Current term:
| Date | Speaker | Title (click for abstract) | Location |
| Jan. 17, 2007 | Nomair Naeem | Precise and Efficient Must-Alias Analysis | DC 3313 |
| Jan. 24, 2007 | Peter Kim | On-Demand Materialization of Aspects for Multilevel Customization | DC 3313 |
| Jan. 31, 2007 | Nomair Naeem | Continuation of Precise and Efficient Must-Alias Analysis | DC 3313 |
| Feb. 14, 2007 | No seminar this week | No seminar this week | DC 3313 |
| Feb. 28, 2007 | Brad Lushman | Unknowns---Not Quite Variables, Not Quite Constants | DC 3313 |
| Mar. 14, 2007 | No seminar this week | No seminar this week | DC 3313 |
| Mar. 21, 2007 | Michael McCool | The RapidMind Development Platform and Parallel Programming Model | DC 3313 |
| Mar. 28, 2007 | No seminar this week | No seminar this week | DC 3313 |
Seminars are Wednesdays at 2:30 pm, unless otherwise indicated.
All are welcome.
Talk announcements are also posted to a public mailing list.
To volunteer to give a talk, contact Ondřej Lhoták. Talks can range from informal to formal, and they can be about any of the following:
The purpose is for us all to see what people are interested in, and motivate discussions.
An inspection of the two approaches shows the importance of modeling
individual heap locations to get efficient must-alias information. We
derive similarities and differences in the two implementations and
come to the conclusion that the two approaches can be easily combined to
produce an even better must-alias analysis.
Unknowns---Not Quite Variables, Not Quite Constants
Brad Lushman
Feb. 28, 2007
Unification is the well-known problem (with well-known solutions) of solving
systems of equations in some term algebra. Semiunification is a more general
problem, in which we also admit term inequalities---an inequality tau <= mu
is considered satisfied if for some substitution S, mu S is a substitution
instance of tau S. In other words, extra substitutions are allowed on the
left-hand side of the inequality. Notable among SUP's applications are its
connections to various polymorphic type systems.
Within semiunification problem (SUP) instances, there are two classes of identifiers: variables and functors. Variables may be the targets of substitutions as part of the solution process for SUP; functors may not. Constants may be viewed as nullary functors. We present a new generalization of SUP, in which we introduce a third class of identifier---the unknown. Unknowns are like variables, in that they may be the targets of substitutions, but like functors, since semantically, they are to be regarded as constants. Moreover, a single identifier may play the role of unknown over part of a SUP instance, and of variable over the rest.
We show that the known decidable subsets of SUP, and indeed even the full
SUP itself, are not appropriate targets for a full syntax-directed translation
from typability in rank-2 System F to a set of unification-style constraints.
Instead, a translation to SUP with unknowns (which we call USUP) provides
a much more natural target. We present a formal definition of USUP, detail
the new reduction rules for solving instances with unknowns, and outline
the connection between unknowns and monomorphic abstractions with broad scope.
The RapidMind Development Platform and Parallel Programming Model
Michael McCool
Mar. 21, 2007
Massively multi-core processors such as GPUs and the Cell BE have the
potential to deliver high performance computation to many applications.
However, these processors require parallel programming on several levels and
have some novel architectural features. The RapidMind development platform
enables portable access to the power of these processors. It provides a
simple, safe, data-parallel model of execution and takes care of many of the
low-level details of mapping this processing model to each hardware target.
It combines a dynamic compiler with a runtime streaming execution manager,
and provides a single system image for computations running on any number of
cores. The interface to this system is embedded inside ISO standard C++,
where it can capture arbitrary computations specified at runtime. The use
of a dynamic compiler means that high-level C++ abstractions can be used
without sacrificing performance.