Azzam HAIDAR

Research activity

Numerical analysis methods :

• Fast Hybrid (direct/iterative) solvers for large-scale scientific applications
• Eigenvalues and eigenvectors problems
• Domain decomposition techniques, Preconditioning techniques
• Mixed-precision linear algebra
• Finite elements, mechanics, structural mechanics, linear elasticity, Lagrange problems

High performance computing:

• High performance scientific libraries
• Massively parallel algorithms
• Hybrid multi-level of parallelsim
• Parallel and distributed environements computing
• Design of new multi-core and hybrid algorithm
• Design and development of portable codes

Large dense matrix calculations:

• Dense librairies (BLAS LAPACK PLASMA MAGMA)
• Dynamic asynchronous out of order scheduling
• Eigenvalues and Eigenvectors solvers, two sided reduction
• Linear Algebra software packages
• Reliability and robutness of numerical software

Large sparse matrix calculations:

• Direct linear solvers
• Iterative linear solvers, preconditioning strategies
• Reliability and robutness of numerical software

Industrial numerical simulation:

• Applications in structural mechanics
• Applications in seismic imaging
• Applications in Electromagnetics
• Applications in semiconductor
• Applications in Computational Fluid Dynamics

Related Projects and Software

• Topic: Distributed Parallel Linear Algebra Software for Multicore Architectures. DPLASMA is the leading implementation of a dense linear algebra package for distributed heterogeneous systems. It is designed to deliver sustained performance for distributed systems where each node featuring multiple sockets of multicore processors, and if available, accelerators like GPUs or Intel Xeon Phi. DPLASMA achieves this objective through the state of the art PaRSEC runtime, porting the Parallel Linear Algebra Software for Multicore Architectures (PLASMA) algorithms to the distributed memory realm.

• Topic: Matrix Algebra on GPU and Multicore Architectures (MAGMA) that is a collection of next generation linear algebra (LA) algorithms for heterogeneous architectures. MAGMA allows scientific applications to fully exploit the power of current heterogeneous computer systems of multi/many-core CPUs and multi-GPUs/coprocessors to deliver the fastest possible time to accurate solution within given energy constraints. The University of Tennessee, the University of California, Berkeley, the University of Colorado Denver and INRI, France contribute to the development of MAGMA.

• Topic: Parallel Runtime Scheduling and Execution Controller. PaRSEC is a generic framework for architecture aware scheduling and management of micro-tasks on distributed many-core heterogeneous architectures. Applications we consider can be expressed as a Direct Acyclic Graph of tasks with labeled edges designating data dependencies. DAGs are represented in a compact problem-size independent format that can be queried on-demand to discover data dependencies in a totally distributed fashion. PaRSEC assigns computation threads to the cores, overlaps communications and computations and uses a dynamic, fully-distributed scheduler based on architectural features such as NUMA nodes and algorithmic features such as data reuse.

• Topic: Parallel Linear Algebra for Scalable Multi-core Architectures. Software frameworks that enable programmers to simplify the process of developing applications that can achieve both high performance and portability across a range of new architectures. The University of Tennessee, the University of California, Berkeley and the University of Colorado Denver develop PLASMA.

• Topic: The Linear Algebra PACKage (LAPACK) and Scalable LAPACK (ScaLAPACK) are widely used libraries for efficiently solving dense linear algebra problems. LAPACK is sequential, relies on the BLAS library, and benefits from the multicore BLAS library. ScaLAPACK is parallel and distributed and relies on BLAS, LAPACK, MPI, and BLACS libraries. The University of Tennessee, the University of California, Berkeley and the University of Colorado Denver have developed these packages.

• MaPHyS: Massively Parallel Hybrid linear Solver.
• Topic: MaPHyS is a scientific library that provides a high performance parallel solver for very large sparse linear systems based on hybrid methods.
• Authors: Azzam Haidar & Luc Giraud.
• Partners: CERFACS-INRIA

• Topic: Study of parallel hybrid sparse linear solvers
• Partners:INRIA, University of Minnesota, TU Brunswick

• ANR-CIS project funded by the french ministry of research
• Topic: Grand challenge simulations and linear solvers
• Academic partners: CERFACS, CNRM-LA, ENSEEIHT-IRIT, INRIA Bordeaux-Sud-Ouest and INRIA Grenoble-Rhône-Alpes
• Industrial partners: CEA-CESTA, EADS-IW, EDF

• Topic: Global offset seismic imaging
• Academic partners: UMR GEOAZUR, IRD, CNRS, Université de Nice
• Industrial partners: BP, CGGVERITAS, EXXON-MOBIL, SHELL, TOTAL

Thesis

Activités de recherche