On Thu, Oct 20, 2005 at 09:17:02AM +0200, Volker Reichelt wrote:
> I just wanted to know what's the state of the gomp branch w.r.t
> bug reports. Does it make sense to already send bug reports to
> you or even add them to bugzilla?
I think so. We are getting near the point where we're looking
for real-life test cases to stress the code more than what we
can do reading the spec and writing synthetic tests.
I've added a libgomp component for reporting bugs against the
runtime, and an openmp keyword for reporting bugs against the
parsers. So if you have a c++ parsing problem, put "c++" in
the "component" field, and "openmp" in the "keywords" field.
Similarly for the Fortran or C front ends. If you get a crash
in "omp-low.c" or something, then it would make sense to put
"middle-end" in the component field.
> We've got a large C++ application that uses OpenMP and we are really
> interested in getting gomp work.
Cool. Note that I havn't implemented constructors and destructors
for variables in data sharing clauses yet. I'm hoping to have
that by Monday, but it is looking to be the most complicated part
of the c++ work.
> Here's one bug for starters:
>
> void foo()
> {
> int i;
> #pragma omp parallel for
> for ( i=0; i<10; ++i )
> continue;
> }
Ya know, I was thinking not a half-hour ago while I was looking
at the spec wrt exceptions in for-loop that "continue" probably
didn't work. I'll fix it tomorrow.
r~
