> support every single stupid thing (among thousands) and even mimic some of the bugs so that models work "the same".

Do you need to do the stupid things performantly, though? Because that sounds like a case for skipping microcode shims, and going straight to instructions that trap into a software implementation. Or just running the whole compute-kernel in a driver-side software emulator that then compiles real sub-kernels for each stretch of non-stupid GPGPU instructions, uploads those, and then calls into them from the software emulation at the right moments. Like a profile-guided JIT, but one that can't actually JIT everything, only some things.