Paul.Zimmermann at loria.fr
Thu Dec 10 15:07:48 CET 2009
> With bdiv_q now working, I wrote a simple divexact function, and I use
> it in mpn/gcdext.c where it needs to compute the second cofactor. As a
> result, the GCDEXT_DC_THRESHOLD on shell was reduced from 496 to around
> 415. I guess it would be reduced a little more more with a more clever
> bidirectional divexact.
> I added summary date at the threshold tables, see
> "cfg/meas avg: 0.80180" means that the average configured threshold
> divided by the average measured threshold is 0.80180. A significant
> move of a threshold!
do you mean meas/cfg instead (I guess "measured" is the threshold measured
by the tuning program, while "configured" is the one in gmp-mparams.h, i.e.,
the threshold from the previous day)?
I notice on the contrary SET_STR_DC_THRESHOLD got multiplied by 1.22 on
What would be nice would be to have those meas/cfg values also in the main
table on http://gmplib.org/devel/thresholds.html, with green colors to
indicate ratios < 0.90, and red colors for ratios > 1.10...
More information about the gmp-devel