When compiling with the Cray compiler I noticed warnings that we are using the error variable before defining it in the methods LU_DECOMPOSE_MATRICES, LU_SOLVE_SPARSE, and LU_DECOMPOSE_SPARSE. This is true because the actual calls to pardiso are commented. I don't think we would get correct results if sparsesolve is set at the moment, although it's possible that this particular code path is never reached.
Do we need the sparsesolve option? If yes, maybe we should add an #ifdef usepardiso around the calls to pardiso (after uncommenting them, of course) and let the methods abort with an error if the code is compiled without usepardiso. I'm happy to make a PR with these changes, but I'll only do it if we actually need the sparsesolve.
When compiling with the Cray compiler I noticed warnings that we are using the
errorvariable before defining it in the methodsLU_DECOMPOSE_MATRICES,LU_SOLVE_SPARSE, andLU_DECOMPOSE_SPARSE. This is true because the actual calls topardisoare commented. I don't think we would get correct results ifsparsesolveis set at the moment, although it's possible that this particular code path is never reached.Do we need the
sparsesolveoption? If yes, maybe we should add an#ifdef usepardisoaround the calls topardiso(after uncommenting them, of course) and let the methods abort with an error if the code is compiled withoutusepardiso. I'm happy to make a PR with these changes, but I'll only do it if we actually need thesparsesolve.