Re: [Gomp-discuss] CVS organization

[Scott Robert Ladd]

Diego Novillo wrote:
> If you were going to submit GOMP for mainline acceptance, then yes, I
> agree.  However, starting on a branch, it doesn't matter if you
> initially set your sights at implementing it on Linux first.  Then you
> generalize your runtime and make it generic.

Hmmm. I remember there being more consternation over the issue, and a 
general sense that we needed soemthign concrete before actually becoming 
a branch. However, I'm agnostic at this point; I just need to know where 
to point CVS. :)

Also, at this point, I believe it makes sense to wait for tree-ssa to be 
merged into mainline; then GOMP would be a direct branch of mainline, as 
opposed to a branch of a soon-to-be-merged branch.

> > I have no problem with revisiting our earlier decisions, but we should 
> > do so now, before we have a design -- and the design should proceed any 
> > implementation work.
>
> Well, if you want.  But don't spend too long designing.  I've always had
> good results with the "design-a-little, implement-a-little" approach. 
> I'm sure the SoftEng world has a term for it.  It lets you evolve and
> learn in a natural way.

Indeed, everything in moderation. I'm looking for a balance between 
diving-in-and-coding and obsessing over the details; we need a happy 
(and productive) medium.

An incremental model is fine; at the moment, I'm experimenting with the 
manual threading of small programs, to see what the code generated by 
OpenMP *should* look like.

Design does not mean a big long document; it means answering questions 
such as "pthreads or Native Posix Threading Library (NPTL) or both?"

--
Scott Robert Ladd
Coyote Gulch Productions (http://www.coyotegulch.com)
Software Invention for High-Performance Computing