Logo

LSSw Sustainability Town Hall - Shared screen with speaker view
Richard Gerber
26:23
How does this relate to the ASCR Workshop on the Science of Scientific-Software Development and Use
Todd Munson
26:57
The workshop concentrates on research into scientific software development.
Mike Heroux
27:27
See questions 4 and 5 on this FAQ: https://leadershipscientificsoftware.github.io/FAQ.html
Richard Gerber
27:50
Perfect, thanks
William Godoy
28:30
Are the slides available? Thanks Hal.
Hal Finkel
30:04
I can send those slides to Mike for distribution and/or please feel free to send me an email.
Mike Heroux
30:14
Yes, please do
Michael Wolf
42:16
Kokkos has Fortran support developed at LANL
Slaven Peles
01:04:29
Both Kokkos and RAJA come from DOE.
Cyrus Harrison
01:08:49
Comment: AMReX and newer versions of SAMARI now support GPUs
James Willenbring
01:12:51
For Shawn, I was wondering why the use of DOE software has decreased so dramatically. Is it due to funding or staffing changes? I suspect the answer is not simple.
Chuck Atkins
01:13:40
A common issue raised amongst many of the panel members is better support for heterogenous compute environments. Can any of the panel comment on that more with some of the specific barriers that currently exist for them?
Jeff Durachta
01:14:41
We are aware of the library. But is anyone aware of an application project that has successfully used it?
Jeff Durachta
01:21:37
And first we'd have to figure out what it is that the ESM stencils actually have to do to get performance out of the platform. We also can't afford to loose performance on our everyday systems.
Cyrus Harrison
01:25:33
think we all feel that both user workflows and developer workflows are much more complex these days, do folks have summaries or references where this has been quantified or simply described?
Cyrus Harrison
01:25:45
no mic right now
Cyrus Harrison
01:27:20
I am thinking about - conveying the pain, and how it has changed the lives of the users and developers
Jeff Durachta
01:30:07
My group shields alot of this from the users as we write the automated workflow infrastructure (that runs between remote model runs on our system at ORNL and our post-processing / archive @GFDL)
Jeff Durachta
01:33:58
Spack also seems to be under very quick development. Our experience is v.N+1 is a lot better than v.N
William Godoy
01:35:34
Jeff mentioned DSLs to close gaps with domain and computational abstractions (correct me if this didn’t come out right). Any vision, experiences, barriers, etc. with closing these gaps? Someone mentioned that the transition to C++ can be very overwhelming. Python numba? Julia? They both build on top of LLVM (Flang was mentioned at some point).
William Godoy
01:38:14
Thanks!
Cyrus Harrison
01:40:35
thanks all!
Bill Hoffman
01:40:41
Thanks.
Julien Jomier
01:40:42
Thanks a lot!