image: add PixOffset methods; use them in image/draw and image/tiff.

image/draw benchmarks show <1% change for the fast paths.
The slow paths got worse by 1-4%, but they're the slow paths.
I don't care so much about them, and presumably compiler improvements
could claw it back.

IIUC 6g's inlining is enabled by default now.

benchmark                        old ns/op    new ns/op    delta
draw.BenchmarkFillOver             2988384      2999624   +0.38%
draw.BenchmarkFillSrc               153141       153262   +0.08%
draw.BenchmarkCopyOver             2155756      2170831   +0.70%
draw.BenchmarkCopySrc                72591        72646   +0.08%
draw.BenchmarkNRGBAOver            2487372      2491576   +0.17%
draw.BenchmarkNRGBASrc             1361306      1409180   +3.52%
draw.BenchmarkYCbCr                2540712      2562359   +0.85%
draw.BenchmarkGlyphOver            1004879      1023308   +1.83%
draw.BenchmarkRGBA                 8746670      8844455   +1.12%
draw.BenchmarkGenericOver         31860960     32512960   +2.05%
draw.BenchmarkGenericMaskOver     16369060     16435720   +0.41%
draw.BenchmarkGenericSrc          13128540     13127810   -0.01%
draw.BenchmarkGenericMaskSrc      30059300     28883210   -3.91%

R=r, gri
CC=golang-dev, rsc
https://golang.org/cl/5536059

src/pkg/image/draw/draw.go

@@ -171,7 +171,7 @@ func drawFillOver(dst *image.RGBA, r image.Rectangle, src *image.Uniform) {
 	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
+	i0 := dst.PixOffset(r.Min.X, r.Min.Y)
 	i1 := i0 + r.Dx()*4
 	for y := r.Min.Y; y != r.Max.Y; y++ {
 		for i := i0; i < i1; i += 4 {
@@ -195,7 +195,7 @@ func drawFillSrc(dst *image.RGBA, r image.Rectangle, src *image.Uniform) {
 	// 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
+	i0 := dst.PixOffset(r.Min.X, r.Min.Y)
 	i1 := i0 + r.Dx()*4
 	for i := i0; i < i1; i += 4 {
 		dst.Pix[i+0] = uint8(sr >> 8)
@@ -213,8 +213,8 @@ func drawFillSrc(dst *image.RGBA, r image.Rectangle, src *image.Uniform) {
 
 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
+	d0 := dst.PixOffset(r.Min.X, r.Min.Y)
+	s0 := src.PixOffset(sp.X, sp.Y)
 	var (
 		ddelta, sdelta int
 		i0, i1, idelta int
@@ -261,8 +261,8 @@ func drawCopyOver(dst *image.RGBA, r image.Rectangle, src *image.RGBA, sp image.
 
 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
+	d0 := dst.PixOffset(r.Min.X, r.Min.Y)
+	s0 := src.PixOffset(sp.X, sp.Y)
 	var ddelta, sdelta int
 	if r.Min.Y <= sp.Y {
 		ddelta = dst.Stride
@@ -348,9 +348,7 @@ func drawNRGBASrc(dst *image.RGBA, r image.Rectangle, src *image.NRGBA, sp image
 func drawYCbCr(dst *image.RGBA, r image.Rectangle, src *image.YCbCr, sp image.Point) {
 	// An image.YCbCr 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
-	)
+	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
@@ -405,9 +403,9 @@ func drawYCbCr(dst *image.RGBA, r image.Rectangle, src *image.YCbCr, sp image.Po
 }
 
 func drawGlyphOver(dst *image.RGBA, r image.Rectangle, src *image.Uniform, mask *image.Alpha, mp image.Point) {
-	i0 := (r.Min.Y-dst.Rect.Min.Y)*dst.Stride + (r.Min.X-dst.Rect.Min.X)*4
+	i0 := dst.PixOffset(r.Min.X, r.Min.Y)
 	i1 := i0 + r.Dx()*4
-	mi0 := (mp.Y-mask.Rect.Min.Y)*mask.Stride + mp.X - mask.Rect.Min.X
+	mi0 := mask.PixOffset(mp.X, mp.Y)
 	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 {
@@ -451,7 +449,7 @@ func drawRGBA(dst *image.RGBA, r image.Rectangle, src image.Image, sp image.Poin
 	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
+	i0 := dst.PixOffset(x0, y0)
 	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 {

src/pkg/image/tiff/reader.go

@@ -223,8 +223,8 @@ func (d *decoder) decode(dst image.Image, ymin, ymax int) error {
 		}
 	case mRGB:
 		img := dst.(*image.RGBA)
-		min := (ymin-img.Rect.Min.Y)*img.Stride - img.Rect.Min.X*4
-		max := (ymax-img.Rect.Min.Y)*img.Stride - img.Rect.Min.X*4
+		min := img.PixOffset(0, ymin)
+		max := img.PixOffset(0, ymax)
 		var off int
 		for i := min; i < max; i += 4 {
 			img.Pix[i+0] = d.buf[off+0]
@@ -235,16 +235,16 @@ func (d *decoder) decode(dst image.Image, ymin, ymax int) error {
 		}
 	case mNRGBA:
 		img := dst.(*image.NRGBA)
-		min := (ymin-img.Rect.Min.Y)*img.Stride - img.Rect.Min.X*4
-		max := (ymax-img.Rect.Min.Y)*img.Stride - img.Rect.Min.X*4
+		min := img.PixOffset(0, ymin)
+		max := img.PixOffset(0, ymax)
 		if len(d.buf) != max-min {
 			return FormatError("short data strip")
 		}
 		copy(img.Pix[min:max], d.buf)
 	case mRGBA:
 		img := dst.(*image.RGBA)
-		min := (ymin-img.Rect.Min.Y)*img.Stride - img.Rect.Min.X*4
-		max := (ymax-img.Rect.Min.Y)*img.Stride - img.Rect.Min.X*4
+		min := img.PixOffset(0, ymin)
+		max := img.PixOffset(0, ymax)
 		if len(d.buf) != max-min {
 			return FormatError("short data strip")
 		}

src/pkg/image/ycbcr.go

@@ -49,28 +49,32 @@ func (p *YCbCr) At(x, y int) color.Color {
 	if !(Point{x, y}.In(p.Rect)) {
 		return color.YCbCr{}
 	}
+	yi := p.YOffset(x, y)
+	ci := p.COffset(x, y)
+	return color.YCbCr{
+		p.Y[yi],
+		p.Cb[ci],
+		p.Cr[ci],
+	}
+}
+
+// YOffset returns the index of the first element of Y that corresponds to
+// the pixel at (x, y).
+func (p *YCbCr) YOffset(x, y int) int {
+	return y*p.YStride + x
+}
+
+// COffset returns the index of the first element of Cb or Cr that corresponds
+// to the pixel at (x, y).
+func (p *YCbCr) COffset(x, y int) int {
 	switch p.SubsampleRatio {
 	case YCbCrSubsampleRatio422:
-		i := x / 2
-		return color.YCbCr{
-			p.Y[y*p.YStride+x],
-			p.Cb[y*p.CStride+i],
-			p.Cr[y*p.CStride+i],
-		}
+		return y*p.CStride + (x / 2)
 	case YCbCrSubsampleRatio420:
-		i, j := x/2, y/2
-		return color.YCbCr{
-			p.Y[y*p.YStride+x],
-			p.Cb[j*p.CStride+i],
-			p.Cr[j*p.CStride+i],
-		}
+		return (y/2)*p.CStride + (x / 2)
 	}
 	// Default to 4:4:4 subsampling.
-	return color.YCbCr{
-		p.Y[y*p.YStride+x],
-		p.Cb[y*p.CStride+x],
-		p.Cr[y*p.CStride+x],
-	}
+	return y*p.CStride + x
 }
 
 // SubImage returns an image representing the portion of the image p visible