I intended to work on shared cache sizing, but ended up spending some time fixing this.

This is useful for sanity-checking memory reports, and we'll also use it for our texture eviction heuristic.

It should really be using the latter, so that it can avoid hand-rolling the frame increments and just use the one already available on frame_builder. In the next patch, we rename the types to make the difference clearer. Depends on D10672

This type identified frames as sequenced by the device layer, whereas render_backend::FrameId identifies frames as sequenced by the render backend. I'm not sure whether there are known cases where these will diverge, but it seems plausible that they might, and so we should more clearly distinguish the types. Depends on D10673

Depends on D10674

Depends on D10675

This simplifies the texture cache code. Depends on D10676

We also switch the boolean from 'in_shared_cache' to 'standalone', since the latter is a bit shorter. Depends on D10677

Depends on D10678

My original try push for this showed a large AWSY regression, so I did a bunch of pushes to show the results for difference values of MAX_FRAME_AGE_WITHOUT_PRESSURE. The result went away (probably a bug in a previous version of the patch), but I wanted to post the results here for reference. Baseline-versus-1: [1] 1-versus-10: [2] 10-verus-25: [3] 25-versus-50: [4] 50-versus-75: [5] 75-versus-100: [6] [1] https://treeherder.mozilla.org/perf.html#/compare?originalProject=mozilla-central&originalRevision=be32f4014f92d0ab717621997e0d36c9bc1c479b&newProject=try&newRevision=72a29fe48ab649bedbe278b2cb09b8fdf79375f8&framework=4&filter=windows%20resident%20qr [2] https://treeherder.mozilla.org/perf.html#/compare?originalProject=try&originalRevision=72a29fe48ab649bedbe278b2cb09b8fdf79375f8&newProject=try&newRevision=ce4f5159f8361c97d9bac896594cd519df86f655&framework=4&filter=resident%20windows10&showOnlyComparable=1&selectedTimeRange=172800 [3] https://treeherder.mozilla.org/perf.html#/compare?originalProject=try&originalRevision=ce4f5159f8361c97d9bac896594cd519df86f655&newProject=try&newRevision=4c1ad0ac2b5e3aeb7f2bac799e7b2c56556162b7&framework=4&filter=resident%20windows10&showOnlyComparable=1&selectedTimeRange=172800 [4] https://treeherder.mozilla.org/perf.html#/compare?originalProject=try&originalRevision=4c1ad0ac2b5e3aeb7f2bac799e7b2c56556162b7&newProject=try&newRevision=9281fe61e658f6ed7b8437908b5f2e528e3ebcc7&framework=4&filter=resident%20windows10&showOnlyComparable=1&selectedTimeRange=172800 [5] https://treeherder.mozilla.org/perf.html#/compare?originalProject=try&originalRevision=9281fe61e658f6ed7b8437908b5f2e528e3ebcc7&newProject=try&newRevision=0894ca6279f6232fb62078b7273b38f787981c14&framework=4&filter=resident%20windows10&showOnlyComparable=1&selectedTimeRange=172800 [6] https://treeherder.mozilla.org/perf.html#/compare?originalProject=try&originalRevision=0894ca6279f6232fb62078b7273b38f787981c14&newProject=try&newRevision=3a2133515aa88fb005a4499f06d651f0a4c11114&framework=4&filter=resident%20windows10&showOnlyComparable=1&selectedTimeRange=172800

(Oh, and the problem turned out to be that I was accidentally looking at the linux numbers on the new push).

Pushed by bholley@mozilla.com: https://hg.mozilla.org/integration/mozilla-inbound/rev/f7087f66bb14 Track the total number of GPU bytes allocated by WebRender. r=gw

Pushed by bholley@mozilla.com: https://hg.mozilla.org/integration/mozilla-inbound/rev/983a36d28484 Missing bits from landed patch. r=gw