Add a new StackFrame class for unoptimized frames (which are either
interpreted or baseline). BaselineFrame becomes a subclass of this
rather than InterpretedFrame, and the various frame constants helpers
are similarly amended.
Bug: v8:11420, v8:11429
Change-Id: I87e9368aef48ef06a39476bf826f379ce1441528
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2692208
Commit-Queue: Leszek Swirski <leszeks@chromium.org>
Auto-Submit: Leszek Swirski <leszeks@chromium.org>
Reviewed-by: Jakob Gruber <jgruber@chromium.org>
Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
Reviewed-by: Ulan Degenbaev <ulan@chromium.org>
Cr-Commit-Position: refs/heads/master@{#72743}
This reflects the actual contents of the type, which is an offset into
the bytecode (or certain marker values). Historically, in the days of
FCG the bailout id used to refer to node ids - this is why certain
tracing output still calls the bailout id 'node id' and 'ast id'.
These spots will be fixed in a follow-up CL.
This change is mechanical:
git grep -l BailoutId | while read f; do \
sed -i 's/BailoutId/BytecodeOffset/g' $f; done
With a manual component of updating the DeoptimizationData method
name from 'BytecodeOffset' to 'GetBytecodeOffset'.
Bug: v8:11332
Change-Id: I956b947a480bf52263159c0eb1e895360bcbe6d2
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2639754
Commit-Queue: Jakob Gruber <jgruber@chromium.org>
Reviewed-by: Nico Hartmann <nicohartmann@chromium.org>
Reviewed-by: Mythri Alle <mythria@chromium.org>
Cr-Commit-Position: refs/heads/master@{#72189}
Since it will be patched in later in the cases where it will be used,
there is no need to have it as a parameter.
Bug: v8:7790
Change-Id: I93b27f3baf8c3841a60f5ac5ed09993d1caf19bc
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2351667
Reviewed-by: Georg Neis <neis@chromium.org>
Commit-Queue: Santiago Aboy Solanes <solanes@chromium.org>
Cr-Commit-Position: refs/heads/master@{#69366}
This is a reland of 13141c8a65
... with a fix for an UB issue of passing null pointers to memcpy()
when size is zero.
TBR=leszeks@chromium.org
Original change's description:
> [zone-compr] Introduce ZoneTypeTraits and ZoneCompression
>
> Also move zone compression flags to src/common/globals.h.
>
> Bug: v8:9923
> Change-Id: Id0a77720e735e2669a1e5eef48e1b4866ad99480
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2324255
> Reviewed-by: Leszek Swirski <leszeks@chromium.org>
> Commit-Queue: Igor Sheludko <ishell@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#69160}
Bug: v8:9923
Change-Id: I2245b81516c39ccea262c282c659ef601af57abf
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2332165
Commit-Queue: Igor Sheludko (OOO Aug 3-17) <ishell@chromium.org>
Reviewed-by: Igor Sheludko (OOO Aug 3-17) <ishell@chromium.org>
Cr-Commit-Position: refs/heads/master@{#69166}
... that controls whether the TF graph zones should support compression.
Bug: v8:9923
Change-Id: Ifbe237b75e9c92e62eb32b69d6b3b1a818269b83
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2308347
Commit-Queue: Igor Sheludko <ishell@chromium.org>
Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
Cr-Commit-Position: refs/heads/master@{#69036}
This CL adds functionality to read the source positions directly
from the JS heap rather than from serialized data.
In order to do this, we create a PersistentHandles container in the
OptimizedCompilationInfo which gets passed onto the JSHeapBroker. This
allows us to create the handles in the main thread and pass them safely
to the background thread.
In order to read safely from the background thread, we need a LocalHeap
which blocks the GC from running and potentially moving the handles.
This LocalHeap is created only when the JSHeapBroker has finalized
serializing and destroyed when retiring it.
Bug: v8:7790
Change-Id: I19f8b08d12e5be0a3df34d6af2043310c0c7b6fe
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2277802
Reviewed-by: Ulan Degenbaev <ulan@chromium.org>
Reviewed-by: Dominik Inführ <dinfuehr@chromium.org>
Reviewed-by: Georg Neis <neis@chromium.org>
Commit-Queue: Santiago Aboy Solanes <solanes@chromium.org>
Cr-Commit-Position: refs/heads/master@{#68836}
... to OptimizedCompilationInfo, BytecodeGraphBuilder, and
JSHeapBroker.
Also add first uses of these flags in pipeline.cc by skipping certain
phases when nci is enabled. With this change, tests in the NCI variant
will start to fail since generic lowering is not fully implemented.
These implementations will follow incrementally in the next days.
Bug: v8:8888
Change-Id: I3f570fb92f09059d1f1f4015f88ffe80ccf746ad
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2239572
Commit-Queue: Jakob Gruber <jgruber@chromium.org>
Reviewed-by: Georg Neis <neis@chromium.org>
Cr-Commit-Position: refs/heads/master@{#68339}
... such that we have only a single representation for special
constants such as undefined, namely the corresponding bitset.
With this CL the following property holds:
t1.IsSingleton() /\ t2.Is(t1) => t1.Is(t2)
Also clean up the Type interface and improve test coverage a little.
Change-Id: I074e20047c92e2c8215c2d438f2627f4ffdbc409
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2096631
Commit-Queue: Tobias Tebbi <tebbi@chromium.org>
Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
Auto-Submit: Georg Neis <neis@chromium.org>
Cr-Commit-Position: refs/heads/master@{#66684}
It could also be a DeadValue.
A regression test will take a while but the fix is straightforward.
Bug: chromium:1027045
Change-Id: I49a66668b7189b7ea7d6d79d514b9e0de3edc966
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1928853
Commit-Queue: Georg Neis <neis@chromium.org>
Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
Cr-Commit-Position: refs/heads/master@{#65094}
E.g. make the setup of the ObjectRef hash table more explicit.
Tbr: jgruber@chromium.org
Bug: v8:7790
Change-Id: I58c03848e7da5c418ff2d6ae1e71b644278f406b
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1776089
Commit-Queue: Georg Neis <neis@chromium.org>
Reviewed-by: Maya Lekova <mslekova@chromium.org>
Cr-Commit-Position: refs/heads/master@{#63475}
This adds a simple counter to Turbofan that's incremented throughout the compilation, hopefully
frequently enough so we can use it to detect divergence and performance bugs.
In addition, we assert that this counter never gets too high. That's the equivalent of a simple
timeout, just more deterministic. The limitations on Turbofan input size should guarantee that
we never exceed this limit. Since we probably do exceed it rarely, this check is only a DCHECK and
intended to detect performance and divergence issues, but not supposed to be performed in release
builds.
In addition, this CL adds UMA stats to observe the real world distribution of the tick measurement.
Bug: v8:9444
Change-Id: I182dac6ecac64715e3f5885ff5c7c17549351cd0
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1695475
Commit-Queue: Tobias Tebbi <tebbi@chromium.org>
Reviewed-by: Georg Neis <neis@chromium.org>
Reviewed-by: Michael Stanton <mvstanton@chromium.org>
Cr-Commit-Position: refs/heads/master@{#62754}
Got rid of the following circular header dependency chains:
compilation-dependencies <-> js-heap-broker <-> access-info
types <-> js-heap-broker <-> access-info
Extracted former CompilationDependencies::Dependency class into its own header.
Extracted *Ref classes into their own header.
This should enable building on older GCC versions, e.g. 5.4.0.
Bug: v8:9440
Change-Id: Ia345bc227d8f7806d0b8622b706346a7ce6d01ea
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1687415
Commit-Queue: Maya Lekova <mslekova@chromium.org>
Reviewed-by: Georg Neis <neis@chromium.org>
Reviewed-by: Michael Stanton <mvstanton@chromium.org>
Cr-Commit-Position: refs/heads/master@{#62541}
In native context specialization, reducing a JSResolvePromise
node requires us to know that there are no "then" properties on
the resolution object's maps. This work must be done at serialization
time.
Bug: v8:7790
Change-Id: If905513a028bc3d71379e2a31e86fff1d3383141
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1666988
Reviewed-by: Georg Neis <neis@chromium.org>
Reviewed-by: Maya Lekova <mslekova@chromium.org>
Commit-Queue: Michael Stanton <mvstanton@chromium.org>
Cr-Commit-Position: refs/heads/master@{#62519}
As a component of the wider Turbofan logging scheme, it makes sense
for JSHeapBroker logging to come through flags specified in the
OptimizedCompilationInfo class, which uses --trace-turbo-filter
to control which functions are logged.
Bug: v8:7790
Change-Id: I3b068d8be78867ab0bd9607dda9eca4123b9d7b1
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1655297
Reviewed-by: Maya Lekova <mslekova@chromium.org>
Commit-Queue: Michael Stanton <mvstanton@chromium.org>
Cr-Commit-Position: refs/heads/master@{#62111}
The call to EnsureSourcePositionsAvailable for a given SharedFunctionInfo
is now done in the serializer for each SFI that is marked as serialized for
compilation. This will enable brokerization of the JSInliner class.
Change-Id: I7821a50fcac8a3e19386e98758f2b0dea3023bb6
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1582400
Commit-Queue: Maya Lekova <mslekova@chromium.org>
Reviewed-by: Georg Neis <neis@chromium.org>
Cr-Commit-Position: refs/heads/master@{#61260}
and use Mixin pattern with linear inheritance instead. This will
allow to customize the way the Isolate is created.
Bug: v8:8238
Change-Id: Ic611df123653af3a0f2271394387492e440b5ea8
Reviewed-on: https://chromium-review.googlesource.com/c/1306433
Reviewed-by: Benedikt Meurer <bmeurer@chromium.org>
Commit-Queue: Igor Sheludko <ishell@chromium.org>
Cr-Commit-Position: refs/heads/master@{#57106}
There's no ambiguity and the shorter name makes things easier to read.
Bug: v8:7790
Change-Id: Ibcf3fd7f38a91e26a83cd335fad0ec80a5fe9be1
Reviewed-on: https://chromium-review.googlesource.com/c/1278392
Commit-Queue: Georg Neis <neis@chromium.org>
Reviewed-by: Jaroslav Sevcik <jarin@chromium.org>
Reviewed-by: Maya Lekova <mslekova@chromium.org>
Cr-Commit-Position: refs/heads/master@{#56623}
- Add a new broker mode kRetired, in which the heap can
again be accessed.
- Change the way modes work. We now always start in kDisabled.
If FLAG_concurrent_compiler_frontend is on, we eventually move
to kSerializing, then to kSerialized, then to kRetired.
- Add an ObjectDataKind to ObjectData that indicates whether the
data is just a dummy (i.e. created while broker was in kDisabled
mode).
This also happens to fix a bug found by clusterfuzz.
Bug: v8:7790, chromium:889722
Change-Id: I38833fe7ad26d2d3efb15ba560576defb82f673a
Reviewed-on: https://chromium-review.googlesource.com/1245425
Commit-Queue: Georg Neis <neis@chromium.org>
Reviewed-by: Maya Lekova <mslekova@chromium.org>
Reviewed-by: Jaroslav Sevcik <jarin@chromium.org>
Cr-Commit-Position: refs/heads/master@{#56260}
This is just a cleanup.
Bug: v8:7790
Change-Id: Ic0114451159b8c504f527f3cf3bdaed6a8cc8741
Reviewed-on: https://chromium-review.googlesource.com/1243103
Commit-Queue: Georg Neis <neis@chromium.org>
Reviewed-by: Jaroslav Sevcik <jarin@chromium.org>
Cr-Commit-Position: refs/heads/master@{#56206}
... if FLAG_concurrent_compiler_frontend is enabled.
Bug: v8:7790
Change-Id: I448560b21d54c8907e8cbf68bdaf8bbdf2b034df
Reviewed-on: https://chromium-review.googlesource.com/1241959
Reviewed-by: Jaroslav Sevcik <jarin@chromium.org>
Commit-Queue: Georg Neis <neis@chromium.org>
Cr-Commit-Position: refs/heads/master@{#56205}
Remove the NumberConstant right hand side limitation for the speculative
number operation optimization, and extend the logic to also deal with
SpeculativeToNumber, which is common when dealing with postfix increment
and array operations.
Also add appropriate tests for all the relevant cases, specifically we
mjsunit tests to increase the general coverage for the various cases
here (in addition to dedicated unittests).
Bug: v8:8015
Change-Id: I8c92f98490c63b07eb19686efd404322979e57c4
Reviewed-on: https://chromium-review.googlesource.com/1235919
Reviewed-by: Jaroslav Sevcik <jarin@chromium.org>
Commit-Queue: Benedikt Meurer <bmeurer@chromium.org>
Cr-Commit-Position: refs/heads/master@{#56072}
The heap broker expects that handles get canonicalized.
R=jarin@chromium.org
Bug: v8:7790
Change-Id: If6162316bb2a256e783a8175ac7d4172d040b28b
Reviewed-on: https://chromium-review.googlesource.com/1155123
Commit-Queue: Georg Neis <neis@chromium.org>
Reviewed-by: Jaroslav Sevcik <jarin@chromium.org>
Cr-Commit-Position: refs/heads/master@{#54823}
We'll soon start collecting data from the JS heap prior to the typed
lowering pass, and then refrain from reading the heap in that pass.
This CL prepares the broker machinery by introducing a hash table that
maps an object (handle) to the corresponding cached data. For the time
being, that cached data is essentially just the handle itself.
Bug: v8:7790
Change-Id: I830e9c72faafb7ae1d10e8a111636b3a3762bbc6
Reviewed-on: https://chromium-review.googlesource.com/1143405
Commit-Queue: Georg Neis <neis@chromium.org>
Reviewed-by: Maya Lekova <mslekova@chromium.org>
Reviewed-by: Jaroslav Sevcik <jarin@chromium.org>
Cr-Commit-Position: refs/heads/master@{#54618}
As a first step towards moving accesses to the broker, this moves
heap accesses from BitsetType::Lub to the broker.
Bug: v8:7790
Change-Id: Ie240b84b979717caae42cb8aa06ee8d9877a446d
Reviewed-on: https://chromium-review.googlesource.com/1088695
Commit-Queue: Jaroslav Sevcik <jarin@chromium.org>
Reviewed-by: Benedikt Meurer <bmeurer@chromium.org>
Cr-Commit-Position: refs/heads/master@{#53571}
Modifies several Type:: methods to take an Isolate to pass through to
BitSetType::Lub as well as their call sites.
Bug: v8:7786
Cq-Include-Trybots: luci.chromium.try:linux_chromium_rel_ng
Change-Id: I9ac769c4c658995421fd28b9b1d77d6f84627116
Reviewed-on: https://chromium-review.googlesource.com/1071515
Commit-Queue: Dan Elphick <delphick@chromium.org>
Reviewed-by: Michael Starzinger <mstarzinger@chromium.org>
Cr-Commit-Position: refs/heads/master@{#53362}
This also includes the precise reducer name. Currently the information
is available in the node tooltip in turbolizer. The new shortcut 's' in
the graph view selects the nodes the currently selected nodes were created
from.
Bug: v8:7327
Change-Id: I7ca7327d0cfa112972e3567df6e4a223c8eff3c0
Reviewed-on: https://chromium-review.googlesource.com/1064059
Commit-Queue: Sigurd Schneider <sigurds@chromium.org>
Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
Cr-Commit-Position: refs/heads/master@{#53258}
This is part of the effort to decrease the amount of undefined behavior.
that v8 relies on.
The main change here is to represent types with class Type rather than
with pointer Type*. To make the CL smaller, I used an operator overload
hack to separate the change from `->` to `.`. I am working on a CL that
will remove the operator and change all those arrows to dots.
Bug: v8:3770
Change-Id: I71a197cb739a1467937bc95c2a757fab0469aa22
Reviewed-on: https://chromium-review.googlesource.com/1032551
Commit-Queue: Jaroslav Sevcik <jarin@chromium.org>
Reviewed-by: Sigurd Schneider <sigurds@chromium.org>
Cr-Commit-Position: refs/heads/master@{#52872}
There is no good reason to have the meat of most objects' initialization
logic in heap.cc, all wrapped by the CALL_HEAP_FUNCTION macro. Instead,
this CL changes the protocol between Heap and Factory to be AllocateRaw,
and all object initialization work after (possibly retried) successful
raw allocation happens in the Factory.
This saves about 20KB of binary size on x64.
Original review: https://chromium-review.googlesource.com/c/v8/v8/+/959533
Originally landed as r52416 / f9a2e24bbc
Cq-Include-Trybots: luci.v8.try:v8_linux_noi18n_rel_ng
Change-Id: Id072cbe6b3ed30afd339c7e502844b99ca12a647
Reviewed-on: https://chromium-review.googlesource.com/1000540
Commit-Queue: Jakob Kummerow <jkummerow@chromium.org>
Reviewed-by: Hannes Payer <hpayer@chromium.org>
Reviewed-by: Michael Starzinger <mstarzinger@chromium.org>
Cr-Commit-Position: refs/heads/master@{#52492}
This reverts commit f9a2e24bbc.
Reason for revert: gc stress failures not all fixed by follow up.
Original change's description:
> [cleanup] Refactor the Factory
>
> There is no good reason to have the meat of most objects' initialization
> logic in heap.cc, all wrapped by the CALL_HEAP_FUNCTION macro. Instead,
> this CL changes the protocol between Heap and Factory to be AllocateRaw,
> and all object initialization work after (possibly retried) successful
> raw allocation happens in the Factory.
>
> This saves about 20KB of binary size on x64.
>
> Cq-Include-Trybots: luci.v8.try:v8_linux_noi18n_rel_ng
> Change-Id: Icbfdc4266d7be8b48d2fe085f03411743dc6a0ca
> Reviewed-on: https://chromium-review.googlesource.com/959533
> Commit-Queue: Jakob Kummerow <jkummerow@chromium.org>
> Reviewed-by: Hannes Payer <hpayer@chromium.org>
> Reviewed-by: Yang Guo <yangguo@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#52416}
TBR=jkummerow@chromium.org,yangguo@chromium.org,mstarzinger@chromium.org,hpayer@chromium.org
Change-Id: Idbbc53478742f3e9525eee83342afc6aedae122f
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Cq-Include-Trybots: luci.v8.try:v8_linux_noi18n_rel_ng
Reviewed-on: https://chromium-review.googlesource.com/999414
Reviewed-by: Michael Achenbach <machenbach@chromium.org>
Commit-Queue: Michael Achenbach <machenbach@chromium.org>
Cr-Commit-Position: refs/heads/master@{#52420}
There is no good reason to have the meat of most objects' initialization
logic in heap.cc, all wrapped by the CALL_HEAP_FUNCTION macro. Instead,
this CL changes the protocol between Heap and Factory to be AllocateRaw,
and all object initialization work after (possibly retried) successful
raw allocation happens in the Factory.
This saves about 20KB of binary size on x64.
Cq-Include-Trybots: luci.v8.try:v8_linux_noi18n_rel_ng
Change-Id: Icbfdc4266d7be8b48d2fe085f03411743dc6a0ca
Reviewed-on: https://chromium-review.googlesource.com/959533
Commit-Queue: Jakob Kummerow <jkummerow@chromium.org>
Reviewed-by: Hannes Payer <hpayer@chromium.org>
Reviewed-by: Yang Guo <yangguo@chromium.org>
Cr-Commit-Position: refs/heads/master@{#52416}
I also used the opportunity to clean up the loop peeler a bit by making the
class stateful, to avoid passing long argument lists around.
Bug: v8:5864
Change-Id: I2e034c6eabd381b01e15cf3e6aa3ce7b14e7b3d8
Reviewed-on: https://chromium-review.googlesource.com/822933
Commit-Queue: Tobias Tebbi <tebbi@chromium.org>
Reviewed-by: Benedikt Meurer <bmeurer@chromium.org>
Cr-Commit-Position: refs/heads/master@{#50067}
Add a more efficient encoding for state values that have a large number of
optimized-out inputs.
Review-Url: https://codereview.chromium.org/2509623002
Cr-Commit-Position: refs/heads/master@{#42088}
Also lower JSToBoolean(x) where x is either some detectable receiver or
null, or any kind of receiver, null or undefined. Also fix a couple of
minor issues with the JSToBoolean lowering and tests.
R=yangguo@chromium.org
BUG=v8:5267
Review-Url: https://codereview.chromium.org/2530773002
Cr-Commit-Position: refs/heads/master@{#41241}
There were once plans to generate cross-context code with TurboFan,
however that doesn't fit into the model anymore, and so all of this
is essentially dead untested code (and thus most likely already broken
in subtle ways). With this mode still in place it would also be a lot
harder to make inlining based on SharedFunctionInfo work.
BUG=v8:2206,v8:5499
R=jarin@chromium.org
Review-Url: https://codereview.chromium.org/2406803002
Cr-Commit-Position: refs/heads/master@{#40109}
With this CL, we devolve all Constants introduced as they are with an object handle into
* Range - for integers
* Nan
* MinusZero
* OtherNumberConstant - for doubles
* HeapConstant
We reduce the amount we have to inspect an object handle during optimization. Also, simplifications result. For example, you never have to check if a Range contains a HeapConstant.
BUG=
Review-Url: https://codereview.chromium.org/2381523002
Cr-Commit-Position: refs/heads/master@{#40041}
Now that it is no longer needed, this also removes the invalid inclusion
of "object-inl.h" within the "unique.h" header file.
Note that this change still leaves 2 violations of that rule in the
code, checked with the "tools/check-inline-includes.sh" tool.
R=bmeurer@chromium.org
Review URL: https://codereview.chromium.org/1321223002
Cr-Commit-Position: refs/heads/master@{#30503}
The usage of Unique<T> throughout the TurboFan IR does not have any
advantage. There is no single point in time when they are initialized
and most use-sites looked through to the underlying Handle<T> anyways.
Also there already was a mixture of Handle<T> versus Unique<T> in the
graph and this unifies the situation to use Handle<T> everywhere.
R=bmeurer@chromium.org,titzer@chromium.org
Review URL: https://codereview.chromium.org/1314473007
Cr-Commit-Position: refs/heads/master@{#30458}
This optimization never triggers currently, and is inherently native
context dependent for no real reason (for example it will not properly
detect those constructors in the case of cross native context inlining),
plus it is slow and awkward. In case we really need this functionality
at some point, we should find a way to make it work with the builtin
function id mechanism that is already in place to match other builtins.
R=jarin@chromium.org,rossberg@chromium.org
Review URL: https://codereview.chromium.org/1221683006
Cr-Commit-Position: refs/heads/master@{#29365}
- Thread Type::FunctionType through stubs and the TF pipeline.
- Augment Typer to decorate parameter nodes with types from
a Type::FunctionType associated with interface descriptors.
- Factor interface descriptors into platform-specific and
platform-independent components so that all descriptors share
a common Type::FunctionType for all platforms.
Review URL: https://codereview.chromium.org/1197703002
Cr-Commit-Position: refs/heads/master@{#29248}
This also threads through the parameter count and local count to the instruction selector. This will be later used to allow merging of various StateValues vector (and prepare for differential encoding which will not distinguish between parameters, locals and expression stack).
BUG=
Review URL: https://codereview.chromium.org/1191243003
Cr-Commit-Position: refs/heads/master@{#29214}
