Re: AC_PROG_CC_C99 and AC_OPENMP

[Bob Friesenhahn]

On Sun, 9 May 2010, Ralf Wildenhues wrote:
> I think Bob refers to different versions of the OpenMP standard,
> as opposed to different versions of some library implementation.
> Requiring one of those versions would make sense, similar to how
> AC_PROG_CC_C99 requests the 1999 version of the ISO C standard
> and not the 1989/90 one.  Of course, the actual test code could,
> besides checking the value of _OPENMP, also test for actual features
> of the respective standard version, similarly to how AC_PROG_CC_C99
> works.  It could do runtime tests when not cross-compiling, and fall
> back to link test/_OPENMP value otherwise.

Yes, exactly.  My own OpenMP-dependent software is only expected to 
work with OpenMP 2.0 (2002 version) or later.  Some other software 
might want to start with OpenMP 2.5 (2005 version) since that is the 
first one officially supported by GCC, or even OpenMP 3.0 (supported 
by GCC 4.4). Luckily, I have found OpenMP portability to good, and 
have only experienced failure with the SGI IRIX compiler.

At the moment, OpenMP feature tests don't seem to be very important 
since implementations should fully support the specification version 
they claim to support (and they seem to do so).  The main issue that I 
have had to resolve outside of AC_OPENMP is determining the name of 
the OpenMP runtime library, since that is compiler/platform dependent.

Bob
--
Bob Friesenhahn
address@hidden, http://www.simplesystems.org/users/bfriesen/
GraphicsMagick Maintainer,    http://www.GraphicsMagick.org/