On Sat, Feb 24, 2001 at 07:53:45PM +0000, Philip Blundell wrote: > >Do you have anything apart from its comment to suggest that it even > >works on arm3? > > Nope.
Ah. > >I don't believe that the non ARM3 code should actually work, and I see no > >reason why (if I've found the bug) the corrected code should not work > >equally well on ARM4 (but more slowly than an umull ?) > > Great. Thanks for the analysis. I'll build a package with that change and > see if it helps. Please beware that I really didn't understand it (or the gcc inline asm code and how exactly it specifies registers, including how to stop input and output values being in the same registers ) well enough to be sure that it was doing exactly what I think it ought to be doing to get the multiply correct. Merely that I found something that seemed to be in error. Hopefully it now works, but there might be 1 in 2**32 (or 1 in 2**64) cases where it still doesn't Nicholas Clark
