HP MLIB for PA-RISC and Itanium HP-UX Version 9.5.6 Release Note

What’s in This Version

• Sparse vector and matrix operations, including the Sparse BLAS

— BLAS 3 routine DGEMM is highly tuned for PA-RISC and Itanium-based platforms.

• Discrete Fourier transforms

• Convolution and correlation

• Miscellaneous tasks such as sorting and generating random numbers

• SCILIB calling conventions

MLIB contains multiple routines to perform the same mathematical operations using

different precisions. For example, matrix multiplication may be performed using the

SGEMM, DGEMM, CGEMM or ZGEMM routines, where S refers to single precision, D

refers to double precision, C refers to single complex precision, and Z refers to double

precision. All of these assume that single precision is represented using 32-bit

floating-point data. All MLIB libraries follow this convention except for the new

VECLIBSC8 and LAPACKSC8 libraries. The new VECLIBSC8 and LAPACKSC8

libraries use 64-bit addresses and 64-bit integer values and assume the single precision is

represented using 64-bit floating-point data. Therefore, the routines whose names begin

with the lettter S use 64-bit floating-point numbers, and the routines whose names begin

with the leter C use complex numbers whose real and imaginary components use 64-bit

floating-point numbers. The D and Z variants of these routines are not included. So

SGEMM in VECLIBSC8 has identical functionality to DGEMM from VECLIB8 and there

is no DGEMM routine in VECLIBSC8.

These libraries are provided for users who prefer Fortran single precision to match the

size of integers. Under this assumption, Fortran integers using 64-bits implies single

precision (S) is represented using 64-bits. By extension, complex single precision (C) has

real and imaginary components each of size 64-bits.

Although VECLIB was designed for use with Fortran programs, C programs can call VECLIB

subprograms as described in Appendix A of the HP MLIB User's Guide.

LAPACK

LAPACK is a collection of Fortran-callable subprograms that provides mathematical software

for application programs involving linear algebra (including linear equations), least squares,

eigenvalue problems, and the singular value decomposition. The name "LAPACK" is an

acronym for "Linear Algebra PACKage". The package was designed to supersede the earlier

LINPACK and EISPACK packages. The National Science Foundation, the Defense Advanced

Research Projects Agency, and the Department of Energy supported the development of the

public domain version of LAPACK, from which HP MLIB LAPACK was derived. HP MLIB

LAPACK is fully compliant with the LAPACK 3.0 Standard.