image/png: speed up PNG decoding for common color models: Gray, NRGBA,

Paletted, RGBA.

benchmark                       old ns/op    new ns/op    delta
BenchmarkDecodeGray               3681144      2536049  -31.11%
BenchmarkDecodeNRGBAGradient     12108660     10020650  -17.24%
BenchmarkDecodeNRGBAOpaque       10699230      8677165  -18.90%
BenchmarkDecodePaletted           2562806      1458798  -43.08%
BenchmarkDecodeRGB                8468175      7180730  -15.20%

benchmark                        old MB/s     new MB/s  speedup
BenchmarkDecodeGray                 17.80        25.84    1.45x
BenchmarkDecodeNRGBAGradient        21.65        26.16    1.21x
BenchmarkDecodeNRGBAOpaque          24.50        30.21    1.23x
BenchmarkDecodePaletted             25.57        44.92    1.76x
BenchmarkDecodeRGB                  30.96        36.51    1.18x

$ file $GOROOT/src/pkg/image/png/testdata/bench*
benchGray.png:           PNG image, 256 x 256, 8-bit grayscale, non-interlaced
benchNRGBA-gradient.png: PNG image, 256 x 256, 8-bit/color RGBA, non-interlaced
benchNRGBA-opaque.png:   PNG image, 256 x 256, 8-bit/color RGBA, non-interlaced
benchPaletted.png:       PNG image, 256 x 256, 8-bit colormap, non-interlaced
benchRGB.png:            PNG image, 256 x 256, 8-bit/color RGB, non-interlaced

R=r
CC=golang-dev
https://golang.org/cl/6127051

src/pkg/image/png/reader.go

@@ -301,6 +301,7 @@ func (d *decoder) decode() (image.Image, error) {
 	defer r.Close()
 	bitsPerPixel := 0
 	maxPalette := uint8(0)
+	pixOffset := 0
 	var (
 		gray     *image.Gray
 		rgba     *image.RGBA
@@ -423,18 +424,24 @@ func (d *decoder) decode() (image.Image, error) {
 				}
 			}
 		case cbG8:
-			for x := 0; x < d.width; x++ {
-				gray.SetGray(x, y, color.Gray{cdat[x]})
-			}
+			copy(gray.Pix[pixOffset:], cdat)
+			pixOffset += gray.Stride
 		case cbGA8:
 			for x := 0; x < d.width; x++ {
 				ycol := cdat[2*x+0]
 				nrgba.SetNRGBA(x, y, color.NRGBA{ycol, ycol, ycol, cdat[2*x+1]})
 			}
 		case cbTC8:
+			pix, i, j := rgba.Pix, pixOffset, 0
 			for x := 0; x < d.width; x++ {
-				rgba.SetRGBA(x, y, color.RGBA{cdat[3*x+0], cdat[3*x+1], cdat[3*x+2], 0xff})
+				pix[i+0] = cdat[j+0]
+				pix[i+1] = cdat[j+1]
+				pix[i+2] = cdat[j+2]
+				pix[i+3] = 0xff
+				i += 4
+				j += 3
 			}
+			pixOffset += rgba.Stride
 		case cbP1:
 			for x := 0; x < d.width; x += 8 {
 				b := cdat[x/8]
@@ -472,16 +479,18 @@ func (d *decoder) decode() (image.Image, error) {
 				}
 			}
 		case cbP8:
-			for x := 0; x < d.width; x++ {
-				if cdat[x] > maxPalette {
-					return nil, FormatError("palette index out of range")
+			if maxPalette != 255 {
+				for x := 0; x < d.width; x++ {
+					if cdat[x] > maxPalette {
+						return nil, FormatError("palette index out of range")
+					}
 				}
-				paletted.SetColorIndex(x, y, cdat[x])
 			}
+			copy(paletted.Pix[pixOffset:], cdat)
+			pixOffset += paletted.Stride
 		case cbTCA8:
-			for x := 0; x < d.width; x++ {
-				nrgba.SetNRGBA(x, y, color.NRGBA{cdat[4*x+0], cdat[4*x+1], cdat[4*x+2], cdat[4*x+3]})
-			}
+			copy(nrgba.Pix[pixOffset:], cdat)
+			pixOffset += nrgba.Stride
 		case cbG16:
 			for x := 0; x < d.width; x++ {
 				ycol := uint16(cdat[2*x+0])<<8 | uint16(cdat[2*x+1])

src/pkg/image/png/reader_test.go

@@ -10,6 +10,7 @@ import (
 	"image"
 	"image/color"
 	"io"
+	"io/ioutil"
 	"os"
 	"strings"
 	"testing"
@@ -267,3 +268,41 @@ func TestReaderError(t *testing.T) {
 		}
 	}
 }
+
+func benchmarkDecode(b *testing.B, filename string, bytesPerPixel int) {
+	b.StopTimer()
+	data, err := ioutil.ReadFile(filename)
+	if err != nil {
+		b.Fatal(err)
+	}
+	s := string(data)
+	cfg, err := DecodeConfig(strings.NewReader(s))
+	if err != nil {
+		b.Fatal(err)
+	}
+	b.SetBytes(int64(cfg.Width * cfg.Height * bytesPerPixel))
+	b.StartTimer()
+	for i := 0; i < b.N; i++ {
+		Decode(strings.NewReader(s))
+	}
+}
+
+func BenchmarkDecodeGray(b *testing.B) {
+	benchmarkDecode(b, "testdata/benchGray.png", 1)
+}
+
+func BenchmarkDecodeNRGBAGradient(b *testing.B) {
+	benchmarkDecode(b, "testdata/benchNRGBA-gradient.png", 4)
+}
+
+func BenchmarkDecodeNRGBAOpaque(b *testing.B) {
+	benchmarkDecode(b, "testdata/benchNRGBA-opaque.png", 4)
+}
+
+func BenchmarkDecodePaletted(b *testing.B) {
+	benchmarkDecode(b, "testdata/benchPaletted.png", 1)
+}
+
+func BenchmarkDecodeRGB(b *testing.B) {
+	benchmarkDecode(b, "testdata/benchRGB.png", 4)
+}