RRQR-MEX a MATLAB mex-interface for the rank revealing QR factorization

RRQR-MEX provides a MATLAB routine rrqr, implementing an interface to the FORTRAN RRQR codes by G. Quintana-Orti und C.H. Bischof.

THIS IS PROVIDED FOR REPRODUCIBILITY ONLY.


In recent Matlab on modern multicore CPUs we recommend qr(.,0) or svd based solutions. This is similarly robust, but usually much faster due to the lack of parallelism in the original Fortran codes.

RRQR-MEX provides a MATLAB routine rrqr, implementing an interface to the FORTRAN RRQR codes by G. Quintana-Orti und C.H. Bischof. The binary MEX-files for Windows 32 bit and Linux 32 bit have been compiled without processor optimizations and linked statically against the reference implementations of BLAS ans LAPACK. Therefore they should run on any MATLAB installation later than R14. More recent MATLAB versions support dynamical linking against special MATLAB LAPACK and BLAS libraries, which can improve the performance. Also processor specific compilation can increase the efficiency of the MEX-files. In those cases, or on other platforms one might wish to compile the MEX-files directly from source.

Downloads

• RRQR-MEX Win32 Windows 32 bit MEX-file plus M-File and users guide.
  (We have received reports, that this version does not work in Matlab r2008a. Try this version in that case.)
• RRQR-MEX Linux 32 bit Linux 32 bit MEX-file plus M-File and users guide.
• RRQR-MEX Linux 64 bit Linux 64 bit MEX-file plus M-File and users guide.
• RRQR-MEX Source the source code for optimized compilation and compilation on other platforms.
• RRQR-MEX users guide
• all binary RRQR-MEX files plus M-File and users guide.
• Reimplementation of the Gateway-function as C-Mex-Function. The source code for optimized compilation.
• RRQR-MEX Linux 32 bit and 64 bit pre-compiled based on the C-Reimplementation.

Due to missing appropriate hardware we can not provide other binary files.

The sources for the RRQR library needed for the compilation from source can be found here.

An updated version with a bit more comfortable Makefiles can be found here.

Related Software

Based on the RRQR Peter Benner implemented a rank revealing tridiagonalization that can be found here.

References

[BisQO98]

C.H. Bischof und G. Quintana-Orti;
Codes for rank-revealing QR factorizations of dense matrices.
ACM Trans. Math. Softw. vol. 24 pp.254-257; 1998

Similar Projects

At Matlab Central alternative RRQR mex-files for Windows are available. http://www.mathworks.com/matlabcentral/fileexchange/18591 Ivo Houtzager bases his gateway functions on the C-mex interface. Thus he avoids problems with the currently unsupported Fortran compilers in Matlab.