[Numpy-discussion] how to vectorize a row of eigenvalue computations ?

[LB]

I've got an array S of shape (N, 6) with N >> 100000 containing the
six components of a stress field given in N points.

I need to make a lot of computation of principal stresses, which are
in fact the eigenvalues of the stress tensors.

I'm using the basic code described below :

import numpy as np

def calc_principal_stresses(S):
    """ Return the principal stress corresponding to the tensor S.
    S is interpreted as an array containing [Sx, Sy, Sz, Sxy, Syz,
Szx]

    Return array([S3, S2, S1])
    """
    p_stresses = np.linalg.eigvalsh( np.array(
	[  [ S[0], S[3], S[5]],
           [ S[3], S[1], S[4]],
           [ S[5], S[4], S[2]],
	]))

    return p_stresses.sort()

p_stresses = array([ calc_principal_stresses(s) for s in S])

Aside putting the sort function outside the loop, is there any way to
optimize or vectorize this kind of operation ?

  Regards,

--
LB