On Sunday 04 January 2009 00:57:44 ja...@njkfrudils.plus.com wrote: > On Sunday 04 January 2009 00:36:46 Jason Martin wrote: > > Alternatively, we could stop trying to identify chips by marketing > > brands and just use the values returned by CPUID. This would create a > > lot of duplicated code in sub-directories, but disk space is cheap. > > So, would something like: > > > > mpn/x86_64/<vendor>/<extended family number/model> > > > > work for our configuration? > > As most of the models are the same , this seems like a waste. > Also this assumes that CPUID is the only differentiator , what about > L2-cachesize (in the future?) , GPU coprocessors > > How about , for each asm file a description of minimum requirements > eg > add_n.asm requires x86_64,LAHF > lshift.adm requires x86_64,SSE4.2 > hamdist.asm requires x86_64,popcnt > > and we only bother with the differences that we use , ie virtualization > instructions we dont bother with. This doesn't help with selecting among > functions that run at different speeds. > > I think what happens at the moment is nearly the best.
Whoops , didn't finish .... When we get a function which splits an existing type into two(or more) subtypes then we duplicate the existing functions between the subtypes , and put the new function_1 into subtype1 and new function_2 into subtype2 > > > Jason Worth Martin > > Asst. Professor of Mathematics > > http://www.math.jmu.edu/~martin > > > > On Sat, Jan 3, 2009 at 5:49 PM, Bill Hart <goodwillh...@googlemail.com> > > wrote: > > > I think that features such as SSE should be tested for after testing > > > for the main chip core. So under /mpn/x86_64/k8 you'd have directories > > > for any features not available on all k8's. > > > > > > Bill. > > > > > > 2009/1/3 mabshoff <michael.absh...@mathematik.uni-dortmund.de>: > > >> On Jan 3, 2:25 pm, jason <ja...@njkfrudils.plus.com> wrote: > > >>> On Jan 3, 9:00 am, "Bill Hart" <goodwillh...@googlemail.com> wrote: > > >> > > >> Hi, > > >> > > >>> > The new intel machines. And I don't know if all Dunnington's use > > >>> > the same family/system CPUID etc. So there might be mutiple CPUID's > > >>> > we need to add to config.guess. > > >>> > > > >>> > Bill. > > >>> > > >>> We should change the lowest common denominator on a x86_64 system to > > >>> something more useful than 486 , say P4 64bit without LAHF ? , then > > >>> people can at least get mpir working on new machines without mucking > > >>> about > > >> > > >> Well, the trouble was that configure believed it was a 32 bit system, > > >> so I don't see much we can do there aside from attempting to compile > > >> things in 64 bit mode. > > >> > > >>> For the K10 , we will need a separate directory for it , I have > > >>> mpn_popcount and mpn_hamdist which will not run on the K8 , requires > > >>> SSE4.1a or whatever it's called ... > > >>> before 7/7.75 c/l now 1.5/1.75 c/l > > >> > > >> Wouldn't it be better to create a SSE4.1a directory and use that > > >> assembly code when SSE 4.1a is available? That seems to be the > > >> prevailing way to do things. > > >> > > >> On second though: according to http://en.wikipedia.org/wiki/SSE4 it > > >> seems that there are three SSE4 flavors: > > >> > > >> * SSE 4.1 > > >> * SSE 4.2 > > >> * SSE 4.1a > > >> > > >> The last one seems to be K10 specific for now, but I would still > > >> recommend to test for SSE 4.1a if your code is that specific. > > >> > > >> <SNIP> > > >> > > >> Cheers, > > >> > > >> Michael > > --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "mpir-devel" group. To post to this group, send email to mpir-devel@googlegroups.com To unsubscribe from this group, send email to mpir-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/mpir-devel?hl=en -~----------~----~----~----~------~----~------~--~---