2^i should be optimized to 1 << i

[hadronized]

Hello,

I came across a weird situation with powers of 2. They’re not optimized away to left shifts. Here’s a Godbolt example.

If you think it should / can be addressed, I’d love to do it! :)

[jonas-schievink]

u32::pow uses hand-rolled exponentiation by squaring, which LLVM probably can't "see through"

[kennytm]

Duplicate of #47234.