In order to achieve practical efficient execution on a parallel architecture, a knowledge of the data dependencies related to the application appears as the key point for building an efficient schedule. By restricting accesses in shared memory, we show that such a data dependency graph can be computed on-line on a distributed architecture. The overhead introduced is bounded with respect to the parallelism expressed by the user: each basic computation corresponds to a user-defined task, each data-dependency to a userdefined data structure. We introduce a language named Athapascan-1 that allows built a graph of dependencies from a strong typing of shared memory accesses. We detail compilation and implementation of the language. Besides, the performance of a code (parallel time, communication and arithmetic works, memory space) are defined from a cost model without the need of a machine model. We exhibit efficient scheduling with respect to these costs on theoretical machine models. Keyw...

Athapascan-1 is a C++ library for multi-threaded parallel programming. It implements a macro data-flow language: both computation and data grains are explicit. Semantics of shared memory access is based on a lexicographic order. The performance of a code (parallel time, communication and arithmetic works, space) can be evaluated from a cost model without need of a machine model.

DECK (Distributed Executive Communication Kernel) is a communication layer that provides support for multithreading and fault tolerance. The approach retained in DECK is close to other distributed communication kernels like PM2, Athapascan, Nexus, TPVM or Chant in its way to integrate communication and multithreading to efficiently overlap communication by computation and provide low latency mechanisms to remote thread creation. However, DECK differs from these communication kernels from the services offered and its modular architecture. The main goal of DECK is to implement a new model for the design of distributed executive kernel to efficiently use the new underlying hardware architectures (SMP architectures and fast communication adapters like Myrinet or memory oriented adapter like SCI) and provide a portable layer that abstract the problems linked with the integration of communication and multithreading while offering support for heterogeneity. A great lack in the cur...

Parallelizing irregular, dynamic data structures can be a very difficult problem. An efficient solution often demands that work on the data structure be divided up among processors, yet despite the large and growing number of such applications there has been little work done on general approaches to such situations. In part this is due to the extreme difficulty of ensuring enough generality to be useful, while still efficiently addressing each individual problem; irregular and dynamic problems can vary dramatically, and direct, efficient solutions simply do not exist. Most attempts have therefore either concentrated on problem-specific areas, where good results can be obtained at the expense of generality, or have defaulted to heuristic methods applicable to almost any possible situation. In this thesis we present a hybrid method, combining a direct and problemspecific approach with a system general enough to be applicable to the majority of applications. Our method is based on local g...

of some characteristics of softwares for parallel computer algebra. SBSH means Sugarbush. PCLBSTM means PACLIB/STURM and PCGVR PAC/GIVARO. Conclusions. Major problems that appears in the design and implementation of parallel computer algebra systems (on-line scheduling of tasks, distributed garbage collection) are studied in the more general context of modern parallel programming languages. Also, parallel computer algebra focus nowdays on two directions. First, a development effort has to be done to integrate a modern parallel language in a general purpose computer algebra system with its huge libraries of complex algorithms. Second, active research should be done in the design of efficient and portable parallel algorithms for more complex problems. Thierry Gautier (INRIA, LMC-IMAG), Hoon Hong (NCSU), Jean-Louis Roch (LMC-IMAG), Gilles Villard (CNRS, LMC-IMAG), Wolfgang Schreiner (RISCLinz) References ...

We present Athapascan-1, a language implemented as a C++ library that enables the on-line computation of the macro-data flow derived from the data-dependencies of a parallel application. The parallelism is explicit but synchronization implicit. The semantics of Athapascan-1 is data driven; it is independent from the scheduling algorithm used to execute the application. The overhead introduced is bounded with respect to the parallelism expressed by the user: each basic computation corresponds to a user-defined task, each data-dependency to a user-defined data structure. Then, the performance of a code (parallel time, communication and arithmetic works, memory space) can be evaluated directly from a cost model without need of a machine model. Keywords: Multithreading, macro-data flow languages, on-line scheduling, parallel complexity. I. Introduction Recent work in the field of parallel programming has resulted in the definition of extensions of sequential languages that can be efficie...

this article we presented several methodologies for solving irregular problems. First we introduced a generic method then we focused on parallel computers using with distributed memory and we finally proposed a methodology for a multithreaded context. A port of the BOB Branch&Bound library on PM

Athapascan-1 is a C++ library for multi-threaded parallel programming. It implements a macro data-flow language: both computation and data grains are explicit. Semantics of shared memory access is based on a lexicographic order. The performance of a code (parallel time, communication and arithmetic works, space) can be evaluated from a cost model without need of a machine model. Its important feature is that, forbidding synchronization, it allows efficient on-line analysis of the data flow describing the execution. This enable to separate the scheduling algorithm, that can take benefit of the knowledge of this graph, from the code. On specific elementary machine models, simple scheduling algorithms can be implemented that provide efficient execution of some specific class of codes. Portability is then achieved by suiting the scheduling algorithm to both the program and the target machine. 1 Introduction Recent work in the field of parallel programming has resulted in the definition o...