1 | Runtime optimizations acceptable (and perhaps best) as "grain size" (i.e. total number elements in array divided by Nproc) is large ==> "Startup" of Runtime Procedure calls is small ==> Can use an interpreter rather efficiently cf. APL and *LISP on CM-2 |
2 | There is only ONE data parallelism ==> Same Runtime library can support C, C++, ADA APL, HPF Interpreter |