The discussion is moot.  The speed-up factor is a
maximum of 2, and you'd have to strain to achieve it:

The logic in inner product is +/@(*"1 _) . This has 
nice cache effects regarding both read and write.  
Which is why the release note 
http://www.jsoftware.com/books/help/release/pdt504.htm
said that the speed-up is larger on CPUs with larger
on-die cache (i.e. larger on newer CPUs).  Moreover,
+/@(*"1 _) allows you to exploit 0s in the left 
argument.

+/@(*"1 _) can not be used in the obvious 
exploitations of symmetry.  So you'd probably lose
both of the above benefits.  So the speed-up factor,
already meagre even at 2, is probably something less
than 2.


----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm

Reply via email to