Basic Linear Algebra Subprograms (BLAS) has emerged as a de-facto standard interface for libraries providing linear algebra functionality. The advent of powerful devices for Internet of Things (IoT) nodes enables the reuse of existing BLAS implementations in these systems. This calls for a discerning evaluation of the properties of these libraries on embedded processors. This work benchmarks and discusses the performance and memory consumption of a wide range of unmodified open-source BLAS libraries. In comparison to related (but partly outdated) publications this evaluation covers the largest set of opensource BLAS libraries, considers memory consumption as well and distinctively focuses on Linux-capable embedded platforms (an ARM-based SoC that contains an SIMD accelerator and one of the first commercial embedded systems based on the emerging RISC-V architecture). Results show that especially for matrix operations and larger problem sizes, optimized BLAS implementations allow for significant performance gains when compared to pure C implementations. Furthermore, the ARM platform outperforms the RISC-V incarnation in our selection of tests.
JavaScript jest wyłączony w Twojej przeglądarce internetowej. Włącz go, a następnie odśwież stronę, aby móc w pełni z niej korzystać.