Even when compiled with -DCUSP switch, the code allocates memory for full K
matrix. For 20-20-100 problem full K needs more than 100 GB of memory, which is
of course unacceptable. Since for these large problems GE solver that works
with full K is very slow, it is useless anyway. A sparse CPU solver should be
used instead of the current GE and everything related to the full K matrix
should be deleted from the code.
Cuneyt
Original issue reported on code.google.com by cuneytsert on 26 Apr 2012 at 12:17
Original issue reported on code.google.com by
cuneytsert
on 26 Apr 2012 at 12:17