This adds the basics necessary to support float32 operations in TurboFan.
The actual functionality required to detect safe float32 operations will
be added based on this later. Therefore this does not affect production
code except for some cleanup/refactoring.
In detail, this patchset contains the following features:
- Add support for float32 operations to arm, arm64, ia32 and x64
backends.
- Add float32 machine operators.
- Add support for float32 constants to simplified lowering.
- Handle float32 representation for phis in simplified lowering.
In addition, contains the following (related) cleanups:
- Fix/unify naming of backend instructions.
- Use AVX comparisons when available.
- Extend ArchOpcodeField to 9 bits (required for arm64).
- Refactor some code duplication in instruction selectors.
BUG=v8:3589
LOG=n
R=dcarney@chromium.org
Review URL: https://codereview.chromium.org/1044793002
Cr-Commit-Position: refs/heads/master@{#27509}
No compiler support for now (BTW: %_NewString, doesn't have that,
either), inline allocation will come later. Hopefully the last
intrisic to add for a StringAddStub POC...
Review URL: https://codereview.chromium.org/1041723002
Cr-Commit-Position: refs/heads/master@{#27508}
And reland "Use a slot that is located on a heap page when removing
invalid entries from the SlotsBuffer."
This reverts commits de018fbda3 and
d23a9f7a3e.
Reason for relanding: looking fine on Canary, let's get these fixes back in.
BUG=chromium:454297,chromium:470801
LOG=y
TBR=ishell@chromium.org
Review URL: https://codereview.chromium.org/1043703003
Cr-Commit-Position: refs/heads/master@{#27507}
The current GC idle time handling heuristics are getting too complicated. Moreover, with longer idle time we are getting more full garbage collections. This CL shrinks the idle round window and reduces complexity in the case where we cause a full garbage collection.
BUG=chromium:468554
LOG=n
Review URL: https://codereview.chromium.org/1024043003
Cr-Commit-Position: refs/heads/master@{#27493}
Reason for revert:
Reverting risky GC changes that block v8 roll.
Original issue's description:
> Reland "Filter invalid slots out from the SlotsBuffer after marking."
>
> > There are two reasons that could cause invalid slots appearance in SlotsBuffer:
> > 1) If GC trims "tail" of an array for which it has already recorded a slots and then migrate another object to the "tail".
> > 2) Tagged slot could become a double slot after migrating of an object to another map with "shifted" fields (for example as a result of generalizing immutable data property to a data field).
>
> > This CL also adds useful machinery that helps triggering incremental write barriers.
>
> > BUG=chromium:454297
> > LOG=Y
>
> NOTRY=true
>
> Committed: https://crrev.com/f86aadd1d45c756467dff8e08a055b462d7a060b
> Cr-Commit-Position: refs/heads/master@{#27433}
TBR=machenbach@chromium.org,ulan@chromium.org
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
Review URL: https://codereview.chromium.org/1041593002
Cr-Commit-Position: refs/heads/master@{#27491}
Before this patch the embedder could assign timestamp to the last interval after calling GetHeapStats. This would be slightly different from the timstamps assigned by v8 internally and written into heap snapshot. This patch allow to avoid this small discrepancy by returning timestamp along with last heap stats update.
BUG=chromium:467222
LOG=Y
Review URL: https://codereview.chromium.org/1037803002
Cr-Commit-Position: refs/heads/master@{#27466}
This adapts the debugger so that the first break event starting the
stepping process can come from optimized code. TurboFan supports a
debugger statement and hence can be the top-most frame whenever the
Debug::HandleDebugBreak handler is triggered.
R=yangguo@chromium.org
TEST=mjsunit/debug,cctest/test-debug
Review URL: https://codereview.chromium.org/1038613002
Cr-Commit-Position: refs/heads/master@{#27447}
This switches full-codegen to no longer push and pop StackHandler
markers onto the operand stack, but relies on a range-based handler
table instead. We only use StackHandlers in JSEntryStubs to mark the
transition from C to JS code.
Note that this makes deoptimization and OSR from within any try-block
work out of the box, makes the non-exception paths faster and should
overall be neutral on the memory footprint (pros).
On the other hand it makes the exception paths slower and actually
throwing and exception more expensive (cons).
R=yangguo@chromium.org
TEST=cctest/test-run-jsexceptions/DeoptTry
Review URL: https://codereview.chromium.org/1010883002
Cr-Commit-Position: refs/heads/master@{#27440}
> There are two reasons that could cause invalid slots appearance in SlotsBuffer:
> 1) If GC trims "tail" of an array for which it has already recorded a slots and then migrate another object to the "tail".
> 2) Tagged slot could become a double slot after migrating of an object to another map with "shifted" fields (for example as a result of generalizing immutable data property to a data field).
> This CL also adds useful machinery that helps triggering incremental write barriers.
> BUG=chromium:454297
> LOG=Y
NOTRY=true
Review URL: https://codereview.chromium.org/1032833002
Cr-Commit-Position: refs/heads/master@{#27433}
Reason for revert:
Need to revert in order to revert https://codereview.chromium.org/1029323003/
Original issue's description:
> Filter invalid slots out from the SlotsBuffer after marking.
>
> There are two reasons that could cause invalid slots appearance in SlotsBuffer:
> 1) If GC trims "tail" of an array for which it has already recorded a slots and then migrate another object to the "tail".
> 2) Tagged slot could become a double slot after migrating of an object to another map with "shifted" fields (for example as a result of generalizing immutable data property to a data field).
>
> This CL also adds useful machinery that helps triggering incremental write barriers.
>
> BUG=chromium:454297
> LOG=Y
>
> Committed: https://crrev.com/5c47c1c0d3e4a488f190c16a64ee02f5a14e6561
> Cr-Commit-Position: refs/heads/master@{#27423}
TBR=hpayer@chromium.org,erik.corry@gmail.com,ishell@chromium.org
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
BUG=chromium:454297
Review URL: https://codereview.chromium.org/1033453005
Cr-Commit-Position: refs/heads/master@{#27426}
There are two reasons that could cause invalid slots appearance in SlotsBuffer:
1) If GC trims "tail" of an array for which it has already recorded a slots and then migrate another object to the "tail".
2) Tagged slot could become a double slot after migrating of an object to another map with "shifted" fields (for example as a result of generalizing immutable data property to a data field).
This CL also adds useful machinery that helps triggering incremental write barriers.
BUG=chromium:454297
LOG=Y
Review URL: https://codereview.chromium.org/1010363005
Cr-Commit-Position: refs/heads/master@{#27423}
Gather references to unbound variables where the reference (VariableProxy) is
inside strong mode. Check them against the global object when a script is bound
to a context (during compilation).
This CL only checks unbound variables which are not inside lazy functions - TBD
how do we solve that; alternatives: add developer mode which disables laziness /
do the check whenever lazy functions are really compiled.
BUG=v8:3956
LOG=N
Review URL: https://codereview.chromium.org/1005063002
Cr-Commit-Position: refs/heads/master@{#27422}
These are needed (among other things) for a TurboFan-generated
StringAddStub. Furthermore, they can be used to nuke the overly
complex %_IsInstanceType intrisic, it's completely expressible in
JavaScript now, but that will be done in a separate CL.
Alpha-sorted things a bit on the way to ease navigation.
Review URL: https://codereview.chromium.org/1010973010
Cr-Commit-Position: refs/heads/master@{#27415}
This removes the CompilationInfoWithZone class from the header file
because it is more than a pure convenience class and shouldn't be used
outside of the compiler at all.
R=titzer@chromium.org
Review URL: https://codereview.chromium.org/1000353004
Cr-Commit-Position: refs/heads/master@{#27411}
it is the last patch of https://codereview.chromium.org/1012633002
All that we need here is to push the collected info to the profiler
and convert it into actionable information about deopt.
On the Next: get the info accessible by embedder.
BUG=chromium:452067
LOG=n
TEST=DeoptAtFirstLevelInlinedSource, DeoptAtSecondLevelInlinedSource, DeoptUntrackedFunction
Review URL: https://codereview.chromium.org/1013143003
Cr-Commit-Position: refs/heads/master@{#27403}
This removes the stub-based constructor from CompilationInfoWithZone
as this class is more than a pure convenience class and only by chance
doesn't have an effect in the destructor.
R=titzer@chromium.org
Review URL: https://codereview.chromium.org/1026513004
Cr-Commit-Position: refs/heads/master@{#27376}
With this change, we remember the types of frame state inputs (in a new
operator, called TypedStateValues). Instead of inferring the value types
when building translations, we used the recorded types.
The original approach was not reliable because the passes after
simplified lowering can change node types, and this in turn confuses
the translation builder.
BUG=chromium:468727
LOG=n
R=bmeurer@chromium.org
Review URL: https://codereview.chromium.org/1015423002
Cr-Commit-Position: refs/heads/master@{#27310}
This reduces the overhead of typed lowering, because we lower
JSToBoolean/JSUnaryNot directly if possible, instead of first lowering
to AnyToBoolean, and then letting the SimplifiedOperatorReducer do the
further lowering.
Also remove some obsolete tests from the cctest suite that have since
been removed by proper unittests. And improve unitttest coverage for the
typed lowering cases.
R=mstarzinger@chromium.org
Review URL: https://codereview.chromium.org/999173003
Cr-Commit-Position: refs/heads/master@{#27295}
Updated excludes to remove SKIPs where possible. Either
removing if no longer needed or replacing with SLOW where
tests just need more time to run
modified: test/cctest/cctest.status
modified: test/mjsunit/mjsunit.status
R=danno@chromium.org, svenpanne@chromium.org
BUG=
Review URL: https://codereview.chromium.org/1017073003
Cr-Commit-Position: refs/heads/master@{#27287}
We were able to remove most of our changes needed to compile
on AIX with an earlier compiler level. These changes are the
remaining ones.
The changes in heap/heap.cc are needed because otherwise the
compiler complains that result is potentially used before
it is initialized.
The changes in heap/mark-compact.cc are required because
AIX supports the full 64 bit address range so the check
being guarded is invalid.
The changes in build/toolchain.gypi and
test/cctest/cctest/gyp are aix only and are adjust the
compile/link options to allow the AIX build to succeed.
modified: build/toolchain.gypi
modified: src/heap/heap.cc
modified: src/heap/mark-compact.cc
modified: test/cctest/cctest.gyp
R=danno@chromium.org, svenpanne@chromium.org
BUG=
Review URL: https://codereview.chromium.org/1013833002
Cr-Commit-Position: refs/heads/master@{#27255}
In constructing the transfer between loop copies, we need to merge the backedges from all the previous copies of the given loop. The control reduction will work out which ones are really reachable.
R=mstarzinger@chromium.org
BUG=
Review URL: https://codereview.chromium.org/1004993004
Cr-Commit-Position: refs/heads/master@{#27246}
The cause was dynamic allocation of an accounting structure used to
create/initialize the type feedback vector, done at the end of the
numbering pass. The solution is to Zone-allocate the structure to
bring it's lifetime in line with the compilation unit.
BUG=
Review URL: https://codereview.chromium.org/1014793003
Cr-Commit-Position: refs/heads/master@{#27241}
A hydrogen code stub is not the best approach because it builds a frame
and doesn't have the technology to discard roots at tail call exits.
Platform-specific stubs provide much better performance at this point.
R=verwaest@chromium.org
BUG=
Review URL: https://codereview.chromium.org/988653003
Cr-Commit-Position: refs/heads/master@{#27235}
The per-isolate compilation cache is a lot faster still than
the serialized code cache. Promote code to compilation cache
after deserialization.
R=vogelheim@chromium.org
BUG=chromium:399580
LOG=N
Review URL: https://codereview.chromium.org/1008363002
Cr-Commit-Position: refs/heads/master@{#27220}
The CL enables the same instructions are selected for Word32 and Word64 compare
operations which is possible due to a fact 32-bit inputs and produced values
are always sign-extended.
TEST=
BUG=
Review URL: https://codereview.chromium.org/1005123002
Cr-Commit-Position: refs/heads/master@{#27212}
Reason for revert:
Increased rate of Chrome crashes. Requires further investigation.
Original issue's description:
> Reland of Remove slots that point to unboxed doubles from the StoreBuffer/SlotsBuffer.
>
> The problem is that tagged slot could become a double slot after migrating of an object to another map with "shifted" fields (for example as a result of generalizing immutable data property to a data field).
> This CL also adds useful machinery that helps triggering incremental write barriers.
>
> BUG=chromium:454297, chromium:465273
> LOG=Y
>
> Committed: https://crrev.com/6d0677d845c47ab9fa297de61d0e3d8e5480a02a
> Cr-Commit-Position: refs/heads/master@{#27141}
TBR=hpayer@chromium.org
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
BUG=chromium:454297, chromium:465273
Review URL: https://codereview.chromium.org/1004623003
Cr-Commit-Position: refs/heads/master@{#27207}
Basically recognize certain x < y ? x : y constructs and turn that into
Float64Min/Float64Max operations, if the target machine supports that.
On x86 we lower to (v)minsd/(v)maxsd.
R=dcarney@chromium.org
Review URL: https://codereview.chromium.org/998283002
Cr-Commit-Position: refs/heads/master@{#27160}
Port 84e90b2d0d
Original commit message:
Four tests are failing due to a problem with no frame ranges.
TEST=cctest/test-cpu-profiler/CollectDeoptEvents
BUG=
Review URL: https://codereview.chromium.org/1000723002
Cr-Commit-Position: refs/heads/master@{#27152}
We can reduce the number of type feedback vector slots required, and
also reduce IC misses.
Review URL: https://codereview.chromium.org/1001533002
Cr-Commit-Position: refs/heads/master@{#27149}
A diamond is unused if the Merge node has no Phi/EffectPhi uses, exactly
two inputs, one IfTrue and one IfFalse, which have the same Branch
control input and no other uses except for the Merge. In this case the
diamond can safely be removed.
R=jarin@chromium.org
Review URL: https://codereview.chromium.org/1000883003
Cr-Commit-Position: refs/heads/master@{#27148}
The test/cctest/cctest.status and test/mjsunit/mjsunit.status
tests now pass under the simulator so removing these
excludes
modified: test/cctest/cctest.status
modified: test/mjsunit/mjsunit.status
R=danno@chromium.org, svenpanne@chromium.org
BUG=
Review URL: https://codereview.chromium.org/995393003
Cr-Commit-Position: refs/heads/master@{#27145}
Since recursive modules are gone, only the top-level scope can have
module inner scopes. Rename Scope::AllocateModulesRecursively to
Scope::AllocateModules, and add test showing the module Variables
are still allocated appropriately in the top level scope.
BUG=v8:1569,v8:3940
LOG=n
Review URL: https://codereview.chromium.org/999893003
Cr-Commit-Position: refs/heads/master@{#27143}
The problem is that tagged slot could become a double slot after migrating of an object to another map with "shifted" fields (for example as a result of generalizing immutable data property to a data field).
This CL also adds useful machinery that helps triggering incremental write barriers.
BUG=chromium:454297, chromium:465273
LOG=Y
Review URL: https://codereview.chromium.org/988363002
Cr-Commit-Position: refs/heads/master@{#27141}
Contribution of PowerPC port (continuation of 422063005,
817143002,866843003, and 901083004). This patch updates
the ppc directories to make them current with changes in
common code as of today.
We had to exclude the test test-serialize/SerializeInternalReference
until we agree on the right way to add those changes for PPC as
outlined in the description in the google doc provided earlier.
We also had to exclude a couple of other tests due to new
failures seen since the last uplevel. We excluded as opposed
to waiting until we could investigate to maximize the chance
of getting PPC compiling in the google repos before new
breaking changes are made.
I'll note that before applying any of our changes the
mozilla part of quickcheck was already broken when using
the lastest repo content so I had to run without that
modified: src/compiler/ppc/code-generator-ppc.cc
modified: src/compiler/ppc/instruction-codes-ppc.h
modified: src/compiler/ppc/instruction-selector-ppc.cc
modified: src/ic/ppc/handler-compiler-ppc.cc
modified: src/ic/ppc/ic-compiler-ppc.cc
modified: src/ppc/assembler-ppc-inl.h
modified: src/ppc/assembler-ppc.cc
modified: src/ppc/assembler-ppc.h
modified: src/ppc/builtins-ppc.cc
modified: src/ppc/code-stubs-ppc.cc
modified: src/ppc/codegen-ppc.cc
modified: src/ppc/full-codegen-ppc.cc
modified: src/ppc/lithium-codegen-ppc.cc
modified: src/ppc/macro-assembler-ppc.cc
modified: src/ppc/macro-assembler-ppc.h
modified: test/cctest/cctest.status
R=danno@chromium.org, svenpanne@chromium.org
BUG=
Review URL: https://codereview.chromium.org/994533004
Cr-Commit-Position: refs/heads/master@{#27125}
None of these fields is used in Blink. Embedder always can implement them using existing API.
BUG=chromium:465651
LOG=Y
Review URL: https://codereview.chromium.org/983833006
Cr-Commit-Position: refs/heads/master@{#27113}
We use slightly different schema for JumpTable on arm64 than for x64.
We do a branch (B) to the JumpTable from the code,
then a branch (B) to the end of jump table code
and then branch to the deoptimizer code with putting
the return address into lr register (Call which is actually Blr).
As a result the 'from' address in Deoptimizer always points to
the end of JumpTable code and we can get nothing from this information.
0) I moved save_doubles and needs_frame code out of for_loop.
1) I replaced B commands with Bl so we put different return addresses
to lr register for the different jump table entries and replaced
the final Call with Br which do not touch lr register.
Also I removed the last_entry check so we will always do the Bl
even for the last entry because we need the right address in lr.
I don't think that this will affect the performance because it
just one more branch for entire deopt mechanics.
BUG=chromium:452067
LOG=n
Review URL: https://codereview.chromium.org/984893003
Cr-Commit-Position: refs/heads/master@{#27094}
Provide an intrinsic %MathFloor / %_MathFloor that is used to optimize
both Math.ceil and Math.floor, and use the JS inlining mechanism to
inline Math.ceil into TurboFan code. Although we need to touch code
outside of TurboFan to make this work, this does not affect the way we
handle Math.ceil and/or Math.floor in CrankShaft, because for CrankShaft
the old-style builtin function id based inlining still kicks in first.
Once this solution is stabilized, we can use it for Math.floor as well.
And once that is settled, we can establish it as the unified way to
inline builtins, and get rid of the specialized builtin function id
based inlining at some point.
Note that "builtin" applies to basically every piece of internal
JavaScript/intrinsics based code, so this also applies to the yet to be
defined JavaScript based code stubs and handlers.
BUG=v8:3953
LOG=n
R=yangguo@chromium.org,svenpanne@chromium.org
Review URL: https://codereview.chromium.org/990963003
Cr-Commit-Position: refs/heads/master@{#27086}
Context specialization enables inlining (at least currently it is the
only enabler for inlining), but inlining enables more possibilities for
context specialization. So we really need to run them together.
This is especially important with the "module based builtins" that we're
working towards.
BUG=v8:3952
LOG=n
Review URL: https://codereview.chromium.org/988423004
Cr-Commit-Position: refs/heads/master@{#27085}
since the old style weakness is slated for removal, we might as well reuse the name to limit confusion. additionally I simplified the callback type to a enum to either get internal field values or not
this should be a non-breaking change with the exception of PhantomPersistentValueMap, which is unused.
R=jochen@chromium.org, erikcorry@chromium.org
BUG=
Review URL: https://codereview.chromium.org/989153003
Cr-Commit-Position: refs/heads/master@{#27084}
Rationale: separate the inputs and outputs of parsing + analysis from the business of compiling (i.e. generating machine code).
BUG=
Review URL: https://codereview.chromium.org/974213002
Cr-Commit-Position: refs/heads/master@{#27078}
The original code always returned the first entry from RelocInfo that matched with
bailout_id. But we may have a few different deopt reasons for one bailout_id.
So we need to get the one which matches with a particular call from JumpTable.
We can do this by checking not 'target_address' (it maps to bailout_id)
but 'from' address which maps to a particular JumpTable entry.
The test was reworked so it tests identical functions against different reasons.
BUG=chromium:452067
LOG=n
Review URL: https://codereview.chromium.org/984773003
Cr-Commit-Position: refs/heads/master@{#27076}
Reason for revert:
It caused a lot of Canary crashes.
Original issue's description:
> Remove slots that point to unboxed doubles from the StoreBuffer/SlotsBuffer.
>
> The problem is that tagged slot could become a double slot after migrating of an object to another map with "shifted" fields (for example as a result of generalizing immutable data property to a data field).
> This CL also adds useful machinery that helps triggering incremental write barriers.
>
> BUG=chromium:454297
> LOG=Y
>
> Committed: https://crrev.com/9633ebabd405c264d33f603f8798c31f59418dcd
> Cr-Commit-Position: refs/heads/master@{#27054}
TBR=verwaest@chromium.org,hpayer@chromium.org
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
BUG=chromium:454297
Review URL: https://codereview.chromium.org/991793002
Cr-Commit-Position: refs/heads/master@{#27063}
We mark certain builtins for inlining, and those should always be
inlined into optimized code (CrankShaft already handles it this way), so
we should support that in TurboFan as well. Currently this mainly
affects a certain set of Math functions, but once have the basics in
place we can extend this to any kind of builtin/code stub/accessor.
This adds a new flag --turbo_builtin_inlining (enabled by default), that
forces the inliner to always inline builtins marked for inlining, but
does not affect inlining of other functions (this is still controlled by
the --turbo-inlining flag).
BUG=v8:3952
LOG=n
R=jarin@chromium.org
Review URL: https://codereview.chromium.org/993473002
Cr-Commit-Position: refs/heads/master@{#27059}
