Mon, 16 Jan 95 15:58:19 MET
Well, what kind of abstraction do you propose for parallel computing ?
Do you have any reference ?
I do not. But I still think higher-order functions and continuations
are a good concept for parallelism, from coarse-grained one down to mid-grain
one. To join/fork would just be a construct to allow continuations to
represent independent evaluation of different arguments of a same function.
Note to JVS: this is no more "Low-level". Join/fork can be just
*so different* according to the underlying hardware, and this is never
a cheap operation; which does not mean it cannot be a basic one in the
HLL, as the dynamic compiler can remove parallelism when too costly (or done
inside CPU pipelines), or inline it to whatever is best suited.