Release v2.9.1

Version 2.9.1

Release v2.9.0

Alfio Lazzaro:
Switch MacOS CI to use MPICH
Update GCC to v15
Install ACC header only for the OpenCL backend
Remove G2G in ROCM (it is enabled), add MPI
Fix FORD documentation for notes
Conditional C/C++ Standard requirements
Add CUDA and HIP C++ Standard
Fix issue on Fedora due to missing FindBlas/Lapack
Add Flang compiler flags

Cristian Le:
Add Fedora packaging via packit

Eisuke Kawashima:
build: add flags for LLVMFlang
build: add flags for IntelLLVM (#928)

Gina Sitaraman:
Add linking with hiprtc library, needed for ROCm 7.x, unused in older ROCm versions
Add gfx950 (Mi350) GPU architecture support
Add Mi300 arch

Hans Pabst:
Fixed looking-up bogus environment variables (#915)
HIP:
- Removed unused hipHostAllocDefault
- Fixed compilation error
CMake cleanup:
- Removed customized/diverged modules (LAPACK/BLAS)
- Allow USE_SMM=auto (#889)
OpenCL:
- Allow CP2K to use DBCSR for DBM (and the other way around)
- Improved matching a device with a parameter set
- Improved script for auto-tuning of parameters
- Updated tuned parameters (Mi250, GH200, PVC, BMG)
- Fixed leaking event objects
LIBXSMM: improved supporting shared library

Mathieu Taillefumier:
CMake: proper handling of the different language standards

Ole Schütt:
Check block sizes are equal in dbcsr_copy_into_existing_sym

Release v2.8.0

Regular end-of-the-year release, with some minor updates and refactoring (mostly due @hfp and @gsitaram ).

Release v2.7.0

Main changes in this release:

• improve OpenCL backend (thanks to @hfp )
• include H100 kernels (thanks to @abussy )
• introduce experimental G2G support (thanks to @gsitaram )
• Initialize member variables of derived types (thanks to @fstein93 )
• some minor bug fixes

Release v2.7.0-rc2

This release introduces a generic kernel if tuned kernels are not available.

Release v2.7.0-rc1

Initial G2G MPI (still experimental), thanks to @gsitaram .
Several OpenCL improvements, thanks to @hfp

Release v2.6.0

• Introducing more A100 kernels
• Improving OpenCL backend
• Support for MPI F08 API

Release v2.5.0

• Support for A100
• Drop support for MPI v2 standard (now at least v3 is required)
• Introducing environment variables to set DBCSR parameters
• Several improvements in the OpenCL backend

Release v2.4.1

Version 2.4.1

Release v2.4.0

Support of AMD Mi250 GPU