Should fortran-lang join the High Performance Software Foundation?

Hello all,

Today I came across an article talking about the creation of the High Performance Software Foundation. My first impression was that fortran-lang could (should?) join this project. After some discussions I got to understand that some of the founders of HPSF might have a certain degree of animosity against Fortran… But that might, given enough support from the community, accept Fortran related projects within. Who knows…

So, I would like to adress the question to some of the fortran-lang members, what are your thoughts on this? Is it worth it to try? If so, how could the broader community help?

3 Likes

I find it interesting that all the U.S. Gov’t lab members are all DoE which went big time C++ several years ago. It does not include DoD HPCMP program (I guess there is still time) so until it does I can’t take it seriously. Still it might be useful just to get the camel’s nose under the tent and be a fly in the ointment :smile:

3 Likes

You have to buy in

… right, since I saw that the objective was to help promote open source tooling for HPC I thought it was something closer to NumFOCUS…

1 Like

And for .not. “premiere members”?

Hi. Co-founder of HPSF here. Just to clarify a bit:

My first impression was that fortran-lang could (should?) join this project.

Do you want to join HPSF as a (supporting) member organization or as a (supported) project? Those are the two ways to join. HPSF exists to support the projects and to foster a portable, accelerated HPC software ecosystem.

Is fortran-lang an open source project? There are projects in LF that are more community-oriented or standards-oriented. If so we could certainly talk about what that might look like. Joining as a project is free, though you do have to contribute the trademark (name/logo) and domain to LF and LF becomes a co-owner on the GitHub org (they don’t want to run it – just to make sure it doesn’t go away if the maintainers do). Projects in LF have their own governance, so you would also need a technical charter for fortran-lang (maybe you already have one?).

@ivanpribec:

You have to buy in

“Members” as described on that page are supporters of the technical projects and working groups. You do not need to buy in to join as a technical project or to participate in any technical projects.

After some discussions I got to understand that some of the founders of HPSF might have a certain degree of animosity against Fortran

I am not sure who is promoting this viewpoint, but Fortran is undoubtedly still important in the HPC software ecosystem, and there’s plenty of Fortran software and Fortran projects that we might want to bring into HPSF. I don’t think there is any opposition to Fortran projects joining.

I find it interesting that all the U.S. Gov’t lab members are all DoE which went big time C++ several years ago. It does not include DoD HPCMP program (I guess there is still time) so until it does I can’t take it seriously.

We just launched on Monday, and have been trying very hard since the intent-to-form announcement at SC23 to get all the initial members onboarded. One of the goals of doing the initial announcement in November was to give others a chance to get in touch and join.

If DoD HPCMP wants to join, they are welcome to do so, and there is a link on hpsf.io where they can sign up as a member. Also, if HPCMP has open source codes they think would belong in HPSF we’d be happy to have them join as projects.

Update: CEA just joined, so there is an existence proof :slight_smile:

9 Likes

See:

They explain about membership and at the end there is the fees for the three levels of membership.

And the presentation:

2 Likes

@tgamblin thank you so much for providing the background information and welcome to the forum!

We are currently part of NumFOCUS, just joined recently. We’ll have to talk if it makes sense to join as a member: I think giving the name and logo to LF might be problematic, I am not currently sure if we want to do that.

2 Likes

Ok so it sounds like you already have a foundation and a legal entity there :slight_smile:

It probably makes sense to talk about how NumFOCUS projects can collaborate with HPSF. e.g., we are thinking we might have a conference for HPC software. I see no reason why NumFOCUS projects couldn’t attend, or why they couldn’t collaborate with other HPSF working groups.

4 Likes

Thanks for the clarifications @tgamblin and welcome to the discourse :slight_smile:

Ok, things are clearer I think. fortran-lang as an org already being supported by NumFOCUS does not need being supported by HPSF directly, but could participate in some of the events and working groups of HPSF. I think that would create a nice synergy with other projects.

Ok, this is the part for which I did not find information on how a technical project/org could join working groups.

If I got it wrong then that’s good news, and good to know :slight_smile:

2 Likes

LLNL seems to be encouraging C/C++ over Fortran with some success, but I don’t see the same at LANL and SNL.

Sandia has been trying to replace their Fortran CTH Hydrocode with C++ Allegra and their old PRONTO3D structural dynamics code with the SIERRA framework (which is all C++) for over 20 years now. CTH is the only Fortran code at Sandia I know that still registers any level of development activity. There may be more but I’m not aware of any. Just about all new code development I’m aware of is C++ and has been for a long time. I’m not sure about LANL but I doubt that other than some long standing codes that still get some use any new development is being done in Fortran.

3 Likes

LANL codes are still mostly Fortran AFAIK, with C++ usage growing. DOE Office of Science labs (ORNL, ANL, LBL) use quite a bit of Fortran but also C++. LLNL and SNL are both very much C+±oriented, and it is true that most LLNL and SNL mission codes are C++.

This is not the same thing as not caring about Fortran. There are at least a couple key codes still developed in Fortran, and there are science codes and models outside of the production environment that are written in Fortran. LLNL also has people working on codes like E3SM.

LLNL has contributed quite a bit of funding to flang and has tried to ensure through the El Capitan procurement that there is a viable 3rd fortran compiler outside of CCE (Cray’s compiler) and gfortran. All of that is because we recognize that there are codes that we need to use that are written in Fortran, for the foreseeable future. Keep in mind, also, that LANL and SNL codes need to run on LLNL machines and vice versa. Our HPC center supports LLNL, LANL, and SNL. I don’t see this changing, and I don’t see DOE ceasing to care about Fortran any time soon.

OTOH, I think it’s accurate to say that there is a shift happening. It has been hard, across DOE codes, to use the same sorts of abstraction layers (RAJA, Kokkos) in Fortran as have been successful in C++ for running efficiently on GPUs. Most of the Fortran codes I know have taken the route of OpenMP for performance portability, and while they’ve been successful, it has been a longer journey to get there, and the compiler and industry support situation for Fortran (see above) is a risk. There simply isn’t the same level of investment in the language as there is in C++.

But again, I don’t think that means that Fortran’s going away, or that we can stop caring about it. Fortran never dies, right?

6 Likes