[llvm] [Exegesis] CPU selection, when native arch and target mismatch (PR #131014)

[via llvm-commits]

AnastasiyaChernikova wrote:

> > In case of cross-compilation select 'generic' by default and raise an error if user manually specified 'native'. Otherwise - in case of host compilation, set cpu to 'native' by default.
> 
> This still doesn't address the problem that I have with the current approach. If we are not cross compiling, we know the target CPU uarch because we can just query it. If we are cross compiling, then we do not know the target CPU. Maybe things outside of analysis mode will mostly work fine with a generic CPU, but I see no reason why we shouldn't require the user to specify an exact CPU (ie error out if `-mcpu` is not specified when cross compiling). This probably improves the exegesis user experience. The user is required to specify a CPU which takes a little bit of extra effort, but won't be wondering about issues like why certain ISA extensions are suspiciously not getting any coverage.

Okay, I agree. Replaced with error output.

https://github.com/llvm/llvm-project/pull/131014