On Tue, Mar 27, 2007 at 11:44:43PM +0200, Peter Kümmel wrote: > Andre Poenitz wrote: > > On Tue, Mar 27, 2007 at 08:31:33PM +0200, Peter Kümmel wrote: > >>> Possibly, because thing get larger, the compiler process is ~520 MB than > >>> which means it goes to swap on this (512 MB) machine. [It is still fine > >>> for the Release version...] > >> OK, I've committed. > >> > >>> Looks like a bit of fine-tuning is needed. > >> I've also added the possibility to build shared libraries: > >> -Dshared=1 > >> > >> On my 512MB machine I could build mit ~90% memory usage > >> (when I interpret top correct) with > >> -Dmerged=1 -Dprofile=1 -Dshared=1 > > > > It is fine for now with 390MB max. (Really good compared with autotools > > btw) > > > > Full build from scratch -Dmerged=1 -Ddebug=1 -Dshared=1: > > real 4m57.772s > > user 3m10.204s > > sys 0m9.133s > > > > Then change of a few files within mathed: > > real 0m28.302s > > user 0m18.629s > > sys 0m1.252s > > > > And what is the influence of the shared library build? > Or doesn't it matter static or shared? > > > > > -Dprofile=1 still goes to swap when compiling src/*.C. > > too much boost templates ;) > > But I have no idea how to prevent this, even when we split it > up into two files, linking will be the same. Only when > it's not ld we have a chance with an split.
It's the compiler as far as I can tell... Andre'
