Tuneup using size_ratio

Niels Möller nisse at lysator.liu.se
Sat Dec 23 21:02:22 CET 2023


[ Resend, due to mailinglist error ] 

marco.bodrato at tutanota.com writes:

> I did try it, and the results are comparable with the results obtained
> with the current code.

Thanks for testing!

>> One thing that looks odd is
>>  /* Use ratio 5/6 when measuring, the middle of the range 2/3 to 1. */
>>
>> for the tuning of MUL_TOOM43_TO_TOOM54_THRESHOLD.

[...]

> Toom54 is not used. And toom43 is used in the range 2/3 .. 6/7.
>
> But if I measure the speed with the ratio 0.7 or 0.8 I get quite
> different thresholds, which one should we use?

Maybe crossover is rather flat, i.e., a wide range with very small
difference in performance? I hope these changes should make it easier to
investigate such things.

> By the way, one should add also the following lines:
>   { "mpn_toom53_mul",    speed_mpn_toom53_mul, FLAG_SR_OPTIONAL },
>   { "mpn_toom54_mul",    speed_mpn_toom54_mul, FLAG_SR_OPTIONAL },

Good catch, adding those.

I intend to push this change in a few days. Have a nice Christmas!

Regards,
/Niels

-- 
Niels Möller. PGP key CB4962D070D77D7FCB8BA36271D8F1FF368C6677.
Internet email is subject to wholesale government surveillance.


More information about the gmp-devel mailing list