MathKeisan 2.1.0 for SX Release Notes

August 2007
NEC Corporation

New in MathKeisan 2.1.0

BLAS routines SNRM2, DNRM2, SCNRM2, DZNRM2 fixed to not underflow on tiny arguments.
BLAS routines CGEMM, ZGEMM improved in performance for even LDA.
BLAS routines SGEMV, DGEMV improved in performance for large |m-n|.
LAPACK library updated to version 3.1.1.
LAPACK library enhanced to avoid bank conflicts on work arrays.
ScaLAPACK library updated to version 1.8.0.
PARPACK library added. It is an MPI parallel version of ARPACK.
Performance improvements for arguments in certain ranges for the FFT subroutines:
- CFFT2, ZFFT2
- CCFFT2D, ZZFFT2D
- CFFT2D, ZFFT2D
- CCFFT3D, ZZFFT3D
- CFFT3D, ZFFT3D
- CCFFTM, ZZFFTM
- MCFFT, MZFFT
- CFFTMLT, ZFFTMLT
- S2DFFT, D2DFFT, C2DFFT, Z2DFFT
- S3DFFT, D3DFFT, C3DFFT, Z3DFFT
- SFFTS, DFFTS, CFFTS, ZFFTS
Size of work arrays reduced for FFT subroutines: ZZFFT3D, CCFFT3D, ZFFT3D, CFFT3D.
Man pages have been added to www.mathkeisan.com.
The English version of the User's Guide is re-arranged, and more information is added on FFT and Solver.

The MathKeisan 2.1.0 distribution for SX contains these files:

For information about using MathKeisan, please see the MathKeisan User's Guide (also available at www.mathkeisan.com).

If you have any questions or feedback about MathKeisan, please contact us at www.mathkeisan.com.