Parallel build on SMP is even more broken.

Marc Espie espie@quatramaran.ens.fr
Wed Sep 6 15:27:00 GMT 2000


In article < 200009061728.KAA20727@kankakee.wrs.com > you write:
>Bear in mind, the wall clock time to build is reduced on a
>uniprocessor machine, by using -j3, on most machines, unless you have
>16Megs of swap or 16M ram.  I suspect for most people that isn't true
>any more.  On my Sun, -j5 is the fastest way to build, I have just one
>processor.  If you aren't using at least -j3, I suspect you are just
>wasting time waiting for builds.  :-)

The timings I have prove this is not always true. On my system, I've
seen a 15 seconds difference out of a 15 minutes compile by switching
from sequential build to a parallel build.

On the other hand, time to write that `nasty make' patch: an option that
scrambles the build order so that things occur in the REVERSE order they
usually do, unless there is an explicit dependency to disallow it (or random
order, whatever).

E.g., in
a: b c

make <-nasty>

would first try to build c, then b, then a :)



More information about the Gcc mailing list