> From: "J.T. Conklin" <jtc@cygnus.com>
> 
> Please excuse the intrusion of a non-pc532 user...

No problem, we take all kinds here. 8-) Interest is the only qualifier,
not hardware...

> I've been examining the code generated for quad arithmetic, both gcc's
> open coding of quad operations and from the functions in libc/quad/*.
> Let's just say the code is suboptimal.

I noticed this about a year ago, and started working on it about 6 months
ago.  Matthias improved my changes some, and adddi3, subdi3, and negdi2
work great for us.  We are kinda working on muldi3 and spinning our wheels
at the moment.  I must admit it's mostly my fault that the patch hasn't
been sent to the FSF maintainer, and I know Phil has it, but perhaps
not the most bug-free version.  I don't have much spare time, but it's
getting better as my daughter is getting into the 6-9 month old range.
 
> I discovered that the ns32k machine description had no patterns for
> DImode integers.  Just for yuks, I quickly hacked functions for add,
> subtract, and negate.  Needless to say, my modified gcc is able to
> generate much smaller (and hopefully faster!) code for those 64 bit
> operations.

Yes on both smaller and faster.  But if you get it wrong, it shows up
in wierd places and at unusual times.  (Examples: telling it subdi3
can exchange operands will mess up the 9660 FS, but you don't see that
part of the insn showing up too often elsewhere.)

> Since I don't have a pc532 of my own (I am working on borrowing one so
> I can do more performance enhancements), I'm wondering if there are
> any NetBSD/pc532 users interested in helping me tune the code generator.

I'll send you our stuff if you like... (Or check the port-pc532
mail archives) The more help we get with making it run faster, the
better.

Jon Buller