CHANGES

Changelog for BOUT++

0.35 (beta) 3/9/2008
	- Layout altered so BOUT++ now a library which is linked to
	  the physics modules (see example directory)
	- Configuration system modified to use autoconf. Can fall back
 	  on manual modification of make.conf if needed though.
	- Submitted to CVS archive on NERSC Jacquard (again) for further
	  development
	- Parallelisation in X now produces the same result as serial
	  (using SPT algorithm for Laplacian inversion). Scaling to
	  512 processors tested on Franklin for small problem
	  (~190x speedup)
	* Still getting occasional segfaults for NXPE > 1
	* Strange segfault on MPI_Bcast on LLNL grendel. Related?

20/8/2008: 
	- Implemented option TwistOrder = 0. If set, uses the same
	  method as ShiftOrder (i.e. usually FFT). This is done
	  in receive() in communicator.cpp. Modified calc_stencil
	  in stencils.cpp
	- Fixed (i think) small bug in the range of twist-shifting
	  e.g. MXG+DDATA_XSPLIT -> DDATA_XSPLIT. This because DDATA_XSPLIT
	  is in processor domain index, not global index. Shouldn't make
	  any difference for current runs (with XSPLIT = 0)

0.3 (July-Aug 2008) release 7th Aug 2008
	Improvements concentrating on improving parallel performance
	- Domain decomposition now in 2D (X and Y).
	  Required changes to communications, boundary conditions, 
	  solver, topology and Laplacian inversion.
	- nx, MX, MXSUB, ngx now all different
	- "Extra" redundant point in Z no longer evolved
	   (still in data classes for now though)
	- X,Y boundary regions only evolved (added to CVODE)
	  if an actual boundary, not if guard cells.
	- New virtual function interface (FieldData) added to
	  scalar & vector classes.
	  Will simplify communicator, solver and datafile classes.
	  Currently only used by communicator class
	- Communicator changed to use ISend, rather than Send
	  depending on input option "async" in section "comms"
	- Communicator now uses MPI_Waitany rather than MPI_Wait
	  to receive data in any order and process asap
	- NPVT removed: Restriction on processor # in inner and outer legs
	- Laplacian inversion now done in parallel using PDD algorithm

	- New version of datafile almost finished (in new_src)
	  - Uses FieldData interface to variables
	  - re-written as a template. Can more easily be adapted
	    to different file formats.
	  - PDB class written to wrap around C library

0.2 ( Nov 2007 - Feb 2008 ):
	- Many bug-fixes
	- Added shifted radial coordinates
	- More efficient (reference counting) data objects
	- Changed FFT routine to use FHT method rather than NR's code
	  Now only solves for +ve frequencies (1/2 the work previously) 

0.1 ( ):
	- Initial version