I am trying to understand why LLVM has the concept of a "vector type". Could LLVM have been designed without one and could vectorization/SIMD be implemented on top of the array type?
It seems to me that a [8 x float] could be mapped to a vector register in the same way as a <8 x float>.