Translating machine code between architectures is a dubious proposal in the first place. Taking compiled cuda shaders, disassembling them and recreating shaders

Translating machine code between architectures is a dubious proposal in the first place. Taking compiled cuda shaders, disassembling them and recreating shaders for some other architecture from the pieces really should be more effort than compiling the source directly to that other architecture.