|
Now instead of working through each commit searially, we build a graph
of steps and work on that by emitting make(1) code to be executed.
May this code prove to be diffucult to modify or extend, this is a valid
commit "checkpoint" to revert to in order to recover the *exact* same
behaviour, but simpler code.
Simple benchmarks suggest that:
- on a single core, the scrit is ~15% slower;
- with 2 cores it is as fast as the original;
- as more cores are added, the faster it gets, up to $(nproc) when
using more cores doesn't change anything.
That means that it is a useful optimization for our desktops, but a
hinderance for our current server, which only has a single core.
|
