Reason for revert:
Speculative revert to see impact on crbug.com/659531
Original issue's description:
> [heap] Add a guard for restarting the memory reducer after mark-compact.
>
> Currently it is possible to get into a cycle of
> mark-compact -> memory reducer -> mark-compact -> memory reducer ...
> where the memory reducer does not free memory.
>
> This patch ensures that the memory reducer restarts only if the
> committed memory increased by sufficient amount after the last run.
>
> BUG=
TBR=hpayer@chromium.org,davidroutier17@gmail.com
# Not skipping CQ checks because original CL landed more than 1 days ago.
BUG=
Review-Url: https://codereview.chromium.org/2472053003
Cr-Commit-Position: refs/heads/master@{#40737}
Exchanged the ZoneList for a ZoneChunkList to avoid
unnecessary growing.
Review-Url: https://codereview.chromium.org/2468183004
Cr-Commit-Position: refs/heads/master@{#40736}
This is preparation for using TF to create builtins that handle variable number of
arguments and have to remove these arguments dynamically from the stack upon
return.
The gist of the changes:
- Added a second argument to the Return node which specifies the number of stack
slots to pop upon return in addition to those specified by the Linkage of the
compiled function.
- Removed Tail -> Non-Tail fallback in the instruction selector. Since TF now should
handles all tail-call cases except where the return value type differs, this fallback
was not really useful and in fact caused unexpected behavior with variable
sized argument popping, since it wasn't possible to materialize a Return node
with the right pop count from the TailCall without additional context.
- Modified existing Return generation to pass a constant zero as the additional
pop argument since the variable pop functionality
LOG=N
Review-Url: https://codereview.chromium.org/2446543002
Cr-Commit-Position: refs/heads/master@{#40699}
Reason for revert:
Seems to break arm64 sim debug and blocks roll:
https://build.chromium.org/p/client.v8.ports/builders/V8%20Linux%20-%20arm64%20-%20sim%20-%20debug/builds/3294
Original issue's description:
> [turbofan] Support variable size argument removal in TF-generated functions
>
> This is preparation for using TF to create builtins that handle variable number of
> arguments and have to remove these arguments dynamically from the stack upon
> return.
>
> The gist of the changes:
> - Added a second argument to the Return node which specifies the number of stack
> slots to pop upon return in addition to those specified by the Linkage of the
> compiled function.
> - Removed Tail -> Non-Tail fallback in the instruction selector. Since TF now should
> handles all tail-call cases except where the return value type differs, this fallback
> was not really useful and in fact caused unexpected behavior with variable
> sized argument popping, since it wasn't possible to materialize a Return node
> with the right pop count from the TailCall without additional context.
> - Modified existing Return generation to pass a constant zero as the additional
> pop argument since the variable pop functionality
>
> LOG=N
TBR=bmeurer@chromium.org,mstarzinger@chromium.org,epertoso@chromium.org,danno@chromium.org
# Not skipping CQ checks because original CL landed more than 1 days ago.
NOPRESUBMIT=true
Review-Url: https://codereview.chromium.org/2473643002
Cr-Commit-Position: refs/heads/master@{#40691}
This is preparation for using TF to create builtins that handle variable number of
arguments and have to remove these arguments dynamically from the stack upon
return.
The gist of the changes:
- Added a second argument to the Return node which specifies the number of stack
slots to pop upon return in addition to those specified by the Linkage of the
compiled function.
- Removed Tail -> Non-Tail fallback in the instruction selector. Since TF now should
handles all tail-call cases except where the return value type differs, this fallback
was not really useful and in fact caused unexpected behavior with variable
sized argument popping, since it wasn't possible to materialize a Return node
with the right pop count from the TailCall without additional context.
- Modified existing Return generation to pass a constant zero as the additional
pop argument since the variable pop functionality
LOG=N
Review-Url: https://codereview.chromium.org/2446543002
Cr-Commit-Position: refs/heads/master@{#40678}
While this seems like it should be true, the array buffer is not actually
neutered until the end of cloning. This is so that, if an exception is thrown
during serialization, the original array buffer is not left neutered. As a
result, Blink will not have neutered the buffer.
This fixes some DCHECK failures during layout tests.
BUG=chromium:148757
Review-Url: https://codereview.chromium.org/2466563002
Cr-Commit-Position: refs/heads/master@{#40675}
The majority of context slot accesses are to the local context (current context
register and depth 0), so this adds bytecodes to optimise for that case.
This cuts down bytecode size by roughly 1% (measured on Octane and Top25).
Review-Url: https://codereview.chromium.org/2459513002
Cr-Commit-Position: refs/heads/master@{#40641}
This is a new bytecode which behaves (for now) exactly like Call,
except that in turbofan graph building we can set the
ConvertReceiverMode to NotNullOrUndefined.
I observe a 1% improvement on Box2D, I'd expect a similar improvement on
other OOP heavy code.
Review-Url: https://codereview.chromium.org/2450243002
Cr-Commit-Position: refs/heads/master@{#40610}
Since ZoneLists are essentially non-standard ZoneVectors and have a bad
growing behaviour (ZoneList-allocations make up ~50% of website parse
zone memory) we should stop using them. The zone-containers are merely
a clean-up, with none of them actually better suited to be used with
zones. This new datastructure allows most operations of a LinkedList (
except pop_first and insertAt/removeAt) but uses about the same memory
as a well-initialized ZoneVector/ZoneList (<3% overhead with reasonably
large lists). It also never attempts to free memory again (which would
not work in zones anyway).
The ZoneChunkList is essentially a doubly-linked-list of arrays of
variable size.
Some test-results where I tried storing 16k pointers in different list
types (lists themselves also zone-allocated):
List type Zone memory used Time taken
-----------------------------------------------------------------------
Zone array (for comparison) 131072 B
Ideally initialized ZoneList 131088 B 0.062ms
ChunkZoneList 134744 B 0.052ms <--new thing
ZoneDeque 141744 B
ZoneLinkedList 393264 B
Initially empty ZoneList 524168 B 0.171ms <--right now
ChunkZoneList only push_front 524320 B
Review-Url: https://codereview.chromium.org/2449383002
Cr-Commit-Position: refs/heads/master@{#40602}
- Modifies RegisterConfiguration to specify complex aliasing on ARM 32.
- Modifies RegisterAllocator to consider aliasing.
- Modifies ParallelMove::PrepareInsertAfter to handle aliasing.
- Modifies GapResolver to split wider register moves when interference
with smaller moves is detected.
- Modifies MoveOptimizer to handle aliasing.
- Adds ARM 32 macro-assembler pseudo move instructions to handle cases where
split moves don't correspond to actual s-registers.
- Modifies CodeGenerator::AssembleMove and AssembleSwap to handle moves of
different widths, and moves involving pseudo-s-registers.
- Adds unit tests for FP operand interference checking and PrepareInsertAfter.
- Adds more tests of FP for the move optimizer and register allocator.
LOG=N
BUG=v8:4124
Review-Url: https://codereview.chromium.org/2410673002
Cr-Commit-Position: refs/heads/master@{#40597}
Modify the Bytecode Register Optimizer to be an independent component
rather than part of the BytecodePipeline. This means the BytecodeArrayBuilder
can explicitly call it with register operands when outputting a bytecode
and the Bytecode Register Optimizer doesn't need to work out which operands
are register operands. This also means we don't need to build BytecodeNodes
for Ldar / Star / Mov bytecodes unless they are actually emitted by the
optimizer.
This change also modifies the way the BytecodeArrayBuilder converts
operands to make use of the OperandTypes specified in bytecodes.h.
This avoids having to individually convert operands to their raw output
value before calling Output(...).
BUG=v8:4280
Review-Url: https://codereview.chromium.org/2393683004
Cr-Commit-Position: refs/heads/master@{#40543}
Additionally, remove all code related to the old-style slots filtering and black area end markers.
BUG=chromium:648568
Review-Url: https://chromiumcodereview.appspot.com/2440683002
Cr-Commit-Position: refs/heads/master@{#40494}
Added a size constraint to the configuration to limit the segment pool.
This will likely fix the memory alerts from small android devices.
BUG=chromium:655129
Review-Url: https://chromiumcodereview.appspot.com/2424393002
Cr-Commit-Position: refs/heads/master@{#40476}
Currently it is possible to get into a cycle of
mark-compact -> memory reducer -> mark-compact -> memory reducer ...
where the memory reducer does not free memory.
This patch ensures that the memory reducer restarts only if the
committed memory increased by sufficient amount after the last run.
BUG=
Review-Url: https://chromiumcodereview.appspot.com/2433933005
Cr-Commit-Position: refs/heads/master@{#40457}
This CL refactors the handling of metadata associated with WebAssembly
modules to reduce the duplicate marshalling of data from the C++ world
to the JavaScript world. It does this by wrapping the C++ WasmModule*
object in a Foreign that is rooted from the on-heap WasmCompiledModule
(which is itself just a FixedArray). Upon serialization, the C++ object
is ignored and the original WASM wire bytes are serialized. Upon
deserialization, the C++ object is reconstituted by reparsing the bytes.
This is motivated by increasing complications in implementing the JS
API, in particular WebAssembly.Table, which must perform signature
canonicalization across instances.
Additionally, this CL implements the proper base + offset initialization
behavior for tables.
R=rossberg@chromium.org,bradnelson@chromium.org,mtrofin@chromium.org,yangguo@chromium.org
BUG=v8:5507, chromium:575167, chromium:657316
Review-Url: https://chromiumcodereview.appspot.com/2424623002
Cr-Commit-Position: refs/heads/master@{#40434}
MIPS64 doesn't support Word32 compare instructions. Instead it relies
that the values in registers are correctly sign-extended and uses
Word64 comparison instead. This behavior is correct in most cases,
but doesn't work when comparing signed with unsigned operands.
The solution proposed here tries to match a comparison of signed
with unsigned operand, and perform Word32Compare simulation only
in those cases. Unfortunately, the solution is not complete because
it might skip cases where Word32 compare simulation is needed, so
basically it is a hack.
BUG=
TEST=mjsunit/compiler/uint32
Review-Url: https://codereview.chromium.org/2391393003
Cr-Commit-Position: refs/heads/master@{#40398}
EffectPhis can cause a cycle in a TurboFan graph. We delay the
processing of EffectPhis in the Int64Lowering to break these cycles. We
do the same already for Phis.
R=titzer@chromium.org
BUG=v8:5518
TEST=unittests/Int64LoweringTest.EffectPhiLoop
Review-Url: https://codereview.chromium.org/2428583002
Cr-Commit-Position: refs/heads/master@{#40378}
This adds more useful information to the v8-heap-stats tool.
BUG=v8:5489
Review-Url: https://codereview.chromium.org/2394213003
Cr-Commit-Position: refs/heads/master@{#40361}
Adds a boolean flag to the liveness analysis which makes it also analyze
the accumulator. This can help prevent the accumulator escaping loops,
as well as decreasing the number of distinct state values nodes in the
graph.
The flag is a kind of ugly way to hack this in, however it is probably
the simplest to add, and (more importantly) to remove once the AST graph
builder is gone.
I measure a 2.6% improvement on Mandreel on my x64 machine, and a ~2%
improvement on Navier-Stokes. Other improvements are expected.
Review-Url: https://codereview.chromium.org/2428503002
Cr-Commit-Position: refs/heads/master@{#40359}
This allows people writing code stubs to just verify the graph of the stub they're working on, at least until we fix all of the issues we have and enable the verification by default.
Also fixes representations in CodeStubAssembler::SmiOr and InterpreterAssembler::StarDispatchLookahead.
R=bmeurer@chromium.org
BUG=
Review-Url: https://codereview.chromium.org/2413653006
Cr-Commit-Position: refs/heads/master@{#40320}
This CL also introduces a NoBarrierAtomicValue with NoBarrier accessors.
BUG=chromium:648568
Review-Url: https://codereview.chromium.org/2408233004
Cr-Commit-Position: refs/heads/master@{#40213}
For the asm.js to WASM pipeline, the current stack traces only show
low-level WASM information.
This CL maps this back to asm.js source positions.
It does so by attaching the asm.js source Script to the compiled WASM
module, and emitting a delta-encoded table which maps from WASM byte
offsets to positions within that Script. As asm.js code does not throw
exceptions, we only store a mapping for call instructions.
The new AsmJsWasmStackFrame implementation inherits from
WasmStackFrame, but contains the logic to provide the source script and
the position inside of it.
What is still missing is the JSFunction object returned by
CallSite.getFunction(). We currently return null.
R=jgruber@chromium.org, titzer@chromium.org
BUG=v8:4203
Review-Url: https://codereview.chromium.org/2404253002
Cr-Commit-Position: refs/heads/master@{#40205}
If passing <nullptr, 0> to the decoder and trying to decode something,
it correctly detects the error and sets an error message, but still
returns true on ok(), and returns a valid result.
I triggered this error by passing a null Vector, returned by FindSection(), to
the decoder.
R=titzer@chromium.org
Review-Url: https://codereview.chromium.org/2410913002
Cr-Commit-Position: refs/heads/master@{#40204}
- Adds an optional representation field to VReg and TestOperand structs.
- Adds a simple FP allocation test to register-allocator-unittest.cc.
- Adds some simple FP tests to move-optimizer-unittest.cc.
LOG=N
BUG=v8:4124
Review-Url: https://codereview.chromium.org/2400513002
Cr-Commit-Position: refs/heads/master@{#40117}
There were once plans to generate cross-context code with TurboFan,
however that doesn't fit into the model anymore, and so all of this
is essentially dead untested code (and thus most likely already broken
in subtle ways). With this mode still in place it would also be a lot
harder to make inlining based on SharedFunctionInfo work.
BUG=v8:2206,v8:5499
R=jarin@chromium.org
Review-Url: https://codereview.chromium.org/2406803002
Cr-Commit-Position: refs/heads/master@{#40109}
Found with libfuzzer. The length is automatically converted to int (thus
large sizes could become negative, even though they are legal "array sizes").
Besides that, the length is coerced to a SMI (which is an even tighter
constraint on 32-bit systems, where it limits the legal sizes to 2^30 - 1).
Add checks that the length of a dense array is below that threshold, and also
fail fast if a length that is provided obviously could not be the correct dense
length (because there isn't enough data left in the buffer to populate such an
array).
BUG=chromium:148757
Review-Url: https://codereview.chromium.org/2399873002
Cr-Commit-Position: refs/heads/master@{#40094}
Previously this would result in applying trying to find a size modulo zero,
which causes SIGFPE. This approach was preferred over adding a default case
to preserve the ability of the compiler to detect unhandled switch cases
(within the valid range of the enum).
BUG=chromium:148757
Review-Url: https://codereview.chromium.org/2395073003
Cr-Commit-Position: refs/heads/master@{#40088}
