John Cremona wrote: > Here's what I did when Atlas failed to recognise my machine: > > 1. Copy the SAGE_ROOT/local directory to (e.g.) /home/john/sagelocal . > 2. Add > > export SAGE_ATLAS_LIB="/home/john/sagelocal" > > to my startup file .bashrc. > > Then in all subsequent builds it finds the existing Atlas and does not > build a new one. > > Of course sagelocal now contains a lot of other stuff too, which you can > delete. > > I think this is exactly what worked, but I stopped needing to use it > when Atlas fixed how it recognised machines (or maybe Michael Abshoff > fixed it, I forget). > > John > > 2009/7/11 Dr. David Kirkby <david.kir...@onetel.net>: >> It's taken more than 12 hours to build ATLAS on t2, and it is still >> going. William said it takes a lot less if some tuning parameters are >> known for the machine, so it does not have to go through that again. I >> asked on the ATLAS help forum, but so far no reply for how to save these >> parameters. >> >> So a few question. >> >> 1) Does anyone know how to save the parameters for a machine, or just >> force ATLAS to use a semi of semi-sensible defaults. >> >> 2) Does anyone know how I can save the binaries and restore them another >> time, rather than rebuild it. Sage releases are comping out quicker than >> I can build it on Sage, due mainly to the very long time to compile >> ATLAS. It takes nowhere near as long on my own SPARC, so I believe it's >> probably since mine is a lot more common than 't2', so perhaps there is >> data on my machine, but not t2. >> >> Any other suggestion - apart from buy a turbo charger? >> >> It would be good if some of the other packages could be built in >> parallel. There is no reason one could not build the next package, >> assuming that does not depend on ATLAS. >> >> Dave
Thanks for that. I have created the tuning data on 't2' but it's not clear how to use it. I think it needs some code written to test for this type of machine, as the architecture is defined as 'UNKNOWN32' on t2, whereas it's defined as UltraSPARC-III or similar on my own machine. (Not that my machine is using the tuning defaults either, but the single threaded performance of my Blade is a lot quicker than t2, so it's less of a pain). I looked at the code for testing different machines on Solaris inside ATLAS. It does not look too sophisticated and appears to make a lot of assumptions which may be inaccurate. I'll try to find out on the solaris newsgroups how to get the type of processor properly on Solaris. Parsing text output of commands is not the best way, if system calls exist to tell one the information. dave Dave --~--~---------~--~----~------------~-------~--~----~ To post to this group, send email to sage-devel@googlegroups.com To unsubscribe from this group, send email to sage-devel-unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URLs: http://www.sagemath.org -~----------~----~----~----~------~----~------~--~---
