The function clock
has been removedfrom python 3.8 and I had a heck of a time installing OSMPS until I added from time import perf_counter as clock
into the file openmps/MPSPyLib/EDLib/exactdiag.py. I don't recall if I had to do this in more than one function.
Thanks for the help
Thanks for posting this issue, Jared! I'll update the appropriate files with an alias so that it always selects the right module.
How helpful did you find our install-from-source guide?
Which steps produced problems for you, and which steps worked well for your use-case? Additionally, would it help you immensely if we supplied conda/pip packages for OpenMPS?
I'll be honest.. I didn't use that guide.
I have installed OpenMPS twice, the first time on my unversity's computing cluster. The guide did not seem to work for it due to my lack of understanding of the cluster's anaconda. Eventually, with the assistence of people who actually understand the backend of the computing clsuter, I got it to compile from the command line using some variant of
python BuildOSMPS.py --option=ARPACK_FLAG:${HOME}/ARPACK/libarpack.a --option=OFLAGS:"-fallow-argument-mismatch"
.After a quick search, apparently there are often complications when trying to add new software to a cluster - so I just tried to figure it out by trial and error, with plenty of both.
I did have some issues using the code to generate .pdf plots on the cluster - and installed OpenMPS on my personal laptop. Since I managed to install it using the terminal on the cluster, I used the same method on my personal laptop, and installed all of the dependencies before unpacking the
.tar.gz
file. Using Redhat'sdnf install
proved to be the easiest method for me to install LAPACK and BLAS on my Fedora 32 machine.Which packages did you use DNF to install?
Most of the packages defined in the environment YML can be installed with
pip
, as well.The only packages that might cause problems would be BLAS and ARPACK.
I installed Anaconda, LAPACK, BLAS with DNF (along with dependencies for OpenFOAM, but that's a different matter, I suppose). If I recall, ARPACK had an error without the -fallow-argument-mismatch on my personal machine, but from reading of other forums that appeard in a web search, this LAPACK issue when building the seemed to be unique to (newer) gfortran. I have gfortran 10.2.1 installed, which came with my Fedora installation.
I don't think that there's an issue with BLAS/ARPACK. It seems to run the pre-built examples including the iMPS examples, which I read somewhere on these forums that the ARPACK is just for the iMPS calculations. If there were issue(s) with linking to ARPACK, wouldn't is show up in the iMPS examples?