[Gomp-discuss] TODO proposal

[Jacob Weismann Poulsen]

Hi all

Browsing the web the other day, I stumbled over the GOMP project.
I am very excited about the idea of bringing openMP 
support into GCC and I would be happy to contribute to it if I can.

I better admit that I haven't read through the entire mail-archive
yet but I have browsed the source code and played a bit with it. 
I have also tried to compile a TODO for a release 0.1 of GOMP (THIS
IS JUST A PROPOSAL). I guess that I'm not the only one who hasn't a 
clear idea of where to begin and a draft for a TODO will at least give 
us something to talk from. The draft for the TODO is attached. 
I have tried to keep the ambitions very low in the hope that a
experimental release within a short time-period (famous last words) 
is indeed feasible. 

The status of my playing around with the current source is:

As for the build environment it now builds a static as well as a 
shared library and installs them properly. Target 'dist' works for
gomp but doesn't construct the gcc-gomp source ball mentioned
in the TODO yet.

As for the samples in misc they now compile and link (target 'check').

As for the gcc front-end I have played around with the pragma-handler
based on the c-openmp.c proposal by Steven Bosscher, cf. 
http://mail.gnu.org/archive/html/gomp-discuss/2003-02/msg00076.html.
I totally agree that putting the openMP pragma handling in the pragma.c
stub is probably not the right thing to do (it is quite clear from the
mail archive that Steven and others came to the same conclusion).
Nevertheless, I have learned from playing with it too and others
might do the same. Moreover, the straightforward approach will give 
us room for experiments. Say that we wish to play with the data-structures 
that will be passed from the front-end on to the back-end, e.g. simply try 
to pass some information all the way, then we have a place to hook these 
things in. That is, the state of my gcc patch is merely for experiments 
and NOT a proposal for a final front-end solution. The current status
is that all the samples from app. A are (hopefully correctly :)) validated
on a line-based basis whenever one passed the 'fgomp' option to the
gomp-patched gcc.

Besides continuing to play with the front-end and library implementation, I 
plan to read through the mail-archive and will then try to boil it down to a 
short README file for the project.

Any comments ?

Best regards,

Jacob Weismann


-- 
<address@hidden>
Fingerprint: 9315 DC43 D2E4 4F70 3AA8  F8F0 9DA0 B765 F5C8 7D26

TODO
Description: Text document