Introduce three calls on GrOp: Make, MakeWithExtraMemory,
and MakeWithProcessorSet. Instead of returning
unique_ptr<GrOp>, they return a type of GrOp::OpOwner.
GrOp::OpOwner safely deletes the op when it goes out
of scope for either new/delete or GrOpMemoryPool
allocations.
In order to make the code easier to refactor, I
eliminated MakeArg from the helpers.
Change-Id: Icfd697906f3147a8734575d08bd7195e7517383a
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/323778
Reviewed-by: Michael Ludwig <michaelludwig@google.com>
Commit-Queue: Herb Derby <herb@google.com>
This CL adds a new type GrDstSampleType to say how we will sample the dst.
We add tracking of the GrDstSampleType in the recording of GrOps and
then during execution passing the information along to the GrPipeline.
In general the tracking of GrDstSampleType is a global state of a GrOpsTask
so it is kept separate fro the DstProxyView which is more specific to a
single Op on the GrOpsTask.
Bug: skia:10409
Change-Id: Ie843c31f2e48a887daf96cee99ed159b196cb545
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/315645
Commit-Queue: Greg Daniel <egdaniel@google.com>
Reviewed-by: Michael Ludwig <michaelludwig@google.com>
Reviewed-by: Robert Phillips <robertphillips@google.com>
Mechanically updated via Xcode "Replace Regular Expression":
typedef (.*) INHERITED;
-->
using INHERITED = $1;
The ClangTidy approach generated an even larger CL which would have
required a significant amount of hand-tweaking to be usable.
Change-Id: I671dc9d9efdf6d60151325c8d4d13fad7e10a15b
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/314999
Commit-Queue: Mike Klein <mtklein@google.com>
Reviewed-by: Mike Klein <mtklein@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
https://clang.llvm.org/extra/clang-tidy/checks/modernize-use-override.html
Adds override (introduced in C++11) to overridden virtual functions and
removes virtual from those functions as it is not required.
virtual on non base class implementations was used to help indicate to
the user that a function was virtual. C++ compilers did not use the
presence of this to signify an overridden function.
Change-Id: If66d8919358f72a4035190caf8d7569268037a9a
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/310160
Commit-Queue: Mike Klein <mtklein@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Mike Klein <mtklein@google.com>
Profiling showed that repeated calls to `input_texel_color` were
actually a very expensive part of the test. Fortunately, we can expose
the texture's pixel buffer to the unit test method and easily reclaim
that performance.
Change-Id: I2c1cdd57a3e14dc859bdf03d8131137ca81364ff
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/309437
Commit-Queue: John Stiles <johnstiles@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
This is a net reduction in code size because this idea was already
implemented separately in two places:
- dump_fragment_processor_tree (GrProcessorSet)
- describe_fp (ProcessorTest)
This consolidates the implementations. This CL also fixes a handful of
dumpInfo() methods that were not sufficiently descriptive--e.g. the FP
name was missing, or the implementation was just buggy.
Change-Id: If34ac46c97e9ae431c7c64b1247fc619703580b2
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/309324
Commit-Queue: Brian Salomon <bsalomon@google.com>
Reviewed-by: Brian Salomon <bsalomon@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
This reverts commit d169e1915c.
Reason for revert: broke chrome via code generator
Original change's description:
> Migrate GrSurfaceContext readPixels to take direct context
>
> After this lands we'll proceed up the stack and add the direct
> context requirement to the public API and SkImage.
>
> Bug: skia:104662
> Change-Id: I4b2d779a7fcd65eec68e631757821ac8e136ddba
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/309044
> Commit-Queue: Adlai Holler <adlai@google.com>
> Reviewed-by: Robert Phillips <robertphillips@google.com>
TBR=robertphillips@google.com,adlai@google.com
Change-Id: I6126f2dca4bc902c903512ac486e22841cc472e5
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug: skia:104662
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/309281
Reviewed-by: Adlai Holler <adlai@google.com>
Commit-Queue: Adlai Holler <adlai@google.com>
After this lands we'll proceed up the stack and add the direct
context requirement to the public API and SkImage.
Bug: skia:104662
Change-Id: I4b2d779a7fcd65eec68e631757821ac8e136ddba
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/309044
Commit-Queue: Adlai Holler <adlai@google.com>
Reviewed-by: Robert Phillips <robertphillips@google.com>
This is causing failures on the status dashboard.
Change-Id: I29e8fb5bef72282dbe6fafb5402607ed48aae707
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/309136
Commit-Queue: John Stiles <johnstiles@google.com>
Commit-Queue: Greg Daniel <egdaniel@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Greg Daniel <egdaniel@google.com>
Opted the Galaxy S20 out of the test to allow Tryjobs to pass.
Change-Id: I8d4637a23f36edb012c96b7059b184c231c0436d
Bug: skia:10384, skia:10595
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/309119
Auto-Submit: John Stiles <johnstiles@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Previously, MakeChildFP avoided infinite recursion by rejecting any FP
that took inputs. MakeChildFP now generates random inputs up to a
user-supplied tree depth.
The ProcessorOptimizationValidationTest test has been updated to
test up to a tree depth of 3. The ProcessorCloneTest has been left at
a tree depth of 1 due to a bug that only appears on Galaxy S20/Mali G77.
The Mali bug doesn't appear to be related to FP cloning, but probably
deserves further analysis. (It appears that on this device, these
processors hooked together in sequence render a tiny bit differently
each time: DitherEffect -> RectBlurEffect -> ImprovedPerlinNoise. By
visual inspection it looks like the dither varies on each draw.)
Change-Id: Ib8f619eb7a8a9c9254080303504c20065ff35453
Bug: skia:10384, skia:10595
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308556
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Previously, this test would synthesize a random FP, call clone(), render
both, and the compare the results and report differences.
Now, if a difference is discovered, this test will re-synthesize the
originally-created random FP from scratch without using clone().
Instead, we call TestCreate again using the same random seed.
Then we can render and compare that output against the original as well.
This will allow to better differentiate failures that were actually
caused by clone(), versus failures caused by other types of
inconsistency.
If the regenerated version still mismatches, there are a variety of
potential explanations:
- the FP's TestCreate() does not always generate the same FP from a
given seed
- the FP's Make() does not always generate the same FP when given the
same inputs
- the FP itself generates inconsistent pixels (shader UB?)
- the driver has a bug
Change-Id: I71701c0f3d33a08f6ee926313782620487d336bd
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/309076
Auto-Submit: John Stiles <johnstiles@google.com>
Commit-Queue: Brian Salomon <bsalomon@google.com>
Reviewed-by: Brian Salomon <bsalomon@google.com>
This fixes a large number of SkSL namespaces which were labeled as if
they were anonymous, and also a handful of other mislabeled namespaces.
Missing namespace-end comments have been added throughout.
A number of diffs are just indentation-related (adjusting 1- or 3-
space indents to 2-space).
Change-Id: I6c62052a0d3aea4ae12ca07e0c2a8587b2fce4ec
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308503
Commit-Queue: John Stiles <johnstiles@google.com>
Reviewed-by: Mike Klein <mtklein@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Change-Id: Ic7799b3c5f4294cba9ff72f8c11a2ad285ab189f
Bug: skia:10217
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/304738
Commit-Queue: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Salomon <bsalomon@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Change-Id: Ia2cfbca8982b57399b6681cbb4501c2933ab4df7
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/304576
Auto-Submit: Brian Salomon <bsalomon@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: Brian Salomon <bsalomon@google.com>
The current algorithm runs an exponentially-increasing number of trials
based on the number of children supported by the fragment processor and
has become a large drag on test times. This version runs a fixed number
of trials to determine which optimization bits are able to appear, and
then continues running trials until each potential optimization has been
demonstrated successfully five times.
The algorithm doesn't attempt to check interactions between the various
optimization bits (e.g. a hypothetical bug that might only occur when
two optimizations interact with one another) but hopefully the minimum
of 100 successful trials is enough to shake out most issues.
Change-Id: I4eba7ace84739027a5aea8f8f895b44c4532b816
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/304059
Commit-Queue: John Stiles <johnstiles@google.com>
Reviewed-by: Greg Daniel <egdaniel@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Previously, this test layered its paints by calling
`addColorFragmentProcessor` multiple times. We intend to remove support
for multiple color FPs on a GrPaint in the near future, so we use input
fragment processors instead to generate the same result.
Change-Id: I33e82ce0067183189e69b2af0fe0c228d1d60f14
Bug: skia:10217
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/303479
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Salomon <bsalomon@google.com>
The helper class allows us to regenerate a "random" fragment processor
with the same seed as many times as desired. This lets us check the
`compatibleWithCoverageAsAlpha` optimization without needing to clone
the input FP; instead, we can actually generate the FP under test three
times in a row, with the same random seed.
In a followup CL (http://review.skia.org/303479/) we will also leverage
this helper class to regenerate the FP under test with the same random
seed, but with a different input.
Change-Id: I1cd83082a949d555f7898970c8a1cc3002818286
Bug: skia:10384
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/303657
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
This simplifies things like ConstantOutputForConstantInput and
invokeChild. It also removes the need for child indices: generated
FPs now directly refer to their children by slot number.
Change-Id: I69bbb042d5d72d21b999256f969c467702d0774d
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/302436
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: Brian Salomon <bsalomon@google.com>
Reviewed-by: John Stiles <johnstiles@google.com>
Bug: skia:10416
Change-Id: I0ca7535a0e6507e6b2a9f4682788826972c5f3b8
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/300296
Reviewed-by: Michael Ludwig <michaelludwig@google.com>
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
This CL makes it explicit that the unit tests always get a direct context.
It is mainly a mechanical CL.
Change-Id: I49e0628851d9c81eb47386ef978edf905c6469d8
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/299866
Reviewed-by: Adlai Holler <adlai@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
This removes the kMixed type of SkSL::SampleMatrix. All analysis of FP
sampling due to parent-child relationships is tracked in flags on
GrFragmentProcessor now.
The sample strategy is tracked as follows:
- An FP marks itself as using the local coordinate builtin directly (automatically done for .fp code based on reference to sk_TransformedCoords2D[0]).
- This state propagates up the parent towards the root, marking FPs as using coordinates indirectly. We stop the propagation when we hit a parent FP that explicitly samples the child because it becomes the source of the child's coordinates.
- If that parent references its local coordinates directly, that kicks off its own upwards propagation.
- Being sampled explicitly propagates down to all children, and effectively disables vertex-shader evaluation of transforms.
- A variable matrix automatically marks this flag as well, since it's essentially a shortcut to (matrix expression) * coords.
- The matrix type also propagates down, but right now that's only for whether or not there's perspective.
- This doesn't affect FS coord evaluation since each FP applies its action independently.
- But for VS-promoted transforms, the child's varying may inherit perspective (or other more general matrix types) from the parent and switch from a float2 to a float3.
- A SampleMatrix no longer tracks a base or owner, GrFragmentProcessor exposes its parent FP. An FP's sample matrix is always owned by its immediate parent.
- This means that you can have a hierarchy from root to leaf like: [uniform, none, none, uses local coords], and that leaf will have a SampleMatrix of kNone type. However, because of parent tracking, the coordinate generation can walk up to the root and detect the proper transform expression it needs to produce, and automatically de-duplicate across children.
Currently, all FP's that are explicitly sampled have a signature of (color, float2 coord). FP's that don't use local coords, or whose coords are promoted to a varying have a signature of (color).
- In this case, the shader builder either updates args.fLocalCoords to point to the varying directly, or adds a float2 local to the function body that includes the perspective divide.
GrFragmentProcessor automatically pretends it has an identity coord transform if the FP is marked as referencing the local coord builtin. This allows these FPs to still be processed as part of GrGLSLGeometryProcessor::collectTransforms, but removes the need for FP implementations to declare an identity GrCoordTransform.
- To test this theory, GrTextureEffect and GrSkSLFP no longer have coord transforms explicitly.
- Later CLs can trivially remove them from a lot of the other effects.
- The coord generation should not change because it detects in both cases that the coord transform matrices were identity.
GrGLSLGeometryProcessor's collectTransforms and emitTransformCode has been completely overhauled to recurse up an FP's parent pointers and collect the expressions that affect the result. It de-duplicates expressions between siblings, and is able to produce a single varying for the base local coord (either when there are no intervening transforms, or the root FP needs an explicit coordinate to start off with).
This also adds the fp_sample_chaining GM from Brian, with a few more configurations to fill out the cells.
Bug: skia:10396
Change-Id: I86acc0c34c9f29d6371b34370bee9a18c2acf1c1
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/297868
Commit-Queue: Michael Ludwig <michaelludwig@google.com>
Reviewed-by: Brian Salomon <bsalomon@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Bug: skia:10396
Change-Id: I0c117ab4d95737b76dec5bce16103b9058218fb8
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/297065
Commit-Queue: Michael Ludwig <michaelludwig@google.com>
Reviewed-by: Brian Salomon <bsalomon@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Previously the test checked two clones against each other. It seems
more useful to test an original against a clone.
Change-Id: Ie49a39829a0f5ee24724e4acefad50e539104f48
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/297460
Commit-Queue: Brian Salomon <bsalomon@google.com>
Reviewed-by: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Also remove unused buffers. FPs no longer support reading from buffers.
Bug: skia:10139
Change-Id: Id1e2d7ef2cfa7f11a95cb0ce448c69af755c2b65
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/297176
Reviewed-by: Robert Phillips <robertphillips@google.com>
Commit-Queue: Brian Salomon <bsalomon@google.com>
Previously, when REPORTER_ASSERT reported a failure, there was no
additional information describing the FP that could not be cloned in the
error log. Now, we print out a very simple tree of the FP and its
children.
Change-Id: I141cfb17ca431864a6f555d56f0335293f259c4e
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/296452
Reviewed-by: Greg Daniel <egdaniel@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
Previously, the GrAllocator header defined three types: GrAllocator,
GrTAllocator, and GrSTAllocator. GrAllocator was generic, and would
manage the blocks of memory (fixed size to hold N * sizeof(item)). It
stored an SkTArray of pointers to each block. GrTAllocator wrapped
the GrAllocator to add template support for a specific T. GrSTAllocator
extended GrTAllocator to include a template arg for inline storage.
In this CL, GrAllocator is no longer defined, and its memory functionality
is replaced with the use of GrBlockAllocator. For the most part, the
implementation of GrTAllocator on top of GrBlockAllocator remains the
same, although there is explicit array to the block pointers so indexing
is slightly different. GrSTAllocator is also removed entirely, so that
GrTAllocator's template includes initial storage size.
The iteration over the allocated items
is updated to wrap GrBlockAllocator's block iterator, and then iterate
by index within each block. Documentation on GrAllocator already recommended
using the iterator instead of random access, and these changes further
reinforce it. It turns out that many of the use cases of GrAllocator
were written to use random access, so many of the files outside of
GrAllocator.h have just been updated to use the new for-range iteration
instead of a random access for loop.
Change-Id: I28b0bc277c323fd7035d4a8442ae67f058b2b64c
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/272058
Commit-Queue: Michael Ludwig <michaelludwig@google.com>
Reviewed-by: Chris Dalton <csmartdalton@google.com>
Also strengthens/adds some guarantees about this function:
* Always returns the same image if the original is texture-backed and
compatible with GrMipMapped (WRT HW MIP support)
* If a new texture backed image is returned it is always an uncached
texture that is not shared with another image or owned by an image
generator.
Adds a GrImageTexGenPolicy that allows control through image/bitmap
GrTextureProducers of whether a new texture must be made and whether
that texture should be budgeted or not.
Increases unit test coverage of this API.
Bug: skia:8669
Change-Id: Ifc0681856114a08fc8cfc57ca83d22efb1c1f166
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/274938
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: Brian Salomon <bsalomon@google.com>
This CL moves the boiler-plate implementation of onPrePrepareDraws up to GrMeshDrawOp. This, unfortunately, required the addition of a 'programInfo' virtual but that may go away in the future.
The only GrMeshDrawOp-derived classes left are:
AAHairlineOp - which has 3 programInfos
AAFlatteningConvexPathOp - which has >=1 mesh (but just 1 programInfo)
GrAtlasTextOp - which has inline uploads
SmallPathOp - which, maybe, has inline uploads
TextureOp - which has chaining
Bug: skia:9455
Change-Id: Id10f70e764054134751545ad38b99f0a0778de76
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/276642
Reviewed-by: Greg Daniel <egdaniel@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
This CL:
Renames the existing createProgramInfo method to onCreateProgramInfo and makes it a virtual on GrMeshDrawOp
Moves the non-virtual createProgramInfo (helper) calls to GrMeshDrawOp
Changes onCreateProgramInfo to not return a ProgramInfo*. This is setting up to handle ops that create >1 programInfo (e.g., AAHairlineOp)
This CL leaves the following 8 ops in need of an onCreateProgramInfo implementation:
AAHairlineOp
AAFlatteningConvexPathOp
GrAtlasTextOp
GrShadowRRectOp
SmallPathOp
GrTextureOp
PrimitiveProcessorTest::Op
VertexColorSpaceBench::Op
Bug: skia:9455
Change-Id: Id7e2b8a40cac86ede6bf3c0e544da5500ff47d8a
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/276403
Commit-Queue: Robert Phillips <robertphillips@google.com>
Reviewed-by: Greg Daniel <egdaniel@google.com>
This reverts commit e5a06ce678.
Reason for revert: Need to make change in google3 first
Original change's description:
> Move GrGpuResource GrSurface and GrTexture into src.
>
> Must land https://chromium-review.googlesource.com/c/chromium/src/+/2087980
> before this can land.
>
> Bug: skia:7966
> Change-Id: I60bbb1765bfbb2c96b2bc0c9826b6b9d57eb2a03
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/275077
> Commit-Queue: Greg Daniel <egdaniel@google.com>
> Reviewed-by: Robert Phillips <robertphillips@google.com>
TBR=egdaniel@google.com,bsalomon@google.com,robertphillips@google.com
Change-Id: Id39e0a351e49a87209de88a6ad9fadb0219db72c
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug: skia:7966
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/275216
Reviewed-by: Greg Daniel <egdaniel@google.com>
Commit-Queue: Greg Daniel <egdaniel@google.com>
The only thing that affects the view returned is whether it ought to
be MIP mapped or not. So don't take a whole GrSamplerState.
The view() function won't ever change the colortype so just query
GrTextureProducer rather than having view() return a tuple.
The rest is transitively reaching through callers and callees of
GrTextureProducer::view() to only pass filter or GrMipMapped instead of
GrSamplerState. Also, some params that indicate whether MIPs are
requested are changed from bool to GrMipMapped. And some minor style
stuff (mainly de-yoda-ifying GrMipMapped checks, using
GrSurfaceProxyView operator bool()).
Change-Id: Ia184aa793cf51d42642ea3bb0521ce06da2efb10
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/274205
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: Brian Salomon <bsalomon@google.com>
Replace with SkISize.
Also change some const SkISize& params to just SkISize.
Change-Id: I3c72d961662eefeda545fba17d63e877cd5ca813
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/269374
Reviewed-by: Greg Daniel <egdaniel@google.com>
Commit-Queue: Brian Salomon <bsalomon@google.com>
Use std::min and std::max everywhere.
SkTPin still exists. We can't use std::clamp yet, and even when
we can, it has undefined behavior with NaN. SkTPin is written
to ensure that we return a value in the [lo, hi] range.
Change-Id: I506852a36e024ae405358d5078a872e2c77fa71e
Docs-Preview: https://skia.org/?cl=269357
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/269357
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: Mike Reed <reed@google.com>
Reviewed-by: Brian Salomon <bsalomon@google.com>
Instead in proxyProvider we just have a create a bitmap call which
does no special fallback or logic. All the callers now go through
GrBitmapTextureMaker which handles and special fallbacks or caching
support that we need.
Change-Id: I71bb896cc78f64f9d6d54b54af2490d48e0f5af5
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/266842
Commit-Queue: Greg Daniel <egdaniel@google.com>
Reviewed-by: Brian Salomon <bsalomon@google.com>
