Hi, Let me clarify a few things: - LinBox's approach to link against numerical BLAS has never changed, and probably will not in the near future. - LAPACK is not BLAS: BLAS provides optimized numerical matrix multiplication kernels (that LinBox uses) and other basic routines, whereas LAPACK provides more advanced routines such as gaussian elimination, QR factorization, etc, designed to work on top of a BLAS. - LinBox links against LAPACK too (in the past and still now), but only for some variants of computations that are AFAIK, not hooked in the sage interface: e.g. computing first invariant of the Smith form.
I also recently remarked that many linalg computations over finite fields and ZZ were shamefully slow. There definitely is a misconfiguration somewhere, as we do not observe any such regression in LinBox. Instead, the library has gotten actually significantly faster. So in the short term: I'm definitely willing to look into improving the sage interface, and fix this misconfiguration. I'll have very little time before the end of september, though, but can start working on it in october. We're also planning on cutting a new release of Linbox this fall (october) an this could be a good time to update the linbox spkg. I also would like to mention that sage rely on Atlas BLAS which is no longer state of the art as far as computation speed is concerned. We consider switching to OpenBLAS, (formerely GOTOBLAS), that is now released under BSD. http://www.openblas.net/ The compile & configuration times are much shorter which would impact sage compile time. Best Clément Le 29/08/2014 10:26, William A Stein a écrit : > On Fri, Aug 29, 2014 at 10:24 AM, Jean-Pierre Flori <jpfl...@gmail.com> wrote: >> I don't know for sure, but I think linbox does not link to LAPACK >> (anymore?). >> Maybe that make a small difference already. > That's precisely what I'm worried about -- not linking lapack would > perhaps change linbox from being world class to absolutely terrible. > >> I opened a ticket about that some monthes ago. > Thanks! > >> -- >> You received this message because you are subscribed to the Google Groups >> "sage-devel" group. >> To unsubscribe from this group and stop receiving emails from it, send an >> email to sage-devel+unsubscr...@googlegroups.com. >> To post to this group, send email to sage-devel@googlegroups.com. >> Visit this group at http://groups.google.com/group/sage-devel. >> For more options, visit https://groups.google.com/d/optout. > > -- Clément Pernet -------------------------------------------------------------------- Lab. de l'Informatique Parallélisme, AriC office LUG 277 ENS de Lyon, CNRS tel: +33 (0)4 37 28 74 75 46 Allée d'Italie fax: +33 (0)4 72 72 80 80 69364 LYON Cedex 07- France http://membres-liglab.imag.fr/pernet http://pgp.mit.edu:11371/pks/lookup?op=get&search=0xAF8B61347AE8C23D -------------------------------------------------------------------- -- You received this message because you are subscribed to the Google Groups "sage-devel" group. To unsubscribe from this group and stop receiving emails from it, send an email to sage-devel+unsubscr...@googlegroups.com. To post to this group, send email to sage-devel@googlegroups.com. Visit this group at http://groups.google.com/group/sage-devel. For more options, visit https://groups.google.com/d/optout.
