Skip to content

/ webrender

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Expire entries in the shared texture cache before allocating another layer #3293

Merged
merged 6 commits into from Nov 9, 2018
Merged

Expire entries in the shared texture cache before allocating another layer #3293

merged 6 commits into from Nov 9, 2018
+297 −293

Commits on Nov 9, 2018

  1. Consolidate driver limits and work around mac bug. 

    This restores non-mac platforms back to the old maximum of 64 layers.

Differential Revision: https://phabricator.services.mozilla.com/D11416
    @bholley
    bholley committed Nov 9, 2018
    2df20cd

  2. Factor texture cache handle lists into a helper struct. 

    This is a bit cleaner and avoids borrow checking problems.

Differential Revision: https://phabricator.services.mozilla.com/D11271
    @bholley
    bholley committed Nov 9, 2018
    ce062a4

  3. Refactor TextureCache allocation with some helpers. 

    The container struct reduces repetition, and the helper method will simplify
future work by allowing early returns.

Differential Revision: https://phabricator.services.mozilla.com/D11272
    @bholley
    bholley committed Nov 9, 2018
    d96a6d7

  4. Unify standalone and shared texture expiration and GC before growing. 

    This patch does a few things:
* Uses the same age-based expiration that we use for the standalone
  cache for the shared cache.
* Tracks the last time we expired shared entries.
* When allocation fails, tries to expire old entries before allocating
  another layer, assuming we haven't done so just a few frames ago.
* Eliminates the size limits on the shared caches, and just grows them
  instead of allocating standalone entries.

The last bit could cause us to get stuck with larger total texture
allocations than we do now, if a lot of entries were used in quick
succession (since we never drop shared entries like we do with
standalone entries). However, I think it's probably unlikely enough
that it's fine to ship it for a few days while I finish up shrinking
support.

The parameters here (75 and 25) still aren't perfect, and I'll tune them going
forward.

Differential Revision: https://phabricator.services.mozilla.com/D11273
    @bholley
    bholley committed Nov 9, 2018
    4cbd157

  5. Rename EntryKind to EntryDetails. 

    We need this so that we can have a detail-less enum called EntryKind to
eliminate the boolean. People on #servo agreed this was an appropriate
naming strategy.
    @bholley
    bholley committed Nov 9, 2018
    a928272

  6. Use EntryKind instead of a boolean.

    @bholley
    bholley committed Nov 9, 2018
    Loading status checks…
    7d33d12
You can’t perform that action at this time.