The signed-distance and coverage area approximations we use for shader anti-aliasing in WR are expensive due to usage of branching and normalization. There are cheaper approximations we can use that give perceptibly similar results so long as we're willing to take the hit in reftest fuzz.

Depends on D104270

The result of compute_aa_range depends on fwidth(local_pos). In the no-perspective case,
the derivatives of local_pos are constant across an entire primitive. SWGL fast-paths only
run in the no-perspective case anyway, so it is convenient to compute the aa_range once
for the entire span and then reuse it, factoring out this per-pixel cost.

https://hg.mozilla.org/mozilla-central/rev/8d1fc04130de
https://hg.mozilla.org/mozilla-central/rev/0f12cc02b12e
https://hg.mozilla.org/mozilla-central/rev/f4f3f317a7d6

== Change summary for alert #28694 (as of Tue, 09 Feb 2021 12:56:20 GMT) ==

Improvements:

For up to date results, see: https://treeherder.mozilla.org/perfherder/alerts?id=28694

== Change summary for alert #28736 (as of Thu, 11 Feb 2021 07:55:31 GMT) ==

Improvements:

For up to date results, see: https://treeherder.mozilla.org/perfherder/alerts?id=28736