[Alignment][NFC] Optimize alignTo

[gchatelet]

Summary

A small optimization suggested by jakehehrlich@ in D64790.

[courbet]

llvm/include/llvm/Support/Alignment.h

174

IIUC, -value is just the same as ~(value - 1). but only works if the representation two's complement. C++ does not guarantee two's complement. What about using ~(value - 1) instead ?

https://godbolt.org/z/8KJBV3

[gchatelet]

• Address comments

llvm/include/llvm/Support/Alignment.h

174

Good catch.

llvm/include/llvm/Support/Alignment.h

 
 /// 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.
 inline uint64_t alignTo(uint64_t Size, Align A) {
-  return (Size + A.value() - 1) / A.value() * A.value();
+  const uint64_t value = A.value();
+  // The following line is equivalent to `(Size + value - 1) / value * value`.
+
+  // The division followed by a multiplication can be thought of as a right
+  // shift followed by a left shift which zeros out the extra bits produced in
+  // the bump; `-value` is a mask where all those bits being zeroed out are just
+  // zero.
+
+  // Most compilers can generate this code but the pattern may be missed when
+  // multiple functions gets inlined.
+  return (Size + value - 1) & -value;
 }
 
 /// Returns a multiple of A needed to store `Size` bytes.
 /// Returns `Size` if current alignment is undefined.
 inline uint64_t alignTo(uint64_t Size, MaybeAlign A) {
   return A ? alignTo(Size, A.getValue()) : Size;
 }