Re: [Gomp-discuss] Somethings to think about ....

[Biagio Lucini]

On Mon, 10 Mar 2003, Lars Segerlund wrote:

>
>   Hi again,
>
>    I did read Diego's thesis rather thoroughly this weekend, and I'm
> curious if anybody else have done this ?

I've read it to some extend. A more accurate reading is in the TODO list
for this week.

>
>   I have also been looking at the linux support for MP, and NUMA ( which
> was added lately ), and linux does support affinity and NUMA in the
> latest kernels, however to take maximum advantage of this it would be
> quite resonable to do a native port to linux ( using clone instead of
> any thread library ), and only implement the synchronization element's
> needed by openMP.
>

Though Linux is the most probable host where people may want to run these
things, we don't want to target just Linux, do we? I think we should stick
with the POSIX compliance (or larger, if somebody has some idea) of gcc.

>   Still the first thing to do is to get openMP running with a threading
> library, and perhaps ( if smp safe ) a semaphore library.
>

We need to do two pretty orthogonal things, in my view:
a) hack the middle end (your suggestion below may be appropriate)
b) create the OpenMP library and all the framework (environment, locks
etc.) needed to get them working in some code.
I am not very well prepared on (a). Besides, I think that the tree-ssa is
still in evolution to bother it at this point with OpenMP directives. So
we can postpone (a) for the moment or better do some work on the tree-ssa
itself while waiting for an idea on how to use this framework for OpenMP.
(by the way, I got the FSF signed copy of my copyright assignment form: in
principle I *could* start to contribute patches to gcc...). Indeed 3 out
of 6 of us are also and probably mainly tree-ssa contributors.

As for (b), I am ready if you are. I think that, while missing the
compiler framework, a sensible thing to do is to assume that the threading
model will be POSIX and to hard code it into the library functions.
So building an OpenMP library that works when called from pthread code
would be a substantial step in the right direction.

>   As for the tasks ahead, I think it's not to hard to use the framework
> in the paper to target the GENERIC tree's ( which is the most resonable
> form to target IMHO ). The algoritms for an rather good implementation
> seem's to all be there, and the nice part is that if we extend the
> pragma handling and add a -fgomp to gcc, we should be able to leave most
> of the regulat stuff in place.
>

Cheers
Biagio