New GMP!!

[BotXXX]

There is a release candidate of GMP 4.2 available. (version number 4.1.99)

The speed increase is quite nice. All checks of gmp and ecm passed without errors. Hopefully the stable 4.2 will be available soon.

BotXXX P4 - GMP-ECM 6.0.1 [powered by GMP 4.1.4] [ECM]
Input number is 190212714790675921146226277918749729345010106518539018612340134123144725152859510370277162195385423282526682481901533977811897620309871580337689953713543555497646216936690778372438516576170136727377556113841523 (210 digits)
Using B1=1000000, B2=839549779, polynomial Dickson(3), sigma=2571227718
Step 1 took 36638ms
Step 2 took 15998ms

BotXXX P4 - GMP-ECM 6.0.1 [powered by GMP 4.1.99] [ECM]
Input number is 190212714790675921146226277918749729345010106518539018612340134123144725152859510370277162195385423282526682481901533977811897620309871580337689953713543555497646216936690778372438516576170136727377556113841523 (210 digits)
Using B1=1000000, B2=839549779, polynomial Dickson(3), sigma=2844247108
Step 1 took 32546ms
Step 2 took 13498ms



BotXXX P4 - GMP-ECM 6.0.1 [powered by GMP 4.1.4] [ECM]
Input number is 190212714790675921146226277918749729345010106518539018612340134123144725152859510370277162195385423282526682481901533977811897620309871580337689953713543555497646216936690778372438516576170136727377556113841523 (210 digits)
Using B1=3000000, B2=4016636513, polynomial Dickson(6), sigma=3250209424
Step 1 took 109222ms
Step 2 took 42896ms

BotXXX P4 - GMP-ECM 6.0.1 [powered by GMP 4.1.99] [ECM]
Input number is 190212714790675921146226277918749729345010106518539018612340134123144725152859510370277162195385423282526682481901533977811897620309871580337689953713543555497646216936690778372438516576170136727377556113841523 (210 digits)
Using B1=3000000, B2=4016636513, polynomial Dickson(6), sigma=4285899244
Step 1 took 98001ms
Step 2 took 35871ms

[smh]

Can you make P4 executables available?

[Phil MjX]

just try gmp-ecm 6.1beta2 with --enable-asm-redc
it rocks > upper B2 bound, 10 to 30% speed increase on a celeron M depending B1 size...
(currently compiling it with gmp 4.1.99 to see what happens)!!!
I'll post my benchs here !

Bye.
Philippe.

[Mystwalker]

On my P4, I get a nice speedup for stage2, but hardly any for stage1:

GMP-ECM 6.0.1 [powered by GMP 4.1.4] [ECM]
Input number is 190212714790675921146226277918749729345010106518539018612340134123144725152859510370
2771621953854232825266824819015339778118976203098715803376899537135435554976462169366907783724385165
76170136727377556113841523 (210 digits)
Using B1=3000000, B2=4016636513, polynomial Dickson(6), sigma=638561964
Step 1 took 116670ms
Step 2 took 45797ms


GMP-ECM 6.1 [powered by GMP 4.2] [ECM]
Input number is 190212714790675921146226277918749729345010106518539018612340134123144725152859510370
2771621953854232825266824819015339778118976203098715803376899537135435554976462169366907783724385165
76170136727377556113841523 (210 digits)
Using B1=3000000, B2=4281751120, polynomial Dickson(6), sigma=2247898896
Step 1 took 112375ms
Step 2 took 55907ms

Interestingly, gmp already reports 4.2...

But my version of 6.1 (as of August 17th) was slower than 6.0.1

With --enable-asm-redc:

GMP-ECM 6.1 [powered by GMP 4.2] [ECM]
Input number is 190212714790675921146226277918749729345010106518539018612340134123144725152859510370
2771621953854232825266824819015339778118976203098715803376899537135435554976462169366907783724385165
76170136727377556113841523 (210 digits)
Using B1=3000000, B2=4281751120, polynomial Dickson(6), sigma=1968267451
Step 1 took 175781ms
Step 2 took 59438ms

Without --enable-asm-redc:

GMP-ECM 6.1 [powered by GMP 4.2] [ECM]
Input number is 190212714790675921146226277918749729345010106518539018612340134123144725152859510370
2771621953854232825266824819015339778118976203098715803376899537135435554976462169366907783724385165
76170136727377556113841523 (210 digits)
Using B1=3000000, B2=4281751120, polynomial Dickson(6), sigma=2247898896
Step 1 took 112375ms
Step 2 took 55907ms

[Phil MjX]

Hi !

my config celeron M 360 (1.4 GHz) 512M ram
Some tests (from the slowest to the fastest):

precompiled pentium M binary (thanks Mystwalker)
GMP-ECM 6.0.1 [powered by GMP 4.1.4] [ECM]
Input number is 491853986948777594673218825453230605859928447069055088395227463891218667208716253406266089004846828871229829486993713464642129 (126 digits)
Using B1=3000000, B2=4016636513, polynomial Dickson(6), sigma=1798624804
Step 1 took 97062ms
Step 2 took 38657ms

precompiled binary for p4northwood
GMP-ECM 6.0.1 [powered by GMP 4.1.4] [ECM]
Input number is 491853986948777594673218825453230605859928447069055088395227463891218667208716253406266089004846828871229829486993713464642129 (126 digits)
Using B1=3000000, B2=4016636513, polynomial Dickson(6), sigma=1692105302
Step 1 took 93874ms
Step 2 took 39431ms

Mingw binary (gmp compiled with pentium-m flags) without --enable-asm-redc
GMP-ECM 6.1-beta2 [powered by GMP 4.2] [ECM]
Input number is 491853986948777594673218825453230605859928447069055088395227463891218667208716253406266089004846828871229829486993713464642129 (126 digits)
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=2088433237
Step 1 took 89765ms
Step 2 took 33657ms

Mingw binary (gmp compiled with pentium-m flags) with --enable-asm-redc
GMP-ECM 6.1-beta2 [powered by GMP 4.1.4] [ECM]
Input number is 491853986948777594673218825453230605859928447069055088395227463891218667208716253406266089004846828871229829486993713464642129 (126 digits)
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=2586761473
Step 1 took 82515ms
Step 2 took 33610ms

Mingw binary (gmp compiled with pentium-m flags) with --enable-asm-redc
GMP-ECM 6.1-beta2 [powered by GMP 4.2] [ECM]
Input number is 491853986948777594673218825453230605859928447069055088395227463891218667208716253406266089004846828871229829486993713464642129 (126 digits)
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=661877002
Step 1 took 76719ms
Step 2 took 32875ms

great speed improvement !!!
(the gmp-ecm 6.1beta2 version is not the same as Mystwalker one : look for B2 bounds)

thanks
best regards

Philippe.

[Mystwalker]

Impressive results.

From when are your 6.1beta2 sources?

[BotXXX]

Originally Posted by smh

Can you make P4 executables available?

I won't distribute the release candidate, but when GMP 4.2 will be final i can release a P4 / MinGW build executable. Have no idea how to compile gmp and ecm for other platforms from mingw.

the release candidate indeed reports itself as 4.2 but i changed it to 4.1.99 since that was the version number from the tarball ;)

[Phil MjX]

Originally Posted by Mystwalker

Impressive results.

From when are your 6.1beta2 sources?

You welcome :

http://ecm.gforge.inria.fr/

and click on "released files" item.

I think gmp-ecm 6.1 will come soon (beginning of april?)...
With gmp 4.2 if will roxxx !

Hope this helps.
Bye.

[Phil MjX]

Originally Posted by BotXXX

I won't distribute the release candidate, but when GMP 4.2 will be final i can release a P4 / MinGW build executable. Have no idea how to compile gmp and ecm for other platforms from mingw.

the release candidate indeed reports itself as 4.2 but i changed it to 4.1.99 since that was the version number from the tarball ;)

PS : I'd also distribute my pentium-m version and, if I have access to my remote pc, an athlon version.

Bye.
Philippe.

[Mystwalker]

Thanks, Philippe!

Just saw that there is another version on the FTP server (not linked to from the homepage): gmp-4.1.999

So let's hope that the final release will be there soon.

[Andi47]

Originally Posted by Phil MjX

just try gmp-ecm 6.1beta2

Where can I download this?