Vectorized CLZ and CTZ

Automated excerpt

By substituting a 32-bit number x into the mantissa of some value 2k, we get a double representing 2k(1+2−52x). When latency-bound, the numbers rise to 2 ns vs 1 ns (but if you’re latency-bound on vectorized ctz, you’re probably doing something wrong). Ian Qvist tested this on Alder Lake (thanks! ) and got 0. 29 ns/iteration, compared to 0. 85 ns for the scalar version, and 1. 3 ns vs 0. 85 ns when latency-bound. On Alder Lake, it’s 0. 35 ns/iteration and 1. 3 ns when latency-bound.

Selected automatically from source text; not independently written or fact-checked. Read the original for full context.