The expression norm 'normalizes' the incoming matrix of the left entrance of the jit.expr object. In other words, if you have a matrix, say created by jit.noise 1 float32 3, and you run that through jit.expr @expr norm, the values will be 0. 0.5 and 1.
Try experimenting with small matrices and jit.print to see the result of the operations in the max window.
A small example (watch the values in the max window):