This means we can write a memset32 (load32 -> store32),
tested explicitly with the new unit test.
Slightly changes to the type protocol,
- load and splat now generate scalars or vectors
depending on how `scalar` is set
- store should no longer have to pay attention to `scalar`;
it's input values will already be the right size
Clean up some of the type declarations where we don't
actually need the subclass types, holding llvm::Type* instead.
This makes using ?: easier.
Change-Id: I2f98701ebdeead0513d355b2666b024794b90193
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/273781
Reviewed-by: Herb Derby <herb@google.com>
Commit-Queue: Mike Klein <mtklein@google.com>
This reverts commit 838007f1ff.
Reason for revert: Significant RAM regression detected by Chrome.
Original change's description:
> Add most important intrinsics to the interpreter
>
> Started with component-wise comparison and mix builtins.
>
> Implemented min, max, clamp, and saturate using those.
> Moved dot to SkSL as well. Because we now have builtins
> implemented using other (intrinsic) builtins, I had to
> split the include file in two - this lets the intrinsics
> be marked so we can call them from the second phase of
> builtins that are written in SkSL.
>
> Given that the comparisons and kSelect are now used by
> these, I added vector versions of those instructions.
> I also switched the kSelect args to match GLSL mix(),
> mostly so the logic of mapping intrinsic arguments to
> instruction register args remains simple.
>
> Inspired by the (never-landed):
> https://skia-review.googlesource.com/c/skia/+/230739
>
> Change-Id: Iecb0a7e8dc633625ff2cada7fb962bf2137fa881
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/272516
> Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
> Commit-Queue: Brian Osman <brianosman@google.com>
TBR=bsalomon@google.com,brianosman@google.com,ethannicholas@google.com,reed@google.com
Change-Id: I931a0ccc254b55339c9b077543a0daaf28146b19
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/273800
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
And rearrange a little so that it's more similar
to the other JIT state we track, so this sort
of bug is less likely in the future.
Change-Id: I64ddb791490efdbbcecb53cce823dda5c8d7c68b
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/273779
Auto-Submit: Mike Klein <mtklein@google.com>
Commit-Queue: Herb Derby <herb@google.com>
Reviewed-by: Herb Derby <herb@google.com>
This reverts commit 915b779f9c.
Reason for revert: finally coming back to this, figure out what's wrong on Android
Original change's description:
> Revert "Track device coordinate space as matrix"
>
> This reverts commit b74d5548a4.
>
> Reason for revert: see if this fixes the android roll
>
> Original change's description:
> > Track device coordinate space as matrix
> >
> > This is a required step to be able to cleanly draw image filtered
> > device layers with arbitrary matrices, instead of relying on
> > SkMatrixImageFilter to apply the transformation.
> >
> > Bug: skia:9545
> > Change-Id: I8d84679a281538875cf4a1b73565294fb7f89c86
> > Reviewed-on: https://skia-review.googlesource.com/c/skia/+/249076
> > Commit-Queue: Michael Ludwig <michaelludwig@google.com>
> > Reviewed-by: Mike Reed <reed@google.com>
>
> TBR=robertphillips@google.com,reed@google.com,michaelludwig@google.com
>
> # Not skipping CQ checks because original CL landed > 1 day ago.
>
> Bug: skia:9545
> Change-Id: Ie374a7500cfbff35cb0782beb863086e118a005a
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/249986
> Reviewed-by: Michael Ludwig <michaelludwig@google.com>
> Commit-Queue: Michael Ludwig <michaelludwig@google.com>
TBR=robertphillips@google.com,reed@google.com,michaelludwig@google.com
# Not skipping CQ checks because original CL landed > 1 day ago.
Bug: skia:9545
Change-Id: If31a9be86cb340a0874533c044c19b6787d5f176
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/272340
Reviewed-by: Mike Reed <reed@google.com>
Commit-Queue: Michael Ludwig <michaelludwig@google.com>
Change-Id: If718decf8867cd66d8394a4f7fc646cf3f0950b3
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/273609
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: Brian Salomon <bsalomon@google.com>
With the removal of rescaling of NPOT textures this no longer matters.
Change-Id: I313e895407c3a2c616e6113a5bde75dc6a167e7c
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/273519
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Auto-Submit: Brian Salomon <bsalomon@google.com>
I've also added in a disabled path to do quasi -O1 / -Os
passes, but mem2reg is really all we need, and I think we
can eliminate the need for even that.
Codegen looks great now, e.g.
(lldb) dis -s fJITEntry -c 40
0x10a48c000: movabsq $0x10a48d000, %rax ; imm = 0x10A48D000
0x10a48c00a: vbroadcastss (%rax), %zmm0
0x10a48c010: cmpq $0xf, %rdi
0x10a48c014: jbe 0x10a48c04d
0x10a48c016: nopw %cs:(%rax,%rax)
0x10a48c020: vmovups %zmm0, (%rsi)
0x10a48c026: addq $-0x10, %rdi
0x10a48c02a: addq $0x40, %rsi
0x10a48c02e: cmpq $0xf, %rdi
0x10a48c032: ja 0x10a48c020
0x10a48c034: jmp 0x10a48c04d
0x10a48c036: nopw %cs:(%rax,%rax)
0x10a48c040: movl $0x2a, (%rsi)
0x10a48c046: decq %rdi
0x10a48c049: addq $0x4, %rsi
0x10a48c04d: testq %rdi, %rdi
0x10a48c050: jne 0x10a48c040
0x10a48c052: vzeroupper
0x10a48c055: retq
Change-Id: Ief5d12548d5b1a683060b2b5d207022d673fe761
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/273606
Reviewed-by: Herb Derby <herb@google.com>
Change-Id: Ie50b15323df0a71c8d4276e3bc603061e469824d
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/273465
Reviewed-by: Jim Van Verth <jvanverth@google.com>
Commit-Queue: Greg Daniel <egdaniel@google.com>
Started with component-wise comparison and mix builtins.
Implemented min, max, clamp, and saturate using those.
Moved dot to SkSL as well. Because we now have builtins
implemented using other (intrinsic) builtins, I had to
split the include file in two - this lets the intrinsics
be marked so we can call them from the second phase of
builtins that are written in SkSL.
Given that the comparisons and kSelect are now used by
these, I added vector versions of those instructions.
I also switched the kSelect args to match GLSL mix(),
mostly so the logic of mapping intrinsic arguments to
instruction register args remains simple.
Inspired by the (never-landed):
https://skia-review.googlesource.com/c/skia/+/230739
Change-Id: Iecb0a7e8dc633625ff2cada7fb962bf2137fa881
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/272516
Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Replaces GrMesh::SendToGpuImpl with direct draw(), drawIndexed(),
drawInstanced(), and drawIndexedInstanced() calls on GrOpsRenderPass.
For now these calls take the index/instance/vertex buffers, but we
will soon switch to binding those in a separate call.
Change-Id: I55d225b5695d88ecc661916c2aeb3f31d21e5585
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/273179
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Chris Dalton <csmartdalton@google.com>
This is a reland of af5f9f008d
This was reverted due to Metal GMs failing. That is a bug in Metal where
pipeline creation fails with "internal error". Reporting to Apple, filing
Skia bug, and moving on.
Original change's description:
> Remove GrDeviceSpaceTextureDecalFragmentProcessor.
>
> It was used to sample clip masks using device coords.
>
> Replace with a more generic GrDeviceSpaceEffect that simply calls a
> child FP with sk_FragCoord.
>
> Also fix issue in GrQuadPerEdgeAA GP. It wouldn't setup coord transforms
> at all if they are all applied in the FS (explicit coords). Moreover,
> the GrGLSLGeometryProcessor::emitTransforms() helper required a valid VS
> var for local coords even when all FPs use explicit coords and wouldn't
> use a local coords var.
>
> Make CPP SkSL code gen for clone copy the explicit coord status of
> children.
>
> Change-Id: Ib8bb36028354405c8012f6e91e9eb46db75d16a6
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/271658
> Commit-Queue: Brian Salomon <bsalomon@google.com>
> Reviewed-by: Brian Osman <brianosman@google.com>
Change-Id: Ia4530e6799019cd92863fe983a2d3c71df6f0620
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/273511
Commit-Queue: Brian Salomon <bsalomon@google.com>
Reviewed-by: Brian Salomon <bsalomon@google.com>
Bug: skia:9935
Change-Id: Id8b851afdd97f8405dbb405e3f142f86dbe1de31
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/273003
Commit-Queue: Jim Van Verth <jvanverth@google.com>
Reviewed-by: Brian Salomon <bsalomon@google.com>
Reviewed-by: Greg Daniel <egdaniel@google.com>
This brings this op into closer correspondence with the other ops.
Change-Id: Idf4f68dcad5492d9ae66c34a9c8a86ff925de26e
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/273492
Reviewed-by: Greg Daniel <egdaniel@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
With the addition of the DDL program pre-compilation we need to know how it is working.
This CL also fixes some threading bugs.
Bug: skia:9455
Change-Id: I20da58a7f1b19685687fae1d159d4e0db8a4964d
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/273001
Reviewed-by: Greg Daniel <egdaniel@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
Change-Id: Id133cb4c97931fb6a8c7d1d6d6b0bb09cbd47d85
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/273486
Commit-Queue: Mike Klein <mtklein@google.com>
Commit-Queue: Mike Reed <reed@google.com>
Auto-Submit: Mike Klein <mtklein@google.com>
Reviewed-by: Mike Reed <reed@google.com>
Change-Id: I0296ff5a68b934f6bc7152a66f57ef045fc94daf
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/272721
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Just the basics of getting existing basic structure in place.
Hoping to flesh this out over the next week or so.
Change-Id: Iba3a3023cc779ffa5fa6f79ebedd53499c076442
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/273323
Commit-Queue: Mike Klein <mtklein@google.com>
Reviewed-by: Herb Derby <herb@google.com>
No color type fallback
Change-Id: I0429491cca1088e877ba2da2ff71ef2219ac9ec7
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/272521
Reviewed-by: Greg Daniel <egdaniel@google.com>
Commit-Queue: Brian Salomon <bsalomon@google.com>
Bug: chromium:1054887
Change-Id: I792a416c7071da75460016574f9aef8832f47e9d
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/273319
Reviewed-by: Mike Klein <mtklein@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Small behavior change: will not check for shader success and program
linking success over the command buffer in a debug build.
Also fail gracefully if stencil renderbuffer allocation fails.
Bug: skia:9938
Bug: chromium:1040186
Change-Id: I623f09d306261d28070078268f6242f92d65fd5f
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/273276
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: Brian Salomon <bsalomon@google.com>
* Rename all the Exclusive things.
Change-Id: If6b5fec5130bf58c396e0a472730efd2ae38c0d2
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/273057
Reviewed-by: Ben Wagner <bungeman@google.com>
Commit-Queue: Herb Derby <herb@google.com>
This debug option was added back in 2011, and the code inside the
#define has not been compilable for many years.
Change-Id: I4366025da6e616f17f75dee14bced2809a90628f
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/273038
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Chris Dalton <csmartdalton@google.com>
Make the ownership explicit by using the hash
table.
Change-Id: I3e1520091e755010c71a9b497af62a64636dc5bd
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/272722
Reviewed-by: Ben Wagner <bungeman@google.com>
Commit-Queue: Herb Derby <herb@google.com>
At the time Chromium is painting, we're passing node IDs
along with painting commands to enable tagging. However,
this assumes that all nodes will end up in the structure
tree, which we might not want.
Instead, allow the client to prune the structure tree
later before telling Skia to generate the PDF, but
keep all of the node IDs to be matched up with.
As an example, suppose the doc looks like this:
root id=1
paragraph id=2
div id=3
text1 id=4
link id=5
text2 id=6
The pruned tree passed to Skia would look like this:
root id=1
paragraph id=2 extra_ids=3,4
link id=5 extra_ids=6
We need to pass the extra node IDs into Skia so
that when content is tagged with id=4, we know to
map that to the paragraph node with id=2 instead.
Note that the resulting PDF document will *not*
have any of these extra IDs, they're all remapped
and consolidated.
While it's not strictly necessary that this is done
in Skia, it's easiest to implement it here. Doing the
same upstream would require replaying an SkPicture
and rewriting all of the node IDs.
Bug: chromium:607777
Change-Id: I0ecb62651e60b84cc5b9d053d7f7d3b9efda1470
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/272462
Reviewed-by: Ben Wagner <bungeman@google.com>
Commit-Queue: Dominic Mazzoni <dmazzoni@chromium.org>
Use absolute error value when doing shader-based interp between all
four subset corner texels.
Add 2x2 image to rectangle_texture GM and draw with all combos of
tile modes.
Change-Id: Ia5856d592b1c9ab8b588301e7b659b05a7439396
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/272920
Reviewed-by: Greg Daniel <egdaniel@google.com>
Commit-Queue: Brian Salomon <bsalomon@google.com>
This is a reland of 29dc430f43
Original change's description:
> Create D3D device and queue
>
> Bug: skia:9935
> Change-Id: Ib6548f413ca3a8befb553d2d47354b400c9162b9
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/272520
> Commit-Queue: Jim Van Verth <jvanverth@google.com>
> Reviewed-by: Greg Daniel <egdaniel@google.com>
Bug: skia:9935
Change-Id: I1c8797e09cdeb3694ea7f47b2236ab7d91d9519f
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/272996
Reviewed-by: Ben Wagner aka dogben <benjaminwagner@google.com>
Commit-Queue: Jim Van Verth <jvanverth@google.com>
This also allows us to remove all the one off Fence code that we
implemented in all the backend TestContexts
Change-Id: I9ff7ba4690cf3f19a180f51fc510991a112bb62c
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/272456
Reviewed-by: Chris Dalton <csmartdalton@google.com>
Commit-Queue: Greg Daniel <egdaniel@google.com>
Previously, all backends were responsible to check for dynamic state
and update it during their internal draw calls. This CL adds
setScissor() and bindTextures() methods to GrOpsRenderTask, and places
this responsibility to update dynamic state on Ganesh instead. It also
replaces the backend-specific "onDrawMeshes" call, with a singular
"onDrawMesh" call that gets called for each mesh.
For now we keep drawMeshes() on GrOpsRenderPass for convenience, but
it will soon be removed and each call site will be responsible to set
its own dynamic state.
Change-Id: If141feae857b22cbf04416557d0c89986eda2a62
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/271976
Commit-Queue: Chris Dalton <csmartdalton@google.com>
Reviewed-by: Brian Salomon <bsalomon@google.com>
Reviewed-by: Jim Van Verth <jvanverth@google.com>
Change-Id: Iaba354865599097d562b29c2ba5e7da8112b22a0
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/272724
Reviewed-by: Jim Van Verth <jvanverth@google.com>
Commit-Queue: Chris Dalton <csmartdalton@google.com>
This reverts commit 29dc430f43.
Reason for revert: This is breaking the Google3 autoroller due to header file d3d12.h not being available on google3.
Original change's description:
> Create D3D device and queue
>
> Bug: skia:9935
> Change-Id: Ib6548f413ca3a8befb553d2d47354b400c9162b9
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/272520
> Commit-Queue: Jim Van Verth <jvanverth@google.com>
> Reviewed-by: Greg Daniel <egdaniel@google.com>
TBR=egdaniel@google.com,jvanverth@google.com,bsalomon@google.com
Change-Id: I3857444cae52cc2338258c46b974ae5496bbaedc
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug: skia:9935
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/272726
Reviewed-by: Leandro Lovisolo <lovisolo@google.com>
Commit-Queue: Leandro Lovisolo <lovisolo@google.com>
The hash table maps descriptors to strikes. So, instead
of searching through the linked list, find the strike
directly.
Change-Id: I53c6fc72c009c35e3476fabd02cee9ee93145c64
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/272465
Commit-Queue: Herb Derby <herb@google.com>
Reviewed-by: Ben Wagner <bungeman@google.com>
Clients will need official access to this class for the compilation iterator.
This CL also hides some of the cruft we don't want exposed.
Bug: skia:9455
Change-Id: I696c058f1c409fb459229552fbbdd935ec112358
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/272643
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
This is a reland of a6cd7c0b1f
Original change's description:
> Multi-threaded strike cache
>
> Allow multiple threads to share the same strike. The old
> system of removing the cache from the linked list is no longer.
> The strikes stay in the list and can be found by other threads.
>
> * Removed strike size verification. There was no way to get the
> locks to work out. The whole point of the change was to have multiple
> threads muting the structure at the same time.
>
> * Strikes are now refed instead of being checked out. Therefore,
> ExclusiveStrikePtr is now just wraps an sk_sp, and should be renamed
> in a future CL.
>
> Change-Id: I832642332a3106e30745f9cdd3156ae72d41fd0b
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/272057
> Reviewed-by: Ben Wagner <bungeman@google.com>
> Commit-Queue: Herb Derby <herb@google.com>
Change-Id: Id02381de93ff82bca58f09e07a457883d57d5565
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/272436
Commit-Queue: Herb Derby <herb@google.com>
Reviewed-by: Ben Wagner <bungeman@google.com>
Our contract is that we will call the finishedCallbacks at some point. So if we
are either abandoning or just deleting the GrGLGpu make sure we call any outstanding
callbacks.
Change-Id: I1425e951185d350a1faa567f0342822c41aafb65
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/272650
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Greg Daniel <egdaniel@google.com>
I've been increasingly concerned about the CPU and memory costs of
creating and holding on to many Compiler instances. Given that they're
use infrequently, having a shared (guarded) instance seems fine.
We can still look at reducing the startup cost of a single instance, but
this removes a large blocker to wider usage of SkRuntimeEffect.
Change-Id: Ia6dc721c73fdf8c9c4c7a8c1af5f350d2c028b22
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/272466
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Fix a compilation error in SkiaDawn by adding an onBindPipeline()
implementation and renaming onDraw() to onDrawMeshes(). onBindPipeline()
does nothing because onDraw() previously did not use the drawBounds
parameter.
Change-Id: Icf3d49296359bfdee8936cc63d6b5d33b597fb4a
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/272511
Commit-Queue: Stephen White <senorblanco@chromium.org>
Reviewed-by: Stephen White <senorblanco@chromium.org>
Bug: skia:9935
Change-Id: Ib6548f413ca3a8befb553d2d47354b400c9162b9
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/272520
Commit-Queue: Jim Van Verth <jvanverth@google.com>
Reviewed-by: Greg Daniel <egdaniel@google.com>
SkSLSlide was triggering this assert - now it prints an error every
frame, but doesn't crash.
Bug skia:9941
Change-Id: I4c02a89c8d824acc71ab595af99e1df2f5fc980d
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/272639
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: Tyler Denniston <tdenniston@google.com>
Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
Auto-Submit: Brian Osman <brianosman@google.com>
This reverts commit ad3b2c9886.
Reason for revert: clang
Original change's description:
> Move SkDeferredDisplayList.h into include\core
>
> Clients will need official access to this class for the compilation iterator.
>
> This CL also hides some of the cruft we don't want exposed.
>
> Bug: skia:9455
> Change-Id: I408c19f9ecd6880a5a7853def591407b0ca43e4e
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/272343
> Commit-Queue: Robert Phillips <robertphillips@google.com>
> Reviewed-by: Brian Salomon <bsalomon@google.com>
TBR=bsalomon@google.com,robertphillips@google.com
Change-Id: Ica80434e7423fb202355eb77a614ece1c4d54726
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug: skia:9455
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/272641
Reviewed-by: Robert Phillips <robertphillips@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
Clients will need official access to this class for the compilation iterator.
This CL also hides some of the cruft we don't want exposed.
Bug: skia:9455
Change-Id: I408c19f9ecd6880a5a7853def591407b0ca43e4e
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/272343
Commit-Queue: Robert Phillips <robertphillips@google.com>
Reviewed-by: Brian Salomon <bsalomon@google.com>
I'm about to add more of these, wanted to land this first.
Change-Id: Ic7ec30f531456075b1f9032f1f07cc54d234af95
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/272527
Auto-Submit: Brian Osman <brianosman@google.com>
Commit-Queue: Ethan Nicholas <ethannicholas@google.com>
Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
