October 5 - 6, 2010

Rice University, Houston, Texas


The advent of high-performance architectures, while potentially more powerful, put an increasing work load on the computational scientist to adapt their code to new programming models and environments. Moreover, the rate at which new architectures arise is increasing, making the problem of adapting their code not just a one time inconvenience but a continual burden.  As a rule, the programmer's own science knowledge base and much of the code is largely unaffected by these upheavals: it is mainly the underlying numerical algorithms and data structures that require the reworking.  Nonetheless these aspects of the code may be pervasive and non-modular: woven densely into every part of the code.  Currently programmers find themselves in an environment for which mature tools do not exist and are forced to do all of the work themselves by hand. For the scientist-developer who's primary concern is science, this is intimidating given the increasing knowledge required to take advantage of these exotic platforms that are changing an increasing pace.  A number of researchers have suggested that compiler-like, possibly domain-specific language tools have a role to play in automating code transformations to more easily

adapt the scientist-developer's code to new architectures.


This workshop is being organized to bring together a diverse set of participants to identify a path forward to address this problem through use of language tools.  The committee has identified three critical questions that should be addressed:


1. Because HPC platforms, platform architectures and programming models are changing at an ever more rapid pace, scientists need a way to keep the science, as expressed in code, separate and independent of these low level implementation details.  What is the transition path to new programming models and/or languages, and what levels of automation are possible and should be pursued? What types and how much user input will be required (e.g., semantical annotations or partial implementations such as templates)?


2. What code transformation tools are needed to support wider adoption of code transformation techniques by the scientist code developer?  What compiler and language-level tools are needed by the *tool* developer to achieve this end?


3. If a majority of the scientist developer's code does not depend on the platform or programming model, is there a way of expressing the scientist's intent as separate from the target environment?



Participation:

-------------------


This workshop will include a small number of invited speakers who will be asked to address these questions.  Participation in the workshop will be open to the wider computing community.  People wishing to participate must SUBMIT A HALF-PAGE SUMMARY of their position on this topic, SPECIFICALLY ADDRESSING THE 3 QUESTIONS ABOVE.  We ask that participants come prepared to think ambitiously about the larger problem rather than to focus on any single existing tool or solution.


The structure of the workshop will be a full day of discussion with a small number of presentations by invited speakers.  This will be followed by a second half day (ending at noon) dedicated to summarizing the outcome of the meeting with the goal of creating a position paper representing the thoughts of the participants.  The whitepaper is primarily to inform DOE Office of Advanced Scientific Computing Research of research needs in the language tools field but may be useful for other agencies.


A CS representative from DOE Office of Advanced Scientific ComputingResearch will also give a perspective of their future needs in this area.


Those interested in participating should e-mail language-tools-2010@cca-forum.org.


Participation Requirements:


* Half page (plain text) answering the 3 topical questions above.

* $175 registration fee.


Please send the half page position statement to language-tools-2010@cca-forum.org.  The registration fee is payable at the meeting.


Logistics:

-----------------------------------------------------------------------


Meeting Dates: 8:00a-5:00p 5 October and 8:00a-12:00p 6 October 2010


Location: Duncan Hall 1049, Rice University

Houston, TX


Accommodations: Hilton Houston Plaza/Medical Center next to the Rice

University Medical Center, URL:

http://www1.hilton.com/en_US/hi/hotel/HOUMCHF-Hilton-Houston-Plaza-Medical-Center-Texas/index.do


Ask for the "LCPC" group identifier when making reservations.




Workshop Committee


R. Armstrong (Sandia NL)

B. Chapman (U. Houston)

J. Feo (PNNL)

J. Mellor-Crummey (Rice U.)

B. Norris (Argonne NL)

M. Sottile (Galois, Inc.)

  1. G.Watson (IBM Watson)


Agenda