Closed micmitch closed 7 years ago
I have removed the mumps stuff. My suggestion is to use Paradiso. The old code is in https://github.com/rowanc1/pymatsolver-mumps
@rowanc1 It's really sad to see Mumps gone. While i did not have success with the setup at first trial, i considered it the most important use-case of this project just because of the performance/license connection.
I have brought this back in #16. I have updated it to work on my current mac.
With Paradiso working well with conda installs, I was prioritizing ease of install over keeping mumps. you are right to note that we might as well keep it in. Thanks for bugging me on this!
Cheers, Rowan
@sschnug this should now be updated. Let me know if it is working for you, or if you have any improvements to suggest for the makefile.
Messy but I am now able to call: from pymatsolver import MumpsSolver
In the f2py call in the makefile my system does not recognize the include file directories. For now I modified mumps_interface.f90 to have full pathnames for include files. This must be a problem with our syntax for the f2py command since these include directories are properly linked in the gfortran calls...
Not sure how to fix or clean this up.
also had to add the following to my system path variable /usr/local/lib/i386
From the pymatsolver directory containing setup.py simply ran python setup.py build_ext --inplace
Ran the SimPEG example EM_TDEM_ID_Inversion.py using the MumpsSolver to test it. Results looked quite reasonable.
from pymatsolver import AvailableSolvers AvailableSolvers {'DiagonalSolver': True, 'Mumps': True, 'TriangleFortran': False, 'TrianglePython': True}
Still not getting 'TriangleFortran' but now have mumps. A step in the right direction.