From: Nigel Tao <nigeltao@golang.org>
Date: Sat, 25 Apr 2020 12:29:02 +0000 (+1000)
Subject: image/draw: optimize paletted dst + uniform src
X-Git-Tag: go1.15beta1~337
X-Git-Url: http://www.git.cypherpunks.su/?a=commitdiff_plain;h=42c48998aada0df10279650d04a018c83cbfa518;p=gostls13.git

image/draw: optimize paletted dst + uniform src

name            old time/op  new time/op  delta
PalettedFill-4  5.74ms ± 1%  0.01ms ± 1%  -99.78%  (p=0.008 n=5+5)
PalettedRGBA-4  3.34ms ± 3%  3.33ms ± 0%     ~     (p=0.690 n=5+5)

Fixes #35938

Thanks to pjbgtnj for the suggestion.

Change-Id: I07b494482cce918f556e196c5a4b481b4c16de3a
Reviewed-on: https://go-review.googlesource.com/c/go/+/230118
Run-TryBot: Nigel Tao <nigeltao@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rob Pike <r@golang.org>
---

diff --git a/src/image/draw/bench_test.go b/src/image/draw/bench_test.go
index a41d7e7dfb..831fd958ba 100644
--- a/src/image/draw/bench_test.go
+++ b/src/image/draw/bench_test.go
@@ -236,7 +236,11 @@ func BenchmarkRGBA(b *testing.B) {
 	bench(b, color.RGBAModel, color.RGBA64Model, nil, Src)
 }
 
-func BenchmarkPaletted(b *testing.B) {
+func BenchmarkPalettedFill(b *testing.B) {
+	bench(b, palette, nil, nil, Src)
+}
+
+func BenchmarkPalettedRGBA(b *testing.B) {
 	bench(b, palette, color.RGBAModel, nil, Src)
 }
 
diff --git a/src/image/draw/draw.go b/src/image/draw/draw.go
index 932a544483..8f96aa2d18 100644
--- a/src/image/draw/draw.go
+++ b/src/image/draw/draw.go
@@ -180,9 +180,25 @@ func DrawMask(dst Image, r image.Rectangle, src image.Image, sp image.Point, mas
 		drawRGBA(dst0, r, src, sp, mask, mp, op)
 		return
 	case *image.Paletted:
-		if op == Src && mask == nil && !processBackward(dst, r, src, sp) {
-			drawPaletted(dst0, r, src, sp, false)
-			return
+		if op == Src && mask == nil {
+			if src0, ok := src.(*image.Uniform); ok {
+				colorIndex := uint8(dst0.Palette.Index(src0.C))
+				i0 := dst0.PixOffset(r.Min.X, r.Min.Y)
+				i1 := i0 + r.Dx()
+				for i := i0; i < i1; i++ {
+					dst0.Pix[i] = colorIndex
+				}
+				firstRow := dst0.Pix[i0:i1]
+				for y := r.Min.Y + 1; y < r.Max.Y; y++ {
+					i0 += dst0.Stride
+					i1 += dst0.Stride
+					copy(dst0.Pix[i0:i1], firstRow)
+				}
+				return
+			} else if !processBackward(dst, r, src, sp) {
+				drawPaletted(dst0, r, src, sp, false)
+				return
+			}
 		}
 	}
 
diff --git a/src/image/draw/draw_test.go b/src/image/draw/draw_test.go
index dea51b6bc5..9c5a118400 100644
--- a/src/image/draw/draw_test.go
+++ b/src/image/draw/draw_test.go
@@ -434,11 +434,11 @@ func TestPaletted(t *testing.T) {
 		t.Fatalf("open: %v", err)
 	}
 	defer f.Close()
-	src, err := png.Decode(f)
+	video001, err := png.Decode(f)
 	if err != nil {
 		t.Fatalf("decode: %v", err)
 	}
-	b := src.Bounds()
+	b := video001.Bounds()
 
 	cgaPalette := color.Palette{
 		color.RGBA{0x00, 0x00, 0x00, 0xff},
@@ -450,19 +450,25 @@ func TestPaletted(t *testing.T) {
 		"src":             Src,
 		"floyd-steinberg": FloydSteinberg,
 	}
+	sources := map[string]image.Image{
+		"uniform":  &image.Uniform{color.RGBA{0xff, 0x7f, 0xff, 0xff}},
+		"video001": video001,
+	}
 
-loop:
 	for dName, d := range drawers {
-		dst0 := image.NewPaletted(b, cgaPalette)
-		dst1 := image.NewPaletted(b, cgaPalette)
-		d.Draw(dst0, b, src, image.Point{})
-		d.Draw(embeddedPaletted{dst1}, b, src, image.Point{})
-		for y := b.Min.Y; y < b.Max.Y; y++ {
-			for x := b.Min.X; x < b.Max.X; x++ {
-				if !eq(dst0.At(x, y), dst1.At(x, y)) {
-					t.Errorf("%s: at (%d, %d), %v versus %v",
-						dName, x, y, dst0.At(x, y), dst1.At(x, y))
-					continue loop
+	loop:
+		for sName, src := range sources {
+			dst0 := image.NewPaletted(b, cgaPalette)
+			dst1 := image.NewPaletted(b, cgaPalette)
+			d.Draw(dst0, b, src, image.Point{})
+			d.Draw(embeddedPaletted{dst1}, b, src, image.Point{})
+			for y := b.Min.Y; y < b.Max.Y; y++ {
+				for x := b.Min.X; x < b.Max.X; x++ {
+					if !eq(dst0.At(x, y), dst1.At(x, y)) {
+						t.Errorf("%s / %s: at (%d, %d), %v versus %v",
+							dName, sName, x, y, dst0.At(x, y), dst1.At(x, y))
+						continue loop
+					}
 				}
 			}
 		}