Open discussion for MAGMA library (Matrix Algebra on GPU and Multicore Architectures)
2 posts • Page 1 of 1
Not in the near-term. The eigensolvers themselves (as opposed to the Hessenberg or tridiagonal reduction) are fairly complicated codes that employ BLAS-1 and BLAS-2 functions, which benefit less from the GPU's parallelism. Possibly accumulation of the Schur vectors Z could be done on the GPU. However, the total geev time is now actually dominated by computing the eigenvectors in trevc, which we are looking at.