relax SVN archive

Merged revisions 15455 via svnmerge from
svn+ssh://bugman@.../svn/relax/1.3

........
r15455 | bugman | 2012-03-07 13:42:05 +0100 (Wed, 07 Mar 2012) | 3 lines

More important import fixes for the multi-processor package.
........

More important import fixes for the multi-processor package.

Merged revisions 15453 via svnmerge from
svn+ssh://bugman@.../svn/relax/1.3

........
r15453 | bugman | 2012-03-07 13:39:06 +0100 (Wed, 07 Mar 2012) | 3 lines

Important import fix!
........

Important import fix!

Shifted the transfer of much of the data to the slave command to the __init__() method.

This should improve the multi-processor scaling by reducing data transfer times per function call.
However, from testing this currently does nothing! The structures now pre-transfered include:

full_in_ref_frame
r_ln_pivot
A
Ri_prime
pcs_theta
pcs_theta_err
missing_pcs

Shifted the Sobol' point subdivision from the target function to the __init__() method.

This should speed up the target function. The data transfer to the slave also occurs at the
__init__() method level, reducing data transfers per function call.

Shifted the subdivide() fn from frame_order_matrix_ops to the frame order target function class.

Shifted the slave command setup from the target function to the frame order __init__() method.

This is in preparation for transfer of data to the slaves prior to the target function calls.

The Processor_box is now initialised in the maths_fns.frame_order.__init__() method.

This is to shift code out of the target functions.

The pcs_numeric_int_pseudo_ellipse_qrint() function has been absorbed into the target function.

With the multi-processor changes, the calculation code from pcs_numeric_int_pseudo_ellipse_qrint is
now in the slave command class. This is also in preparation for reductions in data transfer in the
MPI protocol which is destroying the scaling ability of the parallelised code.