[Simh] Makefile performance comparison
hachti at hachti.de
Tue Mar 10 23:42:43 EDT 2009
> After some inspection, I found that there is one significant
> difference between Philipp's and my makefiles:
> - Philipp does not compile all files for every simulator. Instead
> files like sim_*.c, and others, are compiled only once, and the
> object files are reused.
Yes, that's the way a good makefile should work!?!
> Philipp's approach is of course faster, but requires certainty
> about which source files depend on simulator-specific defines,
> and which do not.
That's correct. But only the common object files like sim_*.c are compiled just once.
Every simulator has a list of sources that is compiled for it. With it's own set of parameters.
Every set of sources for one simulator is compiled into an own directory.
VAX and VAX780 share source files - but no object files!
> On the subject of the number of parallel jobs to use, I have found that
> using twice as many jobs as processors is a good approach. More jobs
> will bring no benefit, and fewer may fail to saturate the CPU.
Hm. j8 is slightly slower than j4 for me - on a 4 core machine.
And between j3 and j4, there's only a minimal gain...
More information about the Simh