ENH: we still have performance warning for our use of outer product

[mathurinm]

the tests on travis still get a lot of:

  /home/travis/build/QB3/CLaR/clar/solvers.py:387: NumbaPerformanceWarning: '@' is faster on contiguous arrays, called on (array(float64, 2d, A), array(float64, 2d, A))
    S_inv_R -= S_inv_X[:, j:j+1] @ B[j:j+1, :]

@QB3 I can't remember, can we fix this on the numba side? Or, given you experience with mne, do you think we can remove the numba in this part and use the dgemm routine for rank one matrix addition ?

[QB3]

1- on my laptoop this error only appear when running pytest, not when running scripts themselves
2- I think indeed that dgemm would be faster, however since we need to jit this part of the code, I do not know if dgemm is compatible with numba

[mathurinm]

The warning appears on the doc https://qb3.github.io/CLaR/auto_examples/plot_covariance_estimate.html#sphx-glr-auto-examples-plot-covariance-estimate-py @QB3