X xzlib: Reuse zlib decoders

name                        old time/op    new time/op    delta
deco/unzlib/py/null-1K        2.11µs ± 1%    2.12µs ± 1%     ~     (p=0.690 n=5+5)
deco/unzlib/go/null-1K        5.87µs ± 1%    1.89µs ± 1%  -67.77%  (p=0.008 n=5+5)
deco/unzlib/py/null-4K        13.3µs ± 1%    13.5µs ± 4%     ~     (p=0.690 n=5+5)
deco/unzlib/go/null-4K        12.6µs ± 0%     8.5µs ± 0%  -32.50%  (p=0.008 n=5+5)
deco/unzlib/py/null-2M        5.17ms ±11%    5.20ms ±10%     ~     (p=1.000 n=5+5)
deco/unzlib/go/null-2M        2.61ms ± 1%    2.58ms ± 1%   -1.32%  (p=0.016 n=5+5)
deco/unzlib/py/wczdata-avg    20.9µs ±14%    24.1µs ± 1%  +15.05%  (p=0.016 n=5+4)
deco/unzlib/go/wczdata-avg    73.0µs ± 1%    68.0µs ± 1%   -6.94%  (p=0.008 n=5+5)
deco/unzlib/py/wczdata-max    23.6µs ± 0%    23.5µs ± 1%     ~     (p=0.905 n=5+4)
deco/unzlib/go/wczdata-max    72.7µs ± 0%    67.8µs ± 0%   -6.79%  (p=0.008 n=5+5)
deco/unzlib/py/prod1-avg      4.53µs ± 3%    4.47µs ± 2%     ~     (p=0.310 n=5+5)
deco/unzlib/go/prod1-avg      15.5µs ± 0%    11.0µs ± 0%  -29.11%  (p=0.008 n=5+5)
deco/unzlib/py/prod1-max       327µs ± 1%     326µs ± 0%     ~     (p=1.000 n=5+5)
deco/unzlib/go/prod1-max       545µs ± 0%     542µs ± 0%   -0.68%  (p=0.008 n=5+5)

still on wczdata and prod1 much slower compared to py/c zlib.

go/neo/internal/xzlib/xzlib.go

@@ -24,6 +24,7 @@ import (
 	"bytes"
 	"compress/zlib"
 	"io"
+	"sync"
 )
 
 
@@ -45,6 +46,41 @@ func Compress(data []byte) (zdata []byte) {
 }
 
 
+// ---- zlib.Reader pool ----
+// (creating zlib.NewReader for every decompress has high overhead for not large blocks)
+
+// znull is a small valid zlib stream.
+// we need it to create new zlib readers under sync.Pool .
+var znull = Compress(nil)
+
+var zrPool = sync.Pool{New: func() interface{} {
+	r, err := zlib.NewReader(bytes.NewReader(znull))
+	if err != nil {
+		panic(err) // must not happen - znull is valid stream
+	}
+	return r
+}}
+
+// interface actually implemented by what zlib.NewReader returns
+type zlibReader interface {
+	io.ReadCloser
+	zlib.Resetter
+}
+
+func zlibNewReader(r io.Reader) (zlibReader, error) {
+	zr := zrPool.Get().(zlibReader)
+	err := zr.Reset(r, nil)
+	if err != nil {
+		zlibFreeReader(zr)
+		return nil, err
+	}
+	return zr, nil
+}
+
+func zlibFreeReader(r zlibReader) {
+	zrPool.Put(r)
+}
+
 // Decompress decompresses data according to zlib encoding.
 //
 // out buffer, if there is enough capacity, is used for decompression destination.
@@ -53,7 +89,7 @@ func Compress(data []byte) (zdata []byte) {
 // return: destination buffer with full decompressed data or error.
 func Decompress(in []byte, out []byte) (data []byte, err error) {
 	bin := bytes.NewReader(in)
-	zr, err := zlib.NewReader(bin)
+	zr, err := zlibNewReader(bin)
 	if err != nil {
 		return nil, err
 	}
@@ -63,6 +99,7 @@ func Decompress(in []byte, out []byte) (data []byte, err error) {
 			err = err2
 			data = nil
 		}
+		zlibFreeReader(zr)
 	}()
 
 	bout := bytes.NewBuffer(out)