Revisiting rc-lookahead, opencl and lookahead-threads

[iAvoe]

I don't post a lot in this forum, so please forgive me if I'm not following some rules/guidelines

--lookahead-threads is to estimate the cost of establishing intra, inter reference, and b-frame prediction;
and the purpose of cost estimation is to reduce uncertainty of compression decision.

--opencl enables "opencl-lookahead", which offloads extra lookahead-threads to GPU
when CPU is working on other things at the same time, but requires estimation to be finished too.

I believe the recommendation for lookahead-threads (enables opencl) should be updated since GPU capability (and VRAM speed) has improved a lot.
Meanwhile posts discussion about these 2 options are dated back to like GTX 480

--rc-lookahead is for estimating VBV, IBP frame and mbtree costs, which is completely handled by CPU fwiw,
and the recommendation for this one is vague, while people often go above 40.
I've also found out that this option is bind-ed to keyint (I'll try find that info if you needs);
which now, I feel the best value is going to be around keyint÷2

[benwaggoner]

My intuition was that that the maximum useful value of --rc-lookahead would be the higher of vbv-maxrate/vbv-bufsize in seconds or keyint - min-keyint.