This reverts commit 662aa425ba.
Reason for revert: Crashing on Canary
BUG=chromium:718891
Original change's description:
> Reland: [TypeFeedbackVector] Store optimized code in the vector
>
> Since the feedback vector is itself a native context structure, why
> not store optimized code for a function in there rather than in
> a map from native context to code? This allows us to get rid of
> the optimized code map in the SharedFunctionInfo, saving a pointer,
> and making lookup of any optimized code quicker.
>
> Original patch by Michael Stanton <mvstanton@chromium.org>
>
> BUG=v8:6246
> TBR=yangguo@chromium.org,ulan@chromium.org
>
> Change-Id: Ic83e4011148164ef080c63215a0c77f1dfb7f327
> Reviewed-on: https://chromium-review.googlesource.com/494487
> Reviewed-by: Jaroslav Sevcik <jarin@chromium.org>
> Commit-Queue: Ross McIlroy <rmcilroy@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#45084}
TBR=ulan@chromium.org,rmcilroy@chromium.org,yangguo@chromium.org,jarin@chromium.org
# Not skipping CQ checks because original CL landed > 1 day ago.
BUG=v8:6246
Change-Id: Idab648d6fe260862c2a0e35366df19dcecf13a82
Reviewed-on: https://chromium-review.googlesource.com/498633
Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
Commit-Queue: Ross McIlroy <rmcilroy@chromium.org>
Cr-Commit-Position: refs/heads/master@{#45174}
This patch expands scope analysis to skip hole initialization
when it can be determined statically that no hole checks will
be generated at runtime.
Two conditions must be met to safely eliminate hole initialization:
- There must not exist a VariableProxy referencing this Variable
whose HoleCheckMode is kRequired
- The Variable must be stack allocated; any other allocation implies
that it may be accessed from not-yet-analyzed scopes (other modules,
inner functions, or eval code) and that code may require
hole checks.
The new logic required removing debug code in full-codegen which is
now incorrect in some cases.
Also fixed Variable's bitfield helpers to take no more space than needed.
Bug: chromium:651637
Change-Id: Ie5ac326af4e05b7a5c3c37cd4d0afba6a51a504d
Reviewed-on: https://chromium-review.googlesource.com/494006
Commit-Queue: Adam Klein <adamk@chromium.org>
Reviewed-by: Georg Neis <neis@chromium.org>
Reviewed-by: Leszek Swirski <leszeks@chromium.org>
Reviewed-by: Michael Starzinger <mstarzinger@chromium.org>
Cr-Commit-Position: refs/heads/master@{#45170}
As a first step towards improving generator creation, create a builtin
that can do it without a call to the runtime. Thread that builtin
into the interpreter via an intrinsic.
BUG=v8:6352
R=bmeurer@chromium.org
Review-Url: https://codereview.chromium.org/2861983002
Cr-Commit-Position: refs/heads/master@{#45145}
Instead of explicitly instantiating six copies of each function for
zero to five arguments, define them as variadic template.
This avoids the arbitrary limit to five arguments, and avoids lots of
code duplication.
Drive-by fix: [iwyu] Add missing include.
R=mstarzinger@chromium.org
Change-Id: Id7edde3611ddfade16c7afbbf4b9000cedd76468
Reviewed-on: https://chromium-review.googlesource.com/496247
Reviewed-by: Michael Starzinger <mstarzinger@chromium.org>
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Cr-Commit-Position: refs/heads/master@{#45130}
New test262 tests bring up a couple cases with async arrow functions
that V8 didn't seem to handle properly; this patch makes those cases
errors:
- async (...x,) => y -- Rest parameter must be last formal parameter
- async (...x = z) => y -- No default value for rest parameter
- async (...x, y) => z -- Rest parameter must be last formal parameter
Bug: v8:4483, v8:5051
Change-Id: I024d9ba0c854e8e5e75283df2ee53127b1be090d
Reviewed-on: https://chromium-review.googlesource.com/496057
Commit-Queue: Daniel Ehrenberg <littledan@chromium.org>
Reviewed-by: Adam Klein <adamk@chromium.org>
Reviewed-by: Caitlin Potter <caitp@igalia.com>
Cr-Commit-Position: refs/heads/master@{#45116}
Removes the do-expression wrapping, modifies BytecodeGenerator change
to enter a class literal's block scope if needed.
This does not solve the actual bug in v8:6322, but helps mitigate it in
simple cases. The bug is caused by BytecodeGenerator not allocating a
large enough array of context registers to hold its entire stack,
allowing non-context registers to be overwritten during PushContext and
PopContext bytecodes.
Nevertheless, I like the idea of not depending on do-expressions when
possible, so I think it's worth doing anyways.
BUG=v8:6322
R=rmcilroy@chromium.org, marja@chromium.org, littledan@chromium.org
Change-Id: I82b7569db2a0eead1694bd04765fc4456c2f1a0a
Reviewed-on: https://chromium-review.googlesource.com/491074
Commit-Queue: Caitlin Potter <caitp@igalia.com>
Reviewed-by: Marja Hölttä <marja@chromium.org>
Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
Reviewed-by: Daniel Ehrenberg <littledan@chromium.org>
Cr-Commit-Position: refs/heads/master@{#45110}
- Removes primitive shuffle opcodes.
- Adds Shuffle opcode for S32x4, S16x8, S8x16.
- Adds code to ARM instruction selector to pick best opcodes for some
common shuffle patterns.
LOG=N
BUG=v8:6020
Review-Url: https://codereview.chromium.org/2847663005
Cr-Commit-Position: refs/heads/master@{#45104}
Since the feedback vector is itself a native context structure, why
not store optimized code for a function in there rather than in
a map from native context to code? This allows us to get rid of
the optimized code map in the SharedFunctionInfo, saving a pointer,
and making lookup of any optimized code quicker.
Original patch by Michael Stanton <mvstanton@chromium.org>
BUG=v8:6246
TBR=yangguo@chromium.org,ulan@chromium.org
Change-Id: Ic83e4011148164ef080c63215a0c77f1dfb7f327
Reviewed-on: https://chromium-review.googlesource.com/494487
Reviewed-by: Jaroslav Sevcik <jarin@chromium.org>
Commit-Queue: Ross McIlroy <rmcilroy@chromium.org>
Cr-Commit-Position: refs/heads/master@{#45084}
1. Generalize context specialization such that the provided context
can be any outer context of the function, not necessarily the
immediate outer context.
2. Based on this: if function specialization is disabled, then
specialize for the module context if there is one.
3. Extend typed lowering of module loads and stores such that if
the operand is a Module constant, we constant-fold the cell load.
That is, a JSLoadModule with a Module HeapConstant input becomes
a LoadField with a Cell HeapConstant input, and similarly for
JSStoreModule.
BUG=v8:1569
Review-Url: https://codereview.chromium.org/2841613002
Cr-Commit-Position: refs/heads/master@{#45083}
In the spirit of the full MC, we evacuate and update pointers in parallel for
the young generation.
The collectors are connected during incremental marking when mark bits are
transferred from the young generation bitmap to the old generation bitmap.
The evacuation phase cannot (yet) move pages and relies completely on copying
objects.
BUG=chromium:651354
Review-Url: https://codereview.chromium.org/2796233003
Cr-Commit-Position: refs/heads/master@{#45074}
Blink uses Isolate::GetEnteredContext() to implement HTML's "entry
context" concept, and thus depends on it not being changed except
explicitly (by Blink.) To support this, stop entering contexts
implicitly in all external API entry points; rather just set the
context as current. The only thing that changes the entered context
is now Context::Enter()/Context::Exit() (and Context::Scope.)
BUG=v8:6307
Review-Url: https://codereview.chromium.org/2862483003
Cr-Commit-Position: refs/heads/master@{#45064}
During computation of the side table, ignore stack effects of
instructions following any unconditional jump in the same block
(|unreachable|, |br|, |br_table| or |return| jump out of the block).
Without this fix, the current stack height might underflow, or we compute an
unnecessarily large max_stack_height_. Note that those instruction will
never get executed anyway.
Hence, we don't need to store any side table information for such
unreachable code.
R=rossberg@chromium.org
BUG=chromium:716936, chromium:715990
Change-Id: I282f7f18ba1b972a112210e692f6cd05cf32308c
Reviewed-on: https://chromium-review.googlesource.com/493266
Reviewed-by: Andreas Rossberg <rossberg@chromium.org>
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Cr-Commit-Position: refs/heads/master@{#45059}
We only need to materialize the existing output register for a given
register transfer if it is in a different equivalence set, otherwise we
already have the value we want in the output register.
BUG=v8:4280
Change-Id: Ic4966590ac10445180aff353940d2c93e6a818aa
Reviewed-on: https://chromium-review.googlesource.com/493168
Reviewed-by: Mythri Alle <mythria@chromium.org>
Commit-Queue: Ross McIlroy <rmcilroy@chromium.org>
Cr-Commit-Position: refs/heads/master@{#45051}
Adds tests for Heap::IsUnmodifiedHeapObject that is used during
scavenge.
Bug:
Change-Id: Ide549a6616101cbd6ed17372ed1ed168c7a76fbd
Reviewed-on: https://chromium-review.googlesource.com/484539
Commit-Queue: Mythri Alle <mythria@chromium.org>
Reviewed-by: Toon Verwaest <verwaest@chromium.org>
Reviewed-by: Michael Lippautz <mlippautz@chromium.org>
Cr-Commit-Position: refs/heads/master@{#45046}
This patch adds a concurrent marking deque that exposes the same interface
for the main thread as the existing marking deque.
The matching interface makes the concurrent marking deque a drop-in
replacement for the sequential marking deque without any change in
mark-compactor and incremental marker.
BUG=chromium:694255
Review-Url: https://codereview.chromium.org/2810893002
Cr-Commit-Position: refs/heads/master@{#45042}
Executing the |end| opcode of a loop assumed that the stack height was
being reset to the height at start of the loop. Hence we were ignoring
the arity of the loop.
During computation of the side table, the arity of the label associated
with the loop was explicitly set to 0, such that a |br| instruction to
that label would not transfer any values.
It turns out though that we need to remember the arity in order to
precompute the correct stack height when executing the |end| opcode of
a loop.
Also, add a regression test.
R=rossberg@chromium.org
BUG=chromium:716936
Change-Id: Ib3a559998f1ce5f8fcd7b94af1426637b3e48f86
Reviewed-on: https://chromium-review.googlesource.com/493286
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Reviewed-by: Andreas Rossberg <rossberg@chromium.org>
Cr-Commit-Position: refs/heads/master@{#45041}
BytecodeRegisterOptimizer had special handling for the case when parameters
is 0. This is not possible from valid javascript. It exists because some
tests do not take this into account. Fixed tests and removed the special
handling.
Also removed a TODO, which is already done here:
https://codereview.chromium.org/2227203002/
Bug: v8:4280,v8:6325
Change-Id: Idc17af12ad9292c13a6677aa4c8b88d21f4adf81
Reviewed-on: https://chromium-review.googlesource.com/490308
Commit-Queue: Mythri Alle <mythria@chromium.org>
Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
Cr-Commit-Position: refs/heads/master@{#45036}
This reverts commit c5ad9c6d8e.
Reason for revert: Fails on gc stress:
https://build.chromium.org/p/client.v8/builders/V8%20Linux64%20GC%20Stress%20-%20custom%20snapshot/builds/12661
Original change's description:
> [TypeFeedbackVector] Store optimized code in the vector
>
> Since the feedback vector is itself a native context structure, why
> not store optimized code for a function in there rather than in
> a map from native context to code? This allows us to get rid of
> the optimized code map in the SharedFunctionInfo, saving a pointer,
> and making lookup of any optimized code quicker.
>
> Original patch by Michael Stanton <mvstanton@chromium.org>
>
> BUG=v8:6246
>
> Change-Id: I60ff8c408c3001bc272b4b198c9cbaea2872a9e5
> Reviewed-on: https://chromium-review.googlesource.com/476891
> Commit-Queue: Ross McIlroy <rmcilroy@chromium.org>
> Reviewed-by: Michael Stanton <mvstanton@chromium.org>
> Reviewed-by: Yang Guo <yangguo@chromium.org>
> Reviewed-by: Jaroslav Sevcik <jarin@chromium.org>
> Reviewed-by: Ulan Degenbaev <ulan@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#45022}
TBR=ulan@chromium.org,rmcilroy@chromium.org,yangguo@chromium.org,mvstanton@chromium.org,jarin@chromium.org
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
BUG=v8:6246
Change-Id: I9cd5735b03898cae6ae7adea0f19d32fceb31619
Reviewed-on: https://chromium-review.googlesource.com/493287
Reviewed-by: Michael Achenbach <machenbach@chromium.org>
Commit-Queue: Michael Achenbach <machenbach@chromium.org>
Cr-Commit-Position: refs/heads/master@{#45027}
Since the feedback vector is itself a native context structure, why
not store optimized code for a function in there rather than in
a map from native context to code? This allows us to get rid of
the optimized code map in the SharedFunctionInfo, saving a pointer,
and making lookup of any optimized code quicker.
Original patch by Michael Stanton <mvstanton@chromium.org>
BUG=v8:6246
Change-Id: I60ff8c408c3001bc272b4b198c9cbaea2872a9e5
Reviewed-on: https://chromium-review.googlesource.com/476891
Commit-Queue: Ross McIlroy <rmcilroy@chromium.org>
Reviewed-by: Michael Stanton <mvstanton@chromium.org>
Reviewed-by: Yang Guo <yangguo@chromium.org>
Reviewed-by: Jaroslav Sevcik <jarin@chromium.org>
Reviewed-by: Ulan Degenbaev <ulan@chromium.org>
Cr-Commit-Position: refs/heads/master@{#45022}
1. Replaces --crankshaft with --opt in tests.
2. Also fixes presubmit to check for --opt flag when
assertOptimized is used.
3. Updates testrunner/local/variants.py and
v8_foozie.py to use --opt flag.
This would mean, nooptimize variant means there are
no optimizations. Not even with %OptimizeFunctionOnNextCall.
Bug:v8:6325
Change-Id: I638e743d0773a6729c6b9749e2ca1e2537f12ce6
Reviewed-on: https://chromium-review.googlesource.com/490206
Commit-Queue: Mythri Alle <mythria@chromium.org>
Reviewed-by: Michael Achenbach <machenbach@chromium.org>
Reviewed-by: Michael Starzinger <mstarzinger@chromium.org>
Cr-Commit-Position: refs/heads/master@{#44985}
... and stop checking that the native contexts of maps recorded in feedback vector
match function's native context - the feedback vector machinery already guarantees
that.
BUG=v8:6325
Change-Id: Iacd3f3a5f703694ff57b774b9658e186ad66641b
Reviewed-on: https://chromium-review.googlesource.com/490084
Reviewed-by: Toon Verwaest <verwaest@chromium.org>
Commit-Queue: Igor Sheludko <ishell@chromium.org>
Cr-Commit-Position: refs/heads/master@{#44982}
This avoids generating redundant code for different template
instantiations.
I also introduce getters instead of accessing the fields directly.
R=ahaas@chromium.org
BUG=v8:6325
Change-Id: I3e0eca9ef6a01e0a3ebb73f4f357bcb59e120f43
Reviewed-on: https://chromium-review.googlesource.com/490166
Reviewed-by: Andreas Haas <ahaas@chromium.org>
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Cr-Commit-Position: refs/heads/master@{#44976}
When a FunctionTemplate-based function is used as a constructor
and returns a JSProxy, we incorrectly treated that result the same
as a non-object result. Now it is treated like any other object
result, i.e., it becomes the result of the constructor call.
R=verwaest@chromium.org
BUG=v8:6294
Review-Url: https://codereview.chromium.org/2845123002
Cr-Commit-Position: refs/heads/master@{#44970}
This reduces the amount of special paths for testing.
Setup the memory used for testing exactly the same way as in real world.
Also, always connect the interpreter to the instance being executed,
and to the existing WasmInstance struct. This keeps information
synchronized between interpreter and test runner.
These changes allow us to execute e.g. GrowMemory from cctests either
in the interpreter or in compiled code.
R=ahaas@chromium.org
Change-Id: Id4726d061f3cdba789275350f500d769d27d2d63
Reviewed-on: https://chromium-review.googlesource.com/488561
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Reviewed-by: Andreas Haas <ahaas@chromium.org>
Cr-Commit-Position: refs/heads/master@{#44966}
It now passes on both 32-bit and 64-bit nosnap bots.
TBR=ulan@chromium.org
NOTREECHECKS=true
Change-Id: Id797c88f1eb32868433e112883c2c64b8640eb2c
Reviewed-on: https://chromium-review.googlesource.com/489682
Reviewed-by: Adam Klein <adamk@chromium.org>
Commit-Queue: Adam Klein <adamk@chromium.org>
Cr-Commit-Position: refs/heads/master@{#44953}
These tests assume that a newly-created Isolate is pristine, but that's
not true for nosnap builds.
TBR=ulan@chromium.org
Change-Id: Ie5d0fb0450f285c8eeb8e088feef6729102c0f14
Reviewed-on: https://chromium-review.googlesource.com/489063
Reviewed-by: Adam Klein <adamk@chromium.org>
Commit-Queue: Adam Klein <adamk@chromium.org>
Cr-Commit-Position: refs/heads/master@{#44952}
Most callers passed kFinalizeIncrementalMarkingMask, so use that as
a default argument (not using default argument syntax to avoid including
heap.h in cctest.h).
Change-Id: I904f1eb3a0f5fdbe63eab16f6a6f01d04618645d
Reviewed-on: https://chromium-review.googlesource.com/488104
Reviewed-by: Ulan Degenbaev <ulan@chromium.org>
Commit-Queue: Adam Klein <adamk@chromium.org>
Cr-Commit-Position: refs/heads/master@{#44950}
When branching to a loop header, we were trying to copy over {arity}
values from the value stack. This is correct for block labels, but not
for loops. When branching back to a loop header, no values need to be
transferred.
R=ahaas@chromium.org
BUG=chromium:715454
Change-Id: I90d806de63d039abf8dcac1abec057860c8f69ca
Reviewed-on: https://chromium-review.googlesource.com/488146
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Reviewed-by: Andreas Haas <ahaas@chromium.org>
Cr-Commit-Position: refs/heads/master@{#44949}
With this CL we reduce the difference between directly using a null prototype
in a literal or using Object.create(null).
- The EmitFastCloneShallowObject builtin now supports cloning slow
object boilerplates.
- Unified behavior to find the matching Map and instantiating it for
Object.create(null) and literals with a null prototype.
- Cleanup of literal type parameter of CompileTimeValue, now in sync with
ObjectLiteral flags.
Review-Url: https://codereview.chromium.org/2445333002
Cr-Commit-Position: refs/heads/master@{#44941}
For int16 imm values Subu would emit addiu with -imm value, but doing
this with min_int16 would overflow and produce incorrect result. This is
fixed by checking if -imm is int16. A test for this case is created.
An optimization is also added for values imm where we cannot just emit
addiu and loading -imm to a register takes one instruction using ori.
Then instead of loading imm with lui;ori and subtracting with subu, we
can load -imm with ori and add with addu.
BUG=
TEST=cctest/test-assembler-mips/Subu
Review-Url: https://codereview.chromium.org/2845043002
Cr-Commit-Position: refs/heads/master@{#44932}
This reverts commit d7cdea6fa2.
Reason for revert: Flakiness on bots
Original change's description:
> [wasm] Add guard pages before Wasm Memory
>
> Although Wasm memory indices are all unsigned, they sometimes get assembled
> as 32-bit signed immediates. Values in the top half of the Wasm memory space
> will then get sign extended, causing Wasm to access in front of its memory
> buffer.
>
> Usually this region is not mapped anyway, so faults still happen as they are
> supposed to. This change protects this region with guard pages so we are
> guaranteed to always fault when this happens.
>
> Bug: v8:5277
> Change-Id: Id791fbe2a5ac1b1d75460e65c72b5b9db2a47ee7
> Reviewed-on: https://chromium-review.googlesource.com/484747
> Commit-Queue: Eric Holk <eholk@chromium.org>
> Reviewed-by: Mircea Trofin <mtrofin@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#44905}
TBR=bradnelson@chromium.org,gdeepti@chromium.org,mtrofin@chromium.org,eholk@chromium.org,mseaborn@chromium.org,adamk@chromium.org,v8-reviews@googlegroups.com,wasm-v8@google.com
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
Change-Id: Ia1d3e5dbf4f518815a9fd4197047077bc8e42816
Reviewed-on: https://chromium-review.googlesource.com/487828
Reviewed-by: Adam Klein <adamk@chromium.org>
Commit-Queue: Adam Klein <adamk@chromium.org>
Cr-Commit-Position: refs/heads/master@{#44907}
Although Wasm memory indices are all unsigned, they sometimes get assembled
as 32-bit signed immediates. Values in the top half of the Wasm memory space
will then get sign extended, causing Wasm to access in front of its memory
buffer.
Usually this region is not mapped anyway, so faults still happen as they are
supposed to. This change protects this region with guard pages so we are
guaranteed to always fault when this happens.
Bug: v8:5277
Change-Id: Id791fbe2a5ac1b1d75460e65c72b5b9db2a47ee7
Reviewed-on: https://chromium-review.googlesource.com/484747
Commit-Queue: Eric Holk <eholk@chromium.org>
Reviewed-by: Mircea Trofin <mtrofin@chromium.org>
Cr-Commit-Position: refs/heads/master@{#44905}
The only users of the LoadStoreOpcodeOf function were a number of
macros in wasm-macro-gen.h, and three test functions using it directly.
This CL refactors those functions to also use the macros.
In one case, this requires storing the value in a local variable first.
R=ahaas@chromium.org
Change-Id: Ia2fbf67a3831fafc9345e155eb240cf1bf6feb5d
Reviewed-on: https://chromium-review.googlesource.com/486842
Reviewed-by: Andreas Haas <ahaas@chromium.org>
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Cr-Commit-Position: refs/heads/master@{#44885}
This patch adds a new interface called RootVisitor and changes the root
iteration functions to accept a RootVisitor instead of an ObjectVisitor.
Future CLs will change ObjectVisitor to provide the host object to all
visiting functions, which will bring it in sync with static visitors.
Having separate visitors for roots and objects removes ambiguity in
VisitPointers and reduces chances of forgetting to record slots.
This is intended as pure refactoring. All places that require behavior
change are marked with TODO and will addressed in future CLs.
BUG=chromium:709075
Review-Url: https://codereview.chromium.org/2801073006
Cr-Commit-Position: refs/heads/master@{#44852}
This header file is only used from tests.
Also, move the LoadStoreOpcodeOf method (only used in tests) from
wasm-opcodes.h to wasm-macro-gen.h.
R=ahaas@chromium.org
Change-Id: I8d4691be494b5c1fbe3084441329850930bad647
Reviewed-on: https://chromium-review.googlesource.com/486861
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Reviewed-by: Andreas Haas <ahaas@chromium.org>
Cr-Commit-Position: refs/heads/master@{#44845}
wasm-macro-gen.h is mainly used from tests, but LocalDeclEncoder is
also used from various other places.
This CL moves the LocalDeclEncoder to an own compilation unit. We want
to later move wasm-macro-gen.h to the tests folder.
It also refactors the LocalDeclEncoder to reuse the
LEBHelper::write_u32v and LEBHelper::sizeof_u32v methods instead of
reimplementing it.
R=ahaas@chromium.org
Change-Id: Ia4651436f0544578da7c1c43596d343571942e97
Reviewed-on: https://chromium-review.googlesource.com/486724
Reviewed-by: Andreas Haas <ahaas@chromium.org>
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Cr-Commit-Position: refs/heads/master@{#44838}
This makes it easier to set the value for embedders where it is
difficult to plumb through to the Isolate constructor.
BUG=chromium:711809
R=jochen@chromium.org
Review-Url: https://codereview.chromium.org/2829223002
Cr-Commit-Position: refs/heads/master@{#44813}
