Now that we've got Flutter builds I'm not very interested in the old
-Mini bot, which was always a proxy for something like Flutter's build.
The NoDEPS bot shouldn't have to disable effects, I think.
Change-Id: I57abca5935b7da3917eb804cda10967af0a18f41
Reviewed-on: https://skia-review.googlesource.com/c/169221
Reviewed-by: Kevin Lubick <kjlubick@google.com>
Commit-Queue: Mike Klein <mtklein@google.com>
We currently have only one bot for these jobs, so we should treat them
as experimental for now.
Bug: 837351
Change-Id: I1ea9f1bf1635c36895eede80df6709509a85df90
Reviewed-on: https://skia-review.googlesource.com/c/168489
Commit-Queue: Ben Wagner <benjaminwagner@google.com>
Auto-Submit: Ben Wagner <benjaminwagner@google.com>
Reviewed-by: Eric Boren <borenet@google.com>
I suspect these were fixed by the driver upgrade in
https://skia-review.googlesource.com/c/skia/+/166621
Bug: skia:6398, skia:6863
Change-Id: I9079454b385b209e0c4bf416063ec01883045ab1
Reviewed-on: https://skia-review.googlesource.com/c/167390
Auto-Submit: Ben Wagner <benjaminwagner@google.com>
Commit-Queue: Greg Daniel <egdaniel@google.com>
Reviewed-by: Greg Daniel <egdaniel@google.com>
Bug: chromium:897031
Bug: chromium:896776
Prior to this fix, we would treat the output from such a JPEG
as if it were a 16 bit per component RGB PNG. We hit an assert
in debug, but in release mode we do the wrong thing.
Split up SkSwizzler::CreateSwizzler into two public factories
(and a private one) based on whether format conversion is desired.
Without format conversion, we may have already converted (as is
the case with this JPEG), so the SkEncodedInfo::Color is not relevant.
That flavor of the factory just needs to know the bytes per pixel,
so provide that info instead.
Add a test file to Google Storage: apron.jpg, from Chromium's
benchmark files.
Change-Id: If1337d58a508466299f9e4666778727c6cdc879a
Reviewed-on: https://skia-review.googlesource.com/c/164619
Auto-Submit: Leon Scroggins <scroggo@google.com>
Commit-Queue: Mike Klein <mtklein@google.com>
Reviewed-by: Mike Klein <mtklein@google.com>
Bug: skia:5774
This recently added image is incomplete, and CG leaves memory
uninitialized. Stop testing it to avoid seeing random images in Gold.
Change-Id: I512c3af6d6df6eedbea96ea5d809cdd5b475833a
Reviewed-on: https://skia-review.googlesource.com/c/165302
Commit-Queue: Leon Scroggins <scroggo@google.com>
Commit-Queue: Eric Boren <borenet@google.com>
Auto-Submit: Leon Scroggins <scroggo@google.com>
Reviewed-by: Eric Boren <borenet@google.com>
Pubsub errors are causing lots of flaky bot failures
Bug: skia:
Change-Id: I2fc6d57d296899faea36d37dd93fa1e381cea199
Reviewed-on: https://skia-review.googlesource.com/c/164048
Commit-Queue: Eric Boren <borenet@google.com>
Commit-Queue: Kevin Lubick <kjlubick@google.com>
Auto-Submit: Eric Boren <borenet@google.com>
Reviewed-by: Kevin Lubick <kjlubick@google.com>
Distance field glyphs draw slightly differently based on where the
glyph lands in the atlas.
Change-Id: If09f95e3a707db03a658b251906c912ba504cab7
Reviewed-on: https://skia-review.googlesource.com/c/163540
Commit-Queue: Mike Klein <mtklein@google.com>
Reviewed-by: Mike Klein <mtklein@google.com>
The bloaty here was compiled with this patch:
https://github.com/google/bloaty/pull/149
Hopefully that lands upstream and we can track master again.
This adds BuildStats.+Debug because we need symbols
to get sensical data. Bloaty's WASM support is experimental
and currently doesn't support having a stripped (Release)
version be profiled using the symbols of a Debug version.
This means that the buildStats for debug will be higher
than actual, but hopefully the absolute positioning will be
the same and thus the outputs useful.
Bug: skia:
Change-Id: Id7bf721843e8c52a0aae2b7e57ff95397693b3dd
Reviewed-on: https://skia-review.googlesource.com/c/163256
Reviewed-by: Mike Klein <mtklein@google.com>
Commit-Queue: Kevin Lubick <kjlubick@google.com>
The platform generators seem to leave memory uninitialized on these
files without reporting an error. Blacklist them so we do not get lots
of random images in Gold
Change-Id: If5f2b65d7a20c8a3c71716591d194556bdbf17a5
Reviewed-on: https://skia-review.googlesource.com/c/163254
Commit-Queue: Leon Scroggins <scroggo@google.com>
Commit-Queue: Kevin Lubick <kjlubick@google.com>
Auto-Submit: Leon Scroggins <scroggo@google.com>
Reviewed-by: Kevin Lubick <kjlubick@google.com>
Of note, the perf results reported are not directly
comparable CPU->GPU because the GPU ones are likely
emulated (no real gpu in the Docker container on an
GCE VM).
Bug: skia:
Change-Id: I52259085f4d9e22c45b67f3e8ce1211a5c6c6d3e
Reviewed-on: https://skia-review.googlesource.com/c/163126
Reviewed-by: Kevin Lubick <kjlubick@google.com>
- Initialize rowsDecoded in SkSampledCodec. Otherwise,
fillIncompleteImage may be called with an uninitialized
value. This change was originally uploaded to AOSP as
https://android-review.googlesource.com/c/platform/external/skia/+/785816
- If SkPngCodec hits an error, still transform from the
interlace buffer (if needed) and set rowsDecoded properly.
- Do not copy uninitialized memory from the interlace buffer.
- Make BRD treat kErrorInInput like kIncompleteInput. The two errors
are different for the purposes of incremental decode. For a direct
decode, they're essentially the same - part was decoded, but then
the decode was interrupted. This allows testing images with
errors on the bots without reporting a failure.
- Add some images to test incomplete pngs
- errorInInput.png: color_wheel.png truncated with extra 0s at the
end to trigger kErrorInInput
- incInterlaced.png: basi3p02.png (from PNGSuite) similarly modified.
This triggers kIncompleteInput. We have several incomplete PNGs that
we are testing on the bots, but this is the first interlaced
- errorInInputInterlaced.png: This is the same as incInterlaced with
more zeroes at the end. The incorrect CRC (all zeroes) results in
kErrorInInput
Change-Id: I25e1ea917e1c1ca212f104cb9d70290265a51b2d
Reviewed-on: https://skia-review.googlesource.com/c/161822
Commit-Queue: Leon Scroggins <scroggo@google.com>
Reviewed-by: Kevin Lubick <kjlubick@google.com>
Also make a CPU only and GPU only build (although
the latter still has a lot of CPU logic).
Bug: skia:
Change-Id: I857c2300021c2adb5344865c28e4ad3e8d332954
Reviewed-on: https://skia-review.googlesource.com/c/162022
Reviewed-by: Kevin Lubick <kjlubick@google.com>
Previously we didn't do this at all for multiple-texture ops.
Improve the test detecting when filtering can be disabled.
Make draw_image_set GM create tiles with pixel of overlap for correct
filtering.
Add draw_image_set_rect_to_rect to exercise filtering/aa disablement
in combination with tiling.
Makes SkGpuDevice filter out inverted src rects (as is done implicitly
in SkBaseDevice by relying on drawImageRect).
Puts GrTextureOp::fFilter in bitfield.
Change-Id: Iee96cb54d665877c7f4aee422a3a7af2b249b1d6
Reviewed-on: https://skia-review.googlesource.com/c/161641
Commit-Queue: Brian Salomon <bsalomon@google.com>
Reviewed-by: Michael Ludwig <michaelludwig@google.com>
That's what their treemap says and we should try to match it.
Bug: skia:
Change-Id: Iefa95c0440eb6b45a94a72c1beffb8a1199152a4
Reviewed-on: https://skia-review.googlesource.com/c/161825
Commit-Queue: Kevin Lubick <kjlubick@google.com>
Reviewed-by: Mike Klein <mtklein@google.com>
We have a similar ingestion strategy to Gold.
I tried to use something off the shelf like benchmark.js
but passing the PathKit context into the benchmarks was
non-trivial. Plus, making a basic benchmarking tool
ended up being not too hard.
We should be able to re-use the docker container/aggregator
for CanvasKit too.
Bug: skia:
Change-Id: I613dfc58ea57c31cf71566a8ac55f8df9272ad25
Reviewed-on: https://skia-review.googlesource.com/c/161620
Commit-Queue: Kevin Lubick <kjlubick@google.com>
Reviewed-by: Joe Gregorio <jcgregorio@google.com>
Reviewed-by: Stephan Altmueller <stephana@google.com>
