• computation of global intermediate trends (instead of x, y, z)
• improvement of nonosc (MIn/MAX search)

This induce computation changes

Improvement of the Conjugate Gradient solver in MPI case. Method : gather the 2 scalary products in one to decrease latency

This changes computation results when using PCG, but does not change the number of iterations.

For algorithm description, see : D’Azevedo, E. F., Eijkhout, V. L., and Romine, C. H. 1993. Conjugate gradient algorithms with reduced synchronization overhead on distributed memory multiprocessors. Computer Science Technical Report CS-93-185, University of Tennessee, Knoxville.

And for additional information : Dongarra, J. and Eijkhout, V. 2003.

Finite-choice algorithm optimization in conjugate gradients. Computer Science

Technical Report UT-CS-03-502, University of Tennessee, Knoxville.

Change averaging window in dynspg_ts. Now it's done between [1/2 rdt, 3/2 rdt] instead of [1, 2 rdt ]