I can program all my non-CUDA GPUs use completely open non-proprietary toolchains. And if that ceases to be the case on one platform I can switch platforms with

I can program all my non-CUDA GPUs use completely open non-proprietary toolchains. And if that ceases to be the case on one platform I can switch platforms without having to rewrite all my code.