Re: [sage-devel] Re: sage -bdist and SAGE_ATLAS_LIB

2012-12-21 Thread Julien Puydt
Le 21/12/2012 15:25, Volker Braun a écrit : Spkg is here if you want to try it: http://trac.sagemath.org/10508 I'm now trying with 5.5.rc1 with atlas+cvxopt replaced by those in the ticket : it looks like using SAGE_ATLAS_LIB really doesn't give a good bdist. Snark on #sagemath -- You rece

[sage-devel] Re: sage -bdist and SAGE_ATLAS_LIB

2012-12-21 Thread P Purkayastha
On 12/21/2012 09:53 PM, Jeroen Demeyer wrote: On 2012-12-21 14:48, Julien Puydt wrote: Something like "export SAGE_ATLAS_DONT_OVERDO_IT=yes" would be wonderful. I think even more stupid is that ATLAS fails to build if it cannot perfectly tune itself. Yes. I have had atlas-3.8.4 build fail mul

Re: [sage-devel] Re: sage -bdist and SAGE_ATLAS_LIB

2012-12-21 Thread Volker Braun
Spkg is here if you want to try it: http://trac.sagemath.org/10508 On Friday, December 21, 2012 2:24:15 PM UTC, Volker Braun wrote: > > Atlas 3.10 adds ARMv7 as machine type. Though it seems we'll never merge > it because it doesn't work on Itanium and the Atlas dev doesn't have enough > time t

Re: [sage-devel] Re: sage -bdist and SAGE_ATLAS_LIB

2012-12-21 Thread Volker Braun
Atlas 3.10 adds ARMv7 as machine type. Though it seems we'll never merge it because it doesn't work on Itanium and the Atlas dev doesn't have enough time to waste on dead architectures. On Friday, December 21, 2012 2:04:15 PM UTC, Snark wrote: > > Le 21/12/2012 14:54, Jean-Pierre Flori a �cr

Re: [sage-devel] Re: sage -bdist and SAGE_ATLAS_LIB

2012-12-21 Thread Julien Puydt
Le 21/12/2012 14:54, Jean-Pierre Flori a écrit : Can't you use the SAGE_ATLAS_ARCH and set it to generic or something like that? Ah, I didn't know about it ; the choice seems to be one of: ATLAS_MACHTYPE = ['UNKNOWN', 'POWER3', 'POWER4', 'POWER5', 'PPCG4', 'PPCG5', 'P5', 'P5M

Re: [sage-devel] Re: sage -bdist and SAGE_ATLAS_LIB

2012-12-21 Thread Jean-Pierre Flori
On Friday, December 21, 2012 2:48:07 PM UTC+1, Snark wrote: > > Le 21/12/2012 14:14, Volker Braun a �crit : > > Setting SAGE_ATLAS_LIB creates a symlink to the underlying atlas > > install. Don't use it if you want to distribute the result. > > > Having said that, I'm thinking about a more f

Re: [sage-devel] Re: sage -bdist and SAGE_ATLAS_LIB

2012-12-21 Thread Jeroen Demeyer
On 2012-12-21 14:48, Julien Puydt wrote: > Le 21/12/2012 14:14, Volker Braun a écrit : >> Setting SAGE_ATLAS_LIB creates a symlink to the underlying atlas >> install. Don't use it if you want to distribute the result. > >> Having said that, I'm thinking about a more flexible system for >> deployin

Re: [sage-devel] Re: sage -bdist and SAGE_ATLAS_LIB

2012-12-21 Thread Julien Puydt
Le 21/12/2012 14:14, Volker Braun a écrit : Setting SAGE_ATLAS_LIB creates a symlink to the underlying atlas install. Don't use it if you want to distribute the result. Having said that, I'm thinking about a more flexible system for deploying and using blas/lapack. I'll probably post to sage-d

Re: [sage-devel] Re: sage -bdist and SAGE_ATLAS_LIB

2012-12-21 Thread Volker Braun
Setting SAGE_ATLAS_LIB creates a symlink to the underlying atlas install. Don't use it if you want to distribute the result. Having said that, I'm thinking about a more flexible system for deploying and using blas/lapack. I'll probably post to sage-devel about that soon. On Friday, December

Re: [sage-devel] Re: sage -bdist and SAGE_ATLAS_LIB

2012-12-21 Thread Julien Puydt
Le 21/12/2012 09:41, P Purkayastha a écrit : On 12/21/2012 03:55 PM, Julien Puydt wrote: Hi, building sage on a box using SAGE_ATLAS_LIB then running sage -bdist to get something to run on another box doesn't work: you'll have to install atlas on the target system too. Is it documented somewhe

[sage-devel] Re: sage -bdist and SAGE_ATLAS_LIB

2012-12-21 Thread P Purkayastha
On 12/21/2012 03:55 PM, Julien Puydt wrote: Hi, building sage on a box using SAGE_ATLAS_LIB then running sage -bdist to get something to run on another box doesn't work: you'll have to install atlas on the target system too. Is it documented somewhere, and if not, where should I document it? S