OpenBLAS 0.3.18 brings a wide assortment of improvements and fixes, some of the highlights include:
- Support for LoongArch (LoongArch64) as the new Loongson MIPS-based CPU architecture.
- Intel Skylake X S/DGEMM kernels for small problem sizes.
- Optimized SBGEMM for Intel Cooper Lake processors.
- Support for run-time CPU core/thread count detection on Haiku OS.
- Better performance for DASUM on IBM POWER10.
- Crash fixes for Apple M1 on macOS.
- Fixed building under BSDs on MIPS/MIPS64.
- OpenBLAS at run-time when exceeding the build-time set number of pre-configured threads, an auxiliary control structure will be allocated with up to an additional 512 threads. The current/prior behavior of OpenBLAS was aborting when exceeding the pre-configured thread count.
Downloads and more details on OpenBLAS 0.3.18 at GitHub .