Add xsync MapOf to benchmarks

[puzpuzpuz]

Adds xsync MapOf to the benchmarks to have one more map in the comparison.

Results on my machine are the following:

$ go test -benchmem -bench .
goos: linux
goarch: amd64
pkg: github.com/alphadose/haxmap/benchmarks
cpu: 11th Gen Intel(R) Core(TM) i7-1185G7 @ 3.00GHz
BenchmarkHaxMapReadsOnly-8             	  174625	      6876 ns/op	       0 B/op	       0 allocs/op
BenchmarkHaxMapReadsWithWrites-8       	  149389	      7895 ns/op	     963 B/op	     120 allocs/op
BenchmarkGoSyncMapReadsOnly-8          	   60259	     19771 ns/op	       0 B/op	       0 allocs/op
BenchmarkGoSyncMapReadsWithWrites-8    	   52824	     22527 ns/op	    4874 B/op	     452 allocs/op
BenchmarkCornelkMapReadsOnly-8         	  178738	      6657 ns/op	       0 B/op	       0 allocs/op
BenchmarkCornelkMapReadsWithWrites-8   	  150156	      8103 ns/op	    1026 B/op	     128 allocs/op
BenchmarkXsyncMapReadsOnly-8           	  294196	      3993 ns/op	       0 B/op	       0 allocs/op
BenchmarkXsyncMapReadsWithWrites-8     	  241662	      4698 ns/op	     922 B/op	      57 allocs/op
PASS
ok  	github.com/alphadose/haxmap/benchmarks	10.413s

[bagualing]

Wow, xsync is knocking on the door.
The latest update of xsync has a significant performance boost.

[puzpuzpuz]

@bagualing thanks. Yeah, MapOf got more efficient in scenarios when the data set is small in the latest version.

[alphadose]

@puzpuzpuz really great work, I will definitely check out xsync map implementation in more depth

I will also see if current haxmap performance can be improved from drawing inspiration from xsync map

[puzpuzpuz]

I will definitely check out xsync map implementation in more depth

That would be awesome. You should be looking into MapOf (generic map) as it's different from Map (non-generic map). MapOf overall design is quite simple: a hash table of cache-line-sized buckets holding key hashes + pointers to immutable key-value pairs. Each bucket is organized in a unrolled linked list and protected with a sync.Mutex. The rehashing involves a bit tricky synchronization, but still not very complicated.