Week 6

  • We added sin, cos, asin, and acos functions to our library.
    • Using a taylor series expansion, and taking advantage of the functional symmetry works very well for sin and cos.
    • asin, and acos were implemented as a binary search that uses sin and cos for guess evaluation.
  • Philip and I tried to integrate our library into the MD Code.
    • We observed that there were several problems with our algorithms when ported onto the MD simulation.
    • Specifically, the code performed much slower than our synthetic results would have otherwise indicated.
    • The problem seems to be particular to Philip's division algorithm (probably because there are many more division operations performed).
    • Philip found that the various conditionals caused severe performance losses.
    • We had decided that it would be best to sacrifice robustness for performance.
    • So, Philip and I are working on replacing these conditionals with logical operations and redundant computations.
    • I also tried to improve the performance of my own NR sqrt by removing conditionals and unrolling loops.
      • The performance difference seems negligible.
  • I submitted my student volunteer application to SC 09
    • Supercomputing 2009 is the premier conference for high performance computing.
    • If accepted, this would be my third time at SC as a student volunteer.
    • We also worked on the extended abstract for our poster for SC (due next Friday!).
  • I was at a talk given by Hitoshi Oi, a visiting professor from The University of Aizu in Japan.
    • He spoke about improving network performance in Xen virtual machines.
    • After the talk, we went to eat pizza at grotto's.