Charm++ parallel programming system

["Bohm, Eric J" <ebohm AT illinois.edu>]

We are proud to announce the (pre) release of Charm++ 8.0.0.

https://charm.cs.illinois.edu/software

We welcome feedback regarding your experiences with Charm 8.0.0 via email, or github discussion.

https://github.com/charmplusplus/charm/discussions/3823

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

What's new in Charm++ 8.0.0

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

This is a feature release, with the following major changes:

Highlights:

- The License has changed to Apache 2.0 with LLVM exception.  This

  change to a popular Open Source license is intended to simplify use,

  collaboration, and greater community involvement in the development

  of Charm++. The NOTICE file contains the pertinent disclaimers.

- The CkIO library (previously only supporting file output) has been

  enhanced to support file input. The input layer enables two-phase,

  collective input from a single file via an array of buffer chares,

  which read from the file system and buffer data until the application 

  requests it. As of this release, the number of buffer chares is 

  not automated by CkIO, and the user is responsible for selecting 

  the input decomposition for best performance.

- The OFI layer has been extended to support the CXI (Cassini)

  extensions for Slingshot-11.  This can be accessed by adding the cxi

  parameter to the build line and allows for greatly improved

  performance on machines such as (Frontier, Perlmutter, Delta) at

  larger node counts.  Adding cxi as a build parameter is not

  necessary on most Slingshot-11 installations, as it will be

  autodetected.

- Added support for NVIDIA's nvc/nvfortran compilers and Intel's new

  icx/ifx compilers.

- Fixed a bug in location management when doing dynamic insertion and

  deletion of chare array elements.

- Deprecated the "atomic" keyword in SDAG code (in favor of "serial").

- Added support for TLSglobals on Apple ARM systems and the ability to

  disable TLSglobals support at build-time.

- Performance optimizations for node group broadcasts.

- Added CMI-SHMEM module for optimizing small/medium-sized messages

  between processes on the same host.

- Fixed demand creation via ckcallback::send and enabled passing

  options to ckcallback::send.

- Improved portability and usability of AMPI's automatic global

  variable privatization methods.

- Transferred organization ownership of the repository on github from

  UIUC-PPL to charmplusplus.