On Tuesday 25 January 2011 21:23:21 jason wrote: > On Dec 5 2010, 11:01 am, Jason <ja...@njkfrudils.plus.com> wrote: > > On Sunday 05 December 2010 07:50:44 Jason wrote: > > > On Sunday 05 December 2010 07:05:40 Jason wrote: > > > > I've added improved versions for > > > > > > > > K8 mpn_nand_n > > > > K8 mpn_nior_n > > > > > > > > These run at the optimal 1.5c/l for ALL alignments , whereas the old > > > > functions run at 1.7c/l to 2.0c/l for some alignments > > > > > > > > So now all the K8 logic functions run at 1.5c/l for ALL alignments. > > > > > > > > Jason > > > > > > I've copyied the new K8 hamdist into the core2 directory as this > > > improves the speed form 6.5c/l to 5.9c/l > > > I did both the linux and windows versions > > > > > > Jason > > > > A new mpn_store for K8/K10/K102 the old one ran like > > > > box12 K8 model 44 .75c/l > > lena K102 model4 0.5c/l > > redhawk K102 model 8 0.75c/l > > flavius K8 mode 5 0.75c/l > > > > and the new one > > > > box12 K8 model 44 .55c/l > > lena K102 model4 0.5c/l > > redhawk K102 model 8 0.5c/l > > flavius K8 mode 5 0.55c/l > > > > You can see from the above that the K102 has at least two variants, looks > > like they are continually tweaking the design(or fixing it) > > > > Jason > > It looks like AMD has just(ish) updated their docs , and we are > missing some models AND familys (besides the obvious bobcat and > bulldozer). There is some interesting stuff there , including the > differences between the revisions , new family 12h with integer > hardware fastpath divider ,family 11h,14h , we currently dont > recognize any of these , so I think we can release a MPIR_2.2.2 when > I've sorted through the variants and I'll check if Intel has updated > theirs , and VIA . > > Jason
For AMD I've added Family 10h models 5,6,10 as K102 Family 12h as K102 , we dont have any specific code for the new hardware divider Family 11h as K8 , it's a fusion of K8 core with GPU Family 14h as K8 , it's a fusion of bobcat core with GPU to mpir-2.2 and trunk , and I've also added sandybridge as westmere to mpir-2.2 Now to see what Intel and VIA have documented. The bobcat core has the ABM unit like the K10 (ie popcount and LZCNT) but the floating point unit is only 64bit like the K8 , it's also missing 3DNOW , and the pipeline/scheduler is completely different. Family 15h appears to be the Bulldozer core , but it's not been documented yet , and wont be out for a while yet. Jason -- 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.
