Week 4

  • I finished the performance analysis of the matrix multiplication code.
    • I collected the results
    • I created various plots
    • I compared and analyzed the data for similarities and differences in performance
  • I began working on the report
    • I created the first draft
    • I discussed the revision with Dr. Taufer.
      • We agreed that the framework should allow for a broader discussion of my work
      • I hope to have a framework for the report finished and ready by next week.
  • I helped Philip with a part of his division algorithm
    • We worked on replacing sprintf with our own code for converting floats to char arrays
  • I began adapting my methodology of analyzing GPU performance for the sqrt algorithm
  • I began work on a test suite that would allow testing various algorithms for performance and accuracy
    • Ideally, I would like to have completed my analysis of the various sqrt algorithms by next week
  • I attended a Ph.D. defense.
    • James, our colleague, worked on distributed constraint optimization algirthms
    • His algorithms outperformed many teams from top schools.
    • I found the presentation to be very interesting -- the material was very well organized and presented
  • I read a paper on Message Passing on Data-Parallel Architectures
    • Lifan presented the paper at our group meeting
    • The talk was about DCGN, an extension to MPI that allowed message passing amoung GPU's and CPU's.
    • His presentation was simple to understand and helped answer my questions
    • I wanted to know if the programming model allowed for a single kernel to be executed on both CPU's and GPU's
    • DCGN does not allow for this, but Dr. Taufer suggested I looked at MCUDA'