RISC-V向量指令集的Compute Library函数库移植

ARM Compute Library是一类针对ARM Cortex-A系列CPU处理器和ARM Mali系列GPU特定优化的软件算法函数库,内部实现了卷积滤波器、卷积神经网络等算法,并且使用Cortex-A CPU NEON、Mali GPU的SIMD技术加速算法运行.RISC-V指令集作为一种开源的指令集,目前发布...

Transplantation of Compute Library of RISC-V Vector Instruction Set

ARM Compute Library is a kind of software algorithm function library specifically optimized for ARM Cortex-A series CPU processors and ARM Mali series GPUs.It internally implements algorithms such as convolution filters and convolutional neural networks,and uses Cortex-A CPU NEON or Mali GPU to accelerate algorithm operation.As an open source instruction set,the RISC-V instruction set has released a relatively stable version of the SIMD instruction set,and C-SKY has open sourced GCC and QEMU that support v0.7.1 intrinsics.On these foundations,this article attempts to port the ARM Compute Library function library to support the RISC-V vector instruction set.The core idea of function migration is to write a header file and use macro definitions without modifying the source files to replace the ARM NEON vector type interface one by one with the vector type and vector function interface defined in RISC-V intrinsics.