Thomas Ritschel has been kind enough to build binaries of Mlucas 2.7c for me. Now I need some guinea pigs, er, I mean, beta testers. Thomas has built both 32-bit and 64-bit executables and verified that they run and done timings on the R10000 and R12000 systems he has access to. These are all fairly recent CPUs (i.e. they can handle 64-bit binaries) with large L2 caches. Now I need to try the code out on some older and smaller-cache systems. If you have access to an SGI which is either pre-R10000 or an R10000 with an L2 cache of 1MB or smaller and you are willing to do some tests for me, the tarball is at
ftp://hogranch.com/pub/mayer/bin/SGI/Mlucas_SGI.tar.gz This contains a 32-bit and a 64-bit executable, and several .cfg files with the optimal radix combinations Thomas found on his systems. In order to run the same suite of timing tests so you can create the .cfg file optimal for your system, you'll need to also grab the source tarball: ftp://hogranch.com/pub/mayer/Mlucas_2.7c.tar.gz This tarball contains a timings.txt file, which has a complete set of entry fileds for doing benchmarks. If your system is pre-R10000, you only need do runs up to an FFT length of 1024K, unless you're curious and willing to burn the CPU time that will be needed to do the larger runlengths. I'll need answers to the following questions: 1) What type of system did you do benchmarks on? ('hinv' will give CPU and cache info; 'uname -a' should tell you which OS.) 2) Did both the 32-bit and 64-bit binaries run on your system? If yes, which is faster? 3) For the faster of the 2 binaries, a .cfg file with the optimal radix set index for each FFT length, and the table of per-iteration timings for all radix sets you tried for each FFT length. Please send these data in simple plaintext mode. Thanks in advance, -Ernst _________________________________________________________________________ Unsubscribe & list info -- http://www.ndatech.com/mersenne/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
