Want to run your Fortran codes on GPUs but don’t want to add external APIs or directives?
Check out our new paper on using Fortran’s “do concurrent” to run on NVIDIA, AMD, and Intel GPUs!
https://arxiv.org/abs/2608.20586
Adding OpenMP data movement directives can help with efficiency (on Intel GPUs) and for some GPUs are required (AMD consumer GPUs), but for many cases, “Fortran-only” works great - even across multiple multi-GPU nodes!
- Ron