Title: Can My Program Be Fast, Energy-efficient, and Simple to Write?

Abstract:

The past decade saw dramatic and disruptive changes to the computing
landscape, with the widespread development of handheld devices, and
with power density and energy considerations becoming the primary
constraints driving technology directions for embedded, mainstream,
and peta/exascale computing at the high end. Non-homogeneous CPU cores
and increasingly complex System-on-Chips are on the roadmap of most
manufacturers. In a word, computing platforms are now heterogeneous,
after decades of mass marketing homogeneous single-core x86
processors.  

These changes have fundamentally impacted the application
development cycle. First, the effort required to efficiently map an
algorithm to a particular hardware has skyrocketed, due to the
increasing complexity of both the target hardware and the software
stack to be used. As a consequence, determining whether the algorithm
is at fault or its implementation when facing sub-par performance is a
considerably time-consuming task, in turn preventing scientists to
quickly evaluate new algorithmic ideas and significantly increasing
the development cost of new consumer applications. Second, the number
of application domains that effectively need to harness massive
computing capabilities keeps increasing: medicine alone drives a
high-impact demand, with automated and computer-assisted diagnosis,
individual tumor-specific cancer treatment, or large-scale exploratory
genome datamining have all become a computing challenge. Our research
must provide usable, practical solutions to address these challenges,
making scientific advances eventually possible via high-performance
computing.  In this talk I will present some current and upcoming
research to dramatically improve productivity and performance
portability on modern computing devices, spanning application
modeling, compiler optimization, and architecture design.  We will see
how a pluri-disciplinary approach is needed to reach the level of
efficiency demanded to address a variety computing challenges, with
illustrations from several key applications such as in-situ automated
lung tumor detection, energy-efficient camera image processing for
handheld devices, or large-scale physics simulations using distributed
computing.