You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
@abuttari
I implemented qrm_spfct_unmqr in #99 and I have a comment for you.
In the real case, you could easily support transp='c' and dispatch to transp='t'.
In the complex case, if transp='t' is not supported, it should return an error.
The flag transp is just ignored when it's not supported and we have the result with transp='n'.
I can do that directly in the Julia interface but it could be relevant for other C or Fortran users.
A quick question, what the difference with qrm_apply?
In general we have another argument side for unmqr.
unmqr now return an error if called with transp=t.
apply is currently only an alias for unmqr for backward compatibility. In the latest version I have renamed a few routines because "apply" (renamed into unmqr) and "solve" (renamed into trsm) don't seem accurate enough.
Yes in LAPACK unmqr there is also a "side" argument but this feature is currently lacking in qrm. It will we added if the need arises.
Float32 + transp='n'
Float32 + transp='t'
Float32 + transp='c'
Float64 + transp='n'
Float64 + transp='t'
Float64 + transp='c'
ComplexF32 + transp='n'
ComplexF32 + transp='t'
ComplexF32 + transp='c'
ComplexF64 + transp='n'
ComplexF64 + transp='t'
ComplexF64 + transp='c'
The text was updated successfully, but these errors were encountered: