ce45f4f3ff
wgengine/filter: inline ip6InList into match.
...
matchIPsOnly gets 5% slower when inlining, despite significantly reduced
memory ops and slightly tighter code.
Part of #19 .
Filter/tcp6_syn_in-8 45.5ns ± 1% 42.4ns ± 2% -6.86% (p=0.000 n=10+10)
Filter/udp6_in-8 107ns ± 2% 94ns ± 2% -11.50% (p=0.000 n=9+10)
Signed-off-by: David Anderson <danderson@tailscale.com>
6 years ago
3fdae12f0c
wgengine/filter: eliminate unnecessary memory loads.
...
Doesn't materially affect benchmarks, but shrinks match6 by 30 instructions
and halves memory loads.
Part of #19 .
Signed-off-by: David Anderson <danderson@tailscale.com>
6 years ago
47380ebcfb
wgengine/filter: twiddle bits to optimize
...
Part of #19 .
name old time/op new time/op delta
Filter/icmp4-8 32.2ns ± 3% 32.5ns ± 2% ~ (p=0.524 n=10+8)
Filter/icmp6-8 49.7ns ± 6% 43.1ns ± 4% -13.12% (p=0.000 n=9+10)
Signed-off-by: David Anderson <danderson@tailscale.com>
6 years ago
2d604b3791
net/packet: represent IP6 as two uint64s.
...
For the operations we perform on these types (mostly net6.Contains),
this encoding is much faster.
Part of #19 .
name old time/op new time/op delta
Filter/icmp4-8 27.5ns ± 1% 28.0ns ± 2% +1.89% (p=0.016 n=5+5)
Filter/tcp4_syn_in-8 38.8ns ± 2% 38.3ns ± 1% -1.24% (p=0.024 n=5+5)
Filter/tcp4_syn_out-8 27.6ns ±12% 24.6ns ± 1% ~ (p=0.063 n=5+5)
Filter/udp4_in-8 71.5ns ± 5% 65.9ns ± 1% -7.94% (p=0.008 n=5+5)
Filter/udp4_out-8 132ns ±13% 119ns ± 1% -10.29% (p=0.008 n=5+5)
Filter/icmp6-8 169ns ±10% 54ns ± 1% -68.35% (p=0.008 n=5+5)
Filter/tcp6_syn_in-8 149ns ± 6% 43ns ± 1% -71.11% (p=0.008 n=5+5)
Filter/tcp6_syn_out-8 37.7ns ± 4% 24.3ns ± 3% -35.51% (p=0.008 n=5+5)
Filter/udp6_in-8 179ns ± 5% 103ns ± 1% -42.75% (p=0.008 n=5+5)
Filter/udp6_out-8 156ns ± 3% 191ns ± 1% +22.54% (p=0.008 n=5+5)
Signed-off-by: David Anderson <danderson@tailscale.com>
6 years ago
04ff3c91ee
wgengine/filter: add full IPv6 support.
...
Part of #19 .
Signed-off-by: David Anderson <danderson@tailscale.com>
6 years ago
David Anderson
Josh Bleecher Snyder