performance - Eigen matrix/array compute distances efficiently -

i need create eigen array distances between each source , recorder position.

i have 3 eigen::array sx, sy , sz denoting source positions , 3 eigen::array rx, ry , rz denoting recorder positions. length of source , recording position arrays not necessary same.

using for-loops can calculate distance matrix follow

eigen::arrayxf  sx, sy, sz; eigen::arrayxf  rx, ry, rz; eigen::arrayxxf dist;  (int s=0; s<nrsources; s++ ) {     (int r=0; r<nrreceivers; r++) {        dist(h,g) = sqrt(pow(rx(h)-sx(g),2.)+                        pow(ry(h)-sy(g),2.)+                        pow(sz(h)-sz(g),2.));    } } 

i need compute dist array 500 x 1000 times experiment. using for-loop obvious works not efficient method since not make use of vectorization.

rewriting sx, sx, sz , hx, hy , hz sxyz , hxyz arrays should possible.

is possible write equation more efficiently!?

you want discard inner loop in favor of expression. doing so, allow calculations lumped , enable vectorization. assuming r , s variables declared eigen::arrayx3f sxyz(nrsources,3), rxyz(nrreceivers,3);, write outer loop as:

for (int s = 0; s < nrsources; s++) {     dist.col(s) =         (rxyz.rowwise() - sxyz.row(s)).matrix().rowwise().norm(); }        

here use rxyz.rowwise() - sxyz.row(s) subtract s'th s rs. matrix() needed give access norm().

benchmark , compare implementation.