I can be very helpful to find out before pardiso procedure call whether I should use int32 or int64 version of PARDISO solver depending on the matrix input. If pardiso routine stores factors indices in int64 on phase=11 I can surely call this routine if the matrix fits in int32, but in case factors are stored in int64, I must use pardiso_64 as soon as there's no guarantee the amount of non-zeros in factors is less than 2^31. I guess factors are stored in int32, but in supernodal format that does not require int64 to be able to store more than 2^31 nnzF, thats why I was able to factorize the matrix Bump2911 from SuiteSparse collection that produces 10 billion nnzF with pardiso int32 sequential out-of-core mode (output attached). So is nnzF<2^31 the only criteria I can use to be sure pardiso int32 version works fine? Lets say I solve the broblem with just 1 rhs.
