Our "header" reading and writing code didn't agree, so we always failed
to recognize cached program binaries. The asserts in the testing sink
failed to notice, because we did get a 100% cache hit rate, but we
immediately discarded the data we received.
We now also check that we didn't insert anything into the cache, as a
proxy for doing any shader compile work. That change, plus the tweak to
set cached=false when the header fields are invalid (like we do if we
encounter problems further in the blob) detected the problem. Adding the
version tag to the start of the encoded blob fixes the test, and means
that program binary caching is actually working again.
This code still looks (and is) fragile. The next CL is going to rewrite
things to use SkReadBuffer and SkWriteBuffer, make the parsing code less
brittle, and give us a more robust way to detect failure anywhere in the
stream.
Bug: skia:9402
Change-Id: I0329f088e0afce3998494d91ef2206e5eb9cac42
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/294599
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Current strategy: everything from the top
Things to look at first are the manual changes:
- added tools/rewrite_includes.py
- removed -Idirectives from BUILD.gn
- various compile.sh simplifications
- tweak tools/embed_resources.py
- update gn/find_headers.py to write paths from the top
- update gn/gn_to_bp.py SkUserConfig.h layout
so that #include "include/config/SkUserConfig.h" always
gets the header we want.
No-Presubmit: true
Change-Id: I73a4b181654e0e38d229bc456c0d0854bae3363e
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/209706
Commit-Queue: Mike Klein <mtklein@google.com>
Reviewed-by: Hal Canary <halcanary@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Reviewed-by: Florin Malita <fmalita@chromium.org>
When in GL backend, adds a "Shaders" section to the debug menu.
"Load" scrapes all of the vertex and fragment shaders being used,
then displays them. They can be edited, and "Save" pushes the
results.
Note: It is trivial to trigger an assert by saving a shader that
doesn't compile. I'd like to make the program builder more robust
in a follow-up CL, to fall back to the "real" SkSL, not draw, or
something along those lines.
Change-Id: I841fe2ee76a3c2eae58b64ef587fcbe25b95cc7e
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/206905
Reviewed-by: Greg Daniel <egdaniel@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
1) Adds a --writeShaders option to fm. When that's included, the GPU
backend is run with a persistent cache (and no binary caching).
Then we dump all of the fragment shaders (GLSL for now) that were
created to the passed-in directory, along with a JSON digest listing
the number of times each one was referenced in the cache.
2) Adds a python script that invokes the Mali Offline Compiler on a
directory generated from #1, scraping the output of each compile for
cycle counts and writing the collated information to stdout.
Change-Id: Ie4b58ddac4f62e936707c6fec44f4fe605c213fa
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/206162
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: Mike Klein <mtklein@google.com>
Uses a new GPU sink that runs each test twice, once to populate the
cache and then again with a new GrContext but a warmed cache. It
verifies that the two generated images are the same.
Change-Id: Iaba195a69751f14ea946afe7174228a813b83a63
Reviewed-on: https://skia-review.googlesource.com/140567
Commit-Queue: Brian Salomon <bsalomon@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
