I am mostly working around the PaStiX sparse direct
solver, implementing functionnalities to enhance the
Block Low-Rank part of the solver.
This software implements dense linear algebra algorithms for
heterogeneous distributed architectures. I have implemented
the reduction from a dense matrix to a bidiagonal or a
tridiagonal matrix. This is a first step when computing
singular values or eigenvalues of dense matrices. I has been
implemented on top of Quark, Parsec and StarPU runtime
systems.
PaStiX is a sparse direct solver that can exploit
distributed heterogeneous architectures. Most of the
developments performed during my thesis were integrated in
this sofwtare and were made available publicy.
PLASMA aims at replacing LAPACK for multi-threaded
architectures. I have integrated an algorithm to compute
eigenpairs of tridiagonal matrices with a Divide & Conquer
method, which is one of the main stages when computing
eigenpairs of dense matrices.