Rcpp performance problems only on Windows using R 4.0.2

[goralBill]

I have a small rcpp function and associated R wrapper function that improved upon the performance of the built-in rowSums function when running under R 3.x. Since moving to R 4.x, it is dramatically slower than the built-in function when running in RStudio under Windows 10 Professional. I contacted Dirk Eddelbuettel (the principal maintainer of Rcpp) via GitHub assuming that there was an issue with R 4.x and Rcpp. He was able to demonstrate that the problem did not exist when running on R 4.x under Linux or MAC OS10. I am left to assume that this is a Windows (or possibly RTools 4) or RStudio problem. If anyone has experienced this issue, I would appreciate some guidance. Test code and results below

Running in my RStudio:

> x = matrix(data = rnorm(1e8), ncol = 1000)

> system.time(rowSums(x))
   user  system elapsed 
   0.23    0.00    0.24 

> system.time(rowSumsA(x))
   user  system elapsed 
   1.74    0.00    1.74

Running on Ubuntu 20.04

> x <- matrix(data = rnorm(1e8), ncol = 1000)

> system.time(rowSums(x))
   user  system elapsed 
  0.167   0.000   0.167

> system.time(rowSumsA(x))
   user  system elapsed
  0.053   0.000   0.053

Code:

#include <Rcpp.h>

// [[Rcpp::export]]
SEXP rowSumsACPP(SEXP x, SEXP na_rm) {
  // Assumes that R wrapper function will only pass in a matrix
  using namespace Rcpp;
  
  NumericMatrix xin(x);
  
  long nRows = xin.nrow();
  long nCols = xin.ncol();
  bool removeNA = as<bool>(na_rm);  
  NumericVector xout = NumericVector(nRows);
  
  if(removeNA) {
    for(long c = 0; c < nCols; c++) {
      for(long r = 0; r < nRows; r++) { 
        // Only non-NA contents of xin will equal itself
        if(xin(r, c) == xin(r, c)) {
          xout[r] += xin(r, c);
        }
      }
    }
  } else {
    for(long c = 0; c < nCols; c++) {
      for(long r = 0; r < nRows; r++) {  
        xout[r] += xin(r, c);
      }
    }
  }
  return(xout);
}


/*** R
rowSumsA <- function(x, na.rm = FALSE) {
  if(mode(na.rm) != "logical") {
    stop("na.rm parameter must be of type logical")
  }
  
  if(any(class(x) == "data.frame")) {
    return(rowSums(x, na.rm))
  } 
  
  if(is.null(dim(x))) {
    # This case is assumed to be a vector, just return it unchanged
    xout = x
  } else {
    if(!(mode(x) %in% c("numeric", "logical"))) {
      stop("x parameter is not numeric or logical")
    }
    # This case is assumed to be a zoo or matrix
    # Rcpp will convert zoo to matrix during assignment, so we don't need to explicitly handle	
    xout = rowSumsACPP(x, na.rm)
  }
  return(xout)	
}
*/