trybots: luci.chromium.try:linux_chromium_headless_rel;luci.v8.try:v8_linux_noi18n_rel_ng;master.tryserver.blink:linux_trusty_blink_rel
Change-Id: Ia4069a068403ce36676c37401d349aefc976b045
Reviewed-on: https://chromium-review.googlesource.com/c/1196693
Commit-Queue: Peter Wong <peter.wm.wong@gmail.com>
Reviewed-by: Jakob Gruber <jgruber@chromium.org>
Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
Cr-Commit-Position: refs/heads/master@{#56699}
Previously when class names were computed and set as part of
StoreDataPropertyInLiteral calls, it was observable to static fields
as these static fields are initialized right after the classes were
constructed but before the class names were installed.
This caused the name property to be undefined for this case.
Instead, this patch always forces the creation of a name property on
the class constructor when static class fields are used. This patch
does kill the class boilerplate optimization, but currently all static
class fields are installed using a runtime call to CreateDataProperty
so this isn't any worse when using static class fields.
In the future, this can be optimized away by storing the name on the
boilerplate.
There is spec discussion here:
https://github.com/tc39/proposal-class-fields/issues/85
There isn't a resolution yet, there's still discussion about whether
to have the name be undefined always for static class field
initializers. But, I don't think that's useful as it would always kill
our boilerplate optimization (like this patch does ..., but without the
future optimization potential).
Bug: v8:5367
Change-Id: I14afdf7ece3f2d9fa3c659d2c0bc3806e0b17abb
Reviewed-on: https://chromium-review.googlesource.com/c/1281002
Reviewed-by: Mythri Alle <mythria@chromium.org>
Reviewed-by: Daniel Ehrenberg <littledan@chromium.org>
Reviewed-by: Igor Sheludko <ishell@chromium.org>
Commit-Queue: Sathya Gunasekaran <gsathya@chromium.org>
Cr-Commit-Position: refs/heads/master@{#56686}
This switches the encoding of the exceptions (in the exceptions as well
as the import section) to use a signature index instead of a flat type
vector encoding. Note that only signatures that have a void return type
can be used for declaring exceptions.
R=clemensh@chromium.org
BUG=v8:8153
Change-Id: I481ccbce9ddf29becdf4ed7ceffe80d6145446e1
Reviewed-on: https://chromium-review.googlesource.com/c/1280323
Commit-Queue: Michael Starzinger <mstarzinger@chromium.org>
Reviewed-by: Clemens Hammacher <clemensh@chromium.org>
Cr-Commit-Position: refs/heads/master@{#56654}
The race happens when the sweeper is looking up the size of an object
that had its map replaced concurrently.
The fix is to load the object map using an acquire load so that the
sweeper observes the initializing stores of the new map.
Bug: v8:8303
Change-Id: Ifaaef06cb815be7d07b6a574085ee61a466bc1d6
Reviewed-on: https://chromium-review.googlesource.com/c/1280310
Reviewed-by: Michael Lippautz <mlippautz@chromium.org>
Reviewed-by: Hannes Payer <hpayer@chromium.org>
Commit-Queue: Ulan Degenbaev <ulan@chromium.org>
Cr-Commit-Position: refs/heads/master@{#56652}
Change-Id: I982f3615136c7a4ba18e4a6d2cc06a3e24e22f54
Reviewed-on: https://chromium-review.googlesource.com/c/1277722
Reviewed-by: Ben Titzer <titzer@chromium.org>
Commit-Queue: Adam Klein <adamk@chromium.org>
Cr-Commit-Position: refs/heads/master@{#56645}
For serialization we are using the code table to find the code of all
functions. We want to serialize compiled code though, not interpreter
entries (we currently fail a DCHECK there).
This CL changes the logic to not update the code table with interpreter
entries but instead keeps a separate bit set of interpreted functions.
R=mstarzinger@chromium.org
Bug: v8:8177, chromium:735509
Change-Id: I69c59f92712135ddef667b54114614fad94cc6fc
Reviewed-on: https://chromium-review.googlesource.com/c/1278794
Reviewed-by: Michael Starzinger <mstarzinger@chromium.org>
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Cr-Commit-Position: refs/heads/master@{#56644}
Split the runtime function for initializing a promise into AwaitPromisesInit
and AwaitPromisesInitOld, the former not firing the INIT hook and being used
by the AwaitOptimized builtin. In addition to this the AsyncHooks now caches
all the previously inited promises and checks that the init hook is not fired
twice for the same promise.
Modified test expectations for the new async ids in the async hooks tests.
Bug: v8:8300
Change-Id: If4a17e501b2a233578fa70b6442f219473f001d9
Reviewed-on: https://chromium-review.googlesource.com/c/1280442
Commit-Queue: Maya Lekova <mslekova@chromium.org>
Reviewed-by: Benedikt Meurer <bmeurer@chromium.org>
Cr-Commit-Position: refs/heads/master@{#56642}
Properly handle the case where the CheckFloat64Hole becomes a
no-op after RETYPE (because the feedback type is already Number).
We always need to pass the Number restriction type here.
Bug: chromium:895199
Change-Id: I96a949ba35db1e6d35abedddc4507c101d95b716
Reviewed-on: https://chromium-review.googlesource.com/c/1278804
Reviewed-by: Jaroslav Sevcik <jarin@chromium.org>
Commit-Queue: Benedikt Meurer <bmeurer@chromium.org>
Cr-Commit-Position: refs/heads/master@{#56622}
This fixes the typing for the case when the call is not lowered to
the simplified operator.
Bug: chromium:880207
Change-Id: Icecf12de77ece0fe9ffec2777874f5f0004a1e97
Reviewed-on: https://chromium-review.googlesource.com/c/1278642
Commit-Queue: Jaroslav Sevcik <jarin@chromium.org>
Commit-Queue: Benedikt Meurer <bmeurer@chromium.org>
Reviewed-by: Benedikt Meurer <bmeurer@chromium.org>
Cr-Commit-Position: refs/heads/master@{#56621}
For the --async-stack-traces we can also look through initial parts of
the promise chain that were created by regular Promise#then() calls to
walk up to the first async function frame. This addresses the missing
support for aforementioned example
```js
(async function() {
await Promise.resolve().then(() =>
console.log(new Error().stack));
})();
```
which now also works.
Bug: v8:7522
Change-Id: I574943c1fc6ee4a1bd56f208dce78eb7506c5c4f
Reviewed-on: https://chromium-review.googlesource.com/c/1278276
Reviewed-by: Maya Lekova <mslekova@chromium.org>
Commit-Queue: Benedikt Meurer <bmeurer@chromium.org>
Cr-Commit-Position: refs/heads/master@{#56613}
Updated the test so that it uses assertPromiseResult, which makes sure that
a promise rejection is not swallowed. The change is reflected in the actual
async ids, checked in the test.
Bug: v8:8300
Change-Id: Ie227ca74a8cf4e0e079809b21c3abc5a5f87c11a
Reviewed-on: https://chromium-review.googlesource.com/c/1278388
Commit-Queue: Maya Lekova <mslekova@chromium.org>
Reviewed-by: Benedikt Meurer <bmeurer@chromium.org>
Cr-Commit-Position: refs/heads/master@{#56608}
Adds support for enqueuing parallel parse / compile tasks for eagerly
compiled IIFEs during parsing. If the --parallel-compile-tasks flag is
enabled, the parser will pre-parse eager top-level IIFEs and enqueue a
task on the compiler dispatcher to do the actual parsing / compilation
on a worker thread.
Currently we always enqueue the task, but we likely want to only
enqueue parallel tasks where the script has multiple IIFEs or a
substantial amount of top-level script code before the IIFE to avoid
the main thread having to immediately block on the parallel task. This
work will be done as a follow-up.
BUG=v8:8041
Change-Id: If68d7c374548cabd4ec32f1fb6752da7d6aaae6b
Reviewed-on: https://chromium-review.googlesource.com/c/1275354
Reviewed-by: Toon Verwaest <verwaest@chromium.org>
Commit-Queue: Ross McIlroy <rmcilroy@chromium.org>
Cr-Commit-Position: refs/heads/master@{#56593}
On ia32, we were pinning too many registers, resulting in no unpinned
byte registers left (we only have three byte registers since {ebx}
is reserved for the root register).
It turns out that on most paths, we don't actually need to pin any
registers, since {Store} is often the last call for an operation (like
any store or set_global). If registers need to be pinned, only pass
those that must be kept alive across the {Store}. This allows to
compute a more narrow set of pinned registers on demand inside {Store}.
Plus minor drive-by changes.
R=titzer@chromium.org
Bug: chromium:894374, chromium:894307, v8:6600
Change-Id: Ic4d7131784c193dc7a2abf0e504d9973f6d5c5f1
Reviewed-on: https://chromium-review.googlesource.com/c/1275819
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Reviewed-by: Ben Titzer <titzer@chromium.org>
Cr-Commit-Position: refs/heads/master@{#56587}
This makes sure that catch blocks that are practically unreachable due
to missing exceptional projections are handled properly. Note that this
is independent of how reachability will be outlined in the final spec
for exception handling. Currently we just assume that all catch blocks
are spec-wise reachable.
R=titzer@chromium.org
TEST=mjsunit/wasm/exceptions-catchall
BUG=v8:8091
Change-Id: I13607a59bd76be146df836e88105a2fbafedb760
Reviewed-on: https://chromium-review.googlesource.com/c/1273018
Reviewed-by: Ben Titzer <titzer@chromium.org>
Commit-Queue: Michael Starzinger <mstarzinger@chromium.org>
Cr-Commit-Position: refs/heads/master@{#56556}
This is behind a new flag --harmony-namespace-exports.
Bug: v8:8101
Cq-Include-Trybots: luci.v8.try:v8_linux_noi18n_rel_ng
Change-Id: I9c252b6de2b08223fcf3296340b78d721471bdb4
Reviewed-on: https://chromium-review.googlesource.com/c/1258004
Commit-Queue: Georg Neis <neis@chromium.org>
Reviewed-by: Adam Klein <adamk@chromium.org>
Reviewed-by: Sathya Gunasekaran <gsathya@chromium.org>
Cr-Commit-Position: refs/heads/master@{#56550}
Use naming similar to the spec: "table" instead of "function table",
"element segment" instead of "function table init".
Change-Id: Ib1b6cdfa566f8bd00017ccedf9440084204f10ff
Reviewed-on: https://chromium-review.googlesource.com/c/1273612
Commit-Queue: Ben Smith <binji@chromium.org>
Reviewed-by: Andreas Haas <ahaas@chromium.org>
Cr-Commit-Position: refs/heads/master@{#56545}
The NumberMultiply typing rule gave up in the presence of NaN inputs,
but we can still infer useful ranges here and just union the result
of that with the NaN propagation (similar for MinusZero propagation).
This way we can still makes sense of these ranges at the uses.
Bug: v8:8015
Change-Id: Ic4c5e8edc6c68776ff3baca9628ad7de0f8e2a92
Reviewed-on: https://chromium-review.googlesource.com/c/1261143
Commit-Queue: Benedikt Meurer <bmeurer@chromium.org>
Reviewed-by: Sigurd Schneider <sigurds@chromium.org>
Cr-Commit-Position: refs/heads/master@{#56539}
Block coverage is based on a system of ranges that can either have
both a start and end position, or only a start position (so-called
singleton ranges). When formatting coverage information, singletons
are expanded until the end of the immediate full parent range. E.g.
in:
{0, 10} // Full range.
{5, -1} // Singleton range.
the singleton range is expanded to {5, 10}.
Singletons are produced mostly for continuation counters that track
whether we execute past a specific language construct.
Unfortunately, continuation counters can turn up in spots that confuse
our post-processing. For example:
if (true) { ... block1 ... } else { ... block2 ... }
If block1 produces a continuation counter, it could end up with the
same start position as the else-branch counter. Since we merge
identical blocks, the else-branch could incorrectly end up with an
execution count of one.
We need to avoid merging such cases. A full range should always take
precedence over a singleton range; a singleton range should never
expand to completely fill a full range. An additional post-processing
pass ensures this.
Bug: v8:8237
Change-Id: Idb3ec7b2feddc0585313810b9c8be1e9f4ec64bf
Reviewed-on: https://chromium-review.googlesource.com/c/1273095
Reviewed-by: Georg Neis <neis@chromium.org>
Reviewed-by: Yang Guo <yangguo@chromium.org>
Commit-Queue: Jakob Gruber <jgruber@chromium.org>
Cr-Commit-Position: refs/heads/master@{#56531}
The MapIterator protector protects the original iteration behaviors of
Map.prototype.keys(), Map.prototype.values(), and Set.prototype.entries().
It does not protect the original iteration behavior of
Map.prototype[Symbol.iterator](). The protector is invalidated when:
* The 'next' property is set on an object where the property holder is the
%MapIteratorPrototype% (e.g. because the object is that very prototype).
* The 'Symbol.iterator' property is set on an object where the property
holder is the %IteratorPrototype%. Note that this also invalidates the
SetIterator protector (see below).
The SetIterator protector protects the original iteration behavior of
Set.prototype.keys(), Set.prototype.values(), Set.prototype.entries(),
and Set.prototype[Symbol.iterator](). The protector is invalidated when:
* The 'next' property is set on an object where the property holder is the
%SetIteratorPrototype% (e.g. because the object is that very prototype).
* The 'Symbol.iterator' property is set on an object where the property
holder is the %SetPrototype% OR %IteratorPrototype%. This means that
setting Symbol.iterator on a MapIterator object can also invalidate the
SetIterator protector, and vice versa, setting Symbol.iterator on a
SetIterator object can also invalidate the MapIterator. This is an over-
approximation for the sake of simplicity.
Bug: v8:7980
Change-Id: I54ad6e4c7f19ccc27d7001f6c4b6c8d6ea4ee871
Reviewed-on: https://chromium-review.googlesource.com/c/1273102
Reviewed-by: Georg Neis <neis@chromium.org>
Reviewed-by: Ulan Degenbaev <ulan@chromium.org>
Reviewed-by: Benedikt Meurer <bmeurer@chromium.org>
Commit-Queue: Hai Dang <dhai@google.com>
Cr-Commit-Position: refs/heads/master@{#56530}
This CL refactors the implementation of WASM->JS import wrappers in order
to make the wrapper code shareable. Instead of specializing to the import
index, we use a tuple as the object ref in the both the import and indirect
tables. The tuple allows the wrapper code to load both the calling
instance and the target callable, rather than relying on code specialization.
This requires some tricky codegen machinery, because WASM call descriptors
expect an instance argument in a given register, yet the wrappers receive
a tuple, the code generator must generate a prologue that loads the
instance (and the callable), since it is not possible to express this at
the graph level.
R=mstarzinger@chromium.orgCC=clemensh@chromium.org
Change-Id: Id67e307f7f5089e776f5439a53b5aee4b76934b6
Reviewed-on: https://chromium-review.googlesource.com/c/1268237
Commit-Queue: Ben Titzer <titzer@chromium.org>
Reviewed-by: Michael Starzinger <mstarzinger@chromium.org>
Cr-Commit-Position: refs/heads/master@{#56520}
This reverts commit 471fef0469.
Reason for revert: A more general fix incoming at https://crrev.com/c/1273095.
Original change's description:
> [coverage] change block range to avoid ambiguity.
>
> By moving the block range end to left of closing bracket,
> we can avoid ambiguity where an open-ended singleton range
> could be both interpreted as inside the parent range, or
> next to it.
>
> R=verwaest@chromium.org
>
> Bug: v8:8237
> Change-Id: Ibc9412b31efe900b6d8bff0d8fa8c52ddfbf460a
> Reviewed-on: https://chromium-review.googlesource.com/1254127
> Reviewed-by: Georg Neis <neis@chromium.org>
> Commit-Queue: Yang Guo <yangguo@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#56347}
TBR=yangguo@chromium.org,neis@chromium.org,verwaest@chromium.org
# Not skipping CQ checks because original CL landed > 1 day ago.
Bug: v8:8237
Change-Id: I39310cf3c2f06a0d98ff314740aaeefbfffc0834
Reviewed-on: https://chromium-review.googlesource.com/c/1273096
Reviewed-by: Jakob Gruber <jgruber@chromium.org>
Reviewed-by: Toon Verwaest <verwaest@chromium.org>
Reviewed-by: Yang Guo <yangguo@chromium.org>
Commit-Queue: Jakob Gruber <jgruber@chromium.org>
Cr-Commit-Position: refs/heads/master@{#56513}
This change introduces new intrinsics used to desugar async functions
in the Parser and the BytecodeGenerator, namely we introduce a new
%_AsyncFunctionEnter intrinsic that constructs the generator object
for the async function (and in the future will also create the outer
promise for the async function). This generator object is internal
and never escapes to user code, plus since async functions don't have
a "prototype" property, we can just a single map here instead of tracking
the prototype/initial_map on every async function. This saves one word
per async function plus one initial_map per async function that was
invoked at least once.
We also introduce two new intrinsics %_AsyncFunctionReject, which
rejects the outer promise with the caught exception, and another
%_AsyncFunctionResolve, which resolves the outer promise with the
right hand side of the `return` statement. These functions also perform
the DevTools part of the job (aka popping from the promise stack and
sending the debug event). This allows us to get rid of the implicit
try-finally from async functions completely; because the finally
block only called to the %AsyncFunctionPromiseRelease builtin, which
was used to inform DevTools.
In essence we now turn an async function like
```js
async function f(x) { return await bar(x); }
```
into something like this (in Parser and BytecodeGenerator respectively):
```
function f(x) {
.generator_object = %_AsyncFunctionEnter(.closure, this);
.promise = %AsyncFunctionCreatePromise();
try {
.tmp = await bar(x);
return %_AsyncFunctionResolve(.promise, .tmp);
} catch (e) {
return %_AsyncFunctionReject(.promise, e);
}
}
```
Overall the bytecode for async functions gets significantly shorter
already (and will get even shorter once we put the outer promise into
the async function generator object). For example the bytecode for a
simple async function
```js
async function f(x) { return await x; }
```
goes from 175 bytes to 110 bytes (a ~38% reduction in size), which
is in particular due to the simplification around the try-finally
removal.
Overall this seems to improve the doxbee-async-es2017-native test by
around 2-3%. On the test case mentioned in v8:8276 we go from
1124ms to 441ms, which corresponds to a 60% reduction in total
execution time!
Tbr: marja@chromium.org
Bug: v8:7253, v8:7522, v8:8276
Cq-Include-Trybots: luci.chromium.try:linux_chromium_headless_rel;luci.chromium.try:linux_chromium_rel_ng;master.tryserver.blink:linux_trusty_blink_rel
Change-Id: Id29dc92de7490b387ff697860c900cee44c9a7a4
Reviewed-on: https://chromium-review.googlesource.com/c/1269041
Commit-Queue: Benedikt Meurer <bmeurer@chromium.org>
Reviewed-by: Benedikt Meurer <bmeurer@chromium.org>
Reviewed-by: Sathya Gunasekaran <gsathya@chromium.org>
Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
Reviewed-by: Jaroslav Sevcik <jarin@chromium.org>
Reviewed-by: Maya Lekova <mslekova@chromium.org>
Reviewed-by: Yang Guo <yangguo@chromium.org>
Cr-Commit-Position: refs/heads/master@{#56502}
The test started flaking on almost every run since 1 day, disabling
until the root cause is triaged.
NOTRY=true
Bug: v8:8219
Change-Id: Id3cf219874e79cefc41bb63a9a4a04b6288d5350
Reviewed-on: https://chromium-review.googlesource.com/c/1270942
Commit-Queue: Sigurd Schneider <sigurds@chromium.org>
Reviewed-by: Sigurd Schneider <sigurds@chromium.org>
Cr-Commit-Position: refs/heads/master@{#56488}
This fixes handling of two corner cases with catch-all blocks:
1) The catch-all blocks are conceptually outside the corresponding try.
2) Reachability of catch-all is determined by parent reachability.
R=clemensh@chromium.org
TEST=mjsunit/wasm/exceptions-catchall
BUG=v8:8091
Change-Id: Idfd8310bc232f3ce389763023c5a33f1ef90e0b5
Reviewed-on: https://chromium-review.googlesource.com/c/1270816
Reviewed-by: Ben Titzer <titzer@chromium.org>
Commit-Queue: Michael Starzinger <mstarzinger@chromium.org>
Cr-Commit-Position: refs/heads/master@{#56486}
This adds support for multiple catch blocks being attached to a single
try block. The implemented semantics are that type checks are performed
in order from top to bottom.
Note that multiple catch blocks of the same type are not prohibited and
will be accepted, making the second such block essentially unreachable.
The current proposal neither explicitly allows nor prohibits it.
R=clemensh@chromium.org
TEST=mjsunit/wasm/exceptions
BUG=v8:8091
Change-Id: I31e7a07a7cffdd909a58342e00f05e52ed1a3182
Reviewed-on: https://chromium-review.googlesource.com/c/1270591
Reviewed-by: Clemens Hammacher <clemensh@chromium.org>
Commit-Queue: Michael Starzinger <mstarzinger@chromium.org>
Cr-Commit-Position: refs/heads/master@{#56478}
This adds support to wire control flow of catch-all expressions into an
existing try-catch cascade. Note that multiple typed catch blocks are
not yet supported, only one typed catch block followed by one catch-all
block is supported.
In case the explicit catch-all block is missing, we emulate the correct
semantics by internally always emitting a catch-all containing a simple
rethrow instruction.
R=clemensh@chromium.org
TEST=mjsunit/wasm/exceptions-catchall
BUG=v8:8091
Change-Id: I6b29a98c4f1a558fabe6012f4ba6c7b7d43529bb
Reviewed-on: https://chromium-review.googlesource.com/c/1270585
Reviewed-by: Clemens Hammacher <clemensh@chromium.org>
Commit-Queue: Michael Starzinger <mstarzinger@chromium.org>
Cr-Commit-Position: refs/heads/master@{#56476}
Now duplicate parameter detection depends on tracking of unresolved references.
This also fixes finding duplicate parameters of arrow functions nested in other
arrow functions.
Change-Id: I644bfdc513244637345c1069e5c7e5fde713da63
Reviewed-on: https://chromium-review.googlesource.com/c/1270578
Commit-Queue: Toon Verwaest <verwaest@chromium.org>
Reviewed-by: Marja Hölttä <marja@chromium.org>
Cr-Commit-Position: refs/heads/master@{#56467}
This adds support to the escape analysis to allow scalar replacement
of (small) FixedArrays with element accesses where the index is not a
compile time constant. This happens quite often when inlining functions
that operate on variable number of arguments. For example consider this
little piece of code:
```js
function sum(...args) {
let s = 0;
for (let i = 0; i < args.length; ++i) s += args[i];
return s;
}
function sum2(x, y) {
return sum(x, y);
}
```
This example is made up, of course, but it shows the problem. Let's
assume that TurboFan inlines the function `sum` into it's call site
at `sum2`. Now it has to materialize the `args` array with the two
values `x` and `y`, and iterate through these `args` to sum them up.
The escape analysis pass figures out that `args` doesn't escape (aka
doesn't outlive) the optimized code for `sum2` now, but TurboFan still
needs to materialize the elements backing store for `args` since there's
a `LoadElement(args.elements,i)` in the graph now, and `i` is not a
compile time constant.
However the escape analysis has more information than just that. In
particular the escape analysis knows exactly how many elements a non
escaping object has, based on the fact that the allocation must be
local to the function and that we only track objects with known size.
So in the case above when we get to `args[i]` in the escape analysis
the relevant part of the graph looks something like this:
```
elements = LoadField[elements](args)
length = LoadField[length](args)
index = CheckBounds(i, length)
value = LoadElement(elements, index)
```
In particular the contract here is that `LoadElement(elements,index)`
is guaranteed to have an `index` that is within the valid bounds for
the `elements` (there must be a preceeding `CheckBounds` or some other
guard in optimized code before it). And since `elements` is allocated
inside of the optimized code object, the escape analysis also knows
that `elements` has exactly two elements inside (namely the values of
`x` and `y`). So we can use that information and replace the access
with a `Select(index===0,x,y)` operation instead, which allows us to
scalar replace the `elements`, since there's no escaping use anymore
in the graph.
We do this for the case that the number of elements is 2, as described
above, but also for the case where elements length is one. In case
of 0, we know that the `LoadElement` must be in dead code, but we can't
just mark it for deletion from the graph (to make sure it doesn't block
scalar replacement of non-dead code), so we don't handle this for now.
And for one element it's even easier, since the `LoadElement` has to
yield exactly said element.
We could generalize this to handle arbitrary lengths, but since there's
a cost to arbitrary decision trees here, it's unclear when this is still
beneficial. Another possible solution for length > 2 would be to have
special stack allocation for these backing stores and do variable index
accesses to these stack areas. But that's way beyond the scope of this
isolated change.
This change shows a ~2% improvement on the EarleyBoyer benchmark in
JetStream, since it benefits a lot from not having to materialize these
small arguments backing stores.
Drive-by-fix: Fix JSCreateLowering to properly initialize "elements"
with StoreElement instead of StoreField (which violates the invariant
in TurboFan that fields and elements never alias).
Bug: v8:5267, v8:6200
Change-Id: Idd464a15a81e7c9653c48c814b406eb859841428
Reviewed-on: https://chromium-review.googlesource.com/c/1267935
Commit-Queue: Benedikt Meurer <bmeurer@chromium.org>
Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
Cr-Commit-Position: refs/heads/master@{#56442}
