Skip to content

Generalized eigenproblem is fully on GPU now

Petr Karpov requested to merge peter_nonhermitian_multiply into master_pre_stage

Introduce new key-value runtime option 'pxtrmm_for_generalized' and alternative GPU-codepath that uses PxTRAN+hermitian_multiply instead of non-GPU ported ScaLAPACK's PxTRMM in generalized eigenproblem. With this, all steps of the generalized eigenproblem are ported to GPU.

Merge request reports