Date: Sun, 25 Mar 2001 20:00 PST
To: [EMAIL PROTECTED]
Subj: small bug in Mlucas 2.7b error checking

In doing some testing of exponents very near the upper
limit of what Mlucas 2.7b allows for 640K FFT length,
I discovered a bug which in some case will allow per-
iteration roundoff checking to be turned off when the
run passes the iteration number (call it ITER_ECHECK_OFF)
set in the mlucas.cfg file, even when the code detects
a dangerous fractional error in the iterations leading
up to ITER_ECHECK_OFF. (In which case roundoff checking
is supposed to remain on for the rest of the LL test
irrespective of the value of ITER_ECHECK_OFF.)

To see whether you need to take any action, do a 'grep
err *stat' in your mersenne test directory. (And until v2.7c
comes out, repeat this whenever your current run has gotten
about a day into its latest exponent.)
If your current exponent has yielded no error warnings, you
need do nothing. If you see one or more error warnings, do
one of the following (your choice):

1) Kill and restart the run. On restart the program parses
the .stat file and if it finds one or more fractional error
warnings it sets the appropriate internal flag which makes
error checking stay on for the entire run.

2) Set the value of ITER_ECHECK_OFF on line 4 of the mlucas.cfg
file to some value >= your current exponent. After the exponent
finishes, you can set it back to the old value. (If there's not
a big speedup from turning error checking off on your system,
you may want to leave it at a large value.)


Cheers,
-Ernst


_________________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
Mersenne Prime FAQ      -- http://www.tasam.com/~lrwiman/FAQ-mers

Reply via email to