Women in Technology

Hear us Roar



Article:
  The AltiVec Difference
Subject:   simd
Date:   2002-04-10 08:03:17
From:   psheldon
I am happy that you illustrated how many algorithms could be parallelized by the G4 instruction set. I had thought that Apple was only selling the cocoa frameworks and would never discuss parallel algorithms constructed in near pseudocode, but Apple is also in the business of selling hardware and it's instruction sets. I have hope that I shall learn how to organize myself into making my own frameworks of objects cotaining algorithms to save my "knowledge work" in science, not merely interface, be interface ever so important for science.


The fact that there is such a modifier (single instruction multiple data) means that there are other modifiers, for example multiple instruction multiple data. Perhaps here I am sensing familiarity in language of someone who consumes and specifies processors just like General Dynamics did.


I might even hope for an article pointing to the extended bacher nauer formalism for cocoa as a translation exercise for some algorithms packaged somewhere else such as cactus? A translation exercise without a nervous standards committee at my home alone might be a blast.


Full Threads Oldest First

Showing messages 1 through 1 of 1.

  • found your NASA trade study asking Jeeves on cactus then fortran90
    2002-04-13 14:02:55  psheldon [View]

    So both cactus and fortran 90 pushing into multiprocessor eventually but not now.
    Seemed to me the major speedup is in Apple's matrix multiplication algorithm exploiting the G4 arrray processor not in the multiprocessor cluster.
    Is that correct?
    So far got to page 14 of the Jy 2000 report "Evaluation of Powermac G4 Systems for Fortran-..."