encoding/gob: Speedup map decoding by reducing the allocations.

The improvementis achieved in encoding/gob/decode.go decodeMap by
allocate keyInstr and elemInstr only once and pass it to
decodeIntoValue, instead of allocating a new instance on every loop
cycle.

name                     old time/op  new time/op  delta
DecodeComplex128Slice-8  64.2µs ±10%  62.2µs ± 8%     ~     (p=0.686 n=4+4)
DecodeFloat64Slice-8     37.1µs ± 3%  36.5µs ± 5%     ~     (p=0.343 n=4+4)
DecodeInt32Slice-8       33.7µs ± 3%  32.7µs ± 4%     ~     (p=0.200 n=4+4)
DecodeStringSlice-8      59.7µs ± 5%  57.3µs ± 1%     ~     (p=0.114 n=4+4)
DecodeInterfaceSlice-8    543µs ± 7%   497µs ± 3%     ~     (p=0.057 n=4+4)
DecodeMap-8              3.78ms ± 8%  2.66ms ± 2%  -29.69%  (p=0.029 n=4+4)

Updates #19525

Change-Id: Iec5fa4530de76f0a70da5de8a129a567b4aa096e
Reviewed-on: https://go-review.googlesource.com/38317
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Rob Pike <r@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

src/encoding/gob/codec_test.go

@@ -545,11 +545,16 @@ func TestEndToEnd(t *testing.T) {
 	type T2 struct {
 		T string
 	}
+	type T3 struct {
+		X float64
+		Z *int
+	}
 	s1 := "string1"
 	s2 := "string2"
 	type T1 struct {
 		A, B, C  int
 		M        map[string]*float64
+		M2       map[int]T3
 		EmptyMap map[string]int // to check that we receive a non-nil map.
 		N        *[3]float64
 		Strs     *[2]string
@@ -561,11 +566,14 @@ func TestEndToEnd(t *testing.T) {
 	}
 	pi := 3.14159
 	e := 2.71828
+	meaning := 42
+	fingers := 5
 	t1 := &T1{
 		A:        17,
 		B:        18,
 		C:        -5,
 		M:        map[string]*float64{"pi": &pi, "e": &e},
+		M2:       map[int]T3{4: T3{X: pi, Z: &meaning}, 10: T3{X: e, Z: &fingers}},
 		EmptyMap: make(map[string]int),
 		N:        &[3]float64{1.5, 2.5, 3.5},
 		Strs:     &[2]string{s1, s2},

src/encoding/gob/decode.go

@@ -542,12 +542,12 @@ func (dec *Decoder) decodeArray(state *decoderState, value reflect.Value, elemOp
 }
 
 // decodeIntoValue is a helper for map decoding.
-func decodeIntoValue(state *decoderState, op decOp, isPtr bool, value reflect.Value, ovfl error) reflect.Value {
+func decodeIntoValue(state *decoderState, op decOp, isPtr bool, value reflect.Value, instr *decInstr) reflect.Value {
-	instr := &decInstr{op, 0, nil, ovfl}
 	v := value
 	if isPtr {
 		v = decAlloc(value)
 	}
+
 	op(instr, state, v)
 	return value
 }
@@ -564,9 +564,11 @@ func (dec *Decoder) decodeMap(mtyp reflect.Type, state *decoderState, value refl
 	n := int(state.decodeUint())
 	keyIsPtr := mtyp.Key().Kind() == reflect.Ptr
 	elemIsPtr := mtyp.Elem().Kind() == reflect.Ptr
+	keyInstr := &decInstr{keyOp, 0, nil, ovfl}
+	elemInstr := &decInstr{elemOp, 0, nil, ovfl}
 	for i := 0; i < n; i++ {
-		key := decodeIntoValue(state, keyOp, keyIsPtr, allocValue(mtyp.Key()), ovfl)
+		key := decodeIntoValue(state, keyOp, keyIsPtr, allocValue(mtyp.Key()), keyInstr)
-		elem := decodeIntoValue(state, elemOp, elemIsPtr, allocValue(mtyp.Elem()), ovfl)
+		elem := decodeIntoValue(state, elemOp, elemIsPtr, allocValue(mtyp.Elem()), elemInstr)
 		value.SetMapIndex(key, elem)
 	}
 }

src/encoding/gob/timing_test.go

@@ -364,3 +364,28 @@ func BenchmarkDecodeInterfaceSlice(b *testing.B) {
 		}
 	}
 }
+
+func BenchmarkDecodeMap(b *testing.B) {
+	count := 10000
+	m := make(map[int]int, count)
+	for i := 0; i < count; i++ {
+		m[i] = i
+	}
+	var buf bytes.Buffer
+	enc := NewEncoder(&buf)
+	err := enc.Encode(m)
+	if err != nil {
+		b.Fatal(err)
+	}
+	bbuf := benchmarkBuf{data: buf.Bytes()}
+	b.ResetTimer()
+	for i := 0; i < b.N; i++ {
+		rm := make(map[int]int, 0)
+		bbuf.reset()
+		dec := NewDecoder(&bbuf)
+		err := dec.Decode(&rm)
+		if err != nil {
+			b.Fatal(i, err)
+		}
+	}
+}