Fast Hash and HashMap in Go

FastGoHash

I have really come to enjoy working in Go. From time to time I expect that the team at Apcera will
need to dip back into C for some things, but I decided to try to push the envelope with Go on making
some faster HashMaps than Go's builtin map. I started with the hash algorithms.

These are some results from my MBA11" core i7.. The fastest Hash algorithms will not work on machines
without support for unaligned access, and I still need to do some cleaning up before we OSS, but so far so good.

I aslo need to implement the fastest one, FNV1A_Hash_Yorikke, from http://http://www.sanmayce.com/Fastest_Hash

I hack SetBytes(1) to get a rough idea of ops/sec, its not actually IO.. So 10 MB/s is 10 Million ops per sec.

======================================

> go test --bench="Hash" -gcflags="-B"
PASS
BenchmarkMurmurHashSmallKey	100000000	        14.8 ns/op	  67.61 MB/s
BenchmarkNewMurmurHashSmallKey	200000000	        9.86 ns/op	 101.43 MB/s
BenchmarkFNV1AHashSmallKey	200000000	        9.18 ns/op	 108.88 MB/s
BenchmarkBernsteinHashSmallKey	200000000	        9.70 ns/op	 103.13 MB/s
BenchmarkMeiyanHashSmallKey	500000000	        7.80 ns/op	 128.17 MB/s
BenchmarkMurmurHashMedKey	50000000	        66.4 ns/op	  15.05 MB/s
BenchmarkNewMurmurHashMedKey	100000000	        27.1 ns/op	  36.94 MB/s
BenchmarkBernsteinHashMedKey	50000000	        47.4 ns/op	  21.11 MB/s
BenchmarkMeiyanHashMedKey	100000000	        11.1 ns/op	  90.38 MB/s
BenchmarkFNV1AHashMedKey	50000000	        46.3 ns/op	  21.59 MB/s

> go test --bench="Map" -gcflags="-B"
PASS
BenchmarkGoMap	                50000000	        48.5 ns/op	  20.61 MB/s
BenchmarkOurMapGet	        100000000	        20.8 ns/op	  48.09 MB/s
BenchmarkOurMapGetLocked	50000000	        35.0 ns/op	  28.58 MB/s
BenchmarkGoMapMedKey	        20000000	        84.5 ns/op	  11.84 MB/s
BenchmarkOurMapGetMedKey	20000000	        72.8 ns/op	  13.73 MB/s
BenchmarkOurMapSet	        10000000	        156 ns/op	   6.39 MB/s

Go's maps aren't the fastest, but they are stable, access and insert times are very predictable from 10 elements to 10 million. Can you share some details about the size of the maps you are benchmarking ?

They are fairly small, which is our use case.

Can you share the code?

Can you share the code?

Champion

They are fairly small, which is our use case.

Thanks