NO SUBJECT

[tprince at cat dot e-mail dot com]

>>Therefore, I think the stuff we'll tend to "fix" by adopting my
proposal will be function return values, offhand, not by what
we think of as "normal" spills -- because it seems that normal
spills in straight-line code are pretty rare.<<

Not so rare that people don't complain about unexpected results
arising from the difference between 80-bit and 64-bit evaluation,
which ought to be corrected by the 80-bit spilling scheme.

>>That's also why I don't think we'll see a big performance hit on
most code, and, of the code that *does* see it, I would't be
surprised if it turned out that a substantial portion of it was
previously producing inadequately correct results in its "fast"
mode.<<

I'd add Toon Moene to the list if I had his address with me.  I see
that he agreed with me that we must be certain to have aligned
storage of spills (128-bit alignment for 80-bit storage).  And, to
agree with a point I think he made, we have tested most of our
single precision codes enough to believe they work fine most of
the time without extra precision, so let's not cut any corners when
we try to make improvements.  Let's be certain that any slightly
different results which do show up are at least as good.

Dr. Timothy C. Prince
Consulting Engineer
Solar Turbines, a Caterpillar Company
alternate e-mail: tprince@computer.org

           To:                                              INTERNET - IBMMAIL
                                                            N3356140 - IBMMAIL