SuiteSparseQR Copyright (c) 2008-2016, Timothy A. Davis,
GPU modules Copyright (c) 2015, Timothy A. Davis, Sencer Nuri Yeralan,
and Sanjay Ranka.
http://www.suitesparse.com

SuiteSparseQR is a multithreaded, multifrontal, rank-revealing sparse QR
factorization method, with optional GPU acceleration using NVIDIA GPUs.

Version 2.0.1 of SuiteSparseQR is an major release, with support for GPU
computing.  See SPQR/Demo/qrdemo_gpu.cpp, and do 'make gpu' in SPQR/Demo to
test it (compare your output with SPQR/Demo/qrdemo_gpu.out.  For more
extensive tests, see SPQR/Demo/go*.m.

NOTE:  you may get an error about a C++-style commment in the C header file,
/usr/local/cude/include/cuComplex.h.  This is a bug in NVIDIA CUDA 7.0.  If
you encounter this error, then either edit the file and delete all lines
beginning with "//", or upgrade to CUDA 7.5.

QUICK START FOR MATLAB USERS (on Windows, Linux, Solaris, or the Mac OS): To
compile and test the MATLAB mexFunctions, do this in the MATLAB command window:

    cd SuiteSparse/SPQR/MATLAB
    spqr_install
    spqr_demo

FOR MORE DETAILS: please see the User Guide in Doc/spqr_user_guide.pdf.

FOR LINUX/UNIX/Mac USERS who want to use the C++ callable library:

    To compile the C++ library and run a short demo, just type "make" in
        the Unix shell.  If you have an NVIDIA GPU, this also compiles
        the GPU accelerated part of SPQR.

    To compile the SuiteSparseQR C++ library, in the Unix shell, do:

        cd Lib ; make

    To compile and test an exhaustive test, edit the Tcov/Makefile to select
    the LAPACK and BLAS libraries, and then do (in the Unix shell):

        cd Tcov ; make

    Compilation options in SuiteSparse_config/SuiteSparse_config.mk,
    SPQR/*/Makefile, or SPQR/MATLAB/spqr_make.m:

        -DNPARTITION    to compile without METIS (default is to use METIS)

        -DNEXPERT       to compile without the min 2-norm solution option
                        (default is to include the Expert routines)

        -DHAVE_TBB      to compile with Intel's Threading Building Blocks
                        (default is to not use Intel TBB)

        -DTIMING        to compile with timing and exact flop counts enabled
                        (default is to not compile with timing and flop counts)


See SPQR/Doc/License.txt for the license.