R57.A6 workaround a performance regression against R54
This release fixes #7, which is significant (10x) performance regression against R54 with certain large temporal radius filtering scripts.
The exact trigger condition is a little involved, as it depends on num_threads
, the cycle
of SelectEvery
and the temporal radius of AverageFrames
for the scripts in #7 (comment).
Whether the source filter is api3 or api4 doesn't matter in this case.