We thought it'd be handy to have swap_src_dst so that it could be used
either to move dst into src or src into dst. Turns out, we already have
a stage that moves dst into src (called "dst", the dst blend mode), so
there's really no reason to have swap_src_dst over the strictly more
efficient move_src_dst.
swap_src_dst is typically 12 register moves, where move_src_dst is 4.
BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=4461
Change-Id: Ib453775f854e313f823851978eaadc3995872312
Reviewed-on: https://skia-review.googlesource.com/4461
Reviewed-by: Matt Sarett <msarett@google.com>
Commit-Queue: Mike Klein <mtklein@chromium.org>
The simplest thing to do here is just run shader+color filter pipeline at
construction time to create a new constant color shader (replacing the paint
color).
This reduces a pipeline like:
- constant_color (paint color)
- matrix_4x5
- clamp_a
- load_d_foo, xfermode, lerp, store_foo
to
- constant_color (paint color -> matrix_4x5 -> clamp_a)
- load_d_foo, xfermode, lerp, store_foo
To implement this all, we add a new store_f32 stage that writes SkPM4f, and
finally get around to implementing Sk8f::Store4() (store while reinterlacing).
Sk4f::Store4() already exists for both SSE and NEON.
Next step: reduce simple constant_color -> store pipelines (src mode, full
coverage) into non-pipeline memsets.
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2480823002
Review-Url: https://codereview.chromium.org/2480823002
Remove subdir thing for Linux. It's incorrect.
Make the script able to take extra GN args.
Default to a custom gn output dir for the command buffer build.
Default to a release build of the command buffer.
Document that the script overwrites the gn args on each run.
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2472343003
Review-Url: https://codereview.chromium.org/2472343003
Adds a clamp for GrTexture filtering that can be set by a subclass at construction. The clamping is performed by GrTextureParams. GrGLTexture limits filtering to bilinear for rectangle and external textures.
Also moves samplerType() to GrTexturePriv from GrTexture.
BUG=skia:5932
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=4352
Change-Id: I1f023d4f4133e7eb393367580c0558257e56c8db
Reviewed-on: https://skia-review.googlesource.com/4352
Commit-Queue: Brian Salomon <bsalomon@google.com>
Reviewed-by: Chris Dalton <csmartdalton@google.com>
This should not change any behavior, but addresses some readability
concerns.
Change-Id: I2614beaed7b5722a3af2c49ab8f4ac16b56aa3a7
Reviewed-on: https://skia-review.googlesource.com/4393
Reviewed-by: Herb Derby <herb@google.com>
Commit-Queue: Ben Wagner <bungeman@google.com>
CQ_INCLUDE_TRYBOTS=master.client.skia.compile:Build-Mac-Clang-Arm7-Debug-iOS-Trybot
BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=4351
Change-Id: I71f77aaad36701eca7a013959e751a843c6f5e3d
Reviewed-on: https://skia-review.googlesource.com/4351
Commit-Queue: Mike Klein <mtklein@chromium.org>
Reviewed-by: Mike Klein <mtklein@chromium.org>
It is no longer used.
Change-Id: Ie2f9a39a4295005cb39bdf2f8fc15542ee75d207
Reviewed-on: https://skia-review.googlesource.com/4386
Commit-Queue: Ben Wagner <bungeman@google.com>
Commit-Queue: Mike Klein <mtklein@chromium.org>
Reviewed-by: Mike Klein <mtklein@chromium.org>
"Remove SkAutoTDelete." did not run trybots on these specific bots.
Change-Id: Ibfa731df387a90a78187b88c75483800981a691c
Reviewed-on: https://skia-review.googlesource.com/4387
Reviewed-by: Ben Wagner <bungeman@google.com>
Commit-Queue: Ben Wagner <bungeman@google.com>
This doesn't create any apps or bundles or sign anything, but it all compiles and links.
Note the awkward transitional hack I used to make each tool's tool_main() serve as the real main() again when built with GN, while keeping the existing setup with GYP. Fun...
BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=4350
Change-Id: I632753d5d8e5848380854f413bf5905d676bfcf4
Reviewed-on: https://skia-review.googlesource.com/4350
Reviewed-by: Jim Van Verth <jvanverth@google.com>
Commit-Queue: Mike Klein <mtklein@chromium.org>
This stems from a behavior difference between Skia and Chrome.
In Skia, we want to write transparent pixels as often as possible.
(It's faster than checking if we should skip each pixel.)
In Chrome, they avoid writing transparent pixels unless
absolutely necessary.
We were cautious about changing behavior when this first landed,
but this is easier to think about in a smaller change (right now).
(1) We can always write transparent pixels when we are writing
an independent frame.
(2) There is no need for the progressiveDisplay() check. We
only ever use progressive display methods on the first
frame - and the first frame is always independent.
BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=4379
Change-Id: I82048a08e2003aac216f483c7db8df997b687149
Reviewed-on: https://skia-review.googlesource.com/4379
Commit-Queue: Matt Sarett <msarett@google.com>
Reviewed-by: Leon Scroggins <scroggo@google.com>
I was looking at the disassembly of matrix_4x5() and noticed it didn't have any FMAs. This makes things that call SkNx_fma() actually use the FMA instruction.
BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=4400
CQ_INCLUDE_TRYBOTS=master.client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot
Change-Id: Ia353a77b0ca14385a43b564997b05586f9472996
Reviewed-on: https://skia-review.googlesource.com/4400
Reviewed-by: Matt Sarett <msarett@google.com>
Commit-Queue: Mike Klein <mtklein@chromium.org>
Replace with std::unique_ptr.
Change-Id: I5806cfbb30515fcb20e5e66ce13fb5f3b8728176
Reviewed-on: https://skia-review.googlesource.com/4381
Commit-Queue: Ben Wagner <bungeman@google.com>
Reviewed-by: Mike Klein <mtklein@chromium.org>
When downscaling with kMediumQuality in non-clamp mode, if we happen to hit a size == mip level, then SkBitmapProcInfo::init takes the trivial matrix path and doesn't set up a matrix for normalized coords on the assumption that we're going to ignore the transform. But kMediumQuality disables allow_ignore_fractional_translate, so we take the filter path after all - but with an incorrect matrix.
R=reed@google.com
BUG=chromium:661180
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2476713002
Review-Url: https://codereview.chromium.org/2476713002
We've just re-noticed this can happen:
1) we have a properly premultiplied linear color;
2) we convert that to sRGB;
3) we convert that back to linear;
4) that color does not appear to be premultiplied.
Removing sk_linear_to_srgb_noclamp(), and thus always clamping to [0,1] here in linear space, does not fix this problem. However, it does help keep it from propagating too badly.
Just double-checked: the older Sk4f pipeline (SkXfermode4f, SkPM4fPriv, etc) already use sk_linear_to_srgb() exclusively, so they're already doing this same clamp.
BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=4314
Change-Id: I6352eb0ba969eb25674e8441e43bb51e1e1c0df3
Reviewed-on: https://skia-review.googlesource.com/4314
Reviewed-by: Yuqian Li <liyuqian@google.com>
Commit-Queue: Mike Klein <mtklein@chromium.org>
Allow configs to be supported as FBO attachments for copies/readbacks without being "renderable" elsewhere in Ganesh.
The motivation for this is to add support for int textures as srcs but not as dsts (at least initially) but to still be able to read them back. This means we don't pay for a unneeded GPU copy when reading back a GrSurface that is a GrTexture and not a GrRenderTarget.
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=4345
Change-Id: I824a73c6b8c1b9634206d76ccf0848d5f9b64441
Reviewed-on: https://skia-review.googlesource.com/4345
Commit-Queue: Brian Salomon <bsalomon@google.com>
Reviewed-by: Robert Phillips <robertphillips@google.com>
This moves the work of finding headers from `gn gen` time into the action itself. We can do this safely now because we're constructing a skia.h.d deps file, which Ninja uses to track if-these-are-dirty-then-this-is-dirty relationships. Everything can now live in one handy find_headers.py.
Upshot is, `gn gen` runs ~50ms faster, and I think the code's clearer this way too.
BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=4065
Change-Id: I2f1e7adf68be8e961eb77596786a795134bbb9a6
Reviewed-on: https://skia-review.googlesource.com/4065
Reviewed-by: Ben Wagner <bungeman@google.com>
Commit-Queue: Mike Klein <mtklein@chromium.org>
- when the shader is opaque, srcover becomes src
- don't load dst when blitting in src mode with full coverage
- fold coverage into src alpha when in srcover mode
It's not obvious that we can fold coverage into src alpha when using a 565 mask, so I've not attempted that. What would we do about alpha?
Over all GMs this causes a single 1-bit difference in sRGB mode. No 565 or f16 diffs.
BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=4349
CQ_INCLUDE_TRYBOTS=master.client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot
Change-Id: Ib9161f498c4efa6b348ca74522166da64d09a7da
Reviewed-on: https://skia-review.googlesource.com/4349
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: Mike Klein <mtklein@chromium.org>
This breaks the color filter down into a couple logical steps:
- go to unpremul
- apply the 4x5 matrix
- clamp to [0,1]
- go to premul
Because we already have handy premul clamp stages, we swap the order of clamp and premul. This is lossless.
While adding our stages to the pipeline, we analyze the matrix to see if we can skip any steps:
- we can skip unpremul if the shader is opaque (alphas are all 1 ~~~> we're already unpremul);
- we can skip the premul back if the color filter always produces opaque (here, are the inputs opaque and do we keep them that way, but we could also check for an explicit 0 0 0 0 1 alpha row);
- we can skip the clamp_0 if the matrix can never produce a value less than 0;
- we can skip the clamp_1 if the matrix can never produce a value greater than 1.
The only thing that should seem missing is per-pixel alpha checks. We don't do those here, but instead make up for it by operating on 4-8 pixels at a time.
We don't split the 4x5 matrix into a 4x4 and 1x4 translate. We could, but when we have FMA (new x86, all ARMv8) we might as well work the translate for free into the FMAs.
This makes gm/fadefilter.cpp draw differently in sRGB and F16 modes, bringing them in line with the GPU sRGB and GPU f16 configs. It's unclear to me what was wrong with the old CPU implementation.
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=4346
CQ_INCLUDE_TRYBOTS=master.client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot
Change-Id: I14082ded8fb8d63354167d9e6b3f8058f840253e
Reviewed-on: https://skia-review.googlesource.com/4346
Reviewed-by: Mike Reed <reed@google.com>
Commit-Queue: Mike Klein <mtklein@chromium.org>
