Along the way:
- Thread isolate parameter explicitly through code that used to
rely on getting it from the zone.
- Canonicalize the parameter position of isolate and zone for
affected code
- Change Hydrogen New<> instruction templates to automatically
pass isolate
R=mstarzinger@chromium.org
LOG=N
Review URL: https://codereview.chromium.org/868883002
Cr-Commit-Position: refs/heads/master@{#26252}
We go back to patching the code for lazy deoptimization because ICs need the on-stack return address to read/update the IC address/state.
The change also fixes bunch of tests, mostly by adding more deoptimization points.
(We still need to add code to ensure lazy deopt patching does not overwrite ICs and other lazy deopts; this is coming next.)
BUG=
R=bmeurer@chromium.org
Review URL: https://codereview.chromium.org/568783002
git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@23934 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Since the deopt patch address needs to be available during GC to
resolve safepoints, we need to move it to the code object (instead of
the deoptimization input data) - accessing a separate fixed array
is not safe during GC. This CL adds a deoptimization_pc field to
each safepoint. The fields points to the deoptimization block.
The CL also fixes wrong register allocator constraints for
frame states on calls. These should always live on the stack
because registers are not preserved during a call.
BUG=
R=bmeurer@chromium.org
Review URL: https://codereview.chromium.org/504493002
git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@23334 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Centralize a register definition in an IC that provides:
1) symbolic names for the register (like, edx == receiver)
2) defines ordering when passed on the stack
Code that implements or uses the IC should use this definition instead of "knowing" what the registers are. Or at least have the definition to validate it's assumptions.
As a side effect of avoiding runtime static initializers (enforced by tools/check-static-initializers.sh, neat), I gave ownership of the registers array to CodeStubInterfaceDescriptor. This prompted a cleanup of that struct.
R=jkummerow@chromium.org
Review URL: https://codereview.chromium.org/352583002
git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@22011 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
I have fixed skipping of the receiver object to materialize captured
objects. This is done with a new DoTranslateSkip method.
We should consider unifying DoTranslateSkip, DoTranslateObject and
DoTranslateCommand as they do the almost the same thing - they only
differ in where they store the result.
The change also turns bunch of ASSERTs into CHECKs.
R=mstarzinger@chromium.org
BUG=359441
TEST=test/mjsunit/regress/regress-359441.js
LOG=N
Review URL: https://codereview.chromium.org/225283006
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@20978 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This is a purely mechanical change, adding an Isolate* to the CodeStub
constructor and a corresponding field plus a getter. A few methods in
CodeStub and its subclasses can be simplified now, but this is done in
a separate CL.
The underlying reason apart from simplicity is that deep down in the
call chain we need to detect if the serializer is active or not. This
information will be part of the Isolate, not a global variable with
funky synchronization primitives around it (which is fundamentally
wrong and the underlying cause for race conditions and a catch-22
during initialization).
BUG=359977
LOG=y
R=mstarzinger@chromium.org
Review URL: https://codereview.chromium.org/246643014
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@20919 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This is an initial step towards tracking the exact types instead of just
the representations of fields. It adds support to track up to one map of
heap object field values, eliminating various map checks on values
loaded from such fields, at the cost of making stores to such fields
slightly more expensive.
Issues with transitioning stores and fast object literals in Crankshaft
fixed.
TEST=mjsunit/field-type-tracking
R=svenpanne@chromium.org
Review URL: https://codereview.chromium.org/238773002
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@20746 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This is an initial step towards tracking the exact types instead of just the representations of fields. It adds support to track up to one map of heap object field values, eliminating various map checks on values loaded from such fields, at the cost of making stores to such fields slightly more expensive.
TEST=mjsunit/field-type-tracking
R=verwaest@chromium.org
Review URL: https://codereview.chromium.org/167303005
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@20701 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
The escape analysis calculates the number of slots in an object as
no-of-slots = object-size / pointer-size. This gives 3 slots for
heap numbers on 32-bit architectures (one slot for the map, two for
the double value); however, my argument materialization code assumed
just two slots (map + value). Since Hydrogen allocates heap numbers
quite rarely, it is hard to produce a more meaningful repro than the
one provided by Clusterfuzz. Any suggestions are welcome.
The fix is simple - we just read out all extra slots (beyond the map
and the double) for heap numbers.
R=mstarzinger@chromium.org
BUG=351315
LOG=N
Review URL: https://codereview.chromium.org/196283004
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@19874 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This adds a pointer to the shared function info into deoptimization data of an optimized code. Whenever the code is deoptimized, it clears the cache in the shared function info.
This fixes the problem when the optimized function dies in new space GC before the code is deoptimized due to code dependency and before the optimized code cache is cleared in old space GC (see mjsunit/regress/regress-343609.js).
This partially reverts r19603 because we need to be able to evict specific code from the optimized code cache.
BUG=343609
LOG=Y
TEST=mjsunit/regress/regress-343609.js
R=yangguo@chromium.org
Review URL: https://codereview.chromium.org/184923002
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@19635 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Instead of tracking simple absolute offset from the start of the script like other places do, track a pair of (inlining id, offset from the start of inlined function).
This enables us to pinpoint with inlining path an instruction came from. Previously in multi-script environments we emitted positions that made very little sense because inside a single optimized function they would point to different scripts without a way to distinguish them.
Start dumping the source of every inlined function to make possible IR viewing tools with integrated source views as there was previously no way to acquire this information from IR dumps. We also dump source position at which each inlining occured.
Tracked positions are written into hydrogen.cfg as pos:<inlining-id>_<offset>.
Flag --emit-opt-code-positions is renamed by this change into --hydrogen-track-positions to better convey it's meaning.
In addition this change assigned global unique identifier to each optimization performed inside isolate. This allows to precisely match compilation artifacts (e.g. IR and disassembly) and deoptimizations.
BUG=
R=yangguo@chromium.org
Review URL: https://codereview.chromium.org/140683011
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@19360 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
version is passing all the existing test + a bunch of new tests
(packaged in the change list, too).
The patch extends the SlotRef object to describe captured and duplicated
objects. Since the SlotRefs are not independent of each other anymore,
there is a new SlotRefValueBuilder class that stores the SlotRefs and
later materializes the objects from the SlotRefs.
Note that unlike the previous implementation of SlotRefs, we now build
the SlotRef entries for the entire frame, not just the particular
function. This is because duplicate objects might refer to previous
captured objects (that might live inside other inlined function's part
of the frame).
We also need to store the materialized objects between other potential
invocations of the same arguments object so that we materialize each
captured object at most once. The materialized objects of frames live
in the new MaterielizedObjectStore object (contained in Isolate),
indexed by the frame's FP address. Each argument materialization (and
deoptimization) tries to lookup its captured objects in the store before
building new ones. Deoptimization also removes the materialized objects
from the store. We also schedule a lazy deopt to be sure that we always
get rid of the materialized objects and that the optmized function
adopts the materialized objects (instead of happily computing with its
captured representations).
Concerns:
- Is the FP address the right key for a frame? (Note that deoptimizer's
representation of frame is different from the argument object
materializer's one - it is not easy to find common ground.)
- Performance is suboptimal in several places, but a quick local run of
benchmarks does not seem to show a perf hit. Examples of possible
improvements: smarter generation of SlotRefs (build other functions'
SlotRefs only for captured objects and only if necessary), smarter
lookup of stored materialized objects.
- Ideally, we would like to share the code for argument materialization
with deoptimizer's materializer. However, the supporting data structures
(mainly the frame descriptor) are quite different in each case, so it
looks more like a separate project.
Thanks for any feedback.
R=danno@chromium.org, mstarzinger@chromium.org
LOG=N
BUG=
Committed: https://code.google.com/p/v8/source/detail?r=18918
Review URL: https://codereview.chromium.org/103243005
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@18936 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
mostly to make sure that it is going in the right direction. The current
version is passing all the existing test + a bunch of new tests
(packaged in the change list, too).
The patch extends the SlotRef object to describe captured and duplicated
objects. Since the SlotRefs are not independent of each other anymore,
there is a new SlotRefValueBuilder class that stores the SlotRefs and
later materializes the objects from the SlotRefs.
Note that unlike the previous implementation of SlotRefs, we now build
the SlotRef entries for the entire frame, not just the particular
function. This is because duplicate objects might refer to previous
captured objects (that might live inside other inlined function's part
of the frame).
We also need to store the materialized objects between other potential
invocations of the same arguments object so that we materialize each
captured object at most once. The materialized objects of frames live
in the new MaterielizedObjectStore object (contained in Isolate),
indexed by the frame's FP address. Each argument materialization (and
deoptimization) tries to lookup its captured objects in the store before
building new ones. Deoptimization also removes the materialized objects
from the store. We also schedule a lazy deopt to be sure that we always
get rid of the materialized objects and that the optmized function
adopts the materialized objects (instead of happily computing with its
captured representations).
Concerns:
- Is there a simpler/more correct way to store the already-materialized
objects? (At the moment there is a custom root reference to JSArray
containing frames' FixedArrays with their captured objects.)
- Is the FP address the right key for a frame? (Note that deoptimizer's
representation of frame is different from the argument object
materializer's one - it is not easy to find common ground.)
- Performance is suboptimal in several places, but a quick local run of
benchmarks does not seem to show a perf hit. Examples of possible
improvements: smarter generation of SlotRefs (build other functions'
SlotRefs only for captured objects and only if necessary), smarter
lookup of stored materialized objects.
- Ideally, we would like to share the code for argument materialization
with deoptimizer's materializer. However, the supporting data structures
(mainly the frame descriptor) are quite different in each case, so it
looks more like a separate project.
Thanks for any feedback.
R=mstarzinger@chromium.org, danno@chromium.org
LOG=N
BUG=
Review URL: https://codereview.chromium.org/103243005
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@18918 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Made operator* return reference to the raw type, not pointer. New method 'get()' should be used when raw pointer is needed.
Also removed useless inline modifier from the SmaprtPointer methods and added const modifier to the methods that don't change smart pointer.
Made ~SmartPointerBase protected to avoid accidental calls of the non-virtual base class's destructor.
drive-by: fixed use after free in src/factory.cc
BUG=None
LOG=N
R=alph@chromium.org, svenpanne@chromium.org
Review URL: https://codereview.chromium.org/101763003
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@18275 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This is controlled by two flags:
--redirect_code_traces
--redirect_code_traces_to=<filename>
When redirection is enabled but --redirect_code_traces_to is not specified traces are written to a file code-<pid>-<isolate>.asm. This mangling scheme matches hydrogen.cfg and allows easy discovery of compilation artifacts in a multi-V8 environment (e.g. when compilation is traced from inside Chromium).
D8 defines --redirect_code_traces_to=code.asm similar to hydrogen.cfg redirection.
BUG=
R=danno@chromium.org
Review URL: https://codereview.chromium.org/43273004
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@17571 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Original descriptions were:
- "Refactor and cleanup VirtualMemory."
- "Fix typo."
- "Deuglify V8_INLINE and V8_NOINLINE."
- "Don't align size on allocation granularity for unaligned ReserveRegion calls."
Reasons for the revert are:
- Our mjsunit test suite slower by a factor of 5(!) in release mode.
- Flaky cctest/test-alloc/CodeRange on all architectures and platforms.
- Tankage of Sunspider by about 6% overall (unverified).
TBR=bmeurer@chromium.org
Review URL: https://codereview.chromium.org/23970004
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@16662 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Remove a lot of platform duplication, and simplify the virtual
memory implementation. Also improve readability by avoiding bool
parameters for executability (use a dedicated Executability type
instead).
Get rid of the Isolate::UncheckedCurrent() call in the platform
code, as part of the Isolate TLS cleanup.
Use a dedicated random number generator for the address
randomization, instead of messing with the per-isolate random
number generators.
TEST=cctest/test-virtual-memory
R=verwaest@chromium.org
Review URL: https://codereview.chromium.org/23641009
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@16637 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
These classes are meant to replace OS::Ticks() and OS::TimeCurrentMillis(),
which are broken in several ways. The ElapsedTimer class implements a
stopwatch using TimeTicks::HighResNow() for high resolution, monotonic
timing.
Also fix the CpuProfile::GetStartTime() and CpuProfile::GetEndTime()
methods to actually return the time relative to the unix epoch as stated
in the documentation (previously that was relative to some arbitrary
point in time, i.e. boot time).
The previous Windows issues have been resolved, and we now use GetTickCount64()
on Windows Vista and later, falling back to timeGetTime() with rollover
protection for earlier Windows versions.
BUG=v8:2853
R=machenbach@chromium.org, yurys@chromium.org
Review URL: https://codereview.chromium.org/23490015
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@16413 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
These classes are meant to replace OS::Ticks() and OS::TimeCurrentMillis(),
which are broken in several ways. The ElapsedTimer class implements a
stopwatch using TimeTicks::HighResNow() for high resolution, monotonic
timing.
Also fix the CpuProfile::GetStartTime() and CpuProfile::GetEndTime()
methods to actually return the time relative to the unix epoch as stated
in the documentation (previously that was relative to some arbitrary
point in time, i.e. boot time).
BUG=v8:2853
R=machenbach@chromium.org
Review URL: https://codereview.chromium.org/23469013
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@16398 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
These classes are meant to replace OS::Ticks() and OS::TimeCurrentMillis(),
which are broken in several ways. The ElapsedTimer class implements a
stopwatch using TimeTicks::HighResNow() for high resolution, monotonic
timing.
Also fix the CpuProfile::GetStartTime() and CpuProfile::GetEndTime()
methods to actually return the time relative to the unix epoch as stated
in the documentation (previously that was relative to some arbitrary
point in time, i.e. boot time).
BUG=v8:2853
R=machenbach@chromium.org, yurys@chromium.org
Review URL: https://codereview.chromium.org/23295034
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@16388 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This change implements a simple data-flow analysis pass over captured
objects to the existing escape analysis. It tracks the state of values
in the Hydrogen graph through CapturedObject marker instructions that
are used to construct an appropriate translation for the deoptimizer to
be able to materialize these objects again.
This can be considered a combination of scalar replacement of loads and
stores on captured objects and sinking of unused allocations.
R=titzer@chromium.org
TEST=mjsunit/compiler/escape-analysis
Review URL: https://codereview.chromium.org/21055011
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@16098 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
