Links to Precompiled GMP-ECM versions - Page 37

ATH · 2015-12-26, 19:51

I got a patch for GMP 6.1.0 on Broadwell:

Broadwell; GMP-ECM7 SVN 2749 with GMP 6.1.0: gmpecm7-svn2749-broadwell.zip (compiled on Core i5 5200U laptop)

ATH · 2016-01-01, 20:21

If anyone is still using Core2 for GMP-ECM:

Core2; GMP-ECM7 SVN 2749 with GMP 6.1.0: gmpecm7-svn2749-core2.zip (compiled on a Core2Quad (Yorksfield) Q9450)

wblipp · 2016-01-20, 16:30

How do I figure out which version to use on a machine? For example, I have occasional access to a machine running Win 7. Control Panel>System says the processor is a Xeon E3-1246 v3

pinhodecarlos · 2016-01-20, 16:47

Quote:

Originally Posted by wblipp

How do I figure out which version to use on a machine? For example, I have occasional access to a machine running Win 7. Control Panel>System says the processor is a Xeon E3-1246 v3

William, use the Haswell one.

ATH · 2016-01-20, 18:04

Carlos is correct, if you look at this list and search for 1246 you will find it under the "Haswell based Xeons"-section:
https://en.wikipedia.org/wiki/List_o...icroprocessors

Belteshazzar · 2016-01-22, 06:12

ATH, any chance you'd be willing to compile one optimized for AMD?

Piledriver (i.e. march/mtune = bdver2) would be the most appropriate target. Though Steamroller and Excavator are newer, Excavator is laptops only, Steamroller (desktop Kaveri APUs) is not much different from Piledriver (instruction wise just adds FSGSBASE), and the main FX line (incl. inexpensive 8-core chips) is still stuck on Piledriver.

I'm a little confused as to why you're using each kind of processor to compile its own binary- is that for testing purposes or is there some reason you're tied to -march=native?

ATH · 2016-01-23, 21:52

I always assumed it was better to compile on the correct hardware instead of cross compiling, and in this case with GMP and GMPECM it seems I was right.

GMP does a lot of checks of the different types of variables which are available on your computer and compiler, so what won't work for cross compiling, and GMPECM has 2 tuning steps for tuning the parameters.

I tried GMP with --build=haswell-w64-mingw32 and --host=piledriver-w64-mingw32 but that failed. It also failed with just --build=piledriver-w64-mingw32. Then I tried using my GMP version and just compiling GMPECM, but that failed as well.

I did consider making a new zip file with MSYS2 and instructions on how to compile both GMP and GMPECM, and if you do not mind, you can try and "beta" test it. I'll send you a pm.

WraithX · 2016-02-27, 20:07

Quote:

Originally Posted by bsquared

Quote:

Originally Posted by jasonp

Unfortunately the GMP-ECM library is not technically multithread-safe, I think it uses a sieve of eratosthenes that has some global state. You could actually do the community a great service if you could fix that and Msieve does have a GMP-ECM driver. Alternately, YAFU can call multiple copies of the GMP-ECM demo binary.

Yep :) The unsafeness of the library is why yafu only allows multithreading with the external binary.

As an FYI to those who may be interested, recent svn changes to GMP-ECM now include a thread safe prime sieve. You can see this mentioned in the NEWS file in svn:

Quote:

Changes between GMP-ECM 6.4.4 and GMP-ECM 7.0:
* GMP-ECM is now thread-safe. In particular the "ecmfactor" binary can be
called with say -t 17 to use 17 threads.

ATH · 2016-02-27, 23:24

Haswell: GMP-ECM7 SVN 2840 with GMP 6.1.0: gmpecm7-svn2840-haswell.zip

Sandy Bridge: GMPECM7 SVN 2840 with GMP 6.1.0: gmpecm7-svn2840-sandybridge.zip

I still had to comment out line 191 and 203 in "bench_mulredc.c", same error as described here in post #50+#52: http://www.mersenneforum.org/showthread.php?t=16444

Also had to comment out line #880+#881 in main.c:

Code:

/*      if (verbose >= 2)
        printf ("Tuned for %s\n", ECM_TUNE_CASE); */

as ECM_TUNE_CASE was not defined. I think is not meant to use that variable in those line since there is a check if it is defined in line #166-#170:

Code:

#ifdef ECM_TUNE_CASE
  printf ("Tuning parameters from %s\n", ECM_TUNE_CASE);
#else
  printf ("ECM_TUNE_CASE not defined.\n");
#endif

WraithX · 2016-02-29, 02:06

Quote:

Originally Posted by ATH

I still had to comment out line 191 and 203 in "bench_mulredc.c", same error as described here in post #50+#52: http://www.mersenneforum.org/showthread.php?t=16444

Also had to comment out line #880+#881 in main.c:

Thanks for mentioning this again. This is now fixed in svn 2841.

Gordon · 2016-02-29, 20:22

Quote:

Originally Posted by ATH

Haswell: GMP-ECM7 SVN 2840 with GMP 6.1.0: gmpecm7-svn2840-haswell.zip

ecm doesn't have the -t parameter and when I run ecmfactor -h it spits out this error

The program can't start because libwinpthread-1.dll is missing....

2016-01-23, 21:52	#403
ATH Einyen Dec 2003 Denmark 2²×863 Posts	I always assumed it was better to compile on the correct hardware instead of cross compiling, and in this case with GMP and GMPECM it seems I was right. GMP does a lot of checks of the different types of variables which are available on your computer and compiler, so what won't work for cross compiling, and GMPECM has 2 tuning steps for tuning the parameters. I tried GMP with --build=haswell-w64-mingw32 and --host=piledriver-w64-mingw32 but that failed. It also failed with just --build=piledriver-w64-mingw32. Then I tried using my GMP version and just compiling GMPECM, but that failed as well. I did consider making a new zip file with MSYS2 and instructions on how to compile both GMP and GMPECM, and if you do not mind, you can try and "beta" test it. I'll send you a pm. Last fiddled with by ATH on 2016-01-23 at 21:52

2016-02-27, 23:24	#405
ATH Einyen Dec 2003 Denmark 2²·863 Posts	Haswell: GMP-ECM7 SVN 2840 with GMP 6.1.0: gmpecm7-svn2840-haswell.zip Sandy Bridge: GMPECM7 SVN 2840 with GMP 6.1.0: gmpecm7-svn2840-sandybridge.zip I still had to comment out line 191 and 203 in "bench_mulredc.c", same error as described here in post #50+#52: http://www.mersenneforum.org/showthread.php?t=16444 Also had to comment out line #880+#881 in main.c: Code: /* if (verbose >= 2) printf ("Tuned for %s\n", ECM_TUNE_CASE); / as ECM_TUNE_CASE was not defined. I think is not meant to use that variable in those line since there is a check if it is defined in line #166-#170: Code: #ifdef ECM_TUNE_CASE printf ("Tuning parameters from %s\n", ECM_TUNE_CASE); #else printf ("ECM_TUNE_CASE not defined.\n"); #endif Last fiddled with by ATH on 2016-02-27 at 23:24*

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Project Links	masser	Sierpinski/Riesel Base 5	25	2011-11-26 09:21
Links to Precompiled Msieve versions	wblipp	Msieve	0	2011-07-17 20:59
Links	davieddy	Information & Answers	9	2010-10-08 14:27
Links question	ET_	PrimeNet	0	2008-01-26 09:35
Links.	Xyzzy	Forum Feedback	2	2007-03-18 02:17

2015-12-26, 19:51	#397
ATH Einyen Dec 2003 Denmark 6574₈ Posts	I got a patch for GMP 6.1.0 on Broadwell: Broadwell; GMP-ECM7 SVN 2749 with GMP 6.1.0: gmpecm7-svn2749-broadwell.zip (compiled on Core i5 5200U laptop)

2016-01-01, 20:21	#398
ATH Einyen Dec 2003 Denmark 2²×863 Posts	If anyone is still using Core2 for GMP-ECM: Core2; GMP-ECM7 SVN 2749 with GMP 6.1.0: gmpecm7-svn2749-core2.zip (compiled on a Core2Quad (Yorksfield) Q9450)

2016-01-20, 16:30	#399
wblipp "William" May 2003 Near Grandkid 947₁₆ Posts	How do I figure out which version to use on a machine? For example, I have occasional access to a machine running Win 7. Control Panel>System says the processor is a Xeon E3-1246 v3

2016-01-20, 18:04	#401
ATH Einyen Dec 2003 Denmark 2²×863 Posts	Carlos is correct, if you look at this list and search for 1246 you will find it under the "Haswell based Xeons"-section: https://en.wikipedia.org/wiki/List_o...icroprocessors

2016-01-22, 06:12	#402
Belteshazzar Feb 2011 3×11 Posts	ATH, any chance you'd be willing to compile one optimized for AMD? Piledriver (i.e. march/mtune = bdver2) would be the most appropriate target. Though Steamroller and Excavator are newer, Excavator is laptops only, Steamroller (desktop Kaveri APUs) is not much different from Piledriver (instruction wise just adds FSGSBASE), and the main FX line (incl. inexpensive 8-core chips) is still stuck on Piledriver. I'm a little confused as to why you're using each kind of processor to compile its own binary- is that for testing purposes or is there some reason you're tied to -march=native?