Dear All: Thanks to Brian Beesley <[EMAIL PROTECTED]> and Francois Jaccard, I have Alpha/Linux binaries of Mlucas 2.7y available: see ftp://209.133.33.182/pub/ mayer/README for details. Brian reports that the binary should run under both Linux V5 and V6. He hasn't yet sent me timings for non-power-of-2 runlengths, but I expect these to be similar trends as the timings I recently reported for Alpha/Unix, meaning a fairly smooth timing progression between any given powers of 2. Here are some power-of-two length timings for Mlucas and MacLucasUNIX on Brian's Alpha PC164 (basically an Alpha 21264A, a.k.a. ev56, with a PC-style configuration): ***ALPHA/LINUX TIMINGS*** Program, platform, cache sizes / per-iteration time in seconds Prime95.19 Mlucas2.7y MacLucasUNIX 6.25 Intel Alpha Alpha PII/400 PC164/533 PC164/533 8KB L1 8KB L1 8KB L1 512KB L2 96KB L2 96KB L2 2MB L3 2MB L3 length ---------- ---------- ---------- 128K .060 .057 (79%) .064 (56%) 256K .132 .147 (67%) .156 (59%) 512K .281 .314 (67%) .320 (61%) 1024K .600 .657 (68%) not run 2048K 1.23 1.56 (59%) not run 4096K 2.60 3.19 (61%) not run Thanks to Bill Rea <[EMAIL PROTECTED]> I also have a binary for SPARC Solaris V7; Bill assures me that one for Solaris 2.6 is coming soon. The above README file says where to get the binary and how to install the f90 RTL files needed to run it. ***SPARC TIMINGS*** Notes: (1) I include timings for the ftp'able Mlucas 2.7y binary compiled without the -xprefetch flag and for an executable compiled with it. As you can see, the -xprefetch flag gives extremely unpredictable results, which are usually worse than default, but at some runlengths it gives a small speedup (indicated with a *) or even substantial speedup (**). If you have a lot of work to do at one of the * or ** runlengths, compile yourself using -prefetch (make sure to compare 100-iteration timings of the resulting binary with the default binary on your machine before starting a long run!) or contact Bill about getting a binary compile with -xprefetch. (2) Times for more-advanced Ultra systems (Ultra2i, E450) tend to be much better than the ones below, even at similar clock rates. For example, Alex Kruppa gets .183 sec/iterations for Mlucas at 256K on his 300 MHz Ultra 2i, and Bill Rea says he gets 0.43 and 0.29 sec/iteration, respectively, for MacLucasUNIX at 512K on his 300MHz/2MBL2 and 400MHz/4MBL2 E450. Abbrevations: n/a = Length not available; must use next-higher power of 2. XXX = test not done due to insufficient host virtual memory. Prime95.19 Mlucas2.7y Mlucas2.7y MacLucasUNIX 6.25 noprefetch -xprefetch Intel SPARC SPARC SPARC PII/400 Ultra5/360 Ultra5/360 Ultra5/360 8KB L1 ? L1 ? L1 ? L1 512KB L2 256KB L2 256KB L2 256KB L2 length ---------- ---------- ---------- ---------- 96K .045 .091 (55%) .158 n/a 112K .055 .130 (47%) .101** n/a 128K .060 .123 (54%) .118* .12 (56%) 160K .083 .184 (50%) .175* n/a 192K .098 .192 (57%) .336 n/a 224K .119 .225 (59%) .228 n/a 256K .132 .260 (56%) .340 .25 (59%) 320K .173 .387 (50%) .369* n/a 384K .211 .410 (57%) .680 n/a 448K .252 .475 (59%) .475 n/a 512K .281 .542 (58%) .710 .51 (61%) 640K .372 .797 (52%) .705** n/a 768K .453 .865 (58%) 1.41 n/a 896K .536 1.00 (60%) .995* n/a 1024K .600 1.10 (61%) 1.44 1.08 (62%) 1280K .776 1.42 (61%) 1.78 n/a 1536K .934 1.83 (57%) 3.01 n/a 1792K 1.11 2.18 (57%) 2.01** n/a 2048K 1.23 2.52 (54%) 2.46* 2.82 (48%) 2560K 1.64 3.55 (51%) 3.22** n/a 3072K 1.99 3.80 (58%) 5.58 n/a 3584K 2.38 4.48 (59%) 4.49 n/a 4096K 2.60 5.12 (56%) 6.44 XXX _________________________________________________________________ Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
