f.luporini12 {@}

     Room 301
     William Penney Laboratory
     South Kensington Campus

What I do and think, in short

Numerical methods for partial differential equations are computationally expensive, often requiring hours, days or even months to produce a result. I'm working on domain specific languages (DSLs), DSL compilers and performance optimizations for finite element and finite difference methods, as well as studying how to leverage domain or application properties for performance optimization. I'm spending quite a lot of time developing tools to make my studies actually usable for real science. Ah, I also believe there's a big distance amongst the computational science academic community, the hpc/compilers/DSL academic community, and industry, simply because people in these fields often don't talk to each other.