Salut Sylvestre! Thanks for the superspeedy follow-up!
On 26 November 2010 at 18:00, Sylvestre Ledru wrote: | Le vendredi 26 novembre 2010 à 10:53 -0600, Dirk Eddelbuettel a écrit : | > Hi science folks, | > | > Ei-ji, who follows this closely and who also wrote a the very nice | > gotoblas2-helper package which turns the (until now non-free) gotoblas2 into | > a rocking .deb package, just posted this to the R list for HPC: | > | > On 27 November 2010 at 01:32, Ei-ji Nakama wrote: | > | Hi, | > | | > | Please look at the following. | > | http://www.tacc.utexas.edu/tacc-projects/gotoblas2/ | > | | > | comparatively new CPU should force it with `make TARGET=NEHALEM' | > | > Wouldn't we want GotoBLAS2 in Debian now that it is BSD-licensed? I checked | > the link and there is in fact a new download link -- no more registration at | > TACC and all that. | > | > Sylvestre, any interest and equally importantly, spare cycles on your part? | That is excellent news !! Especially since, thanks to your document, I | am aware of the great performances of GotoBLAS2. | Especially since we have now in Debian a quick and easy way to switch | between BLAS implementations. | | I am ready to upload this in Debian if it is OK with you. That would be wonderful. I'd be happy test source packages etc pp | Sylvestre | PS: Can you confirm that all the performances optimisation are done at | runtime (like the MKL) and not build time (like ATLAS) ? Not sure I fully understand the question. A lot of _tuning_ happens at _build time_ just like Atlas. That is where the tuning comes from. What Goto BLAS shares with MKL (and does better than Atlas) is the ability to _select the number of cores at run-time_. Is that what you meant? All 'hard' questions I suggest you direct in a friendly email to Ei-ji who knows so much more about all this than I do... Cheers, Dirk -- Dirk Eddelbuettel | e...@debian.org | http://dirk.eddelbuettel.com -- To UNSUBSCRIBE, email to debian-science-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/19695.60017.105032.21...@max.nulle.part