[llvm][Support] Replace `%` operator with `&` in `Align::isAligned(...)`

[Stoorx]

Summary

Since the function Align::value() ALWAYS returns power of 2,
it is easier to check reminder by conjunction with mask, which is (Pow2Value - 1).

[fhahn]

it is easier to check reminder by conjunction with mask, which is (Pow2Value - 1).

Hmm, easier in what respect? Isn't the original code more straight-forward?

[Stoorx]

easier in what respect?

I mean, easier for CPU :-)

The division operation reasonably seems to be more complex than bitwise AND. And since this function is pretty hot it can save a bit of performance.
(Actually that's just my assumption. I did not make benchmark really.)

Side thought: I suspect some advanced CPUs can optimize such kind of divisions by powers of 2 by themselves.

llvm/include/llvm/Support/Alignment.h

public:
   }
 
   /// For convenience, returns a valid alignment or 1 if undefined.
   Align valueOrOne() const { return value_or(Align()); }
 };
 
 /// Checks that SizeInBytes is a multiple of the alignment.
 inline bool isAligned(Align Lhs, uint64_t SizeInBytes) {
-  return SizeInBytes % Lhs.value() == 0;
+  return !(SizeInBytes & (Lhs.value() - 1));
 }
 
 /// Checks that Addr is a multiple of the alignment.
 inline bool isAddrAligned(Align Lhs, const void *Addr) {
   return isAligned(Lhs, reinterpret_cast<uintptr_t>(Addr));
 }
 
 /// Returns a multiple of A needed to store `Size` bytes.