A very fast Pari program (spin-off from 'New Mersenne software... Android')

[science_man_88]

Originally Posted by axn

3 min 7 sec / 11 sec = 17x
23min 7 sec / 1 min 38.5 sec = 14x

OP's code has better scaling than GMP-based one !

Paridroid gave 4.66 and 28.66 seconds with a halfass script:

try(p)=my(x=Mod(4,2^p-1));for(y=3,p,sqr(x)-2);x==0

[paulunderwood]

Originally Posted by science_man_88

Paridroid gave 4.66 and 28.66 seconds with a halfass script:

try(p)=my(x=Mod(4,2^p-1));for(y=3,p,sqr(x)-2);x==0

We can easily make SM's code quicker with RG's code posted elsewhere which does the mod reduction efficiently.

[science_man_88]

Originally Posted by paulunderwood

We can easily make SM's code quicker with RG's code posted elsewhere which does the mod reduction efficiently.

my paridroid stops working if I allocate more memory, so it'll depend on it not crashing with that, and such.

[ewmayer]

Originally Posted by science_man_88

Paridroid gave 4.66 and 28.66 seconds with a halfass script:

try(p)=my(x=Mod(4,2^p-1));for(y=3,p,sqr(x)-2);x==0

What exponents are those timings for, and on what compute hardware?

[science_man_88]

Originally Posted by thorken

Let's see colleague who is very good at insulting you without even being able to execute the application given your extensive experience. You say my program reads numbers from somewhere? Finally apprentice magician is very good insult without having much idea of ​​programming so it seems, make your an app for Movil that calculates numbers of Mersenne, here you are making comparisons with i7 and the cpu of a mobile. This program is for mobile and is the fastest on Android, the only one I've seen is a Google Play apk and it's very slow, this one is much faster and I'm talking about Android and mobile cpus, prime95 in AtoM n270 It is slower than this app in a GT-I9195. Greetings.

They were just saying that's one possible way some program without source given could be both fast and accurate to what's already known. As shown a interpretted script can run faster. Also I got that lucas Lehmer app you talked about, when it takes 27-28 seconds for 12711 then yes many programs could run faster. that's why coders use asymptotics.

Originally Posted by ewmayer

What exponents are those timings for, and on what compute hardware?

same exponents as OP, on the only working hardware I have at last check an android phone. I think it has a snapdragon processor, and PARIdroid was 2.11.0.1.5 edit2: seems my PARIdroid thinks they aren't prime though...

[ewmayer]

Originally Posted by science_man_88

same exponents as OP, on the only working hardware I have at last check an android phone. I think it has a snapdragon processor, and PARIdroid was 2.11.0.1.5 edit2: seems my PARIdroid thinks they aren't prime though...

OP listed timings for 3 exponents in 2 separate posts: 23209,86243 and 216091. Which 2 of these are you talking about?

[science_man_88]

Originally Posted by ewmayer

OP listed timings for 3 exponents in 2 separate posts: 23209,86243 and 216091. Which 2 of these are you talking about?

the latter 2 , but I realize why it was so quick forgot an x= part.

[ewmayer]

Originally Posted by science_man_88

the latter 2 , but I realize why it was so quick forgot an x= part.

So your simple Pari script running on Android is nearly 4x faster than GP2's GMP-based LL tester running on an AWS Skylake core?

[science_man_88]

Originally Posted by ewmayer

So your simple Pari script running on Android is nearly 4x faster than GP2's GMP-based LL tester running on an AWS Skylake core?

I typoed, so no. in fact without being able to not crash Paridroid with more memory allocated or just stalling I may not even be to even properly run the test.

[ewmayer]

Originally Posted by science_man_88

I typoed, so no. in fact without being able to not crash Paridroid with more memory allocated or just stalling I may not even be to even properly run the test.

So instead of repeatedly computing x = x^2-2 (mod 2^p-1), you set x = 4 and repeatedly computed x^2-2 without updating x ... so your final result should be x = 4, which means your x==0 of the final result should have come up 'false'. You didn't notice that?

[science_man_88]

Originally Posted by ewmayer

So instead of repeatedly computing x = x^2-2 (mod 2^p-1), you set x = 4 and repeatedly computed x^2-2 without updating x ... so your final result should be x = 4, which means your x==0 of the final result should have come up 'false'. You didn't notice that?

I didn't know the exponents created primes. I haven't put CRG's MeVec script back on my device.