Mprime on Solaris for ECM?

[D. B. Staple]

Hello,

I am wondering if compiling mprime for Solaris is feasible. On mersenne.org I find the discouraging comment: "The program is also completely non-portable."

The only references I can find to this are along the tune of:
http://www.mersenneforum.org/showthr...hlight=solaris
which suggests using mlucas/glucas instead.

Since I want to do ECM and not LL testing these programs are not really an option, and GMP-ECM is way slower than mprime for the numbers I am running.

Thanks,

Doug

[rogue]

Originally Posted by D. B. Staple

Hello,

I am wondering if compiling mprime for Solaris is feasible. On mersenne.org I find the discouraging comment: "The program is also completely non-portable."

The only references I can find to this are along the tune of:
http://www.mersenneforum.org/showthr...hlight=solaris
which suggests using mlucas/glucas instead.

Since I want to do ECM and not LL testing these programs are not really an option, and GMP-ECM is way slower than mprime for the numbers I am running.

Thanks,

Doug

I assume you are referring to a Sparc CPU, not Solaris on x86. The Sparc CPU has a completely different architecture than x86. prime95 is optimized by using thousands (if not tens of thousands) of lines of x86 assembly. This means that it cannot be ported to run on Sparc. mlucas and glucas are written entirely in C (except for some small pieces of code, such as prefetches that use assembly) and can easily be ported to most architectures.

[D. B. Staple]

I am in fact referring to Solaris on x86 Opteron CPUs; I understand that changing architecture is not feasible or necessarily even desirable.

I would expect that the linux version of mprime would be close to being able to compile for x86 Solaris, but in practice this may not be the case.

[rogue]

Originally Posted by D. B. Staple

I am in fact referring to Solaris on x86 Opteron CPUs; I understand that changing architecture is not feasible or necessarily even desirable.

I would expect that the linux version of mprime would be close to being able to compile for x86 Solaris, but in practice this may not be the case.

I would expect a port of mprime to be relatively easy, although you might need to use the GNU gcc compiler instead of the Solaris CC compiler. You also might need a tool to convert the binaries to the correct format. I used objconv to port LLR to MacIntel.

[sylvester]

I have a Solaris on Intel version running since February 2005. I took the Linux source/binary tarball, recompiled the C source and linked with the Linux .o files. Works just fine with two caveats:
1) measured CPU speed was off, overrode it with settings in local.ini
2) no security module is available, but PrimeNet seems to happily accept the submited double-checks and credits my account.
It passed all self-tests and is running fine on Athlon XP, Solaris 9u7, recompiled with Sun's Workshop 6u2 (or maybe 8cc) compilers.

Have fun.

[D. B. Staple]

I tried briefly to compile the source and failed, and can't invest too much time in making it work... Out of curiosity, did you manage to use the provided makefile to compile the source, or did you simply figure out what the relevant files and compiler flags were and then go for it? It's possible that I don't know what I'm doing; it would be nice if I am missing something simple and could get this to work.

In the meantime, could you send me your binary?

[sylvester]

I put it in the attachment to this message. I had to rename it to .pdf from .zip. Download the attachment and rename it back to .zip. The forum software has different size limits for various file types.

I don't remember what exactly did I do to make this work. Looking at the files in the above .zip I certainly changed the makefile and recompiled two source files prime.c* and menu.c*. I do remember that most of the Linux *.o files simply linked in correctly to create the Solaris executable. I don't think I had to make any changes to the source code. If I did change something, the changes were trivial. Otherwise I would have saved the changed files in the .zip.

The resultant executable passed all 37 self-tests on the Athlon XP and dual-Opteron machines. Sometime in 2007 I had to do emergency install of Linux on the dual-Opteron machine and I no longer have the Solaris build environment available. The Athlon XP machine is a pure NFS file server and will stay that way.

The attached binary doesn't have the proper security hash function. I don't think it is useable for a first-time LL test. But it seems that the PrimeNet happily accepts the double-check LL tests from this executable.

I don't know if the above is by design or by accident. Could somebody who works closely with George ask him if he is OK with submitting double-check from a non-blessed software? I do know that George had asked Guillermo Ballester Valor to remove the PrimeNet connectivity from his (non-blessed) software. I really don't want to offend anyone or contribute to corrupting the integrity of the whole project.

Thanks and have fun.

[D. B. Staple]

Thanks,

I tried your binary on my system and it worked fine. I'll have to try recreating what you have done with a new version some time I get a chance.

For the record, I remember reading somewhere that it was intentional that a copy of mprime compiled from scratch would not be able to communicate with primenet, but I can't find the reference for it. This is ok for me, because I am using mprime to do ECM (which does not use primenet).

Thanks again,

Doug