I've been working (branch cuda_MLMD) on adding some machine learning
special sauce via the interface defined in "external.F90". Although this
is supposed to be painless drop-in of libraries, without even a recompile,
I've had to loosen the checks on some mdin_ctrl flags, add some cmake
rules, and generally poke at a few things that will need integrating back
to the main branch.
Not sure this counts as a "significant addition" that has to be flagged,
but I remember people were working on neural nets and RE sampling with
similar philosophy, so please talk to me so we can do some integration
engineering.
Josh Berryman
On Wed, 9 Feb 2022 at 15:50, David A Case <david.case.rutgers.edu> wrote:
> Hi everyone:
>
> Following our usual schedule, we plan to release AmberTools22 and Amber22
> in
> April. Here are some comments and requests to help smooth the process.
>
> 1. The deadline for all code submission is March 11 -- that's only a month
> away, so get busy with the polishing of things you want to include. Please
> reply by email if you are planning significant additions.
>
> 2. The plan is to have Amber and AmberTools be independent packages, built
> in
> separate directory hierarchies that don't depend on each other. This
> doesn't affect the development repository, which will continue to have a
> merged system. Rather, the tarballs for release will do the separation.
> The main advantage comes for future releases of AmberTools, which can be
> installed without affecting any existing Amber release.
>
> a. If you wish, you can view how this is done in the "release_pmemd"
> branch, which is where the Amber22 tarballs will be made.
> (Folder names will probably change before the release.)
> b. If testing shows unexpected problems, we have the fallback of doing
> things the old way.
>
> 3. The "legacy" build system will be disabled, with an error message given
> if people try to use it. Desparate users can comment out the error
> stop,
> but we know it doesn't really work anymore, and we are not supporting
> it.
>
> 4. We need volunteers, even starting now, for a number of special needs.
> Please us the master development branch:
>
> a. Build on OSX 12 (Monterey), with Intel hardware. If you've already
> done this, please report details.
> b. Build on OSX 11 or 12 on a machine with a M1 chip. Again, please
> report details of how this was done.
> c. If anyone can get parallel builds to work on CentOS/RHEL using their
> standard (yum-installed) MPI stacks, please let us know. Right now,
> users seem to have to compile their own MPI.
> d. Try the Intel compilers (including Intel MPI) on CentOS. (Intel
> compilers are now free and easy to install:
>
>
> https://www.intel.com/content/www/us/en/developer/tools/oneapi/toolkits.html
>
> You will need the Base and HPC toolkits. The Intel python has all
> the
> modules that Amber needs, so you should be able to skip the
> miniconda
> download step.)
>
> e. Build on Windows 11/WSL.
>
> <insert usual apologies for being so long-winded/> ....dac
>
>
> _______________________________________________
> AMBER-Developers mailing list
> AMBER-Developers.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber-developers
>
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Thu Feb 10 2022 - 01:30:02 PST
