0

Sorry if the question is naive or obvious: I am the user, not developer. The question is whether LSODA as implemented in ODEPACK (FORTRAN code) takes advantage of AVX option of Xeon processors, and how much performance improvement relative to no-AVX can I expect? Same for AVX512: will it be further speed up?

Vladimir F Героям слава
  • 57,977
  • 4
  • 76
  • 119
user8423358
  • 1
  • 1
  • if i understand correctly, ODEPACK is a library written in Fortran. in this case, rebuild it with the right compiler flags and the Fortran compiler will do its best to vectorize it. if your application is working on double precision, the best speedup versus non vectorized version you can expect is 4x for AVX and 8x for AVX512 (8x and 16x for single precision) – Gilles Gouaillardet Aug 06 '17 at 04:46
  • One should stress that even on a non-AVX CPU one will likely have vectorization using SSE or SSE2 or similar – Vladimir F Героям слава Aug 06 '17 at 17:30
  • Where can we see the code? Does it work on arrays? – Vladimir F Героям слава Aug 06 '17 at 17:33
  • 1
    General discussion can be found here http://lh3lh3.users.sourceforge.net/solveode.shtml ODEPACK description is here http://people.sc.fsu.edu/~jburkardt/f77_src/odepack/odepack.html and the code is here http://www.netlib.org/odepack/opkdmain.f Looks like it may result in large gain in speed, I will try it, thanks for your input – user8423358 Aug 07 '17 at 15:11

0 Answers0