Imported Upstream version 60upstream/60

5 files changed, 1257 insertions, 0 deletions

diff --git a/src/pkg/image/draw/Makefile b/src/pkg/image/draw/Makefile
new file mode 100644
index 000000000..2ba6e7b51
--- /dev/null
+++ b/src/pkg/image/draw/Makefile
@@ -0,0 +1,11 @@
+# Copyright 2009 The Go Authors.  All rights reserved.
+# Use of this source code is governed by a BSD-style
+# license that can be found in the LICENSE file.
+
+include ../../../Make.inc
+
+TARG=image/draw
+GOFILES=\
+	draw.go\
+
+include ../../../Make.pkg
diff --git a/src/pkg/image/draw/bench_test.go b/src/pkg/image/draw/bench_test.go
new file mode 100644
index 000000000..a99b40814
--- /dev/null
+++ b/src/pkg/image/draw/bench_test.go
@@ -0,0 +1,206 @@
+// Copyright 2011 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package draw
+
+import (
+	"image"
+	"image/ycbcr"
+	"testing"
+)
+
+const (
+	dstw, dsth = 640, 480
+	srcw, srch = 400, 300
+)
+
+// bench benchmarks drawing src and mask images onto a dst image with the
+// given op and the color models to create those images from.
+// The created images' pixels are initialized to non-zero values.
+func bench(b *testing.B, dcm, scm, mcm image.ColorModel, op Op) {
+	b.StopTimer()
+
+	var dst Image
+	switch dcm {
+	case image.RGBAColorModel:
+		dst1 := image.NewRGBA(dstw, dsth)
+		for y := 0; y < dsth; y++ {
+			for x := 0; x < dstw; x++ {
+				dst1.SetRGBA(x, y, image.RGBAColor{
+					uint8(5 * x % 0x100),
+					uint8(7 * y % 0x100),
+					uint8((7*x + 5*y) % 0x100),
+					0xff,
+				})
+			}
+		}
+		dst = dst1
+	case image.RGBA64ColorModel:
+		dst1 := image.NewRGBA64(dstw, dsth)
+		for y := 0; y < dsth; y++ {
+			for x := 0; x < dstw; x++ {
+				dst1.SetRGBA64(x, y, image.RGBA64Color{
+					uint16(53 * x % 0x10000),
+					uint16(59 * y % 0x10000),
+					uint16((59*x + 53*y) % 0x10000),
+					0xffff,
+				})
+			}
+		}
+		dst = dst1
+	default:
+		panic("unreachable")
+	}
+
+	var src image.Image
+	switch scm {
+	case nil:
+		src = &image.ColorImage{image.RGBAColor{0x11, 0x22, 0x33, 0xff}}
+	case image.RGBAColorModel:
+		src1 := image.NewRGBA(srcw, srch)
+		for y := 0; y < srch; y++ {
+			for x := 0; x < srcw; x++ {
+				src1.SetRGBA(x, y, image.RGBAColor{
+					uint8(13 * x % 0x80),
+					uint8(11 * y % 0x80),
+					uint8((11*x + 13*y) % 0x80),
+					0x7f,
+				})
+			}
+		}
+		src = src1
+	case image.RGBA64ColorModel:
+		src1 := image.NewRGBA64(srcw, srch)
+		for y := 0; y < srch; y++ {
+			for x := 0; x < srcw; x++ {
+				src1.SetRGBA64(x, y, image.RGBA64Color{
+					uint16(103 * x % 0x8000),
+					uint16(101 * y % 0x8000),
+					uint16((101*x + 103*y) % 0x8000),
+					0x7fff,
+				})
+			}
+		}
+		src = src1
+	case image.NRGBAColorModel:
+		src1 := image.NewNRGBA(srcw, srch)
+		for y := 0; y < srch; y++ {
+			for x := 0; x < srcw; x++ {
+				src1.SetNRGBA(x, y, image.NRGBAColor{
+					uint8(13 * x % 0x100),
+					uint8(11 * y % 0x100),
+					uint8((11*x + 13*y) % 0x100),
+					0x7f,
+				})
+			}
+		}
+		src = src1
+	case ycbcr.YCbCrColorModel:
+		yy := make([]uint8, srcw*srch)
+		cb := make([]uint8, srcw*srch)
+		cr := make([]uint8, srcw*srch)
+		for i := range yy {
+			yy[i] = uint8(3 * i % 0x100)
+			cb[i] = uint8(5 * i % 0x100)
+			cr[i] = uint8(7 * i % 0x100)
+		}
+		src = &ycbcr.YCbCr{
+			Y:              yy,
+			Cb:             cb,
+			Cr:             cr,
+			YStride:        srcw,
+			CStride:        srcw,
+			SubsampleRatio: ycbcr.SubsampleRatio444,
+			Rect:           image.Rect(0, 0, srcw, srch),
+		}
+	default:
+		panic("unreachable")
+	}
+
+	var mask image.Image
+	switch mcm {
+	case nil:
+		// No-op.
+	case image.AlphaColorModel:
+		mask1 := image.NewAlpha(srcw, srch)
+		for y := 0; y < srch; y++ {
+			for x := 0; x < srcw; x++ {
+				a := uint8((23*x + 29*y) % 0x100)
+				// Glyph masks are typically mostly zero,
+				// so we only set a quarter of mask1's pixels.
+				if a >= 0xc0 {
+					mask1.SetAlpha(x, y, image.AlphaColor{a})
+				}
+			}
+		}
+		mask = mask1
+	default:
+		panic("unreachable")
+	}
+
+	b.StartTimer()
+	for i := 0; i < b.N; i++ {
+		// Scatter the destination rectangle to draw into.
+		x := 3 * i % (dstw - srcw)
+		y := 7 * i % (dsth - srch)
+
+		DrawMask(dst, dst.Bounds().Add(image.Point{x, y}), src, image.ZP, mask, image.ZP, op)
+	}
+}
+
+// The BenchmarkFoo functions exercise a drawFoo fast-path function in draw.go.
+
+func BenchmarkFillOver(b *testing.B) {
+	bench(b, image.RGBAColorModel, nil, nil, Over)
+}
+
+func BenchmarkFillSrc(b *testing.B) {
+	bench(b, image.RGBAColorModel, nil, nil, Src)
+}
+
+func BenchmarkCopyOver(b *testing.B) {
+	bench(b, image.RGBAColorModel, image.RGBAColorModel, nil, Over)
+}
+
+func BenchmarkCopySrc(b *testing.B) {
+	bench(b, image.RGBAColorModel, image.RGBAColorModel, nil, Src)
+}
+
+func BenchmarkNRGBAOver(b *testing.B) {
+	bench(b, image.RGBAColorModel, image.NRGBAColorModel, nil, Over)
+}
+
+func BenchmarkNRGBASrc(b *testing.B) {
+	bench(b, image.RGBAColorModel, image.NRGBAColorModel, nil, Src)
+}
+
+func BenchmarkYCbCr(b *testing.B) {
+	bench(b, image.RGBAColorModel, ycbcr.YCbCrColorModel, nil, Over)
+}
+
+func BenchmarkGlyphOver(b *testing.B) {
+	bench(b, image.RGBAColorModel, nil, image.AlphaColorModel, Over)
+}
+
+func BenchmarkRGBA(b *testing.B) {
+	bench(b, image.RGBAColorModel, image.RGBA64ColorModel, nil, Src)
+}
+
+// The BenchmarkGenericFoo functions exercise the generic, slow-path code.
+
+func BenchmarkGenericOver(b *testing.B) {
+	bench(b, image.RGBA64ColorModel, image.RGBA64ColorModel, nil, Over)
+}
+
+func BenchmarkGenericMaskOver(b *testing.B) {
+	bench(b, image.RGBA64ColorModel, image.RGBA64ColorModel, image.AlphaColorModel, Over)
+}
+
+func BenchmarkGenericSrc(b *testing.B) {
+	bench(b, image.RGBA64ColorModel, image.RGBA64ColorModel, nil, Src)
+}
+
+func BenchmarkGenericMaskSrc(b *testing.B) {
+	bench(b, image.RGBA64ColorModel, image.RGBA64ColorModel, image.AlphaColorModel, Src)
+}
diff --git a/src/pkg/image/draw/clip_test.go b/src/pkg/image/draw/clip_test.go
new file mode 100644
index 000000000..db40d82f5
--- /dev/null
+++ b/src/pkg/image/draw/clip_test.go
@@ -0,0 +1,193 @@
+// Copyright 2011 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package draw
+
+import (
+	"image"
+	"testing"
+)
+
+type clipTest struct {
+	desc          string
+	r, dr, sr, mr image.Rectangle
+	sp, mp        image.Point
+	nilMask       bool
+	r0            image.Rectangle
+	sp0, mp0      image.Point
+}
+
+var clipTests = []clipTest{
+	// The following tests all have a nil mask.
+	{
+		"basic",
+		image.Rect(0, 0, 100, 100),
+		image.Rect(0, 0, 100, 100),
+		image.Rect(0, 0, 100, 100),
+		image.ZR,
+		image.ZP,
+		image.ZP,
+		true,
+		image.Rect(0, 0, 100, 100),
+		image.ZP,
+		image.ZP,
+	},
+	{
+		"clip dr",
+		image.Rect(0, 0, 100, 100),
+		image.Rect(40, 40, 60, 60),
+		image.Rect(0, 0, 100, 100),
+		image.ZR,
+		image.ZP,
+		image.ZP,
+		true,
+		image.Rect(40, 40, 60, 60),
+		image.Pt(40, 40),
+		image.ZP,
+	},
+	{
+		"clip sr",
+		image.Rect(0, 0, 100, 100),
+		image.Rect(0, 0, 100, 100),
+		image.Rect(20, 20, 80, 80),
+		image.ZR,
+		image.ZP,
+		image.ZP,
+		true,
+		image.Rect(20, 20, 80, 80),
+		image.Pt(20, 20),
+		image.ZP,
+	},
+	{
+		"clip dr and sr",
+		image.Rect(0, 0, 100, 100),
+		image.Rect(0, 0, 50, 100),
+		image.Rect(20, 20, 80, 80),
+		image.ZR,
+		image.ZP,
+		image.ZP,
+		true,
+		image.Rect(20, 20, 50, 80),
+		image.Pt(20, 20),
+		image.ZP,
+	},
+	{
+		"clip dr and sr, sp outside sr (top-left)",
+		image.Rect(0, 0, 100, 100),
+		image.Rect(0, 0, 50, 100),
+		image.Rect(20, 20, 80, 80),
+		image.ZR,
+		image.Pt(15, 8),
+		image.ZP,
+		true,
+		image.Rect(5, 12, 50, 72),
+		image.Pt(20, 20),
+		image.ZP,
+	},
+	{
+		"clip dr and sr, sp outside sr (middle-left)",
+		image.Rect(0, 0, 100, 100),
+		image.Rect(0, 0, 50, 100),
+		image.Rect(20, 20, 80, 80),
+		image.ZR,
+		image.Pt(15, 66),
+		image.ZP,
+		true,
+		image.Rect(5, 0, 50, 14),
+		image.Pt(20, 66),
+		image.ZP,
+	},
+	{
+		"clip dr and sr, sp outside sr (bottom-left)",
+		image.Rect(0, 0, 100, 100),
+		image.Rect(0, 0, 50, 100),
+		image.Rect(20, 20, 80, 80),
+		image.ZR,
+		image.Pt(15, 91),
+		image.ZP,
+		true,
+		image.ZR,
+		image.Pt(15, 91),
+		image.ZP,
+	},
+	{
+		"clip dr and sr, sp inside sr",
+		image.Rect(0, 0, 100, 100),
+		image.Rect(0, 0, 50, 100),
+		image.Rect(20, 20, 80, 80),
+		image.ZR,
+		image.Pt(44, 33),
+		image.ZP,
+		true,
+		image.Rect(0, 0, 36, 47),
+		image.Pt(44, 33),
+		image.ZP,
+	},
+
+	// The following tests all have a non-nil mask.
+	{
+		"basic mask",
+		image.Rect(0, 0, 80, 80),
+		image.Rect(20, 0, 100, 80),
+		image.Rect(0, 0, 50, 49),
+		image.Rect(0, 0, 46, 47),
+		image.ZP,
+		image.ZP,
+		false,
+		image.Rect(20, 0, 46, 47),
+		image.Pt(20, 0),
+		image.Pt(20, 0),
+	},
+	// TODO(nigeltao): write more tests.
+}
+
+func TestClip(t *testing.T) {
+	dst0 := image.NewRGBA(100, 100)
+	src0 := image.NewRGBA(100, 100)
+	mask0 := image.NewRGBA(100, 100)
+	for _, c := range clipTests {
+		dst := dst0.SubImage(c.dr).(*image.RGBA)
+		src := src0.SubImage(c.sr).(*image.RGBA)
+		var mask image.Image
+		if !c.nilMask {
+			mask = mask0.SubImage(c.mr)
+		}
+		r, sp, mp := c.r, c.sp, c.mp
+		clip(dst, &r, src, &sp, mask, &mp)
+
+		// Check that the actual results equal the expected results.
+		if !c.r0.Eq(r) {
+			t.Errorf("%s: clip rectangle want %v got %v", c.desc, c.r0, r)
+			continue
+		}
+		if !c.sp0.Eq(sp) {
+			t.Errorf("%s: sp want %v got %v", c.desc, c.sp0, sp)
+			continue
+		}
+		if !c.nilMask {
+			if !c.mp0.Eq(mp) {
+				t.Errorf("%s: mp want %v got %v", c.desc, c.mp0, mp)
+				continue
+			}
+		}
+
+		// Check that the clipped rectangle is contained by the dst / src / mask
+		// rectangles, in their respective co-ordinate spaces.
+		if !r.In(c.dr) {
+			t.Errorf("%s: c.dr %v does not contain r %v", c.desc, c.dr, r)
+		}
+		// sr is r translated into src's co-ordinate space.
+		sr := r.Add(c.sp.Sub(c.dr.Min))
+		if !sr.In(c.sr) {
+			t.Errorf("%s: c.sr %v does not contain sr %v", c.desc, c.sr, sr)
+		}
+		if !c.nilMask {
+			// mr is r translated into mask's co-ordinate space.
+			mr := r.Add(c.mp.Sub(c.dr.Min))
+			if !mr.In(c.mr) {
+				t.Errorf("%s: c.mr %v does not contain mr %v", c.desc, c.mr, mr)
+			}
+		}
+	}
+}
diff --git a/src/pkg/image/draw/draw.go b/src/pkg/image/draw/draw.go
new file mode 100644
index 000000000..a748ff8c7
--- /dev/null
+++ b/src/pkg/image/draw/draw.go
@@ -0,0 +1,493 @@
+// Copyright 2009 The Go Authors.  All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Package draw provides image composition functions
+// in the style of the Plan 9 graphics library
+// (see http://plan9.bell-labs.com/magic/man2html/2/draw)
+// and the X Render extension.
+package draw
+
+import (
+	"image"
+	"image/ycbcr"
+)
+
+// m is the maximum color value returned by image.Color.RGBA.
+const m = 1<<16 - 1
+
+// Op is a Porter-Duff compositing operator.
+type Op int
+
+const (
+	// Over specifies ``(src in mask) over dst''.
+	Over Op = iota
+	// Src specifies ``src in mask''.
+	Src
+)
+
+var zeroColor image.Color = image.AlphaColor{0}
+
+// A draw.Image is an image.Image with a Set method to change a single pixel.
+type Image interface {
+	image.Image
+	Set(x, y int, c image.Color)
+}
+
+// Draw calls DrawMask with a nil mask.
+func Draw(dst Image, r image.Rectangle, src image.Image, sp image.Point, op Op) {
+	DrawMask(dst, r, src, sp, nil, image.ZP, op)
+}
+
+// clip clips r against each image's bounds (after translating into the
+// destination image's co-ordinate space) and shifts the points sp and mp by
+// the same amount as the change in r.Min.
+func clip(dst Image, r *image.Rectangle, src image.Image, sp *image.Point, mask image.Image, mp *image.Point) {
+	orig := r.Min
+	*r = r.Intersect(dst.Bounds())
+	*r = r.Intersect(src.Bounds().Add(orig.Sub(*sp)))
+	if mask != nil {
+		*r = r.Intersect(mask.Bounds().Add(orig.Sub(*mp)))
+	}
+	dx := r.Min.X - orig.X
+	dy := r.Min.Y - orig.Y
+	if dx == 0 && dy == 0 {
+		return
+	}
+	(*sp).X += dx
+	(*sp).Y += dy
+	(*mp).X += dx
+	(*mp).Y += dy
+}
+
+// DrawMask aligns r.Min in dst with sp in src and mp in mask and then replaces the rectangle r
+// in dst with the result of a Porter-Duff composition. A nil mask is treated as opaque.
+func DrawMask(dst Image, r image.Rectangle, src image.Image, sp image.Point, mask image.Image, mp image.Point, op Op) {
+	clip(dst, &r, src, &sp, mask, &mp)
+	if r.Empty() {
+		return
+	}
+
+	// Fast paths for special cases. If none of them apply, then we fall back to a general but slow implementation.
+	if dst0, ok := dst.(*image.RGBA); ok {
+		if op == Over {
+			if mask == nil {
+				switch src0 := src.(type) {
+				case *image.ColorImage:
+					drawFillOver(dst0, r, src0)
+					return
+				case *image.RGBA:
+					drawCopyOver(dst0, r, src0, sp)
+					return
+				case *image.NRGBA:
+					drawNRGBAOver(dst0, r, src0, sp)
+					return
+				case *ycbcr.YCbCr:
+					drawYCbCr(dst0, r, src0, sp)
+					return
+				}
+			} else if mask0, ok := mask.(*image.Alpha); ok {
+				switch src0 := src.(type) {
+				case *image.ColorImage:
+					drawGlyphOver(dst0, r, src0, mask0, mp)
+					return
+				}
+			}
+		} else {
+			if mask == nil {
+				switch src0 := src.(type) {
+				case *image.ColorImage:
+					drawFillSrc(dst0, r, src0)
+					return
+				case *image.RGBA:
+					drawCopySrc(dst0, r, src0, sp)
+					return
+				case *image.NRGBA:
+					drawNRGBASrc(dst0, r, src0, sp)
+					return
+				case *ycbcr.YCbCr:
+					drawYCbCr(dst0, r, src0, sp)
+					return
+				}
+			}
+		}
+		drawRGBA(dst0, r, src, sp, mask, mp, op)
+		return
+	}
+
+	x0, x1, dx := r.Min.X, r.Max.X, 1
+	y0, y1, dy := r.Min.Y, r.Max.Y, 1
+	if image.Image(dst) == src && r.Overlaps(r.Add(sp.Sub(r.Min))) {
+		// Rectangles overlap: process backward?
+		if sp.Y < r.Min.Y || sp.Y == r.Min.Y && sp.X < r.Min.X {
+			x0, x1, dx = x1-1, x0-1, -1
+			y0, y1, dy = y1-1, y0-1, -1
+		}
+	}
+
+	var out *image.RGBA64Color
+	sy := sp.Y + y0 - r.Min.Y
+	my := mp.Y + y0 - r.Min.Y
+	for y := y0; y != y1; y, sy, my = y+dy, sy+dy, my+dy {
+		sx := sp.X + x0 - r.Min.X
+		mx := mp.X + x0 - r.Min.X
+		for x := x0; x != x1; x, sx, mx = x+dx, sx+dx, mx+dx {
+			ma := uint32(m)
+			if mask != nil {
+				_, _, _, ma = mask.At(mx, my).RGBA()
+			}
+			switch {
+			case ma == 0:
+				if op == Over {
+					// No-op.
+				} else {
+					dst.Set(x, y, zeroColor)
+				}
+			case ma == m && op == Src:
+				dst.Set(x, y, src.At(sx, sy))
+			default:
+				sr, sg, sb, sa := src.At(sx, sy).RGBA()
+				if out == nil {
+					out = new(image.RGBA64Color)
+				}
+				if op == Over {
+					dr, dg, db, da := dst.At(x, y).RGBA()
+					a := m - (sa * ma / m)
+					out.R = uint16((dr*a + sr*ma) / m)
+					out.G = uint16((dg*a + sg*ma) / m)
+					out.B = uint16((db*a + sb*ma) / m)
+					out.A = uint16((da*a + sa*ma) / m)
+				} else {
+					out.R = uint16(sr * ma / m)
+					out.G = uint16(sg * ma / m)
+					out.B = uint16(sb * ma / m)
+					out.A = uint16(sa * ma / m)
+				}
+				dst.Set(x, y, out)
+			}
+		}
+	}
+}
+
+func drawFillOver(dst *image.RGBA, r image.Rectangle, src *image.ColorImage) {
+	sr, sg, sb, sa := src.RGBA()
+	// The 0x101 is here for the same reason as in drawRGBA.
+	a := (m - sa) * 0x101
+	i0 := (r.Min.Y-dst.Rect.Min.Y)*dst.Stride + (r.Min.X-dst.Rect.Min.X)*4
+	i1 := i0 + r.Dx()*4
+	for y := r.Min.Y; y != r.Max.Y; y++ {
+		for i := i0; i < i1; i += 4 {
+			dr := uint32(dst.Pix[i+0])
+			dg := uint32(dst.Pix[i+1])
+			db := uint32(dst.Pix[i+2])
+			da := uint32(dst.Pix[i+3])
+
+			dst.Pix[i+0] = uint8((dr*a/m + sr) >> 8)
+			dst.Pix[i+1] = uint8((dg*a/m + sg) >> 8)
+			dst.Pix[i+2] = uint8((db*a/m + sb) >> 8)
+			dst.Pix[i+3] = uint8((da*a/m + sa) >> 8)
+		}
+		i0 += dst.Stride
+		i1 += dst.Stride
+	}
+}
+
+func drawFillSrc(dst *image.RGBA, r image.Rectangle, src *image.ColorImage) {
+	sr, sg, sb, sa := src.RGBA()
+	// The built-in copy function is faster than a straightforward for loop to fill the destination with
+	// the color, but copy requires a slice source. We therefore use a for loop to fill the first row, and
+	// then use the first row as the slice source for the remaining rows.
+	i0 := (r.Min.Y-dst.Rect.Min.Y)*dst.Stride + (r.Min.X-dst.Rect.Min.X)*4
+	i1 := i0 + r.Dx()*4
+	for i := i0; i < i1; i += 4 {
+		dst.Pix[i+0] = uint8(sr >> 8)
+		dst.Pix[i+1] = uint8(sg >> 8)
+		dst.Pix[i+2] = uint8(sb >> 8)
+		dst.Pix[i+3] = uint8(sa >> 8)
+	}
+	firstRow := dst.Pix[i0:i1]
+	for y := r.Min.Y + 1; y < r.Max.Y; y++ {
+		i0 += dst.Stride
+		i1 += dst.Stride
+		copy(dst.Pix[i0:i1], firstRow)
+	}
+}
+
+func drawCopyOver(dst *image.RGBA, r image.Rectangle, src *image.RGBA, sp image.Point) {
+	dx, dy := r.Dx(), r.Dy()
+	d0 := (r.Min.Y-dst.Rect.Min.Y)*dst.Stride + (r.Min.X-dst.Rect.Min.X)*4
+	s0 := (sp.Y-src.Rect.Min.Y)*src.Stride + (sp.X-src.Rect.Min.X)*4
+	var (
+		ddelta, sdelta int
+		i0, i1, idelta int
+	)
+	if r.Min.Y < sp.Y || r.Min.Y == sp.Y && r.Min.X <= sp.X {
+		ddelta = dst.Stride
+		sdelta = src.Stride
+		i0, i1, idelta = 0, dx*4, +4
+	} else {
+		// If the source start point is higher than the destination start point, or equal height but to the left,
+		// then we compose the rows in right-to-left, bottom-up order instead of left-to-right, top-down.
+		d0 += (dy - 1) * dst.Stride
+		s0 += (dy - 1) * src.Stride
+		ddelta = -dst.Stride
+		sdelta = -src.Stride
+		i0, i1, idelta = (dx-1)*4, -4, -4
+	}
+	for ; dy > 0; dy-- {
+		dpix := dst.Pix[d0:]
+		spix := src.Pix[s0:]
+		for i := i0; i != i1; i += idelta {
+			sr := uint32(spix[i+0]) * 0x101
+			sg := uint32(spix[i+1]) * 0x101
+			sb := uint32(spix[i+2]) * 0x101
+			sa := uint32(spix[i+3]) * 0x101
+
+			dr := uint32(dpix[i+0])
+			dg := uint32(dpix[i+1])
+			db := uint32(dpix[i+2])
+			da := uint32(dpix[i+3])
+
+			// The 0x101 is here for the same reason as in drawRGBA.
+			a := (m - sa) * 0x101
+
+			dpix[i+0] = uint8((dr*a/m + sr) >> 8)
+			dpix[i+1] = uint8((dg*a/m + sg) >> 8)
+			dpix[i+2] = uint8((db*a/m + sb) >> 8)
+			dpix[i+3] = uint8((da*a/m + sa) >> 8)
+		}
+		d0 += ddelta
+		s0 += sdelta
+	}
+}
+
+func drawCopySrc(dst *image.RGBA, r image.Rectangle, src *image.RGBA, sp image.Point) {
+	n, dy := 4*r.Dx(), r.Dy()
+	d0 := (r.Min.Y-dst.Rect.Min.Y)*dst.Stride + (r.Min.X-dst.Rect.Min.X)*4
+	s0 := (sp.Y-src.Rect.Min.Y)*src.Stride + (sp.X-src.Rect.Min.X)*4
+	var ddelta, sdelta int
+	if r.Min.Y <= sp.Y {
+		ddelta = dst.Stride
+		sdelta = src.Stride
+	} else {
+		// If the source start point is higher than the destination start point, then we compose the rows
+		// in bottom-up order instead of top-down. Unlike the drawCopyOver function, we don't have to
+		// check the x co-ordinates because the built-in copy function can handle overlapping slices.
+		d0 += (dy - 1) * dst.Stride
+		s0 += (dy - 1) * src.Stride
+		ddelta = -dst.Stride
+		sdelta = -src.Stride
+	}
+	for ; dy > 0; dy-- {
+		copy(dst.Pix[d0:d0+n], src.Pix[s0:s0+n])
+		d0 += ddelta
+		s0 += sdelta
+	}
+}
+
+func drawNRGBAOver(dst *image.RGBA, r image.Rectangle, src *image.NRGBA, sp image.Point) {
+	i0 := (r.Min.X - dst.Rect.Min.X) * 4
+	i1 := (r.Max.X - dst.Rect.Min.X) * 4
+	si0 := (sp.X - src.Rect.Min.X) * 4
+	yMax := r.Max.Y - dst.Rect.Min.Y
+
+	y := r.Min.Y - dst.Rect.Min.Y
+	sy := sp.Y - src.Rect.Min.Y
+	for ; y != yMax; y, sy = y+1, sy+1 {
+		dpix := dst.Pix[y*dst.Stride:]
+		spix := src.Pix[sy*src.Stride:]
+
+		for i, si := i0, si0; i < i1; i, si = i+4, si+4 {
+			// Convert from non-premultiplied color to pre-multiplied color.
+			sa := uint32(spix[si+3]) * 0x101
+			sr := uint32(spix[si+0]) * sa / 0xff
+			sg := uint32(spix[si+1]) * sa / 0xff
+			sb := uint32(spix[si+2]) * sa / 0xff
+
+			dr := uint32(dpix[i+0])
+			dg := uint32(dpix[i+1])
+			db := uint32(dpix[i+2])
+			da := uint32(dpix[i+3])
+
+			// The 0x101 is here for the same reason as in drawRGBA.
+			a := (m - sa) * 0x101
+
+			dpix[i+0] = uint8((dr*a/m + sr) >> 8)
+			dpix[i+1] = uint8((dg*a/m + sg) >> 8)
+			dpix[i+2] = uint8((db*a/m + sb) >> 8)
+			dpix[i+3] = uint8((da*a/m + sa) >> 8)
+		}
+	}
+}
+
+func drawNRGBASrc(dst *image.RGBA, r image.Rectangle, src *image.NRGBA, sp image.Point) {
+	i0 := (r.Min.X - dst.Rect.Min.X) * 4
+	i1 := (r.Max.X - dst.Rect.Min.X) * 4
+	si0 := (sp.X - src.Rect.Min.X) * 4
+	yMax := r.Max.Y - dst.Rect.Min.Y
+
+	y := r.Min.Y - dst.Rect.Min.Y
+	sy := sp.Y - src.Rect.Min.Y
+	for ; y != yMax; y, sy = y+1, sy+1 {
+		dpix := dst.Pix[y*dst.Stride:]
+		spix := src.Pix[sy*src.Stride:]
+
+		for i, si := i0, si0; i < i1; i, si = i+4, si+4 {
+			// Convert from non-premultiplied color to pre-multiplied color.
+			sa := uint32(spix[si+3]) * 0x101
+			sr := uint32(spix[si+0]) * sa / 0xff
+			sg := uint32(spix[si+1]) * sa / 0xff
+			sb := uint32(spix[si+2]) * sa / 0xff
+
+			dpix[i+0] = uint8(sr >> 8)
+			dpix[i+1] = uint8(sg >> 8)
+			dpix[i+2] = uint8(sb >> 8)
+			dpix[i+3] = uint8(sa >> 8)
+		}
+	}
+}
+
+func drawYCbCr(dst *image.RGBA, r image.Rectangle, src *ycbcr.YCbCr, sp image.Point) {
+	// A YCbCr image is always fully opaque, and so if the mask is implicitly nil
+	// (i.e. fully opaque) then the op is effectively always Src.
+	var (
+		yy, cb, cr uint8
+	)
+	x0 := (r.Min.X - dst.Rect.Min.X) * 4
+	x1 := (r.Max.X - dst.Rect.Min.X) * 4
+	y0 := r.Min.Y - dst.Rect.Min.Y
+	y1 := r.Max.Y - dst.Rect.Min.Y
+	switch src.SubsampleRatio {
+	case ycbcr.SubsampleRatio422:
+		for y, sy := y0, sp.Y; y != y1; y, sy = y+1, sy+1 {
+			dpix := dst.Pix[y*dst.Stride:]
+			for x, sx := x0, sp.X; x != x1; x, sx = x+4, sx+1 {
+				i := sx / 2
+				yy = src.Y[sy*src.YStride+sx]
+				cb = src.Cb[sy*src.CStride+i]
+				cr = src.Cr[sy*src.CStride+i]
+				rr, gg, bb := ycbcr.YCbCrToRGB(yy, cb, cr)
+				dpix[x+0] = rr
+				dpix[x+1] = gg
+				dpix[x+2] = bb
+				dpix[x+3] = 255
+			}
+		}
+	case ycbcr.SubsampleRatio420:
+		for y, sy := y0, sp.Y; y != y1; y, sy = y+1, sy+1 {
+			dpix := dst.Pix[y*dst.Stride:]
+			for x, sx := x0, sp.X; x != x1; x, sx = x+4, sx+1 {
+				i, j := sx/2, sy/2
+				yy = src.Y[sy*src.YStride+sx]
+				cb = src.Cb[j*src.CStride+i]
+				cr = src.Cr[j*src.CStride+i]
+				rr, gg, bb := ycbcr.YCbCrToRGB(yy, cb, cr)
+				dpix[x+0] = rr
+				dpix[x+1] = gg
+				dpix[x+2] = bb
+				dpix[x+3] = 255
+			}
+		}
+	default:
+		// Default to 4:4:4 subsampling.
+		for y, sy := y0, sp.Y; y != y1; y, sy = y+1, sy+1 {
+			dpix := dst.Pix[y*dst.Stride:]
+			for x, sx := x0, sp.X; x != x1; x, sx = x+4, sx+1 {
+				yy = src.Y[sy*src.YStride+sx]
+				cb = src.Cb[sy*src.CStride+sx]
+				cr = src.Cr[sy*src.CStride+sx]
+				rr, gg, bb := ycbcr.YCbCrToRGB(yy, cb, cr)
+				dpix[x+0] = rr
+				dpix[x+1] = gg
+				dpix[x+2] = bb
+				dpix[x+3] = 255
+			}
+		}
+	}
+}
+
+func drawGlyphOver(dst *image.RGBA, r image.Rectangle, src *image.ColorImage, mask *image.Alpha, mp image.Point) {
+	i0 := (r.Min.Y-dst.Rect.Min.Y)*dst.Stride + (r.Min.X-dst.Rect.Min.X)*4
+	i1 := i0 + r.Dx()*4
+	mi0 := (mp.Y-mask.Rect.Min.Y)*mask.Stride + mp.X - mask.Rect.Min.X
+	sr, sg, sb, sa := src.RGBA()
+	for y, my := r.Min.Y, mp.Y; y != r.Max.Y; y, my = y+1, my+1 {
+		for i, mi := i0, mi0; i < i1; i, mi = i+4, mi+1 {
+			ma := uint32(mask.Pix[mi])
+			if ma == 0 {
+				continue
+			}
+			ma |= ma << 8
+
+			dr := uint32(dst.Pix[i+0])
+			dg := uint32(dst.Pix[i+1])
+			db := uint32(dst.Pix[i+2])
+			da := uint32(dst.Pix[i+3])
+
+			// The 0x101 is here for the same reason as in drawRGBA.
+			a := (m - (sa * ma / m)) * 0x101
+
+			dst.Pix[i+0] = uint8((dr*a + sr*ma) / m >> 8)
+			dst.Pix[i+1] = uint8((dg*a + sg*ma) / m >> 8)
+			dst.Pix[i+2] = uint8((db*a + sb*ma) / m >> 8)
+			dst.Pix[i+3] = uint8((da*a + sa*ma) / m >> 8)
+		}
+		i0 += dst.Stride
+		i1 += dst.Stride
+		mi0 += mask.Stride
+	}
+}
+
+func drawRGBA(dst *image.RGBA, r image.Rectangle, src image.Image, sp image.Point, mask image.Image, mp image.Point, op Op) {
+	x0, x1, dx := r.Min.X, r.Max.X, 1
+	y0, y1, dy := r.Min.Y, r.Max.Y, 1
+	if image.Image(dst) == src && r.Overlaps(r.Add(sp.Sub(r.Min))) {
+		if sp.Y < r.Min.Y || sp.Y == r.Min.Y && sp.X < r.Min.X {
+			x0, x1, dx = x1-1, x0-1, -1
+			y0, y1, dy = y1-1, y0-1, -1
+		}
+	}
+
+	sy := sp.Y + y0 - r.Min.Y
+	my := mp.Y + y0 - r.Min.Y
+	sx0 := sp.X + x0 - r.Min.X
+	mx0 := mp.X + x0 - r.Min.X
+	sx1 := sx0 + (x1 - x0)
+	i0 := (y0-dst.Rect.Min.Y)*dst.Stride + (x0-dst.Rect.Min.X)*4
+	di := dx * 4
+	for y := y0; y != y1; y, sy, my = y+dy, sy+dy, my+dy {
+		for i, sx, mx := i0, sx0, mx0; sx != sx1; i, sx, mx = i+di, sx+dx, mx+dx {
+			ma := uint32(m)
+			if mask != nil {
+				_, _, _, ma = mask.At(mx, my).RGBA()
+			}
+			sr, sg, sb, sa := src.At(sx, sy).RGBA()
+			if op == Over {
+				dr := uint32(dst.Pix[i+0])
+				dg := uint32(dst.Pix[i+1])
+				db := uint32(dst.Pix[i+2])
+				da := uint32(dst.Pix[i+3])
+
+				// dr, dg, db and da are all 8-bit color at the moment, ranging in [0,255].
+				// We work in 16-bit color, and so would normally do:
+				// dr |= dr << 8
+				// and similarly for dg, db and da, but instead we multiply a
+				// (which is a 16-bit color, ranging in [0,65535]) by 0x101.
+				// This yields the same result, but is fewer arithmetic operations.
+				a := (m - (sa * ma / m)) * 0x101
+
+				dst.Pix[i+0] = uint8((dr*a + sr*ma) / m >> 8)
+				dst.Pix[i+1] = uint8((dg*a + sg*ma) / m >> 8)
+				dst.Pix[i+2] = uint8((db*a + sb*ma) / m >> 8)
+				dst.Pix[i+3] = uint8((da*a + sa*ma) / m >> 8)
+
+			} else {
+				dst.Pix[i+0] = uint8(sr * ma / m >> 8)
+				dst.Pix[i+1] = uint8(sg * ma / m >> 8)
+				dst.Pix[i+2] = uint8(sb * ma / m >> 8)
+				dst.Pix[i+3] = uint8(sa * ma / m >> 8)
+			}
+		}
+		i0 += dy * dst.Stride
+	}
+}
diff --git a/src/pkg/image/draw/draw_test.go b/src/pkg/image/draw/draw_test.go
new file mode 100644
index 000000000..55435cc27
--- /dev/null
+++ b/src/pkg/image/draw/draw_test.go
@@ -0,0 +1,354 @@
+// Copyright 2010 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package draw
+
+import (
+	"image"
+	"image/ycbcr"
+	"testing"
+)
+
+func eq(c0, c1 image.Color) bool {
+	r0, g0, b0, a0 := c0.RGBA()
+	r1, g1, b1, a1 := c1.RGBA()
+	return r0 == r1 && g0 == g1 && b0 == b1 && a0 == a1
+}
+
+func fillBlue(alpha int) image.Image {
+	return image.NewColorImage(image.RGBAColor{0, 0, uint8(alpha), uint8(alpha)})
+}
+
+func fillAlpha(alpha int) image.Image {
+	return image.NewColorImage(image.AlphaColor{uint8(alpha)})
+}
+
+func vgradGreen(alpha int) image.Image {
+	m := image.NewRGBA(16, 16)
+	for y := 0; y < 16; y++ {
+		for x := 0; x < 16; x++ {
+			m.Set(x, y, image.RGBAColor{0, uint8(y * alpha / 15), 0, uint8(alpha)})
+		}
+	}
+	return m
+}
+
+func vgradAlpha(alpha int) image.Image {
+	m := image.NewAlpha(16, 16)
+	for y := 0; y < 16; y++ {
+		for x := 0; x < 16; x++ {
+			m.Set(x, y, image.AlphaColor{uint8(y * alpha / 15)})
+		}
+	}
+	return m
+}
+
+func vgradGreenNRGBA(alpha int) image.Image {
+	m := image.NewNRGBA(16, 16)
+	for y := 0; y < 16; y++ {
+		for x := 0; x < 16; x++ {
+			m.Set(x, y, image.RGBAColor{0, uint8(y * 0x11), 0, uint8(alpha)})
+		}
+	}
+	return m
+}
+
+func vgradCr() image.Image {
+	m := &ycbcr.YCbCr{
+		Y:              make([]byte, 16*16),
+		Cb:             make([]byte, 16*16),
+		Cr:             make([]byte, 16*16),
+		YStride:        16,
+		CStride:        16,
+		SubsampleRatio: ycbcr.SubsampleRatio444,
+		Rect:           image.Rect(0, 0, 16, 16),
+	}
+	for y := 0; y < 16; y++ {
+		for x := 0; x < 16; x++ {
+			m.Cr[y*m.CStride+x] = uint8(y * 0x11)
+		}
+	}
+	return m
+}
+
+func hgradRed(alpha int) Image {
+	m := image.NewRGBA(16, 16)
+	for y := 0; y < 16; y++ {
+		for x := 0; x < 16; x++ {
+			m.Set(x, y, image.RGBAColor{uint8(x * alpha / 15), 0, 0, uint8(alpha)})
+		}
+	}
+	return m
+}
+
+func gradYellow(alpha int) Image {
+	m := image.NewRGBA(16, 16)
+	for y := 0; y < 16; y++ {
+		for x := 0; x < 16; x++ {
+			m.Set(x, y, image.RGBAColor{uint8(x * alpha / 15), uint8(y * alpha / 15), 0, uint8(alpha)})
+		}
+	}
+	return m
+}
+
+type drawTest struct {
+	desc     string
+	src      image.Image
+	mask     image.Image
+	op       Op
+	expected image.Color
+}
+
+var drawTests = []drawTest{
+	// Uniform mask (0% opaque).
+	{"nop", vgradGreen(255), fillAlpha(0), Over, image.RGBAColor{136, 0, 0, 255}},
+	{"clear", vgradGreen(255), fillAlpha(0), Src, image.RGBAColor{0, 0, 0, 0}},
+	// Uniform mask (100%, 75%, nil) and uniform source.
+	// At (x, y) == (8, 8):
+	// The destination pixel is {136, 0, 0, 255}.
+	// The source pixel is {0, 0, 90, 90}.
+	{"fill", fillBlue(90), fillAlpha(255), Over, image.RGBAColor{88, 0, 90, 255}},
+	{"fillSrc", fillBlue(90), fillAlpha(255), Src, image.RGBAColor{0, 0, 90, 90}},
+	{"fillAlpha", fillBlue(90), fillAlpha(192), Over, image.RGBAColor{100, 0, 68, 255}},
+	{"fillAlphaSrc", fillBlue(90), fillAlpha(192), Src, image.RGBAColor{0, 0, 68, 68}},
+	{"fillNil", fillBlue(90), nil, Over, image.RGBAColor{88, 0, 90, 255}},
+	{"fillNilSrc", fillBlue(90), nil, Src, image.RGBAColor{0, 0, 90, 90}},
+	// Uniform mask (100%, 75%, nil) and variable source.
+	// At (x, y) == (8, 8):
+	// The destination pixel is {136, 0, 0, 255}.
+	// The source pixel is {0, 48, 0, 90}.
+	{"copy", vgradGreen(90), fillAlpha(255), Over, image.RGBAColor{88, 48, 0, 255}},
+	{"copySrc", vgradGreen(90), fillAlpha(255), Src, image.RGBAColor{0, 48, 0, 90}},
+	{"copyAlpha", vgradGreen(90), fillAlpha(192), Over, image.RGBAColor{100, 36, 0, 255}},
+	{"copyAlphaSrc", vgradGreen(90), fillAlpha(192), Src, image.RGBAColor{0, 36, 0, 68}},
+	{"copyNil", vgradGreen(90), nil, Over, image.RGBAColor{88, 48, 0, 255}},
+	{"copyNilSrc", vgradGreen(90), nil, Src, image.RGBAColor{0, 48, 0, 90}},
+	// Uniform mask (100%, 75%, nil) and variable NRGBA source.
+	// At (x, y) == (8, 8):
+	// The destination pixel is {136, 0, 0, 255}.
+	// The source pixel is {0, 136, 0, 90} in NRGBA-space, which is {0, 48, 0, 90} in RGBA-space.
+	// The result pixel is different than in the "copy*" test cases because of rounding errors.
+	{"nrgba", vgradGreenNRGBA(90), fillAlpha(255), Over, image.RGBAColor{88, 46, 0, 255}},
+	{"nrgbaSrc", vgradGreenNRGBA(90), fillAlpha(255), Src, image.RGBAColor{0, 46, 0, 90}},
+	{"nrgbaAlpha", vgradGreenNRGBA(90), fillAlpha(192), Over, image.RGBAColor{100, 34, 0, 255}},
+	{"nrgbaAlphaSrc", vgradGreenNRGBA(90), fillAlpha(192), Src, image.RGBAColor{0, 34, 0, 68}},
+	{"nrgbaNil", vgradGreenNRGBA(90), nil, Over, image.RGBAColor{88, 46, 0, 255}},
+	{"nrgbaNilSrc", vgradGreenNRGBA(90), nil, Src, image.RGBAColor{0, 46, 0, 90}},
+	// Uniform mask (100%, 75%, nil) and variable YCbCr source.
+	// At (x, y) == (8, 8):
+	// The destination pixel is {136, 0, 0, 255}.
+	// The source pixel is {0, 0, 136} in YCbCr-space, which is {11, 38, 0, 255} in RGB-space.
+	{"ycbcr", vgradCr(), fillAlpha(255), Over, image.RGBAColor{11, 38, 0, 255}},
+	{"ycbcrSrc", vgradCr(), fillAlpha(255), Src, image.RGBAColor{11, 38, 0, 255}},
+	{"ycbcrAlpha", vgradCr(), fillAlpha(192), Over, image.RGBAColor{42, 28, 0, 255}},
+	{"ycbcrAlphaSrc", vgradCr(), fillAlpha(192), Src, image.RGBAColor{8, 28, 0, 192}},
+	{"ycbcrNil", vgradCr(), nil, Over, image.RGBAColor{11, 38, 0, 255}},
+	{"ycbcrNilSrc", vgradCr(), nil, Src, image.RGBAColor{11, 38, 0, 255}},
+	// Variable mask and variable source.
+	// At (x, y) == (8, 8):
+	// The destination pixel is {136, 0, 0, 255}.
+	// The source pixel is {0, 0, 255, 255}.
+	// The mask pixel's alpha is 102, or 40%.
+	{"generic", fillBlue(255), vgradAlpha(192), Over, image.RGBAColor{81, 0, 102, 255}},
+	{"genericSrc", fillBlue(255), vgradAlpha(192), Src, image.RGBAColor{0, 0, 102, 102}},
+}
+
+func makeGolden(dst image.Image, r image.Rectangle, src image.Image, sp image.Point, mask image.Image, mp image.Point, op Op) image.Image {
+	// Since golden is a newly allocated image, we don't have to check if the
+	// input source and mask images and the output golden image overlap.
+	b := dst.Bounds()
+	sb := src.Bounds()
+	mb := image.Rect(-1e9, -1e9, 1e9, 1e9)
+	if mask != nil {
+		mb = mask.Bounds()
+	}
+	golden := image.NewRGBA(b.Max.X, b.Max.Y)
+	for y := r.Min.Y; y < r.Max.Y; y++ {
+		sy := y + sp.Y - r.Min.Y
+		my := y + mp.Y - r.Min.Y
+		for x := r.Min.X; x < r.Max.X; x++ {
+			if !(image.Point{x, y}.In(b)) {
+				continue
+			}
+			sx := x + sp.X - r.Min.X
+			if !(image.Point{sx, sy}.In(sb)) {
+				continue
+			}
+			mx := x + mp.X - r.Min.X
+			if !(image.Point{mx, my}.In(mb)) {
+				continue
+			}
+
+			const M = 1<<16 - 1
+			var dr, dg, db, da uint32
+			if op == Over {
+				dr, dg, db, da = dst.At(x, y).RGBA()
+			}
+			sr, sg, sb, sa := src.At(sx, sy).RGBA()
+			ma := uint32(M)
+			if mask != nil {
+				_, _, _, ma = mask.At(mx, my).RGBA()
+			}
+			a := M - (sa * ma / M)
+			golden.Set(x, y, image.RGBA64Color{
+				uint16((dr*a + sr*ma) / M),
+				uint16((dg*a + sg*ma) / M),
+				uint16((db*a + sb*ma) / M),
+				uint16((da*a + sa*ma) / M),
+			})
+		}
+	}
+	return golden.SubImage(b)
+}
+
+func TestDraw(t *testing.T) {
+	rr := []image.Rectangle{
+		image.Rect(0, 0, 0, 0),
+		image.Rect(0, 0, 16, 16),
+		image.Rect(3, 5, 12, 10),
+		image.Rect(0, 0, 9, 9),
+		image.Rect(8, 8, 16, 16),
+		image.Rect(8, 0, 9, 16),
+		image.Rect(0, 8, 16, 9),
+		image.Rect(8, 8, 9, 9),
+		image.Rect(8, 8, 8, 8),
+	}
+	for _, r := range rr {
+	loop:
+		for _, test := range drawTests {
+			dst := hgradRed(255).(*image.RGBA).SubImage(r).(Image)
+			// Draw the (src, mask, op) onto a copy of dst using a slow but obviously correct implementation.
+			golden := makeGolden(dst, image.Rect(0, 0, 16, 16), test.src, image.ZP, test.mask, image.ZP, test.op)
+			b := dst.Bounds()
+			if !b.Eq(golden.Bounds()) {
+				t.Errorf("draw %v %s: bounds %v versus %v", r, test.desc, dst.Bounds(), golden.Bounds())
+				continue
+			}
+			// Draw the same combination onto the actual dst using the optimized DrawMask implementation.
+			DrawMask(dst, image.Rect(0, 0, 16, 16), test.src, image.ZP, test.mask, image.ZP, test.op)
+			if image.Pt(8, 8).In(r) {
+				// Check that the resultant pixel at (8, 8) matches what we expect
+				// (the expected value can be verified by hand).
+				if !eq(dst.At(8, 8), test.expected) {
+					t.Errorf("draw %v %s: at (8, 8) %v versus %v", r, test.desc, dst.At(8, 8), test.expected)
+					continue
+				}
+			}
+			// Check that the resultant dst image matches the golden output.
+			for y := b.Min.Y; y < b.Max.Y; y++ {
+				for x := b.Min.X; x < b.Max.X; x++ {
+					if !eq(dst.At(x, y), golden.At(x, y)) {
+						t.Errorf("draw %v %s: at (%d, %d), %v versus golden %v", r, test.desc, x, y, dst.At(x, y), golden.At(x, y))
+						continue loop
+					}
+				}
+			}
+		}
+	}
+}
+
+func TestDrawOverlap(t *testing.T) {
+	for _, op := range []Op{Over, Src} {
+		for yoff := -2; yoff <= 2; yoff++ {
+		loop:
+			for xoff := -2; xoff <= 2; xoff++ {
+				m := gradYellow(127).(*image.RGBA)
+				dst := m.SubImage(image.Rect(5, 5, 10, 10)).(*image.RGBA)
+				src := m.SubImage(image.Rect(5+xoff, 5+yoff, 10+xoff, 10+yoff)).(*image.RGBA)
+				b := dst.Bounds()
+				// Draw the (src, mask, op) onto a copy of dst using a slow but obviously correct implementation.
+				golden := makeGolden(dst, b, src, src.Bounds().Min, nil, image.ZP, op)
+				if !b.Eq(golden.Bounds()) {
+					t.Errorf("drawOverlap xoff=%d,yoff=%d: bounds %v versus %v", xoff, yoff, dst.Bounds(), golden.Bounds())
+					continue
+				}
+				// Draw the same combination onto the actual dst using the optimized DrawMask implementation.
+				DrawMask(dst, b, src, src.Bounds().Min, nil, image.ZP, op)
+				// Check that the resultant dst image matches the golden output.
+				for y := b.Min.Y; y < b.Max.Y; y++ {
+					for x := b.Min.X; x < b.Max.X; x++ {
+						if !eq(dst.At(x, y), golden.At(x, y)) {
+							t.Errorf("drawOverlap xoff=%d,yoff=%d: at (%d, %d), %v versus golden %v", xoff, yoff, x, y, dst.At(x, y), golden.At(x, y))
+							continue loop
+						}
+					}
+				}
+			}
+		}
+	}
+}
+
+// TestNonZeroSrcPt checks drawing with a non-zero src point parameter.
+func TestNonZeroSrcPt(t *testing.T) {
+	a := image.NewRGBA(1, 1)
+	b := image.NewRGBA(2, 2)
+	b.Set(0, 0, image.RGBAColor{0, 0, 0, 5})
+	b.Set(1, 0, image.RGBAColor{0, 0, 5, 5})
+	b.Set(0, 1, image.RGBAColor{0, 5, 0, 5})
+	b.Set(1, 1, image.RGBAColor{5, 0, 0, 5})
+	Draw(a, image.Rect(0, 0, 1, 1), b, image.Pt(1, 1), Over)
+	if !eq(image.RGBAColor{5, 0, 0, 5}, a.At(0, 0)) {
+		t.Errorf("non-zero src pt: want %v got %v", image.RGBAColor{5, 0, 0, 5}, a.At(0, 0))
+	}
+}
+
+func TestFill(t *testing.T) {
+	rr := []image.Rectangle{
+		image.Rect(0, 0, 0, 0),
+		image.Rect(0, 0, 40, 30),
+		image.Rect(10, 0, 40, 30),
+		image.Rect(0, 20, 40, 30),
+		image.Rect(10, 20, 40, 30),
+		image.Rect(10, 20, 15, 25),
+		image.Rect(10, 0, 35, 30),
+		image.Rect(0, 15, 40, 16),
+		image.Rect(24, 24, 25, 25),
+		image.Rect(23, 23, 26, 26),
+		image.Rect(22, 22, 27, 27),
+		image.Rect(21, 21, 28, 28),
+		image.Rect(20, 20, 29, 29),
+	}
+	for _, r := range rr {
+		m := image.NewRGBA(40, 30).SubImage(r).(*image.RGBA)
+		b := m.Bounds()
+		c := image.RGBAColor{11, 0, 0, 255}
+		src := &image.ColorImage{c}
+		check := func(desc string) {
+			for y := b.Min.Y; y < b.Max.Y; y++ {
+				for x := b.Min.X; x < b.Max.X; x++ {
+					if !eq(c, m.At(x, y)) {
+						t.Errorf("%s fill: at (%d, %d), sub-image bounds=%v: want %v got %v", desc, x, y, r, c, m.At(x, y))
+						return
+					}
+				}
+			}
+		}
+		// Draw 1 pixel at a time.
+		for y := b.Min.Y; y < b.Max.Y; y++ {
+			for x := b.Min.X; x < b.Max.X; x++ {
+				DrawMask(m, image.Rect(x, y, x+1, y+1), src, image.ZP, nil, image.ZP, Src)
+			}
+		}
+		check("pixel")
+		// Draw 1 row at a time.
+		c = image.RGBAColor{0, 22, 0, 255}
+		src = &image.ColorImage{c}
+		for y := b.Min.Y; y < b.Max.Y; y++ {
+			DrawMask(m, image.Rect(b.Min.X, y, b.Max.X, y+1), src, image.ZP, nil, image.ZP, Src)
+		}
+		check("row")
+		// Draw 1 column at a time.
+		c = image.RGBAColor{0, 0, 33, 255}
+		src = &image.ColorImage{c}
+		for x := b.Min.X; x < b.Max.X; x++ {
+			DrawMask(m, image.Rect(x, b.Min.Y, x+1, b.Max.Y), src, image.ZP, nil, image.ZP, Src)
+		}
+		check("column")
+		// Draw the whole image at once.
+		c = image.RGBAColor{44, 55, 66, 77}
+		src = &image.ColorImage{c}
+		DrawMask(m, b, src, image.ZP, nil, image.ZP, Src)
+		check("whole")
+	}
+}