[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Gomp-discuss] Questions, questions
From: |
Scott Robert Ladd |
Subject: |
Re: [Gomp-discuss] Questions, questions |
Date: |
Tue, 23 Mar 2004 09:51:58 -0500 |
User-agent: |
Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.6) Gecko/20040312 Debian/1.6-3 |
Lars Segerlund wrote:
g95 is quite good now, and it's frontend is VERY easy to work with,
so I think it would be safe to say it's ready for OpenMP.
Wonderful. I've been studying gfortran for some time now, using it on a
regular basis and working on some patches (integer ranges, structure
returns). I'm waiting to submit the patches until the integration of
tree-ssa into GCC mainline is complete.
I have much less familiarity with the C and C++ front ends -- even
though those are my primary langauges!
Which brings up a question: Will we need to implement *three* versions
of OpenMP, given that C and C++ have different frontends?
The thing about academic fortran programmers are that often they are
afraid of computers :-) ... so they want a magic 'parrallell thingie'
to solve their parallellization issues for them. Which is good and
bad, and then there are the exeptions which forces progress on the
mediocre crowd :-) ( Pun intended ).
C++ programmers aren't much different. And the "magic" has value in
terms of portability and simplicity; I can create an OpenMP application
that scales automatically and cleanly, whereas to do with same thing
manually is rather more cumbersome.
I'm in the midst of manually parallelizing an application, making me
quite aware of these realities.
As for your point 3. we decided to use Posix threads, and abstract
the rest in the lib. ( thus different thread implementation and so on
different lib, no code change), i think :-) ..
Sounds good.
--
Scott Robert Ladd
Coyote Gulch Productions (http://www.coyotegulch.com)
Software Invention for High-Performance Computing