@rdmark Just noticed one more thing: This should be 519168, not 512000.

@uweseimet FWIW, it seems like low-end RPis actually report less than 512MB MemTotal in /proc/meminfo, according to some random google search results.

F.e. for RPi Zero: 445032KB

https://inokara.hateblo.jp/entry/2016/02/08/004354

It would probably be good to benchmark MemTotal on some key RPi models and adjust the value later.

@rdmark Would this matter in our case? The current change always assumes at least 1 core, and if less memory is reported than is actually available we would still be better off than before.
Also, this report is from 2016. It may be wrong or outdated, like so many stuff you find in internet forums.

Edit: It SHOULD always assume at least one core. BTW, the report definitely does not apply to buster, bullseye or bookworm, it's too old.

@uweseimet It would matter for f.e. the RPi 3B, which has 1GB of memory. This has been my primary development RPi for several years. It benefits hugely from using 2+ cores for compilation. But if MemTotal is less than 2x512MB then it would only get one core by default, defeating the goal of this change.

@rdmark Can you test this? If not and you have doubts we should discard the PR and the ticket, because it would not be resolvable if one assumes that wrong memory amounts are reported.

I can test it in 2 months. My RPis are all on a long ocean journey right now.

I think we should merge as-is for now. It's an improvement for high end RPis regardless.

My hypothesis is that ~400MB is a better number. In the past, I've used 3 cores with clang and 2 cores with gcc for compiling piscsi on my RPi3B+. But some benchmarking is preferable over anecdotal data.

@rdmark I'm fine with changing the value from 512 MB to 400 MB or something in between.

Using 450MB as divisor now.

@rdmark Sounds good. I have already approved, please merge any time.