[ENH] linalg.hessenberg: use orghr #3506
Conversation
|
Changes Unknown when pulling 8c458a4 on andrenarchy:orghr into * on scipy:master*. |
|
I'm tempted to just merge this but I'll wait for someone who knows something about fortran. |
| integer min_lwork,max_lwork,n,lo,hi | ||
| character prefix | ||
| c | ||
| c Returned maxwrk is acctually optimal lwork. |
For n<=2, the input matrix already is in Hessenberg form. Without seperate treatment, LAPACK returns an error.
| callstatement { hi++; lo++; (*f2py_func)(&n,&lo,&hi,a,&n,tau,work,&lwork,&info); } | ||
| callprotoargument int*,int*,int*,<ctype2>*,int*,<ctype2>*,<ctype2>*,int*,int* | ||
| integer intent(hide),depend(a) :: n = shape(a,0) | ||
| <ftype2> dimension(n,n),intent(in,out,copy,out=ht,aligned8),check(shape(a,0)==shape(a,1)) :: a |
There was a problem hiding this comment.
Why forcing alignment here? --- I presume that's what aligned8 attribute does, but I'm not sure.
|
I'd like to see a test for a small matrix path, and some more tests, esp for complex matrices and weirder orderings (C, F, sliced). EDIT: Had a look at the existing tests. Complex matrices are tested already --- so, it's only 2x2 shortcut which needs an extra test. |
|
Changes Unknown when pulling 21244a9 on andrenarchy:orghr into * on scipy:master*. |
|
In any case this needs a rebase now that gh-3871 changed/fixed the lwork calculations. @andrenarchy would you rebase? |
|
Closing this. Continued in #4490. |
fixes #3502