On Intel, xchg stalls the pipeline. We use xchg to implement swap
moves. In a separate exploration, the presence of xchg in a very hot
loop, due to a change in register allocation, lead to over 20% regression.
Simply changing that instruction with push/mov/pop (almost) eliminated
the regression.
In light of that, I removed uses of xchg. This leads to more instructions,
though. That is particularly problematic for long cycles, which, today,
we translate to successions of swaps.
I plan to address this cycle issue in a separate change. For now, the
goal is to unblock the initial work that lead here.
Review URL: https://codereview.chromium.org/1580233003
Cr-Commit-Position: refs/heads/master@{#33282}
port 322ffda30d (r33265)
original commit message:
Also migrate the Number constructor to a native builtin, using the
same mechanism already used by the String constructor. Otherwise just
parsing and compiling the Number constructor to optimized code already
eats 2ms on desktop for no good reason, and the resulting optimized
code is not even close to awesome.
Drive-by-fix: Use correct context for the [[Construct]] case of the
String constructor as well, and share some code with it.
BUG=
Review URL: https://codereview.chromium.org/1581313002
Cr-Commit-Position: refs/heads/master@{#33280}
port 12bcba1543 (r33258)
original commit message:
The API functions are always in sloppy mode, so receiver is always a
JSReceiver once the actual call trampoline runs, no need to check again
in various places.
BUG=
Review URL: https://codereview.chromium.org/1582173002
Cr-Commit-Position: refs/heads/master@{#33278}
The bulk of these were marked simply '[PASS]', while one was marked
'[PASS, FAIL]' but seems to always pass.
BUG=v8:3305, v8:4405
LOG=n
Review URL: https://codereview.chromium.org/1584863003
Cr-Commit-Position: refs/heads/master@{#33277}
Port 322ffda30d
Original commit message:
Also migrate the Number constructor to a native builtin, using the
same mechanism already used by the String constructor. Otherwise just
parsing and compiling the Number constructor to optimized code already
eats 2ms on desktop for no good reason, and the resulting optimized
code is not even close to awesome.
Drive-by-fix: Use correct context for the [[Construct]] case of the
String constructor as well, and share some code with it.
R=bmeurer@chromium.org, joransiu@ca.ibm.com, jyan@ca.ibm.com, michael_dawson@ca.ibm.com
BUG=
Review URL: https://codereview.chromium.org/1583903002
Cr-Commit-Position: refs/heads/master@{#33272}
exception operand in a handler block. This avoids confusing
unrelated ranges, which may happen if escape analysis elides
the exception operand.
BUG=chromium:576195
LOG=N
Review URL: https://codereview.chromium.org/1579603005
Cr-Commit-Position: refs/heads/master@{#33271}
Removes assignment hazard scope. Reverts back to the naive scheme of
allocating a temporary for every variable load. It was decided to revert it
because the current implementation does not handle logical expressions,
ternary operators, visiting objects in named/keyed loads. Also, we wanted
to evaluate alternate approaches and choose one when we have a mechanism
to measure performance.
BUG=v8:4280
LOG=N
Review URL: https://codereview.chromium.org/1576403004
Cr-Commit-Position: refs/heads/master@{#33269}
Once we use type feedback, we need to reflect the feedback in the types, propagate
the new narrower types forward and use them in the subsequent
representation inference. This CL propagates and uses the recomputed types
for Phi and Select nodes (rather than using the types from the typer).
Review URL: https://codereview.chromium.org/1580353003
Cr-Commit-Position: refs/heads/master@{#33268}
This bug improves performance of escape analysis.
* A allocation discovery phase (EscapeAnalysis::AssignAliases)
ensures compact representation of virtual state
* Node revisiting in EscapeStatusAnalysis has been improved
* Escape analysis no longer requires a trimmed graph
BUG=v8:4586
LOG=n
Review URL: https://codereview.chromium.org/1559123003
Cr-Commit-Position: refs/heads/master@{#33267}
Also migrate the Number constructor to a native builtin, using the
same mechanism already used by the String constructor. Otherwise just
parsing and compiling the Number constructor to optimized code already
eats 2ms on desktop for no good reason, and the resulting optimized
code is not even close to awesome.
Drive-by-fix: Use correct context for the [[Construct]] case of the
String constructor as well, and share some code with it.
R=jarin@chromium.org
Review URL: https://codereview.chromium.org/1573243009
Cr-Commit-Position: refs/heads/master@{#33265}
ReportFailedAccessCheck() sometimes doesn't throw an exception (it's up
to the embedder). Pretend that OrdinaryDefineOwnProperty() succeeded in
that case. This is consistent with existing behavior of other methods:
- JSObject::DefineOwnPropertyIgnoreAttributes
- JSObject::SetPropertyWithFailedAccessCheck
- Object::SetPropertyWithAccessor
See also commit f66c3f5c35.
BUG=chromium:574217
LOG=n
R=neis@chromium.org
Review URL: https://codereview.chromium.org/1580823002
Cr-Commit-Position: refs/heads/master@{#33262}
The API functions are always in sloppy mode, so receiver is always a
JSReceiver once the actual call trampoline runs, no need to check again
in various places.
R=verwaest@chromium.org
Review URL: https://codereview.chromium.org/1575973006
Cr-Commit-Position: refs/heads/master@{#33258}
Failing to do so results in out-of-date marking information, because live bytes
is not properly adjusted.
This CL adds support for right trimming ByteArray and properly DCHECKs that we
do not left trim ByteArray (as we already do for FixedTypedArrayBase).
BUG=
Review URL: https://codereview.chromium.org/1577263003
Cr-Commit-Position: refs/heads/master@{#33252}
The hash calculation was dependent on upper part of |inner_pointer| and caused non-deterministic cache miss events which in turn caused non-deterministic progress of pages sweeping (see GcSafeFindCodeForInnerPointer()).
Review URL: https://codereview.chromium.org/1582573002
Cr-Commit-Position: refs/heads/master@{#33246}
There's no need to mark these as possibly-direct-eval, since all such an
eval-tagged string will ever get passed is the array of string parts, which
will be immediately returns (since it's not a string). It will
never do a lookup in the current scope, nor (in sloppy mode) introduce
new declarations.
This patch is not intended to change behavior, but I've added tests that
demonstrate the stuff explained in the preceding paragraph.
Review URL: https://codereview.chromium.org/1580463003
Cr-Commit-Position: refs/heads/master@{#33245}
This fixes a number of crashes where other code was assuming there would
be at least one deopt point in all optimized functions (i.e., the
StackCheck) but we weren't producing any.
BUG=v8:4280
LOG=N
Review URL: https://codereview.chromium.org/1572333002
Cr-Commit-Position: refs/heads/master@{#33243}
No need to distribute the setup of the Function global property across
three different places, instead do everything in a single place during
bootstrapping.
R=verwaest@chromium.org
Review URL: https://codereview.chromium.org/1577703005
Cr-Commit-Position: refs/heads/master@{#33242}
That will allow for adding private symbols to JSProxies in a follow-up
change
BUG=chromium:571365
R=neis@chromium.org,verwaest@chromium.org,rossberg@chromium.org
LOG=n
Review URL: https://codereview.chromium.org/1575423002
Cr-Commit-Position: refs/heads/master@{#33241}
Adds support for LOOKUP_SLOT_CALL calls to the interpreter. Also changes
VisitCall to keep callee and reciever consecutive to avoid register
shuffles when performing LOOKUP_SLOT_CALL calls. Adds tests for the
interpreter and bytecode graph generator.
BUG=v8:4280
LOG=N
Review URL: https://codereview.chromium.org/1568323002
Cr-Commit-Position: refs/heads/master@{#33237}
- Simplify the variable-length pop sequence on entry. (It now uses
smaller code with no branches.)
- Use conditional compare to merge branches where appropriate.
- Make use of Ldrsw + UntagSmiFieldMemOperand to load smis more
efficiently.
- Only load 'undefined' and 'null' once per builtin.
- A few other small improvements.
BUG=
Review URL: https://codereview.chromium.org/1576403002
Cr-Commit-Position: refs/heads/master@{#33235}
We use a scratchpad to remember visited allocation sites for post processing
(making tenure decisions). The previous implementation used a rooted FixedArray
with constant length (256) to remember all sites. Updating the scratchpad is a
bottleneck in any parallel/concurrent implementation of newspace evacuation.
The new implementation uses a HashMap with allocation sites as keys and
temporary counts as values. During evacuation we collect a local hashmap of
visited allocation sites. Upon merging the local hashmap back into a global one
we update potential forward pointers of compacted allocation sites. The
scavenger can directly enter its entries into the global hashmap. Note that the
actual memento found count is still kept on the AllocationSite as it needs to
survive scavenges and full GCs.
BUG=chromium:524425
LOG=N
R=hpayer@chromium.org
Review URL: https://codereview.chromium.org/1535723002
Cr-Commit-Position: refs/heads/master@{#33233}
