ScaLAPACK, pzheev work-size query and doc
Posted: Fri Jul 14, 2017 6:16 amI have now tested all the ScaLAPACK (and LAPACK) eigenvalue routines (dsy and zhe) to assert that they return the correct work-size for successful runs.
All but pzheev return the correct sizes.
For JOBZ='N' the worksize `nint(rwork(1))` is correct (>= 2*n)
For JOBZ='V' the worksize `nint(rwork(1))` is incorrect (< 2*n + 2*n-2)
Hence there is something wrong with the pzheev work-size query.
As for the documentation it is stated that the rwork array is a complex*16, however, it should merely be a double precision array.
(line 182 in pzheev.f)
All but pzheev return the correct sizes.
For JOBZ='N' the worksize `nint(rwork(1))` is correct (>= 2*n)
For JOBZ='V' the worksize `nint(rwork(1))` is incorrect (< 2*n + 2*n-2)
Hence there is something wrong with the pzheev work-size query.
As for the documentation it is stated that the rwork array is a complex*16, however, it should merely be a double precision array.
(line 182 in pzheev.f)