This ties the caps to the compiler instance, paving the way for
pre-optimizing the shared code. Most of the time, the compiler is
created and owned the GPU instance, so this is fine. For runtime
effects, we now use the shared (device-agnostic) compiler instance
for the first compile, even on GPU. It's configured with caps that
apply no workarounds. We pass the user's SkSL to the backend as
cleanly as possible, and then apply any workarounds once it's part
of the full program.
Bug: skia:10905
Bug: skia:10868
Change-Id: Ifcf8d7ebda5d43ad8e180f06700a261811da83de
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/331493
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: Brian Salomon <bsalomon@google.com>
Reviewed-by: John Stiles <johnstiles@google.com>
Bug: skia:9000
This constructor is already declared; provide a definition.
Add a test.
Change-Id: Ic54ccffe0b1adf7c8c411e19a3fc9d825a5c9619
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/331799
Commit-Queue: Leon Scroggins <scroggo@google.com>
Reviewed-by: Mike Reed <reed@google.com>
RenderPassEncoder::SetIndexBuffer() is deprecated.
Change-Id: Ie7b357df8bbee16fd08fa1bdf5e9cbedf11b259d
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/331495
Reviewed-by: Greg Daniel <egdaniel@google.com>
Commit-Queue: Stephen White <senorblanco@google.com>
There's really no reason for these classes to be separate at this
point. Also extracts a "GrStrokeOp" base class that has the
functionality that will be shared with indirect stroking.
Bug: skia:10419
Change-Id: I960d5e6d64f0814ccb4a3852bc627af2b8082a1f
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/331860
Reviewed-by: Greg Daniel <egdaniel@google.com>
Commit-Queue: Chris Dalton <csmartdalton@google.com>
Previously, temp variables created by sample() calls were named after
the offset of the sample() call within the code. This was
straightforward but would fail if the sample() call were duplicated via
inlining of helper functions.
FP sample() temp variables are now named using a counter, starting from
zero and counting upwards.
Change-Id: I16f9a3426117677c0df13d15772320def99cc0d6
Bug: skia:10858
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/331415
Commit-Queue: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
This will allow ASAN to detect use-after-free errors in pooled memory,
enabling our fuzzers to catch errors sooner.
Testing with oss-fuzz:26942 : http://screen/C5TEbu3CJvHzRqA
Change-Id: Ic47d6b043998e5069525490cd25b2390cad94360
Bug: skia:10885
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/331482
Reviewed-by: Michael Ludwig <michaelludwig@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Previously, when a prototype was parsed, this added a function
declaration to the symbol table, but the prototype itself was not
re-emitted during code generation. This meant that the final code might
not be valid, since the absence of prototypes meant that the code might
attempt to invoke a function before its declaration. Now, prototypes are
stored in the ProgramElement list and re-emitted during code generation
for GLSL/Metal/CPP. (SPIR-V doesn't name its functions at all.)
Change-Id: I76446c796000eb0b56f964d82457122182c28b87
Bug: skia:10872
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/331136
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
f01a9d9020
is the culprit
Revert "GrRefCntedCallback has Make function."
This reverts commit b2c42140ea.
Revert "Add SkImage::MakeFromYUVATexturesCopyToExternal"
This reverts commit f01a9d9020.
Bug: skia:10632
Change-Id: Ief076f168b63ff8ca15b607163a13d5f52a733d2
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/331798
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Brian Salomon <bsalomon@google.com>
If a glyph's position is > 32K or < -32K, then it is not
on the device. Don't bother adding it to the GrOp. This
reduces the glyph position from SkIPoint to
Vec<2, int16_t>.
But, if some of the glyphs are dropped, and the blob is
scrolled, then the glyphs may be in range. Add a flag
to track if glyphs were dropped, and force a redraw if
not exactly the same matrix is used to draw.
In addition,
* Rename VertexData to DevicePosition
* Add needed calls to GrGlyphRect
Change-Id: Ifdb8aa86d3da0b69d46c11fce5f33803781d880c
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330622
Commit-Queue: Herb Derby <herb@google.com>
Reviewed-by: Ben Wagner <bungeman@google.com>
This reverts commit 28eb592735.
Reason for revert: std::optional use is breaking bots, e.g.:
https://ci.chromium.org/raw/build/logs.chromium.org/skia/4fa8175011a1c411/+/annotations
Original change's description:
> Reland "heif: Add AVIF type and AVIF sniffing"
>
> This is a reland of 3ab5b73649
>
> Fix: Chromium does not use the heif decoder. So hide it behind
> an ifdef.
>
> Original change's description:
> > heif: Add AVIF type and AVIF sniffing
> >
> > AVIF is the image format based on the AV1 video codec. The
> > container for AVIF is very similar to that of HEIF. Add type
> > definitions for AVIF and sniffing code for detecting AVIF images.
> >
> > The underlying android platform's HEIF decoder implementation will
> > also support AVIF decoding.
> >
> > Bug: b/141654151
> > Change-Id: I7e31f4cedf0bffb8920ddf880a26601e48d0e833
> > Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330059
> > Reviewed-by: Leon Scroggins <scroggo@google.com>
> > Reviewed-by: Derek Sollenberger <djsollen@google.com>
> > Reviewed-by: Chong Zhang <chz@google.com>
> > Commit-Queue: Leon Scroggins <scroggo@google.com>
>
> Bug: b/141654151
> Change-Id: Ie6840d6fc97789be619924fc60df8683cd58430f
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/331496
> Reviewed-by: Derek Sollenberger <djsollen@google.com>
> Reviewed-by: Leon Scroggins <scroggo@google.com>
> Commit-Queue: Leon Scroggins <scroggo@google.com>
TBR=djsollen@google.com,scroggo@google.com,chz@google.com,vigneshv@google.com
Change-Id: I25df9f65b9d56d81be70a5fa650c3d841bd5a15e
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug: b/141654151
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/331800
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Brian Salomon <bsalomon@google.com>
Previously, if this was a DF-based text op and the shader had local coords, onPrepare() and setupDfProcessor() would both invert the same matrix.
Change-Id: I1cf989045791aabf9223728d485ce09ffb6412bb
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330939
Commit-Queue: Michael Ludwig <michaelludwig@google.com>
Reviewed-by: Herb Derby <herb@google.com>
Reviewed-by: Brian Salomon <bsalomon@google.com>
When eliminating a CFG node, we now flag its exit nodes; if our
optimization pass reaches one of those flagged nodes, we stop the
current optimization process in its tracks and initiate a rescan.
We do NOT recursively mark the exits of the exit nodes, so this fix is
reliant on the CFG being ordered in a non-chaotic fashion, but in
practice this seems to be sufficient for the CFGs we generate today.
Change-Id: I892805361c5f4297e02146f37a759dfda83f5488
Bug: oss-fuzz:26942
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/331597
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
This is a reland of 3ab5b73649
Fix: Chromium does not use the heif decoder. So hide it behind
an ifdef.
Original change's description:
> heif: Add AVIF type and AVIF sniffing
>
> AVIF is the image format based on the AV1 video codec. The
> container for AVIF is very similar to that of HEIF. Add type
> definitions for AVIF and sniffing code for detecting AVIF images.
>
> The underlying android platform's HEIF decoder implementation will
> also support AVIF decoding.
>
> Bug: b/141654151
> Change-Id: I7e31f4cedf0bffb8920ddf880a26601e48d0e833
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330059
> Reviewed-by: Leon Scroggins <scroggo@google.com>
> Reviewed-by: Derek Sollenberger <djsollen@google.com>
> Reviewed-by: Chong Zhang <chz@google.com>
> Commit-Queue: Leon Scroggins <scroggo@google.com>
Bug: b/141654151
Change-Id: Ie6840d6fc97789be619924fc60df8683cd58430f
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/331496
Reviewed-by: Derek Sollenberger <djsollen@google.com>
Reviewed-by: Leon Scroggins <scroggo@google.com>
Commit-Queue: Leon Scroggins <scroggo@google.com>
Change-Id: Ief57d9c102b3c7658738920cdf54ccd4d21c5c5e
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/331656
Commit-Queue: Brian Osman <brianosman@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: Brian Osman <brianosman@google.com>
Reviewed-by: John Stiles <johnstiles@google.com>
Change-Id: I16a5938f5156fac2e93b8cd0b2a8e2ed45b46386
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330938
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Michael Ludwig <michaelludwig@google.com>
Each caller does not need to check for null function.
Also inherit from SkNVRefCnt.
Change-Id: I9a53c3c8e9dec1361db6cfb21b97d178173b23ac
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/331490
Reviewed-by: Robert Phillips <robertphillips@google.com>
Commit-Queue: Brian Salomon <bsalomon@google.com>
This brings back the 2px wide picture frame approach described in these
slides: go/thin-line-aa. This has been in place when the per-edge-aa
only needed to support rectangles but was scrapped when the code path
was updated to support arbitrary quadrilaterals.
I opted to have the GrQuadPerEdgeAA logic check for degeneracy and
update what it requests for the outset. This scale factor out to a 2px
wide shape makes sense in the context of anti-aliasing, but not so for
the generalized inset/outset logic defined in GrQuadUtils. It would have
been more efficient to implement it there, but would have locked it in
to being just AA inset/outset.
I also updated SkGpuDevice's drawStrokedLine to construct the quad of
the line directly, and to always turn a line path into a rect, instead
of restricting it based on matrix or stroked width. With this new change
the quality of the fill rect is much higher under rotations and
perspective compared to the hairline.
See rect case: https://drive.google.com/file/d/1xwgG5heADcdXYShsDodgbuv2tbHfuBNt/view?usp=sharing
Hairline case: https://drive.google.com/file/d/1duNLxiYLLJhsJ94Uc01rSxjB4ar6_Ud9/view?usp=sharing
Bug: chromium:820987
Change-Id: Ibd58b89a467ad5a61c5479d11259024259f1bb47
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329418
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Michael Ludwig <michaelludwig@google.com>
Bug: skia:10828
Change-Id: Ifab9c2b179fa981a94d03ff96718c3b31f745fee
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/331338
Reviewed-by: Jim Van Verth <jvanverth@google.com>
Commit-Queue: Greg Daniel <egdaniel@google.com>
Updated API for creating a RGBA texture-backed image from RGBA texture
planes with user-provided backing store using GrYUVABackendTextures.
Ultimately we would like to remove all such APIs and have the client
make a SkSurface and draw a YUVA texture image to the surface but
a recent attempt to do that in Chrome caused a not yet understood
perf regression.
Add wacky_yuv_formats variation that tests new api.
Bug: skia:10632
Change-Id: I89411216948682f13281a91a7575d5f345badda7
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329956
Commit-Queue: Brian Salomon <bsalomon@google.com>
Reviewed-by: Robert Phillips <robertphillips@google.com>
We need to abandon all the GPU resources before dropping any
refs the thread-safe cache may be holding.
Change-Id: Id1a06adf9e0241bfaf55e3f58bf8c2db928ea141
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/331536
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
This method is only valid in the range 2^(+/-30) due to fp32 overflow.
Adds a comment to the function and updates its test.
TBR=bsalomon@google.com
Change-Id: Ifa2fc0ed4a7f9123f0bebaa02c666c61e06e62a6
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/331481
Reviewed-by: Chris Dalton <csmartdalton@google.com>
Commit-Queue: Chris Dalton <csmartdalton@google.com>
This method finds the locations a cubic needs to be chopped at before
it can be passed to the stroke tessellation shader. It's an integral
part of CPU stroke preparation and therefore extremely perf sensitive.
Bug: skia:10419
Change-Id: Ib23c2583b8cfc78814ce52425f7af2c8b2f8b420
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330314
Commit-Queue: Chris Dalton <csmartdalton@google.com>
Reviewed-by: Michael Ludwig <michaelludwig@google.com>
Bug: chromium:1139750, skia:8389
Cq-Include-Trybots: luci.chromium.try:linux-blink-rel,win10-blink-rel
Change-Id: I69c55f505947fdec5d9d391d2b2d2d3ff6dec9b8
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330216
Reviewed-by: Jim Van Verth <jvanverth@google.com>
Reviewed-by: Michael Ludwig <michaelludwig@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Some time ago FontConfig stopped being thread antagonistic and
documented that it was thread safe (or at least any thread un-safety was
a bug). However, until
447b9ccc7d
at least one of the basic atomic primitives was implemented incorrectly.
Resume using a mutex to serialize access to FontConfig until 2.13.93.
Bug: cl/339089311
Change-Id: I632f03bc575a37b5391390a0868aef256e3aacda
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/331339
Reviewed-by: Mike Klein <mtklein@google.com>
Commit-Queue: Ben Wagner <bungeman@google.com>
This has two advantages:
- sksl_gpu was on the verge of overflowing 255 program elements anyway
- this makes it easier to skip ProgramElements when generating the
dehydrated data; in particular, we don't have any need to dehydrate
function prototypes, but if we just skip them, the count would be
wrong
Change-Id: Idbcdec53518e9e6f42473a73a53dae408ce7c980
Bug: skia:10872
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/331282
Commit-Queue: John Stiles <johnstiles@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Adds grvx, Ganesh's addendum to skvx. Here we introduce familiar names
and operations from GPU languages, as well as functions that are
approximate and/or have LSB differences from platform to platform.
The initial implementation has: fast_fma, fast_acos, and
fast_angle_between_vectors. When a function is approximate, its error
range is well documented and tested.
Also establishes GrWangsFormula as the first user of grvx.
Bug: skia:10419
Change-Id: Id0682599cf9c0303eff386095afc3ef9f3a7fa1b
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330119
Commit-Queue: Chris Dalton <csmartdalton@google.com>
Reviewed-by: Brian Salomon <bsalomon@google.com>
We finally have a reference and derivation of Wang's formula, thanks to
tdenniston@. And it turns out that the formula we had been using for
cubics wasn't quite right. It was overly conservative for certain types
of curves.
This CL fixes the incorrect cubic formulas and adds a citation to the
"Pyramid Algorithms" book. We should now be getting by with fewer linear
segments.
Bug: skia:10419
Change-Id: Ib850c7b4d17b8d9f9abed800cc7cb5f074df6e17
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/331156
Reviewed-by: Tyler Denniston <tdenniston@google.com>
Commit-Queue: Chris Dalton <csmartdalton@google.com>
This reverts commit 3ab5b73649.
Reason for revert: <optional> header is C++17 only
Original change's description:
> heif: Add AVIF type and AVIF sniffing
>
> AVIF is the image format based on the AV1 video codec. The
> container for AVIF is very similar to that of HEIF. Add type
> definitions for AVIF and sniffing code for detecting AVIF images.
>
> The underlying android platform's HEIF decoder implementation will
> also support AVIF decoding.
>
> Bug: b/141654151
> Change-Id: I7e31f4cedf0bffb8920ddf880a26601e48d0e833
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330059
> Reviewed-by: Leon Scroggins <scroggo@google.com>
> Reviewed-by: Derek Sollenberger <djsollen@google.com>
> Reviewed-by: Chong Zhang <chz@google.com>
> Commit-Queue: Leon Scroggins <scroggo@google.com>
TBR=djsollen@google.com,scroggo@google.com,chz@google.com,vigneshv@google.com
Change-Id: I9c9cd00af1a41bffa37725f39afc9bc0e504d616
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug: b/141654151
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/331336
Reviewed-by: John Stiles <johnstiles@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
AVIF is the image format based on the AV1 video codec. The
container for AVIF is very similar to that of HEIF. Add type
definitions for AVIF and sniffing code for detecting AVIF images.
The underlying android platform's HEIF decoder implementation will
also support AVIF decoding.
Bug: b/141654151
Change-Id: I7e31f4cedf0bffb8920ddf880a26601e48d0e833
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330059
Reviewed-by: Leon Scroggins <scroggo@google.com>
Reviewed-by: Derek Sollenberger <djsollen@google.com>
Reviewed-by: Chong Zhang <chz@google.com>
Commit-Queue: Leon Scroggins <scroggo@google.com>
The triangulating path renderer can generate multiple triangulations of
the same path at different levels of precision. As previously
implemented the more precise triangulations would steal the unique
key from prior triangulations. This new callback will allow us to
replicate this behavior in the thread-safe cache.
Bug: 1108408
Change-Id: I8b445ca1e503b2fd78727a23d9376a2cf77f291c
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330562
Reviewed-by: Adlai Holler <adlai@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
Motivated by investigating
https://skia-review.googlesource.com/c/skia/+/330696
This CL does not fix anything, but is meant to better document
the current behavior.
Change-Id: I62b8cbfb39e05404f0f5303f024e1f56fc32b7e5
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330937
Commit-Queue: Mike Reed <reed@google.com>
Reviewed-by: Florin Malita <fmalita@chromium.org>
Reviewed-by: Tyler Denniston <tdenniston@google.com>
This error was caused by an unbalanced symbol table push. This could
occur when an interface block encountered an error while parsing its
var-decls.
Change-Id: I910a980ac92fac7c0786c48b8dc3003ee3e75e5b
Bug: oss-fuzz:26700
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330896
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
Profiling sksl_large showed a non-trivial amount of time was spent on
allocating vector<int>s related to swizzle components. This CL
essentially eliminates that cost, for a ~2% savings.
Nanobench before: http://screen/35m2rfy5B8h9eyg
Nanobench after: http://screen/4GuW6ipodyBL34h
Change-Id: I653b69bf1bfbcfdf048987edd23e6f14a5ef3fbc
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330336
Commit-Queue: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Change-Id: I50467eb42528d2109bf7e0d682b3cc4ae892b53b
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330817
Commit-Queue: Greg Daniel <egdaniel@google.com>
Commit-Queue: Mike Klein <mtklein@google.com>
Auto-Submit: Greg Daniel <egdaniel@google.com>
Reviewed-by: Mike Klein <mtklein@google.com>
This is a reland of f71e0be970
Original change's description:
> Moved SkSL data back into node classes
>
> The original goal of this rearchitecture had been to move all of the
> data into IRNode so that we could manage IRNode objects directly rather
> than std::unique_ptr<IRNode>. Other changes have rendered that original
> goal obsolete, so this is undoing most of the work that was done during
> this rearchitecture.
>
> Change-Id: Ic56ffb17bb013c8b4884d710215f5345a481468a
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330297
> Commit-Queue: Ethan Nicholas <ethannicholas@google.com>
> Commit-Queue: John Stiles <johnstiles@google.com>
> Reviewed-by: John Stiles <johnstiles@google.com>
Change-Id: Ifec4777a42ef0f95f6edc418dcd46fd38c856fa5
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330739
Commit-Queue: John Stiles <johnstiles@google.com>
Commit-Queue: Ethan Nicholas <ethannicholas@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
This reverts commit f71e0be970.
Reason for revert: breaking Build-Debian10-EMCC-wasm-Release-WasmGMTests
Original change's description:
> Moved SkSL data back into node classes
>
> The original goal of this rearchitecture had been to move all of the
> data into IRNode so that we could manage IRNode objects directly rather
> than std::unique_ptr<IRNode>. Other changes have rendered that original
> goal obsolete, so this is undoing most of the work that was done during
> this rearchitecture.
>
> Change-Id: Ic56ffb17bb013c8b4884d710215f5345a481468a
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330297
> Commit-Queue: Ethan Nicholas <ethannicholas@google.com>
> Commit-Queue: John Stiles <johnstiles@google.com>
> Reviewed-by: John Stiles <johnstiles@google.com>
TBR=brianosman@google.com,ethannicholas@google.com,johnstiles@google.com
Change-Id: I7a043c8e3e5c711164303cf160846d7cf20ddfbe
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330736
Reviewed-by: John Stiles <johnstiles@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
Change-Id: Idfbe128978575ab84b54485bffe2d82570ee099f
Bug: skia:10870
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330620
Commit-Queue: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Change-Id: I84de1cc582c6b3392881d376794703c61995b20f
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330221
Reviewed-by: John Stiles <johnstiles@google.com>
Reviewed-by: Chris Dalton <csmartdalton@google.com>
Commit-Queue: Michael Ludwig <michaelludwig@google.com>
The original goal of this rearchitecture had been to move all of the
data into IRNode so that we could manage IRNode objects directly rather
than std::unique_ptr<IRNode>. Other changes have rendered that original
goal obsolete, so this is undoing most of the work that was done during
this rearchitecture.
Change-Id: Ic56ffb17bb013c8b4884d710215f5345a481468a
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330297
Commit-Queue: Ethan Nicholas <ethannicholas@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
Reviewed-by: John Stiles <johnstiles@google.com>
This was declared correctly, but implemented with a typo.
That means, of course, it really can't be called.
Bug: skia:9584
Change-Id: Ibd652067d6b3bafcc7a5e67c367b48896b431204
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330277
Auto-Submit: Mike Klein <mtklein@google.com>
Commit-Queue: Mike Reed <reed@google.com>
Reviewed-by: Mike Reed <reed@google.com>
This is an experiment to see if Windows performance numbers are impacted
by removing the recycler entirely. When pools were > 300K, the cost of
continually creating and destroying the pool in a tight loop was
prohibitive, but now that they are 64K, it's possible that the pool
allocation is less expensive and recycling them might not be worth the
downsides.
Change-Id: I7cf4f00e539310da2f852f36419f12d6904727cf
Bug: skia:10865
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330436
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
We track the total amount of memory allocated and the percentage of the
allocated memory that is used.
Bug: skia:10871
Change-Id: I4aa120a3545d215cf42430aa6a73e924118f1dbc
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329963
Commit-Queue: Greg Daniel <egdaniel@google.com>
Reviewed-by: Brian Salomon <bsalomon@google.com>
GrBlockAllocators will hang onto an extra scratch node unless you ask
them to reset. When a Pool is detached from its thread, we know that
we're done with whatever work we needed to do, so it's a good time to
reset any scratch space and reclaim that memory.
Change-Id: I0ff2c4fad9f51a2f3dc911730abad77c06af8d57
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330276
Commit-Queue: John Stiles <johnstiles@google.com>
Reviewed-by: Michael Ludwig <michaelludwig@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
This is a reland of 67e1cf4b1d
The iOS 8 code path now compiles normally
Original change's description:
> Replace pooling mechanism with GrMemoryPool.
>
> This change is a wash for tests that could fit inside the previous
> hard-coded pool (512 nodes) and appears to be a 5% improvement for
> sksl_large. Larger programs would hypothetically show an even more
> significant improvement.
>
> When SK_SUPPORT_GPU is disabled, we disable pooling entirely and fall
> back to the system allocator. This is necessary because SkSL can exist
> without Ganesh (such as in the wasm+CanvasKit build).
>
> Nanobench: http://screen/4xJEzdGducRxGeq
>
> Change-Id: I71dc702a84ab5c163673e35ec651003d7d45dacd
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330219
> Commit-Queue: John Stiles <johnstiles@google.com>
> Reviewed-by: Brian Osman <brianosman@google.com>
> Auto-Submit: John Stiles <johnstiles@google.com>
Change-Id: Iced330084f1ed8997e19bbee585422cb89e1c6b0
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330404
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>
The key for the cached mask allows some slop in the subpixel
positioning of the mask. This means that, in rare circumstances,
the cached, integer draw-rect can differ from what would've been
computed directly.
Bug: 1143674
Change-Id: I1a0f47c0f8030c692bb737f8386da77348046af5
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330410
Reviewed-by: Adlai Holler <adlai@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
Add checks for compatible channels and format types.
Bug:skia:10863
Change-Id: I44f9310097007c3ed70f94e949b10612a2fe2643
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329421
Reviewed-by: Brian Salomon <bsalomon@google.com>
Reviewed-by: Alexis Hetu <sugoi@google.com>
Commit-Queue: Greg Daniel <egdaniel@google.com>
This reverts commit 67e1cf4b1d.
Reason for revert: iOS 8
Original change's description:
> Replace pooling mechanism with GrMemoryPool.
>
> This change is a wash for tests that could fit inside the previous
> hard-coded pool (512 nodes) and appears to be a 5% improvement for
> sksl_large. Larger programs would hypothetically show an even more
> significant improvement.
>
> When SK_SUPPORT_GPU is disabled, we disable pooling entirely and fall
> back to the system allocator. This is necessary because SkSL can exist
> without Ganesh (such as in the wasm+CanvasKit build).
>
> Nanobench: http://screen/4xJEzdGducRxGeq
>
> Change-Id: I71dc702a84ab5c163673e35ec651003d7d45dacd
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330219
> Commit-Queue: John Stiles <johnstiles@google.com>
> Reviewed-by: Brian Osman <brianosman@google.com>
> Auto-Submit: John Stiles <johnstiles@google.com>
TBR=brianosman@google.com,ethannicholas@google.com,johnstiles@google.com
Change-Id: I26dbd7f2d5348dd717c39fd0780ee5d140292e9a
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330416
Reviewed-by: John Stiles <johnstiles@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
This takes the "one pass" code path more often, using less memory, as it
does not have to allocate an intermediate width*height pixel buffer.
Wuffs v0.2 did not support RGB 565 color but Wuffs v0.3 does.
The Codec_AnimatedTransparentGif test passes with skia_use_wuffs true or
false.
Change-Id: Id569fc0bf62e614fa881cb235a9a6a1ca340dcb0
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329916
Commit-Queue: Leon Scroggins <scroggo@google.com>
Reviewed-by: Leon Scroggins <scroggo@google.com>
With the addition of vertex data to the thread safe cache we also have
to handle the case where a given SkPath becomes inaccessible and
proactively invalidate the matching entry.
Bug: 1108408
Change-Id: Id11ce2aa10517f7c0772a253634d3c0d13e13460
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330261
Reviewed-by: Adlai Holler <adlai@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
This change is a wash for tests that could fit inside the previous
hard-coded pool (512 nodes) and appears to be a 5% improvement for
sksl_large. Larger programs would hypothetically show an even more
significant improvement.
When SK_SUPPORT_GPU is disabled, we disable pooling entirely and fall
back to the system allocator. This is necessary because SkSL can exist
without Ganesh (such as in the wasm+CanvasKit build).
Nanobench: http://screen/4xJEzdGducRxGeq
Change-Id: I71dc702a84ab5c163673e35ec651003d7d45dacd
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330219
Commit-Queue: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
This provides the optimizer with hints so that it can avoid virtual
calls in scenarios where the type is known.
(The ExternalValue leaf class is excluded from this change; unlike
most IRNodes, it is meant to be subclassed.)
Change-Id: I68ff3e2336cb478bcc546c9fe3640719a01d8ea7
Bug: skia:10886
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330223
Commit-Queue: John Stiles <johnstiles@google.com>
Commit-Queue: Herb Derby <herb@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Herb Derby <herb@google.com>
This is trying to dynamically set the array bounds for sk_in, based on
the primitive type. However, the sizes() field of the interface block
isn't even used for that - it's done with similar logic in
writeInterfaceBlock. Also, this was using the wrong size for several
primitive types.
Bug: skia:7733
Change-Id: Ic0925b51d86dcaea718373bb16d633cc2cd3398f
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330222
Commit-Queue: Brian Osman <brianosman@google.com>
Commit-Queue: Ethan Nicholas <ethannicholas@google.com>
Auto-Submit: Brian Osman <brianosman@google.com>
Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
Now that GrOps no longer needs a GrMemoryPool to be
deleted, just pass down the the SkArenaAlloc instead
of both the GrMemoryPool and SkArenaAlloc.
The alloc is only used for two ops, but we pass it
to all the ops most of which don't use it.
Change-Id: I873efcdfe44b446f2ac5089ea8425dff257e318c
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330118
Reviewed-by: Michael Ludwig <michaelludwig@google.com>
Commit-Queue: Herb Derby <herb@google.com>
This reverts commit 292f7777da.
Reason for revert: Oddly this is causing issues in vulkan chrome linux with specific window managers, see crbug.com/1141831
Original change's description:
> In Vk don't set dynamic blend constant on Pipeline if we don't use it.
>
> We already don't call the dynamic update for the blend constant if it's
> not used. This change makes it so we don't even tell the VkPipeline
> to have support for a dynamic blend constant. This allows the driver
> to make the pipline objects slightly more optimal.
>
> Change-Id: Ib7ea4e0ec7eeb8e7c1cf165d426ee863fa0a3a00
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329036
> Reviewed-by: Jim Van Verth <jvanverth@google.com>
> Commit-Queue: Greg Daniel <egdaniel@google.com>
TBR=egdaniel@google.com,jvanverth@google.com
# Not skipping CQ checks because original CL landed > 1 day ago.
Change-Id: Id8b74889b87c6402982fd4c8dcda3a5876005890
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330218
Reviewed-by: Greg Daniel <egdaniel@google.com>
Commit-Queue: Greg Daniel <egdaniel@google.com>
This doesn't change any behavior; it just exposes existing code that
previously could only run at ~GrMemoryPool time. Now you can report
leaks on a memory pool without actually destroying it.
Change-Id: I35bf6c7d886827031cce0e2b045775207062c121
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330217
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>
We were getting lucky with intersect in that the tested shaders always
ended up multiplying by the input alpha at the end. This adds a
dedicated GM that tests the difference op for clip shaders, across
a variety of shape types.
Bug: skia:10879
Change-Id: I45fdaad27f05ae7a74dbdc79eece2e9688806568
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330123
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Michael Ludwig <michaelludwig@google.com>
This reverts commit 375721d7bb.
Some manual edits needed due to time elapsed.
Bug: chromium:1141332, skia:10566
Change-Id: Iadb15d3f5334d9eed4e7053e9c19d75a0bbeb9de
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330196
Reviewed-by: Mike Reed <reed@google.com>
Commit-Queue: Mike Klein <mtklein@google.com>
To make this possible, the logic for maintaining a pool of a given node
size has been split into a separate `Subpool` class. We now have two
Subpool objects in the PoolData class; they are always adjacent in
memory, and we can still identify a pooled node just as easily by
verifying that the pointer address is within the range of one of our
Subpools. (Identifying which subpool requires one extra check.)
This change will allow us to pool a handful of larger allocations, which
will benefit almost every program since a handful of variables and
functions are universally necessary, even in simple code.
For now, the chosen size of a large node is just a guess (2x the small
node size); this can be adjusted easily once we have a better idea of
the expected sizes for our larger IRNodes, but for now this successfully
catches a significant number of allocations in `sksl_medium` and
`sksl_large`.
Change-Id: I08fb9c55c02162d4b56e72ff4e0923f2cf258651
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330124
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
This is a reland of 22ef2257c8
This reland fixes an issue with ExternalValue nodes on 32-bit builds;
these should never be pooled, because we can't control their lifetimes.
Original change's description:
> Update the SkSL pool interface to take an allocation size.
>
> Since our end goal no longer has all IRNodes ending up as the exact same
> size in memory, it makes sense for the allocator function to take in a
> desired size. This also opens the door to separate "small" and "large"
> pools, if we want to add pooling support for large but semi-common
> things like Variables.
>
> Change-Id: If3dbe31588adeedede327c5967c344a19507b6fa
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329961
> Reviewed-by: Brian Osman <brianosman@google.com>
> Commit-Queue: Brian Osman <brianosman@google.com>
> Auto-Submit: John Stiles <johnstiles@google.com>
Change-Id: If701ae4a5e18b66d4138bc09c9c8dc1a60579c90
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330105
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
I switched GrOps from using GrOpMemoryPool to
GrMemoryPool, but I forgot to switch over the
GrOp::Owner. Make sure the GrOp::Owner works for
both new/delete and GrMemoryPool.
In addition, convert one last unique_ptr<GrOp> to
GrOp::Owner.
Change-Id: I660ad77bee0f060f263ff2ed07974afb83063441
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/330097
Commit-Queue: Herb Derby <herb@google.com>
Reviewed-by: Mike Klein <mtklein@google.com>
This reverts commit 22ef2257c8.
Reason for revert: mysterious tree breakage
Original change's description:
> Update the SkSL pool interface to take an allocation size.
>
> Since our end goal no longer has all IRNodes ending up as the exact same
> size in memory, it makes sense for the allocator function to take in a
> desired size. This also opens the door to separate "small" and "large"
> pools, if we want to add pooling support for large but semi-common
> things like Variables.
>
> Change-Id: If3dbe31588adeedede327c5967c344a19507b6fa
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329961
> Reviewed-by: Brian Osman <brianosman@google.com>
> Commit-Queue: Brian Osman <brianosman@google.com>
> Auto-Submit: John Stiles <johnstiles@google.com>
TBR=brianosman@google.com,ethannicholas@google.com,johnstiles@google.com
Change-Id: I346948aeb407dc6e6e84835f68d0353e4869ddf7
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329965
Reviewed-by: John Stiles <johnstiles@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
Change-Id: I93fdeaa3ea6be73619f82859bb53aa88fae3262b
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329962
Reviewed-by: Michael Ludwig <michaelludwig@google.com>
Reviewed-by: Mike Klein <mtklein@google.com>
Commit-Queue: Herb Derby <herb@google.com>
Since our end goal no longer has all IRNodes ending up as the exact same
size in memory, it makes sense for the allocator function to take in a
desired size. This also opens the door to separate "small" and "large"
pools, if we want to add pooling support for large but semi-common
things like Variables.
Change-Id: If3dbe31588adeedede327c5967c344a19507b6fa
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329961
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
When the test expression has a side-effect, but both the true and else
blocks are empty, the optimizer moves the test out to a standalone
ExpressionStatement. Updating the usage in that situation involves
traversing an IfStatement with no test.
Bug: oss-fuzz:26666
Change-Id: I2fb4004f2401784402040345df49a7d42e4aab5e
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329960
Reviewed-by: John Stiles <johnstiles@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
There's no functional change here; it's just using a slightly higher-
level abstraction to pass the same payload.
Change-Id: Ife7efa038db5d6dbde5decae2be79ad9db877aba
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329782
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
This uma stat can allow us to look a few different things. First we can
look at the distribution of render passes per submit (including mean,
median, ect.). Additionally we can do things like sum the total number
of renderpasses from all submits and divide by the total number of
frames (Compositing.Display.DrawToSwapUs::TotalCount) to get the
average number of render passes per frame and not just submit.
Technically under the hood the ENUMERATION and LINEAR_EXACT macros will
do the same thing and we could just use ENUMERATION here. However to
match how chrome uses their UMA macros it is more correct to use the
LINEAR_EXACT since we aren't counting enum values. Chromes macros
actually have static asserts the values are or are not enums, but the
skia implementations do not.
Includes some minor updates to names to match chromes UMA macro values.
This still requires the chrome implementation of the new macro.
Bug: skia:10871
Change-Id: Idbc4d2fc649bbdefd0952a002c3327cb700b552b
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329776
Commit-Queue: Greg Daniel <egdaniel@google.com>
Reviewed-by: Brian Salomon <bsalomon@google.com>
Reviewed-by: Adlai Holler <adlai@google.com>
GLSL requires that functions are declared before first use. In most
cases, we avoid the requirement for explicit prototypes by this by
strategically ordering our functions within the emitted code, but an FP
file might reference its helper functions in any order or have helper
functions that cross-invoke each other, so prototypes should be emitted.
Change-Id: I3b9e9c9ec4bd5be90b4f71f8165af45364facf30
Bug: skia:10872
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329781
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
This is necessary to support function calls in FP files properly; in
some cases, functions can be referenced before they have been emitted,
and we need to be able to name them.
This CL resolves the remaining errors in GrRecursion.cpp. There are
still additional errors in GrNestedCall.cpp that will be fixed in
followup CLs.
Change-Id: Iec98ef02ea6a98a9945a4e0e3cfa3537dff01305
Bug: skia:10684, skia:10872
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329676
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Change-Id: I7a07790a7991bc3ec0f9ebfe4ff28e88f00ec0a7
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329936
Reviewed-by: Mike Klein <mtklein@google.com>
Commit-Queue: Mike Klein <mtklein@google.com>
This backs off a bit on the "stuff everything into IRNode" push, moving
the data of the big nodes back into them and significantly cutting the
size of IRNode. This isn't a complete revert of the changes to these
nodes - now that all of the fields have been hidden behind accessor
methods, it's merely an implementation detail where the data actually
lives, so these changes are much smaller and more targeted.
Change-Id: I84c770245f26dfe36651d0f482543505bd7f71ef
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329629
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: Ethan Nicholas <ethannicholas@google.com>
Removes unused code, including utilities for dealing with KLM
functionals for the implicit cubic function. The implicit has proven
to not be a very good tool for rendering cubics.
Change-Id: I577b50a9eb296c52dc0101a20394480a4a008654
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329440
Reviewed-by: Greg Daniel <egdaniel@google.com>
Commit-Queue: Chris Dalton <csmartdalton@google.com>
Adds an SkChopCubicAt overload that performs two chops at once in
SIMD. Also updates SkChopCubicAt to accept T values of 0 and 1. This
has been the source of bugs in the past.
Bug: skia:10419
Change-Id: Ic8a482a69192fb1685f3766411cbdceed830f9b7
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/327436
Reviewed-by: Mike Reed <reed@google.com>
Commit-Queue: Chris Dalton <csmartdalton@google.com>
and flush out some non-substantive changes; namely:
whitespace changes
tweak what is stored in TessInfo
change createMesh to be CreateMesh
Bug: 1108408
Change-Id: Ife9a0e500b9f51db597350b0257b20efece03bdc
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329633
Reviewed-by: Adlai Holler <adlai@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
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>
Previously, the generated name was passed back in a SkString-pointer
output parameter. Directly returning the SkString simplified nearly
every call site, and made none of them worse.
Change-Id: Ied57adcf8fb087a301641adad437d8a68f8d3297
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329622
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Previously, we'd just emit functions with `%s` in their function bodies.
Also, the fFormatArgs wouldn't get cleaned up, so subsequent code would
fill in the wrong format arguments in each place.
There are still problems with function calls (`sample` is broken;
function names are accessed before they've been declared or initialized)
but this is a step in the right direction.
Bug: skia:10684
Change-Id: I7399a71d44ebba5049703484ec9933dffbe8e2b9
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329417
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Automatically computed once IR generation is complete, and then used
throughout optimization and inlining, and the backend generators.
Optimization and inlining are reworked to keep the data up-to-date as
they edit the IR.
Bug: skia:10589
Change-Id: Iaded42d8157732dd6fe05f74c5b7bb8366916635
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328382
Reviewed-by: John Stiles <johnstiles@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Bug: skia:10866
Change-Id: I06c675a423380d30017ebcc5485d556211d403fa
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329390
Commit-Queue: Michael Ludwig <michaelludwig@google.com>
Auto-Submit: Mike Klein <mtklein@google.com>
Reviewed-by: Michael Ludwig <michaelludwig@google.com>
Creating the next command buffer immediately seems to confuse
the Metal GPU capture command in Xcode. If the command buffer isn't
used until the next frame it apparently doesn't include it in the
capture. The solution is to delay creation until we first need it.
Change-Id: I95fea78a68356d319418ca15bd852e34d9efa12b
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329416
Reviewed-by: Adlai Holler <adlai@google.com>
Commit-Queue: Jim Van Verth <jvanverth@google.com>
This is large enough to hold sksl_medium in its entirety; sksl_large
overflows somewhat, but it's still much better than nothing.
Change-Id: Ief22a94cc7f554ecc289905e9ccc20594eab6819
Bug: chromium:1141863
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329456
Auto-Submit: John Stiles <johnstiles@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
This removes the last of the SkFontHost LCD globals and the
SkSurfaceProps::kLegacyFontHost_InitType.
Bug: skia:3934
Change-Id: Ic2342a3ea3dbcd075d6817cbd3fc27274e376b8d
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329364
Commit-Queue: Ben Wagner <bungeman@google.com>
Reviewed-by: Mike Reed <reed@google.com>
GrStrokePatchBuilder had the wrong formula. This CL moves the formula to
GrStrokeTessellateOp and calculates it in one spot for the builder and
tessellator both. It also fixes a bug that was hiding behind this
formula error and updates some variable names.
Bug: skia:10419
Change-Id: I908d3960b16cac8dca0d40ff4116a3f5c5beed06
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328416
Reviewed-by: Jim Van Verth <jvanverth@google.com>
Commit-Queue: Chris Dalton <csmartdalton@google.com>
In cases where multiple variables were declared on a single line, it is
legal for variable initialization-expressions to reference variables
declared earlier in the var-decl statement. It is NOT legal for the
inliner to move those references up to the previous statement, where the
variable doesn't exist yet.
This is mitigated by disabling the IRGenerator inliner for var-decls
past the first one in a var-decls statement. (The optimizer will still
pass over this code later and is able to inline it correctly, if it is
worth doing.)
Change-Id: I7a0d45eab20e30ed9f6b2f5c1251b6e0d8eeaea3
Bug: oss-fuzz:26167
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329357
Auto-Submit: John Stiles <johnstiles@google.com>
Commit-Queue: Ethan Nicholas <ethannicholas@google.com>
Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
When a Program is freed, rather than immediately disposing of its Pool,
it now sends it to Pool::Recycle, which holds onto it. If Pool::Create
is called, it satisfies the request by simply handing back the recycled
pool. Only one pool is kept in recycle storage at a time--recycling
more than one pool in a row will cause all but one to be freed. To avoid
holding onto Pool memory indefinitely, pool recycle storage is cleaned
up whenever a Compiler is destroyed.
Change-Id: I21c1ccde84507e344102d05506d869e62ca095a6
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329175
Reviewed-by: Brian Osman <brianosman@google.com>
Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Change-Id: I1b14e4c8de0ca60a0ebbdbc225befc5074e8de22
Bug: skia:10862
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329177
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 simplifies the signatures of several methods and makes it easier to
ignore errors we have already reported.
Change-Id: I767ca22a66e69fe72557c2560ef84f5dbe339eb8
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329220
Commit-Queue: Ethan Nicholas <ethannicholas@google.com>
Reviewed-by: John Stiles <johnstiles@google.com>
This reverts commit f73091bb81.
Reason for revert: rolling forward after fixing prior CL
Original change's description:
> Revert "Add pooling support on iOS."
>
> This reverts commit 38a93e622b.
>
> Reason for revert: need revert first pool change
>
> Original change's description:
> > Add pooling support on iOS.
> >
> > This replaces the `thread_local` attribute with `pthread_setspecific`
> > and `pthread_getspecific`. I don't have easy access to iOS 8/9 for
> > testing purposes, but on Mac OS X, this implementation works and
> > benchmarks the same as the `thread_local` implementation.
> >
> > Change-Id: I86db88c24d59d946adb66141b32733ebf5261c76
> > Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328837
> > Reviewed-by: Brian Osman <brianosman@google.com>
> > Commit-Queue: Brian Osman <brianosman@google.com>
> > Auto-Submit: John Stiles <johnstiles@google.com>
>
> TBR=brianosman@google.com,adlai@google.com,johnstiles@google.com
>
> Change-Id: Ic06f9e32e524b2be601ee21a5da605fd19aaa64b
> No-Presubmit: true
> No-Tree-Checks: true
> No-Try: true
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329164
> Reviewed-by: Greg Daniel <egdaniel@google.com>
> Commit-Queue: Greg Daniel <egdaniel@google.com>
TBR=egdaniel@google.com,brianosman@google.com,adlai@google.com,johnstiles@google.com
Change-Id: I0e021e9304ee88d6a29739c287eb515abff8b8a4
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329173
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: John Stiles <johnstiles@google.com>
This is a reland of e16eca95f5
This fixes the no-op (iOS) implementation of CreatePoolOnThread.
Original change's description:
> Create a basic IRNode pooling system.
>
> Allocations are redirected by overriding `operator new` and `operator
> delete` on the IRNode class. This allows us to use our existing
> `unique_ptr` and `make_unique` calls as-is. The Pool class is simple;
> it holds a fixed number of nodes and recycles them as they are returned.
>
> A fixed pool size of 2000 nodes was chosen. That is large enough to hold
> the contents of `sksl_large` during compilation, but it can be
> overflowed by very large shaders, or if multiple programs are converted
> at the same time. Exhausting the pool is not a problem; if this happens,
> additional nodes will be allocated via the system allocator as usual.
> More elaborate schemes are possible but might not add a lot of value.
>
> Thread safety is accomplished by placing the pool in a `thread_local`
> static during a Program's creation and destruction; the pool is freed
> when the program is destroyed. One important consequence of this
> strategy is that a program must free every node that it allocated during
> its creation, or else the node will be leaked. In debug, leaking a node
> will be detected and causes a DEBUGFAIL. In release, the pool will be
> freed despite having a live node in it, and if that node is later freed,
> that pointer will be passed to the system `free` (which is likely to
> cause a crash).
>
> In this CL, iOS does not support pooling, since support for
> `thread_local` was only added on iOS 9. This is fixed in the followup
> CL, http://review.skia.org/328837, which uses pthread keys on iOS.
>
> Nanobench shows ~15% improvement:
> (last week) http://screen/5CNBhTaZApcDA8h
> (today) http://screen/8ti5Rymvf6LUs8i
>
> Change-Id: I559de73606ee1be54e5eae7f82129dc928a63e3c
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/326876
> Commit-Queue: John Stiles <johnstiles@google.com>
> Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
> Auto-Submit: John Stiles <johnstiles@google.com>
Change-Id: I8623a574a7e92332ff00b83982497863c8953929
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329171
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 reverts commit 5b09e6a007.
Reason for revert: breaking g3
Original change's description:
> Reland "Create a basic IRNode pooling system."
>
> This is a reland of e16eca95f5
>
> Original change's description:
> > Create a basic IRNode pooling system.
> >
> > Allocations are redirected by overriding `operator new` and `operator
> > delete` on the IRNode class. This allows us to use our existing
> > `unique_ptr` and `make_unique` calls as-is. The Pool class is simple;
> > it holds a fixed number of nodes and recycles them as they are returned.
> >
> > A fixed pool size of 2000 nodes was chosen. That is large enough to hold
> > the contents of `sksl_large` during compilation, but it can be
> > overflowed by very large shaders, or if multiple programs are converted
> > at the same time. Exhausting the pool is not a problem; if this happens,
> > additional nodes will be allocated via the system allocator as usual.
> > More elaborate schemes are possible but might not add a lot of value.
> >
> > Thread safety is accomplished by placing the pool in a `thread_local`
> > static during a Program's creation and destruction; the pool is freed
> > when the program is destroyed. One important consequence of this
> > strategy is that a program must free every node that it allocated during
> > its creation, or else the node will be leaked. In debug, leaking a node
> > will be detected and causes a DEBUGFAIL. In release, the pool will be
> > freed despite having a live node in it, and if that node is later freed,
> > that pointer will be passed to the system `free` (which is likely to
> > cause a crash).
> >
> > In this CL, iOS does not support pooling, since support for
> > `thread_local` was only added on iOS 9. This is fixed in the followup
> > CL, http://review.skia.org/328837, which uses pthread keys on iOS.
> >
> > Nanobench shows ~15% improvement:
> > (last week) http://screen/5CNBhTaZApcDA8h
> > (today) http://screen/8ti5Rymvf6LUs8i
> >
> > Change-Id: I559de73606ee1be54e5eae7f82129dc928a63e3c
> > Reviewed-on: https://skia-review.googlesource.com/c/skia/+/326876
> > Commit-Queue: John Stiles <johnstiles@google.com>
> > Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
> > Auto-Submit: John Stiles <johnstiles@google.com>
>
> Change-Id: I114971e8e7ac0fabaf26216ae8813eeeaad0d4a2
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329086
> Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
> Commit-Queue: John Stiles <johnstiles@google.com>
TBR=brianosman@google.com,ethannicholas@google.com,johnstiles@google.com
Change-Id: Ie77a23366f2ba52fcbb0a751d11ca2792790a30c
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329165
Reviewed-by: Greg Daniel <egdaniel@google.com>
Commit-Queue: Greg Daniel <egdaniel@google.com>
This reverts commit 38a93e622b.
Reason for revert: need revert first pool change
Original change's description:
> Add pooling support on iOS.
>
> This replaces the `thread_local` attribute with `pthread_setspecific`
> and `pthread_getspecific`. I don't have easy access to iOS 8/9 for
> testing purposes, but on Mac OS X, this implementation works and
> benchmarks the same as the `thread_local` implementation.
>
> Change-Id: I86db88c24d59d946adb66141b32733ebf5261c76
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328837
> Reviewed-by: Brian Osman <brianosman@google.com>
> Commit-Queue: Brian Osman <brianosman@google.com>
> Auto-Submit: John Stiles <johnstiles@google.com>
TBR=brianosman@google.com,adlai@google.com,johnstiles@google.com
Change-Id: Ic06f9e32e524b2be601ee21a5da605fd19aaa64b
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329164
Reviewed-by: Greg Daniel <egdaniel@google.com>
Commit-Queue: Greg Daniel <egdaniel@google.com>
This replaces the `thread_local` attribute with `pthread_setspecific`
and `pthread_getspecific`. I don't have easy access to iOS 8/9 for
testing purposes, but on Mac OS X, this implementation works and
benchmarks the same as the `thread_local` implementation.
Change-Id: I86db88c24d59d946adb66141b32733ebf5261c76
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328837
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Change-Id: I433159717b831e9dbfc251c38687572bdc45c959
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329037
Auto-Submit: Jim Van Verth <jvanverth@google.com>
Commit-Queue: Ethan Nicholas <ethannicholas@google.com>
Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
This is a reland of e16eca95f5
Original change's description:
> Create a basic IRNode pooling system.
>
> Allocations are redirected by overriding `operator new` and `operator
> delete` on the IRNode class. This allows us to use our existing
> `unique_ptr` and `make_unique` calls as-is. The Pool class is simple;
> it holds a fixed number of nodes and recycles them as they are returned.
>
> A fixed pool size of 2000 nodes was chosen. That is large enough to hold
> the contents of `sksl_large` during compilation, but it can be
> overflowed by very large shaders, or if multiple programs are converted
> at the same time. Exhausting the pool is not a problem; if this happens,
> additional nodes will be allocated via the system allocator as usual.
> More elaborate schemes are possible but might not add a lot of value.
>
> Thread safety is accomplished by placing the pool in a `thread_local`
> static during a Program's creation and destruction; the pool is freed
> when the program is destroyed. One important consequence of this
> strategy is that a program must free every node that it allocated during
> its creation, or else the node will be leaked. In debug, leaking a node
> will be detected and causes a DEBUGFAIL. In release, the pool will be
> freed despite having a live node in it, and if that node is later freed,
> that pointer will be passed to the system `free` (which is likely to
> cause a crash).
>
> In this CL, iOS does not support pooling, since support for
> `thread_local` was only added on iOS 9. This is fixed in the followup
> CL, http://review.skia.org/328837, which uses pthread keys on iOS.
>
> Nanobench shows ~15% improvement:
> (last week) http://screen/5CNBhTaZApcDA8h
> (today) http://screen/8ti5Rymvf6LUs8i
>
> Change-Id: I559de73606ee1be54e5eae7f82129dc928a63e3c
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/326876
> Commit-Queue: John Stiles <johnstiles@google.com>
> Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
> Auto-Submit: John Stiles <johnstiles@google.com>
Change-Id: I114971e8e7ac0fabaf26216ae8813eeeaad0d4a2
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329086
Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
This fixes a potential leak of the symbol table. (The leak would
eventually resolve itself once the IRGenerator were destroyed, but it
does outlast the compilation step.)
Change-Id: I6b2e303f00a3331fccbd8421a5173defd352f022
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328985
Auto-Submit: John Stiles <johnstiles@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
We already don't call the dynamic update for the blend constant if it's
not used. This change makes it so we don't even tell the VkPipeline
to have support for a dynamic blend constant. This allows the driver
to make the pipline objects slightly more optimal.
Change-Id: Ib7ea4e0ec7eeb8e7c1cf165d426ee863fa0a3a00
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/329036
Reviewed-by: Jim Van Verth <jvanverth@google.com>
Commit-Queue: Greg Daniel <egdaniel@google.com>
This reverts commit 07829f27ec.
Reason for revert: actually looks like skp gone good
Original change's description:
> Revert "Perform bounding rect-relative calcs in full float in GrRRectBlurEffect"
>
> This reverts commit f4594d1d5b.
>
> Reason for revert: skp gone bad
>
> Original change's description:
> > Perform bounding rect-relative calcs in full float in GrRRectBlurEffect
> >
> > Add GM that tests very wide/tall SkRRects with blurs
> >
> > Bug: chromium:1138810
> > Change-Id: Ib5a2e04de50c441f57f5e4b6194c3f9829323dc9
> > Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328383
> > Commit-Queue: Brian Salomon <bsalomon@google.com>
> > Reviewed-by: Michael Ludwig <michaelludwig@google.com>
>
> TBR=bsalomon@google.com,michaelludwig@google.com
>
> Change-Id: I3633efd4802a0a9493831ec471304333272fa87e
> No-Presubmit: true
> No-Tree-Checks: true
> No-Try: true
> Bug: chromium:1138810
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328905
> Reviewed-by: Brian Salomon <bsalomon@google.com>
> Commit-Queue: Brian Salomon <bsalomon@google.com>
TBR=bsalomon@google.com,michaelludwig@google.com
# Not skipping CQ checks because this is a reland.
Bug: chromium:1138810
Change-Id: Ic061849c410f43411d03972cdfadc5f0610efe6c
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328907
Reviewed-by: Brian Salomon <bsalomon@google.com>
Reviewed-by: Michael Ludwig <michaelludwig@google.com>
Commit-Queue: Brian Salomon <bsalomon@google.com>
This also has several other bug fixes and refactorings within it that
I realized were possible while updating every where that had checked
sigma > 0 to be sigma > kEffectivelyZeroSigma.
The big things are that SkBlurPriv.h goes away and its functions are
just moved into SkGpuBlurUtils since they were only used by the GPU.
The implementations of those functions are also collected into
SkGpuBlurUtils.cpp. I removed the GrMatrixConvolution::MakeGaussian,
in favor of SkGpuBlurUtils filling in the kernel itself and then calling
the regular Make. This let me consolidate two different 1D kernel
computing functions, and remove the 1D fallback code from the 2D kernel
calculation because GaussianBlur() can detect that earlier.
The new GM, BlurSigmaSmall, originally drew incorrectly on the GPU
backend because it's small but non-zero sigma would trick the sigma > 0
checks in various places so we'd do a full 2 pass X/Y blur. However,
when the sigma was too small, the kernel was just filled with 0s so the
Y pass would effectively clear everything. While I could have just fixed
that to be a [0, 1, 0] kernel, updating the blur pipeline to compare
against integer radii seems more robust.
Change-Id: I3c41e0235a27615a9056b25e627ffedd995264bd
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328797
Commit-Queue: Michael Ludwig <michaelludwig@google.com>
Reviewed-by: Brian Salomon <bsalomon@google.com>
This reverts commit e16eca95f5.
Reason for revert: ASAN error on fuzzer
https://status.skia.org/logs/snBeMRUkDrwDYbnm2SAG/7ad38736-d579-4e94-bc10-87c002f3f7d6/fd7b6ea1-5d36-4612-85d1-88462a5271f7
Original change's description:
> Create a basic IRNode pooling system.
>
> Allocations are redirected by overriding `operator new` and `operator
> delete` on the IRNode class. This allows us to use our existing
> `unique_ptr` and `make_unique` calls as-is. The Pool class is simple;
> it holds a fixed number of nodes and recycles them as they are returned.
>
> A fixed pool size of 2000 nodes was chosen. That is large enough to hold
> the contents of `sksl_large` during compilation, but it can be
> overflowed by very large shaders, or if multiple programs are converted
> at the same time. Exhausting the pool is not a problem; if this happens,
> additional nodes will be allocated via the system allocator as usual.
> More elaborate schemes are possible but might not add a lot of value.
>
> Thread safety is accomplished by placing the pool in a `thread_local`
> static during a Program's creation and destruction; the pool is freed
> when the program is destroyed. One important consequence of this
> strategy is that a program must free every node that it allocated during
> its creation, or else the node will be leaked. In debug, leaking a node
> will be detected and causes a DEBUGFAIL. In release, the pool will be
> freed despite having a live node in it, and if that node is later freed,
> that pointer will be passed to the system `free` (which is likely to
> cause a crash).
>
> In this CL, iOS does not support pooling, since support for
> `thread_local` was only added on iOS 9. This is fixed in the followup
> CL, http://review.skia.org/328837, which uses pthread keys on iOS.
>
> Nanobench shows ~15% improvement:
> (last week) http://screen/5CNBhTaZApcDA8h
> (today) http://screen/8ti5Rymvf6LUs8i
>
> Change-Id: I559de73606ee1be54e5eae7f82129dc928a63e3c
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/326876
> Commit-Queue: John Stiles <johnstiles@google.com>
> Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
> Auto-Submit: John Stiles <johnstiles@google.com>
TBR=brianosman@google.com,ethannicholas@google.com,johnstiles@google.com
Change-Id: I625d95a14057727b297c0bfc5b98bcd78ad8572c
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328906
Reviewed-by: John Stiles <johnstiles@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
This reverts commit f4594d1d5b.
Reason for revert: skp gone bad
Original change's description:
> Perform bounding rect-relative calcs in full float in GrRRectBlurEffect
>
> Add GM that tests very wide/tall SkRRects with blurs
>
> Bug: chromium:1138810
> Change-Id: Ib5a2e04de50c441f57f5e4b6194c3f9829323dc9
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328383
> Commit-Queue: Brian Salomon <bsalomon@google.com>
> Reviewed-by: Michael Ludwig <michaelludwig@google.com>
TBR=bsalomon@google.com,michaelludwig@google.com
Change-Id: I3633efd4802a0a9493831ec471304333272fa87e
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug: chromium:1138810
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328905
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Brian Salomon <bsalomon@google.com>
These changes cause fewer unnecessary temporary variables to be emitted
in the final gencode and give the optimizer less work to do.
Change-Id: Ied0e83904e2d108382666d18ab733334292e5a63
Bug: skia:10549
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328838
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Switch statements were not caught when originally implementing
http://review.skia.org/328384.
Change-Id: Iff21e5743bf6a604e13c45a736799b4929844472
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328900
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>
Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
Add GM that tests very wide/tall SkRRects with blurs
Bug: chromium:1138810
Change-Id: Ib5a2e04de50c441f57f5e4b6194c3f9829323dc9
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328383
Commit-Queue: Brian Salomon <bsalomon@google.com>
Reviewed-by: Michael Ludwig <michaelludwig@google.com>
Allocations are redirected by overriding `operator new` and `operator
delete` on the IRNode class. This allows us to use our existing
`unique_ptr` and `make_unique` calls as-is. The Pool class is simple;
it holds a fixed number of nodes and recycles them as they are returned.
A fixed pool size of 2000 nodes was chosen. That is large enough to hold
the contents of `sksl_large` during compilation, but it can be
overflowed by very large shaders, or if multiple programs are converted
at the same time. Exhausting the pool is not a problem; if this happens,
additional nodes will be allocated via the system allocator as usual.
More elaborate schemes are possible but might not add a lot of value.
Thread safety is accomplished by placing the pool in a `thread_local`
static during a Program's creation and destruction; the pool is freed
when the program is destroyed. One important consequence of this
strategy is that a program must free every node that it allocated during
its creation, or else the node will be leaked. In debug, leaking a node
will be detected and causes a DEBUGFAIL. In release, the pool will be
freed despite having a live node in it, and if that node is later freed,
that pointer will be passed to the system `free` (which is likely to
cause a crash).
In this CL, iOS does not support pooling, since support for
`thread_local` was only added on iOS 9. This is fixed in the followup
CL, http://review.skia.org/328837, which uses pthread keys on iOS.
Nanobench shows ~15% improvement:
(last week) http://screen/5CNBhTaZApcDA8h
(today) http://screen/8ti5Rymvf6LUs8i
Change-Id: I559de73606ee1be54e5eae7f82129dc928a63e3c
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/326876
Commit-Queue: John Stiles <johnstiles@google.com>
Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Prepare this class to hold either a proxyView or a vertex blob
Bug: 1108408
Change-Id: Ib6abb4da64ccc70b9e2af2546e1b071396dd42cb
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328836
Reviewed-by: Adlai Holler <adlai@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
PLS and discard-only shaders are the only time this has an impact,
and it doesn't seem like a problem to have the declaration?
Removes one use of variable reference counts, which are going to be
refactored.
Change-Id: Idb8d06087eed56070252ee02dcf907bf0d24c5a1
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328796
Reviewed-by: John Stiles <johnstiles@google.com>
Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Not yet used as of this CL.
Change-Id: Ic82ab5e2e2ca17fb11c16e22cfa6b7ad5ff74c77
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328657
Reviewed-by: John Stiles <johnstiles@google.com>
Commit-Queue: Ethan Nicholas <ethannicholas@google.com>
This is conceptually very similar to http://review.skia.org/328384, but
the inliner doesn't use `clone()` when it clones a node.
Change-Id: I7456b63687ce2f93a7980fb101dfc97e143a378f
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328817
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>
After an IRNode is cloned, callers expect to be able to safely mutate
its SymbolTable, so it can't be left with a built-in one.
Change-Id: If658fd11ad580da552f9d689edeeed4c842b38c9
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328384
Commit-Queue: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Change-Id: I28b01a83960b6f7c8e715a817dc75a2408465a26
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328658
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 new method receives its arguments in the same order as the
underlying C system call. This CL should not change behavior.
Change-Id: I6fd1e497b19d38d1133b1b8187146b98131093a8
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328624
Commit-Queue: Adam Barth <abarth@google.com>
Commit-Queue: Mike Klein <mtklein@google.com>
Auto-Submit: Adam Barth <abarth@google.com>
Reviewed-by: Mike Klein <mtklein@google.com>
This functionality will also be needed in TriangulatingPathOp::onPrePrepareDraws
Bug: 1108408
Change-Id: I0bf3f3c14820b946f4c4bc738c55e7a5eb2e7ced
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328556
Reviewed-by: Adlai Holler <adlai@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
This has a slight ripple effect into Enum, as it was using the builtin
status as an indicator that the enum was shared with C++ code. This now
has a dedicated bool flag.
Change-Id: Id03efa902546775666acd031e6d57123e02b6c6e
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328381
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
This is everything except for literally removing the class.
Change-Id: I2f16caf865d1bcf9c0f267aed73313c0676a73bb
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/327222
Commit-Queue: Adlai Holler <adlai@google.com>
Reviewed-by: Robert Phillips <robertphillips@google.com>
Reviewed-by: Brian Salomon <bsalomon@google.com>
Our Metal codegen assumes that out params are pointers, but Metal's
built-in frexp actually takes a reference for the exponent, not a
pointer. We now add in a helper function to translate.
Change-Id: I24686347d07151dd99a1ff1c43aff2b35c3181e5
Bug: skia:10762
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328387
Reviewed-by: Jim Van Verth <jvanverth@google.com>
Commit-Queue: Jim Van Verth <jvanverth@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Change-Id: Iff707ae968b84609acde89ba94dec7a5d53c7525
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328576
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 should fix the macOS fuzzer build.
Bug: chromium:1139725
Change-Id: I14090da4ee7d9d0a6e515b05c23c0a1e50ca4e5f
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328385
Auto-Submit: Michael Ludwig <michaelludwig@google.com>
Reviewed-by: Florin Malita <fmalita@chromium.org>
Commit-Queue: Michael Ludwig <michaelludwig@google.com>
For posterity, here's my initial, wrong thinking:
If we squint at "return foo" and read it as "result = foo; goto
end_of_program", and then remember we can always skip forward jumps (and
where's further forward than end of program?), early returns turn out to
work just like a store.
The reason this is wrong is that by the time we reach a final return,
the entire mask stack has been popped back down to its original default
ffffffff (active) state. But that return shouldn't override any prior
returns. So that scheme isn't quite right.
Instead we accumulate the result by disabling updates to lanes that have
already returned. By the time we're done, all lanes should have hit
_some_ active return, now asserted.
Bug: skia:10852
Change-Id: I27b05f04a60ff4a5f2fe5f59bf398c3f7224a41b
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/327457
Commit-Queue: Mike Klein <mtklein@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
We will need this functionality in TriangulatingPathOp::onPrePrepareDraws as well as at the
current locations
Bug: 1108408
Change-Id: I2a65f07f47a549531d84dbf2afd82ad9d9b35225
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328536
Reviewed-by: Adlai Holler <adlai@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
This is useful because we can clone FunctionDefinitions without cloning
the matching FunctionDeclaration. The FunctionDeclaration will remain a
builtin, but the definition should be a malleable clone.
Change-Id: Icfc1e0855fb8fcd6914a5d657f5098986fcf19ea
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328396
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Using `return` instead of assigning to the output color removes an
unnecessary temporary variable from the output.
Change-Id: Ica31e290f8745a7309ae32c7148516d2189308ea
Bug: skia:10549
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328386
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
This adds the budgeted parameter to the GrAttachment ctors. Currently
we only have stencil and msaa attachments which are always budgeted
but this will soon change as more things get added.
Along the same lines this fixes the gpu memory size calculate on
render target. The msaa attachment was getting double counted in
the RT and the attachment itself.
Bug: skia:10727
Change-Id: I3520de9627eadaa4074f7425df509a6c1ccbe07f
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/327337
Reviewed-by: Robert Phillips <robertphillips@google.com>
Commit-Queue: Greg Daniel <egdaniel@google.com>
Bug: skia:10419
Change-Id: I74429ceb1fc03e68872c74eb62f64675e1bd55cc
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/327476
Reviewed-by: Michael Ludwig <michaelludwig@google.com>
Commit-Queue: Chris Dalton <csmartdalton@google.com>
Was missing this virtual so when we were registering attachments in the
cache we were never setting a scratch key on them.
Bug: skia:10727
Change-Id: Id505fcea17ed83b9f9f33739ded518adbcc2acb7
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/326440
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Greg Daniel <egdaniel@google.com>
This causes a ~4% regression on sksl_large, but some of that
can be bought back in two ways:
1) Removing (now unnecessary) cloning of program elements
2) Hoisting the new analysis passes, with (nontrivial)
logic to update/maintain the call counts as we edit IR.
Also, this fixes bugs where we were emitting functions that
had "calls" from no-longer called functions.
Bug: skia:10776
Change-Id: I4f8c29957be2e4233a883c9a1125f363b82ee40c
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/327198
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: John Stiles <johnstiles@google.com>
Previously, we'd add UnresolvedFunction symbols wherever the original
FunctionDeclaration was found, but that can alter shared SymbolTables
(e.g. the root table) and the node will not be cleaned up when the
program is deleted. The pooling system expects that any node created as
part of a Program will go away when the Program does.
This change puts the UnresolvedFunction symbol as close to the
FunctionDeclaration as possible without changing builtin symbol tables.
Change-Id: I4c80cfba734047f11b50c86829d95d8e393c3060
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/327921
Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
This is desired for a short term memory improvement on low memory
devices.
Bug: chromium:1138979
Change-Id: I7df41a9c4d21b7a7f62b738e6358b36dd262f77d
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/327345
Commit-Queue: Kyle Charbonneau <kylechar@chromium.org>
Commit-Queue: Brian Salomon <bsalomon@google.com>
Reviewed-by: Brian Salomon <bsalomon@google.com>
This flag is set on anything rehydrated and on the root symbol table;
that is, anything that will outlive your Program's tree.
Change-Id: Idd9a167ee69f1fb9e526aecbee733ce1ccb5d265
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/327920
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
By the time we do this check, we've consumed all the bytes for the
branchIfAllFalse instruction, so ip is pointing at the next instruction.
If that next instruction is itself the jump target, we classify that as
a backward jump instead of the zero-sized forward jump that it is.
Switching to >= fixes that.
Change-Id: I537131bfb0d213c8407734184b78a510624d60c5
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/327458
Auto-Submit: Mike Klein <mtklein@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
