ZCC toolchain from Terapines has fully supported SG2042 RISC-V chip
For More Info, Visit ZCC Introduction
Terapines have also benchmarked some popular AI kernels in addition to SPECInt2006. As shown in below, we have evaluated the auto-vectorization performance of zcc, LLVM, EPI LLVM, GCC against 4 kernels includes psroi, warp, resize and correlation, we have also compared the dynamic instruction count agaist hand optimized kernels written with RVV builtins. zcc shows 18% average better performance against hand optimized kernels, at most 10 times faster than EPI LLVM (9cfcff6873), 100 times faster than LLVM 16 and 69 times faster than GCC 12.2. You can find the source code and build commands of these 4 kernels from https://github.com/dodohack/rv_lib.
If you wish to have a try of zcc toolchain, please send mail to [email protected]