On Sun, 2011-12-18 at 10:30 -0500, dsimcha wrote: [...] > Please elaborate. I think D for HPC is a terrific idea. It's the only > language I know of with all of the following four attributes:
For years HPC was Fortran only (*). Many of the codes used today were written in the 1970s and Fortran compiler technology revolves around making those sequential codes run on parallel kit. During the 1990s there was a move to writing new codes in C++ -- especially in the HEP area. Of course many of these codes rely on being able to call into the Fortran codes. With the rise of data visualization there was a need for a language for writing GUIs and the like, and they have decided on Python since it connects to C and C++ trivially. They also found SciPy and NumPy which actually leads to prototyping new codes in Python and then rewriting in Fortran or C++ if there is a need to -- which often there is not. With PyPy now running faster than CPython there is a possibility that new codes other than core computation frameworks will always be in Python. Given the time and effort it took to get C++ accepted by this Fortran oriented community. > 1. Allows you to program straight down to the bare metal with zero or > close to zero overhead, like C and C++. Or Fortran. > 2. Interfaces with C and Fortran legacy code with minimal or no overhead. This is an essential. This is Go's biggest problems, it cannot link easily to existing codes. > 3. Has modern convenience/productivity features like GC, a real module > system and structural typing via templates. The bulk of the HEP community might know the words but only if it is in Fortran -- or C++. > 4. Has support for parallelism in the standard library. (I'm aware of > OpenMP, but in my admittedly biased opinion std.parallelism is orders of > magnitude more flexible and easier to use.) The main current tool is MPI for cluster and multicore parallelism using SPMD and MIMD models, and OpenMP for thread management on multicore systems. Fortran, C, C++, and Python all have MPI capability and Fortran, C, and C++ have OpenMP -- OpenMP isn't relevant to Python. This has been the model for 20 years and is likely immovable. No matter how good a different model is, it is fighting a war on an entrenched technology. The issue is not really a technical one, it is an inertia and expectation, even political, one. C++ broke into the Fortran dominated area by fiat of a group of people in the HEP community. Also Barton & Nackman "Scientific and Engineering C++: An Introduction with Advanced Techniques and Examples" was an important factor. The question is what is the need and the vector for D to gain traction in this area where it really does have a USP? (*) FORTRAN and Fortran are different languages. The case was changed formally with the Fortran 1995 standard. -- Russel. ============================================================================= Dr Russel Winder t: +44 20 7585 2200 voip: sip:russel.win...@ekiga.net 41 Buckmaster Road m: +44 7770 465 077 xmpp: rus...@russel.org.uk London SW11 1EN, UK w: www.russel.org.uk skype: russel_winder
signature.asc
Description: This is a digitally signed message part
