Skip to content
dybber edited this page Oct 29, 2012 · 10 revisions

Common Log Of Human Work

29-10-2012

  • 9:00-11:00

    Discussion on how to proceed

16-10-2012

  • 10:00-14:30

    Individual activities. (See individual logs)

  • 14:30-16:30

    Pair debugging: Eliminated rounding errors in CUDA pricer and reduced memory consumption considerably.

15-10-2012

Individual activities. (See individual logs)

12-10-2012

  • 09:30-09:59

    Revising schedule and searching for literature.

  • 10:00-11:59

    Individual activities. (See individual logs)

11-10-2012

  • 10:00-12:00

    Debugging cuda. Martin knows sufficiently cuda voodoo to make kernels print. Discovered that all threads read optionData[tid] rather than optionData[blockDim.x]!.

  • 13:00-13:59

    "printf" in Cuda has a re-ordering, call excluding, life on its own. Caught an off-by-one indexing error (not completely sure why it was there).

  • 14:00-14:40

    Last (apparent) error in cuda implementation found! It seems that the binomial reducer must have "cache-Δ" devisible by "n", otherwise it seems to compute "too much". We celebrate by having a ~5 minute fresh air break with victory Tic-Tacs. And we decide that the calendar now has 256 days in it.

  • 15:00-15:59

    Trimming Nvidia sdk-sample cruft from Cuda example. Currently undecided about the calendar issue in Cuda. Maintaining wiki. (Checking that every implementation executes on napoleon || checking survey literature) ("||" is the CSP parallel process operator) Setting up plc computer to use diku proxy for literature search.

  • 16:00-16:30

    Literature search. Done for today.

Survey "VectorMARK" in progress

Clone this wiki locally