News for hackers. Stuff that matters.

A comment on Exploiting Vector Instructions with Generalized Stream Fusion [pdf].

Re:Exploiting Vector Instructions with Generalized Stream Fu... by im3w1l on Friday December 22, 2017 @02:33PM (#15990211)
> Benchmarks show that high-level Haskell code written using our compiler and libraries can produce code that is faster than both compiler- and hand-vectorized C.

Big, if it holds up.

[ Parent | on HN ]
Re:Exploiting Vector Instructions with Generalized Stream Fu... by Veedrac on Friday December 22, 2017 @05:00PM (#15991254)
We're talking <10% difference on one particular toy example exclusively on arrays at least ¼MiB large.

Given how similar the code produced is from both compilers (literally just movapd-mulpd-addpd), the only relevant difference seems to be usage of prefetching, which C compilers don't do much because the common wisdom is that it hurts in most scenarios.

[ Parent | on HN ]