OpenBLAS 0.3.21 Brings Support For Newer Arm CPUs, More Optimizations

OpenBLAS as the high performance, open-source BLAS / LAPACK implementation debuted a new version on Sunday with more CPU optimizations and expanded processor coverage.

OpenBLAS 0.3.21 features support for more CPUs, particularly on the Arm side. There is also support for more compilers, build system enhancements, and more. The OpenBLAS 0.3.21 release highlights include:

- Support for building OpenBLAS with the Intel IFX, Fujitsu FCC, and Cray C/Fortran compilers are now supported.

- Initial support for Zhaoxin/Centaur KH40000 processors.

- OpenBLAS' CMake build system now supports cross-compiling for individual Intel and AMD x86_64 targets. There are now CMake targets exposed ranging from Intel Prescott to Sapphire Rapids and then on the AMD side from Barcelona to Zen.

- Various IBM POWER fixes, including a number of Power10 fixes. The OpenBLAS POWER build now also enables compiling the BFLOAT16 kernels by default.

- OpenBLAS RISC-V CPU auto-detection logic has been fixed.

- An SBGEMM kernel for the Arm Neoverse-N2 is added.

- Support for 64-bit Arm systems running Microsoft Windows.

- Initial support for the Apple M1 CPU on Linux.

- Initial support for the Phytium FT2000 processor.

- Initial support for the Arm Cortex A510 / A710 / X1 / X2 processors.

- Fixing compilation of OpenBLAS on various x86_64 CPU targets under different conditions.

- Initial support for the Loongson 2K1000 processor.

All the details and downloads via OpenBLAS on GitHub.