aom encoder: expose auto-tiles boolean option (since v3.10.0)

[lovell]

Setting this option allows libaom to determine the optimum tile count for the given output dimensions and available threads.

This typically reduces encoding (wall clock) time and/or peak RAM consumption at the cost of slightly increased file sizes. I see a ~30-50% speed up on many-core systems.

$ /usr/bin/time -v heif-enc in.jpg -o out.avif
	User time (seconds): 1.71
	System time (seconds): 0.03
	Percent of CPU this job got: 258%
	Elapsed (wall clock) time (h:mm:ss or m:ss): 0:00.67
	Maximum resident set size (kbytes): 50432

$ /usr/bin/time -v heif-enc -p auto-tiles=true in.jpg -o out.tiled.avif
	User time (seconds): 2.23
	System time (seconds): 0.04
	Percent of CPU this job got: 588%
	Elapsed (wall clock) time (h:mm:ss or m:ss): 0:00.38
	Maximum resident set size (kbytes): 52864

 847374 in.jpg
 117500 out.avif
 118148 out.tiled.avif

[farindk]

Very nice.

Note that I am currently finalizing writing/reading individual tiles:
https://github.com/strukturag/libheif/wiki/Read-and-Writing-Tiled-Images

This might also be interesting for libvips. Not just because of the parallelization possibilities, but also because of the support for much larger images.