Matrix-matrix multiply

• 

• For example, consider 9x9 matrix on 9 PEs
  • column-major

  • at most two copies of subdomains' data fits on 1 PE

• Compare combinations of domain decompositions
  • Like HPF DISTRIBUTION's

• For message-passing time estimates, use:
  • Message latency =

  • Transfer speed for 1W (64-bit word) = 10nsec

Generated with CERN WebMaker