changing the inline representation of RawSmallVec allows us to use array indexing which is faster than direct pointer arithmetic
some time ago I built a prototype for a custom array stored in the stack that was evaluable at compile-time, it's usually a few percentage points faster (code here)
it seems it's because llvm optimizes it better but I have no idea
it makes handling it much easier as well
changing the inline representation of
RawSmallVecallows us to use array indexing which is faster than direct pointer arithmeticsome time ago I built a prototype for a custom array stored in the stack that was evaluable at compile-time, it's usually a few percentage points faster (code here)
it seems it's because llvm optimizes it better but I have no idea
it makes handling it much easier as well