4 ms·
Except that actually does apply to X86. As an example with AVX instructions you can process 8 floats at the same time, compared to 4 doubles. So if float is en
by sharpneli 13y ago
Except that actually does apply to X86.
As an example with AVX instructions you can process 8 floats at the same time, compared to 4 doubles. So if float is enough for you you can expect double performance in either memory transfer bound or ideally vectorizable algorithms.
And in mobile computer graphics 16bit values are common.