The assumption that every function body produces a value does not hold
for functions that e.g. unconditionally throw or endlessly loop. This
fixes the inlining logic to handle such cases.
R=bmeurer@chromium.org
TEST=mjsunit/regress/regress-crbug-530598
BUG=chromium:530598
LOG=n
Review URL: https://codereview.chromium.org/1333193005
Cr-Commit-Position: refs/heads/master@{#30738}
(reason for revert/reland: patch incorrectly left --vector-stores flag
on, helpfully revealing some gcstress issues to look at, but they
don't need to block this CL).
Some pretty hacky code was used to carry out the tail-call
handler dispatch on ia32 vector stores due to a lack
of free registers. It really tanks performance. A better
approach is to use a virtual register on the isolate.
BUG=
TBR=jkummerow@chromium.org, vogelheim@chromium.org
Review URL: https://codereview.chromium.org/1346573002
Cr-Commit-Position: refs/heads/master@{#30737}
Returning a result array with holey elements kind in where we actually have a packed kind causes performance regressions.
LOG=N
BUG=chromium:531357
Review URL: https://codereview.chromium.org/1340033002
Cr-Commit-Position: refs/heads/master@{#30736}
This models the materialization of arguments objects in the prologue
within the IR graph. It will in turn allow us to optimize access to
these objects and also correctly handle them with inlining.
R=bmeurer@chromium.org,mvstanton@chromium.org
TEST=cctest/test-run-jsobjects/Arguments*
Review URL: https://codereview.chromium.org/1344553003
Cr-Commit-Position: refs/heads/master@{#30735}
The conditionals were added to facilitate a two-sided build GN config update
where the default optimization config was renamed. This is complete so the
transitional code can be removed.
Review URL: https://codereview.chromium.org/1338163003
Cr-Commit-Position: refs/heads/master@{#30733}
Remove incorrect usage of callee-saved FPU regs (f20 and above).
Also remove unnecessary push/pop which were occasionally unpaired,
and caused crash.
TEST=cctest/test-assembler-mips[64]
BUG=
Review URL: https://codereview.chromium.org/1338713004
Cr-Commit-Position: refs/heads/master@{#30729}
Reason for revert:
[Sheriff] Fails mozilla with deadcode:
http://build.chromium.org/p/client.v8/builders/V8%20Linux%20-%20deadcode/builds/5357
Original issue's description:
> [crankshaft] Re-add fast-case for string add left/right.
>
> Now the StringAddStub can optionally convert it's parameters to strings
> (following the rules for the addition operator). This could be further
> optimized once we have a ToPrimitiveStub, but it should be sufficient
> for the moment.
>
> Also removed the unused Strength parameter to the HStringAdd operator,
> because string addition does not depend on language mode.
>
> CQ_INCLUDE_TRYBOTS=tryserver.v8:v8_win_nosnap_shared_rel,v8_linux_nosnap_dbg
> R=mstarzinger@chromium.org
> BUG=v8:4307
> LOG=n
>
> Committed: https://crrev.com/d261849e53fbf8c36efae42d478271f87acff70f
> Cr-Commit-Position: refs/heads/master@{#30726}
TBR=mstarzinger@chromium.org,jarin@chromium.org,bmeurer@chromium.org
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
BUG=v8:4307
Review URL: https://codereview.chromium.org/1347473003
Cr-Commit-Position: refs/heads/master@{#30727}
Now the StringAddStub can optionally convert it's parameters to strings
(following the rules for the addition operator). This could be further
optimized once we have a ToPrimitiveStub, but it should be sufficient
for the moment.
Also removed the unused Strength parameter to the HStringAdd operator,
because string addition does not depend on language mode.
CQ_INCLUDE_TRYBOTS=tryserver.v8:v8_win_nosnap_shared_rel,v8_linux_nosnap_dbg
R=mstarzinger@chromium.org
BUG=v8:4307
LOG=n
Review URL: https://codereview.chromium.org/1339053002
Cr-Commit-Position: refs/heads/master@{#30726}
Filtering out functions does not allow sample processor
to figure out if a sample hit a non-frame region of a function.
The filtering should be done in the frontend.
BUG=chromium:529931
LOG=N
Review URL: https://codereview.chromium.org/1337743002
Cr-Commit-Position: refs/heads/master@{#30723}
Port eadfd66631
Original commit message:
The String constructor was somewhat complex with a lot of micro
optimizations that are not relevant or even misguided. It would be
really hard to port that code to ES6, which requires String to be
subclassable. So as a first step we reduced the necessary complexity
to the bare minimum (also removing the last user of the fairly complex
MacroAssembler::LookupNumberStringCache method).
This also removes the counters for the String constructor, which
were not properly exposed anymore (and not kept in sync with inlined
versions of the String constructor anyway).
R=bmeurer@chromium.org, joransiu@ca.ibm.com, jyan@ca.ibm.com, michael_dawson@ca.ibm.com, dstence@us.ibm.com
BUG=
Review URL: https://codereview.chromium.org/1339143002
Cr-Commit-Position: refs/heads/master@{#30722}
Enhance the perf runner to run with the profiler and print the summary for each d8 run. This automates running the profiler with multiple benchmarks.
BUG=None
LOG=N
TBR=machenbach@chromium.org
Review URL: https://codereview.chromium.org/1327033003
Cr-Commit-Position: refs/heads/master@{#30720}
Some pretty hacky code was used to carry out the tail-call
handler dispatch on ia32 vector stores due to a lack
of free registers. It really tanks performance. A better
approach is to use a virtual register on the isolate.
BUG=
Review URL: https://codereview.chromium.org/1336313002
Cr-Commit-Position: refs/heads/master@{#30718}
This prevents the internal objects-visiting.h to be usable outisde of
the "heap" directory. The static object visitation is only usefull
within the GC and is now properly encapsulated.
R=hpayer@chromium.org,mlippautz@chromium.org
Review URL: https://codereview.chromium.org/1328003002
Cr-Commit-Position: refs/heads/master@{#30716}
This moves scavenging functionality into a separate component so that
neither the scavenger nor objects-visiting need to be exposed outside
the heap.
R=hpayer@chromium.org,mlippautz@chromium.org
Review URL: https://codereview.chromium.org/1323993004
Cr-Commit-Position: refs/heads/master@{#30712}
Adds support for JS calls to the interpreter. In order to support
calls from the interpreter, the PushArgsAndCall builtin is added
which pushes a sequence of arguments onto the stack and calls
builtin::Call.
Adds the Call bytecode.
MIPS port contributed by akos.palfi@imgtec.com in https://codereview.chromium.org/1334873002/
BUG=v8:4280
LOG=N
Review URL: https://codereview.chromium.org/1323463005
Cr-Commit-Position: refs/heads/master@{#30710}
These builtins present an optimization for the general addition case,
where one side is already known to be a string. Unfortunately this
optimization is wrong in the presence of @@toPrimitive (there are some
ideas how to implement a similar optimization using the prototype
backpointer mechanism that jkummerow@ introduced earlier). So this
also removes the broken %_IsStringWrapperSafeForDefaultValueOf, which is
the key part of the optimization mentioned above.
R=mstarzinger@chromium.org
BUG=v8:4307
LOG=n
Review URL: https://codereview.chromium.org/1336273002
Cr-Commit-Position: refs/heads/master@{#30707}
The String constructor was somewhat complex with a lot of micro
optimizations that are not relevant or even misguided. It would be
really hard to port that code to ES6, which requires String to be
subclassable. So as a first step we reduced the necessary complexity
to the bare minimum (also removing the last user of the fairly complex
MacroAssembler::LookupNumberStringCache method).
This also removes the counters for the String constructor, which
were not properly exposed anymore (and not kept in sync with inlined
versions of the String constructor anyway).
R=jarin@chromium.org
Review URL: https://codereview.chromium.org/1335193002
Cr-Commit-Position: refs/heads/master@{#30706}
In certiain cases the ArgumentsIteratorSetter would trigger an invalid
state in the LookupIterator when being overridden. This is now solved
by bypassing the SetDataProperty and directly using
DefinePropertyOrElementIgnoringAttributes since we know exactly which
property we're going to install
LOG=N
BUG=chromium:521484
Review URL: https://codereview.chromium.org/1332873002
Cr-Commit-Position: refs/heads/master@{#30705}
The GN default optimization flag is changing from being specific to the
debug/release status of the build to always being "default_optimization" so
it's easier to override without being conditional on the exact setup in
BUILDCONFIG.gn. See https://codereview.chromium.org/1324623005/
Since V8 is DEPS-ed in, it will need to support both modes to allow a landing.
This patch uses a temporary transitional flag I added to BUILDCONFIG to
indicate which variant should be used. After the patch is landed, we can remove
the new conditions and just remove default_optimization.
Review URL: https://codereview.chromium.org/1333963002
Cr-Commit-Position: refs/heads/master@{#30703}
Correctly save and restore FP registers in cctest/ConvertDToI to
avoid accidental register overwriting by the generated code.
Note: This failure is manifested only in R6 mode.
TEST=test/cctest/interpreter/test-interpreter
BUG=
Review URL: https://codereview.chromium.org/1337023002
Cr-Commit-Position: refs/heads/master@{#30700}
Port 622fa0ea21
Original commit message:
Currently we do this dance between the CallConstructStub, the
CALL_* builtins and the %GetConstructorDelegate, %GetProxyTrap,
and %Apply runtime functions for every [[Construct]] operation on
non-function callables. This is complexity is unnecessary, and can
be simplified to work without any JS builtin. This will also make it
a lot easier to implement ES6 compliant [[Construct]] for proxies.
Also sanitize the invariant for CallConstructStub, which up until now
always restored the context itself, but that force us to always create
another copy of all arguments in case of proxies and other callables,
so we can relax that constraint by making the caller restore the context
(this only affects fullcodegen, since the optimizing compilers already
properly restore the context anyway).
BUG=
Review URL: https://codereview.chromium.org/1340553002
Cr-Commit-Position: refs/heads/master@{#30699}
Port 622fa0ea21
Original commit message:
Currently we do this dance between the CallConstructStub, the
CALL_* builtins and the %GetConstructorDelegate, %GetProxyTrap,
and %Apply runtime functions for every [[Construct]] operation on
non-function callables. This is complexity is unnecessary, and can
be simplified to work without any JS builtin. This will also make it
a lot easier to implement ES6 compliant [[Construct]] for proxies.
Also sanitize the invariant for CallConstructStub, which up until now
always restored the context itself, but that force us to always create
another copy of all arguments in case of proxies and other callables,
so we can relax that constraint by making the caller restore the context
(this only affects fullcodegen, since the optimizing compilers already
properly restore the context anyway).
R=bmeurer@chromium.org, joransiu@ca.ibm.com, jyan@ca.ibm.com, michael_dawson@ca.ibm.com, dstence@us.ibm.com
BUG=
Review URL: https://codereview.chromium.org/1338663003
Cr-Commit-Position: refs/heads/master@{#30696}
Just use a %ThrowStackOverflow runtime function instead, which
does the trick, especially since the Isolate already has a
preallocated StackOverflow error for that.
R=mstarzinger@chromium.org
Review URL: https://codereview.chromium.org/1337883002
Cr-Commit-Position: refs/heads/master@{#30693}
Currently we do this dance between the CallConstructStub, the
CALL_* builtins and the %GetConstructorDelegate, %GetProxyTrap,
and %Apply runtime functions for every [[Construct]] operation on
non-function callables. This is complexity is unnecessary, and can
be simplified to work without any JS builtin. This will also make it
a lot easier to implement ES6 compliant [[Construct]] for proxies.
Also sanitize the invariant for CallConstructStub, which up until now
always restored the context itself, but that force us to always create
another copy of all arguments in case of proxies and other callables,
so we can relax that constraint by making the caller restore the context
(this only affects fullcodegen, since the optimizing compilers already
properly restore the context anyway).
R=mstarzinger@chromium.org
Review URL: https://codereview.chromium.org/1335723002
Cr-Commit-Position: refs/heads/master@{#30691}
Grouping of live ranges that would be beneficial if
allocated on the same register. Currently, that means phi
outputs and inputs.
Review URL: https://codereview.chromium.org/1312473018
Cr-Commit-Position: refs/heads/master@{#30690}
Port 752b0308df
Original commit message:
The refactoring is because it's awkward and error-prone to deterimine which IC slot an
ObjectLiteralProperty uses for feedback. The fix is for each one to know it's own slot. In the
numbering pass, we allocate slots for the ObjectLiteral, then hand out those slots into the
properties.
It adds one word to the ObjectLiteralProperty expression - I'm investigating if thats a
problem.
This changes makes compiling the object literal cleaner across the three compilers. Also, the
slot allocation logic in ObjectLiteral::ComputeFeedbackRequirements() was refactoring to mimic
the style in full-codegen. This is useful since it must remain in sync with
FullCodegen::VisitObjectLiteral().
R=mvstanton@chromium.org, joransiu@ca.ibm.com, jyan@ca.ibm.com, michael_dawson@ca.ibm.com, dstence@us.ibm.com
BUG=
Review URL: https://codereview.chromium.org/1331893003
Cr-Commit-Position: refs/heads/master@{#30689}