[AMBER] more on building Amber18 on Mojave

From: David A Case <david.case.rutgers.edu>
Date: Mon, 15 Oct 2018 10:39:27 -0400

As promised, over the weekend, I played with Mojave, and seemed to have
some sucess. Three steps are different from previous versions of MacOSX:

1. As noted before (by Andrew Schaub), you need to get the headers in the
proper location. Try this:

   sudo xcode-select --install (to get the command-line tool installed)

   open /Library/Developer/CommandLineTools/Packages/macOS_SDK_headers_for_macOS_10.14.pkg
       (to get headers in the expected locations)

For me, this gets rid of the "cannot compile C and Fortran together"
error at the configure stage.

2. For some reason, you have to tell the miniconda installation what
version of the OS you have:

   export MACOSX_DEPLOYMENT_TARGET='10.14' (to get correct miniconda behavior)

For me, this gets rid of the "cannot find <include> file" error when
building parmed.

3. If you get "token not found errors" with -macAccelerate, try this:

       sudo port install ld64 +ld64_xcode (this is specific to macports)


With these changes, the "standard" build (allowing the configure script
to download miniconda) worked for me. FWIW, I'm using gcc8.2 from
macports, and the corresponding openmpi (also from macports.) FWIW, I
also tried mpich from macports, but that failed: haven't had time to
investigate that yet.

     ./configure -macAccelerate gnu
     make install
     make test
     ./configure -mpi -macAccelerate gnu
     make install
     export DO_PARALLEL='mpirun -np 4'
     make test

One remaining problem: test cases that exercise "ncdump" in parallel are
failing. Also, I don't have access to any Mac's with GPUs, so I haven't
done any testing of pmemd.cuda.

I'm guessing there will be more things to learn about Mojave, but I hope
this helps.


AMBER mailing list
Received on Mon Oct 15 2018 - 08:00:03 PDT
Custom Search