03-07-2012 10:59 AM
I'm using the PARDISO solver as part of an iterative algorithm that requires a linear solve of a sparse system at each iteration. Moreover, while executing the algorithm with the Totalview debugger, the execution is halted at the third iteration because of a floating point exception thrown by the PARDISO call to release internal memory. The curious thing is that the PARDISO call to solve the system completes successfully (I checked the residual of the linear system, and it is in the order of machine precision). Nevertheless, I cannot explain the nature of this floating point exception and am afraid it is indicative of potentially spurious behavior such as a segmentation fault. I would greatly appreciate any help or information related to this issue.
06-07-2012 07:25 AM
You may want to try and see if the floating point exception still occures when compiling with the debug (i.e., -g) option. Sometimes, when optimizing for speed (e.g., -fast option) the compiler will take shortcuts that lead to floating point exceptions.