I think J has it right. Infinity is not qualitatively different from any other number, except in the sense that it's too tedious to be worth dealing with.
So if BLAS treats it differently, my idea of the correct solution would be to modify BLAS. As for distributing it, I'd probably document my changes, making sure that it compiles as expected on various machines, and then advertise the github location. I do not anticipate any licensing issues, and if someone at ISI felt the changes were worthwhile they do the official release thing. -- Raul On Tue, May 7, 2013 at 4:28 PM, Henry Rich <[email protected]> wrote: > You would also have agonize over the fact that in J, > > 0 * _ > 0 > > which is probably not what BLAS does. > > And anyway, if you did modify cip.c, how would you distribute the fruits of > your labor? > > Henry Rich > > > On 5/7/2013 4:10 PM, Raul Miller wrote: >> >> On Tue, May 7, 2013 at 3:43 PM, Konrad Hinsen >> <[email protected]> wrote: >>> >>> BLAS is very hard to beat, so the practical question is: could one make J >>> use BLAS for matrix multiplication (automatically, from standard J verbs) >>> and other operations where BLAS offers support? >> >> >> You would modify cip.c (conjunctions: inner product). >> >> Presumably you would be interested in the cases where both matrices >> were rank 2 (the AR() macro) and 3!:0 is 8 or 16 (case FL: and case >> CMPX: of the AT() macro). >> > ---------------------------------------------------------------------- > For information about J forums see http://www.jsoftware.com/forums.htm ---------------------------------------------------------------------- For information about J forums see http://www.jsoftware.com/forums.htm
