https://github.com/skywind3000/FastMemcpy/blob/8fea5f666be174c6548d0ae4010e81b0a742c853/FastMemcpy.h#L644 Hi, it bugs me as 128 seems to be a reasonable choice. Is that derived from experiments? Or something related to the mechanism of prefetching itself?