OpenBLAS 0.3.32 is now available for this optimized open-source Basic Linear Algebra Subprograms “BLAS” library. Notable with the OpenBLAS 0.3.32 release is improving CPU auto-detection for newer Intel processors.
OpenBLAS 0.3.32 adds automatic detection for Intel Arrow Lake H/U processors as well as the new Core Ultra Series 3 Panther Lake processors. Intel Jasper Lake detection was also missing but now in place as was Xeon Emerald Rapids support too. Plus code restructuring is making the auto detection more aligned for future Intel CPU models.
The new OpenBLAS release also updates its CPU cache size translation table, improved CPU detection fallback for older non-AVX Intel CPUs, and brings fixes for the GCC 15 compiler and other x86_64 fixes.
OpenBLAS 0.3.32 also brings fixes and optimizations for RISC-V 64-bit, POWER, LoongArch64, and ARM64. On the ARM64 side is now having auto-detection for Cortex A725 and X925 cores as well as initial support for Neoverse V3 CPUs. There is also new optimized kernels for Neoverse N1 and improved kernel performance for Neoverse V2.
The OpenBLAS 0.3.32 release also improved its WebAssembly (WASM) target with optimized intrinsics kernels for SGEMM and DGEMM. There are also various other fixes and enhancements as outlined with the GitHub project announcement.
