rcpp matrix multiplication speed improvement - performance

Question

I am using Rcpp for some heavy computation. However, my code doesn't run as fast as expected. I know I must have done something wrong because the same computation is faster in MATLAB. I have a hard time figuring out how I could improve my code, so I'm genuinely asking for help here.
I have a 12 by 5000^2/2 matrix called mX, and a 5000^2/2 by 1 matrix called w.
I would like to calculate A=mX * diag(w) * t(mX).
I have written the following four functions to do it.
#include <RcppArmadillo.h>
#include <RcppEigen.h>
#include <omp.h>
// [[Rcpp::depends("RcppArmadillo")]]
// [[Rcpp::depends("RcppEigen")]]
// [[Rcpp::plugins(openmp)]]
using namespace Rcpp;
using Eigen::MatrixXd;
// [[Rcpp::export]]
arma::mat MultiplyArma1(arma::mat mX, arma::colvec w){
return mX*arma::diagmat(w)*arma::trans(mX);
}
// [[Rcpp::export]]
arma::mat MultiplyArma2(arma::mat mX, arma::colvec w){
return mX*(arma::repmat(w,1,12)%arma::trans(mX));
}
// [[Rcpp::export]]
arma::mat MultiplyCpp(arma::mat mX, arma::colvec w){
omp_set_num_threads(4);
int N=mX.n_rows;
int K=mX.n_cols;
arma::mat tmX=arma::zeros<arma::mat>(K,N);
arma::mat ProductF=arma::zeros<arma::mat>(N,N);
arma::mat ProductM=arma::zeros<arma::mat>(K,N);
#pragma omp parallel for schedule(static)
for (int i=0; i<K; ++i){
for (int j=0; j<N; ++j){
tmX(i,j)=mX(j,i);
}
}
#pragma omp parallel for schedule(static)
for (int j=0; j<N; ++j){
for (int i=0; i<K; ++i){
ProductM(i,j)=tmX(i,j)*w(i);
}
}
#pragma omp parallel for schedule(static)
for (int j=0; j<N; ++j){
for (int i=0; i<N; ++i){
for (int k=0; k<K; ++k){
ProductF(i,j) +=tmX(k,i)*ProductM(k,i);
}
}
}
return ProductF;
}
// [[Rcpp::export]]
MatrixXd MultiplyEigen(const MatrixXd mX, const MatrixXd Dw){
return mX* ( Dw.cwiseProduct( mX.transpose() ) );
}
The 1st and 2nd both use RcppArmadillo, and only differs in representation of diag(w). The 3rd makes use of openmp, and I code up everthing element wise. The 4th uses RcppEigen. I don't know how to generate diag(w) in RcppEigen, so I have the input be Dw=rep(w,1,12).
mX=matrix(0.7, 12, 5000^2/2)
w=matrix(0.2, 5000^2/2,1)
Dw=repmat(w, 1,12)
tic()
a1=MultiplyArma1(mX,w)
toc()
tic()
a2=MultiplyArma2(mX,w)
toc()
tic()
a3=MultiplyCpp(mX,w)
toc()
tic()
a4=MultiplyEigen(mX, Dw)
toc()
Then as I time the speed, They all take more than 3 seconds. However, when I do the same calculation in MATLAB A=mX*(repmat(w,1,12).*mX'), it takes only 1.5 seconds.
It makes me feel that there's still room for improving my Rcpp function, but I honestly don't know how. I would very much appreciate your help!