I find the example is incredibly confusing. I expected this to be an optimization of a dot product or matrix multiplication or similar but as it stands it seems

I find the example is incredibly confusing. I expected this to be an optimization of a dot product or matrix multiplication or similar but as it stands it seems to be an element-wise square of a VECTOR_LEN * N matrix with subsequent accumulation of columns (not the most common operation in typical signal processing). It should be also noted that this can be trivially implemented in direct SIMD using GCC/clang vector extensions, without having the compiler 'guess' the SIMD part, which btw would be also make the code much easier to read.