Skip to content

Comment on Summing ASCII encoded integers on Haswell at almost the speed of memcpyparent

Comments

Intel's inconsistent SIMD support across its x86 chips seems crazy to me.

It's already a hard sell to get programmers to write architecture -specific SIMD code.

But Intel then made different subsets of SIMD instructions available on different models. And muddied the terminology by making "AVX-512" be ambiguous.

Maybe it was somehow a win in terms of profitability, but it sure made me reluctant to write code to exploit the new instructions.

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.