Invited to report on RISC-V calculations

The Xiamen Open Source Chip Industry Promotion Association is established under the joint guidance and support of the Xiamen Torch High tech Zone Management Committee and the Xiamen Jimei District Government, led by the China Science and Technology (Xiamen) Data Intelligence Research Institute. Its purpose is to provide high-quality enterprises in the integrated circuit industry with information, talent, technology, funds, and resources in the development, production, application, project investment and financing, resource docking and integration of open source chips Policies and other related support and services.

Time: March 7, 2022 Report Introduction With the expansion of the RISC-V Vector instruction set and the development of multi-core technology, RISC-V has gradually entered the server field, such as the recently launched 64 core RISC-V server CPU chip. In the field of high-performance computing, only chips and server hardware are not enough, and a rich computing software ecosystem needs to be built to support more applications running well and efficiently. This report introduces the relevant progress of RISC-V computing software stack and its future development goals, including OpenBLAS open source matrix computing library, OpenFFT library, OpenVML vector math library, heterogeneous computing framework PerfXAPI, and high-performance Python computing platform PerfXPy. In addition, the development of the RISC-V high-density server Hengshan series and the adaptation of computing software will also be introduced. ChatGPT introduced that OpenBLAS is an open source high-performance mathematics library based on BLAS (Basic Linear Algebra Subprograms), which provides a series of matrix operations and linear algebra functions, including vector addition, matrix multiplication, matrix decomposition, eigenvalue calculation, etc. The goal of OpenBLAS is to provide efficient mathematical computing capabilities to meet the needs of scientific computing, data analysis, machine learning, and other fields. The characteristics of OpenBLAS include: High performance: OpenBLAS uses various optimization techniques such as SIMD instructions, multithreading, etc. to improve computational efficiency. Cross platform: OpenBLAS supports multiple operating systems and hardware platforms, including Linux, Windows, Mac OS X, and more. Easy to use: OpenBLAS provides a simple and easy-to-use API that can be easily integrated into various programming languages. Open source: OpenBLAS is an open source project that anyone can use, modify, and distribute for free. OpenBLAS has a wide range of applications, including scientific computing, data analysis, machine learning, image processing, and other fields. In many open source projects, OpenBLAS is also a commonly used math library, such as NumPy, SciPy, Matlab, etc.

A certain customer: Although Intel's MKL is easy to use, its drawbacks are obvious: 1 Our software and algorithms will only run on x86 and cannot run on the client's ARM and RISC-V hosts. If we need to make targeted adjustments to the performance of the computing library, MKL closed source cannot allow you to modify it, but OpenBLAS can. Thank you to OpenBLAS for providing us with another option, which is also very stable and has excellent performance.