Most proof-of-work cpu miners use this optimization. My own Cuckoo Cycle cpu solver supports 4-way, 8-way, and even 16-way siphash computation, where the latter

Most proof-of-work cpu miners use this optimization. My own Cuckoo Cycle cpu solver supports 4-way, 8-way, and even 16-way siphash computation, where the latter use multiple sets of avx registers to allow for more interleaving of operations. On recent iCore i7, the 8-way performs best. The Makefile at https://github.com/tromp/cuckoo/blob/master/src/Makefile https://github.com/tromp/cuckoo/blob/master/src/Makefile has targets cuckoo28stx, cuckoo28stx4, cuckoo28stx8 and cuckoo28stx16 for these different levels of parallellization.