This reverts commit 85ba94f28c.
All parallelism can be turned off using --predictable, or --noparallel-compaction.
This patch completely parallelizes
- semispace copy: from space -> to space (within newspace)
- newspace evacuation: newspace -> oldspace
- oldspace compaction: oldspace -> oldspace
Previously newspace has been handled sequentially (semispace copy, newspace
evacuation) before compacting oldspace in parallel. However, on a high level
there are no dependencies between those two actions, hence we parallelize them
altogether. We base the number of evacuation tasks on the overall set of
to-be-processed pages (newspace + oldspace compaction pages).
Some low-level details:
- The hard cap on number of tasks has been lifted
- We cache store buffer entries locally before merging them back into the global
StoreBuffer in a finalization phase.
- We cache AllocationSite operations locally before merging them back into the
global pretenuring storage in a finalization phase.
- AllocationSite might be compacted while they would be needed for newspace
evacuation. To mitigate any problems we defer checking allocation sites for
newspace till merging locally buffered data.
CQ_EXTRA_TRYBOTS=tryserver.v8:v8_linux_arm64_gc_stress_dbg,v8_linux_gc_stress_dbg,v8_mac_gc_stress_dbg,v8_linux64_asan_rel,v8_linux64_tsan_rel,v8_mac64_asan_rel
BUG=chromium:524425
LOG=N
R=hpayer@chromium.org, ulan@chromium.org
Review URL: https://codereview.chromium.org/1640563004
Cr-Commit-Position: refs/heads/master@{#33552}
(Fixed a bug found by nosnap builds.)
We get less "pollution" of type feedback if we have one vector per native
context, rather than one for the whole system. This CL moves the vector
appropriately.
We rely more heavily on the Optimized Code Map in the SharedFunctionInfo. The
vector actually lives in the first slot of the literals array (indeed there is
great commonality between those arrays, they can be thought of as the same
thing). So we make greater effort to ensure there is a valid literals array
after compilation.
This meant, for performance reasons, that we needed to extend
FastNewClosureStub to support creating closures with literals. And ultimately,
it drove us to move the optimized code map lookup out of FastNewClosureStub
and into the compile lazy builtin.
The heap change is trivial so I TBR Hannes for it...
TBR=hpayer@chromium.org
BUG=
Review URL: https://codereview.chromium.org/1642613002
Cr-Commit-Position: refs/heads/master@{#33548}
This ensures that the BytecodeGraphBuilder can generate correct graphs
even when deoptimization has not been enabled. This configuration is not
enabled in production, and we might eventually decide to deprecate it
for good. Until then, this is a quick fix.
R=jarin@chromium.org
TEST=cctest/test-pipeline
Review URL: https://codereview.chromium.org/1640683002
Cr-Commit-Position: refs/heads/master@{#33545}
Introduces the concept of transfer direction to register operands. This
enables the register translator to emit exactly the moves that a
bytecode having it's register operands translated needs.
BUG=v8:4280,v8:4675
LOG=N
Review URL: https://codereview.chromium.org/1633153002
Cr-Commit-Position: refs/heads/master@{#33544}
Reason for revert:
[Sheriff] Leads to crashes on all webrtc chromium testers, e.g.:
https://build.chromium.org/p/chromium.webrtc/builders/Mac%20Tester/builds/49664
Original issue's description:
> [heap] Parallel newspace evacuation, semispace copy, and compaction \o/
>
> All parallelism can be turned off using --predictable, or --noparallel-compaction.
>
> This patch completely parallelizes
> - semispace copy: from space -> to space (within newspace)
> - newspace evacuation: newspace -> oldspace
> - oldspace compaction: oldspace -> oldspace
>
> Previously newspace has been handled sequentially (semispace copy, newspace
> evacuation) before compacting oldspace in parallel. However, on a high level
> there are no dependencies between those two actions, hence we parallelize them
> altogether. We base the number of evacuation tasks on the overall set of
> to-be-processed pages (newspace + oldspace compaction pages).
>
> Some low-level details:
> - The hard cap on number of tasks has been lifted
> - We cache store buffer entries locally before merging them back into the global
> StoreBuffer in a finalization phase.
> - We cache AllocationSite operations locally before merging them back into the
> global pretenuring storage in a finalization phase.
> - AllocationSite might be compacted while they would be needed for newspace
> evacuation. To mitigate any problems we defer checking allocation sites for
> newspace till merging locally buffered data.
>
> CQ_EXTRA_TRYBOTS=tryserver.v8:v8_linux_arm64_gc_stress_dbg,v8_linux_gc_stress_dbg,v8_mac_gc_stress_dbg,v8_linux64_asan_rel,v8_linux64_tsan_rel,v8_mac64_asan_rel
> BUG=chromium:524425
> LOG=N
> R=hpayer@chromium.org, ulan@chromium.org
>
> Committed: https://crrev.com/8f0fd8c0370ae8c5aab56491b879d7e30c329062
> Cr-Commit-Position: refs/heads/master@{#33523}
TBR=hpayer@chromium.org,ulan@chromium.org,mlippautz@chromium.org
# Skipping CQ checks because original CL landed less than 1 days ago.
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
BUG=chromium:524425
Review URL: https://codereview.chromium.org/1643473002
Cr-Commit-Position: refs/heads/master@{#33539}
In a generator function, the parser rewrites a return statement into a "final"
yield. A final yield used to close the generator, which was incorrect because
the return may occur inside a try-finally clause and so the generator may not
yet terminate.
BUG=
Review URL: https://codereview.chromium.org/1634553002
Cr-Commit-Position: refs/heads/master@{#33537}
They were already treated as a BindingPattern error; this patch simply
replaces that call with one marking them as both a binding and assignment
error, and adds parsing tests for both cases.
BUG=v8:4707
LOG=n
Review URL: https://codereview.chromium.org/1632303002
Cr-Commit-Position: refs/heads/master@{#33528}
It allows embedder to inject a stack sample on demand.
BUG=chromium:579191
LOG=N
Review URL: https://codereview.chromium.org/1631043002
Cr-Commit-Position: refs/heads/master@{#33527}
All parallelism can be turned off using --predictable, or --noparallel-compaction.
This patch completely parallelizes
- semispace copy: from space -> to space (within newspace)
- newspace evacuation: newspace -> oldspace
- oldspace compaction: oldspace -> oldspace
Previously newspace has been handled sequentially (semispace copy, newspace
evacuation) before compacting oldspace in parallel. However, on a high level
there are no dependencies between those two actions, hence we parallelize them
altogether. We base the number of evacuation tasks on the overall set of
to-be-processed pages (newspace + oldspace compaction pages).
Some low-level details:
- The hard cap on number of tasks has been lifted
- We cache store buffer entries locally before merging them back into the global
StoreBuffer in a finalization phase.
- We cache AllocationSite operations locally before merging them back into the
global pretenuring storage in a finalization phase.
- AllocationSite might be compacted while they would be needed for newspace
evacuation. To mitigate any problems we defer checking allocation sites for
newspace till merging locally buffered data.
CQ_EXTRA_TRYBOTS=tryserver.v8:v8_linux_arm64_gc_stress_dbg,v8_linux_gc_stress_dbg,v8_mac_gc_stress_dbg,v8_linux64_asan_rel,v8_linux64_tsan_rel,v8_mac64_asan_rel
BUG=chromium:524425
LOG=N
R=hpayer@chromium.org, ulan@chromium.org
Review URL: https://codereview.chromium.org/1577853007
Cr-Commit-Position: refs/heads/master@{#33523}
This replace HeapType with a dedicated class that implements just what we need for field type tracking. In the next CL, I plan to remove FieldType::Iterator because FieldType can iterate over at most one map.
The ultimate plan is to get rid of templates in types.(h|cc) and remove type-inl.h.
TBR=rossberg@chromium.org
Review URL: https://codereview.chromium.org/1636013002
Cr-Commit-Position: refs/heads/master@{#33521}
Reason for revert:
FAilure on win32 bot, need to investigate webkit failures.
Original issue's description:
> Type Feedback Vector lives in the closure
>
> We get less "pollution" of type feedback if we have one vector per native
> context, rather than one for the whole system. This CL moves the vector
> appropriately.
>
> We rely more heavily on the Optimized Code Map in the SharedFunctionInfo. The
> vector actually lives in the first slot of the literals array (indeed there is
> great commonality between those arrays, they can be thought of as the same
> thing). So we make greater effort to ensure there is a valid literals array
> after compilation.
>
> This meant, for performance reasons, that we needed to extend
> FastNewClosureStub to support creating closures with literals. And ultimately,
> it drove us to move the optimized code map lookup out of FastNewClosureStub
> and into the compile lazy builtin.
>
> The heap change is trivial so I TBR Hannes for it...
>
> TBR=hpayer@chromium.org
>
> BUG=
>
> Committed: https://crrev.com/a5200f7ed4d11c6b882fa667da7a1864226544b4
> Cr-Commit-Position: refs/heads/master@{#33518}
TBR=bmeurer@chromium.org,akos.palfi@imgtec.com
# Skipping CQ checks because original CL landed less than 1 days ago.
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
BUG=
Review URL: https://codereview.chromium.org/1632993003
Cr-Commit-Position: refs/heads/master@{#33520}
We get less "pollution" of type feedback if we have one vector per native
context, rather than one for the whole system. This CL moves the vector
appropriately.
We rely more heavily on the Optimized Code Map in the SharedFunctionInfo. The
vector actually lives in the first slot of the literals array (indeed there is
great commonality between those arrays, they can be thought of as the same
thing). So we make greater effort to ensure there is a valid literals array
after compilation.
This meant, for performance reasons, that we needed to extend
FastNewClosureStub to support creating closures with literals. And ultimately,
it drove us to move the optimized code map lookup out of FastNewClosureStub
and into the compile lazy builtin.
The heap change is trivial so I TBR Hannes for it...
TBR=hpayer@chromium.org
BUG=
Review URL: https://codereview.chromium.org/1563213002
Cr-Commit-Position: refs/heads/master@{#33518}
This increases the size of register operands to be 16-bit.
Not all bytecodes have wide register variants, so when they are
needed a register translator will copy them into a small area
reserved at the top of the 8-bit register range and these registers
are supplied as arguments to the bytecode with 8-bit operands.
This is non-intrusive for typical bytecode where the number of
registers is less than 120. For bytecodes with wide register
operands (above the window) their index needs to be translated
to avoid the reserved translation window.
Enables splay.js to run in Octane and a handful of mjsunit tests.
BUG=v8:4280,v8:4675
LOG=NO
Review URL: https://codereview.chromium.org/1613163002
Cr-Commit-Position: refs/heads/master@{#33516}
- Remove semispace target capacity: It's unused and adds some unneeded
complexity
- Enforcing decl order for SemiSpace
- Move forward declarations in spaces.h to top
- Add all members to default constructor
BUG=chromium:581076
LOG=N
Review URL: https://codereview.chromium.org/1631713002
Cr-Commit-Position: refs/heads/master@{#33515}
This fixes the translation of 'throw' bytecodes to TurboFan graphs. The
correct runtime function is being used now, also the frame states are
attached to the correct nodes now.
R=mythria@chromium.org
TEST=cctest/test-run-jsexceptions/ThrowMessageIndirectly
BUG=v8:4674
LOG=n
Review URL: https://codereview.chromium.org/1636033002
Cr-Commit-Position: refs/heads/master@{#33513}
Adds support for calling native function literals. Moves the logic for building
the native function's SharedFunctionInfo out of full-codegen into compiler.cc
to allow it to be shared between fullcodegen and Ignition.
BUG=v8:4686
LOG=N
Review URL: https://codereview.chromium.org/1635553002
Cr-Commit-Position: refs/heads/master@{#33510}
The current support for try-catch in the interpreter can handle most of
the cases appearing in our test suite. Also the flag in question did not
detect try-finally constructs. This removes the flag and instead extends
the test expectations.
R=rmcilroy@chromium.org
BUG=v8:4674
LOG=n
Review URL: https://codereview.chromium.org/1631593003
Cr-Commit-Position: refs/heads/master@{#33494}
If it's Smi::FromInt(0), the NULL check would trigger. Instead, use the
handle-zap value to mean "not set".
BUG=v8:3647,chromium:580651
R=vogelheim@chromium.org
LOG=y
Review URL: https://codereview.chromium.org/1628173002
Cr-Commit-Position: refs/heads/master@{#33492}
- Completely rely on the concurrent sweeping state for SweepingCompleted()
- Rename the state accordingly.
CQ_EXTRA_TRYBOTS=tryserver.v8:v8_linux_arm64_gc_stress_dbg,v8_linux_gc_stress_dbg,v8_mac_gc_stress_dbg,v8_linux64_asan_rel,v8_linux64_tsan_rel,v8_mac64_asan_rel
R=hpayer@chromium.org
Review URL: https://codereview.chromium.org/1614953002
Cr-Commit-Position: refs/heads/master@{#33490}
Change the interpreter to always store the current context in the frame's
context slot instead of the function context. This makes it possible to
restore the correct context during deopt.
BUG=v8:4678,v8:4280
LOG=N
Review URL: https://codereview.chromium.org/1604923002
Cr-Commit-Position: refs/heads/master@{#33477}
These tests have been disabled while support for try-finally was work in
progress. By now control flow is connected properly and the tests pass.
R=oth@chromium.org
BUG=v8:4674
LOG=n
Review URL: https://codereview.chromium.org/1626503002
Cr-Commit-Position: refs/heads/master@{#33474}
This adds an explicit ReThrow bytecode to be used in the modelling of
try-finally statements. An exception that is being re-thrown should not
trigger message object creation or location computation and hence cannot
use the existing Throw bytecode.
R=rmcilroy@chromium.org
TEST=cctest/test-interpreter/InterpreterTryFinally
BUG=v8:4674
LOG=n
Review URL: https://codereview.chromium.org/1621673002
Cr-Commit-Position: refs/heads/master@{#33472}
This models function local control flow through try-finally constructs
using a token dispatch mechanism. All paths through the finally block
are assigned a token, at the end of the finally block a switch construct
dispatches according to this token.
R=oth@chromium.org,rmcilroy@chromium.org
BUG=v8:4674
LOG=n
Review URL: https://codereview.chromium.org/1613443002
Cr-Commit-Position: refs/heads/master@{#33465}
Break and continue operations need to pop the context chain to the
correct context before jumping to the target.
BUG=v8:4280,v8:4678
LOG=N
Review URL: https://codereview.chromium.org/1618693002
Cr-Commit-Position: refs/heads/master@{#33464}
This fixes the broken return address when the exception handler within
interpreted bytecode is being entered via stack unwinding. The address
in question will never actually be taken, but our stack walker uses this
address to determine whether a frame is interpreted.
R=rmcilroy@chromium.org
TEST=cctest/test-interpreter/InterpreterTryCatch
BUG=v8:4674
LOG=n
Review URL: https://codereview.chromium.org/1615063002
Cr-Commit-Position: refs/heads/master@{#33463}
When accessor getter callback is called the v8::PropertyCallbackInfo::ShouldThrowOnError() is always false, since according to ES6 there's no difference between strict and non-strict property loads. For the setter case the v8::PropertyCallbackInfo::ShouldThrowOnError() returns true if the property is set in strict context.
Interceptors follow same idea: for getter, enumerator and query callbacks the v8::PropertyCallbackInfo::ShouldThrowOnError() is always false, and for setter and deleter callback the v8::PropertyCallbackInfo::ShouldThrowOnError() returns true in strict context.
This CL also cleans up the CallApiGetterStub and removes bogus asserts from [arm] Push(reg1, reg2, ..., regN) that prevented from pushing a set of registers containing duplicates.
BUG=v8:4267
LOG=Y
Committed: https://crrev.com/1d3e837fcbbd9d9fd5e72dfe85dfd47c025f3c9f
Cr-Commit-Position: refs/heads/master@{#33438}
Review URL: https://codereview.chromium.org/1587073003
Cr-Commit-Position: refs/heads/master@{#33461}
Reason for revert:
The random nature of the tests caused the following buildbot to fail: https://build.chromium.org/p/client.v8/builders/V8%20Linux%20gcc%204.8/builds/4724/steps/Check/logs/stdio
Original issue's description:
> [profiler] Implement POC Sampling Heap Profiler
>
> This implements a proof-of-concept sampling based heap profiler inspired by
> tcmalloc's heap profiler [1] and Go's mprof/memprofile [2].
>
> The basic idea is the sample allocations using a randomized Poisson process. At
> any point in time we can cheaply request the set of live sample objects that
> should be a representative sample of heap. Samples include stack-traces from the
> allocation sites, making this an effective tool for memory leak debugging.
>
> Unlike AllocationTracking, this is intended to be cheap and usable online in
> production.
>
> The proof-of-concept is only sampling new-space allocations at this point.
> Support for sampling paged space and native allocations is anticipated in the
> future.
>
> [1] http://goog-perftools.sourceforge.net/doc/heap_profiler.html
> [2] http://blog.golang.org/profiling-go-programs
>
> Committed: https://crrev.com/e5a9947811db9c9e23557dbad27f8b8a349b3262
> Cr-Commit-Position: refs/heads/master@{#33448}
TBR=jochen@chromium.org,alph@chromium.org,hpayer@chromium.org,yangguo@chromium.org
# Skipping CQ checks because original CL landed less than 1 days ago.
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
Review URL: https://codereview.chromium.org/1615173002
Cr-Commit-Position: refs/heads/master@{#33449}
This implements a proof-of-concept sampling based heap profiler inspired by
tcmalloc's heap profiler [1] and Go's mprof/memprofile [2].
The basic idea is the sample allocations using a randomized Poisson process. At
any point in time we can cheaply request the set of live sample objects that
should be a representative sample of heap. Samples include stack-traces from the
allocation sites, making this an effective tool for memory leak debugging.
Unlike AllocationTracking, this is intended to be cheap and usable online in
production.
The proof-of-concept is only sampling new-space allocations at this point.
Support for sampling paged space and native allocations is anticipated in the
future.
[1] http://goog-perftools.sourceforge.net/doc/heap_profiler.html
[2] http://blog.golang.org/profiling-go-programs
Review URL: https://codereview.chromium.org/1555553002
Cr-Commit-Position: refs/heads/master@{#33448}
Reason for revert:
[Sheriff] Breaks layout tests. Please fix upstream.
https://build.chromium.org/p/client.v8.fyi/builders/V8-Blink%20Linux%2064/builds/4077
Original issue's description:
> Array length reduction should throw in strict mode if it can't delete an element.
>
> When accessor getter callback is called the v8::PropertyCallbackInfo::ShouldThrowOnError() is always false, since according to ES6 there's no difference between strict and non-strict property loads. For the setter case the v8::PropertyCallbackInfo::ShouldThrowOnError() returns true if the property is set in strict context.
>
> Interceptors follow same idea: for getter, enumerator and query callbacks the v8::PropertyCallbackInfo::ShouldThrowOnError() is always false, and for setter and deleter callback the v8::PropertyCallbackInfo::ShouldThrowOnError() returns true in strict context.
>
> This CL also cleans up the CallApiGetterStub and removes bogus asserts from [arm] Push(reg1, reg2, ..., regN) that prevented from pushing a set of registers containing duplicates.
>
> BUG=v8:4267
> LOG=Y
>
> Committed: https://crrev.com/1d3e837fcbbd9d9fd5e72dfe85dfd47c025f3c9f
> Cr-Commit-Position: refs/heads/master@{#33438}
TBR=verwaest@chromium.org,ishell@chromium.org
# Skipping CQ checks because original CL landed less than 1 days ago.
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
BUG=v8:4267
Review URL: https://codereview.chromium.org/1611313003
Cr-Commit-Position: refs/heads/master@{#33444}
Motivated by finding a bug in a larger module, this CL adds the ability
to dump out a byte-by-byte, nested view of the decoded AST. This
byte-by-byte output uses the opcode enum to make it readable, but is
suitable for pasting into a byte[] in C or JS and thus making a regression
test.
Also fix a bug; the case of running out of registers for indirect calls.
R=ahaas@chromium.org
BUG=
Review URL: https://codereview.chromium.org/1616973004
Cr-Commit-Position: refs/heads/master@{#33442}
When accessor getter callback is called the v8::PropertyCallbackInfo::ShouldThrowOnError() is always false, since according to ES6 there's no difference between strict and non-strict property loads. For the setter case the v8::PropertyCallbackInfo::ShouldThrowOnError() returns true if the property is set in strict context.
Interceptors follow same idea: for getter, enumerator and query callbacks the v8::PropertyCallbackInfo::ShouldThrowOnError() is always false, and for setter and deleter callback the v8::PropertyCallbackInfo::ShouldThrowOnError() returns true in strict context.
This CL also cleans up the CallApiGetterStub and removes bogus asserts from [arm] Push(reg1, reg2, ..., regN) that prevented from pushing a set of registers containing duplicates.
BUG=v8:4267
LOG=Y
Review URL: https://codereview.chromium.org/1587073003
Cr-Commit-Position: refs/heads/master@{#33438}
These features are not used by devtools and consequently not
exposed through the devtools protocol. They make the debugger
unnecessarily complex. If we decide that we need this, we should
implement this on a higher layer.
R=bmeurer@chromium.org
Review URL: https://codereview.chromium.org/1607193003
Cr-Commit-Position: refs/heads/master@{#33436}
We divide character ranges into
- BMP, matched normally.
- non-BMP, matched as alternatives of surrogate pair ranges.
- lone surrogates, matched with lookaround assertion that its indeed lone.
R=erik.corry@gmail.com
BUG=v8:2952
LOG=N
Review URL: https://codereview.chromium.org/1578253005
Cr-Commit-Position: refs/heads/master@{#33432}
Although the `for..in` statement allows Expressions to define the
iterator, only an AssignmentExpression may occupy this position in the
`for..of` statement.
BUG=v8:4692
LOG=N
R=adamk@chromium.org
Review URL: https://codereview.chromium.org/1602823003
Cr-Commit-Position: refs/heads/master@{#33420}
This implements a first prototype of stack unwinding for interpreted
frames. The unwinding machinery performs a range-based lookup in the
given handler table and potentially continues dispatching at the handler
offset. Note that this does not yet correctly restore the context to the
correct value when the handler is being entered.
R=rmcilroy@chromium.org,oth@chromium.org
BUG=v8:4674
LOG=n
Review URL: https://codereview.chromium.org/1605633003
Cr-Commit-Position: refs/heads/master@{#33414}
Platforms which do not provide rounding instructions (like x64 without
sse4.1, arm before v8) fall back to this new soft float inplementation.
BUG=575379
LOG=Y
R=titzer@chromium.org
Review URL: https://codereview.chromium.org/1611513003
Cr-Commit-Position: refs/heads/master@{#33412}
The motivation for this is that CompilationInfo really shouldn't
explicitly know anything about CodeStubs. This is evident in
the TurboFan stubs pipeline, which only needs to pass down
information about Code::Flags to the code generator and not
any of the CallInterfaceDescriptor silliness that Hydrogen has
to push around, since TF has the Linkage class that
encapsulates everything that is needed for the stub ABI. So,
instead of threading CodeStub machinery through the TF stub
pipeline, it is now removed from CompilationInfo and replaced
by only the explicit bits needed both by the Crankshaft and
TF pipelines in code generation.
Review URL: https://codereview.chromium.org/1604543002
Cr-Commit-Position: refs/heads/master@{#33410}
This removes the above flag definition. The flag is no longer needed as
the default implementation is more than capable of faking presence of
handling of try-catch and try-finally constructs by now.
R=rmcilroy@chromium.org
BUG=v8:4674
LOG=n
Review URL: https://codereview.chromium.org/1603063003
Cr-Commit-Position: refs/heads/master@{#33402}
This implements a first version of exception handler table construction
within the interpreter. Note that the local control flow for try-catch
and try-finally statements is still off, and also stack unwinding does
not yet respect interpreter frames. But generated handler tables should
be populated correctly already.
R=oth@chromium.org
BUG=v8:4674
LOG=n
Review URL: https://codereview.chromium.org/1607433005
Cr-Commit-Position: refs/heads/master@{#33400}
The old handling of escaped keywords erroneously treated escaped versions
of "let" and "static" as ESCAPED_KEYWORD, leading to erroneous errors in
sloppy mode. Moreover, though the class literal parsing code attempted
to fix up the parsing of escaped versions of "static" to allow it in the
right places, that code wasn't complete.
Fixing the scanner to mark escaped "static" as ESCAPED_STRICT_RESERVED_WORD
allows simplifying the class literal parsing code. A little extra code
was needed to properly handle the new treatment of escaped "let".
Note that "yield" is still broken (that is, we're overly restrictive of
escaped "yield" in sloppy mode).
Review URL: https://codereview.chromium.org/1602013007
Cr-Commit-Position: refs/heads/master@{#33396}
