2 ms·
The loop was unrolled with a large factor. The loop was also vectorized, so that each multiply is a 4-element vector multiplication, with the final result bein
by jcranmer 10y ago
The loop was unrolled with a large factor.
The loop was also vectorized, so that each multiply is a 4-element vector multiplication, with the final result being a horizontal reduction.
The rest of the mess is trying to account for all of the cases where the input power is not a multiple of 128.