history

[!] Important
[+] New
[-] Fixed error
[*] Changes

------------------------------------------------------------------------------------------------------------------------
Ver. 0.73 - 02.03.06

! first stable release that has a manual and other documentation

------------------------------------------------------------------------------------------------------------------------
Ver. 0.74 - 29.03.06

+ Slightly increased the speed for scattered field calculation.
+ User manual improved.
* 'buggy' beam type removed (obsolete).
* '-nosym' and '-sym_enf' options combined to '-sym {no|enf}'.
- In documentation there was option '-pol ldr avg' while actually it was '-pol ldr avgpol'. It is averaging over
  incident polarizations (not directions). Documentation corrected.
+ Internal fields are now automatically collected into one file, when running in parallel mode ('IntField-X' and
  'IntField-Y')
- Saving of geometry in parallel mode could produce incomplete results (parts of some processors could be lost). Fixed.
+ MPI-specific command line options are enabled. Before they could cause segmentation fault.
* Geometry filename given to the command line option '-save_geom' is now specified relative to the output directory
  (i.e. if no path is given, it is saved into this directory).
+ Some warnings are now duplicated in the log.
+ In parallel mode each processor (except root) produces a separate logfile, if any errors or warnings were detected
  (on this specific processor). These files are named 'logerr.n' where n - is a number of processor.
* Minor changes to output format to stdout.
+ Added source files 'param.c/h'.
+ Overall robustness of the code is improved (added many consistency checks).
+ Help system added (new command line option -h).
- Fixed small bug in file locking (it could fail unexpectedly under certain circumstances).

------------------------------------------------------------------------------------------------------------------------
Ver. 0.74.2 - 05.04.06

+ Makefiles are more robust now. Option of using Intel compiler added.
+ Release option added to Makefiles (turns on/of all warnings automatically).
- A number of minor inaccuracies in the code fixed.
+ Section describing FFTW 3 installation added to the manual.

------------------------------------------------------------------------------------------------------------------------
Ver. 0.75 - 24.05.06

- Small bug in logging of command line fixed.
- Function 'system' (more precisely everything that calls 'fork') could fail unexpectedly under certain MPI
  implementations. All system calls removed.
+ Added source file 'os.h'.
+ Checking of errors in stdout has been added.
+ Possible size of FFT-grid (and non-weird number of processors) has been extended to full range of FFTW. One possible
  divisor of either 11 or 13 has been added.
+ Increased robustness of makefiles with Intel compilers 8.1 and 9.0 in conjunction with MPICH.
* Sample file 'batch' renamed to 'test.pbs' in 'sample/'.
+ Added sample script for SGE 'sample/test.sge'.
+ Different descriptions of Gaussian beam are now fully operational. Help, references added. Evaluation of extinction
  crosssection is modified to be compliant with Gaussian beams.
- Origin is now always in the center of the computational box (before it was not so when jagged>1).
- When x_0=E_inc was used, more iterations than necessary to reach epsilon were performed. Fixed.
* Progress reported in the first iteration after checkpoint loading was incorrect in certain cases. To fix it required
  loss of checkpoint compatibility (!!!) with earlier versions.  Some additional information is shown after checkpoint
  loading.
- Sometimes in parallel mode program could die of some error before producing an error message. Fixed.
* Copyright statement changed in code description and output. Now it is with University of Amsterdam.
+ Precision increased to 10 in ouput of several variables into log.
- Calculation of Csca or g caused both polarizations to be calculated even for particles that are symmetric for 90
  degree rotation over the z-axis. That is not necessary - fixed.
+ Integration module significantly improved. It automatically uses trapezoid rule for periodic function. Error estimates
  now seem to be reliable.
* Format of parameters for integration module changed because of improvement. It affects parameter files
  alldir_params.dat, avg_params.dat, scat_params.dat, making them incompatible with previous versions.
- Integration logs for Csca and g.x-g.z were saved in the same file for both incident polarizations. Now suffix "-X" or
  "-Y" is added to file names.
+ www page of ADDA is now available at http://www.science.uva.nl/research/scs/Software/adda/

------------------------------------------------------------------------------------------------------------------------
Ver. 0.76 - 14.10.06

+ Makefiles modified to include new compiler 'compaq'. New variables added to set path for FFTW3 header and library.
* Default dpl is now calculated based on the maximum absolute value among given refractive indices, not the first one.
  dpl=10*max(|m|)
- Removed requirement of full set of refractive indices to be given for multi-domain scatterers, if '-prognose' option
  is used. Produces information message when relevant.
+ Added information about distribution of occupied dipoles among different domains to the log file for multi-domain
  particles.
* GPL license changed to version 2 - file 'copyleft' replaced.
+ New option '-V' added to show version, compiler used, and copyright information. A head with information and copyright
  is no more shown at every run (redundant).
+ All memory allocation routines are now called by similar-typed macros - much more cleaner. Robustness of memory
  handling improved.
+ Timing is now produced also when '-prognose' option is used. Communication times are shown only in parallel mode.
  Redundant calls to calculate communication timing in sequential mode removed.
* All timing in parallel mode is now done through MPI_Wtime, which measures wall time. No more timer overflows in
  parallel runs. New header 'timing.h' for this task.
+ Function attributes moved from 'const.h' to new header 'function.h'. Just for convenience.
- Communication timing now measures exactly communication times (synchronization before some communications is added for
  that).
- Memory requirement in the manual slightly corrected. The values produced by the program itself were correct.
+ New option '-granul <vol_frac> <diam> [<dom_number>]' to granulate one of the domains. When this option is used
  '_gran' is added to shapename. Currently it may easily fail for volume fraction larger than 30-40%. Separate timing is
  produced for granule generator. Two files added - 'mt19937ar.c/h' - for a random generator (Mersenne Twister). The
  header file slightly modified from the original.
- Bug in checkpointer fixed. Was not portable to systems where size_t!=int.
+ Added check if file system does not support file locking; error output of file locking and file removal is now more
  informative (errno string is shown).
- Orientation averaging did not work in parallel (reported by Antti Penttila). Fixed.
- Symmetries were not switched off by non-symmetric Gaussian beams. Fixed. Improved hadling of symmetries.
+ Computation of scattering of a tightly focused Gaussian beam was compared to multiple multipole method (data by Roman
  Schuh and Thomas Wriedt). Perfect aggreement. Results of the comparison are added to the manual.
+ Added chapter to the manual to facilitate adding a new predefined shape.
+ New file 'doc/faq' added for frequently asked questions.
+ New option '-store_beam' to save incident electric fields to a file.
+ Starting from this release ADDA executable for WIN32 is available. It is compiled with MinGW, and distributed with
  FFTW3 dll. Thanks to Alexander Shvalov for information on MinGW.
- ADDA crashed when one of the given refractive indices was equal to one. Check for this was added, now error message is
  produced in this case.
+ Added support for anisotropic refractive index (option '-anisotr'). Currently it is limited to refractive index
  tensors that are diagonal in the particle reference frame. Thanks to Michiel Min for fruitful discussions.

------------------------------------------------------------------------------------------------------------------------
Ver. 0.77 - 05.06.07

- Calculation of radiation forces gave segmentation fault. The bug was introduced in version 0.76. Fixed.
+ Improved handling of large integers throughout the program. 'size_t' type is used wherever possible, so it should work
  for any problem that would fit into memory. Checks of integer overflow where necessary to avoid crashes.
+ Improved handling of input parameters to check for possible very large numbers, etc. Overall robustness improved.
+ Added checks to indicate incompatibility of orientation averaging and checkpoints with some other features.
- Consistency check for coated sphere was inaccurate. Fixed.
- Lack of EOL in the end of the input files could cause crash, as well as EOL conversion in geometry files. Fixed.
+ Makefiles improved (overall clarity + more flexibility for MPI compiling), included option for intel compilers without
  use of static linking.
+ Command line option '-store_dip_pol' was added to save dipole polarizations to file.
- Computer name was not properly shown when running MPI version on Windows. Fixed.
+ Improved the code that detects breakdowns of iterative solvers. Now it should be much more sensitive. Thanks to Sorin
  Pulbere for reporting a test problem.
- A minor bug in Romberg integration for angles that are sampled uniformly in their cos values, if both min and max
  values are exactly equal to 90 degrees. Reported by Antti Penttila. Fixed.
+ Locking of files made more flexible. Added compile option to independently turn off the advanced file locking.
- Minor change in comm.c to remove a warning.
+ Manual was significantly improved: "DDA formulation" section was rewritten to be self-contained, sections "Near-field"
  and "Comparison with other DDA codes" were added. Sections on compiling and running the code were extended to discuss
  in detail multi-core PCs. Sections "Applicability of DDA" and "System requirements" were extended to include recent
  benchmarking results and discussion. Thanks to Vitezslav Karasek and Liviu Clime for their feedback.

------------------------------------------------------------------------------------------------------------------------
Ver. 0.78 - 19.03.08

* Makefiles were improved to enable automatic compilation of both sequential and MPI versions from the same source
  folder. Name of MPI executable is now adda_mpi.
+ New file parbas.h added. MPI version is now checked for conformity both during compilation and at runtime.
+ Parallel executable for Win32, compiled with MPICH 2, is included in the package.
+ sample/test.sge file was updated to run on DAS-3 cluster.
+ Added explicit description of the used notation for the Euler angles in the manual, source code, help system, and
  parameter file.
+ sample/test.pbs was slightly improved.
+ File sinint.c added to calculate sine and cosine integrals.
+ Function cEqual was added to cmplx.h, it replaces memcpy calls in many other files.
+ Filtered coupled dipoles implemented (both full and quasistatic version). New arguments 'fcd' and 'fcd_st' for
  interaction term ('-int' command line option) and 'fcd' for polarization term ('-pol' command line option).
+ To-do list added (both in xls and pdf format).
+ Hands-on tutorial was added to the manual. It is based on the one carried out during the DDA Workshop in Bremen.
- ADDA was crashing for certain incorrect parameters given with the -h command line option. Fixed.
+ A few explicit statements that angles are specified in degrees were added to the help and input files.
+ New command line option '-opt {speed|mem}' added. It allows to choose whether ADDA will optimize itself for maximum
  speed or for minimum memory usage.
+ Stability of complex arithmetics slightly improved.
+ Calculation of scattered fields is accelerated by more than 4 times (compiled with gcc). Thanks to R. Scott Brock for
  this idea. This may also significantly accelerate orientation averaging, depending on the problem. Intel compiler is
  now only 10-20% faster than gcc, for all parts of the code.
- Minor bug in symmetry initialization of shape 'spherebox' is fixed.
- Information messages, when no real dipoles were allocated to one of the processors, did not work. Fixed.
* Information messages ("INFO:...") are now given without references to source files, which looks simpler.
+ New command line option '-eq_rad <arg>' was added, it allows one to specify volume-equivalent radius.
* Format of shape definitions in 'make_particle.c' was slightly changed.
* Shapes 'box' and 'spherebox' no longer depend on the second and third argument of '-grid' option.
+ Shape 'box' now accepts two optional arguments (y/x and z/x aspect ratios) defining a rectangular parallelepiped.
+ Automatic line wrapping was added for most output to stdout and sderr. ADDA tries to get terminal width from COLUMNS
  environmental variables and, if fails, uses the default value.
* '-pedantic' option for gcc was removed from Makefile, since it gave warnings about long predefined strings.
- Shape 'line' was not working correcty in combination with '-jagged' or xplicitely specified grid sizes along y- and
  z-axes. Fixed.
+ Two new shapes were added: capsule and egg. They were implemented by Daniel Hahn and Richard I. Joseph.
+ Description of predefined shapes in the manual was improved.
* Now, if grid size is auto-initialized from default dpl, ADDA ensures that it is not smaller than 16.
+ Limitation on positiveness of dipole coordinates in shape file was removed. Now ADDA automatically determines the
  minimum box around the particle and centers it as usual. It also ignores blank lines in the middle or in the end of
  shape file.
+ Orientation averaging was optimized. When beta=0 or 180, gamma angle is not relevant but only combinations
  alpha+-gamma. If alpha is varied in a full range [0,360), only one value of gamma for these specific beta values is
  calculated, saving a few evaluations of internal fields. Thanks to Antti Penttila for this idea.
+ Limited two-way compatibility with DDSCAT 6.1 geometry format was added, corresponding to its shape option FRMFIL and
  output of 'calltarget' utility. ADDA automatically detects DDSCAT format during reading of dipole file. When saving
  geometry to the file, the format is determined by new command line option '-sg_format {text|text_ext|ddscat}'.
+ New argument 'auto' was added to '-sym' command line option to correspond to the general rule that one of the possible
  arguments is the default one. No functionality is changed for other arguments.

------------------------------------------------------------------------------------------------------------------------
Ver. 0.78.1 - 08.04.08

- Critical bug in cDiv and cDivSelf functions in source file cmplx.h was fixed (introduced in version 0.78). This bug
  made BiCGSTAB and BiCG iterative solvers to fail. However, even when other two iterative solvers were used, erroneous
  results were calculated for certain values of refractive index.

------------------------------------------------------------------------------------------------------------------------
Ver. 0.78.2 - 11.04.08

- An implementation of a new FCD polarization prescription (-pol fcd) was somewhat faulty. Correction terms of orders
 (kd)^2 and ln(...)(kd)^3 were interchanged. Fixed.

------------------------------------------------------------------------------------------------------------------------
Ver. 0.79 - 29.05.09

+ Added using SLURM job ID, when available.
- Corrected inaccuracies in passing grid... to FFTW planning functions by making explicit conversion from size_t to
  int *.
- Warnings were produced about anisotropic refractive index in the input file for any file in DDSCAT format. Fixed.
+ Saving of granule coordinates to a file was implemented.
+ Discussion of difference between two ways to calculate Csca, possibilities to improve the accuracy of these
  calculations, and connection between differential cross section and the Mueller matrix has been added to the manual.
+ Tested for 64bit, implementing a number of corrections/improvements.
+ Support for new compiler (IBM XLC/XLF) added to the makefiles.
- Warnings for grid adjustments (e.g. from odd to even) were extended to cover all possible ambiguities, including
  combination with specified '-size' command line option. A discussion of these issues is added to the manual.
* Affiliation of one of the author (Maxim Yurkin) has changed to Institute of Chemical Kinetics & Combustion
 (Novosibirsk, Russia). This does not change the global strategy of ADDA development.
+ New shape 'axisymmetric' was added. The internals of shape generation routines were adjusted so that shape can now
  define the absolute size of the particle.
* Full name of the package was changed from 'Amsterdam DDA' to 'ADDA'.
* Code for calculating and saving different variants of Mueller matrix integrated over phi was shortened by implementing
  repeated blocks as inline function.
* Code repository has been moved to Google Code: http://a-dda.googlecode.com
+ Google groups adda-announce, adda-discuss and adda-develop were created to facilitate interaction with users. This
  replaces the previous mailing list.
* Heads of source and a few other files were changed using subversion keywords.
* Copyright was updated to GPL 3.
* Exit codes inside the code were changed to standard ones (EXIT_FAILURE, EXIT_SUCCESS).
* The special code in exit sequence for Borland compiler was removed (no more relevant).
+ Short history of earlier ADDA development was added to the manual.
* To-do list was replaced by issue tracker http://code.google.com/p/a-dda/issues
- avg_inc_pol was not initialized in param.c, which could potentially lead to undetermined behavior of LDR
  polarizability prescription (whether it is averaged over incident polaeization or not). Fixed.
- MakeParticle() was referenced as int instead of void in ADDAmain.c. This should not cause any problems, but was fixed
  anyway.
+ A description how to add new beam type and new command line options were added to the code and manual.
* All 'intel...' compilation options in makefiles were combined into one. Tested with icc 10.1 and 11.0.
+ Information on how to cite the manual was added to its first page. It includes a direct link to the particular version
  of the manual on Google code.
* PBS script was updated according to recent changes of module structure and 64-bit OS on LISA . An extensive
  description of this script was removed from the manual.
+ Manual license is now explicitly set to Creative Commons Attribution 3.0 license.
+ A small summary of the comparison of the DDA with the FDTD was added to the manual.
* '-prognose' command line option was replaced by '-prognosis' (correct spelling). The former is still operational but
  marked as deprecated.
+ A new option '-scat fin' was implemented, which uses finite dipole correction for calculation of Cabs and Cext.
+ Breakdown detection of iterative solvers was improved.
* The code was transformed to C99 standard without using its parts, which are not yet widely supported by compilers
  (such as inline and complex). Explicit support of this standard is required during compilation. Some style changes
  were done, and some work in this direction is still due.

------------------------------------------------------------------------------------------------------------------------
Ver. 1.0 - 10.09.2010

* doc/faq has been moved to a wiki page http://code.google.com/p/a-dda/wiki/FAQ
+ Calculation of Green's tensor was accelerated by doing all 6 components at once (issue 58).
+ IGT was implemented (issue 84).
+ Added information on extra build options (e.g. DEBUG) to the output of '-V' command line option.
+ New shapes: bisphere, bicoated, biellipsoid.
* The code was refactored: 1) all Fortran sources were moved to fort/; 2) make_mpi and make_seq were moved (renamed) to
  mpi/Makefile and seq/Makefile respectively; 3) now compilation goes on in folders mpi/ and seq/ completely
  independently.
+ Overall Makefiles were significantly modified: 1) targets 'all', 'cleanseq', and 'cleanmpi' were added; 2) informing
  about compilation process was implemented using $(info ...) directives; 3) most of the analysis in the Makefile is now
  skipped for non-compilation targets (like 'clean').
+ Flag NO_FORTRAN was implemented in Makefile, which can be used to disable all Fortran sources.
+ Three debug levels can now be easily specified in Makefile (issue 65). All the compilation options can be specified
  through the command line. And Makefile will automatically determine whether recompilation is required based on the
  differences in the final command line with which the compiler is invoked (issue 81).
* Format descriptors for output all over the code were replaced by a number of string macros. Should be easier to
  maintain. The visible output has stayed the same apart for a few minor changes.
* Code for saving Mueller matrix and cross sections after orientation averaging was moved from calculator.c to
  CalculateE.c.
- Minor bugs were fixed (issue 93, issue 96, issue 97, issue 107,  issue 110).
* Global constants MPI and ROOT were changed to ADDA_MPI and ADDA_ROOT respectively (issue 86).
+ New command line option '-scat_matr {muel|ampl|both|none}' to specify which scattering matrix (amplitude and/or
  Mueller) need to be stored to file (issue 7).
+ Iterative solvers were significantly refactored to move all common parts to one place. Added description of how to add
  a new iterative solver (issue 73).
+ Timing was slightly reworked. In particular, new timing variable was added to gather total communication time for one
  run of the iterative solver.
+ Significant changes towards using C99 standard (issue 64).
- Several bugs in granule generator were fixed (issue 108, issue 109).
+ Warning settings for the compiler was tightened (now it watches for unused and uninitialized variables). Corresponding
  changes were made in the code to suppress "legitimate" warnings.
+ Fabio Della Sala and Stefania D'Agostino contributed package to calculate near-field with ADDA (misc/near_field).
+ Several wiki pages were added and ADDA homepage at Google Code was completely rewritten.
- Fixed bug in Frp_mat (calculation of radiative forces) - r972.

Changes to the manual:
+ Section "System Requirements": Added description of the maximum number of processors that can be used for a given
  problem and corresponding minimum memory required by a single processor in a parallel mode.
+ Section "Predefined shapes": Added description of 3 new shapes.
+ Section "Partition over processors in parallel mode": Added description of the dependence of parallel efficiency on
  the order (orientation) of the axes for prolate or oblate particles.
+ Section "Mueller matrix": Added description of the difference between the Mueller matrix and the Stokes phase and
  scattering matrices.
- Section "Mueller matrix": Fixed typo in the last equation of this section (32) - last 'cos' should be 'sin' instead
  (issue 87).
* Section "Finale": Removed a paragraph, duplicating the info from Introduction.
* Section "Command Line Options": Description of the '-shape' command line option was simplified removing full list of
  options.
* Sections "What's New", "Obtaining the Source Code", "Compiling and Linking", "Acknowledgements", "Comparison with
  Other Codes", "Tutorial", "How to Modify the Source Code",  and "A Brief History of Long ADDA Development" were moved
  to the wiki pages.
+ A new section "How to Use the Manual" were added, in particular to point to existing wiki pages. Some text was also
  added to section "Finale".
* Section "Near-field": Added a short reference to the near_field package.
* Section "Applicability of DDA": Added discussion of possible ways to effectively simulate scattering by particles in
  the presence of substrate.
* Added sections "Amplitude matrix" and "ampl" describing use of the corresponding new command line option.

------------------------------------------------------------------------------------------------------------------------
Ver. 1.1 - 05.06.2012

+ New OpenCL-based version of ADDA to effectively use GPUs for simulations (issue 118). However, issues 144, 145 remain.
  Also new command line option '-gpu <index>' to choose the GPU to run on. Makefiles were changed to incorporate *.cpp
  sources.
+ Calculation of radiation forces was largely improved, fixing several bugs. Now it is well-tested with only two
  remaining limitations (issues 14, 135).
* Output files 'VisFrp-?.dat' were renamed into 'RadForce-?'.
* Command line option '-Cpr_mat' was renamed into '-Cpr'. The former was marked deprecated.
+ Three new 'misc/' packages: hyperfun and pip (issue 34), parse_log (issue 82). In particular, it greatly simplifies
  using many common 3D shape formats together with ADDA.
+ Added support for DDSCAT7 shape format, the previous one was labeled DDSCAT6. Command line option '-sg_format ...' was
  updated accordingly (issue 105).
+ When using '-shape read', recognized shape format is now saved to log.
+ New shapes: chebyshev, plate, and prism.
+ New option to read incident beam from file: '-beam read <filenameY> [<filenameX>]' (issue 2)
+ New polarizability formulations: DGF and LAK (issue 13).
+ New formulation IGT_SO, which is a much faster version of IGT. It applies to polarizability, interaction term, and
  scattering quantities (issue 84).
+ New iterative solvers: CSYM and QMR2 (issue 24).
+ New command line option to choose how to initialize the iterative solver: '-init_field {auto|zero|inc|wkb}'. First
  implementation of the WKB incident field is done, but it has a limitation (issue 59).
+ New command line option '-recalc_resid' to directly recalculate residual at the end of the iterative solver
  (issue 25).
* Updated the default build platform to gcc v.4.x (tested with 4.5.2). Many connected changes, in particular issue 122.
+ 64-bit executables of ADDA for Windows are now available (issue 66).
+ Windows executables for routines from misc/ are now available (both 32- and 64-bit) (issue 104).
+ Added automatic tests of ADDA executables to '/tests/2exec' (issue 6).
+ Added explicit tests that arguments, which have to be passed to MPI functions as int, do not exceed INT_MAX
  (issue 137).
- Minor bugs were fixed (issues 123, 124, 128, 136, 139, 146).
- Minor bug in handling input shape filename was fixed (r1141).
+ Several wiki pages were added: AddingShapeFileFormat, Awards, InstallingMinGW, InstructionCommitters,
  LargestSimulations, OpenCL, Publications, and References (issue 117).
* Copyright information was changed to "(C) ADDA contributors". '$Auth' field was removed from headers.
* Deprecated command line option '-prognose' was finally removed.
* Automatic names for output directory were slightly changed (r1106).
* Minor changes in stdout and log.
+ Two new development branches - weighted discretization (issue 12) and sparse (non-FFT) DDA (issue 98).

Changes to the manual:
+ Added section "OpenCL (GPU) mode" describing executing of adda_ocl.
+ Section "Applicability of the DDA" was divided into two subsections: "General applicability" and "Extensions of the
  DDA"; some text was added.
+ Section "System Requirements": Updated descripton to include new iterative solvers and OpenCL mode. Two figures were
  combined into one.
+ Section "Construction of a dipole set": Three figures were combined into one.
+ Section "Predefined shapes": Added description of 3 new shapes
+ Section "Single orientation": Added figure describing Euler angles.
+ Section "Beam type": Added expression for beam power.
+ Section "DDA Formulation" significantly changed. Many new formulations were added. All polarizability expressions were
  rewritten in terms of term M. Difererences for Cabs between '-scat dr' and '-scat fin' are now explained in more
  details.
+ Section "Mueller matrix" was renamed into "Mueller matrix and its derivatives" and added expression for parallel and
  perpendicular intensities (or horizontal and vertical) and backscattering cross section.
+ Completely rewritten section "Radiation forces".
+ Section "Iterative solver" significantly changed according to new features.
+ Added sections "Incident beam file" describing format of corresponding input file.
+ Section "VisFrp" was renamed into "RadForce" and significantly changed.
+ Section "Geometry files": added warning about shown box size.

------------------------------------------------------------------------------------------------------------------------
Ver. 1.2 - 08.06.2013

+ Sparse mode is now fully integrated into ADDA and significantly improved (issue 98, SparseMode). It is designed for
  very porous particles. New file sparse_ops.h.
+ Linked to new OpenCL FFT library from AMD (clAmdFft), it is now the default one in OpenCL mode (issue 144,
  InstallingclAmdFft). Speed improvement is a few times depending on the GPU. A few minor improvements of handling FFTs.
+ New iterative solver BCGS2 - an improved version of BICGSTAB (issue 24).
+ New command line option argument -init_field read <filenameY> [<filenameX>]' to initialize an iterative solver by the
  field read from file (issue 2).
* Function to read field was moved from GenerateB.c to comm.c, now it is called ReadFields(). It was also optimized in
  MPI mode.
* Behavior of WKB initial field was slightly improved. Before WKB result was assigned to E_exc, now it is assigned to E
 (similar to the behavior of '-init_field read').
+ Subversion revision number was added to the output of -V command line option (issue 125). New files svnrev.h and
  updsvnrev.sh. 
+ Several improvements of tests/2exec to accomodate OpenCL and sparse modes, as well as comparison of different FFT
  implementations (r1192, r1200, r1203).
+ Added tests/2exec/README with basic description of using this test suite.
* Window width is now 120 throughout the code.
+ New files interaction.c/h, which hold all code concerning the calculation of Green's tensor.
* Intermediate vector DipoleCoord_tmp is no more used in make_particle.c. This somewhat decreases memory usage during
  prognosis. Overall, the code in MakeParticle() was significantly changed (a lot of moving parts around).
* Finally removed deprecated options -Cpr_mat and '-sg_format ddscat'.
* Outdated INLINE macro removed throughout the code.
* Many if-else-if sequences in the code were changed to switch.
* The code for CalcField and SaveGeometry, has been simplified by moving conditions (if and switch) inside the loops.
+ Added test for duplicate dipoles in input shape files.
+ Added function FinalizeSymmetry(), which is called immediately after InitShape().
+ Timing for generation of incident beam was added.
- Fixed inaccuracy in input/scat_params.dat (r1159).
- A few minor bugs fixed (issues 83, 159, 162).
+ Added wiki pages: SparseMode, InstallingclAmdFft, CodeDesign (issue 75), and CodeStyleGuide (issue 74, includes code
  style scheme for Eclipse CDT).
* Replaced most png images with svg ones in ComparisonOtherCodes.
+ Added new folder devtools. It contains a couple of text files, bash scripts for preparing files for release and other
  tasks, and default set of svn-properties for new files (to be used with TortoiseSVN).
+ Added doxygen config file (doc/doxy/Doxyfile) for automatic generation of extensive docs and GraphViz files (doc/gv/)
  with schemes describing ADDA design (for svg files in CodeDesign).

Changes to Makefiles:
+ Full support for compilation in 32-bit mode on 64-bit OS, using 'EXTRA_FLAGS=-m32' (see also 
  InstallingMinGW#Multilib_configuration_on_Windows).
+ Added comment on possible use of 'CC=...' to specify a different compiler.
* Improved the automatic dependency generation in Makefile. Now it happens during the compilation (in one go) and also
  works for C++ sources. It should be faster and produce immediate break if, e.g., header file is missing.
+ Added new targets to Makefile - cleanfull, cleanruns, cleanrunsseq, cleanrunsmpi, cleanrunsocl. They enable cleaning
  output from ADDA runs (in addition to the output of ADDA compilation, which is cleaned by clean).
* debug.c and prec_time.c are now only compiled when needed.
+ Makefiles in misc/*/ were improved to allow compilation from another folder (path) and with extra flags (e.g. -m32).

Changes to the manual:
+ Changed "authors" to "developers" throughout the manual. Added link to developer list in Section "Introduction".
+ Section "OpenCL (GPU) mode": added link to wiki page OpenCL.
* Colors in Fig.1a were adjusted because they caused problems during conversion to pdf. 
+ Added figure with examples of predefined shapes (issue 140).
+ Section "Beam type": added note that '-beam read' cancels symmetry.
+ Section "How to calculate scattering quantities": added discussion of symmetry-based tests of DDA results (issue 143).
+ Section "Integral scattering quantities": added note that the result for unpolarized incident light is equal to that
  for full-range averaging over alpha.
+ Section "Iterative solver": added description of new iterative solver bcgs2 and new option '-init_field read'. Also
  the intercomparison of iterative solvers was rewritten in terms of matrix-vector products, and a summary table was
  added. Section "System Requirements" was updated to refer to this table.
+ Section "Fast Fourier transform": added description of a clAmdFft library.
+ Added section "Sparse mode", a few other sections were also updated.
+ Section "Miscellanea": added link to wrapper script in R language by Baptiste Auguie.
+ Section "References": DOIs and URLs were made into hyperlinks.

------------------------------------------------------------------------------------------------------------------------
Ver. 1.3b4 - 20.02.2014

+ Major changes to the code due to switch to complex types of C99 (issue 70).
+ Major new feature: DDA for particle near surface (substrate) - issue 101. Command line options: '-surf ...', 
  '-int_surf ...'.
+ New command line option -scat_plane, which defines that the scattering should be calculate in the scattering plane
  (issue 170).
+ Implementation of point-dipole incident field (issue 149). Should work both for free space and in the presence of
  surface (dipole should be above the surface). Controlled by new beam type: '-beam dipole ...'.
+ New DDA formulation for non-local (spatially dispersive) materials (two different flavors: point-value and averaged 
  over dipole volume) - issue 190. Command line options '-int {nloc|nloc_av} ...' and '-pol {nloc|nloc_av} ...'.
  Currently assumes quasi-static case (particles are much smaller than the wavelength).
 
* E2_alldir is now used to store only the final result, while (intermediary) complex fields are stored in a new vector
  E_ad. This slightly increases required memory (proportional to number of angles).
+ MPI_C_DOUBLE_COMPLEX and MPI_C_BOOL are used, where possible (and supported).
+ Accumulate() now operates in place and operates on any datatype. Corresponding buffers Egrid_buffer, Eplane_buffer,
  and E2alldir_buffer were removed (somewhat decreasing the memory footprint in MPI mode).
* MatVec in sparse mode was updated to use standard functions from linalg.c.
* Calling of IGT Fortran routines was slightly changed, conversion from double to complex has been moved inside the C
  file from Fortran routine.
+ Added defines to simplify adding pragmas to ignore warnings in gcc and loop_count in icc.
+ Several improvements of tests/2exec to accomodate new features and test modes (r1254, r1255, r1263, r1265, r1285).
+ Several functions have been added to cmplx.h.
* Handling of scattering directions based on angles theta and phi (alldir and scat_grid) was localized to a separate
  function SetScatPlane in crosssec.c.
* The logic in interaction.c/h was significantly modified. Now two types of functions are defined (visible outside) for
  both direct and reflected interaction, accepting either integers or real values.
+ Parsing and testing of command line arguments was simplified, by adding several utility functions
+ Function CoupleConstant in calculator.c was rewritten to separate the code for different formulations.
+ Removed compensation of Cext for '-scat fin' (issue 188).
+ README files in win32/ and win64/ were improved to refer to the main package.
+ Added a test to timing.c, which will notify the user of (potential) timer overflow in timing results in log.
+ Symmetries are now mentioned in the log not only in 'no' or 'enf' regime, but also in 'auto'. Then the specific set
  of used symmetries is given.
- A few minor bugs fixed or addressed (issues 172, 174, 182).
+ Added wiki pages: AddingInteraction, AddingReflection, AddingPolarizability (issue 106), Links.

Changes to Makefiles:
+ Added test of gcc version, '-pedantic' is used only for versions >= 4.7.2.
+ Added test to require version 3.81 of make, including support for else-if (r1285).
+ Now compilation of *.f90 is allowed in the general framework (not yet used).

Changes to the manual:
+ New Sections: "Surface Mode", "Incident Beam", "Reflection term", "Decay rate enhancement", "Sommerfeld integrals".
* Section "System Requirements": memory and processor time requirements updated, including surface mode.
+ Section "Reference frames": added discussion on different definition of scattering angles for free-space and surface
  modes.
* Section "Partition over processors in parallel mode": better explained the partition, which is performed along both 
  the z- and x-axes. Related changes were made in Section "Fast Fourier transform".
+ Section "Particle symmetries": added explanation of symmetries codes as saved to log.
+ Section "Single orientation": mentioned incompatibility of rotation with -surf.
+ Section "Propagation direction": added note of different definition of scattering angles and about special meaning of
  -prop for dipole incident beam (or the one read from file).
+ Section "Beam type": added explicit formulae for incident field of the plane wave, added explanation of incident wave
  in surface mode and of point-dipole field, added a table with all possible beam types.
+ Section "DDA formulation": added discussion of differences between direct and total interaction terms (related to
  reflection term).
+ Section "Polarizability Prescription": added a brief note on non-local formulations.
+ Section "Interaction Term": added a brief note on non-local formulations.
+ Section "How to calculate scattering quantities": extended formulae to include surface-mode, added expression for
  Poynting vector, removed compensation of Cext for '-scat fin'.
+ New Section "Definition of scattering plane and angles". Some material was moved here from several other sections.
  Also numerous changes throughout the manual to refer to this section for scattering plane definition.
+ Section "Mueller matrix and its derivatives": added explicit definition of the Mueller matrix with discussion of
  peculiarities related to the surface mode, added expression for circular depolarization ratio, formulae updated to
  comply with both definitions of the scattering plane.
+ Section "Amplitude matrix": added explicit definition of the amplitude scattering matrix.
- Section "Radiation forces": removed incorrect statement on calculation of optical torques. Instead it now refers to a
  workaround. Described incompatibilities with the surface mode.
+ Section "Iterative Solver": mentioned incompatibility of 'init_field wkb' with -surf. In formulae matrix G is
  replaced by H.
+ Section "Fast Fourier transform": specified additional time for FFTs in surface mode.
+ Section "Parallel performance": minor changes, it now explicitly refers to optimal partition of FFT grid.

------------------------------------------------------------------------------------------------------------------------
Ver. 1.4.0 - 15.12.2020

* ADDA moved to GitHub with a lot of related changes to documentation (#210):
  * `doc/copyleft` was renamed/moved to `./LICENSE`
  * top-level `README.md` replaces previous homepage.
  * wiki pages ReleaseNotes and Downloads (created recently - #189, #72) were deprecated in favor of GitHub releases.
  * wiki page GettingStarted renamed into Home
  * wiki page PackageDescription was deprecated in favor of `README.md` in each directory
  * all links to Google Code updated to point to GitHub in source files, manual, wikis, and even some adda-discuss
    messages.
  * Changed machinery of `svnrev` to `githash` (#237).
  * Changed svn:ignore to .gitignore with some optimizations.
  * Changed template for comments in the top of source files, removed svn fields.
  * Updated release scripts 

* Many changes to OpenCL part:
  * Significant speed improvements (especially for relatively small problems). It is partly based on additional GPU
    memory consumption (if available), controlled by `-opt ...` command line option.
  * Switched from discontinued library clAmdFft to clFFT (#204). It supports the same sizes of transforms as FFTW3. 
  * Test implementation of the whole iterative solver (currently only BiCG) on the GPU using clBLAS (#199).
  + Added rigorous accounting (in OpenCL memory) for clFFT internal buffers.
  * OpenCL part of `matvec.c` moved into seperate file `oclmatvec.c`
  * Function wrappers for clBLAS (#213).
  * Added error checks (`CL_CH_ERR`) to many ocl function calls. Special error checks for clFFT and clBLAS calls.
  * Improved compilation flags for MacOS - now works out of the box (#279).
  * Fixed a few bugs on the way (#157, #207). 

+ Non-cubical (rectangular cuboid) dipoles - a major new feature (#196). Lattice spacings are now hardcoded into the
  saved geometry files. Dipole size is shown in the log.
+ Remove limitations of grid size dimensions to be even (#95).
* Improvement of `tests/2exec` to ignore most of the legitimate messages. Compatible with both bash 3.1 and later 
  versions. Enhanced syntax of suites for fine tuning. New suite `suite_rd` and modes for rectangular dipoles. Added 
  `test_all` for systematic testing. `diff_numeric.awk` now can compare complex numbers.
- Windows 32-bit executables are no more provided.
* Adopted semantic versioning.  
* Minimum MPI version requirement is now 2.1.
* Improved detection of optional MPI features (#201).
* Improved description and version detection for modern MPI implementations (#203, #246).
* Added support for Microsoft MPI - works out of the box on Windows (d5814aa).
* Behavior of '-maxiter ...' command line option was changed. Now when given number is reached ADDA produces a warning
  and continues to calculate scattering quantities (#111).
+ Added separate timing for matrix-vector products (#16). Presicion of total wall time was improved on most operating
  systems (Windows and POSIX).
* Optimized computation of exp(i*x) for real x using tables (#169). It accelerates sparse mode by 30-50% (similar to 
  more specialized SSE3 code), but can be turned off by defining NO_IMEXPTABLE in cmplx.h.
* Fixed redundant IFAIL errors in IGT routines (f6d6400).
- Several minor bugs fixed or addressed (#198, #215, #216, #228, #266, #269, #270, #271, #276).
* Minor changes to output - couple of spaces added/removed (758fea0).
+ Added wiki pages: UsingEclipse, InstallingclFFT, and InstallingclBLAS.
* Deprecated wiki pages InstallingclAmdFft and InstallingclAmdBlas (the latter newly created).
* Updated wiki pages CompilingADDA, InstallingFFTW3, InstallingMinGW, OpenCL, SparseMode, InstallingMPI, Acknowledgements.
* Build scripts improved and updated. One wrapper can be used to build and test at once.
+ New command line option: -so_buf {no|line|full} for manually specifying the buffering of stdout. On Widows line 
  buffering and default POSIX behavior (when command line is not used) is emulated, which improves performance in some
  cases (#277).
* Removed 6 dots in stdout after the "Fourier transform of Dmatrix" (or Rmatrix). They were not properly buffered in
  many cases.
+ Set the official pronounciation of ADDA - /'adda/.

Changes to `misc/`:
* `near_field` made robust with respect to comments in shape files. Now also works with inhomogeneous particles (#259).
* `parse_log` updated (bb2efa6)
* `pip` updated for massive simulation (many files, billions of dipoles) (#262).

Changes to Makefiles:
* Fixed bug preventing use of LDLIBS from environment (b6e54fb)
* Updated comments to recommend linking to DLLs directly
* Changed optimization level during debug build from -O2 to -O1 (or -Og) (c9ad8bf)

Changes to the manual:
+ Section "Introduction": specified pronounciation of ADDA.
+ Section "Sequential mode": mentioned that ADDA is a console application.
* Section "How to use the manual" renamed into "Using the manual".
+ Section "General applicability": added statement that the DDA is a "numerically exact" method and more clear 
  description for particles smaller than wavelength.
* Section "Extensions of the DDA": made it explicit that all refractive indices need to be scaled by m0.  
* Section "System Requirements": mentioned extra memory that can be allocated in OpenCL mode. Eq.(4) for Mtot was made
  less ambiguous with respect to the part "only if parallel" (#206), also the range before Nreal now uses less confusing
  notation. 
+ Section "Reference frames": added advice when unsure about dipole coordinates.
+ Section "The computational grid": mentioned that the default unit of length is 1/k, removed mentions that grid sizes are required to be even. Added description of rectangular dipoles (also affects many other parts of the manual).
+ Section "Construction of a dipole set": recommended disabling volume correction in a number of computational scenarios.
+ Section "Predefined shapes": added clarification and reference to description of RBC shape, added explicit mention of argument bounds for many shapes. 
+ Section "Beam type": added explicit statements that beam arguments are in laboratory reference frame. Recommended use 
  of -no_vol_cor for point-dipole excitation.
+ Section "DDA formulation": the sum in the main equations was changed to include Rii (assuming Gii=0). Added a note about notation (G and H instead of G0 and G).
+ Section "Polarizability prescription": added a lot of formulae for rectangular dipoles, reworked usage recommendations.
+ Section "Interaction term": explained differences for rectangular dipoles, reworked usage recommendations.
+ Section "How to calculate scattering quantities": added rationale for using a specific definition of the scattering 
  amplitude F(n)
+ Section "Mueller matrix and its derivatives": Footnote describing Stokes scattering matrix was corrected (integrated
  -> averaged). Corrected special symbols (||) in Eq.(42) and "phi" symbols in Eq.(41).
+ Section "Integral scattering quantities": added a note that Cext is directly related to a molar extinction coefficient
  of a suspension of particles. Added note that definition of efficiencies is different from those based on
  geometric-shadow cross section.
+ Section "Decay-rate enhancement": added specific formulae in the case when intrinsic losses are present (corresponding intrinsic quantum yield q0<1), based on a recent publication.  
+ Section "Internal fields and dipole polarizations": recommended the use of -no_vol_cor.
+ Section "Near-field": recommended the use of -no_vol_cor, mentioned that outer-box workaround can be used to compute
  near field in the surface mode.
+ Section "Iterative Solver": changed the desciption of `-maxiter` option. Added formula to estimate Niter for
  nanoparticles. Threshold for convergence 𝜀 renamed into 𝜀_iter to avoid confusion with dielectric permittivity.
+ Section "Basic timing": added reference to #197 (problem with clock in OpenCL mode), updated the description of wall
  timers (precise ones are used when available).
+ Section "Precise timing": added a caveat for OpenCL mode.
+ Section "Geometry files": added mention that sample files can be produced by ADDA itself.
+ Section "stdout": added explanation of '-buf_so' command line option.
* Chanded font for hyperlinks back to Times.
* Changed source file format to .docx
* Changed all equations to new Microsoft Word format