问题
is there a package for matrix inversion in R using parallel computation?
Thanks!
Hello. I am having trouble installing the HiPLARb package, here is what I did:
Download auto-installer script: http://www.hiplar.org/downloads/HiPLARb.Installer
./HiPLARb.Installer --with-openblas --no-gpu --prefix=/home/kaiyin/mylib
everything went ok, all prerequesite libraries installed succesully, also the patched version of R 2.15.2
Download the package from http://www.hiplar.org/downloads/HiPLARb_0.1.3.tar.gz
Modify the shell PATH to prepend the patched R bin directory.
R CMD INSTALL --configure-args="--with-lapack= \
-L/home/kaiyin/mylib/lib\ -lopenblas \
--with-plasma-lib=/home/kaiyin/mylib \
--with-magma-lib=/home/kaiyin/mylib" HiPLARb_0.1.3.tar.gz
I got the following error:
configure: error: unrecognized option: `-L/home/kaiyin/mylib/lib -lopenblas'
Try `./configure --help' for more information
ERROR: configuration failed for package HiPLARb
Please help, thanks!
Since -lopenblas is not a valid configure option, I removed it and retried:
R CMD INSTALL --configure-args="--with-lapack=/home/kaiyin/mylib/lib\
--with-plasma-lib=/home/kaiyin/mylib \
--with-magma-lib=/home/kaiyin/mylib" ../HiPLARb_0.1.3.tar.gz
This time a new error:
configure: error: invalid variable name: ` --with-magma-lib'
3rd try:
R CMD INSTALL --configure-args="--with-lapack=/home/kaiyin/mylib/lib\
--with-plasma\
--with-magma\
--with-plasma-lib=/home/kaiyin/mylib \
--with-magma-lib=/home/kaiyin/mylib" ../HiPLARb_0.1.3.tar.gz
configure: WARNING: you should use --build, --host, --target
configure: WARNING: invalid host type:
checking "if PLASMA libraries exist"... configure: WARNING: "No PLASMA option given."
configure: WARNING: "No CUDA option given."
configure: WARNING: "No MAGMA option given."
configure: error: Cannot find MAGMA or PLASMA Libraries. Please install MAGMA, PLASMA or both. Please see the install guide for more details.
回答1:
Yes, several; but it depends on what you mean by parallel.
If you mean very large scale parallelism, like won't fit on a single machine and you need a cluster to deal with it, then there are the PBD packages. These require packages like MPI and Scalapack t be installed, but they will typically be available on any cluster used for big technical computing already.
If the matrix will fit on a single node and you just want to throw all of your CPUs at it, the HiPLAR package can make use of multiple cpus or even GPUs based on the PLASMA or MAGMA libraries (which again need to be installed).
More generally, the CRAN HPC task view is a useful listing of these sorts of packages.
回答2:
in order to parallel computation in R there are several packages
give a look of this PDF : Getting the most out of your CPUs Parallel computing strategies in R
here a website High-Performance and Parallel Computing with R where are report several packages about parallel computing in R
来源：https://stackoverflow.com/questions/15508811/is-there-a-package-for-parallel-matrix-inversion-in-r