Copyright (c) 2011-2014 Inria,   All rights reserved.


===========================================================================

This file contains the main features as well as overviews of specific
bug fixes (and other actions) for each version of ScalFMM since
version 1.1

2.0
-----
- New API for Kernels (The same kernel can be used witth all Octree (Classical, block, Adaptive)
- Adaptive FMM is available in shared memory paradigm (OpenMP and StarPU)
- scoped enum (FParticleTYpe (target,source), OPM4_Priorities, ...) is used
- StarPU works with Duplicated and Let trees with MPI (STF approach)
- Remove all warnings with gcc 7.2 still some with 8.1
- std::complex is used for equispaced points interpolation
- Fix bug with intel compiler and target/source 
- Improvement in CmakeLists  - FUSE is working again -  
- Compile with Intel and CLang (with a lot of warnings)
- Now morse_cmake is a git submodule 
- Fix several bugs in pseudo-periodic algorithms (sequential and parallel)
- Improve install step


1.5
-----
- improvements of the MPI version of ScalFMM 
- ScalFMM on StarPU (runtime system) is available 
- Add GPU version for ufmm (interpolation FMM based on equispaced grid points)
- Fix bug in distribution generation
-

1.4
-----
- Add interpolation FMM based on equispaced grid points (ufmm)
- Add the group tree and the things related to starpu (runtime system)
- Change the P2P (it is now done by tree functions in the abstract kernel) refer to the P2P page of the doc
- Update the P2P/M2L prototype, the position of the elements is known from an array position
- Fix bug in SSE, AVX, ...  kernel 1/r 

1.3
-----
- The choice of floating point values is done by template (by passing template to all classes) and no more at configure time
- There is no more ScalFMM variables and defines but only SCALFMM (all letters in upper case)
- We start to use unique ptr (and advise to use it) in the main files
- Add blocked version of the algorithm to increase the granularity (task-based approach)
- Fix bug with Chebyschev kernel and multi RHS.
- Improvements in unit tests (works with Linux and MacOs X for intel and gcc)
- Use MORSE distribution to detect BLAS, FFTW, ....
- BLAS and FFTW are turned on by default, and the compilation continue even if they are not found

1.2.1
-----
- Bug fix : Support for huge MPI message in tree construction and Parallel QuickSort (count can be greater than Int32.MaxValue)
- Bug fix : Data sharing attribute clauses for omp in Core/FAlgorithmThreadProc.hpp

1.2
-----
- New FMA format to read/write particles
- Add examples repository
- Add distributions generator
- Fix bugs in periodic model
- Fix bug and performance improvements in MPI thread algorithms
- Now we use MPI_THREAD_SERIALIZED rather than MPI_THREAD_MULTIPLE in MPI+OpenMP algorithm.
- Improve tests -> now we use CTest and CDash (Thanks to  F Rue  (Inria))
- Add support for Intel icl++ compiler (Intel 2015 compiler) on MacOS X and Clang on linux system
- Add SSE and AVX support in CMake
- Add SSE and AVX code for 1/r kernel
- CMake improvements