OpenBLAS 0.3.34 Improves Multi-Threading, Support For A Memory-Safe C Toolchain

OpenBLAS 0.3.34 released on Thursday as this popular, open-source BLAS library providing optimized support for a variety of CPUs/architectures. OpenBLAS 0.3.34 continues working on squeezing more performance out of today's processors as well as delivering other new features. The OpenBLAS 0.3.34 release has added or improved multi-threading thresholds for a number of kernels. Affected by these multi-threading improvements are SPR, SPR2, SYR, SYR2, TBMV, TPMV, GETRS, LAED3, LASWP, LAUUM, TRTRS, and ZTRTRI. OpenBLAS 0.3.34 also added support for compiling with Fil-C, which is an experimental LLVM-based C toolchain focused on memory safety. The Fil-C compiler fork of LLVM aims for providing a memory-safe implementation of C and C++. OpenBLAS 0.3.34 also brings changes to align with the upcoming Reference-LAPACK 3.13 release, various fixes, and a number of CPU/architecture specific optimizations. OpenBLAS now provides a "significantly faster" SGEMM kernel on ARM, more optimized ARM64 kernels, some newly-optimized kernels for RISC-V 64-bit, improved Zhaoxin x86_64 CPU checks, and some fixes for old AMD Opteron/Barcelona platforms. Downloads and more details on the new OpenBLAS 0.3.34 release at GitHub.