[llvm] [RISCV] Allow crypto features to imply dependents (PR #112659)

[Craig Topper via llvm-commits]

topperc wrote:

> > Do we need to change `RISCVISAInfo::checkDependency()`?
> 
> ...Probably? But because LLVM believes "negative features" are a sensible notion, I know there are some absolutely _wacky_ edge-cases in feature resolution which prevent me from feeling very confident in removing any code from that. The fact that in _normal_ usage the errors are simply naturally silenced by correctly resolving the features makes the 5 LOC that would be clawed back seem basically inconsequential.
> 
> Believe me, the thought of deleting more code greatly tempts me.

Let's go ahead and land this. I'll consider whether we can simplify `RISCVISAInfo::checkDependency()` separately.

https://github.com/llvm/llvm-project/pull/112659