CHANGES

# changes since the last release

# 18.03

  -- STL support for EB

# 18.02

  -- Fortran interface: added new functions amrex_multifab_build_alias
     and amrex_imultifab_build_alias.

  -- Documentation in reST/Sphinx.  See Docs/Readme.sphinx for more
     details.

  -- Src/LinearSolvers/MLMG now supports nodal based data.

  -- Fortran module AMReX_fillpatch_module takes a user provided
     callback function for filling physical boundary conditions.  That
     function used to use 0-based index for component (i.e., the last
     index).  For consistence with 1-based index for component in
     other Fortran modules, this has been changed to 1-based index for
     component.

# 18.01

  -- New linear solver MLMG at Src/LinearSolvers/MLMG/.  Currently it
     supports cell-centered single level and multi-level composite
     solve.  Fortran interface is also available at
     Src/F_Interfaces/LinearSolvers/.  See Tutorials/LinearSolvers for
     tutorials.

# 17.12

  -- Ported more features (including dynamic scheduling and explicit
     tile size) of MFIter from C++ to Fortran.

  -- Added procedures to Fortran amrex_multifab_module to add,
     subtract, multiply, divide, saxpy and lincomb (linear
     combination).  Note that we call it saxpy, but it doesn't mean
     single precision.

  -- The `configure; make; make install` build approach enables linear
     solvers by default now.

  -- MultiFab and iMultiFab now support move assignment operator.
     This allows for swapping two MultiFabs without deep copying.  For
     instance, std::swap(mf1, mf2);

  -- Tutorials/Amr/Advection_AmrCore changes from using
     `Vector<unique_ptr<MultiFab>>` to `Vector<MultiFab>`.

# 17.11

  -- AMREX_ASSERT_WITH_MESSAGE and AMREX_ALWAYS_ASSERT_WITH_MESSAGE
     are new macros for assertion when it fails additional message
     will be printed.   For example:

         AMREX_ASSERT_WITH_MESSAGE(x > y, "reason why x > y must be true");

  -- amrex::Array is deprecated and replaced by amrex::Vector.
     Currently Array is an alias to Vector.  So this should not break
     any codes.  In the future, we will remove Array completely.
     Therefore it is recommended that application codes should start
     replacing Array with Vector.  There is a script,
     amrex/Tools/Migration/amrex_array_to_vector.sh, to help.  But use
     it with caution.   If it is completely gibberish to you, you
     probably shouldn't use the script. 

  -- functions for printing to a file in a parallel and thread-safe manner
     have been added. For example:

        amrex::AllPrintToFile("output") << "Hello from rank " << rank << 
                                           " and thread " << thread << "\n";

     will print out the requested information to separate files for each 
     rank and thread.

# 17.10

  -- Added a load balance approach to Amr/AmrLevel.  The runtime
     parameters for this is amr.loadbalance_with_workestimates and
     amr.loadbalance_level0_int.  The latter is for single level
     (i.e., amr.max_level=0) only.  For multi-level runs, load balance
     is done during regrid and thus the load balance interval is
     controlled by amr.regrid_int.  To use this approach, AmrLevel
     derivative must has a StateData for work estimates and its index
     is returned by virtual function int WorkEstType().

  -- MFIter now supports dynamic scheduling of OpenMP threads.  For
     example,
 
       for (MFIter mfi(mf, MFItInfo().SetDynamic(true).EnableTiling(tile_size);
            mfi.isValid(); ++mfi)
       { ... }

  -- Added EBFluxRegister to Src/EB for refluxing and
     re-redistribution in EB.

  -- amrex.signal_handling = 1 is a new runtime parameter that can be
     used to control whether AMReX should handle signals like SIGSEGV
     etc.