Re: [Gomp-discuss] Start of an OpenMP Implementation Spec

[Ross Towle]

On Thu, 26 Aug 2004 11:35:49 -0400, Scott Robert Ladd
<address@hidden> wrote:

> One initial "decision" was to implement for Linux, and then port later
> to other architectures. I disagree; in keeping with the principles of
> GNU and GCC, OpenMP should be based on an abstract model with
> platform-specific implementations. Opinions?
> 

Scott,  I agree wholeheartedly with you.  The abstract model is
expressed by the OpenMP
helper routines (_OMP_MasterFunction, _OMP_SlaveFunction,
_OMP_Barrier, ...)  The
OpenMP helper routines follow the abstract model described in the
OpenMP2.0 Spec.

Some helper routines (_OMP_Barrier) will have the underlying platform
specific code in it
because the routine is so simple.  Other helper routines
(_OMP_MasterFunction) will call
additional helper function where the platform specific code will be found.

This moves the platform-specific implimentation out of the compilers
and into some
helper routines in the OpenMP runtime library.

Having OpenMP helper routines is a key change from previous proposals
before this
group.  I t makes the transformations in the "middle end" (before
gimplification) more stylized.
The platform-specific stuff is moved into runtime routines.  And there
will be less code-bloat.

-Ross A Towle