ChangeLog

Overview of changes

Approximately most to least significant

SEE EOF for changes to IDL routines needed....................


Makefile

  Added makefile for compilation
  
    Type "make" to compile
    Faster recompilation if change only some files
    Allows to compile in parallel if wanted
    'make clean' removes module files etc

File format 

  Changed to unformatted. Use readu in IDL
  (Note: if running on IBM architecture may have to provide switch to IDL for byte ordering)

    Reduces size by around factor 3
    Allows use of mpiIO routines to write from processors individually.
    Reduces memory requirement on root processor by factor n where n is number of processors
    Speeds up IO by factor propto n


MPI communication

  Blocking send/recieve changed to MPI_SENDRECV which handles both ends of communication
    MPI communication no longer serialised
    Speeds up communication by factor propto n


Whole domain variables

  Variables no longer exist across the entire space domain on root processor
  All definitions and evolution are done on individual processor's sections
    Memory requirement on root reduced by factor n
    Speed-up in initialization propto factor n
  

Subroutine calls
  
  No longer calls each routine n_xp times on array subsections, instead calls once on whole array
    Array subsection passing is slow process requiring making a copy of relevant part
    Speeds up calls by factor unknown


FORALL loops

  Returned to DO loops rather than FORALL
    FORALL allows more complex indices which aren't required and complicate compiled code unnecessarily
    Speeds up loops by around 10%


Restart routine

  Re-implemented, to run from specific restart data dumps
  Rolling restart added, deletes older restart files to reduce total output
  Also reads original density profile, including fluctuations if added, so result will continue the same

Calculation of wave wave coupling
    
  Better method to interpolate between different k grids introduced
    Ensures energy conservation in this process
    More stable with respect to coarse k-grid
    Simpler, more transparent code
    
Global variables

  Loop indices are no longer global
    Led to pathologies on Intel compiler

  Velocity, k_x and dk_x are now global as are k_f, k_h etc
    Reduces array passing to subroutines


Timestep determination
  
  Makes use of MPI functionality to find maximum wave energy density
    Faster, more transparent code

  Propagation timestep for EM waves used to determine code main step, dt


IF statements

  IF statement use has been reduced where possible, replacing with MAX functions or [0,1] variables
    Allows better optimisation by compiler


Bounds on passed arrays

  Arrays passed to subroutines have been given implied bounds


Allocation of arrays in subroutines

  Arrays used in subroutines with non-constant size (e.g. n_xp) are formally allocated/deallocated


F77 type syntax

  Updated syntax for e.g. alog-> log, double precision -> real(kind=8)
    

Divide-by-zero

  Potential divide-by-zero errors have been removed to protect against compiler differences


Input value checking
    
  More checking that input values are sensible and error messages or amendments where necessary


Code structure
  
  Divided and renamed subroutines by purpose
  Rearranged module file to group routines by purpose


Init params file versus constants
  
  Moved some key variables to the init.par file so they may be changed without recompiling
  Changed method of reading init.par file to Namelists so can easily add more items
  Only root reads file: multiple parallel reads can cause problems on some systems


Progress file and wallclock file

  Uses MPI_Wtime to record a wallclock time elapsed on each save
  Ditto to record total running time


Added outputs
  
  Velocity, k_x, k_f, k_h output to file. See below for IDL instructions to read them
  TO BE DONE- indices file, timesteps etc
  Warnings file to record if certain condtions arise (not yet completed)


Output files

  Dump everything in full precision only for restart purposes
  Dump only necessary data for science, controlled by input file
    Choice of single or double precision for this
  Dump edges of EM domains to make dynamic spectra. One file per time to allow restarting
  

ADDITIONS

Rejigged outputs to include:

Velocity and wavenumbers
Smooth background density profile

Progress and warnings files 


CHANGES TO IDL READING PROCEDURES

readf should be changed to readu
fwd and xdf are now written transposed, reverse indices of xdf before reading and transpose it after
write an extra file containing velocities and wavenumbers
If using the single-precision data dumps, change reading arrays to flt instead of dbl 

ADD to drst the following:

k_x_in=dblarr(kvc)
vely_in=dblarr(vc)
k_f_in= etc.....

fname = string(path_start,path,'velocity.dat')
openr,1,fname
readf,1,vely_in
readf,1,k_x_in
 readf,1,k_f_in
 readf,1,k_h_in
close,1

print,'Read velocity.dat'

.............................................................

required parameters now written into params.dat rather than drawing from init.par
In DRST replace lines between 'params	= dblarr(4)' and 'xc=fix(init(9))-3' by
......................................................................
params	= dblarr(10)

fname = string(path_start,path,'params.dat')
openr,1,fname
readf,1,params
close,1

print,'Read params.dat'

kvc = params(2)
vc = params(3)

xc=fix(params(4))-3
....................................................................


and in infovk change definitions:
d	= params(7)*1d+9
x_0	= params(8)
x_min	= params(9)
tsave	= params(5)*x2
n_beam = params(6)

dn_dx(*) = xTau(2,*)
n_e(*) = xdf(1,*)
n_e_smth(*) = xdf(4,*)
omega(*) = xdf(2,*)  ;*1e6*2*!pi
freq(*) = xdf(2,*)