[SciPy-user] Confusing BLAS/LAPACK situation on Mac OS X
bgranger at scu.edu
Sat Jan 28 16:54:40 CST 2006
Brian Granger, Ph.D.
Assistant Professor of Physics
Santa Clara University
bgranger at scu.edu
>>> robert.kern at gmail.com 01/28/06 1:43 PM >>>
Brian Granger wrote:
> What about:
> Also, Apple's docs on Accelerate claim that they include clapack.
> Are these incomplete/unusable for some reason? It seems silly to build ATLAS if it is already there.
Look at the symbols in libBLAS.dylib and libLAPACK.dylib. You can see the CBLAS
versions in libBLAS.dylib, but there are only the FORTRAN versions in
libLAPACK.dylib. You are right that the docs claim (correctly) that
Accelerate.framework contains CLAPACK. However, I goofed and referred to things
incorrectly. CLAPACK is simply an implementation of FORTRAN LAPACK in C. It uses
FORTRAN column-major arrays. The flapack module links against these. However,
ATLAS also provides row-major versions of some(?) LAPACK functions to match the
row-major CBLAS routines. The clapack module wraps these. These are not included
in Accelerate.framework, although the row-major CBLAS versions are.
I have been reading the header files in the Accelerate framework to get a better understanding of all this.
Apple's implementation of BLAS is the usual CBLAS, which can work with matrices in either row or column major storage. The storage order is determined at runtime by an argument passed to the functions.
Are these wrapped and available in scipy? If so, then where?
The LAPACK implmentation, as you assert, includes only the fortran style functions that require columnn major arrays. So even though these are called "CLAPACK" by Apple, scipy refers to them as "FLAPACK" because they are not the row major versions that ATLAS exposes. Is this correct?
robert.kern at gmail.com
"In the fields of hell where the grass grows high
Are the graves of dreams allowed to die."
-- Richard Harter
SciPy-user mailing list
SciPy-user at scipy.net
This message scanned for viruses and SPAM at SCU (MGW2)
More information about the SciPy-user