This is the mail archive of the
mailing list for the libstdc++ project.
Re: std::norm improvement
On Wed, Mar 4, 2009 at 9:32 AM, Jaroslav Hajek <email@example.com> wrote:
> On Wed, Mar 4, 2009 at 9:22 AM, Andrew Pinski <firstname.lastname@example.org> wrote:
>> " Still, I think
>> gcc is pointlessly (is that an English word?) decreasing performance
>> if ffast-math is not on (it's not on even with -O3).
>> No it is not, this is the problem here. That is from the thread you
>> wrote in.
> Yes, I've learned afterwards in the same thread that it is not pointless.
>> You did not test every single input to figure out if your
>> implementation is better or worse. ?I think it might be best if you
>> tested every input include NaNs and infs. ?That is where some problems
>> come into play also I think.
> You mean like every pair of real numbers? Isn't there about 2^128 of
> them (for double)? Even if a single test took a mere picosecond, that
> would still eat up a trillion years...
> I can certainly test all combinations of Infs and NaNs, if that's what
> you meant. I'll report back with the results.
Hmm, it seems that there's no simple fix to achieve (Inf, NaN) giving
Inf and (x, NaN) giving NaN, which is probably what is wanted. Sorry
for the noise.
RNDr. Jaroslav Hajek
computing expert & GNU Octave developer
Aeronautical Research and Test Institute (VZLU)
Prague, Czech Republic