This change significantly improves the performance of string
concatenation in optimized code for the case where the resulting string
is represented as a ConsString. On the relevant test cases we go from
serializeNaive: 10762 ms.
serializeClever: 7813 ms.
serializeConcat: 10271 ms.
to
serializeNaive: 10278 ms.
serializeClever: 5533 ms.
serializeConcat: 10310 ms.
which represents a 30% improvement on the "clever" benchmark, which
tests specifically the ConsString creation performance.
This was accomplished via a couple of different steps, which are briefly
outlined here:
1. The empty_string gets its own map, so that we can easily recognize
and handle it appropriately in the TurboFan type system. This
allows us to express (and assert) that the inputs to NewConsString
are non-empty strings, making sure that TurboFan no longer creates
"crippled ConsStrings" with empty left or right hand sides.
2. Further split the existing String types in TurboFan to be able to
distinguish between OneByte and TwoByte strings on the type system
level. This allows us to avoid having to dynamically lookup the
resulting ConsString map in case of ConsString creation (i.e. when
we know that both input strings are OneByte strings or at least
one of the input strings is TwoByte).
3. We also introduced more finegrained feedback for the Add bytecode
in the interpreter, having it collect feedback about ConsStrings,
specifically ConsOneByteString and ConsTwoByteString. This feedback
can be used by TurboFan to only inline the relevant code for what
was seen so far. This allows us to remove the Octane/Splay specific
magic in JSTypedLowering to detect ConsString creation, and instead
purely rely on the feedback of what was seen so far (also making it
possible to change the semantics of NewConsString to be a low-level
operator, which is only introduced in SimplifiedLowering by looking
at the input types of StringConcat).
4. On top of the before mentioned type and interpreter changes we added
new operators CheckNonEmptyString, CheckNonEmptyOneByteString, and
CheckNonEmptyTwoByteString, which perform the appropriate (dynamic)
checks.
There are several more improvements that are possible based on this, but
since the change was already quite big, we decided not to put everything
into the first change, but do some follow up tweaks to the type system,
and builtin optimizations later.
Tbr: mstarzinger@chromium.org
Bug: v8:8834, v8:8931, v8:8939, v8:8951
Change-Id: Ia24e17c6048bf2b04df966d3cd441f0edda05c93
Cq-Include-Trybots: luci.chromium.try:linux-blink-rel
Doc: https://bit.ly/fast-string-concatenation-in-javascript
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1499497
Commit-Queue: Michael Achenbach <machenbach@chromium.org>
Reviewed-by: Yang Guo <yangguo@chromium.org>
Reviewed-by: Jaroslav Sevcik <jarin@chromium.org>
Reviewed-by: Mythri Alle <mythria@chromium.org>
Reviewed-by: Benedikt Meurer <bmeurer@chromium.org>
Cr-Commit-Position: refs/heads/master@{#60318}
This is a reland of 35269f77f8
Switches on an expression that unconditionally throws would have all their
case statements dead, causing a DCHECK error in the SwitchBuilder. This
fixes up the DCHECK to allow dead labels.
Original change's description:
> [ignition] Skip binding dead labels
>
> BytecodeLabels for forward jumps may create a dead basic block if their
> corresponding jump was elided (due to it dead code elimination). We can
> avoid generating such dead basic blocks by skipping the label bind when
> no corresponding jump has been observed. This works because all jumps
> except JumpLoop are forward jumps, so we only have to special case one
> Bind for loop headers to bind unconditionally.
>
> Since Binds are now conditional on a jump existing, we can no longer rely
> on using Bind to get the current offset (e.g. at the beginning of a try
> block). Instead, we now expose the current offset in the bytecode array
> writer. Conveniently, this means that we can be a bit smarter about basic
> blocks around these statements.
>
> As a drive-by, remove the unused Bind(target,label) function.
>
> Bug: chromium:934166
> Change-Id: I532aa452fb083560d07b90da99caca0b1d082aa3
> Reviewed-on: https://chromium-review.googlesource.com/c/1488763
> Commit-Queue: Leszek Swirski <leszeks@chromium.org>
> Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#59942}
TBR=rmcilroy@chromium.org
Bug: chromium:934166
Change-Id: If6eab4162106717ce64a2dc477000c6a76354cb4
Reviewed-on: https://chromium-review.googlesource.com/c/1494535
Reviewed-by: Leszek Swirski <leszeks@chromium.org>
Commit-Queue: Leszek Swirski <leszeks@chromium.org>
Cr-Commit-Position: refs/heads/master@{#59948}
This reverts commit 35269f77f8.
Reason for revert: Fuzzer unhappy: https://ci.chromium.org/p/v8/builders/ci/V8%20Fuzzer/29792
Original change's description:
> [ignition] Skip binding dead labels
>
> BytecodeLabels for forward jumps may create a dead basic block if their
> corresponding jump was elided (due to it dead code elimination). We can
> avoid generating such dead basic blocks by skipping the label bind when
> no corresponding jump has been observed. This works because all jumps
> except JumpLoop are forward jumps, so we only have to special case one
> Bind for loop headers to bind unconditionally.
>
> Since Binds are now conditional on a jump existing, we can no longer rely
> on using Bind to get the current offset (e.g. at the beginning of a try
> block). Instead, we now expose the current offset in the bytecode array
> writer. Conveniently, this means that we can be a bit smarter about basic
> blocks around these statements.
>
> As a drive-by, remove the unused Bind(target,label) function.
>
> Bug: chromium:934166
> Change-Id: I532aa452fb083560d07b90da99caca0b1d082aa3
> Reviewed-on: https://chromium-review.googlesource.com/c/1488763
> Commit-Queue: Leszek Swirski <leszeks@chromium.org>
> Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#59942}
TBR=rmcilroy@chromium.org,leszeks@chromium.org
Change-Id: I8118e54e0afa5e08b0a0a874c952f8a01f1c3242
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug: chromium:934166
Reviewed-on: https://chromium-review.googlesource.com/c/1494534
Reviewed-by: Sigurd Schneider <sigurds@chromium.org>
Commit-Queue: Sigurd Schneider <sigurds@chromium.org>
Cr-Commit-Position: refs/heads/master@{#59947}
BytecodeLabels for forward jumps may create a dead basic block if their
corresponding jump was elided (due to it dead code elimination). We can
avoid generating such dead basic blocks by skipping the label bind when
no corresponding jump has been observed. This works because all jumps
except JumpLoop are forward jumps, so we only have to special case one
Bind for loop headers to bind unconditionally.
Since Binds are now conditional on a jump existing, we can no longer rely
on using Bind to get the current offset (e.g. at the beginning of a try
block). Instead, we now expose the current offset in the bytecode array
writer. Conveniently, this means that we can be a bit smarter about basic
blocks around these statements.
As a drive-by, remove the unused Bind(target,label) function.
Bug: chromium:934166
Change-Id: I532aa452fb083560d07b90da99caca0b1d082aa3
Reviewed-on: https://chromium-review.googlesource.com/c/1488763
Commit-Queue: Leszek Swirski <leszeks@chromium.org>
Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
Cr-Commit-Position: refs/heads/master@{#59942}
For desrtucturing assignments from null/undefined, we throw an error
that references the destructuring object literal's property name, e.g.
for
var { x } = null;
we report that we cannot destructure 'x' from null.
Rather than calculating this property during bytecode generation (and
including it in the bytecode as an argument to the type error
constructor), we can calculate it at exception throwing time, by
re-parsing the source in a similar way to the existing call site
rendering.
This slightly decreases bytecode size and slightly decreases the amount
of work the bytecode compiler needs to do. In the future, it could also
allow us to give more detailed error messages, as we now have access to
the entire AST and are on the slow path anyway.
Bug: v8:6499
Change-Id: Icdbd4667db548b4e5e62ef97797a3771b5c1bf72
Reviewed-on: https://chromium-review.googlesource.com/c/1396080
Reviewed-by: Toon Verwaest <verwaest@chromium.org>
Reviewed-by: Sathya Gunasekaran <gsathya@chromium.org>
Commit-Queue: Leszek Swirski <leszeks@chromium.org>
Cr-Commit-Position: refs/heads/master@{#58706}
The 'done' setting dance in BuildFillArrayWithIterator turned out to
not be useful, as the StoreInArrayLiteral call could not ever throw an
exception. Since iterator exceptions count as done, we are guarnteed to
be done as soon as we enter the loop.
Change-Id: Ibe2ba1fcbe383bfcfedb185169890b6931cc7884
Reviewed-on: https://chromium-review.googlesource.com/c/1402792
Commit-Queue: Leszek Swirski <leszeks@chromium.org>
Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
Cr-Commit-Position: refs/heads/master@{#58695}
The IteratorClose spec specifies that exceptions in
%GetMethod(iterator.return) are not suppressed by exceptions in the
given continuation (body of a loop, assignments in destructuring),
while exceptions in the execution of iterator.return() are.
This means that we have to split out the property access + a typeof
check to be outside the try-catch, and keep the call inside of it.
The non-split version is only for cases when there is no 'throws'
continuation (as is the case for yield* calling IteratorClose), so
the existing BuildIteratorClose can be renamed to reflect this.
Change-Id: Id71aea4fddd6ffb986bd9aaa09d29615a8800f71
Reviewed-on: https://chromium-review.googlesource.com/c/1402789
Reviewed-by: Georg Neis <neis@chromium.org>
Commit-Queue: Leszek Swirski <leszeks@chromium.org>
Cr-Commit-Position: refs/heads/master@{#58694}
Emit a single destructuring assignment for destructuring declarations,
which can be desugared by the bytecode generator. This allows us to
remove destructuring desugaring from the parser (specifically, the
pattern rewriter) entirely.
The pattern "rewriter" is now only responsible for walking the
destructuring pattern to declare variables, mark them assigned, and
potentially rewrite scopes for the edge case of parameters with a sloppy
eval.
Note that since the rewriter is no longer rewriting, we have to flip the
VariableProxy copying logic for var re-lookup, so that we now pass the
new VariableProxy to the variable declaration and leave the original
unresolved (rather than passing the original through and rewriting to a
new unresolved VariableProxy).
This change does have some effect on breakpoint locations, due to some
of the available information changing between the parser and bytecode
generator, however the new locations appear to be more consistent
between assignments and declarations.
Change-Id: I3a58dd0a387d2bfb8e5e9e22dde0acc5f440cb82
Reviewed-on: https://chromium-review.googlesource.com/c/1382462
Commit-Queue: Leszek Swirski <leszeks@chromium.org>
Reviewed-by: Adam Klein <adamk@chromium.org>
Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
Reviewed-by: Yang Guo <yangguo@chromium.org>
Reviewed-by: Toon Verwaest <verwaest@chromium.org>
Cr-Commit-Position: refs/heads/master@{#58670}
Instead of de-sugaring destructuring assignment in the parser (using the
pattern rewriter), pass the Object/ArrayLiterals through to the bytecode
generator, which can desugar them in-place.
This allows us to decrease the amount of AST node creation, and improve
the generated bytecode using domain-specific knowledge. As a side effect
we partially fix an old execution ordering spec bug.
Currently only implemented for assignments, not declarations, as the
latter has some additional complexity.
Bug: v8:4951
Change-Id: I3d69d232bea2968ef20df68a74014d9e05808cfe
Reviewed-on: https://chromium-review.googlesource.com/c/1375660
Commit-Queue: Leszek Swirski <leszeks@chromium.org>
Reviewed-by: Adam Klein <adamk@chromium.org>
Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
Reviewed-by: Toon Verwaest <verwaest@chromium.org>
Cr-Commit-Position: refs/heads/master@{#58512}
