Re: Makefile suggestion

[R Clint Whaley <rwhaley@cs.utk.edu>]

Andy,

>I've noticed that there are four unused source files in
>ATLAS/interfaces/blas/C/src:
>cblas_zrotg.c
>cblas_crotg.c
>cblas_csrot.c
>cblas_zdrot.c
>
>They are not mentioned in the Makefile.
>
>(this is all with respect to the atlas3.1.1D.tgz tarball)
>
>By just editting the Makefile, I can build cblas_zrotg and cblas_crotg:

Thanks for pointing out my error, see below for the feeble excuse :)

>As an aside, (please forgive the ignorance) are the routines crotg and
>zrotg "officially" part of BLAS? They're not on my BLAS quick reference
>sheet, but they're listed on netlib as part of BLAS.
>
>The same question also goes for csrot and zdrot, though I've run into a
>little more trouble compiling cblas_csrot and cblas_zdrot...

None of these routines are part of the "official" blas standard (which is
defined by a series of published papers).  The level 1 standard is defined
in:
@ARTICLE
{BLAS1,
      AUTHOR = "C. Lawson and R. Hanson and D. Kincaid and F. Krogh",
      TITLE = "``{Basic Linear Algebra Subprograms for Fortran Usage}''",
      JOURNAL = toms ,
      YEAR = "1979",
      VOLUME = "5",
      NUMBER = "3",
      PAGES = "308--323"
}

However, we wanted ATLAS to a complete drop-in replacement for the Fortran77
reference BLAS, which most people think of as the standard anyway, so we
provide these "non-standard" routines.  Unfortunately, the Level 1 BLAS
tester only tests the routines in the official "paper" standard, and I have
been too lazy to add the tests for the C interface (Antoine wrote ATLAS's f77
interface to the BLAS, and he tested these routines as I should have done, so
I think the F77 interface, and ATLAS's underlying computational routines, 
are correct) but my lack of rigor has allowed these problems to remain in the
C interface.  I plan to add the required tests, but it is on a long list
of things that I plan to do :)

I hope to release another developer release soon, which will include the
corrected makefile, at least . . .

Thanks,
Clint