Draft "Unsafe fp optimizations" project description.

[dewar@gnat.com]

>>Do you have any pointers to the literature for these discussions?

Sorry, no, I would start by going to a good computer science library. This
is one time when online searching is unlikely to be very helpful, since
much of this material is quite old. 

One source of useful input is the original discussions surrounding the
adoption of IEEE, and there are definitely a number of published papers
here (at least some of these appear as references in Sam's thesis). In
particular there is extensive discussion of the denormal-vs-flush issue,
which is relevant, given that archiectures like Alpha and MIPS have
backed off from the denormal orthodoxy recently.