CCX 2.19 with Pardiso solver runs for about 20 hours and then starts to print numerous messages “** On entry to DGEMM parameter number 13 had an illegal value”.
I guess this is an integer parameter in a BLAS DGEMM subroutine. Should I compile with i8? Any ideas?
Do you get this error message for a particular model ? I would try removing features (such as contact, plasticity and so on) to see if it’s caused by some specific characteristic of the model.
No contact, no plasticity in the model. Just some buckling computation, about 2 million nodes, 6 million equations, 230 million nonzero lower triangular matrix elements. Previously I ran the model with a coarser mesh and had no problems. This is why I suspect that i4 is not enough anymore.
I think you’re probably right about needing i8. 2 millions nodes could easily have more than 4 billion nonzeros after fill-in even if it’s only 230 million initially.