proto, runtime/protoiface, internal/impl: add fast-path Merge

Comparing -tags=protoreflect to fast-path:

name                              old time/op    new time/op    delta
pkg:google.golang.org/protobuf/internal/benchmarks goos:linux goarch:amd64
/Clone/google_message1_proto2-12    1.70µs ± 1%    0.30µs ± 1%  -82.64%  (p=0.001 n=7+7)
/Clone/google_message1_proto3-12    1.01µs ± 1%    0.19µs ± 1%  -80.77%  (p=0.000 n=7+8)
/Clone/google_message2-12            818µs ± 8%     141µs ± 6%  -82.78%  (p=0.000 n=8+8)
pkg:google.golang.org/protobuf/internal/benchmarks/micro goos:linux goarch:amd64
EmptyMessage/Clone-12               51.1ns ± 1%    39.3ns ± 3%  -23.03%  (p=0.000 n=7+8)
RepeatedInt32/Clone-12              24.5µs ± 1%     1.1µs ± 3%  -95.64%  (p=0.000 n=8+8)
Required/Clone-12                    978ns ± 1%     132ns ± 2%  -86.46%  (p=0.000 n=8+8)

name                              old alloc/op   new alloc/op   delta
pkg:google.golang.org/protobuf/internal/benchmarks goos:linux goarch:amd64
/Clone/google_message1_proto2-12    1.08kB ± 0%    0.74kB ± 0%  -31.85%  (p=0.000 n=8+8)
/Clone/google_message1_proto3-12      872B ± 0%      544B ± 0%  -37.61%  (p=0.000 n=8+8)
/Clone/google_message2-12            602kB ± 0%     411kB ± 0%  -31.65%  (p=0.000 n=8+8)
pkg:google.golang.org/protobuf/internal/benchmarks/micro goos:linux goarch:amd64
EmptyMessage/Clone-12                96.0B ± 0%     64.0B ± 0%  -33.33%  (p=0.000 n=8+8)
RepeatedInt32/Clone-12              25.4kB ± 0%     3.2kB ± 0%  -87.33%  (p=0.000 n=8+8)
Required/Clone-12                     416B ± 0%      256B ± 0%  -38.46%  (p=0.000 n=8+8)

name                              old allocs/op  new allocs/op  delta
pkg:google.golang.org/protobuf/internal/benchmarks goos:linux goarch:amd64
/Clone/google_message1_proto2-12      52.0 ± 0%      21.0 ± 0%  -59.62%  (p=0.000 n=8+8)
/Clone/google_message1_proto3-12      33.0 ± 0%       3.0 ± 0%  -90.91%  (p=0.000 n=8+8)
/Clone/google_message2-12            22.3k ± 0%      7.5k ± 0%  -66.41%  (p=0.000 n=8+8)
pkg:google.golang.org/protobuf/internal/benchmarks/micro goos:linux goarch:amd64
EmptyMessage/Clone-12                 3.00 ± 0%      2.00 ± 0%  -33.33%  (p=0.000 n=8+8)
RepeatedInt32/Clone-12               1.51k ± 0%     0.00k ± 0%  -99.80%  (p=0.000 n=8+8)
Required/Clone-12                     51.0 ± 0%      18.0 ± 0%  -64.71%  (p=0.000 n=8+8)

Change-Id: Ife9018097c34cb025dc9c4fdd9a61b2f947853c6
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/219147
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
diff --git a/internal/impl/codec_map.go b/internal/impl/codec_map.go
index b319583..dfeb944 100644
--- a/internal/impl/codec_map.go
+++ b/internal/impl/codec_map.go
@@ -67,6 +67,14 @@
 			}
 		},
 	}
+	switch valField.Kind() {
+	case pref.MessageKind:
+		funcs.merge = mergeMapOfMessage
+	case pref.BytesKind:
+		funcs.merge = mergeMapOfBytes
+	default:
+		funcs.merge = mergeMap
+	}
 	if valFuncs.isInit != nil {
 		funcs.isInit = func(p pointer, f *coderFieldInfo) error {
 			return isInitMap(p.AsValueOf(ft).Elem(), mapi, f)
@@ -327,3 +335,54 @@
 	}
 	return nil
 }
+
+func mergeMap(dst, src pointer, f *coderFieldInfo, opts mergeOptions) {
+	dstm := dst.AsValueOf(f.ft).Elem()
+	srcm := src.AsValueOf(f.ft).Elem()
+	if srcm.Len() == 0 {
+		return
+	}
+	if dstm.IsNil() {
+		dstm.Set(reflect.MakeMap(f.ft))
+	}
+	iter := mapRange(srcm)
+	for iter.Next() {
+		dstm.SetMapIndex(iter.Key(), iter.Value())
+	}
+}
+
+func mergeMapOfBytes(dst, src pointer, f *coderFieldInfo, opts mergeOptions) {
+	dstm := dst.AsValueOf(f.ft).Elem()
+	srcm := src.AsValueOf(f.ft).Elem()
+	if srcm.Len() == 0 {
+		return
+	}
+	if dstm.IsNil() {
+		dstm.Set(reflect.MakeMap(f.ft))
+	}
+	iter := mapRange(srcm)
+	for iter.Next() {
+		dstm.SetMapIndex(iter.Key(), reflect.ValueOf(append(emptyBuf[:], iter.Value().Bytes()...)))
+	}
+}
+
+func mergeMapOfMessage(dst, src pointer, f *coderFieldInfo, opts mergeOptions) {
+	dstm := dst.AsValueOf(f.ft).Elem()
+	srcm := src.AsValueOf(f.ft).Elem()
+	if srcm.Len() == 0 {
+		return
+	}
+	if dstm.IsNil() {
+		dstm.Set(reflect.MakeMap(f.ft))
+	}
+	iter := mapRange(srcm)
+	for iter.Next() {
+		val := reflect.New(f.ft.Elem().Elem())
+		if f.mi != nil {
+			f.mi.mergePointer(pointerOfValue(val), pointerOfValue(iter.Value()), opts)
+		} else {
+			opts.Merge(asMessage(val), asMessage(iter.Value()))
+		}
+		dstm.SetMapIndex(iter.Key(), val)
+	}
+}