Optimizing 128-bit Division (2020)

This is not directly relevant, but Power ISA 3.1 (from May 2020) has 128 bit division operations (vdivuq and vdivsq) with a maximum latency of 61 cycles. I don&#x27;t have access to a Power 10 machine to see how it compares to what&#x27;s presented here, but I thought it was an interesting addition to the ISA.<p><a href="https:&#x2F;&#x2F;wiki.raptorcs.com&#x2F;w&#x2F;images&#x2F;f&#x2F;f5&#x2F;PowerISA_public.v3.1.pdf" rel="nofollow">https:&#x2F;&#x2F;wiki.raptorcs.com&#x2F;w&#x2F;images&#x2F;f&#x2F;f5&#x2F;PowerISA_public.v3.1...</a> <a href="https:&#x2F;&#x2F;files.openpower.foundation&#x2F;s&#x2F;EgCy7C43p2NSRfR" rel="nofollow">https:&#x2F;&#x2F;files.openpower.foundation&#x2F;s&#x2F;EgCy7C43p2NSRfR</a>

Wasn&#x27;t there a nice post somewhere talking about using anything but division for the task of splitting stuff into buckets?

Some major blunders in the first paragraphs. Iterating over all 32 bits is trivial, over 64 bit not possible in normal time. She mixed up the numbers there.<p>Thus I don&#x27;t see the real need to extend to 128 bits for the birthday paradox. At Google scale there might be, but not for pedestrians.