Saturday, June 13, 2009

Can Computers be Programmed Productively in the Post-Dividend Era?

by Rastislav Bodík, Justin Bonnar (University of California Berkeley, USA) and Doug Kimelman (IBM T.J. Watson Research Center, USA)

Summary of review comments:
The authors argue that programming abstractions that have lead to programmer productivity are not necessarily energy efficient. So in the context of energy constrained computing devices (i.e., when battery life is short), there must be a better way of implementing the intended computation. They outline areas of focus that they are pursuing, which can be explained as (static) specialization, direct refinement from high-level abstraction, and ML-style static languages.

2 comments:

  1. I liked this talk. It seems to have some synergies with the "bloat" work, although it is only 1 example of the abstraction tax.

    ReplyDelete
  2. This brings to mind Stepanov's abstraction penalty benchmark,
    which can be found at:

    http://www.stepanovpapers.com/

    While at SGI, Stepanov encouraged the compiler team to reduce
    the abstraction penalty for the SGI compiler. Optimizations
    like scalar replacement of aggregates and, of course, procedure
    inlining, played an important role in that work.

    ReplyDelete