Skip to content

Latest commit

 

History

History
392 lines (282 loc) · 17 KB

CHANGELOG.md

File metadata and controls

392 lines (282 loc) · 17 KB
Raw

Change Log

All notable changes to this project will be documented in this file.

OpenFPM 5.1.0 - Jun 2024

  • Refactor implementations of cell list and Verlet list CellList, CellList_gpu,VerletList and all neighborhood iterators. Move from keeping two sets (unordered and ordered) of positions/property vectors to reordering explicitly before launching CUDA kernels that utilize this feature

Added

  • Add adaptive cut-off radius Verlet list on CPU. Particles individual radii are passed as a vector to getVerletAdaptRCut of openfpm_pdata. Time complexity: O(N^2)

Changes

  • cuda_launch is a function instead of a macro
  • Move project c++ standard to c++17
  • Add bit-wise option flags that control lell list behaviour. Pass as a parameter opt to getCellList of openfpm_pdata or set directly on cell list via setOpt
  • Add support for the following cell list bit-wise option flags: CL_SYMMETRIC, CL_NON_SYMMETRIC, CL_LOCAL_SYMMETRIC, CL_LINEAR_CELL_KEYS, CL_HILBERT_CELL_KEYS, CL_GPU_REORDER, where for the last one CL_GPU_REORDER control options for reordering/restoring operations could be more fine-tuned: CL_GPU_REORDER_POSITION, CL_GPU_REORDER_PROPERTY, CL_GPU_RESTORE_POSITION, CL_GPU_RESTORE_PROPERTY
  • Add bit-wise option flags that control Verlet list behaviour. Pass as a template parameter opt in VerletList or pass Verlet list type with opt set to getVerlet of openfpm_pdata. The difference in how opt is set for cell list and Verlet list is due to legacy code
  • Add support for the following Verlet list bit-wise option flags: VL_NON_SYMMETRIC, VL_SYMMETRIC, VL_CRS_SYMMETRIC, VL_ADAPTIVE_RCUT, VL_NMAX_NEIGHBOR, VL_SKIP_REF_PART
  • getCellListGPU of openfpm_pdata doesn't fill a cell list with particle locations. Additionally, updateCellListGPU has to be called to perform this operation. The reason is to remove two identical fill operations in a row in simulations
  • To reorder vector_dist for GPU coalesced memory access dictated by the cell list structure, cell list on GPU has to be constructed with one of the following flags enabled: CL_GPU_REORDER, CL_GPU_REORDER_POSITION, CL_GPU_REORDER_PROPERTY, CL_GPU_RESTORE_POSITION, CL_GPU_RESTORE_PROPERTY. Properties to be copied to reordered copies of position/property vectors have to be passed to updateCellListGPU<PROPERTIES>(CELL LIST) of vector_dist. Properties that undergone changes in reordered copies and have to be restored to the original position/property vectors have to be passed to restoreOrder<PROPERTIES>(CELL LIST) of vector_dist

OpenFPM 5.0.0 - Feb 2024

  • Move to openfpm meta OpenFPM project structure with git subprojects openfpm_data, openfpm_devices, openfpm_io, openfpm_vcluster, openfpm_pdata, openfpm_numerics. Example codes, installation scripts for dependencies, configuration scripts moved to openfpm. Only source code of openfpm_pdata kept in openfpm_pdata

Added

  • Add DC-PSE gpu implementation
  • Add support for minter header-only library. Installed automatically if openfpm_numerics is enabled
  • Add Odeint gpu support

Fixed

  • Fix remove_marked when number of particles is decreased and memory is reused
  • Fix host_to_device_impl when allocated size > actual size
  • Fix DC-PSE operator with Lagrange multiplier for number of variables > 1
  • Fix template substitution failure for Graph_CSR with no edge
  • Fix Cell List getNNIterator giving different neighborhood lists for dense and sparse versions
  • Fix sendrecvMultipleMessagesNBX for message size > 2 GB

Changes

  • Get rid of autotools emulation for installation procedure. Remove install/update/upgrade scripts. Move to CMake with manually building dependencies and manualy calling configuration scripts for CMake, example.mk, openfpm_vars
  • Remove gdbui from list of git subprojects
  • Move from moderngpu as it's no longer supported to cub
  • Refactor Cell List implementation, change code structure and variable naming. Split dense and sparse Cell List into partial template specializations
  • Move all backends for CUDA emulation, CUDA headers and parallel primitives to openfpm_devices from openfpm_data
  • Remove default parameter from NO_CHECK from Cell List iterators (e.g. getNNIterator). cl.template getNNIterator<NO_CHECK>(...) is replaced by cl.getNNIterator(...)
  • Pass Cell List NNIteratorBox parameter to getCellList/getCellListGPU of vector_dist. getNNIteratorBox has to be set when initializing the Cell List compared to getNNIteratorRadius, which could be set later, due to sparse Cell List on gpu using it in construct(). If not set, the dafault value of 1 would have been used by construct()

[4.1.0] (Codename Thesis)

  • On a general base the code should not use CUDA_ON_CPU but if it does CUDA_ON_CPU macro now cover both SEQUENTIAL and OpenMP backend. The macros CUDIFY_USE_CUDA,CUDIFY_USE_HIP,CUDIFY_USE_OPENMP,CUDIFY_USE_SEQUENTIAL,CUDIFY_USE_NONE can be checked to control which CUDA backend is used

Fixed

  • Minors bug

[4.0.0] September 2021 (Codename Heisenberg)

  • Adding DCPSE, Level-set based numerics (Closest-point)

Changes

  • Particles writer use now the new XML pvtp Paraview format
  • Particles constructor does not accept discordance precision on space example: If I create a Particle set with vector_dist<3,double,.......> I MUST give in the constructor a domain Box<3,double> and a Ghost<3,double>, giving a Ghost<3,float> will produce an error

[3.3.0] April 2021 (Codename Vega)

  • Adding support for HIP and AMD GPU. (Only particles) 1_gpu_first/7_sph_dlb_gpu/7_sph_dlb_gpu_opt are compatible with HIP Additional Notes:
  • WARNING: AMD GPUs are tested manually and not in CI. This mean that out this release stuff can break at least until I do not convince my working place to buy one for me ... and is gonna be hard because or rule here and there ... or who is reading this message does not want to buy one for me :-)
  • SparseGridGPU are unsupported untill AMD does not fix the bug reported here: ROCm/HIP#2260

Changes

  • None

Fixed

  • uninitialized variables in the SPH example on GPU, and other fixes necessary for AMD gpus

[3.2.0] January 2021 (Codename Hopper)

  • Adding CUDA_ON_CPU option to run CUDA code on CPU
  • Adding gdb-gui debugger

Fixed

  • Minors bugs

Changes

  • In order to compile OpenFPM is now required a compiler implementing C++14 Standard

[3.1.0] October 2020 (Codename New Horizons)

  • Adding GPU support for ghost_put
  • Adding support for CUDA 11

Fixed

  • Few details in the installation system of linear algebra for OSX

Changed

  • None

[3.0.0] July 2020 (Codename Sparsity)

  • Upgrading all the dependencies: BOOST,PETSC,SUITESPARSE,OPENBLAS
  • Adding CPU and GPU sparse grids. Look at the examples SparseGrid in the forlder examples
  • Improving performance of GPU function remove_marked

Fixed

  • Several installation bugs on PETSC installation

Changed

  • Name for GoogleCharts has been changed please look at the examples

[1.X ] End of life (Theese versions are not enymore supported)

[2.0.0] February 2019 (Codename Elisa)

Added

  • Adding GPU support (see example 1_gpu_first_step 3_molecular_dynamic_gpu 7_sph_dlb_gpu 7_sph_dlb_gpu_opt)

Fixed

  • Detection of clang 10.0.0 on mac-osx mojave
  • In VTK binary format all 64 bit types are casted to 32 bit. Either the long/unsigned_long are bugged in Paraview we tested, either I do not understand how they work.

Changed

  • The type Vcluster now is templated and the standard Vcluster is Vcluster<> Most probably you have to change in your code from Vcluster to Vcluster<>

[1.1.1] December 2018 (Codename Poisson)

Fixed

  • Detection of clang 10.0.0 on mac-osx mojave

[1.0.X ] End of life (Theese versions are not enymore supported)

[1.1.0] February 2018 (Condename Ring)

Added

  • Interface for Multi-vector dynamic load balancing
  • Increaded performance for grid ghost get
  • Introduced forms to increase the performance of the grid iterator in case of stencil code (see example 5_GrayScott)
  • EMatrix wrapped eigen matrices compatibles with vector_dist_id
  • General tuning for high dimension vector_dist_id (up to 50 dimensions) + PS_CMA_ES (Particle-Swarm Covariant Matrix Adaptation Evolution Strategy) example in Numerics
  • Added Discrete element Method example (8_DEM)
  • Added serial_to_parallel example VCluster (2_serial_to_parallel). The example it show how to port a serial example into openfpm gradually swtiching from a serial section to a parallel section
  • Introduced map(LOCAL) for fast communication in case we have small movement

Fixed

  • Installation/detection of PETSC
  • CRITICAL-BUG scalar product in combination with vector product is broken (it return 0)
  • Fixing 2D IO in binary for vector
  • Fixing 1D grid writer in ASCII mode
  • Fixing Intel compilation of Linear algebra

[1.0.0] 13 September 2017 (Codename: Vortex)

Added

  • Introduced getDomainIterator for Cell-list
  • New dynamic load balancing scheme to 7_SPH_opt (see Vector/7_SPH_opt)
  • Increased performance of 7_SPH_opt
  • Vortex in Cell example Numerics/Vortex_in_cell
  • Interpolation functions (see Numerics/vortex_in_cell example)
  • Gray-scott 3D example with stencil iterator optimization (see Grid/gray_scott_3d example)
  • HDF5 Check point restart for vector_dist particles (see Vector/1_HDF5_save_and_load)
  • Raw reader for grid (see ...)
  • Added setPropNames to give names to properties (Grid and Vector see Vector/0_simple)
  • Ghost put on grid (see Numerics/Vortex_in_Cell example)
  • Stencil iterators for faster stencil codes see (Test3D_stencil function in Units tests src/Grid/Iterators/grid_dist_id_iterators_unit_tests.hpp)
  • Algebraic multigrid solvers interface for linear systems (see Vortex in Cell example)
  • Added setPropNames in vector_dist see Vector/0_simple
  • Support for Windows with CYGWIN

Fixed

  • Bug fixes in installation of PETSC
  • 2 Bugs in 7_SPH_opt and 7_SPH_opt error in Kernel and update for boundary particles
  • Bug in VTK writer binary in case of vectors
  • Bug in VTK writer binary: long int are not supported removing output
  • Bug in FDScheme in the constructor with stencil bigger than one
  • Bug Fixed Memory leak in petsc solver
  • Bug Performance bug in the grid iterator

Changed

  • CAREFULL: write("output",frame) now has changed to write_frame("output",frame) write() with two arguments has a different meanings write("output",options)

  • getCellList and getCellListSym now return respectively CellList_gen<dim, St, Process_keys_lin, Mem_fast, shift<dim, St>> CellList<dim, St, Mem_fast, shift<dim, St>>

  • getIterator in CellList changed getCellIterator

  • Grid iterator types has changes (one additional template parameter)

  • FDScheme the constructor now has one parameter less (Parameter number 4 has been removed) (see Stokes_Flow examples in Numerics)

  • MPI,PETSC,SUPERLU,Trilinos,MUMPS,SuiteSparse has been upgraded

[0.8.0] 28 February 2017

Added

  • Dynamic Load balancing
  • Added SPH Dam break with Dynamic load balancing (7_sph_dlb)(7_sph_dlb_opt)
  • Added automatic procedure for update ./install --update and --upgrade (From 0.8.0 version will have a long term support for bug fixing. End-of-life of 0.8.0 is not decided yet, it should be still supported for bug fixing after release 0.9.0)
  • Added video lessons for Dynamic load balancing (openfpm.mpi-cbg.de) (website officially open)
  • Added for debugging the options PRINT_STACKTRACE, CHECKFOR_POSNAN, CHECKFOR_POSINF, CHECKFOR_PROPINF, CHECKFOR_PROPNAN, SE_CLASS3.
  • Added the possibility to write binary VTK files using VTK_WRITER and FORMAT_BINARY see 0_simple_vector for an example

Fixed

  • Installation of PETSC with MUMPS

Changed

  • BOOST updated to 1.63
  • Eigen updated to 3.3.7

[0.7.1] 28 January 2017

Fixed

  • Multiphase verlet single to all case generate overflow

[0.7.0] 15 December 2016

Added

  • Symmetric cell-list/verlet list Crossing scheme
  • VCluster examples
  • Cell-list crossing scheme

Fixed

  • CRITICAL BUG: OpenFPM has a bug handling decomposition when a processor has a disconnected domains (By experience this case has been seen on big number of processors).
  • Found and fixed a memory leak when using complex properties

Changed

  • The file VCluster has been mooved #include "VCluster.hpp" must be changed to #include "VCluster/VCluster.hpp" BECAUSE OF THIS, PLEASE CLEAN THE OPENFPM FOLDER OTHERWISE YOU WILL END TO HAVE 2 VCLUSTER.HPP

[0.6.0] - 5 November 2016

Added

  • Symmetric cell-list/verlet list
  • Multi-phase cell-list and Multi-phase cell-list
  • Added ghost_get that keep properties
  • Examples: 1_ghost_get_put it show how to use ghost_get and put with the new options 4_multiphase_celllist_verlet completely rewritten for new Cell-list and multiphase verlet 5_molecular_dynamic use case of symmetric cell-list and verlet list with ghost put 6_complex_usage It show how the flexibility of openfpm can be used to debug your program
  • Plotting system can export graph in svg (to be included in the paper)

Fixed

  • Option NO_POSITION was untested
  • Regression: Examples code compilation was broken on OSX (Affect only 0.5.1) (Internal: Added OSX examples compilarion/running test in the release pipeline)
  • gray_scott example code (variable not initialized)

Changes

[0.5.1] - 27 September 2016

Added

  • ghost_put support for particles
  • Full-Support for complex property on vector_dist (Serialization)
  • Added examples for serialization of complex properties 4_Vector
  • improved speed of the iterators

Fixed

  • Installation PETSC installation fail in case of preinstalled MPI
  • Miss-compilation of SUITESPARSE on gcc-6.2
  • vector_dist with negative domain (Now supported)
  • Grid 1D has been fixed
  • One constructor of Box had arguments inverted. PLEASE CAREFULL ON THIS BUG float xmin[] = {0.0,0.0}; float xmax[] = {1.0,1.0}; // Box<2,float> box(xmax,xmin) BUG IT WAS xmax,xmin Box<2,float> box(xmin,xmax) <--- NOW IT IS xmin,xmax Box<2,float> box({0.0,0.0},{1.0,1.0}) <---- This constructor is not affected by the BUG

Changed

  • On gcc the -fext-numeric-literals compilation flag is now mandatory

[0.5.0] - 15 August 2016

Added

  • map communicate particles across processors mooving the information of all the particle map_list give the possibility to give a list of property to move from one to another processor
  • Numeric: Finite Differences discretization with matrix contruction and parallel solvers (See example ... )
  • vector_dist now support complex object like Point VectorS Box ... , with no limitation and more generic object like std::vector ... (WARNING TEMPORARY LIMITATION: Communication is not supported property must be excluded from communication using map_list and ghost_get)
  • vector_dist support expressions (See example ...)
  • No limit to ghost extension (they can be arbitrary extended)
  • Multi-phase CellList
  • Hilber curve data and computation reordering for cache firndliness

Fixed

  • Removed small crash for small grid and big number of processors

Changed

Known Bugs

  • On gcc 6.1 the project does not compile
  • Distributed grids on 1D do not work

[0.4.0] - 26-05-2016

Added

  • Grid with periodic boundary conditions
  • VTK Writer for distributed vector, now is the default writer
  • Installation of linear algebra packages
  • More user friendly installation (No environment variables to add in your bashrc, installation report less verbose)

Fixed

  • GPU compilation
  • PARMetis automated installation
  • Critical Bug in getCellList, it was producing Celllist with smaller spacing

Changed

[0.3.0] - 16-04-2016

Added

  • Molacular Dynamic example
  • addUpdateCell list for more optimal update of the cell list instead of recreate the CellList

Fixed

  • Nothing to report

Changed

  • Eliminated global_v_cluster, init_global_v_cluster, delete_global_v_cluster, substituted by create_vcluster, openfpm_init, openfpm_finalize
  • CartDecomposition parameter for the distributed structures is now optional
  • template getPos<0>(), substituted by getPos()

[0.2.1] - 01-04-2016

Changed

  • GoogleChart name function changed: AddPointGraph to AddLinesGraph and AddColumsGraph to AddHistGraph

[0.2.0] - 2016-03-25

Added

  • Added Load Balancing and Dynamic Load Balancing on Beta
  • PSE 1D example with multiple precision
  • Plot example for GoogleChart plotting
  • Distributed data structure now support 128bit floating point precision (on Beta)

Fixed

  • Detection 32 bit system and report as an error
  • Bug in rounding off for periodic boundary condition

Changed

  • Nothing to report

[0.1.0] - 2016-02-05

Added

  • PSE 1D example
  • Cell list example
  • Verlet list example
  • Kickstart for OpenFPM_numeric
  • Automated dependency installation for SUITESPRASE EIGEN OPENBLAS(LAPACK)

Fixed

  • CRITICAL BUG in periodic bondary condition
  • BOOST auto updated to 1.60
  • Compilation with multiple .cpp files

Changed

  • Nothing to report