--- Gio 3/9/09, Dave Korn ha scritto: > Da: Dave Korn > Oggetto: Re: std::arg() bug : not repetitive ? > A: cygwin-apps cygwin.com > Data: Giovedì 3 settembre 2009, 06:07 > Brian Ford wrote: > > On Wed, 2 Sep 2009, Marco Atzeri wrote: > > > >> I have no clue if -march=pentium4 is acceptable > for AMD cpu's. > > > > FWIW, we have been using -march=pentium4 -mfpmath=sse > successfully in a > > very large software project since November 2003 > without incident. We have > > run on AMD Opterons and a variety of Intel processors > without issue > > (given that they were at least Pentium 4's, of > course). > > > > In http://lkml.indiana.edu/hypermail/linux/kernel/0904.2/00818.html > it is > argued that: > > > Nvidia started requiring SSE in their binary drivers, > but have since > > backed out that requirement, so it is possible to make > people realize > > their code has unreasonable requirements. > > If we turn on SSE in the distro, we block anyone > using Pentium2 or early > (pre-XP) Athlon CPUs from using Cygwin. I think that > might be a step too far. > In the place where I was working until only last November, > we had a ton of > old P2 boxes around the place, doing useful jobs, and they > all had Cygwin on > them. It would have crippled us if we suddenly had to > upgrade them all one day. > > OTOH maybe we could hack up a software SSE emulation > for those users... > > cheers, > DaveK > Hi Dave,
I am trying to convince octave developers to add an additional z1!=z2 before the comparison of abs(z1) < abs(z2) and arg(z1) < arg(z2). This should resolve the truncation problem of arg and abs when comparing two equal complex numbers. Eventually I will add that only for cygwin, it seems simpler than non standard arch settings for the package. Thanks Marco
