> Or is an ISA more than just a different 'language'?
It tends to be more like going from C89 to Haskell. You're not just switching the keywords around, but also fundamental architectural concepts. There's still some parts you can recycle and some skills that transfer, but less than you'd like.
> Or is hardware not really the biggest problem, but rather Software like compilers, kernels, etc.?
That's the next problem. Kernels, device drivers, support hardware, a lot of low level stuff needs to be adapted, and even a company the size of Qualcomm doesn't necessarily do everything inhouse, there will be lots of external IPs involved and all those partners need to also be willing to migrate over to a different ISA.
I'd perhaps put ARM vs RISC-V closer to like C# vs Java (or maybe JS vs Python) - on a high level rather similar, but a majority of the specifics don't really line up (especially at the uop level requiring dumping a bunch of previous optimized cases, and adding a bunch of new ones).
It tends to be more like going from C89 to Haskell. You're not just switching the keywords around, but also fundamental architectural concepts. There's still some parts you can recycle and some skills that transfer, but less than you'd like.
> Or is hardware not really the biggest problem, but rather Software like compilers, kernels, etc.?
That's the next problem. Kernels, device drivers, support hardware, a lot of low level stuff needs to be adapted, and even a company the size of Qualcomm doesn't necessarily do everything inhouse, there will be lots of external IPs involved and all those partners need to also be willing to migrate over to a different ISA.