GMP on PowerPC 970

[Torbjorn Granlund]

Brendan Younger <bry at mandli.com> writes:

  I didn't use any of the mpn/powerpc64 code aside from umul_ppmm(), primarily
  since it was written for assemblers which support the POWER mnemonics and
  Apple's version of "as" doesn't.  Hence, everything compiled was generic
  C-code.

I understood that.  But you might have used longlong.h macros,
correctly or not.

  Also, there is no 64bit/32bit dichotomy on the PowerPC 970; any
  program can call a 64-bit instruction at any time so there is no slow-down
  when using 64-bit code.

Except that instructions like adde and subfe change behaviour,
which is a really major problem to GMP.

  I would be interested in fixing GMP to work with
  "longlong" limbs, but I'd like to have the support of someone who knows more
  about the internals of GMP than I do.

It is not terribly hard, and in fact we have it working in the
current CVS sources.

-- 
Torbjörn