This code originally contained 16 extremely similar versions of the same
logic. (set1i, 1iv, 1f, 1fv, 2i, 2iv...)
Now the core logic has been boiled down to two (set, setv) templated
helper functions, one for array and one for non-array uniforms. Template
arguments fill in the minor variations between each type.
Structuring the logic in this way makes it easier to perform sweeping
changes as needed, and guarantees that the setters all work the same.
Change-Id: Iaa74d25c612b12ab26d7f32879c174bca6c850c1
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/462099
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
This allows GrQuad to be written as a quad without the header needing
knowledge of it.
Bug: skia:12524
Change-Id: I22f67021877de426f710981d4fa2bd48c88a4d52
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/462319
Commit-Queue: Chris Dalton <csmartdalton@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Bug: skia:12559
Change-Id: I5775120a826e582e73aab83273e59bc48c057a5b
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/462077
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: John Stiles <johnstiles@google.com>
The lexer uses a matrix of state transition data to parse text. This
matrix was highly sparse and contained large runs of zeros and repeated
values. I found that by transposing the matrix, it became extremely
compressible. In its transposed state, most slices had only 1-3 unique
nonzero values.
This CL leverages this finding to reduce the matrix data from ~55K to
~10K. A handful of slices do contain a large number of unique values and
continue to be represented as plain 16-bit arrays. Some slices contain
no data at all and were eliminated entirely. The majority of slices are
now represented by a compact two-bit array. Bit pattern 00 always
represents zero. Bit patterns 01, 10 and 11 are translated into values
v0, v1 and v2, which are stored per slice (bit-packed to save a few
extra bytes). This transformation involves a fair amount of bit shifting
and masking, but the generated code will be quite efficient:
https://godbolt.org/z/eTvjr96ez
Change-Id: Iecc67aadd510ccf63b4bcb11ed861d703efefaae
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/461356
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
Bug: skia:12466
Change-Id: I5c0c919d4523258a11b4545346093953a4f09c7e
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/461597
Reviewed-by: Michael Ludwig <michaelludwig@google.com>
Commit-Queue: Jim Van Verth <jvanverth@google.com>
We can use `if constexpr` to avoid the need for a helper function that
uses template specialization, and we can use `getBufferPtrAndMarkDirty`
instead of performing the equivalent operations by hand.
Change-Id: I548c719bae4f215e0a9c658a942deaec38ce27ad
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/462098
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Jim Van Verth <jvanverth@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
This helps remove all Ganesh dependencies from GrVertexWriter so it can
be shared with Graphite.
Bug: skia:12466
Change-Id: I1ede1cfc48d501cb1e028bca1e8590e6c92f4e4d
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/462316
Commit-Queue: Chris Dalton <csmartdalton@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Jul started. It does not work yet.
Change-Id: Ie2c3d943f281dae86befff64f7342c097a5bc7b9
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/455257
Bot-Commit: Rubber Stamper <rubber-stamper@appspot.gserviceaccount.com>
Commit-Queue: Julia Lavrova <jlavrova@google.com>
At this point, it seems like this was a mis-diagnosis of the underlying
issue around dual-source blending (and its interaction with other blend
state).
Change-Id: I11af0c9b70c32e14c353848db3d6adbfe5f08225
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/462176
Reviewed-by: Greg Daniel <egdaniel@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Design doc: http://go/skia-android-hwasan.
This CL defines the following tasks:
- Build-Debian10-Clang-arm64-Debug-Android_HWASAN
- Test-Android-Clang-Pixel4a-CPU-Snapdragon730G-arm64-Debug-All-Android_HWASAN
- Test-Android-Clang-Pixel4a-GPU-Adreno618-arm64-Debug-All-Android_HWASAN
The HWAddress Sanitizer is only available on 64-bit Arm devices, and requires Android 10+ built with HWAddress Sanitizer support. There are prebuilt HWASan images for supported Pixel devices at https://ci.android.com (instructions at https://developer.android.com/ndk/guides/hwasan).
To support these tasks, we added two Pixel 4a devices to the tree, flashed with the corresponding prebuilt HWASan image. The specific Pixel devices were chosen based on stock availability.
See patchset 11 for an example error I introduced in //gm/3d.cpp, which was caught by the HWAddress Sanitizer.
Bug: skia:12099
Change-Id: Ic3847abc61c8a1bf686d29b97ef9d89b9bb5eb86
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/436572
Reviewed-by: Joe Gregorio <jcgregorio@google.com>
Commit-Queue: Leandro Lovisolo <lovisolo@google.com>
Bug: skia:12466
Change-Id: I7be58aca7eedf18e4157d9473216785eead8c89a
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/461700
Commit-Queue: Jim Van Verth <jvanverth@google.com>
Reviewed-by: Robert Phillips <robertphillips@google.com>
After reporting the error, we convert the reserved word to an identifier
as this led to the best error reporting. (This avoids double error
reporting or strange cascading errors.)
Change-Id: I67209bc342fe794287baeaaaf34fa77afd4ac26b
Bug: skia:12560
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/462096
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
This can successfully build a C library:
bazel build --config=clang //third_party:libpng
This can build and run a statically-linked executable:
bazel test --config=clang //:bazel_test
For more verbose compile and linking output, add the
`--features diagnostic`
flag to a Bazel command (see _make_diagnostic_flags() in
toolchain/clang_toolchain_config.bzl. Similarly, a
`--features print_search_dirs` can be used to show where
clang is looking for libraries etc to link against.
These features are made available for easier debugging.
Suggested review order:
- Read https://docs.bazel.build/versions/4.2.1/tutorial/cc-toolchain-config.html
if unfamiliar with setting up C++ toolchains in Bazel
- .bazelrc and WORKSPACE.bazel that configure use and download
of the toolchain (Clang 13, musl 1.2.2)
- toolchain/build_toolchain.bzl which downloads and assembles
the toolchain (w/o installing anything on the host machine)
- toolchain/BUILD.bazel and toolchain/*trampoline.sh to see
the setup of the toolchain rules.
- toolchain/clang_toolchain_config.bzl to see the configuration
of the toolchain. Pay special attention to the various
command line flags that are set.
- See that tools/bazel_test.cc has made a new home in
experimental/bazel_test/bazel_test.cpp, with a companion
BUILD.bazel. Note the addition of some function calls
that test use of the C++ standard library.
The number being used to test the PNG library is the latest
and greatest that verifies we are compiling the one brought
in via DEPS (and not a local one).
- third_party/* to see how png (and its dependent zlib) have
been built. Pay special attention to the musl_compat hack
to fix static linking (any idea what the real cause is?)
- //BUILD.bazel to see definition of the bazel_test executable.
Change-Id: I7b0922d0d45cb9be8df2fd5fa5a1f48492654d5f
Bug: skia:12541
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/461178
Reviewed-by: Ben Wagner <bungeman@google.com>
Reviewed-by: Leandro Lovisolo <lovisolo@google.com>
Bug: skia:12559
Change-Id: I91e0b46702ca361737459d64d1cdb57f8dd2f64e
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/461697
Reviewed-by: Greg Daniel <egdaniel@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
In places where we use GR_METAL_SDK_VERSION to determine the contents of
a class, it's important to consistently #include the header which sets
the value of GR_METAL_SDK_VERSION.
Change-Id: Ic4824ff36c982d3493ebec03dd38465bb90b287a
Bug: skia:12513
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/461836
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Jim Van Verth <jvanverth@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
Commit-Queue: Jim Van Verth <jvanverth@google.com>
Bug: skia:12466
Change-Id: I7eb240985ce0c267bbc7be040e0f812407d31435
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/461699
Commit-Queue: Jim Van Verth <jvanverth@google.com>
Commit-Queue: Greg Daniel <egdaniel@google.com>
Auto-Submit: Jim Van Verth <jvanverth@google.com>
Reviewed-by: Greg Daniel <egdaniel@google.com>
Except that applyClipToDraw doesn't actually evaluate any clip shapes,
and the BoundsManager is the slow/naive one, I think all the changes to
Device fully implement the tracking/work needed for color/depth sorting.
Bug: skia:12466
Change-Id: I74f38e56d34d822c2be35f1ebedfb7c00b87f04c
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/460616
Reviewed-by: Chris Dalton <csmartdalton@google.com>
Commit-Queue: Michael Ludwig <michaelludwig@google.com>
Bug: skia:12561
Change-Id: I0a739dcfd4214ee4855f04a70fb6f66ab638bf2e
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/461478
Reviewed-by: Robert Phillips <robertphillips@google.com>
Commit-Queue: Jim Van Verth <jvanverth@google.com>
This class is only meant to be used with Wang's formulas, so it
belongs in their name namespace.
Bug: skia:12524
Change-Id: Ib1667781b947a7f052af103b7bf9084d09a690de
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/461567
Commit-Queue: Chris Dalton <csmartdalton@google.com>
Reviewed-by: Robert Phillips <robertphillips@google.com>
Tessellation code can just exist in the skgpu namespace.
Bug: skia:12524
Change-Id: I44d473c5f82d608bb6cd3bb9df239c4f4028da4f
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/461557
Reviewed-by: Greg Daniel <egdaniel@google.com>
Commit-Queue: Chris Dalton <csmartdalton@google.com>
Thise should have been aligned properly already due to their
placement, but it doesn't hurt to make it more explicit.
Change-Id: Idb6a03e93d86df7727324987f166d197d5e708a4
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/461377
Commit-Queue: Chris Dalton <csmartdalton@google.com>
Reviewed-by: Herb Derby <herb@google.com>
Bug: skia:12466
Change-Id: Id0419836ca96d99836879d509ca2cd1c9244d76f
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/461598
Commit-Queue: Greg Daniel <egdaniel@google.com>
Commit-Queue: Jim Van Verth <jvanverth@google.com>
Auto-Submit: Greg Daniel <egdaniel@google.com>
Reviewed-by: Jim Van Verth <jvanverth@google.com>
This lets clients access and use the same RGB to YUV matrices that we
use internally.
Inspired by https://crrev.com/c/3223983
Bug: skia:12545
Change-Id: I7f70f56e721819c3d33da447eff2e572b7774c51
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/461476
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: Robert Phillips <robertphillips@google.com>
Use the coverage from the mask to avoid or reduce blitting.
nanobench
desk_chalkboard 51ms -> 38ms.
Change-Id: I58e3ca5f9ef59dbd5b9ceb1a905ea1f3d948fd89
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/461296
Reviewed-by: Mike Reed <reed@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: Herb Derby <herb@google.com>
Bug: skia:12466
Change-Id: I2e0421191d05c3e0f457425e0dab49c2afb9c2b9
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/461477
Commit-Queue: Jim Van Verth <jvanverth@google.com>
Commit-Queue: Chris Dalton <csmartdalton@google.com>
Auto-Submit: Jim Van Verth <jvanverth@google.com>
Reviewed-by: Chris Dalton <csmartdalton@google.com>
Bug: skia:12466
Change-Id: I44bd3899dea817318f83444f023c1c8adcfc2d1e
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/459881
Commit-Queue: Greg Daniel <egdaniel@google.com>
Reviewed-by: Jim Van Verth <jvanverth@google.com>
While we won't be transforming shapes, we do need an accurate way to
transform bounding boxes to device space. Right now this just uses
SkM44's mapRect function and converts to/from an SkRect. However,
under the hood SkM44 uses a similar negation trick as skgpu::Rect so
I need to explore if using Rect's internal structure directly will let
us compute the bounding box with fewer operations.
Additionally, SkM44's mapRect clips to a fixed w=epsilon plane, and I
think Transform::mapRect should just do the right thing and handle
w clipping/projection as accurately as possible.
Bug: skia:12466
Change-Id: I08c9df6c40cbcdcd8f622baae222bdf361d59070
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/460578
Commit-Queue: Michael Ludwig <michaelludwig@google.com>
Reviewed-by: Chris Dalton <csmartdalton@google.com>
Switches Shape to store kRect types as skgpu::Rect instead of SkRect,
although it can be constructed or set from an SkRect. It returns a Rect
from rect() and bounds() now.
Also switches out the line to be constructed from float2's, and re-use
the fRect storage to store the first segment vertex in topLeft and
the second vertex in botRight.
Bug: skia:12466
Change-Id: I1032fb6f222b0853305186c0e34091b014ae0cea
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/460577
Reviewed-by: Chris Dalton <csmartdalton@google.com>
Commit-Queue: Michael Ludwig <michaelludwig@google.com>
This step was missing from the previous CL.
Bug: skia:12466
Change-Id: I3da0082efefbabf6008fc8962d706f75f0743ecc
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/461176
Auto-Submit: Jim Van Verth <jvanverth@google.com>
Reviewed-by: Greg Daniel <egdaniel@google.com>
Commit-Queue: Jim Van Verth <jvanverth@google.com>
Implements a BSP tree with NEON/SSE optimizations that tracks
non-overlapping regions. This object can be used batch sets of paths
into non-overlapping draws. Performance on AppleM1 with our existing
benchmarks looks very promising:
desk_motionmarkarcs.skp 1227 paths -> 69 draws 450us
desk_motionmarksuits.skp 1417 paths -> 26 draws 201us
desk_chalkboard.skp 1940 paths -> 11 draws 84us
desk_ynevsvg.skp 859 paths -> 10 draws 31us
desk_micrographygirl.skp 318 paths -> 29 draws 11us
Bug: skia:12466
Change-Id: I847a93ed726dea10cb403cb76e578bd81eb920d2
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/460298
Commit-Queue: Chris Dalton <csmartdalton@google.com>
Reviewed-by: Herb Derby <herb@google.com>
Halfs are lacking sufficient precision for this effect.
Change-Id: If3c2cac34d465d74b7ad8228417c42950f48adfd
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/461177
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: Florin Malita <fmalita@google.com>
Bug: skia:12466
Change-Id: I89d41509b8158aabac5250f035115954e8ae7f4b
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/460655
Commit-Queue: Chris Dalton <csmartdalton@google.com>
Reviewed-by: Michael Ludwig <michaelludwig@google.com>
Makes the path sniffing code from BisectSlide reusable.
Change-Id: I21c1e752590359c557c35804e5d5044d8041a308
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/460637
Reviewed-by: Jim Van Verth <jvanverth@google.com>
Commit-Queue: Chris Dalton <csmartdalton@google.com>
