Hard-coded fil-in estimation causes errors

[djturizo]

Using the library to factor a square test matrix of size 377938 yields the following error:

Storage for L subscripts exceeded; Current column 146522; Need at least 75175526;
You may set it by the 8-th parameter in routine sp_ienv().
Memory allocation failed at line 231 in file /home/daniel/superlu_mt-4.0.2/SRC/pmemory.c

Reading the code I believe the issue comes from the function pdgstrf_MemInit in the file pdmemory.c, where we have:

...
    int_t      FILL_LUSUP = sp_ienv(6); /* Guess the fill-in growth for LUSUP */
    int_t      FILL_UCOL = sp_ienv(7); /* Guess the fill-in growth for UCOL */
    int_t      FILL_LSUB = sp_ienv(8); /* Guess the fill-in growth for LSUB */
...

These fill-in growth estimate are hard-coded values, which may be insufficient for some matrices. A way to solve this issue would be to enable the user to provide the fill-in estimates, so that the factorization can be repeated with larger allocation in case of failure.

[xiaoyeli]

You can set a positive number as an actual fill-in estimate. See comment in sp_ienv.c

[djturizo]

And when a negative number is set it is used as a fill-in factor, right? I tried increasing the factors and I would always get the same error, even with a factor of 300 or more. There seems to be a bug hidden deeper down.