Re: [AMBER] Improving our options for pre-compiled AmberTools?

From: Matias Machado <>
Date: Fri, 17 Dec 2021 15:42:12 -0300 (UYT)

Dear dac,

The same issue may arise in supercomputing centers, where some packages requires certain privileges to be installed by a simple yum/apt/pacman/etc command...

Indeed several versions of CentOS, which is the reference OS in HPC places, lack many of the requires packages...

Frequently, to overcome this issue I make local (/home/xxx) compilations of the requires libs (like bison, flex and cmake3)... and then I use "module" to set and load the env variables...

It's more work but it works...

Simplifying the AMBER installation would be very welcome...

All the best,


----- Mensaje original -----
De: "David A Case" <>
Para: "AMBER Mailing List" <>
Enviados: Viernes, 17 de Diciembre 2021 12:01:49
Asunto: [AMBER] Improving our options for pre-compiled AmberTools?

On Fri, Dec 17, 2021, Vaibhav Dixit wrote:

>That suggestion did work and I got past the miniconda installation.
>But now I'm getting an error related to bison for which I tried using conda
>to install bison, but that also gave me permission related errors since by
>default it tried to install packages where I don't have write permissions
>on the cluster.

It's certainly true that our Installation instructions for Linux assume that
people can use standard package managers to install packages. For Ubuntu,
for example, we recommend this:

    apt -y install tcsh make gcc gfortran flex bison patch \
                bc xorg-dev libbz2-dev wget

But this indeed requires sudo privileges. I'm not anywhere near enough of a
conda guru to know if one could create a conda environment that would work
for compiling Amber.

So: I'm asking for advice from the list: how do people compile Amber in
places that don't have standard modules for the needed third-party stuff,
and for users that don't have sudo privileges? Possibilities:

    a. detailed instructions on how to compile missing pieces from scratch?
       For example, more things like the existing configure_mpich script,
       which doesn't touch anything outside the Amber folder hierarchy?
    b. Use singularity containers? Or something similar?
    c. Ask people to install the AmberTools conda package, then provide
       instructions for how to make just the things that are missing (mainly
       MPI/GPU codes?
    d. A better pre-compiled release that includes MPI/GPU functionality?
    e. other ideas?

Don't let the "perfect" get in the way of the "better"! We can't solve all
problems. For example, Vaibhav's environment wouldn't even allow a
miniconda download. But if we could get closer to allowing folks to avoid
compilation, that would be a plus.


AMBER mailing list

AMBER mailing list
Received on Fri Dec 17 2021 - 11:00:02 PST
Custom Search