strings: optimize Count for amd64
Move optimized Count implementation from bytes to runtime. Use in
both bytes and strings packages.
Add CountByte benchmark to strings.
Strings benchmarks:
name old time/op new time/op delta
CountHard1-4 226µs ± 1% 226µs ± 2% ~ (p=0.247 n=10+10)
CountHard2-4 316µs ± 1% 315µs ± 0% ~ (p=0.133 n=9+10)
CountHard3-4 919µs ± 1% 920µs ± 1% ~ (p=0.968 n=10+9)
CountTorture-4 15.4µs ± 1% 15.7µs ± 1% +2.47% (p=0.000 n=10+9)
CountTortureOverlapping-4 9.60ms ± 0% 9.65ms ± 1% ~ (p=0.247 n=10+10)
CountByte/10-4 26.3ns ± 1% 10.9ns ± 1% -58.71% (p=0.000 n=9+9)
CountByte/32-4 42.7ns ± 0% 14.2ns ± 0% -66.64% (p=0.000 n=10+10)
CountByte/4096-4 3.07µs ± 0% 0.31µs ± 2% -89.99% (p=0.000 n=9+10)
CountByte/
4194304-4 3.48ms ± 1% 0.34ms ± 1% -90.09% (p=0.000 n=10+9)
CountByte/
67108864-4 55.6ms ± 1% 7.0ms ± 0% -87.49% (p=0.000 n=9+8)
name old speed new speed delta
CountByte/10-4 380MB/s ± 1% 919MB/s ± 1% +142.21% (p=0.000 n=9+9)
CountByte/32-4 750MB/s ± 0% 2247MB/s ± 0% +199.62% (p=0.000 n=10+10)
CountByte/4096-4 1.33GB/s ± 0% 13.32GB/s ± 2% +898.13% (p=0.000 n=9+10)
CountByte/
4194304-4 1.21GB/s ± 1% 12.17GB/s ± 1% +908.87% (p=0.000 n=10+9)
CountByte/
67108864-4 1.21GB/s ± 1% 9.65GB/s ± 0% +699.29% (p=0.000 n=9+8)
Fixes #19411
Change-Id: I8d2d409f0fa6df6d03b60790aa86e540b4a4e3b0
Reviewed-on: https://go-review.googlesource.com/38693
Reviewed-by: Keith Randall <khr@golang.org>