Yet another surprising finding when looking at ARM code generation is
that passing these values to functions by const& does make a difference,
even when fully inlined. I can only guess that the compiler's somehow
more sure that way that the values won't change? Anyway, convert all
skvx functions that take Vec arguments to take const Vec& instead.
This tweak is enough to let the natural implementation of mull()
actually produce good code generation, so I've promoted that to SkVx.h
and added a unit test. Notice in the NEON case we've got a base case at
N=8 and two recursive cases, one down to 8 as usual when N > 8, but also
one up to 8 when N < 8.
This also is another big speedup for ARMv7 NEON, bringing it to nearly
the same speed as ARMv8 NEON on the same device.
Bug: chromium:952502
Change-Id: I0f19bab45cf02222ccc8090053ea2a4a380f1dfe
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/208582
Commit-Queue: Michael Ludwig <michaelludwig@google.com>
Auto-Submit: Mike Klein <mtklein@google.com>
Reviewed-by: Michael Ludwig <michaelludwig@google.com>
This reverts commit e3b110dc6e.
Reason for revert: bot failures
Original change's description:
> align skvx::Vec<N,T> to N*sizeof(T)
>
> This increases the alignment of these vector types. I would have liked
> to keep the alignment minimal, but it's probably no big deal either way.
>
> In terms of code generation, it doesn't make much difference for x86 or
> ARMv8, but it seems hugely important for good ARMv7 NEON code. It's a
> ~10x difference for the bench I've been playing around with that spends
> most of its time in that SkOpts::blit_row_color32 routine.
>
> Bug: chromium:952502
> Change-Id: Ib12caad6b9b3f3f6e821ed70bfb57099db37b15f
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/208581
> Commit-Queue: Michael Ludwig <michaelludwig@google.com>
> Reviewed-by: Michael Ludwig <michaelludwig@google.com>
> Auto-Submit: Mike Klein <mtklein@google.com>
TBR=mtklein@google.com,michaelludwig@google.com
Change-Id: I72357b9775685efcc2cd75db220711c8145b8ac4
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug: chromium:952502
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/208680
Reviewed-by: Mike Klein <mtklein@google.com>
Commit-Queue: Mike Klein <mtklein@google.com>
This increases the alignment of these vector types. I would have liked
to keep the alignment minimal, but it's probably no big deal either way.
In terms of code generation, it doesn't make much difference for x86 or
ARMv8, but it seems hugely important for good ARMv7 NEON code. It's a
~10x difference for the bench I've been playing around with that spends
most of its time in that SkOpts::blit_row_color32 routine.
Bug: chromium:952502
Change-Id: Ib12caad6b9b3f3f6e821ed70bfb57099db37b15f
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/208581
Commit-Queue: Michael Ludwig <michaelludwig@google.com>
Reviewed-by: Michael Ludwig <michaelludwig@google.com>
Auto-Submit: Mike Klein <mtklein@google.com>
This is preparing the ground for having proxies be atomically reffed
Change-Id: Ice48c41120d4ec10a758f696a69028b6363e8a7e
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/207762
Reviewed-by: Greg Daniel <egdaniel@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
Try: out/release/nanobench --match _charToGlyph
Pseudo plan to use this:
- attach to whatever typeface backends need it (probably just freetype)
- have a purge/limiting scheme (e.g. only cache N entries)
- if we care, make the search fancier (e.g. binary, slope, etc.)
Bug: 951647
Change-Id: Ib1042ca5891d2742499faf1314579c402121a855
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/207703
Reviewed-by: Herb Derby <herb@google.com>
Commit-Queue: Mike Reed <reed@google.com>
These replicate the base logic of Sk4px::Wide::div255() and
Sk4px::approxMulDiv255(), and will come in handy replacing them.
No platform specializations yet... want to remind myself what
codegen they get from these vanilla versions first, and then
I'll fill in the platform specific stuff as needed. The tests
should cover everything pretty exhaustively.
Change-Id: I5854d1bc0902a85cbb2351f669c4da7cc31a8775
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/207683
Commit-Queue: Mike Klein <mtklein@google.com>
Commit-Queue: Michael Ludwig <michaelludwig@google.com>
Reviewed-by: Michael Ludwig <michaelludwig@google.com>
Auto-Submit: Mike Klein <mtklein@google.com>
For some reason, Clang can infer <N,T> but GCC can't.
No big deal... we know exactly the ones we want anyway.
Change-Id: I15ba4d4edbd3bc0f37ebe3c2b6e411726cd9fb69
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/207341
Commit-Queue: Mike Klein <mtklein@google.com>
Reviewed-by: Michael Ludwig <michaelludwig@google.com>
Was starting to use this and ran into a few problems with clashing
symbols, namely SI and cast(). Seemed simple enough to not use SI,
and to move all the free-standing types into skvx: skvx::cast,
skvx::shuffle, etc.
Change-Id: Ia5d8ef6d0ae5375bf80d76be88d16f0c9cde56e7
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/207340
Commit-Queue: Mike Klein <mtklein@google.com>
Auto-Submit: Mike Klein <mtklein@google.com>
Reviewed-by: Michael Ludwig <michaelludwig@google.com>
This is rather ham-fisted but I would like to have a short-term fix for the "always-flush" perf regressions.
Change-Id: I508e2d725ac18f50318baf31bc3243fe932c724b
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/206697
Reviewed-by: Greg Daniel <egdaniel@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
This reverts commit 4187ac5e47.
Reason for revert: test if this is causing the angle GM image diffs
Original change's description:
> Skip flush in GrDrawingManager if the specifed proxy doesn't have any work
>
> This is rather ham-fisted but I would like to have a short-term fix for the "always-flush" perf regressions.
>
> Change-Id: Id359ad5a01a290e7e6c06f7ccc1c385ad47d2c06
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/206277
> Reviewed-by: Brian Salomon <bsalomon@google.com>
> Commit-Queue: Robert Phillips <robertphillips@google.com>
TBR=egdaniel@google.com,bsalomon@google.com,robertphillips@google.com
Change-Id: I19ea23a6b8e3985353fc1ebaf974513bb922e9f2
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/206694
Reviewed-by: Robert Phillips <robertphillips@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
This is rather ham-fisted but I would like to have a short-term fix for the "always-flush" perf regressions.
Change-Id: Id359ad5a01a290e7e6c06f7ccc1c385ad47d2c06
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/206277
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
We want GrResourceCache to be able to track which resources are held by
refs (as opposed to pending IOs) so that it can track the affect of
flushing on resource purgeability. Therefore, all cases that can add the
first ref to a GrGpuResource must funnel through GrResourceCache. This
lays the groundwork by restricting initial refs.
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug: skia:8927
Change-Id: I1213c3db258d2412df6666e3222419211ceaa192
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/205482
Commit-Queue: Brian Salomon <bsalomon@google.com>
Reviewed-by: Robert Phillips <robertphillips@google.com>
kSynced: Proxy and GrSurface key kept in sync.
kUnsynced: Proxy and GrSurface keys are unrelated.
This will allow cross-context image generators' lazy instantiation
callbacks to use unique keys to find any pre-existing backing GrTexture
rather than keeping an unref'ed bare pointer to the GrTexture.
Bug: skia:8927
Change-Id: Id15e2a64e8d2e56c4ce70b9399eb1d8bcea6ac9a
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/204723
Commit-Queue: Brian Salomon <bsalomon@google.com>
Reviewed-by: Robert Phillips <robertphillips@google.com>
In the interest of only changing one thing at a time, this should fix the putImageData perf regression in Chrome. It will be followed up by a CL to change prepareSurfaceForExternalIO to always flush (which could have a different perf impact).
On Chrome's putImageData benchmark we get:
w/o this CL: avg 293.1247823176651 runs/s
w/ this CL: avg 374.8427288397461 runs/s
Bug: 942538
Change-Id: I9b9bc752532890f4313a45e5a6aa34b915d5e43d
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/205003
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
There is a lot more cleanup associated w/ removing this but this is the minimum that will un-block Flutter.
Change-Id: I7b8595007f912d9ddbb2c9d7a9989e10101f9c92
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/201647
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
This ensures that we stay in [0,1], except for plus mode, which
requires a larger and more invasive change.
Bug: skia:
Change-Id: I97f6bcea8b10e70e55ba24bcff759ddbb1761794
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/201460
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Guarding the implict constructors and scalar/vector
operations with std::is_convertible ought to make SkVx
types feel more like normal C types, allowing implicit
conversions exactly when the scalar equivalents would.
This shouldn't change the behavior of any code, or make
anything new possible... just nicer to read and write.
Change-Id: Iff4b89012c5b8c7f7933e6841c925b81186bc614
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/201402
Commit-Queue: Mike Klein <mtklein@google.com>
Commit-Queue: Michael Ludwig <michaelludwig@google.com>
Reviewed-by: Michael Ludwig <michaelludwig@google.com>
Auto-Submit: Mike Klein <mtklein@google.com>
This is a reland of 8b40ac35b2
Original change's description:
> Reintroduce deinstantiate lazy proxy types and use for promise images.
>
> This reverts a fraction of b2c5dae65d to
> restore the deinstantiate lazy proxy type, supporting implementation,
> and tests.
>
> Use them for promise images to avoid thread safety issues for promise
> image resources. Makes promise image instantiation callbacks do a thread
> safe unref of their fulfilled GrTexture in GrResourceCache. The
> GrResourceCache mechanism for receiving unref messages is extended to
> allow multiple pending unrefs. All this is new.
>
>
> Bug: skia:8800
> Change-Id: I7b1d4fea13c053b6fbbd39c0c6eaf567b8bf81f1
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/199002
> Reviewed-by: Brian Salomon <bsalomon@google.com>
> Reviewed-by: Robert Phillips <robertphillips@google.com>
> Commit-Queue: Brian Salomon <bsalomon@google.com>
Bug: skia:8800
Change-Id: Ib939fc5c19edf0c6b965c9f6adf0afedd4267703
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/201220
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Brian Salomon <bsalomon@google.com>
Our only use case for mixed samples is stencil-then-cover. This mode
is now handled by AATypeFlags::kMixedSampledStencilThenCover.
Bug: skia:
Change-Id: Id7431cf83ccb20752d1bc85c6ad41efe408e0359
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/200841
Commit-Queue: Chris Dalton <csmartdalton@google.com>
Reviewed-by: Brian Salomon <bsalomon@google.com>
This will turn on explicit allocation (w/o opList sorting) in Chrome. It leaves the old allocation system in place in Android Framework and some of Skia's older bots.
We want:
Flutter and Chrome to always explicitly allocate but not sort opLists outside of DDLs
Android to never explicitly allocate and, thus, automatically never sort opLists
This needs the following Chrome suppression CL to land first:
https://chromium-review.googlesource.com/c/chromium/src/+/15182 (Add flag to skia/config/SkUserConfig.h to unblock Skia roll)
TBR=bsalomon@google.com
Change-Id: I3f51005ebc975ec754c2e0d2c646c0c324b02158
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/200507
Reviewed-by: Robert Phillips <robertphillips@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
This reverts commit 88b93da63d.
Reason for revert: Chrome
Original change's description:
> Always explicitly allocate except in Android Framework (take 2)
>
> This will turn on explicit allocation (w/o opList sorting) in Chrome. It leaves the old allocation system in place in Android Framework and some of Skia's older bots.
>
> We want:
> Flutter and Chrome to always explicitly allocate but not sort opLists outside of DDLs
> Android to never explicitly allocate and, thus, automatically never sort opLists
>
> This cannot land until after the following Chrome CL lands:
>
> https://chromium-review.googlesource.com/c/chromium/src/+/1516096 (Disable opList sorting within Skia)
>
>
> Change-Id: Ic7d6a1a77a08f2fe42324773f62cccf8175ab3d7
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/199931
> Reviewed-by: Brian Salomon <bsalomon@google.com>
> Commit-Queue: Robert Phillips <robertphillips@google.com>
TBR=bsalomon@google.com,robertphillips@google.com
Change-Id: Ida481ee9833d6db366b3d315fb4e9850d7c005ab
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/200506
Reviewed-by: Robert Phillips <robertphillips@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
This reverts commit 2cce805f1f.
Reason for revert: Chrome
Original change's description:
> Always try to reduce opList splitting in DDL contexts/drawingManagers
>
> This may get us in trouble w/ local DDL testing (since we run all our GMs through DDLs). For Chrome this shouldn't yet be a problem (since they are only using DDLs for compositing).
>
> This does mean we're on a tight timeline to land predictive intermediate flushes before Chrome starts using DDLs for rasterization.
>
> Change-Id: I0bb95c075cff3ee49498ff267d76c3a61d16373e
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/199722
> Reviewed-by: Brian Salomon <bsalomon@google.com>
> Commit-Queue: Robert Phillips <robertphillips@google.com>
TBR=bsalomon@google.com,robertphillips@google.com
Change-Id: Idb2dbda1a41844b2541526d504b117fd4cd628cc
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/200505
Reviewed-by: Robert Phillips <robertphillips@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
This may get us in trouble w/ local DDL testing (since we run all our GMs through DDLs). For Chrome this shouldn't yet be a problem (since they are only using DDLs for compositing).
This does mean we're on a tight timeline to land predictive intermediate flushes before Chrome starts using DDLs for rasterization.
Change-Id: I0bb95c075cff3ee49498ff267d76c3a61d16373e
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/199722
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
This reverts commit 8b40ac35b2.
Reason for revert: breaks viz_unittests
Original change's description:
> Reintroduce deinstantiate lazy proxy types and use for promise images.
>
> This reverts a fraction of b2c5dae65d to
> restore the deinstantiate lazy proxy type, supporting implementation,
> and tests.
>
> Use them for promise images to avoid thread safety issues for promise
> image resources. Makes promise image instantiation callbacks do a thread
> safe unref of their fulfilled GrTexture in GrResourceCache. The
> GrResourceCache mechanism for receiving unref messages is extended to
> allow multiple pending unrefs. All this is new.
>
>
> Bug: skia:8800
> Change-Id: I7b1d4fea13c053b6fbbd39c0c6eaf567b8bf81f1
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/199002
> Reviewed-by: Brian Salomon <bsalomon@google.com>
> Reviewed-by: Robert Phillips <robertphillips@google.com>
> Commit-Queue: Brian Salomon <bsalomon@google.com>
TBR=bsalomon@google.com,robertphillips@google.com
Change-Id: Iba960efba4290a284294c62d0470ad7e932c174a
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug: skia:8800
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/200460
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Brian Salomon <bsalomon@google.com>
This will turn on explicit allocation (w/o opList sorting) in Chrome. It leaves the old allocation system in place in Android Framework and some of Skia's older bots.
We want:
Flutter and Chrome to always explicitly allocate but not sort opLists outside of DDLs
Android to never explicitly allocate and, thus, automatically never sort opLists
This cannot land until after the following Chrome CL lands:
https://chromium-review.googlesource.com/c/chromium/src/+/1516096 (Disable opList sorting within Skia)
Change-Id: Ic7d6a1a77a08f2fe42324773f62cccf8175ab3d7
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/199931
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
This reverts a fraction of b2c5dae65d to
restore the deinstantiate lazy proxy type, supporting implementation,
and tests.
Use them for promise images to avoid thread safety issues for promise
image resources. Makes promise image instantiation callbacks do a thread
safe unref of their fulfilled GrTexture in GrResourceCache. The
GrResourceCache mechanism for receiving unref messages is extended to
allow multiple pending unrefs. All this is new.
Bug: skia:8800
Change-Id: I7b1d4fea13c053b6fbbd39c0c6eaf567b8bf81f1
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/199002
Reviewed-by: Brian Salomon <bsalomon@google.com>
Reviewed-by: Robert Phillips <robertphillips@google.com>
Commit-Queue: Brian Salomon <bsalomon@google.com>
Currently doesn't do any clamping, but connects the new config to the
recently added SkColorType. Behavioral changes coming in future CLs.
Bug: skia:
Change-Id: I907396030c435d0aa5931063b3dc6f1b60c661af
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/199980
Reviewed-by: Robert Phillips <robertphillips@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
This reverts commit e157745dfc.
Reason for revert: Too exciting
Original change's description:
> Always explicitly allocate except in Android Framework
>
> This will turn on explicit allocation (w/o opList sorting) in Chrome. It leaves the old allocation system in place in Android Framework and some of Skia's older bots.
>
>
> Change-Id: Idc02985e52f074894a251c7335ef00b009c72ccd
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/199725
> Reviewed-by: Brian Salomon <bsalomon@google.com>
> Commit-Queue: Robert Phillips <robertphillips@google.com>
TBR=bsalomon@google.com,robertphillips@google.com
Change-Id: Icb097844de6db92e8151c81616a758837ecd9dfc
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/199929
Reviewed-by: Robert Phillips <robertphillips@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
This will turn on explicit allocation (w/o opList sorting) in Chrome. It leaves the old allocation system in place in Android Framework and some of Skia's older bots.
Change-Id: Idc02985e52f074894a251c7335ef00b009c72ccd
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/199725
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
We want:
DDLs to always be sorted
Live rendering to still be sorted w/in Skia
Live rendering to still not be sorted in Chrome and Android
Additionally, we want reduced-opList-splitting to only be enabled on some of Skia's bots.
Change-Id: I15e7d69c7e109749665a86a0169ad918c993dc77
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/199244
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
- Transforming a path in place wasn't updating the gen ID of the path
- Transforming a path into another (uniquely held) path wasn't calling
gen ID change listeners.
Bug: skia:
Change-Id: I9e244725d9bd5776d203ce6b12698cee09d0b714
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/199003
Auto-Submit: Brian Osman <brianosman@google.com>
Commit-Queue: Mike Reed <reed@google.com>
Reviewed-by: Mike Reed <reed@google.com>
This reverts commit 9ac0407006.
Reason for revert: Breaking DDL Win10 skpbench bot
Original change's description:
> Distinguish between "flushed" and "finished" idle state callbacks on GrTexture.
>
> This is necessary to convert the promise image API to call Release when all
> work is flushed and Done when all work is complete (future work).
>
> Change-Id: I9745952bb0978ca2aaa79aeed460730b2fea856e
> Bug: skia:8800
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/197163
> Commit-Queue: Brian Salomon <bsalomon@google.com>
> Reviewed-by: Robert Phillips <robertphillips@google.com>
TBR=egdaniel@google.com,bsalomon@google.com,robertphillips@google.com
# Not skipping CQ checks because original CL landed > 1 day ago.
Bug: skia:8800
Change-Id: I5e6c4ea072beb4fb67a53d2ea2b007a7d201799d
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/198603
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Brian Salomon <bsalomon@google.com>
This will allow all the glyph drawing closures to live on the
GrTextBlob.
Other cleanup
* Rename glyphCache things to grStrikeCache.
Change-Id: I03e0353a1434230086a08184221272f1e5751ae6
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/197244
Commit-Queue: Herb Derby <herb@google.com>
Reviewed-by: Mike Klein <mtklein@google.com>
This is necessary to convert the promise image API to call Release when all
work is flushed and Done when all work is complete (future work).
Change-Id: I9745952bb0978ca2aaa79aeed460730b2fea856e
Bug: skia:8800
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/197163
Commit-Queue: Brian Salomon <bsalomon@google.com>
Reviewed-by: Robert Phillips <robertphillips@google.com>
Now that we never re-fulfill a promise image we no longer need to deinstantiate
promise image proxies. They now can use kSingleUse callback semantics.
This was the only usage of the kDeinstantiate lazy callback type so it is
removed. The DeinstantiateProxyTracker is also no longer required and is
removed.
The GrTexture idle callback mechanism now uses GrReleaseProcHelper, which has
been extended to support chaining multiple callbacks together and an abandon()
method that aborts calling the callback in the destructor. It has been renamed
GrRefCntedCallback to reflect its more general usage.
Bug: skia:8800
Change-Id: I857c9eec57fdf706631a266ec8bea682d6657a7c
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/196500
Commit-Queue: Brian Salomon <bsalomon@google.com>
Reviewed-by: Robert Phillips <robertphillips@google.com>
For now this is distinct from kRGBA_F16_SkColorType but treated the
same. Next steps are to see if we can keep it clamped to [0,1].
Switched a few switches away from default to exhaustive.
Took away any explicit SW clamps for now except the one we definitely
want in append_gamut_clamp_if_normalized().
Skip F16Norm in the DDL test because we can't yet distinguish it from
F16.
Change-Id: I021a864fe078e4fa4e2b399982e6c38350e10d74
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/196371
Commit-Queue: Mike Klein <mtklein@google.com>
Commit-Queue: Mike Reed <reed@google.com>
Reviewed-by: Mike Reed <reed@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Auto-Submit: Mike Klein <mtklein@google.com>
Prior to this CL, we always used raster-pipeline if the device had a
colorspace. In this CL, we defer to individual shaders to decide if they can still
use the legacy path (onMakeContext).
The motivation is purely performance (and perhaps short-term backwards
compat): the legacy blits are usually faster, so staying in them
(as long as we still draw correctly) is preferable.
Causes some (mostly unimportant) rebaselines, esp. around bilerp, as the
rasterpipeline does a much better (but slower) job of it, and now we will
use the legacy bilerp more often.
Bug: skia:8793
Change-Id: I1e7e482a863a1a09ffef86a87e2aa9a4a8d1b9b5
Reviewed-on: https://skia-review.googlesource.com/c/195888
Commit-Queue: Mike Reed <reed@google.com>
Reviewed-by: Mike Klein <mtklein@google.com>
Use SkDEBUGCODE(...;) instead of SkDEBUGCODE(...);
clang-format has trouble formatting the new spelling, but
it's already commonly used in skia, and if you have something
SkDEBUGCODE instead of just checking `#if SK_DEBUG` then
there isn't all that much you can do.
(You could make SkDEBUGCODE() insert a static_assert(true, "")
at the end to eat semicolons after it, but that's also
a bit gross. And many SkDEBUGCODE()s already put the ; inside
the parens.)
Bug: chromium:926235
Change-Id: I57bc800309283b7206082d8a7acc4b462a3b4cfa
Reviewed-on: https://skia-review.googlesource.com/c/196421
Reviewed-by: Mike Klein <mtklein@google.com>
Commit-Queue: Mike Klein <mtklein@google.com>
Auto-Submit: Nico Weber <thakis@chromium.org>
Bug: skia:
Change-Id: I9bab3ff73d3934786d7457c4b1bcf67d01c653f3
Reviewed-on: https://skia-review.googlesource.com/c/196060
Commit-Queue: Mike Reed <reed@google.com>
Reviewed-by: Mike Klein <mtklein@google.com>
Since these are all already static, it doesn't have any real functional
impact in terms of linking or codegen. But it does supress unused
function warnings in compilation units that don't use everything.
Add a new SI boilerplate macro to go along with SINT and SIT.
Change-Id: If2c09951b7453338dd20a3a88e3abbee5eefcd27
Reviewed-on: https://skia-review.googlesource.com/c/195921
Commit-Queue: Mike Klein <mtklein@google.com>
Reviewed-by: Michael Ludwig <michaelludwig@google.com>
