From: minforth@gmx.net   
      
   Am 19.07.2025 um 16:51 schrieb Anton Ertl:   
   > minforth  writes:   
   >> Am 19.07.2025 um 12:18 schrieb Anton Ertl:   
   >>   
   >>> One way to deal with all that would be to have a long-vector stack and   
   >>> have something like my vector wordset   
   >>> , where the sum of a vector   
   >>> would be a word that is implemented in some lower-level way (e.g.,   
   >>> assembly language); the sum of a vector is actually a planned, but not   
   >>> yet existing feature of this wordset.   
   >>>   
   >>   
   >> Not wanting to sound negative, but who in practice adds up long   
   >> vectors, apart from testing compilers and fp-arithmetic?   
   >   
   > Everyone who does dot-products.   
   >   
   >> Dot products, on the other hand, are fundamental for many linear   
   >> algebra algorithms, eg. matrix multiplication and AI.   
   >   
   > If I add a vector-sum word   
   >   
   > df+red ( dfv -- r )   
   > \ r is the sum of the elements of dfv   
   >   
   > to the vector wordset, then the dot-product is:   
   >   
   > : dot-product ( dfv1 dfv2 -- r )   
   >    df*v df+red ;   
      
   Sure, slow hand is not just for guitar players.   
   With FMA, one could traverse the vectors in one go.   
      
   https://docs.nvidia.com/cuda/floating-point/index.html   
      
   > Concerning matrix multiplication, while you can use the dot-product   
   > for it, there are many other ways to do it, and some are more   
   > efficient (although, admittedly, I have not used pairwise addition for   
   > these ways).   
      
   There are tons of algorithms depending on various matrix properties.   
      
   Then, given a desktop and a fat CPU, LAPACK et al. are your friends.   
   Embedded or special CPU .. is a different story.   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)