On 05/31/10 05:52 AM, William Stein wrote:
I've never had this problem building on 32-bit SPARC, or 64-bit OpenSolaris x64. I only see it on 64-bit Solaris 10 SPARC. Perhaps the behavior is undefined if the dependency is missing, and it just happens by luck to work on the other systems. Either that, or I've mis-understood it.I think you've found a bug in the makefile. We have LINBOX depends on ATLAS ATLAS depends on LAPACK and PYTHON LAPACK depends on FORTRAN It is just a *lucky coincidence* that the Sage build has actually worked. We need to add BLAS as a dependency of LAPACK, e.g,. $(INST)/$(LAPACK): $(INST)/$(FORTRAN) $(INST)/$(BLAS) $(SAGE_SPKG) $(INST)/$(LAPACK) 2>&1 Then everything should work fine, since depence is transitive. William
That may be so, (Francois seems to have another opinion though). But it is certainly not the only issue on Solaris. If, as Francois says, Linbox prefers ATLAS, then I think I know why Linbox is not finding ATLAS on Solaris 10 with a SPARC processor in 64-bit mode.
1) On Solaris 10 SPARC, *without* SAGE64 being set to "yes" (the normal build-mode on 't2'), ATLAS builds a 32-bit library, as one would expect.
2) On OpenSolaris x64, with SAGE64 set to "yes" (what I believe will be the normal build mode for OpenSolaris x64), ATLAS builds 64-bit as one would want.
3) However, on Solaris 10 SPARC, with SAGE64 set to "yes", ATLAS is building 32-bit, despite most other packages building 64-bit on that system. I expect linbox reports the message as it can't find a suitable 64-bit BLAS library, only an unsuitable 32-bit one.
'cblas' is also building 32-bit when it should be building 64-bit, so Linbox will not find a suitable 'cblas' either.
So MartinX is not the only one with ATLAS problems with his i5-750 processor - I seem to have found a problem on the SPARC processor if one tries to build 64-bit code.
It might be different on 't2' (I've not tried), since ATLAS of the T2+ processors on 't2.math', whereas ATLAS is aware of the older UltraSPARC III+ processors I have. ATLAS *might* build differently on two otherwise similar SPARC systems. It certianly spends an eternity building on 't2', whereas it is nothing like as slow on my SPARC system.
Dave -- To post to this group, send an email to [email protected] To unsubscribe from this group, send an email to [email protected] For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
