Dear Ray,
thanks for comments
> > Then, if the case, you have to clean the lapack source tree from fasl's
> > files and start again with load(lapack). This procedure need enough
> > memory be sure
>
> You shouldn't have to remove all fasls. Just the last one that failed
> to compile.
> > that nothing else is concurrent with compilation. BTW for cmucl the
> > compilation is more smooth on my pc.
>
> There is (was?) some "optimization" for cmucl. One of the files is
> 6000+ lines all in one function. It takes a huge amount of memory and a
> very long time to compile, so I modified it so that cmucl byte-compiles
> that one file. It compiles much faster, doesn't use as much memory, and
> doesn't seem to impact runtime too much.
I observe that this time cmucl and sbcl run the eigen-value problem
equally fast while earlier cmucl was superior. They work reasonably
fast , 1-5min for one set of eigen-vectors for 200x200 matrix in my problem.
In practice, however, the most time-consuming operation will be the
constructing of such matrices from PDE (say via Bubnov-Galerkin procedure)
with in maxima.
best regards
Valery