First: we need to finally stop the harmful myth that floating point multiplication is slower than addition. This has not been true for a long while. Second: wh

First: we need to finally stop the harmful myth that floating point multiplication is slower than addition. This has not been true for a long while. Second: why are so many people insisting that the loop is auto-vectorised? Is there any evidence to that? Data dependencies alone explain the observed performance delta. Auto-vectorization would have resulted in a higher speedup.