In preparation for moving dm_flags out. Important behavioral change to
note: now we're passing issue, patchset, and patch_storage flags for all
runs of DM, including non-tryjobs. I don't think it should change any
behavior, but it's something to watch out for.
Change-Id: Ie45dbfa606b4325dc94c855dbc3531c7eeaacf9b
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/274051
Commit-Queue: Eric Boren <borenet@google.com>
Reviewed-by: Ben Wagner aka dogben <benjaminwagner@google.com>
Change-Id: Ic4b0d800ba38bd0d9548b6323ac845c5aa398e6d
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/274050
Reviewed-by: Ben Wagner aka dogben <benjaminwagner@google.com>
Commit-Queue: Eric Boren <borenet@google.com>
This generally makes things tidier (e.g. perf_skottietrace.py no longer
needs special logic so that the app is only installed once), and it
means iOS can install only the app that's going to be run instead of
installing everything.
Change-Id: Ic07c56e724d0379255c8da5f204d9d6379b646ec
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/273002
Commit-Queue: Ben Wagner aka dogben <benjaminwagner@google.com>
Auto-Submit: Ben Wagner aka dogben <benjaminwagner@google.com>
Reviewed-by: Eric Boren <borenet@google.com>
Reviewed-by: Kevin Lubick <kjlubick@google.com>
- add Mac PDF rasterizer
- create a mac pdf bot
- blacklist some GMs that CG takes forever to rasterize
- remove obsolete --dont_write pdf feature
This new Mac -PDF bot is now the _only_ bot running PDF code.
Obviously we can expand from here.
pdfium was looking like a pain to build and/or integrate
into our own build. pdftocairo looked promising, and is
already installed on the bots, but it is very slow and
prone to error. CG was the next thing I tried.
Change-Id: I82b04121f484e7dd78f60a648485a09218dd5279
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/267810
Commit-Queue: Mike Klein <mtklein@google.com>
Reviewed-by: Kevin Lubick <kjlubick@google.com>
Reviewed-by: Mike Reed <reed@google.com>
An error from addr2line (e.g. unknown binary type) can truncate the
output, hiding the remainder of the stack trace. Instead print the
original line.
Change-Id: I563aae4333a79a17560378e399e79b60c79f9ac7
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/261288
Commit-Queue: Kevin Lubick <kjlubick@google.com>
Auto-Submit: Ben Wagner aka dogben <benjaminwagner@google.com>
Reviewed-by: Kevin Lubick <kjlubick@google.com>
It seems like we see quite a few "force quarantined" bots when pushing
resources to the device fails. Rewrite copy_directory_contents_to_device
to use the _adb method, which automatically retries.
Change-Id: Iae36f8e42e85a58fdddc8c9dc80e693a371fab8b
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/248560
Commit-Queue: Eric Boren <borenet@google.com>
Auto-Submit: Ben Wagner aka dogben <benjaminwagner@google.com>
Reviewed-by: Eric Boren <borenet@google.com>
We need to specify attribute locations and (sometimes) frag shader
output locations. Desktop GL worked fine without this, but Flutter
ran into this problem, and a Pixel 2 reproduced the issue.
Note that both APIs (BindFragDataLocation and BindAttribLocation)
don't take effect until the next time the program is linked, so
we have to relink the program after applying those changes.
I was afraid that re-linking would eliminate the perf benefits of
pre-compiling the shaders, but (at least on Pixel 2) that's not
the case. I traced the life of a single program, and the initial
link (during precompile) was 4.4 ms. The re-link took 0.23 ms.
Change-Id: Iadb3b425a8cf9f6a52e015c2e37f875c0fd73d6d
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/241758
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
