pdgbtrf potential bug under edge cases

[TLCFEM]

When BWL=BWU=0, LDBB evaluates to 0.

scalapack/SRC/pdgbtrf.f

Line 829 in a23c2cd

LDBB = 2*BW + BWU

Later when it is passed to dgetrf, it triggers an error as the lead is expect to be positive.

scalapack/SRC/pdgbtrf.f

Lines 860 to 861 in a23c2cd

	CALL DGETRF( N-LN, N-LN, AF( BBPTR+BW*LDBB ), LDBB,
	$ IPIV( LN+1 ), INFO )

 ** On entry to DGETRF parameter number  4 had an illegal value

pdgbtrs will further raise a few errors when BWL=BWU=0.

Here is a MWE.

program gbsv_test
    implicit none
    integer ictxt

    integer n, kl, ku, ja, laf, lwork, info
    integer, dimension(10) :: desca, ipiv, af, work
    double precision a

    call blacs_get(-1, 0, ictxt)
    call blacs_gridinit(ictxt, 'R', 1, 1)

    n = 1
    kl = 0
    ku = 0
    a = 1.0
    ja = 1
    laf = 10
    lwork = 10
    info = 0

    desca(1) = 501
    desca(2) = ictxt
    desca(3) = 1
    desca(4) = 1
    desca(5) = 0
    desca(6) = 1

    call pdgbtrf(n, kl, ku, a, ja, desca, ipiv, af, laf, work, lwork, info)

    call blacs_gridexit(ictxt)
    call blacs_exit(0)

end program gbsv_test

[TLCFEM]

Also, I noticed that the parameter IPIV, the documentation states it shall >= DESCA( NB ), which is the block size if I understand correctly.

However, doing so sometimes would lead to overflow.

It appears it shall be >= LOCc( N ) + BWL + BWU.