Description:

New Version:

In this cluster we will focus on algorithms and applications that could enable
the use of heterogeneous architectures for embedded systems. Application areas
of interest include:

 * radio/video streaming
 * cellular phones
 * television
 * content analysis/search/pattern recognition
 * bioinformatics
 * graphics/games

Algorithms already under study apply to high resolution digital video
(H.264/AVC), and bioinformatics (BLAST and Fasta).

Algorithms and applications will be included in a software repository, as
source code, with documentation and performance analysis on the different
architectures under study.

The architectures under consideration will be the STI Cell BE architecture, the
Philips TriMedia architecture, and the STMicroelectronics xSTream architecture.
In this cluster, we will use the currently available toolchains (possibly
based on GCC), to exploit the SIMD, vector, and streaming characteristics of
such architectures. We expect that this study will help in the development of
new approaches and interesting ideas to be applied to the runtime systems for
these architectures. The goal will be to achieve high efficiency, adapting the
coding of the algorithms to the different architectures. An initial target is
obtaining 80% of peak performance.

The fellowship is planned for UPC and he/she will be in charge of collecting
the applications,  use the compiler and runtime infrastructures available on
the different architectures to optimize the algorithms, and document the
transformations done to the code and the performance obtained.

After an initial period of 18 months collecting algorithms and applications for
such architectures, we expect to extend the cluster for the next 18 months
period, adapting the topic to the future research trends on this kind of
architectures.

 ------------------------------------

Old Version: 

In this cluster we will focus on algorithms and applications that could enable the use of the Cell BE processor for embedded systems. Application areas of interest include:

 * radio/video streaming
 * cellular phones
 * television
 * content analysis/search/pattern recognition
 * bioinformatics
 * graphics/games

 Algorithms already under study apply to high resolution digital video (H.264/AVC), and bioinformatics (BLAST and Fasta).

Algorithms and applications will be included in a software repository, as source code, with documentation and performance analysis on the Cell BE processor, and possibly other architectures.

The Cell BE processor has 2 hardware contexts (PE, compatible with the PowerPC architecture) and 8 SPEs. Each SPE is a simple processor with its own small local memory (256 Kb.) that acts as a coprocessor. The instruction set of the SPEs is not compatible with the PowerPC architecture, except that the SIMD instructions are represented by a similar set of compiler intrisics. So, it is easy to port a code with SIMD instructions from the PowerPC architecture to the SPE instruction set.

The current compiler infrastructures for the Cell BE processor include a modified GCC compiler from Sony and the XL compiler from IBM. We plan to use both infrastructures to optimize the applications using, where applicable, vectorization and parallelization.

Vectorization on the Cell processor can be achieved by using SIMD instructions either in the PE or the SPEs.  Parallelization usually will exploit some or all of the 8 SPEs attached to each PE.

The fellowship is planned for UPC and he/she will be in charge of collecting the applications,  use the compiler infrastructures available on the Cell BE processor to optimize them, and document the transformations done to the code and the performance obtained.

After this initial period of 18 months collecting algorithms and applications for the Cell BE processor, we expect to extend the cluster for the next 18 months period, adapting the topic to the future research trends on this kind of architecture.

Nature:
Research cluster

Total funding:
Requested: € 43200

Including fellowship funding:
Requested: € 21600

Description of how the funding will be used:

Travel: we plan to have meetings every 3 months, possibly in conjunction
with other Hipeac events:

   6 people traveling x 4 meetings/year x 600 euros/trip = 14400 euros

Fellowships: one fellowship for a person working on the topic at UPC:

   1 fellowship: 12 months x 1200 euros/month = 14400 euros

Total for 12 months (travel + fellowship): 14400 + 14400 = 28800

Total for 18 months (travel + fellowship):  21600 + 21600 = 43200

At the end of the period, we plan to extend this cluster for the next 18 months period. 

Duration of the funding:
Requested: 18 month(s)

Participating members:
BERNSTEIN David (IBM) (--member--)
BILAS Angelos (FORTH) (--member--)
GIL Marisa (UPC) (--member--)
MARTORELL Xavier (UPC) (--member--)
MENDELSON Bilha (IBM) (--member--)
MORANCHO Enric (UPC) (--member--)
NAVARRO Nacho (UPC) (--member--)
RAMIREZ Alex (UPC) (--member--)
ZAKS Ayal (IBM) (--member--)

Other people collaborating:

Eduard Ayguade, Mauricio Alvarez, Friman Sanchez, Carlos Villavieja,
Javier Cabezas