langreiter.com plain, simple: comments-2008-01-04

LANGREITER.COM plain, simple

START

Freitag, 4. Januar 2008

Anwar Ghuloum: C for Throughput Computing (a.k.a. Ct)

It's certainly interesting to compare the Intel team's incrementalist vision ("work with [...] prevailing tools", "adoptable incrementally", don't "force the programmer to fundamentally change the algorithmic and data-container idioms") with the radical approach exemplified by Kx's K (re-write everything for every major version^* and [usually] enjoy simplification and code mass reduction in the process [albeit often there wasn't much code mass to begin with]).

Future-Proof Data Parallel Algorithms and Software on Intel Multi-Core Architecture — Recommended reading.

Ct seems to trace its intellectual roots primarily to one of my favourite widely-unknown languages, NESL, and a variant of Lisp I heard of for the first time today: Paralation Lisp (paralation standing for "parallel relation"), developed by Gary Sabot while at Thinking Machines; if you have a look at the references section of the above-mentioned paper, you'll find APL well-represented as well.

^* Admittedly, those come in at about ¾ per decade.

2008-01-04-incr

earl 5955 days AGO
Jay Sipelstein and Guy Blelloch also look at Paralation Lisp (amongst [create Fortran], APL, [create CM-Lisp] and SETL) in their 1991 paper "Collection-Oriented Languages". They have side-by-side examples stated in each of those languages; e.g. APL's A+B+2 for element-wise summing of two vectors and a scalar becomes (elwise ((e1 A) (e2 B)) (+ (+ e1 e2) 2)) in Paralation Lisp.