encoding/hex: don't overallocate memory in DecodeString

Now as []byte(string) doesn't always cause heap allocation (CL 520599, #2205)
we can make DecodeString simpler and more performant, by not allocating
x2 the required memory.

goos: linux
goarch: amd64
pkg: encoding/hex
cpu: AMD Ryzen 5 4600G with Radeon Graphics
                      │  beforehex   │              afterhex               │
                      │    sec/op    │   sec/op     vs base                │
DecodeString/256-12      197.9n ± 1%   172.2n ± 1%  -13.01% (p=0.000 n=10)
DecodeString/1024-12     684.9n ± 1%   598.5n ± 1%  -12.61% (p=0.000 n=10)
DecodeString/4096-12     2.764µ ± 0%   2.343µ ± 1%  -15.23% (p=0.000 n=10)
DecodeString/16384-12   10.774µ ± 1%   9.348µ ± 1%  -13.23% (p=0.000 n=10)
geomean                  1.417µ        1.226µ       -13.53%

                      │  beforehex   │               afterhex               │
                      │     B/s      │     B/s       vs base                │
DecodeString/256-12     1.205Gi ± 1%   1.385Gi ± 1%  +14.94% (p=0.000 n=10)
DecodeString/1024-12    1.393Gi ± 1%   1.593Gi ± 1%  +14.42% (p=0.000 n=10)
DecodeString/4096-12    1.380Gi ± 0%   1.628Gi ± 1%  +17.97% (p=0.000 n=10)
DecodeString/16384-12   1.416Gi ± 1%   1.632Gi ± 1%  +15.25% (p=0.000 n=10)
geomean                 1.346Gi        1.556Gi       +15.64%

                      │   beforehex   │               afterhex               │
                      │     B/op      │     B/op      vs base                │
DecodeString/256-12        256.0 ± 0%     128.0 ± 0%  -50.00% (p=0.000 n=10)
DecodeString/1024-12      1024.0 ± 0%     512.0 ± 0%  -50.00% (p=0.000 n=10)
DecodeString/4096-12     4.000Ki ± 0%   2.000Ki ± 0%  -50.00% (p=0.000 n=10)
DecodeString/16384-12   16.000Ki ± 0%   8.000Ki ± 0%  -50.00% (p=0.000 n=10)
geomean                  2.000Ki        1.000Ki       -50.00%

                      │ beforehex  │              afterhex               │
                      │ allocs/op  │ allocs/op   vs base                 │
DecodeString/256-12     1.000 ± 0%   1.000 ± 0%       ~ (p=1.000 n=10) ¹
DecodeString/1024-12    1.000 ± 0%   1.000 ± 0%       ~ (p=1.000 n=10) ¹
DecodeString/4096-12    1.000 ± 0%   1.000 ± 0%       ~ (p=1.000 n=10) ¹
DecodeString/16384-12   1.000 ± 0%   1.000 ± 0%       ~ (p=1.000 n=10) ¹
geomean                 1.000        1.000       +0.00%

Change-Id: I5676e48f222d90786ea18e808cb4ecde9de82597
GitHub-Last-Rev: aeedf3f
GitHub-Pull-Request: #67259
Reviewed-on: https://go-review.googlesource.com/c/go/+/584118
Auto-Submit: Ian Lance Taylor <iant@google.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Ian Lance Taylor <iant@google.com>

src/encoding/hex/hex.go

@@ -136,11 +136,9 @@ func EncodeToString(src []byte) string {
 // If the input is malformed, DecodeString returns
 // the bytes decoded before the error.
 func DecodeString(s string) ([]byte, error) {
-	src := []byte(s)
-	// We can use the source slice itself as the destination
-	// because the decode loop increments by one and then the 'seen' byte is not used anymore.
-	n, err := Decode(src, src)
-	return src[:n], err
+	dst := make([]byte, DecodedLen(len(s)))
+	n, err := Decode(dst, []byte(s))
+	return dst[:n], err
 }
 
 // Dump returns a string that contains a hex dump of the given data. The format

src/encoding/hex/hex_test.go

@@ -275,6 +275,18 @@ func BenchmarkDecode(b *testing.B) {
 	}
 }
 
+func BenchmarkDecodeString(b *testing.B) {
+	for _, size := range []int{256, 1024, 4096, 16384} {
+		src := strings.Repeat("2b744faa", size/8)
+		b.Run(fmt.Sprintf("%v", size), func(b *testing.B) {
+			b.SetBytes(int64(size))
+			for i := 0; i < b.N; i++ {
+				sink, _ = DecodeString(src)
+			}
+		})
+	}
+}
+
 func BenchmarkDump(b *testing.B) {
 	for _, size := range []int{256, 1024, 4096, 16384} {
 		src := bytes.Repeat([]byte{2, 3, 5, 7, 9, 11, 13, 17}, size/8)