save log waste and tag caching

[suhuaqin]

Hi, here's a merge request on performance optimization with two main changes:

1. save log consumption when log printing is not needed
2. cache tag parsing results

Both of these changes are very simple, here is my benchmark code and the results:

package ngap

import (
	"github.com/free5gc/ngap/ngapType"
	"testing"
)

var by = []byte{0, 14, 0, 128, 168, 0, 0, 9, 0, 10, 0, 5, 96, 248, 4, 128, 182, 0, 85, 0, 3, 64, 48, 57, 0, 18, 64, 10, 3, 255, 5, 0, 100, 240, 0, 0, 0, 1, 0, 28, 0, 7, 0, 100, 240, 0, 1, 0, 65, 0, 0, 0, 5, 2, 1, 1, 1, 1, 0, 119, 0, 9, 0, 0, 4, 0, 0, 0, 0, 0, 0, 0, 94, 0, 32, 24, 215, 2, 95, 204, 253, 169, 14, 166, 251, 21, 39, 237, 201, 125, 92, 17, 166, 65, 16, 192, 153, 171, 157, 76, 45, 186, 236, 224, 124, 7, 35, 0, 31, 64, 2, 0, 0, 0, 38, 64, 56, 55, 126, 2, 110, 230, 241, 215, 1, 126, 0, 66, 1, 1, 119, 0, 11, 242, 100, 240, 0, 1, 0, 65, 192, 22, 70, 161, 84, 7, 64, 100, 240, 0, 0, 0, 1, 21, 5, 4, 1, 1, 1, 1, 49, 5, 4, 1, 1, 1, 1, 33, 1, 1, 94, 1, 5}

var pdu *ngapType.NGAPPDU

func TestMain(m *testing.M) {
	var err error
	pdu, err = Decoder(by)
	if err != nil {
		panic(err)
	}
	m.Run()
}

func BenchmarkNgap(b *testing.B) {
	b.Run("encoder", benchEncoder)
	b.Run("decoder", benchDecoder)
}

func benchEncoder(b *testing.B) {
	for i := 0; i < b.N; i++ {
		_, err := Encoder(*pdu)
		if err != nil {
			panic(err)
		}
	}
}

func benchDecoder(b *testing.B) {
	for i := 0; i < b.N; i++ {
		_, err := Decoder(by)
		if err != nil {
			panic(err)
		}
	}
}

goos: darwin
goarch: arm64
pkg: github.com/free5gc/ngap
               │     v0.txt     │               v3.txt                │
               │     sec/op     │    sec/op     vs base               │
Ngap/encoder-8    395.80µ ± ∞ ¹   11.30µ ± ∞ ¹  -97.15% (p=0.008 n=5)
Ngap/decoder-8   369.736µ ± ∞ ¹   9.619µ ± ∞ ¹  -97.40% (p=0.008 n=5)
geomean            382.5µ         10.42µ        -97.28%
¹ need >= 6 samples for confidence interval at level 0.95

               │     v0.txt      │                v3.txt                │
               │      B/op       │     B/op       vs base               │
Ngap/encoder-8   194.858Ki ± ∞ ¹   5.461Ki ± ∞ ¹  -97.20% (p=0.008 n=5)
Ngap/decoder-8   186.561Ki ± ∞ ¹   5.719Ki ± ∞ ¹  -96.93% (p=0.008 n=5)
geomean            190.7Ki         5.588Ki        -97.07%
¹ need >= 6 samples for confidence interval at level 0.95

               │    v0.txt    │               v3.txt               │
               │  allocs/op   │  allocs/op   vs base               │
Ngap/encoder-8   3710.0 ± ∞ ¹   335.0 ± ∞ ¹  -90.97% (p=0.008 n=5)
Ngap/decoder-8   3540.0 ± ∞ ¹   253.0 ± ∞ ¹  -92.85% (p=0.008 n=5)
geomean          3.624k         291.1        -91.97%
¹ need >= 6 samples for confidence interval at level 0.95

[ss920386]

@suhuaqin Thanks for the PR. May I ask will this have a risk of memory leak? I didn't see sync.Map.Delete() called in this PR. Will the cache grow forever?

[suhuaqin]

@suhuaqin Thanks for the PR. May I ask will this have a risk of memory leak? I didn't see sync.Map.Delete() called in this PR. Will the cache grow forever?

The maximum size of the cache depends on the number of structures to be parsed and the number of members of the structure. If you are concerned about memory wastage caused by the small frequency of use of certain caches, consider introducing a cache elimination mechanism