[llvm] [RISCV] Introduce a new tune feature string syntax and its parser (PR #168160)

[Sam Elliott via llvm-commits]

lenary wrote:

> Just had some discussions with Craig on naming and we're wondering: "no-no-xyz" sounds weird and so does "no-disable-xyz". Should we either give those features a better name or using "-" (and potentially "+" for positive features) instead of "no-"? This won't impact the core part of this PR, but just curious about folks' naming preferences.

Double negatives are very confusing, you're right.

We probably want:
```
  # Ones already with `no-` prefix lose it for the positive version
  -mtune=<cpu>:no-default-unroll # causes no-default-unroll
  -mtune=<cpu>:default-unroll   # removes no-default-unroll

  # Ones with `disable-`:
  -mtune=<cpu>:x    # removes disable-x
  -mtune=<cpu>:no-x # causes disable-x
```

We might have to have a new tablegen class and instances, to make this happen, rather than using the existing SubtargetFeatures (and their currently-internal names). I don't have a problem with adding that, if it means the interface can be clearer.


https://github.com/llvm/llvm-project/pull/168160