master (v.1.1.0) vs develop (3/12/2020)

[mdehoogh]

I'm using libtommath to some extent in an interpreter I'm writing, and because the current develop branch was not giving me the error on mp_set_double compilation that master (v.1.1.0) branch was giving me, I was intent on switching over to using develop instead. However, some of the functions I was using have turned into defines (mp_incr, mp_decr, mp_isodd, mp_sqr); mp_n_root was renamed (I guess) to mp_root_n and mp_toradix to mp_to_radix; MP_YES and MP_NO were removed. Using the develop tommath.h was much worse than an adaptation of the old tommath.h, so I switched to using the latter. One problem I encountered was that the type of the size parameter in mp_radix_size changed from int to size_t, causing serious problems when passing in a local int (overwriting a pointer). For now I've added a compiler flag to check whether to use size_t or int but it is a nuisance to have to put that in a C source file:

#ifdef M_MP_DEVELOP size_t arepsize=0; // MDH@13MAR2020: changing type int to size_t (which is larger), so that we should be able to use it with libtommath-develop #else int arepsize=0; #endif

Furthermore, when I use the develop version my factorial algorithm is say 15% slower with develop than with master. Any idea why?

Kind regards, Marc P. de Hoogh The Netherlands

[czurnieden]

[…]the error on mp_set_double compilation that master (v.1.1.0) branch was giving[…]

What error do you get? There is a problem with MacOS, see #159 for the details. The actual commit to fix it is in #476

Please be aware that the current version in master is 1.2.0. I don't know if the fix from #476 gets backported in the near future, we are currently a bit understaffed and overworked here, so it might take a little more time to get 1.2.1 out.

Furthermore, when I use the develop version my factorial algorithm is say 15% slower with develop than with master. Any idea why?

15% is quite significant! What algorithm? Naive, binary splitting, Borwein, Borwein-Schönberg? Do you have the implementation public? (I apologize in advance if you have it in your github repository and I was too blind to find it)

[mdehoogh]

Will have a look.

Op 13 mrt. 2020 18:36 schreef Christoph Zurnieden [email protected]:

[…]the error on mp_set_double compilation that master (v.1.1.0) branch was giving[…]

What error do you get? There is a problem with MacOS, see #159https://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_libtom_libtommath_issues_159&d=DwMFaQ&c=XYzUhXBD2cD-CornpT4QE19xOJBbRy-TBPLK0X9U2o8&r=5ZdsQceb57-oiBH6dKANmB5EQ29Bw79DlNhWrSmRRjI&m=omJsp-ldy4OV2mrliQj2joS3O44BCBbVKuibFpzSCWs&s=3_NMkdM9Tov0fb3qrIXAobpsFJQUpQFBnDXRi_q9vsk&e= for the details. The actual commit to fix it is in #476https://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_libtom_libtommath_pull_476&d=DwMFaQ&c=XYzUhXBD2cD-CornpT4QE19xOJBbRy-TBPLK0X9U2o8&r=5ZdsQceb57-oiBH6dKANmB5EQ29Bw79DlNhWrSmRRjI&m=omJsp-ldy4OV2mrliQj2joS3O44BCBbVKuibFpzSCWs&s=r4NRl2shp6-WzRFsptEzOUcFewnYb5mUSZxgGZ81Bms&e=

Please be aware that the current version in master is 1.2.0. I don't know if the fix from #476https://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_libtom_libtommath_pull_476&d=DwMFaQ&c=XYzUhXBD2cD-CornpT4QE19xOJBbRy-TBPLK0X9U2o8&r=5ZdsQceb57-oiBH6dKANmB5EQ29Bw79DlNhWrSmRRjI&m=omJsp-ldy4OV2mrliQj2joS3O44BCBbVKuibFpzSCWs&s=r4NRl2shp6-WzRFsptEzOUcFewnYb5mUSZxgGZ81Bms&e= gets backported in the near future, we are currently a bit understaffed and overworked here, so it might take a little more time to get 1.2.1 out.

Furthermore, when I use the develop version my factorial algorithm is say 15% slower with develop than with master. Any idea why?

15% is quite significant! What algorithm? Naive, binary splitting, Borwein, Borwein-Schönberg? Do you have the implementation public? (I apologize in advance if you have it in your github repository and I was too blind to find it)

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHubhttps://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_libtom_libtommath_issues_479-23issuecomment-2D598836331&d=DwMFaQ&c=XYzUhXBD2cD-CornpT4QE19xOJBbRy-TBPLK0X9U2o8&r=5ZdsQceb57-oiBH6dKANmB5EQ29Bw79DlNhWrSmRRjI&m=omJsp-ldy4OV2mrliQj2joS3O44BCBbVKuibFpzSCWs&s=D2_S-msH9vmTO_22LZXJ59w448CQfXWIo8tvV0LKVM4&e=, or unsubscribehttps://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_notifications_unsubscribe-2Dauth_ACELO24SOJFVGXUSQLU3QYLRHJVIPANCNFSM4LHCLBNQ&d=DwMFaQ&c=XYzUhXBD2cD-CornpT4QE19xOJBbRy-TBPLK0X9U2o8&r=5ZdsQceb57-oiBH6dKANmB5EQ29Bw79DlNhWrSmRRjI&m=omJsp-ldy4OV2mrliQj2joS3O44BCBbVKuibFpzSCWs&s=Mgs3rcqzJuviYOT_ntUD3SxG-hTETgCW2LKsANLtV7Y&e=.

[mdehoogh]

See comments below.

Op 13 mrt. 2020, om 18:36 heeft Christoph Zurnieden <[email protected]mailto:[email protected]> het volgende geschreven:

[…]the error on mp_set_double compilation that master (v.1.1.0) branch was giving[…]

What error do you get?

I guess it’s the MacOS issue.

There is a problem with MacOS, see #159https://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_libtom_libtommath_issues_159&d=DwMFaQ&c=XYzUhXBD2cD-CornpT4QE19xOJBbRy-TBPLK0X9U2o8&r=5ZdsQceb57-oiBH6dKANmB5EQ29Bw79DlNhWrSmRRjI&m=omJsp-ldy4OV2mrliQj2joS3O44BCBbVKuibFpzSCWs&s=3_NMkdM9Tov0fb3qrIXAobpsFJQUpQFBnDXRi_q9vsk&e= for the details. The actual commit to fix it is in #476https://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_libtom_libtommath_pull_476&d=DwMFaQ&c=XYzUhXBD2cD-CornpT4QE19xOJBbRy-TBPLK0X9U2o8&r=5ZdsQceb57-oiBH6dKANmB5EQ29Bw79DlNhWrSmRRjI&m=omJsp-ldy4OV2mrliQj2joS3O44BCBbVKuibFpzSCWs&s=r4NRl2shp6-WzRFsptEzOUcFewnYb5mUSZxgGZ81Bms&e=

Ok, I’ll have a look.

Please be aware that the current version in master is 1.2.0. I don't know if the fix from #476https://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_libtom_libtommath_pull_476&d=DwMFaQ&c=XYzUhXBD2cD-CornpT4QE19xOJBbRy-TBPLK0X9U2o8&r=5ZdsQceb57-oiBH6dKANmB5EQ29Bw79DlNhWrSmRRjI&m=omJsp-ldy4OV2mrliQj2joS3O44BCBbVKuibFpzSCWs&s=r4NRl2shp6-WzRFsptEzOUcFewnYb5mUSZxgGZ81Bms&e= gets backported in the near future, we are currently a bit understaffed and overworked here, so it might take a little more time to get 1.2.1 out.

Will check against 1.2.0 as well.

Furthermore, when I use the develop version my factorial algorithm is say 15% slower with develop than with master. Any idea why?

15% is quite significant

What algorithm? Naive, binary splitting, Borwein, Borwein-Schönberg?

Nothing fancy, ordinary multiplication, but I’ll investigate further.

See snippet below (Mbiginteger is defined as mp_int)

Mbiginteger* _result=_getBiginteger(6); // the smallest value to return
if(_result){
    // we could store fac values in a special list with index equal to the argument, in which case we could look up the starting value
    // we could start at some intermediate value????
    Mbiginteger *_multiplier=_getBiginteger(3);
    if(_multiplier){
        while(mp_cmp(_multiplier,_finalmultiplier)==MP_LT){
            if(mp_incr(_multiplier)!=MP_OKAY){outputError("Failed to increment a big integer");_result=NULL;break;} // if we fail to increment break
            if(mp_mul(_result,_multiplier,_result)!=MP_OKAY){outputError("Failed to multiply a big integer");_result=NULL;break;}
            //////////if(amVerbose())outputBigInteger("Result so far: '",result,"'.");
        }
        // get rid of intermediate big integers
        free_biginteger(_multiplier);
    }else
        outputError("Failed to create big integer 3");
}else
    outputError("Failed to create big integer 6");
free_biginteger(_finalmultiplier);
if(amVerbose())outputBiginteger("Result of applying the factorial() function: '",_result,"'.\n");
return (_result?_getBigintegerValue(_result,true):NULL);

Do you have the implementation public?

No, it’s still private.

(I apologize in advance if you have it in your github repository and I was too blind to find it)

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHubhttps://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_libtom_libtommath_issues_479-23issuecomment-2D598836331&d=DwMFaQ&c=XYzUhXBD2cD-CornpT4QE19xOJBbRy-TBPLK0X9U2o8&r=5ZdsQceb57-oiBH6dKANmB5EQ29Bw79DlNhWrSmRRjI&m=omJsp-ldy4OV2mrliQj2joS3O44BCBbVKuibFpzSCWs&s=D2_S-msH9vmTO_22LZXJ59w448CQfXWIo8tvV0LKVM4&e=, or unsubscribehttps://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_notifications_unsubscribe-2Dauth_ACELO24SOJFVGXUSQLU3QYLRHJVIPANCNFSM4LHCLBNQ&d=DwMFaQ&c=XYzUhXBD2cD-CornpT4QE19xOJBbRy-TBPLK0X9U2o8&r=5ZdsQceb57-oiBH6dKANmB5EQ29Bw79DlNhWrSmRRjI&m=omJsp-ldy4OV2mrliQj2joS3O44BCBbVKuibFpzSCWs&s=Mgs3rcqzJuviYOT_ntUD3SxG-hTETgCW2LKsANLtV7Y&e=.

[sjaeckel]

I'm closing this as there was no action for too long. Please re-open if you have further questions/comments.