1

I am wondering why the FLOP-count for division is treated different in literature (and internet).

I have found this definition here at stackoverflow (1div = 4flop): https://stackoverflow.com/a/329243/6059576

and another one in the book "Matrix Calculations" by Golub & Van Loan (1div = 1 flop) https://books.google.de/books?id=X5YfsuCWpxMC&lpg=PA12&ots=91CiEj4dn-&dq=matrix%20computations%20flop&hl=de&pg=PA12#v=onepage&q=matrix%20computations%20flop&f=false

and another in "Numerical Computation" by Ueberhuber (1div = 10-30 flop) https://books.google.de/books?id=giH7CAAAQBAJ&lpg=PA21&ots=DtPHO06Eta&dq=numerical%20computation1&hl=de&pg=PA193#v=onepage&q&f=false

Is there any reference (book,other publication,...) for the first one? Because the Author says

..sort of a defacto standard in the HPC community

Community
  • 1
  • 1
boschika
  • 7
  • 2
  • This question is not really about benchmarking, but about estimating flop count for complex FP instructions, such as division. May be a more informative title would benefit. – Dima Chubarov Mar 14 '16 at 11:58

0 Answers0