to be able to use OpenMP parallelism on the node, and MPI between nodes. We describe the challenges of converting MILC MPI code to using a second level of OpenMP parallelism, and benchmarks on IBM and Sun computers.

MPI / OpenMP mixed mode codes could potentially offer the most effective parallelisation strategy for an SMP cluster, as well as allowing the different characteristics of both paradigms to be exploited to give the best performance on a single SMP. This paper discusses the implementation

movements inside the CLUMP 3) to limit the effort of the programmer while ensuring the portability of the codes on a wide vaxiety of CLUMP configurations. We investigate an approach based on the MPI and OpenMP standaxds. The approach consists in the intra-node paxallelization of the MPI programs

of MPI calls related to thread-safety violations, then replaces them with our own MPI wrappers, which involve accesses to some specific shared variables. The static analysis avoids instrumenting unrelated code, which significantly reduces runtime overhead. In the dynamic analysis, both happen

hardware point of view, we will propose a general model for estimating overhead in OpenMP loops. We will parallelize a physics simulation to do large scale simulations as efficient as possible. We will look at typical physics simulation codes to parallelize and optimize them with OpenMP and MPI in a mixed

implementation in GCC, providing code trans-formations in the GIMPLE representation and compare them to those necessary for transforming a shared-memory program (annotated with OpenMP pragmas) to a message-passing MPI-based program. In order to test some ideas, we have developed a limited prototype. The result

Introduction Despite the progress in numerical algorithm and computer power, reliable CFD prediction of complex flows as turbulent flows is still a challenge. Nu- merical turbulent prediction (Large Eddy Simulation, see [7] for more details) computes the three-dimensional, unsteady turbulent flow directly by use of the Navier-Stokes equations, with a model utilized only for the small scales of motion that are unresolved by the computational mesh (subgrid-scale model). The application of LES to industrial flows is highly attractive as the present increase in computing power of computers brings the necessary computational res-sources within the reach of industrial research and design groups. However, many situations still require High-Performance Computers. This requirement is due to the use of fine meshes with a huge number of points. A typical simulation the requires around 106 to 107 grid points. In turn, this implies a small size of the time step, increasing the number of iterations

installation, and the combination of which present significant logistical challenges for end users. Building upon prior research, and influenced by experience gained from the code bases of the LAM/MPI, LA-MPI, and FT-MPI projects, Open MPI is an all-new, productionquality MPI-2 implementation

We describe an implementation of a sizable subset of OpenMP on networks of workstations (NOWs). By extending the availability of OpenMP to NOWs, we overcome one of its primary drawbacks compared to MPI, namely lack of portability to environments other than hardware shared memory machines

, that will accept OpenMP and will target TreadMarks, a Software Distributed Shared Memory System (SDSM), and Message-Passing Interface (MPI) library routines. A second part of the paper presents ideas for OpenMP extensions that enable the programmer to override the compiler whenever automatic methods fail