Not memory speed but extremely fast: https://github.com/powturbo/TurboHist

vardump · on Nov 6, 2019

Yeah, it's nice, about what you can achieve. I'd be happy if it was about 8 times faster. Less than 0.2 cycles per byte would be good.

Unfortunately, that's just not achievable with current x86 instruction set.

powturbo · on Nov 6, 2019

There is new AVX512 instruction (_mm512_conflict_epi32) supposed to solve this, but it can't make the histogram construction faster than the scalar functions.