On 12/16/2016 07:01 PM, Ondrej wrote: > Hey, > I had int as a key in my map and later realised I only used 100 or so > values, so swapped it for a uint8, expecting a minor speed increase, but > saw a major slowdown instead. To recreate this, I tried this simple > microbenchmark and was surprised by the difference between (u)int8/16 > and int/int32/int64. > > https://play.golang.org/p/RxmtASRiS4 > > In my actual code I have the key in a slice, so I expected []uint8 to be > more cache friendly than []int, but I guess any potential cache > friendliness was outweighed by this hashing performance difference. >
It looks like any type that isn't the size of a word is slower. Probably due to all sort of masking and maybe shifting it has to do. -ayan -- You received this message because you are subscribed to the Google Groups "golang-nuts" group. To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
