TERRY%PAGES.GATNET.MFENET@LLL-MFE.ARPA (11/06/85)
In response to Paul Schauble's query about vectorizing computations: I don't know how deeply you want to delve into this; let me just answer your questions directly. FORTRAN is still the language of choice. I am not aware of any imple- mentations that extend the language for vectorizing - the compiler is simply made smart enough to rearrange the calculations for efficient vectorization (oops, I take that back - I know of one case where there are compiler directives to say "be careful; don't try to vectorize here"). The impression I get (I am a systems programmer, not a scientific programmer) is that some of the folks would LIKE to see other languages, but few if any are available. I know of some who are pushing for C and Pascal ... Just as an additional aside - there is also a lot of interest these days in PARALLEL processing; i. e., getting more than one CPU to work on the problem at the same time. Most implementations of this require a LOT of intervention on the part of the programmer, and are easy to screw up. Parallelizing (what a word!) requires extensions, in the form of subrou- tine libraries and/or compiler directives. There is at least one language, called Occam, which was designed especially for parallel programming. Usual disclaimer about expressing my own opinions here ------------------- Here at the San Diego Supercomputer Center, we are a week away from acquiring a Cray X-MP/48. This is a 4-CPU machine with vector capabilities, and the FORTRAN compiler(s) do automagic vectorization. Several of us arer VERY interested in the parallel processing problem, and we will be working on a couple of different approaches. This will enable users to write code to potentially use all 4 CPUs simultaneously, but do it EASILY, rather than with lots of wailing and gnashing of teeth. End disclaimer section --------------------------------------------------- Anyway, vectorizing is much better understood than parallelizing, as far as automatic compilation goes, and so that is pretty well handled without a great deal of thought on the part of the programmer (though a good programmer can help the compiler out a great deal). Terry San Diego Supercomputer Center