table1

gistfile1.rebol

library(Rcpp)
library(microbenchmark)

cppFunction('
  IntegerVector table1(const CharacterVector x) {
    std::map<std::string, int> counts;
    std::vector<std::string> vec = as<std::vector<std::string> >(x);
  
    int n = x.length();
      for (int i = 0; i < n; i++) {
      counts[vec[i]]++;
    }
  
    // Loop through each element of map and output into named vector
    IntegerVector out(counts.size());
    CharacterVector names(counts.size());
  
    std::map<std::string, int>::iterator it;
    int i;
    for (i = 0, it = counts.begin(); it != counts.end(); i++, it++) {
      names[i] = it->first;
      out[i] = it->second;
    }
    out.attr("names") = names;
    return(out);
  }
')

# Take advantage of Rcpp auto-magic as/wrap for parameters and return
cppFunction('
  std::map<std::string, int> table2(const std::vector<std::string>& x) {
    std::map<std::string, int> counts;
    for (int i = 0; i < x.size(); i++) {
      counts[x[i]]++;
    }  
    return counts;
  }
')

# Eliminate the copying of inbound strings
cppFunction('
  std::map<std::string, int> table3(CharacterVector x) {
    std::map<std::string, int> counts;
    for (int i = 0; i < x.size(); i++) {
      const char* name = x[i];
      counts[name]++;
    }  
    return counts;
  }
')

# Eliminate map insertion overhead by using char* as key
cppFunction('
  std::map<std::string, int> table4(CharacterVector x) {
    
    // perform the count
    std::map<const char*, int> counts;
    for (int i = 0; i < x.size(); i++) {
      const char* name = x[i];
      counts[name]++;
    }  
  
    // creating a new map keyed by std::string
    std::map<std::string,int> result;
    for (std::map<const char*, int>::const_iterator 
          it = counts.begin(); it != counts.end(); ++it) {
      result[it->first] = result[it->first] + it->second;
    }
    return result;
  }
')

x <- sample(letters, 1e4, rep = T)
microbenchmark(
  table(x),
  table1(x),
  table2(x),
  table3(x),
  table4(x)
)