Nvidia's CUDA framework enables GPU programming and is designed for use with C, C++, and Fortran programming languages. Our goal is to make this framework accessible in the Go programming language as well. To achieve this, we have developed library wrappers and created the CudaGo source-to-source compiler. CudaGo operates via the command line, converting CUDA C++ kernels into a format compatible with Go. Performance analysis shows that our solution delivers execution speeds comparable to traditional methods while introducing only minimal overhead.
|