To support OS with SMP (Symmetric Multiprocessing), besides Linux SMP which has been well supported in RISC-V upstream and Andes platforms, Andes is now offering the first RISC-V port for SMP Zephyr RTOS and Zephyr’s driver subsystem.
Verified on Andes RISC-V multicore platform developers will now only need to focus on the applications themselves without worrying about the underlying system software. As to AMP (Asymmetric Multiprocessing) demands, AndeSight integrates the OpenAMP which provides communication infrastructure between heterogeneous systems and enables AMP applications to leverage parallelism offered by multiprocessor systems.
AndeSight provides the user-friendly multicore debugging feature for both SMP and AMP systems in one IDE interface. It is also offering offers a new feature “Core Grouping” which send debug commands to a set of cores in the same debug session. AndeSight also offers the record-and-replay scripting capability to save interactive steps for easy issue reproduction and automatic testing.
The RVP extension enables multiple data in integer registers to be processed in one single cycle, helping to efficiently boost the performance with low power consumption for TinyML, AIoT, and signal processing applications on edge and endpoint. Moreover, the RVV extension targets high-volume data computations with a configurable vector processing architecture.
To simplify the software development of powerful ISA extensions, AndeSight also offers full support from toolchains for RVP and RVV, their respective intrinsic functions, highly optimised DSP and Vector libraries, and sample codes to guide code optimisation. In addition, AndeSight IDE v5.1 enables auto-vectorization to generate RVV instructions automatically by the compiler.
Users will be able to dramatically speed up the development of Neural Network algorithms with the AndeSoft NN Library. The NN library is optimised for RVP and RVV instructions for INT8 and FP16 data types. It also supports several quantization methods: shift-based, symmetry, and asymmetry. For MobileNet-v1 inference using NN library with FP16 data type, the NX27V processor with 512-bit SIMD width and 512-bit vector length achieves a 96x speedup over itself executing only RISC-V baseline extensions.
Moreover, TensorFlow Lite for Microcontroller can execute TensorFlow models with AndeSoft NN Library on development boards.
“The AndeSight IDE v5.1 simplifies software development with handy features and optimised tools and helps accelerate the completion of highly competitive products,” said Andes President and CTO Dr. Charlie Su. “Processors cannot deliver their full capabilities without the matching software solutions. With comprehensive AndeSight IDE, customers can release their software with more features, better performance and higher quality in a shorter time.”
