Drop deprecated LAPACK apis

[jschueller]

dgelsx, dgegv, dgeqpf are deprecated since lapack 3.1.1 (2007), remove them

[beutlich]

Thanks for reporting and fixing this. We are in the preparations for new major MSL (where we are actually allowed to break backwards-compatibility).

As these functions are not used within the MSL it seems sufficient. However, instead of removing these four functions, we require to

• move them to a (to be created) package ObsoleteModelica4.Math.Matrices.LAPACK,
• provide conversion scripts for automatic updating of user libraries (by either using the moved functions or the replacement functions dgelsy, dggev, dgeqp3)
• test models to test the conversion.

Saying this, we missed to mark them as obsolete now, such that users of the latest MSL v3.2.3 are aware of the deprecated functions and the change does not come as unexpected surprise. Not sure, how strict we should be here.

[jschueller]

I dont know how to create packages, can you provide details ?

[beutlich]

@HansOlsson @MartinOtter This PR is now ready for your review.

[HansOlsson]

The work-areas for dgeqp3 does not seem ideal since:

• It is the minimum possible; which will not utilize the level-3 block algorithm in a good way. Would it be possible to have lwork as an optional input (with suitable default)?
• The work-area length is duplicated. If we cannot use work[lwork] would it be possible to not have "lwork" as a variable and instead replace lwork in the argument list by size(work,1)?

[beutlich]

The work-areas for dgeqp3 does not seem ideal since:

* It is the minimum possible; which will not utilize the level-3 block algorithm in a good way. Would it be possible to have lwork as an optional input (with suitable default)?

Hm, could do so. What would be a suitable default for lwork then? Leave it as max(1, 3*size(A, 2) + 1)?

* The work-area length is duplicated. If we cannot use work[lwork] would it be possible to not have "lwork" as a variable and instead replace lwork in the argument list by size(work,1)?

Many LAPACK functions do so (for example dgesvd_sigma) and I just adopted this style of implementation.

[HansOlsson]

Hm, could do so. What would be a suitable default for lwork then? Leave it as max(1, 3*size(A, 2) + 1)?

Yes.

Many LAPACK functions do so (for example dgesvd_sigma) and I just adopted this style of implementation.

You're right. That should likely be improved for more functions.
However, it is related to the change in this PR, since level-3 routines need larger work-areas and we here switch to a level-3 implementation.

[beutlich]

* The work-area length is duplicated. If we cannot use work[lwork] would it be possible to not have "lwork" as a variable and instead replace lwork in the argument list by size(work,1)?

Note, that a an Integer variable lwork is mapped to int type in C, but size(work, 1) would map to size_t in C, but impossible in Fortran. See also #1774.

[HansOlsson]

Looks ok.