Most parallel moves do not actually require the heavy analysis in the
GapResolver since there is no overlap between source and destination
registers. Handle some of these cases by a fast-path to avoid the
quadratic behaviour in the general case.
The fast path is taken for about 2/3 of the non-empty parallel moves on
wasm workloads.
R=herhut@chromium.org, titzer@chromium.org
Bug: v8:8423
Change-Id: I026256fd6cf268f176fccc97f0f427e07629aeb3
Reviewed-on: https://chromium-review.googlesource.com/c/1400410
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Reviewed-by: Stephan Herhut <herhut@chromium.org>
Reviewed-by: Ben Titzer <titzer@chromium.org>
Cr-Commit-Position: refs/heads/master@{#58644}
These flakes can't be reproduced locally so we need more information
when they fail. Add some logging so that we can figure out why they
are breaking.
Bug: v8:8649, v8:8648
Change-Id: I2fb1384bb7592c6fc68c08952505e79329f00bec
Reviewed-on: https://chromium-review.googlesource.com/c/1400418
Commit-Queue: Peter Marshall <petermarshall@chromium.org>
Reviewed-by: Yang Guo <yangguo@chromium.org>
Cr-Commit-Position: refs/heads/master@{#58641}
Within an inline stack we would have multiple copies of the exact same
CodeEntry object to represent an inline frame. We had one copy for every
time that the frame appeared in an inline stack. One CodeEntry can have
multiple inline stacks and each stack can have multiple inline frames.
In the common case, the stacks overlap and repeat frames.
This CL creates a single CodeEntry object to represent each inlined
function as an inline frame (for a given CodeEntry with inlinings). This
removes most of the duplication of inline CodeEntry objects. We still
have some duplication, e.g. if we inline bar() into foo() and foo2() but
they are not themselves inlined into anything, then we will have two
inline CodeEntry objects for bar(). Removing all duplication is harder
to achieve because the lifetime of the inlined frame CodeEntry is now no
longer tied to the inliner.
Get rid of the InlineEntry struct as it is now indentical to
CodeEntryAndLineNumber.
We store the list of canonical inline CodeEntry objects on the
CodeObject of the inlining function so that it can own the lifetimes of
inlined frames.
Also rename inline_locations_ to inline_stacks_ to be clearer.
Bug: v8:7719
Change-Id: Ied765b4cce7fd33f3290798331f1e6767cc42e8c
Reviewed-on: https://chromium-review.googlesource.com/c/1396086
Commit-Queue: Peter Marshall <petermarshall@chromium.org>
Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
Reviewed-by: Alexei Filippov <alph@chromium.org>
Cr-Commit-Position: refs/heads/master@{#58639}
Remove the use of a jump table in the prologue of the deopt entries
and instead pass the bailout id explicitly in a register when calling
the deopt entry routine from optimized code. This unifies the logic
with the way the Arm64 code works. It saves the following amount of
memory in code stubs:
- arm: 384KB
- ia32: 480KB
- x64: 240KB
This could be offset by a slight increase in the size of optimized code
for loading the immediate, however this impact should be minimal and
will scale with the maximum number of bailout ids (e.g., the size of
code will increase by one instruction per bailout id on Arm, therefore
~98,000 bailouts will be needed before the overhead is greater than
the current fixed table size).
Change-Id: I838604b48fa04cbd45320c7b9dac0de08fd8eb25
Reviewed-on: https://chromium-review.googlesource.com/c/1398224
Commit-Queue: Ross McIlroy <rmcilroy@chromium.org>
Reviewed-by: Jaroslav Sevcik <jarin@chromium.org>
Cr-Commit-Position: refs/heads/master@{#58636}
Always precheck that the PreparseData has data before serializing it.
Drive-by-fix:
- rename preparsed_scope_data_builder_ to preparse_data_builder_
Change-Id: I8e709af8f69db44e03caa9132f06a7b8c906bfdb
Reviewed-on: https://chromium-review.googlesource.com/c/1387305
Reviewed-by: Toon Verwaest <verwaest@chromium.org>
Commit-Queue: Camillo Bruni <cbruni@chromium.org>
Cr-Commit-Position: refs/heads/master@{#58635}
This reverts commit ac2fb66b65.
Reason for revert: Flakily crashes on several bots:
https://ci.chromium.org/p/v8/builders/luci.v8.ci/V8%20Win32/18524https://ci.chromium.org/p/v8/builders/luci.v8.ci/V8%20Win64%20-%20msvc/6824https://ci.chromium.org/p/v8/builders/luci.v8.ci/V8%20Linux64%20-%20internal%20snapshot/19766
Original change's description:
> [wasm] Remove finisher task
>
> This removes the finisher task and instead finishes compilation units
> from the background.
> It also changes ownership of the AsyncCompileJob to be shared among all
> tasks that still operate on it. The AsyncCompileJob dies when the last
> reference dies.
>
> R=ahaas@chromium.org
> CC=mstarzinger@chromium.org
>
> Bug: v8:7921, v8:8423
> Change-Id: Id09378327dfc146459ef41bc97176a8716756ae4
> Cq-Include-Trybots: luci.v8.try:v8_linux64_tsan_rel
> Reviewed-on: https://chromium-review.googlesource.com/c/1335553
> Reviewed-by: Andreas Haas <ahaas@chromium.org>
> Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#58630}
TBR=ahaas@chromium.org,clemensh@chromium.org
Change-Id: I6b332b66adaec8f713fb31f4c8517cae7ebb4645
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug: v8:7921, v8:8423
Cq-Include-Trybots: luci.v8.try:v8_linux64_tsan_rel
Reviewed-on: https://chromium-review.googlesource.com/c/1400420
Reviewed-by: Michael Achenbach <machenbach@chromium.org>
Commit-Queue: Michael Achenbach <machenbach@chromium.org>
Cr-Commit-Position: refs/heads/master@{#58634}
Simple experiment that prefers free registers that are not used for
arguments to prevent cases where we allocate a free register without
hint and thereby block later uses of that register for no good
reason.
Bug: v8:8311
Change-Id: I95e96b150410e97937cb72d575ae6bece9ee08f9
Reviewed-on: https://chromium-review.googlesource.com/c/1397668
Reviewed-by: Sigurd Schneider <sigurds@chromium.org>
Commit-Queue: Stephan Herhut <herhut@chromium.org>
Cr-Commit-Position: refs/heads/master@{#58633}
This removes the finisher task and instead finishes compilation units
from the background.
It also changes ownership of the AsyncCompileJob to be shared among all
tasks that still operate on it. The AsyncCompileJob dies when the last
reference dies.
R=ahaas@chromium.org
CC=mstarzinger@chromium.org
Bug: v8:7921, v8:8423
Change-Id: Id09378327dfc146459ef41bc97176a8716756ae4
Cq-Include-Trybots: luci.v8.try:v8_linux64_tsan_rel
Reviewed-on: https://chromium-review.googlesource.com/c/1335553
Reviewed-by: Andreas Haas <ahaas@chromium.org>
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Cr-Commit-Position: refs/heads/master@{#58630}
Previously we'd always push variable proxies into the unresolved list of the
current scope, and possibly delete them from the list later in case they end up
being declarations. If variables become assigned, there were two ways to mark
them as such: The preparser would marked the variables tracked on the
PreParserExpression, and the parser would traverse the LHS AST to find and mark
all variables.
After this CL, if the scope already knows it's tracking declarations, the
variables are never added to the unresolved list in the first place. If the
scope is ambigous, it tracks the variable proxies on the side and only adds
them to the unresolved list if they end up being references rather than
declarations. The same list is now used to bulk mark all LHS variables as
assigned; uniformely for both the parser and the preparser.
In a next step we'll also use the scope to create declarations. That way we can
stop tracking variables_ on PreParserExpression altogether.
Change-Id: I6ada37006cc2e066731f29cd4ea314550fc7959f
Reviewed-on: https://chromium-review.googlesource.com/c/1397669
Commit-Queue: Toon Verwaest <verwaest@chromium.org>
Reviewed-by: Leszek Swirski <leszeks@chromium.org>
Cr-Commit-Position: refs/heads/master@{#58629}
Print all the mismatch failures in the bytecode rather than aborting at
the first mismatch.
R=rmcilroy
Change-Id: Id572ead5fdc4d126ac9a05942f940b0eaef7150f
Reviewed-on: https://chromium-review.googlesource.com/c/1400412
Commit-Queue: Dan Elphick <delphick@chromium.org>
Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
Cr-Commit-Position: refs/heads/master@{#58628}
Backend is split out, so remove backend owners from src/compiler. Also,
mention explicitly in src/compiler/backend that also all src/compiler
owners are owners there.
R=titzer@chromium.org
No-Try: true
Change-Id: I5409946f65bf27337b715af555083a4804fbb8dd
Reviewed-on: https://chromium-review.googlesource.com/c/1400411
Reviewed-by: Ben Titzer <titzer@chromium.org>
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Cr-Commit-Position: refs/heads/master@{#58626}
This CL adds support for anyref in WebAssembly.Global objects. Note
that the specification is not complete yet in this area.
I did the following changes:
- I renamed the `array_buffer` field of WasmGlobalObject to
`untagged_buffer`
- I added an additional field of type FixedArray, `tagged_buffer`.
- In the constructor of WasmGlobalObject I allocate either the former
or the latter, but not both.
- In the WebAssembly.Global constructor I added special handling for
the case where no initial value is provided. In that case I set the
inital value to `null` and not to `undefined`.
R=titzer@chromium.org
Bug: v8:7581
Change-Id: I7e4855d7e6c04a9bcdc7ebd450caca5819d060e2
Reviewed-on: https://chromium-review.googlesource.com/c/1398226
Commit-Queue: Andreas Haas <ahaas@chromium.org>
Reviewed-by: Ben Titzer <titzer@chromium.org>
Cr-Commit-Position: refs/heads/master@{#58625}
If we have both f32 and f64 locals, we use the same register to hold
their zero value. On stack transfers, we might thus encounter the same
fp register with both the f32 and f64 type. Explicitly allow that case
to happen.
R=ahaas@chromium.org
Bug: chromium:918917, v8:6600
Change-Id: I6937008d38853fe2bdccd9715e1a2499cf6bf7c6
Reviewed-on: https://chromium-review.googlesource.com/c/1398225
Reviewed-by: Andreas Haas <ahaas@chromium.org>
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Cr-Commit-Position: refs/heads/master@{#58623}
On ia32, the instruction selector uses movsx_b to compile the wasm
SignExtendWord8ToInt32 instruction. movsx_b requires a byte register
as input. However, not all allocatable registers on ia32 are. As we
cannot currently express constraints on subsets of registers, this
change now forces the input to movsx_b into eax.
Bug: chromium:919572
Change-Id: I39bd391974954ec9044940c3164398109eb78908
Reviewed-on: https://chromium-review.googlesource.com/c/1400409
Reviewed-by: Michael Starzinger <mstarzinger@chromium.org>
Commit-Queue: Stephan Herhut <herhut@chromium.org>
Cr-Commit-Position: refs/heads/master@{#58622}
Only look 5 frames up the stack when looking for a DCHECK to move the
frame to to prevent excessive iteration especially after a stack
overflow.
Change-Id: I227c46596f09c9af0a47e6673d3165eaccb75163
Reviewed-on: https://chromium-review.googlesource.com/c/1400408
Reviewed-by: Leszek Swirski <leszeks@chromium.org>
Commit-Queue: Dan Elphick <delphick@chromium.org>
Cr-Commit-Position: refs/heads/master@{#58621}
Two uses in the API needed adaptation; all other uses have already
been subsumed by the new implementation (previously known as
NeverReadOnlySpaceObjectPtr, here renamed to NeverReadOnlySpaceObject).
Bug: v8:3770
Change-Id: Idf0e4a98a407b9afea22e8790da34cf017b892a5
Reviewed-on: https://chromium-review.googlesource.com/c/1397671
Commit-Queue: Jakob Kummerow <jkummerow@chromium.org>
Reviewed-by: Sigurd Schneider <sigurds@chromium.org>
Reviewed-by: Yang Guo <yangguo@chromium.org>
Reviewed-by: Leszek Swirski <leszeks@chromium.org>
Cr-Commit-Position: refs/heads/master@{#58620}
StackHandlers form a chain, where the last element is nullptr,
so calling "handler->next()->foo()" is unsafe because "foo"
might see "this == nullptr".
Bug: v8:3770
Change-Id: Ic989384fa192e29d4d8cb76ff01b32173bf55fd9
Reviewed-on: https://chromium-review.googlesource.com/c/1400406
Reviewed-by: Yang Guo <yangguo@chromium.org>
Commit-Queue: Jakob Kummerow <jkummerow@chromium.org>
Cr-Commit-Position: refs/heads/master@{#58619}
Graph width is now managed by the Graph instead of the GraphView,
which simplifies some interfaces.
Change-Id: If78bc9a469cc8369bc75695a6612627103036bc8
Notry: true
Bug: v8:7327
Reviewed-on: https://chromium-review.googlesource.com/c/1398227
Reviewed-by: Georg Neis <neis@chromium.org>
Commit-Queue: Sigurd Schneider <sigurds@chromium.org>
Cr-Commit-Position: refs/heads/master@{#58618}
This CL prevents redundancy elimination from widening types, which
can cause problems if the input of a DeadValue (which has type None)
is replaced by an equivalent node that does not have type None. This
can happen because load elimination does not re-type nodes, for
example.
Bug: chromium:919340
Change-Id: I89e872412edbcdc610e70ae160cde56cd045006c
Reviewed-on: https://chromium-review.googlesource.com/c/1397709
Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
Reviewed-by: Jaroslav Sevcik <jarin@chromium.org>
Commit-Queue: Sigurd Schneider <sigurds@chromium.org>
Cr-Commit-Position: refs/heads/master@{#58617}
Prior to this CL compilation fails with:
- 'error: offset of on non-standard-layout type' due to offsetof()
- 'Assertion failed: vector subscript out of range' due to the OOB vector subscripts
Change-Id: I8751fafd1058ca839de832267811f8f1f47c53fe
Reviewed-on: https://chromium-review.googlesource.com/c/1400404
Reviewed-by: Michael Lippautz <mlippautz@chromium.org>
Commit-Queue: Jakob Gruber <jgruber@chromium.org>
Cr-Commit-Position: refs/heads/master@{#58616}
The two names refer to the same thing by now, so this patch is
entirely mechanical.
Bug: v8:3770
Change-Id: Ia360c06c89af6b3da27fd21bbcaeb2bdaa28ce22
Reviewed-on: https://chromium-review.googlesource.com/c/1397705
Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
Reviewed-by: Jakob Gruber <jgruber@chromium.org>
Reviewed-by: Michael Lippautz <mlippautz@chromium.org>
Commit-Queue: Jakob Kummerow <jkummerow@chromium.org>
Cr-Commit-Position: refs/heads/master@{#58615}
|argc| parameter of JSEntry is passed as int from C++ code, and loaded
into a register on the asm code. As int is 32 bit, and registers are
64 bit on 64 bit platforms, upper 32 bits of the loaded value may be
contaminated by a random value if it's passed as a stack parameter.
For now, |argc| is passed as a register parameter on all platforms, and
the upper 32 bits of |argc| is filled by zero, fortunately. However, if
we shuffle the order of parameters, |argc| can be passed as a stack
parameter and its value may be broken.
Specifically on x64 Windows, the first 4 parameters are passed as
register parameters and the rest are stack parameters. As |argc| is the
4th parameter, if we prepend another parameter and shift |argc| to
the 5th parameter, |argc| will become a stack parameter and its load
to 64 bit register breaks the value.
This CL converts the type of the |argc| parameter to intptr_t, so that
it's safe to load from stack to full width registers.
Bug: v8:8124
Change-Id: Ie7407cf5e6252ed7323a9c42389db387b0064673
Reviewed-on: https://chromium-review.googlesource.com/c/1400326
Reviewed-by: Jakob Gruber <jgruber@chromium.org>
Commit-Queue: Taiju Tsuiki <tzik@chromium.org>
Cr-Commit-Position: refs/heads/master@{#58614}
This is a reland of part of
https://chromium-review.googlesource.com/c/v8/v8/+/1397664.
It drops the explicit fni_.Infer() call after parsing arrow functions. We'll
want to avoid inferring if the arrow function is an argument to a function
call.
It also avoids adding the single argument of "name => " to the inferred name.
Bug: chromium:916975
Change-Id: I96a934408113483d73eba14073fe21e8cfe2ada6
Reviewed-on: https://chromium-review.googlesource.com/c/1397665
Commit-Queue: Toon Verwaest <verwaest@chromium.org>
Reviewed-by: Maya Lekova <mslekova@chromium.org>
Cr-Commit-Position: refs/heads/master@{#58613}
The callback to AsyncStreamingProcessor::OnFinishedStream would
call into v8 internals without restoring the current context if
the processed module had no code. Instead, now always restore the
context before doing any finishing work.
Bug: chromium:915493
Change-Id: Ib779df81301ad1e3597515a4173c9a57efc593ac
Reviewed-on: https://chromium-review.googlesource.com/c/1397672
Reviewed-by: Andreas Haas <ahaas@chromium.org>
Commit-Queue: Stephan Herhut <herhut@chromium.org>
Cr-Commit-Position: refs/heads/master@{#58612}
We currently hold the register list for {move_src_regs_} in the
{StackTransferRecipe} class, and only compute src use counts for
executing the moves. Since the use counts and the register list are
redundant, just compute the use counts right away, and get rid of the
register list.
This saves one iteration of the register list in {ExecuteMoves}.
R=ahaas@chromium.org
Bug: v8:6600, v8:8423
Change-Id: I832fb0d1c1d3afe536289162a81a49b73313e7f4
Reviewed-on: https://chromium-review.googlesource.com/c/1397670
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Reviewed-by: Andreas Haas <ahaas@chromium.org>
Cr-Commit-Position: refs/heads/master@{#58611}
noexcept keyword needs to be consistent between the declaration and
implementation in C++17.
Bug: v8:8616, chromium:752720
Change-Id: Iff4022c8c4b861ebdbe8e08995af1bc4da866dae
Reviewed-on: https://chromium-review.googlesource.com/c/1396459
Reviewed-by: Clemens Hammacher <clemensh@chromium.org>
Reviewed-by: Michael Lippautz <mlippautz@chromium.org>
Commit-Queue: Taiju Tsuiki <tzik@chromium.org>
Cr-Commit-Position: refs/heads/master@{#58609}
Fix WebAssembly's global/constructor js-api. Globals with a value
of i64 is now valid even if Wasm BigInt feature isn't activated.
Bug: v8:8319
Cq-Include-Trybots: luci.chromium.try:linux-blink-rel
Change-Id: Ia41ad69efa5253064ecdb8f59b149393cd672b68
Reviewed-on: https://chromium-review.googlesource.com/c/1382747
Commit-Queue: Ben Smith <binji@chromium.org>
Reviewed-by: Adam Klein <adamk@chromium.org>
Reviewed-by: Ben Smith <binji@chromium.org>
Cr-Commit-Position: refs/heads/master@{#58603}
This patch sets the name slot of the private name symbols for
private fields and display the names in error messages of invalid
private field accesses.
TBR: adamk@chromium.org
Bug: v8:8144
Change-Id: Id34c468e2bddd1c3001517b4d447c7497402df76
Reviewed-on: https://chromium-review.googlesource.com/c/1374332
Reviewed-by: Camillo Bruni <cbruni@chromium.org>
Reviewed-by: Igor Sheludko <ishell@chromium.org>
Reviewed-by: Mathias Bynens <mathias@chromium.org>
Reviewed-by: Ulan Degenbaev <ulan@chromium.org>
Reviewed-by: Mythri Alle <mythria@chromium.org>
Reviewed-by: Sathya Gunasekaran <gsathya@chromium.org>
Commit-Queue: Joyee Cheung <joyee@igalia.com>
Cr-Commit-Position: refs/heads/master@{#58601}
This queue is used for transferring tick samples between the foreground
thread which creates them and the background thread which processes
them.
I've tested this on the node server example that we are using to
measure memory use and found that we never fill the queue at this size.
The load factor of the queue is basically a measure of how fast the
producer pushes to it and how fast the consumer processes samples from
it. To load test the configuration a bit I reduced the sampling
interval from 1000us (1000 samples/sec) to 50us (20,000 samples/sec).
At this rate we still only use 196/251 available slots in the queue at
peak load (measurement taken by keeping a running max of #slots used,
taken at StartEnqueue()).
The default sampling interval is 1000us. 512 KiB ought be enough for
anybody!
Bug: v8:7719
Change-Id: I93cc1119d3549a319d2db8b831781712bfb88613
Reviewed-on: https://chromium-review.googlesource.com/c/1397704
Reviewed-by: Alexei Filippov <alph@chromium.org>
Reviewed-by: Yang Guo <yangguo@chromium.org>
Commit-Queue: Peter Marshall <petermarshall@chromium.org>
Cr-Commit-Position: refs/heads/master@{#58600}
