a perlin noise simd implementation (processes 8 positions at once)
the benchmark consists of generating size * size noise values and assigning them to an array of image data
size testing range goes from 8 -> 4096 in x2 increments, each size is ran 100 iterations for sisd and simd to get a more accurate result
results seem to vary a bit between runs, but on average simd seems to be about 130% faster than the sisd implementation