Re: parallel bnfinit

[Bill Allombert <Bill.Allombert@math.u-bordeaux.fr>]

On Fri, Dec 11, 2020 at 09:34:29AM +0100, Bill Allombert wrote:
> Dear PARI developers,
> 
> With Loïc Grenié, we have experimented with a parallel version
> of bnfinit (the git branch bill-bnfinit-parallel).
> 
> I managed to compute bnfinit(x^97-x-1) in 7h, 16min on 128 cores.
> (518h, 21min of total time).
> The regulator is 2.422012163387226001 e57

Using the same branch, I managed to compute:

bnfinit(x^101-x-1) in 23h, 26min
on 128 cores. (1821h, 17min of total time).
The regulator is  6.157929707093319559395563978 e60

bnfinit(x^104-4*x^79+x^78+6*x^54-3*x^53+x^52-4*x^29+3*x^28-2*x^27+x^26+x^4-x^3+x^2-x+1)
in 5h, 18min on 128 cores. (275h, 6min of total time).
The regulator is 5.429592086253841570125004087 e59

bnfinit(polsubcyclo(251,125)) in 5h, 48min on 128 cores. 
(11h, 37min of total time).
The regulator is 2.735957758094112219294067232 e110
The degree is 125
The discriminant is 3.6269491237345921613053902415015024776E297
 
This one is both the highest degree and highest discriminant for which I
managed to compute the bnfinit so far. We see that this example would
have worked without parallelism in less than 12 hours.

Cheers,
Bill.