relax SVN archive

Merged revisions 15440-15445 via svnmerge from
svn+ssh://bugman@.../svn/relax/1.3

........
r15440 | bugman | 2012-03-06 14:32:08 +0100 (Tue, 06 Mar 2012) | 11 lines

Bug fix for relax not starting on 32-bit machines.

This was first reported by Hugh Dannatt <h dott dannatt att sheffield dot ac dot uk> in the post at
https://mail.gna.org/public/relax-users/2012-03/msg00011.html.

The numpy.float128 object appears to be missing on 32-bit numpy versions (at least on old numpy
versions). In such cases, float128 is aliased to float64. This should only affect relax users who
place special high precision (float128) objects into the relax data store, when loading a save file
with such an object on a 32-bit machine. This should practically never happen!
........
r15441 | bugman | 2012-03-06 15:15:10 +0100 (Tue, 06 Mar 2012) | 5 lines

Expansion and improvement of the information printed out by 'relax --info'.

A number of details are now taken from the sys module.
........
r15442 | bugman | 2012-03-06 15:48:22 +0100 (Tue, 06 Mar 2012) | 3 lines

For Unix and GNU/Linux systems, the relax info printout now shows the ram and swap size.
........
r15443 | bugman | 2012-03-06 15:57:04 +0100 (Tue, 06 Mar 2012) | 3 lines

Updated the value.set user function unknown parameter error message to list the known ones.
........
r15444 | bugman | 2012-03-06 17:11:02 +0100 (Tue, 06 Mar 2012) | 3 lines

Added support for the memory size on MS Windows to the relax info print out.
........
r15445 | bugman | 2012-03-06 18:29:47 +0100 (Tue, 06 Mar 2012) | 3 lines

Updated the multi-processor package __all__ list.
........

Updated the multi-processor package __all__ list.

Added support for the memory size on MS Windows to the relax info print out.

Updated the value.set user function unknown parameter error message to list the known ones.

For Unix and GNU/Linux systems, the relax info printout now shows the ram and swap size.

Expansion and improvement of the information printed out by 'relax --info'.

A number of details are now taken from the sys module.

Bug fix for relax not starting on 32-bit machines.

This was first reported by Hugh Dannatt <h dott dannatt att sheffield dot ac dot uk> in the post at
https://mail.gna.org/public/relax-users/2012-03/msg00011.html.

The numpy.float128 object appears to be missing on 32-bit numpy versions (at least on old numpy
versions). In such cases, float128 is aliased to float64. This should only affect relax users who
place special high precision (float128) objects into the relax data store, when loading a save file
with such an object on a 32-bit machine. This should practically never happen!

Full implementation of the pseudo-ellipse frame order target function for multi-processor execution.

Unfortunately using 1e5 integration points for the system test, on a dual core machine the average
time for the different fabrics for 3 different runs are:

- Uni-processor fabric: 23.297s
- Mpi4py fabric with 2 slaves: 23.339s

Increased the pseudo-ellipse frame order system test integration points from 50 to 1e3.

This increases the system test time by 0.15 seconds on one test machine, while increasing the number
of integrated points within the cone from 1 to 23. This will be more useful for debugging the
multi-processor code.

Initial parallelisation of the pseudo-ellipse frame order target function for the multi-processor.

This currently functions only for the uni-processor fabric, more work on the data structures within
the memo is required before the mpi4py fabric is functional. A number of classes and functions have
been added:

- The Slave_command_pcs_pseudo_ellipse_qrint class: the pcs_numeric_int_pseudo_ellipse_qrint()
function has been split up so that its numerical integration code is within this class.
- The Memo_pcs_pseudo_ellipse_qrint class: Currently being used to hold the number of
integration points.
- The Result_command_pcs_pseudo_ellipse_qrint class: Is currently being used to transfer the
number of integration points into the memo.
- The subdivide() function: Used to split up the points into blocks for each slave processor.
It is currently broken for >1 slaves.

These changes do not increase the Frame_order.test_cam_pseudo_ellipse system test timings.