[AArch64] Use string comparison for ArchInfo equality

[tmatheson]

Summary

Tests are failing with LLVM_LINK_LLVM_DYLIB because copies of ArchInfo are
being created somehow, so using address for comparison is failing. Change
to use string comparison based on the name until I can figure out what the
issue with address comparison is.

[Hahnfeld]

I can confirm it solves the failing tests I'm seeing with LLVM_LINK_LLVM_DYLIB, but I'm not versed in this part of the code or the recent changes.

[mgorny]

Well, it fixes the problem for me as well. I don't know the code but FWICS you've added the original assumption, so I guess you know best.

My personal preference would be to revert the change and combine this into the original patch but I'm happy with whatever fixes the problem.

[tmatheson]

Landed as part of f4225d325c19ae0e5dbe39faa900d81e24559da0

llvm/include/llvm/Support/AArch64TargetParser.h

struct ArchInfo {
   uint64_t DefaultExts;  // bitfield of default extensions ArchExtKind
 
   // These are not intended to be copied or created outside of this file.
   ArchInfo(const ArchInfo &) = delete;
   ArchInfo(const ArchInfo &&) = delete;
   ArchInfo &operator=(const ArchInfo &rhs) = delete;
   ArchInfo &&operator=(const ArchInfo &&rhs) = delete;
 
-  // Comparison is done by address. Copies should not exist.
-  bool operator==(const ArchInfo &Other) const { return this == &Other; }
-  bool operator!=(const ArchInfo &Other) const { return this != &Other; }
+  bool operator==(const ArchInfo &Other) const {
+    return this->Name == Other.Name;
+  }
+  bool operator!=(const ArchInfo &Other) const {
+    return this->Name != Other.Name;
+  }
 
   // Defines the following partial order, indicating when an architecture is
   // a superset of another:
   //
   //     v9.4a > v9.3a > v9.3a > v9.3a > v9a;
   //       v       v       v       v       v
   //     v8.9a > v8.8a > v8.7a > v8.6a > v8.5a > v8.4a > ... > v8a;
   //