[clang] [X86][clang-cl] Add CL option /vlen (PR #166375)

[Hans Wennborg via cfe-commits]

================
@@ -9218,6 +9218,10 @@ def : CLFlag<"Qscatter-">, Alias<mno_scatter>,
 
 def _SLASH_arch : CLCompileJoined<"arch:">,
   HelpText<"Set architecture for code generation">;
+def _SLASH_vlen : CLCompileJoined<"vlen">,
+  HelpText<"Set vector length for autovectorization and other optimizations">;
+def _SLASH_vlen_default : CLFlag<"vlen">,
+  HelpText<"Set default vector length for autovectorization and other optimizations">;
----------------
zmodem wrote:

I think these should be `_SLASH_vlen` (for the default) and `_SLASH_vlen_EQ` for the one ending in = (and that should be included in the flag spelling.

But, since only two values are allowed, it might be simpler to do something like:

```
def _SLASH_vlen : CLFLag<"vlen"> ...
def _SLASH_vlen_EQ_256 : CLFLag<"vlen=256"> ...
def _SLASH_vlen_EQ_512 : CLFlag<"vlen=512"> ...
```

Does Clang provide good diagnostics when passing an `-mprefer-vector-width` that doesn't fit the target architecture?

If so, could we make `_SLASH_vlen_EQ_256` an alias for `-mprefer-vector-width=256` and so on, and not have to add any new driver logic at all?

https://github.com/llvm/llvm-project/pull/166375