Templated function for sparse and dense matrices in RcppArmadillo - c++

Question

I'm trying to define a templated function that can handle both sparse and dense matrix inputs using RcppArmadillo. I got the very simple case of sending a dense or sparse matrix to C++ and back to R to work like this:
library(inline); library(Rcpp); library(RcppArmadillo)
sourceCpp(code = "
// [[Rcpp::depends(RcppArmadillo)]]
#include <RcppArmadillo.h>
using namespace Rcpp ;
using namespace arma ;
template <typename T> T importexport_template(const T X) {
T ret = X ;
return ret ;
};
//[[Rcpp::export]]
SEXP importexport(SEXP X) {
return wrap( importexport_template(X) ) ;
}")
library(Matrix)
X <- diag(3)
X_sp <- as(X, "dgCMatrix")
importexport(X)
## [,1] [,2] [,3]
##[1,] 1 0 0
##[2,] 0 1 0
##[3,] 0 0 1
importexport(X_sp)
##3 x 3 sparse Matrix of class "dgCMatrix"
##
##[1,] 1 . .
##[2,] . 1 .
##[3,] . . 1
and I interpret that to mean that the templating basically works (i.e., a dense R-matrix gets turned into a arma::mat, while a sparse R-matrix gets turned into a arma::sp_mat-object by the implicit calls to Rcpp::as, and the corresponding impliict Rcpp:wraps then do the right thing as well and return dense for dense and sparse for sparse).
The actual function I try to write needs multiple arguments of course, and that's where I fail -- doing something like:
sourceCpp(code ="
// [[Rcpp::depends(RcppArmadillo)]]
#include <RcppArmadillo.h>
using namespace Rcpp ;
using namespace arma ;
template <typename T> T scalarmult_template(const T X, double scale) {
T ret = X * scale;
return ret;
};
//[[Rcpp::export]]
SEXP scalarmult(SEXP X, double scale) {
return wrap(scalarmult_template(X, scale) ) ;
}")
fails because the compiler doesn't know how to resolve * at compile time for SEXPREC* const.
So I guess I need something like the switch-statement in this Rcpp Gallery snippet to properly dispatch to specific template functions, but I don't know how to write that for types that seem more complicated than INTSXP etc.
I think I know how to access the type I would need for such a switch statement, e.g.:
sourceCpp(code ="
// [[Rcpp::depends(RcppArmadillo)]]
#include <RcppArmadillo.h>
using namespace Rcpp ;
using namespace arma ;
//[[Rcpp::export]]
SEXP printtype(SEXP Xr) {
Rcpp::Rcout << TYPEOF(Xr) << std::endl ;
return R_NilValue;
}")
printtype(X)
##14
##NULL
printtype(X_sp)
##25
##NULL
but I don't understand how to proceed from there. What would a version of scalarmult_template that works for sparse and dense matrices look like?

Answer 1

Answering my own question based on #KevinUshey's comment. I do matrix multiplication for 3 cases: dense-dense, sparse-dense, and "indMatrix"-dense:
library(inline)
library(Rcpp)
library(RcppArmadillo)
library(Matrix)
library(rbenchmark)
sourceCpp(code="
// [[Rcpp::depends(RcppArmadillo)]]
#include <RcppArmadillo.h>
using namespace Rcpp ;
using namespace arma ;
arma::mat matmult_sp(const arma::sp_mat X, const arma::mat Y){
arma::mat ret = X * Y;
return ret;
};
arma::mat matmult_dense(const arma::mat X, const arma::mat Y){
arma::mat ret = X * Y;
return ret;
};
arma::mat matmult_ind(const SEXP Xr, const arma::mat Y){
// pre-multplication with index matrix is a permutation of Y's rows:
S4 X(Xr);
arma::uvec perm = X.slot("perm");
arma::mat ret = Y.rows(perm - 1);
return ret;
};
//[[Rcpp::export]]
arma::mat matmult_cpp(SEXP Xr, const arma::mat Y) {
if (Rf_isS4(Xr)) {
if(Rf_inherits(Xr, "dgCMatrix")) {
return matmult_sp(as<arma::sp_mat>(Xr), Y) ;
} ;
if(Rf_inherits(Xr, "indMatrix")) {
return matmult_ind(Xr, Y) ;
} ;
stop("unknown class of Xr") ;
} else {
return matmult_dense(as<arma::mat>(Xr), Y) ;
}
}")
n <- 10000
d <- 20
p <- 30
X <- matrix(rnorm(n*d), n, d)
X_sp <- as(diag(n)[,1:d], "dgCMatrix")
X_ind <- as(sample(1:d, n, rep=TRUE), "indMatrix")
Y <- matrix(1:(d*p), d, p)
matmult_cpp(as(X_ind, "ngTMatrix"), Y)
## Error: unknown class of Xr
all.equal(X%*%Y, matmult_cpp(X, Y))
## [1] TRUE
all.equal(as.vector(X_sp%*%Y),
as.vector(matmult_cpp(X_sp, Y)))
## [1] TRUE
all.equal(X_ind%*%Y, matmult_cpp(X_ind, Y))
## [1] TRUE
EDIT: This has been turned into an Rcpp Gallery post.