p.s: Re. my previous message about roundoff errors, you
Prime95 users may be wondering why M18899929 would cause
roundoff warnings - after all, it's well within the safe
range of exponents for 1024K FFT length. The reason is
that I was using FFT length 960K, which Mlucas supports
as an intermediate runlength between 896K and 1024K.
I tested this exponent in order to see if the program
might be able to handle an upper exponent limit of 18.9M
for this FFT length. Based on the number of RO warnings
I got and the number of times I had to restart the run
to work around fatal errors, the answer is clearly no -
but that's precisely why we do this kind of QA testing.
Also, since Brian was using Prime95 for his run of the
same exponent, he was using FFT length 1024K, so in fact
it wouldn't have mattered whether he was running on a P4
or not - at 1024K, either x86 register-double precision
or the less-precise SSE2 instructions would be more than
adequate for this exponent.
-Ernst
