SUPERLU support about rcpparmadillo HOT 7 CLOSED

@ivan-krukov After enabling ARMA_USE_SUPERLU, you also need to pass the -lsuperlu option to R.

#include <RcppArmadillo.h>
// [[Rcpp::depends(RcppArmadillo)]]

using namespace Rcpp ;
using namespace arma;

// [[Rcpp::export]]
void test () {
  sp_mat A = sprandu<sp_mat>(1000, 1000, 0.1);

  vec b = randu<vec>(1000);
  mat B = randu<mat>(1000, 5);

  vec x = spsolve(A, b);  // solve one system
  mat X = spsolve(A, B);  // solve several systems

  bool status = spsolve(x, A, b);  // use default solver
  if(status == false)  { cout << "no solution" << endl; }

  spsolve(x, A, b, "lapack");   // use LAPACK  solver
  spsolve(x, A, b, "superlu");  // use SuperLU solver

}

In R:

> Sys.setenv("PKG_LIBS"="-lsuperlu")
> Rcpp::sourceCpp("test.cpp")

If you didn't pass -lsuperlu, you might get some errors like undefined symbol: set_default_options.

from rcpparmadillo.

I wrote this up as a new piece on the Rcpp Gallery.

from rcpparmadillo.

Good question. I am not sure.

Context, as you may know: RcppArmadillo takes a plain copy, no cmake, no options. Works great with basic BLAS and LAPACK which we get from R.

Extensions are harder. We are learning that now with MLPACK (over here) which uses linking against external Boost (as it needs serialization) and Armadillo (as that is how it is build).

You could try something narrower and half way. Ie define the required tokens in your source file (or src/Makevars), then load RcppArmadillo to have them take effect. Then you "only" need to make sure you supply SUPERLU itself.

There are a few people interested in better sparse support, maybe time to discuss on rcpp-devel, or here.

from rcpparmadillo.

Just checked, and Conrad's documentation leans the same way:

ARMA_USE_SUPERLU Enable use of SuperLU, which is used by spsolve() for finding the solutions of sparse systems; you will need to link with the superlu library, for example -lsuperlu
Caveat: Armadillo 6.x requires SuperLU 4.3, while Armadillo 7.x requires SuperLU 5.2

from rcpparmadillo.

Perfect, was just thinking I should write an Rcpp Gallery post along those line.

The default libsuperlu-dev on Ubuntu is currenty behind, the one in Debian is good enough so easy enough to test in a Docker container, or a locally rebuilt packager for superlu.

from rcpparmadillo.

It should all be headers ... until the call to superlu is made at which point object code is needed.

Not any different from when LAPACK is called, which get for free from R which is why it seems so seamless.

from rcpparmadillo.

Just checked everything works, no need to recompile RcppArmadillo.

This is excellent. Thank you!

from rcpparmadillo.