lapackers

["Deaglan Halligan" <deaglanhalligan@berkeley.edu>]

Odd about the LAPACK routine, I'll take a look later. I know that xGEES
outputs standard Schur form, as well as xGEEV.

I've finished debugging and testing the P{s,d,c,z}GEES codes. Right now
I'm just doing comments and cosmetic changes. I can post some of these
routines online.

I can also start making P{c,z}GEEV, which is just a matter of gluing
together some high-level routines for the top level driver, and then
translating the LAPACK test codes for {c,z}GEEV into a Sca/LAPACK driver.
Shouldn't take too long.

As I was saying before the real case for P{s,d}GEEV will be more
complicated because we don't have Sca/LAPACK code to get the eigenvectors
from Schur form, "P{s,d}TREVC" routines. The complex ones P{c,z}GEEV
already exist. I'm not sure yet how hard this will be to write.

I agree we should put together these design decisions into a LAWN.

Deaglan

>
>
> On Mon, 29 Jun 2009, James Demmel wrote:
>
>> I also took a very quick look at SHGEQZ (1243 lines), and don't
>> immediately
>> see where it does what the comments claim: basically do a 2x2 SVD
>> to "diagonalize" the 2x2 blocks in the output B matrix.
>
> ok, so forget everything I said before. What you wrote is what I am
> speaking about. (well, actually, what Vasile is speaking about)
> --julien.
>
>