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
by escape analysis). Added several tests that expose the bug.
Summary:
LCodegen::AddToTranslation assumes that Lithium environments are
generated by depth-first traversal, but LChunkBuilder::CreateEnvironment
was generating them in breadth-first fashion. This fixes the
CreateEnvironment to traverse the captured objects depth-first.
Note:
It might be worth considering representing LEnvironment by a list
with the same order as the serialized translation representation
rather than having two lists with a subtle relationship between
them (and then serialize in a slightly different order again).
R=titzer@chromium.org, mstarzinger@chromium.org
LOG=N
BUG=
Review URL: https://codereview.chromium.org/93803003
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@18470 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
The current usage of this runtime function is broken as it does not
prevent inlining of the affected function but rather bails out from the
whole unit of compilation after trying to inline affected functions.
This simplifies said runtime function to avoid accidental misuse.
R=titzer@chromium.org
TEST=mjsunit/never-optimize
Review URL: https://codereview.chromium.org/19776006
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@15762 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This allows the deoptimizer to materialize objects (e.g. the arguments
object) while deopting without having a consective stack area holding
the object values. The LEnvironment explicitly tracks locations for
these values and preserves them in the translation.
R=svenpanne@chromium.org
TEST=mjsunit/compiler/inline-arguments
Review URL: https://codereview.chromium.org/16779004
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@15087 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
The predicate CanBeSpilled had a bug, prohibiting the necessary spilling and
correct splitting of live ranges. Removed a redundant assertion immediately done
by the callee anyway.
Thanks to Slava for help with that issue and the entertaining historical
background of the whole story... ;-)
BUG=177883
Review URL: https://codereview.chromium.org/12631012
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@13891 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This allows Crankshaft to completely inline a f.apply() dispatch if the
exact number of arguments is known and the function is constant. The
deoptimizer doesn't generate the f.apply() frame during deoptimization,
so the materialized frames look like f.apply() did a tailcall.
R=jkummerow@chromium.org
TEST=mjsunit/compiler/inline-function-apply
Review URL: https://codereview.chromium.org/12263004
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@13665 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
HCheckPrototypeMaps currently records the prototype and the holder of the
prototype chain (both ends of the chain) and assumes that the chain elements
and their maps did not change in during the entirety of Crankshaft. The actual
traversal of the prototype chain happens in Lithium at code generation.
With parallel compilation, this assumption is not longer correct.
R=mstarzinger@chromium.org
BUG=
Review URL: https://chromiumcodereview.appspot.com/11864013
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@13454 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Previously Crankshaft emitted a generic load for these, now we emit a load of a
named field, guarded by a proto chain check.
LCheckPrototypeMaps now returns the holder, which is for free, because it
already had to check its map as the last step, anyway. This is in sync with what
StubCompiler::CheckPrototype does.
Review URL: https://codereview.chromium.org/11338030
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@12847 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This fixes materialization of arguments objects for strict mode functions during
deoptimization. We materialize arguments from the stack area where optimized
code pushes the arguments when entering the inlined environment. For adapted
invocations we use the arguments adaptor frame for materialization.
R=svenpanne@chromium.org
BUG=v8:2261
TEST=mjsunit/regress/regress-2261,mjsunit/compiler/inline-arguments
Review URL: https://chromiumcodereview.appspot.com/10908194
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@12489 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
xmm0 is not saved across runtime call on x64 because MacroAssembler::EnterExitFrameEpilogue preserves only allocatable XMM registers unlike on ia32 where it preserves all registers.
Cleanup handling of shifts: SHR can deoptimize only when its a shift by 0, all other shift never deoptimize.
Fix type inference for i-to-t change instruction. On X64 this ensures that write-barrier is generated correctly.
R=danno@chromium.org
Review URL: https://chromiumcodereview.appspot.com/10868032
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@12373 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Safe operations are those that either do not observe unsignedness or have special support for uint32 values:
- all binary bitwise operations: they perform ToInt32 on inputs;
- >> and << shifts: they perform ToInt32 on left hand side and ToUint32 on right hand side;
- >>> shift: it performs ToUint32 on both inputs;
- stores to integer external arrays (not pixel, float or double ones): these stores are "bitwise";
- HChange: special support added for conversions of uint32 values to double and tagged values;
- HSimulate: special support added for deoptimization with uint32 values in registers and stack slots;
- HPhi: phis that have only safe uses and only uint32 operands are uint32 themselves.
BUG=v8:2097
TEST=test/mjsunit/compiler/uint32.js
Review URL: https://chromiumcodereview.appspot.com/10778029
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@12367 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Highlights of this CL:
* Introduced a new opcode in the deoptimizer for a setter stub frame.
* Added a global setter stub for returning after deoptimizing a setter.
* We do not need special deopt support for getters, although the getter stub creates an internal frame. The normal machinery works just right for this case, although we generate a stack that can never occur during normal fullcode execution. If this hurts us one day, we can parameterize and reuse the setter deopt machinery.
Review URL: https://chromiumcodereview.appspot.com/10855098
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@12328 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Currently we inline functions with different contexts only on ia32, so we have
to move the helper functions for the various contexts to the top level. Further
more, "new Object()" seems to prevent inlining, too, so we us a simple object
literal.
Although things get consistently inlined now, something strange seems to happen
in test/effect contexts: The DEOPT output seems to contain too few frames, and
we don't get any DEOPT ouput after the first time for those contexts. This has
to be investigated...
TBR=mstarzinger@chromium.org
Review URL: https://chromiumcodereview.appspot.com/10836258
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@12312 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Currently only simple setter calls are handled (i.e. no calls in count
operations or compound assignments), and deoptimization in the setter is not
handled at all. Because of the latter, we temporarily hide this feature behind
the --inline-accessors flag, just like inlining getters.
We now use an enum everywhere we depend on the handling of a return value,
passing around several boolean would be more confusing.
Made VisitReturnStatement and the final parts of TryInline more similar, so
matching them visually is a bit easier now.
Simplified the signature of AddLeaveInlined, the target of the HGoto can simply
be retrieved from the function state.
Review URL: https://chromiumcodereview.appspot.com/10836133
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@12286 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
