> I found a small bug in Jon's negdi2 code yesterday. It will eventually
> produce something like the following:
> 
> 	negd	r0,r0
> 	movqd	0,r1
> 	subcd	r1,r1
> 
> :-) I changed this to do
> 
> 	negd	r1,r1
> 	negd	r0,r0
> 	subcd	0,r1
> 
> when %0 and %1 refer to the same object. So here comes yet another
> version of the patch. I hope I'm the only one who has already installed
> a full set of libraries compiled with the new compiler :-( If you have done,
> this as well, you will have to replace at least vfprintf.so...
> I was really glad that my cc1 was statically linked against the old version
> of libc.a.

Oops... I wasn't thinking about that one happening in place.  I
was more worried about putting a pair of subtracts in the wrong
order, or putting the args to the subtracts in the wrong order.
It was only after I saw "0 - x" turn into a library call that I
did that one at all.

Sorry.  Guess it's a good thing that I've been having problems
building my last SUP...  I've also found it handy to have the
binary tar files from 1.1 sitting on my disk beside my CVS
repository.  It made it easier to recover from a screwed up
/usr/libexec/cc1.  8-)

Jon Buller