Better support for SONATA and GPU bug fixes

• Bug fixes for multi-gpu execution (#330)
• Add support for sonata reports multipopulation (#334)
• Flush reports when having gap junctions (#342)
• Fixed for building on IBM Power 9 with GPU Support (#326)
• NMODL integration and build related fixes

Support for SONATA reports

• Support for SONATA reports (#225)
• Bugfix: num_segments() always returns 0 (#315)
• Update nmodl to latest master (#314)
• Add nmodl python lib to PYTHONPATH of nrnivmodl_core_makefile (#317)

Improved CLI with CLI11, use submodules and GPU build enhancements

Most of the improvements are related to the legacy code cleanup

• Improved Command Line with CLI11 (#179
• Code cleanup : #259, #257,
• Random123 as a submodule (#258)
• Fixes for NMODL/MOD2C binary usage through nrnivmodl-core
• Improvements for GPU build with PGI
• Improve GPU device selection in multi-gpu environment
• Update NMODL to latest master
• Bugfix : Support for using CoreNEURON in embedded mode with external mechanisms (#275)
• Fix issues while NMODL is built with sympy and analytical passes (#280)
• Disable vectorize in eigen in nmodl when gpu is enabled (#286)
• Refactoring setup for multisend method of spike exchange (#273)

NMODL integration and repo code reorganisation

• NMODL integration in CoreNeuron nrnivmodl-core workflow (#193)
• CoreNEURON CMake refactoring for NEURON integration (#198)
• Enable nmodl as submodule to CoreNEURON (#208)
• Enable GPU builds under Jenkins (#212)
• Gap junction fix (#213)
• Add travis CI configuration (#205)
• Add --pic default for ISPC compilation (#265)
• Refactoring / Cleanup of code : mainly scoping variables (#246)
• Coreneuron source code reorganization (#229)

Support for membrane current report and NMODL in nrnivmodl-core

• Membrane current report generation
• Support for membrane current calculation in NEURON and CoreNEURON bi-directional in-memory transfer mode
• NMODL integration in CoreNeuron nrnivmodl-core workflow

NEURON and CoreNEURON bi-directional in-memory transfer support

• spike sorting and forwardskip spike filtering improvement
• initial support for cuda unified memory
• added more ci and memory error fixes
• Bi-directional in-memory transfer between NEURON and CoreNEURON
• support for profiler specific instrumentors (Caliper, tau etc)
• ISPC target integration with NMODL

Support for nrnivmodl-core

Similar to nrnivmodl of NEURON, CoreNEURON now supports nrnivmodl-core.

Direct memory transfer support between NEURON and CoreNEURON

Allow using coreneuron as shared library with in-memory transfer between neuron and coreneuron.

Support for coreneuron library within NEURON

• Move to C++ (adding coreneuron as namespace)
• Mod2c now generates .cpp file
• Allows coreneuron library to be linked into NEURON (for single execution workflow)

Support for checkpoint-restart and multi-report

Major changes:

• support for multi-report (#74)
• checkpoint restart implementation (#76)
• support for vectorplay on GPU (#84)

Minor changes:

• centralize global variables definitions (#79)
• enforce c++ rules before migration (#80)
• centralise includes and extern declarations to visible for mod2c (#81)
• add ETA while execution of psolve (#87)

Bug fixes :

• OpenACC pragma annotation was missing for kernels using euler (#85)
• Bug fix in gpu implementation : global_vars.cpp was compiled with (#83)
• Bug fix: skip initial block for capac.c and eion.c (necessary for checkpoint-restore) (#77)

Also, this is last release before moving to C++.