Primo/FastECPP reservation thread

[paulunderwood]

Originally Posted by sweety439

OK, also you can see https://oeis.org/A092559, current the larger prime factor of 2^A092559(n)+1 have been proven prime for all a(n) up to a(52) = 117239, why you skip a(53) = 120788 and prove a(54) = 127031?

https://primes.utm.edu/top20/page.php?id=67 was used as a source, not OEIS.

[R. Gerbicz]

Originally Posted by paulunderwood

This number failed at Size [8] 215618 bits with: mu_div_qr GNU MP assertion failed cy == 0.

I am now attempting to certify the cofactor of M221509 (prp66673).

Originally Posted by paulunderwood

The same error has occurred at Size [2]. Where is this bug?

Taking W127031.

Have you used for this my corn code ?
Notice that even if in some stages the code fails then likely it is a persistent bug in gmp and will fail later. In later stages you are doing the same algorithms just with a barely random smaller/larger numbers [Where the first stage could be a little exceptional if N has a special form].

[paulunderwood]

Originally Posted by R. Gerbicz

Have you used for this my corn code ?
Notice that even if in some stages the code fails then likely it is a persistent bug in gmp and will fail later. In later stages you are doing the same algorithms just with a barely random smaller/larger numbers [Where the first stage could be a little exceptional if N has a special form].

Of course I am using your corn.c. I'd be a fool not to utilize the huge speed up.

I had R86453 fail with this error besides these two, before corn.c. I think the problem lies with GMP and sheer size of the numbers being certified, but I could be wrong.

[chalsall]

Originally Posted by paulunderwood

Of course I am using your corn.c. I'd be a fool not to utilize the huge speed up.

Guys... Yoga breaths...

Could we please just try to press the reset button from time to time?

We are all on the same side. Or, at least, most of us. Correct?

On the same page. Noise is something to be expected.

Correct?

P.S. I sometimes talk to the Universe. I find it catharitic.

[paulunderwood]

Originally Posted by chalsall

Guys... Yoga breaths...

Could we please just try to press the reset button from time to time?

We are all on the same side. Or, at least, most of us. Correct?

On the same page. Noise is something to be expected.

Correct?

P.S. I sometimes talk to the Universe. I find it catharitic.

There was no animosity intended. I was merely conveying how much I admire Robert's work that is corn.c, and agreeing with him that the error is most likely with GMP's mpn/mu_div_qr.c somewhere. That is one hell of a complicated bit of code and should be left to the authorities to debug.

I am hoping this bug is fixed because I really want to have another go at certifying R86453 even if that takes 2 years!

[chalsall]

Originally Posted by paulunderwood

There was no animosity intended. I am hoping this bug is fixed because I really want to have another go at certifying R86453 even if that takes 2 years!

Super cool.

Some take questioning as an attack. When the exact opposite is intended.

[chalsall]

Originally Posted by chalsall

Super cool. Some take questioning as an attack. When the exact opposite is intended.

Could I please ask a sincere question?

Why do people assume (Ass U Me) a question means an attack?

Like... Seriously... Get over yourself... You are not that important...

I almost feel like I resonate with Ricky Gervais...

[VBCurtis]

Originally Posted by chalsall

Like... Seriously... Get over yourself... You are not that important...

Please take this part of your own advice, and stop posting noise in thread after thread. We're tired of your non sequiturs.
Give us a break for a month: If your post doesn't have any math content, delete it instead of posting it.

Give it a try. Help us cut the noise in the forum. Muzzle yourself.

[gorignak]

I'd like to reserve M130439 cofactor PRP39261.

[R. Gerbicz]

Originally Posted by paulunderwood

I had R86453 fail with this error besides these two, before corn.c. I think the problem lies with GMP and sheer size of the numbers being certified, but I could be wrong.

What I can see even running on an 8k digits input that mpn_preinv_mu_div_qr is called only in trial factorization step.
And most likely only on mpz_mod.
Could you run this tiny test, testing mpz_mod (directly not on random numbers, but it is fine since there is no special code for p^e mod q^f, where p,q is odd):

#include <stdio.h>
#include <time.h>
#include <math.h>
#include "gmp.h"
#define mult 1.4 // for the next bitlevel: nbits=mult*nbits.
#define div 2 // to divide n by roughly n^(1/d)
#define L 3 // to see more tests increase it, for given nbits it will do L^2-L tests, for L<=7
#define maxnbits 12e9

int main(void){
    
    int ptab[]={3,5,7,11,13,17,19};
    mpz_t n,d,r;
    FILE* fout;
    fout=fopen("tests.txt","a+");
    
    mpz_init(n);
    mpz_init(d);
    mpz_init(r);
    
    for(long nbits=1024;nbits<=maxnbits;nbits=(long)nbits*mult){  
        printf("Testing %ld bits n numbers.\n",nbits);
        for(int i=0;i<L&&i<7;i++){
            long E1=(double)nbits*(log(2)/log(ptab[i]));
            mpz_ui_pow_ui(n,ptab[i],E1);
            
            for(int j=0;j<L&&j<7;j++)if(i!=j){
                long E2=((double)1.0/div)*(double)nbits*(log(2)/log(ptab[j]));
                
                mpz_ui_pow_ui(d,ptab[j],E2);
                time_t sec=time(NULL);
                mpz_mod(r,n,d);
                
                int parity=(mpz_even_p(r)!=0?0:1);
                printf("calc parity(%d^%ld mod %d^%ld)=%d, time=%ld sec.\n",
                       ptab[i],E1,ptab[j],E2,parity,time(NULL)-sec);
                fprintf(fout,"calc parity(%d^%ld mod %d^%ld)=%d, time=%ld sec.\n",
                        ptab[i],E1,ptab[j],E2,parity,time(NULL)-sec);
                fflush(fout);
            }
        }
    }
    fclose(fout);
    mpz_clear(n);
    mpz_clear(d);
    mpz_clear(r);
    
    return 0;
}

It will save also the output. Maybe this is already an overshoot because on the code you'll use smaller numbers due to constants. (With 16 GB Ram it could run, for smaller machines it could die).

[paulunderwood]

The results of the tests are attached.