feat(lang/fastrand): add Read([]byte) function

[ziposcar]

What

This PR add a new function func Read([]byte) (n int, err error) for fastrand package. It has the same signature with math/rand.Read and math/rand.(*Rand).Read.

Why

fastrand is so good that I want to use it everywhere~
And in my benchmark, it's at least 30% faster than math/rand.(*Rand).Read in single goroutine.

How

Just like math/rand.Read.

[zhangyunhao116]

Thank you!

Guess we could use the unsafe to accelerate this function, 32-bit memory copy will be much faster.

Here is an example.

// Read generates len(p) random bytes and writes them into p.
// It always returns len(p) and a nil error. And it is safe
// for concurrent use.
func Read(p []byte) (n int, err error) {
	l := len(p)

	if l >= 4 {
		i := 0
		uint32p := *(*[]uint32)(unsafe.Pointer(&p))
		for l >= 4 {
			uint32p[i] = Uint32()
			i++
			l -= 4
		}
	}

	if l > 0 {
		r := Uint32()
		for l > 0 {
			p[len(p)-l] = byte(r >> (l * 8))
			l--
		}
	}

	return len(p), nil
}

Performance comparison. Read2 is the new one.

name           time/op
10Read-16      8.66ns ± 0%
10Read2-16     7.29ns ± 0%
100Read-16     68.6ns ± 0%
100Read2-16    42.7ns ± 0%
1000Read-16     652ns ± 0%
1000Read2-16    362ns ± 0%
10000Read-16   6.42µs ± 0%
10000Read2-16  3.60µs ± 0%

Use local XORSHIFT looks like a better idea, but it is too complicated, we could implement it in the next PR.
Let me know if anything can help :)

[ziposcar]

Great! So what should I do? Close this MR or just update my commit ?

[zhangyunhao116]

Just update this PR please, I can implement the complicated version in the next PR.

[ziposcar]

Updated. Thanks for review

[zhangyunhao116]

Thanks!