floating point war casualty list

[Roman Lechtchinsky]

On Mon, 3 Jun 2002, Stephen L Moshier wrote:

> Again if they are not going to revert the changes, then the cray
> is still broken on the CVS trunk.  The fix is not trivial.

The fix is trivial (for a sufficiently broad definition of trivial). It
could probably be hacked within a couple of days. I don't think this is
feasible, though, because it will make the code even less maintainable
than it is now. The code in real.c should be simplified before attempting
that.

> They are thinking about a nice table-driven program for the bit formats
> and that will help.  But overall it is a longer term project because
> it affects all targets and entails considerable effort to test all
> the different formats.

Zack Weinberg has suggested to create a branch for this because it might
not be a good idea to do this on the mainline at this point in the release
cycle. The branch would have the following goals:

  - Refactor, simplify and document the code in real.c.
  - Fix the various problems identified in this and other threads.
  - Add support for 32/64-bit arithmetic (which is necessary for native
    builds on the Cray and desirable for speed reasons on other
    architectures).
  - Make TARGET_FLOAT_FORMAT a run-time parameter (which seems to be
    desirable for some targets).

Of course, this is only feasible if there are enough people who are
interested in this and who will be willing to help. I'm not sure if that
is the case at the moment.

Bye

Roman