Move hashn() calculations inline to callers

[dgryski]

Currently, hashn() allocates a new slice of the hash values and returns it. This causes an allocation that we could get rid of if we instead chose to return h1 and h2 and let the caller compute the different hash values.

This changes the callers from

for i, pos := range hashn(h, d, w) {

to

h1, h2 := hashn(h)
for i := 0; i < d; i++ {
    pos := (h1 + uint32(i)*h2) % uint32(w)

Removing these allocations In my tests this speeds up the hashing code by a factor of 2 (~1500ns/op to ~750ns/op).

I have a patch that does this, but it touches a bunch of code so I thought I'd clear this with you before pushing.

[dustin]

Sure, speeding stuff up sounds like a great idea. :)

[dgryski]

Looks like I didn't get the magic words right in the commit message to auto-close this.