git.lucas.co / go_mono
git clone https://git.lucas.co/go_mono.git

commit67c770d218f41829dd283ff7993b04cbce136e84
parent65a798f031
authorNigel Tao <nigeltao@golang.org>
date2015-04-08 20:47
draw: substitute Src for Over when the source image is completely opaque.

The two Ops are equivalent, but Src is faster.

Change-Id: I2c73a13755047c224c71fb5af786875f02681de9
Reviewed-on: https://go-review.googlesource.com/8640
Reviewed-by: Rob Pike <r@golang.org>

 draw/gen.go   |   54 +-
 draw/impl.go  | 4906 ++++++++++++++-------------------------------------------
 draw/scale.go |    7 +
 3 files changed, 1234 insertions(+), 3733 deletions(-)

diff --git a/draw/gen.go b/draw/gen.go
index 658270e..1e335e0 100644
--- a/draw/gen.go
+++ b/draw/gen.go
@@ -55,6 +55,7 @@ var (
 	// should be the fallback pair ("Image", "image.Image").
 	//
 	// TODO: add *image.CMYK src type after Go 1.5 is released.
+	// An *image.CMYK is also alwaysOpaque.
 	dsTypes = []struct{ dType, sType string }{
 		{"*image.RGBA", "*image.Gray"},
 		{"*image.RGBA", "*image.NRGBA"},
@@ -72,6 +73,13 @@ var (
 		"440",
 	}
 	ops = []string{"Over", "Src"}
+	// alwaysOpaque are those image.Image implementations that are always
+	// opaque. For these types, Over is equivalent to the faster Src, in the
+	// absence of a source mask.
+	alwaysOpaque = map[string]bool{
+		"*image.Gray":  true,
+		"*image.YCbCr": true,
+	}
 )
 
 func init() {
@@ -104,6 +112,9 @@ func gen(w *bytes.Buffer, receiver string, codes ...string) {
 	for _, code := range codes {
 		for _, t := range dsTypes {
 			for _, op := range ops {
+				if op == "Over" && alwaysOpaque[t.sType] {
+					continue
+				}
 				expn(w, code, &data{
 					dType:    t.dType,
 					sType:    t.sType,
@@ -132,6 +143,9 @@ func genKernel(w *bytes.Buffer) {
 	}
 	for _, t := range dsTypes {
 		for _, op := range ops {
+			if op == "Over" && alwaysOpaque[t.sType] {
+				continue
+			}
 			expn(w, codeKernelTransformLeaf, &data{
 				dType: t.dType,
 				sType: t.sType,
@@ -585,7 +599,7 @@ func expnDollar(prefix, dollar, suffix string, d *data) string {
 
 func expnSwitch(op, dType string, expandBoth bool, template string) string {
 	if op == "" && dType != "anyDType" {
-		lines := []string{"switch opts.op() {"}
+		lines := []string{"switch op {"}
 		for _, op = range ops {
 			lines = append(lines,
 				fmt.Sprintf("case %s:", op),
@@ -607,6 +621,14 @@ func expnSwitch(op, dType string, expandBoth bool, template string) string {
 		fallback, values = "image.Image", sTypesForDType[dType]
 	}
 	for _, v := range values {
+		if dType != "" {
+			// v is the sType. Skip those always-opaque sTypes, where Over is
+			// equivalent to Src.
+			if op == "Over" && alwaysOpaque[v] {
+				continue
+			}
+		}
+
 		if v == fallback {
 			lines = append(lines, "default:")
 		} else {
@@ -745,18 +767,22 @@ const (
 			if adr.Empty() || sr.Empty() {
 				return
 			}
+			op := opts.op()
+			if op == Over && opaque(src) { // TODO: also check that opts.SrcMask == nil.
+				op = Src
+			}
 			// sr is the source pixels. If it extends beyond the src bounds,
 			// we cannot use the type-specific fast paths, as they access
 			// the Pix fields directly without bounds checking.
 			if !sr.In(src.Bounds()) {
-				switch opts.op() {
+				switch op {
 				case Over:
 					z.scale_Image_Image_Over(dst, dr, adr, src, sr)
 				case Src:
 					z.scale_Image_Image_Src(dst, dr, adr, src, sr)
 				}
 			} else if _, ok := src.(*image.Uniform); ok {
-				Draw(dst, dr, src, src.Bounds().Min, opts.op())
+				Draw(dst, dr, src, src.Bounds().Min, op)
 			} else {
 				$switch z.scale_$dTypeRN_$sTypeRN$sratio_$op(dst, dr, adr, src, sr)
 			}
@@ -769,6 +795,10 @@ const (
 			if adr.Empty() || sr.Empty() {
 				return
 			}
+			op := opts.op()
+			if op == Over && opaque(src) { // TODO: also check that opts.SrcMask == nil.
+				op = Src
+			}
 			d2s := invert(s2d)
 			// bias is a translation of the mapping from dst co-ordinates to
 			// src co-ordinates such that the latter temporarily have
@@ -788,14 +818,14 @@ const (
 			// we cannot use the type-specific fast paths, as they access
 			// the Pix fields directly without bounds checking.
 			if !sr.In(src.Bounds()) {
-				switch opts.op() {
+				switch op {
 				case Over:
 					z.transform_Image_Image_Over(dst, dr, adr, &d2s, src, sr, bias)
 				case Src:
 					z.transform_Image_Image_Src(dst, dr, adr, &d2s, src, sr, bias)
 				}
 			} else if u, ok := src.(*image.Uniform); ok {
-				transform_Uniform(dst, dr, adr, &d2s, u, sr, bias, opts.op())
+				transform_Uniform(dst, dr, adr, &d2s, u, sr, bias, op)
 			} else {
 				$switch z.transform_$dTypeRN_$sTypeRN$sratio_$op(dst, dr, adr, &d2s, src, sr, bias)
 			}
@@ -961,9 +991,13 @@ const (
 			if adr.Empty() || sr.Empty() {
 				return
 			}
+			op := opts.op()
+			if op == Over && opaque(src) { // TODO: also check that opts.SrcMask == nil.
+				op = Src
+			}
 
 			if _, ok := src.(*image.Uniform); ok && sr.In(src.Bounds()) {
-				Draw(dst, dr, src, src.Bounds().Min, opts.op())
+				Draw(dst, dr, src, src.Bounds().Min, op)
 				return
 			}
 
@@ -998,6 +1032,10 @@ const (
 			if adr.Empty() || sr.Empty() {
 				return
 			}
+			op := opts.op()
+			if op == Over && opaque(src) { // TODO: also check that opts.SrcMask == nil.
+				op = Src
+			}
 			d2s := invert(s2d)
 			// bias is a translation of the mapping from dst co-ordinates to
 			// src co-ordinates such that the latter temporarily have
@@ -1015,7 +1053,7 @@ const (
 			adr = adr.Sub(dr.Min)
 
 			if u, ok := src.(*image.Uniform); ok && sr.In(src.Bounds()) {
-				transform_Uniform(dst, dr, adr, &d2s, u, sr, bias, opts.op())
+				transform_Uniform(dst, dr, adr, &d2s, u, sr, bias, op)
 				return
 			}
 
@@ -1032,7 +1070,7 @@ const (
 			// we cannot use the type-specific fast paths, as they access
 			// the Pix fields directly without bounds checking.
 			if !sr.In(src.Bounds()) {
-				switch opts.op() {
+				switch op {
 				case Over:
 					q.transform_Image_Image_Over(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
 				case Src:
diff --git a/draw/impl.go b/draw/impl.go
index efe65d2..8fed0bd 100644
--- a/draw/impl.go
+++ b/draw/impl.go
@@ -16,43 +16,32 @@ func (z nnInterpolator) Scale(dst Image, dr image.Rectangle, src image.Image, sr
 	if adr.Empty() || sr.Empty() {
 		return
 	}
+	op := opts.op()
+	if op == Over && opaque(src) { // TODO: also check that opts.SrcMask == nil.
+		op = Src
+	}
 	// sr is the source pixels. If it extends beyond the src bounds,
 	// we cannot use the type-specific fast paths, as they access
 	// the Pix fields directly without bounds checking.
 	if !sr.In(src.Bounds()) {
-		switch opts.op() {
+		switch op {
 		case Over:
 			z.scale_Image_Image_Over(dst, dr, adr, src, sr)
 		case Src:
 			z.scale_Image_Image_Src(dst, dr, adr, src, sr)
 		}
 	} else if _, ok := src.(*image.Uniform); ok {
-		Draw(dst, dr, src, src.Bounds().Min, opts.op())
+		Draw(dst, dr, src, src.Bounds().Min, op)
 	} else {
-		switch opts.op() {
+		switch op {
 		case Over:
 			switch dst := dst.(type) {
 			case *image.RGBA:
 				switch src := src.(type) {
-				case *image.Gray:
-					z.scale_RGBA_Gray_Over(dst, dr, adr, src, sr)
 				case *image.NRGBA:
 					z.scale_RGBA_NRGBA_Over(dst, dr, adr, src, sr)
 				case *image.RGBA:
 					z.scale_RGBA_RGBA_Over(dst, dr, adr, src, sr)
-				case *image.YCbCr:
-					switch src.SubsampleRatio {
-					default:
-						z.scale_RGBA_Image_Over(dst, dr, adr, src, sr)
-					case image.YCbCrSubsampleRatio444:
-						z.scale_RGBA_YCbCr444_Over(dst, dr, adr, src, sr)
-					case image.YCbCrSubsampleRatio422:
-						z.scale_RGBA_YCbCr422_Over(dst, dr, adr, src, sr)
-					case image.YCbCrSubsampleRatio420:
-						z.scale_RGBA_YCbCr420_Over(dst, dr, adr, src, sr)
-					case image.YCbCrSubsampleRatio440:
-						z.scale_RGBA_YCbCr440_Over(dst, dr, adr, src, sr)
-					}
 				default:
 					z.scale_RGBA_Image_Over(dst, dr, adr, src, sr)
 				}
@@ -105,6 +94,10 @@ func (z nnInterpolator) Transform(dst Image, s2d *f64.Aff3, src image.Image, sr
 	if adr.Empty() || sr.Empty() {
 		return
 	}
+	op := opts.op()
+	if op == Over && opaque(src) { // TODO: also check that opts.SrcMask == nil.
+		op = Src
+	}
 	d2s := invert(s2d)
 	// bias is a translation of the mapping from dst co-ordinates to
 	// src co-ordinates such that the latter temporarily have
@@ -124,39 +117,24 @@ func (z nnInterpolator) Transform(dst Image, s2d *f64.Aff3, src image.Image, sr
 	// we cannot use the type-specific fast paths, as they access
 	// the Pix fields directly without bounds checking.
 	if !sr.In(src.Bounds()) {
-		switch opts.op() {
+		switch op {
 		case Over:
 			z.transform_Image_Image_Over(dst, dr, adr, &d2s, src, sr, bias)
 		case Src:
 			z.transform_Image_Image_Src(dst, dr, adr, &d2s, src, sr, bias)
 		}
 	} else if u, ok := src.(*image.Uniform); ok {
-		transform_Uniform(dst, dr, adr, &d2s, u, sr, bias, opts.op())
+		transform_Uniform(dst, dr, adr, &d2s, u, sr, bias, op)
 	} else {
-		switch opts.op() {
+		switch op {
 		case Over:
 			switch dst := dst.(type) {
 			case *image.RGBA:
 				switch src := src.(type) {
-				case *image.Gray:
-					z.transform_RGBA_Gray_Over(dst, dr, adr, &d2s, src, sr, bias)
 				case *image.NRGBA:
 					z.transform_RGBA_NRGBA_Over(dst, dr, adr, &d2s, src, sr, bias)
 				case *image.RGBA:
 					z.transform_RGBA_RGBA_Over(dst, dr, adr, &d2s, src, sr, bias)
-				case *image.YCbCr:
-					switch src.SubsampleRatio {
-					default:
-						z.transform_RGBA_Image_Over(dst, dr, adr, &d2s, src, sr, bias)
-					case image.YCbCrSubsampleRatio444:
-						z.transform_RGBA_YCbCr444_Over(dst, dr, adr, &d2s, src, sr, bias)
-					case image.YCbCrSubsampleRatio422:
-						z.transform_RGBA_YCbCr422_Over(dst, dr, adr, &d2s, src, sr, bias)
-					case image.YCbCrSubsampleRatio420:
-						z.transform_RGBA_YCbCr420_Over(dst, dr, adr, &d2s, src, sr, bias)
-					case image.YCbCrSubsampleRatio440:
-						z.transform_RGBA_YCbCr440_Over(dst, dr, adr, &d2s, src, sr, bias)
-					}
 				default:
 					z.transform_RGBA_Image_Over(dst, dr, adr, &d2s, src, sr, bias)
 				}
@@ -202,27 +180,6 @@ func (z nnInterpolator) Transform(dst Image, s2d *f64.Aff3, src image.Image, sr
 	}
 }
 
-func (nnInterpolator) scale_RGBA_Gray_Over(dst *image.RGBA, dr, adr image.Rectangle, src *image.Gray, sr image.Rectangle) {
-	dw2 := uint64(dr.Dx()) * 2
-	dh2 := uint64(dr.Dy()) * 2
-	sw := uint64(sr.Dx())
-	sh := uint64(sr.Dy())
-	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
-		sy := (2*uint64(dy) + 1) * sh / dh2
-		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
-		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx, d = dx+1, d+4 {
-			sx := (2*uint64(dx) + 1) * sw / dw2
-			pi := (sr.Min.Y+int(sy)-src.Rect.Min.Y)*src.Stride + (sr.Min.X + int(sx) - src.Rect.Min.X)
-			pr := uint32(src.Pix[pi]) * 0x101
-			out := uint8(uint32(pr) >> 8)
-			dst.Pix[d+0] = out
-			dst.Pix[d+1] = out
-			dst.Pix[d+2] = out
-			dst.Pix[d+3] = 0xff
-		}
-	}
-}
-
 func (nnInterpolator) scale_RGBA_Gray_Src(dst *image.RGBA, dr, adr image.Rectangle, src *image.Gray, sr image.Rectangle) {
 	dw2 := uint64(dr.Dx()) * 2
 	dh2 := uint64(dr.Dy()) * 2
@@ -336,178 +293,6 @@ func (nnInterpolator) scale_RGBA_RGBA_Src(dst *image.RGBA, dr, adr image.Rectang
 	}
 }
 
-func (nnInterpolator) scale_RGBA_YCbCr444_Over(dst *image.RGBA, dr, adr image.Rectangle, src *image.YCbCr, sr image.Rectangle) {
-	dw2 := uint64(dr.Dx()) * 2
-	dh2 := uint64(dr.Dy()) * 2
-	sw := uint64(sr.Dx())
-	sh := uint64(sr.Dy())
-	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
-		sy := (2*uint64(dy) + 1) * sh / dh2
-		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
-		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx, d = dx+1, d+4 {
-			sx := (2*uint64(dx) + 1) * sw / dw2
-			pi := (sr.Min.Y+int(sy)-src.Rect.Min.Y)*src.YStride + (sr.Min.X + int(sx) - src.Rect.Min.X)
-			pj := (sr.Min.Y+int(sy)-src.Rect.Min.Y)*src.CStride + (sr.Min.X + int(sx) - src.Rect.Min.X)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			pyy1 := int(src.Y[pi])<<16 + 1<<15
-			pcb1 := int(src.Cb[pj]) - 128
-			pcr1 := int(src.Cr[pj]) - 128
-			pr := (pyy1 + 91881*pcr1) >> 8
-			pg := (pyy1 - 22554*pcb1 - 46802*pcr1) >> 8
-			pb := (pyy1 + 116130*pcb1) >> 8
-			if pr < 0 {
-				pr = 0
-			} else if pr > 0xffff {
-				pr = 0xffff
-			}
-			if pg < 0 {
-				pg = 0
-			} else if pg > 0xffff {
-				pg = 0xffff
-			}
-			if pb < 0 {
-				pb = 0
-			} else if pb > 0xffff {
-				pb = 0xffff
-			}
-			dst.Pix[d+0] = uint8(uint32(pr) >> 8)
-			dst.Pix[d+1] = uint8(uint32(pg) >> 8)
-			dst.Pix[d+2] = uint8(uint32(pb) >> 8)
-			dst.Pix[d+3] = 0xff
-		}
-	}
-}
-
-func (nnInterpolator) scale_RGBA_YCbCr422_Over(dst *image.RGBA, dr, adr image.Rectangle, src *image.YCbCr, sr image.Rectangle) {
-	dw2 := uint64(dr.Dx()) * 2
-	dh2 := uint64(dr.Dy()) * 2
-	sw := uint64(sr.Dx())
-	sh := uint64(sr.Dy())
-	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
-		sy := (2*uint64(dy) + 1) * sh / dh2
-		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
-		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx, d = dx+1, d+4 {
-			sx := (2*uint64(dx) + 1) * sw / dw2
-			pi := (sr.Min.Y+int(sy)-src.Rect.Min.Y)*src.YStride + (sr.Min.X + int(sx) - src.Rect.Min.X)
-			pj := (sr.Min.Y+int(sy)-src.Rect.Min.Y)*src.CStride + ((sr.Min.X+int(sx))/2 - src.Rect.Min.X/2)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			pyy1 := int(src.Y[pi])<<16 + 1<<15
-			pcb1 := int(src.Cb[pj]) - 128
-			pcr1 := int(src.Cr[pj]) - 128
-			pr := (pyy1 + 91881*pcr1) >> 8
-			pg := (pyy1 - 22554*pcb1 - 46802*pcr1) >> 8
-			pb := (pyy1 + 116130*pcb1) >> 8
-			if pr < 0 {
-				pr = 0
-			} else if pr > 0xffff {
-				pr = 0xffff
-			}
-			if pg < 0 {
-				pg = 0
-			} else if pg > 0xffff {
-				pg = 0xffff
-			}
-			if pb < 0 {
-				pb = 0
-			} else if pb > 0xffff {
-				pb = 0xffff
-			}
-			dst.Pix[d+0] = uint8(uint32(pr) >> 8)
-			dst.Pix[d+1] = uint8(uint32(pg) >> 8)
-			dst.Pix[d+2] = uint8(uint32(pb) >> 8)
-			dst.Pix[d+3] = 0xff
-		}
-	}
-}
-
-func (nnInterpolator) scale_RGBA_YCbCr420_Over(dst *image.RGBA, dr, adr image.Rectangle, src *image.YCbCr, sr image.Rectangle) {
-	dw2 := uint64(dr.Dx()) * 2
-	dh2 := uint64(dr.Dy()) * 2
-	sw := uint64(sr.Dx())
-	sh := uint64(sr.Dy())
-	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
-		sy := (2*uint64(dy) + 1) * sh / dh2
-		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
-		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx, d = dx+1, d+4 {
-			sx := (2*uint64(dx) + 1) * sw / dw2
-			pi := (sr.Min.Y+int(sy)-src.Rect.Min.Y)*src.YStride + (sr.Min.X + int(sx) - src.Rect.Min.X)
-			pj := ((sr.Min.Y+int(sy))/2-src.Rect.Min.Y/2)*src.CStride + ((sr.Min.X+int(sx))/2 - src.Rect.Min.X/2)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			pyy1 := int(src.Y[pi])<<16 + 1<<15
-			pcb1 := int(src.Cb[pj]) - 128
-			pcr1 := int(src.Cr[pj]) - 128
-			pr := (pyy1 + 91881*pcr1) >> 8
-			pg := (pyy1 - 22554*pcb1 - 46802*pcr1) >> 8
-			pb := (pyy1 + 116130*pcb1) >> 8
-			if pr < 0 {
-				pr = 0
-			} else if pr > 0xffff {
-				pr = 0xffff
-			}
-			if pg < 0 {
-				pg = 0
-			} else if pg > 0xffff {
-				pg = 0xffff
-			}
-			if pb < 0 {
-				pb = 0
-			} else if pb > 0xffff {
-				pb = 0xffff
-			}
-			dst.Pix[d+0] = uint8(uint32(pr) >> 8)
-			dst.Pix[d+1] = uint8(uint32(pg) >> 8)
-			dst.Pix[d+2] = uint8(uint32(pb) >> 8)
-			dst.Pix[d+3] = 0xff
-		}
-	}
-}
-
-func (nnInterpolator) scale_RGBA_YCbCr440_Over(dst *image.RGBA, dr, adr image.Rectangle, src *image.YCbCr, sr image.Rectangle) {
-	dw2 := uint64(dr.Dx()) * 2
-	dh2 := uint64(dr.Dy()) * 2
-	sw := uint64(sr.Dx())
-	sh := uint64(sr.Dy())
-	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
-		sy := (2*uint64(dy) + 1) * sh / dh2
-		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
-		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx, d = dx+1, d+4 {
-			sx := (2*uint64(dx) + 1) * sw / dw2
-			pi := (sr.Min.Y+int(sy)-src.Rect.Min.Y)*src.YStride + (sr.Min.X + int(sx) - src.Rect.Min.X)
-			pj := ((sr.Min.Y+int(sy))/2-src.Rect.Min.Y/2)*src.CStride + (sr.Min.X + int(sx) - src.Rect.Min.X)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			pyy1 := int(src.Y[pi])<<16 + 1<<15
-			pcb1 := int(src.Cb[pj]) - 128
-			pcr1 := int(src.Cr[pj]) - 128
-			pr := (pyy1 + 91881*pcr1) >> 8
-			pg := (pyy1 - 22554*pcb1 - 46802*pcr1) >> 8
-			pb := (pyy1 + 116130*pcb1) >> 8
-			if pr < 0 {
-				pr = 0
-			} else if pr > 0xffff {
-				pr = 0xffff
-			}
-			if pg < 0 {
-				pg = 0
-			} else if pg > 0xffff {
-				pg = 0xffff
-			}
-			if pb < 0 {
-				pb = 0
-			} else if pb > 0xffff {
-				pb = 0xffff
-			}
-			dst.Pix[d+0] = uint8(uint32(pr) >> 8)
-			dst.Pix[d+1] = uint8(uint32(pg) >> 8)
-			dst.Pix[d+2] = uint8(uint32(pb) >> 8)
-			dst.Pix[d+3] = 0xff
-		}
-	}
-}
-
 func (nnInterpolator) scale_RGBA_YCbCr444_Src(dst *image.RGBA, dr, adr image.Rectangle, src *image.YCbCr, sr image.Rectangle) {
 	dw2 := uint64(dr.Dx()) * 2
 	dh2 := uint64(dr.Dy()) * 2
@@ -760,28 +545,6 @@ func (nnInterpolator) scale_Image_Image_Src(dst Image, dr, adr image.Rectangle,
 	}
 }
 
-func (nnInterpolator) transform_RGBA_Gray_Over(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.Gray, sr image.Rectangle, bias image.Point) {
-	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
-		dyf := float64(dr.Min.Y+int(dy)) + 0.5
-		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
-		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx, d = dx+1, d+4 {
-			dxf := float64(dr.Min.X+int(dx)) + 0.5
-			sx0 := int(d2s[0]*dxf+d2s[1]*dyf+d2s[2]) + bias.X
-			sy0 := int(d2s[3]*dxf+d2s[4]*dyf+d2s[5]) + bias.Y
-			if !(image.Point{sx0, sy0}).In(sr) {
-				continue
-			}
-			pi := (sy0-src.Rect.Min.Y)*src.Stride + (sx0 - src.Rect.Min.X)
-			pr := uint32(src.Pix[pi]) * 0x101
-			out := uint8(uint32(pr) >> 8)
-			dst.Pix[d+0] = out
-			dst.Pix[d+1] = out
-			dst.Pix[d+2] = out
-			dst.Pix[d+3] = 0xff
-		}
-	}
-}
-
 func (nnInterpolator) transform_RGBA_Gray_Src(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.Gray, sr image.Rectangle, bias image.Point) {
 	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
 		dyf := float64(dr.Min.Y+int(dy)) + 0.5
@@ -900,7 +663,7 @@ func (nnInterpolator) transform_RGBA_RGBA_Src(dst *image.RGBA, dr, adr image.Rec
 	}
 }
 
-func (nnInterpolator) transform_RGBA_YCbCr444_Over(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.YCbCr, sr image.Rectangle, bias image.Point) {
+func (nnInterpolator) transform_RGBA_YCbCr444_Src(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.YCbCr, sr image.Rectangle, bias image.Point) {
 	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
 		dyf := float64(dr.Min.Y+int(dy)) + 0.5
 		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
@@ -944,7 +707,7 @@ func (nnInterpolator) transform_RGBA_YCbCr444_Over(dst *image.RGBA, dr, adr imag
 	}
 }
 
-func (nnInterpolator) transform_RGBA_YCbCr422_Over(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.YCbCr, sr image.Rectangle, bias image.Point) {
+func (nnInterpolator) transform_RGBA_YCbCr422_Src(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.YCbCr, sr image.Rectangle, bias image.Point) {
 	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
 		dyf := float64(dr.Min.Y+int(dy)) + 0.5
 		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
@@ -988,7 +751,7 @@ func (nnInterpolator) transform_RGBA_YCbCr422_Over(dst *image.RGBA, dr, adr imag
 	}
 }
 
-func (nnInterpolator) transform_RGBA_YCbCr420_Over(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.YCbCr, sr image.Rectangle, bias image.Point) {
+func (nnInterpolator) transform_RGBA_YCbCr420_Src(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.YCbCr, sr image.Rectangle, bias image.Point) {
 	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
 		dyf := float64(dr.Min.Y+int(dy)) + 0.5
 		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
@@ -1032,7 +795,7 @@ func (nnInterpolator) transform_RGBA_YCbCr420_Over(dst *image.RGBA, dr, adr imag
 	}
 }
 
-func (nnInterpolator) transform_RGBA_YCbCr440_Over(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.YCbCr, sr image.Rectangle, bias image.Point) {
+func (nnInterpolator) transform_RGBA_YCbCr440_Src(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.YCbCr, sr image.Rectangle, bias image.Point) {
 	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
 		dyf := float64(dr.Min.Y+int(dy)) + 0.5
 		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
@@ -1076,7 +839,7 @@ func (nnInterpolator) transform_RGBA_YCbCr440_Over(dst *image.RGBA, dr, adr imag
 	}
 }
 
-func (nnInterpolator) transform_RGBA_YCbCr444_Src(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.YCbCr, sr image.Rectangle, bias image.Point) {
+func (nnInterpolator) transform_RGBA_Image_Over(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src image.Image, sr image.Rectangle, bias image.Point) {
 	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
 		dyf := float64(dr.Min.Y+int(dy)) + 0.5
 		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
@@ -1087,190 +850,14 @@ func (nnInterpolator) transform_RGBA_YCbCr444_Src(dst *image.RGBA, dr, adr image
 			if !(image.Point{sx0, sy0}).In(sr) {
 				continue
 			}
-			pi := (sy0-src.Rect.Min.Y)*src.YStride + (sx0 - src.Rect.Min.X)
-			pj := (sy0-src.Rect.Min.Y)*src.CStride + (sx0 - src.Rect.Min.X)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			pyy1 := int(src.Y[pi])<<16 + 1<<15
-			pcb1 := int(src.Cb[pj]) - 128
-			pcr1 := int(src.Cr[pj]) - 128
-			pr := (pyy1 + 91881*pcr1) >> 8
-			pg := (pyy1 - 22554*pcb1 - 46802*pcr1) >> 8
-			pb := (pyy1 + 116130*pcb1) >> 8
-			if pr < 0 {
-				pr = 0
-			} else if pr > 0xffff {
-				pr = 0xffff
-			}
-			if pg < 0 {
-				pg = 0
-			} else if pg > 0xffff {
-				pg = 0xffff
-			}
-			if pb < 0 {
-				pb = 0
-			} else if pb > 0xffff {
-				pb = 0xffff
-			}
-			dst.Pix[d+0] = uint8(uint32(pr) >> 8)
-			dst.Pix[d+1] = uint8(uint32(pg) >> 8)
-			dst.Pix[d+2] = uint8(uint32(pb) >> 8)
-			dst.Pix[d+3] = 0xff
-		}
-	}
-}
-
-func (nnInterpolator) transform_RGBA_YCbCr422_Src(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.YCbCr, sr image.Rectangle, bias image.Point) {
-	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
-		dyf := float64(dr.Min.Y+int(dy)) + 0.5
-		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
-		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx, d = dx+1, d+4 {
-			dxf := float64(dr.Min.X+int(dx)) + 0.5
-			sx0 := int(d2s[0]*dxf+d2s[1]*dyf+d2s[2]) + bias.X
-			sy0 := int(d2s[3]*dxf+d2s[4]*dyf+d2s[5]) + bias.Y
-			if !(image.Point{sx0, sy0}).In(sr) {
-				continue
-			}
-			pi := (sy0-src.Rect.Min.Y)*src.YStride + (sx0 - src.Rect.Min.X)
-			pj := (sy0-src.Rect.Min.Y)*src.CStride + ((sx0)/2 - src.Rect.Min.X/2)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			pyy1 := int(src.Y[pi])<<16 + 1<<15
-			pcb1 := int(src.Cb[pj]) - 128
-			pcr1 := int(src.Cr[pj]) - 128
-			pr := (pyy1 + 91881*pcr1) >> 8
-			pg := (pyy1 - 22554*pcb1 - 46802*pcr1) >> 8
-			pb := (pyy1 + 116130*pcb1) >> 8
-			if pr < 0 {
-				pr = 0
-			} else if pr > 0xffff {
-				pr = 0xffff
-			}
-			if pg < 0 {
-				pg = 0
-			} else if pg > 0xffff {
-				pg = 0xffff
-			}
-			if pb < 0 {
-				pb = 0
-			} else if pb > 0xffff {
-				pb = 0xffff
-			}
-			dst.Pix[d+0] = uint8(uint32(pr) >> 8)
-			dst.Pix[d+1] = uint8(uint32(pg) >> 8)
-			dst.Pix[d+2] = uint8(uint32(pb) >> 8)
-			dst.Pix[d+3] = 0xff
-		}
-	}
-}
-
-func (nnInterpolator) transform_RGBA_YCbCr420_Src(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.YCbCr, sr image.Rectangle, bias image.Point) {
-	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
-		dyf := float64(dr.Min.Y+int(dy)) + 0.5
-		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
-		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx, d = dx+1, d+4 {
-			dxf := float64(dr.Min.X+int(dx)) + 0.5
-			sx0 := int(d2s[0]*dxf+d2s[1]*dyf+d2s[2]) + bias.X
-			sy0 := int(d2s[3]*dxf+d2s[4]*dyf+d2s[5]) + bias.Y
-			if !(image.Point{sx0, sy0}).In(sr) {
-				continue
-			}
-			pi := (sy0-src.Rect.Min.Y)*src.YStride + (sx0 - src.Rect.Min.X)
-			pj := ((sy0)/2-src.Rect.Min.Y/2)*src.CStride + ((sx0)/2 - src.Rect.Min.X/2)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			pyy1 := int(src.Y[pi])<<16 + 1<<15
-			pcb1 := int(src.Cb[pj]) - 128
-			pcr1 := int(src.Cr[pj]) - 128
-			pr := (pyy1 + 91881*pcr1) >> 8
-			pg := (pyy1 - 22554*pcb1 - 46802*pcr1) >> 8
-			pb := (pyy1 + 116130*pcb1) >> 8
-			if pr < 0 {
-				pr = 0
-			} else if pr > 0xffff {
-				pr = 0xffff
-			}
-			if pg < 0 {
-				pg = 0
-			} else if pg > 0xffff {
-				pg = 0xffff
-			}
-			if pb < 0 {
-				pb = 0
-			} else if pb > 0xffff {
-				pb = 0xffff
-			}
-			dst.Pix[d+0] = uint8(uint32(pr) >> 8)
-			dst.Pix[d+1] = uint8(uint32(pg) >> 8)
-			dst.Pix[d+2] = uint8(uint32(pb) >> 8)
-			dst.Pix[d+3] = 0xff
-		}
-	}
-}
-
-func (nnInterpolator) transform_RGBA_YCbCr440_Src(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.YCbCr, sr image.Rectangle, bias image.Point) {
-	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
-		dyf := float64(dr.Min.Y+int(dy)) + 0.5
-		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
-		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx, d = dx+1, d+4 {
-			dxf := float64(dr.Min.X+int(dx)) + 0.5
-			sx0 := int(d2s[0]*dxf+d2s[1]*dyf+d2s[2]) + bias.X
-			sy0 := int(d2s[3]*dxf+d2s[4]*dyf+d2s[5]) + bias.Y
-			if !(image.Point{sx0, sy0}).In(sr) {
-				continue
-			}
-			pi := (sy0-src.Rect.Min.Y)*src.YStride + (sx0 - src.Rect.Min.X)
-			pj := ((sy0)/2-src.Rect.Min.Y/2)*src.CStride + (sx0 - src.Rect.Min.X)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			pyy1 := int(src.Y[pi])<<16 + 1<<15
-			pcb1 := int(src.Cb[pj]) - 128
-			pcr1 := int(src.Cr[pj]) - 128
-			pr := (pyy1 + 91881*pcr1) >> 8
-			pg := (pyy1 - 22554*pcb1 - 46802*pcr1) >> 8
-			pb := (pyy1 + 116130*pcb1) >> 8
-			if pr < 0 {
-				pr = 0
-			} else if pr > 0xffff {
-				pr = 0xffff
-			}
-			if pg < 0 {
-				pg = 0
-			} else if pg > 0xffff {
-				pg = 0xffff
-			}
-			if pb < 0 {
-				pb = 0
-			} else if pb > 0xffff {
-				pb = 0xffff
-			}
-			dst.Pix[d+0] = uint8(uint32(pr) >> 8)
-			dst.Pix[d+1] = uint8(uint32(pg) >> 8)
-			dst.Pix[d+2] = uint8(uint32(pb) >> 8)
-			dst.Pix[d+3] = 0xff
-		}
-	}
-}
-
-func (nnInterpolator) transform_RGBA_Image_Over(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src image.Image, sr image.Rectangle, bias image.Point) {
-	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
-		dyf := float64(dr.Min.Y+int(dy)) + 0.5
-		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
-		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx, d = dx+1, d+4 {
-			dxf := float64(dr.Min.X+int(dx)) + 0.5
-			sx0 := int(d2s[0]*dxf+d2s[1]*dyf+d2s[2]) + bias.X
-			sy0 := int(d2s[3]*dxf+d2s[4]*dyf+d2s[5]) + bias.Y
-			if !(image.Point{sx0, sy0}).In(sr) {
-				continue
-			}
-			pr, pg, pb, pa := src.At(sx0, sy0).RGBA()
-			dst.Pix[d+0] = uint8(uint32(pr) >> 8)
-			dst.Pix[d+1] = uint8(uint32(pg) >> 8)
-			dst.Pix[d+2] = uint8(uint32(pb) >> 8)
-			dst.Pix[d+3] = uint8(uint32(pa) >> 8)
-		}
-	}
-}
+			pr, pg, pb, pa := src.At(sx0, sy0).RGBA()
+			dst.Pix[d+0] = uint8(uint32(pr) >> 8)
+			dst.Pix[d+1] = uint8(uint32(pg) >> 8)
+			dst.Pix[d+2] = uint8(uint32(pb) >> 8)
+			dst.Pix[d+3] = uint8(uint32(pa) >> 8)
+		}
+	}
+}
 
 func (nnInterpolator) transform_RGBA_Image_Src(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src image.Image, sr image.Rectangle, bias image.Point) {
 	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
@@ -1342,43 +929,32 @@ func (z ablInterpolator) Scale(dst Image, dr image.Rectangle, src image.Image, s
 	if adr.Empty() || sr.Empty() {
 		return
 	}
+	op := opts.op()
+	if op == Over && opaque(src) { // TODO: also check that opts.SrcMask == nil.
+		op = Src
+	}
 	// sr is the source pixels. If it extends beyond the src bounds,
 	// we cannot use the type-specific fast paths, as they access
 	// the Pix fields directly without bounds checking.
 	if !sr.In(src.Bounds()) {
-		switch opts.op() {
+		switch op {
 		case Over:
 			z.scale_Image_Image_Over(dst, dr, adr, src, sr)
 		case Src:
 			z.scale_Image_Image_Src(dst, dr, adr, src, sr)
 		}
 	} else if _, ok := src.(*image.Uniform); ok {
-		Draw(dst, dr, src, src.Bounds().Min, opts.op())
+		Draw(dst, dr, src, src.Bounds().Min, op)
 	} else {
-		switch opts.op() {
+		switch op {
 		case Over:
 			switch dst := dst.(type) {
 			case *image.RGBA:
 				switch src := src.(type) {
-				case *image.Gray:
-					z.scale_RGBA_Gray_Over(dst, dr, adr, src, sr)
 				case *image.NRGBA:
 					z.scale_RGBA_NRGBA_Over(dst, dr, adr, src, sr)
 				case *image.RGBA:
 					z.scale_RGBA_RGBA_Over(dst, dr, adr, src, sr)
-				case *image.YCbCr:
-					switch src.SubsampleRatio {
-					default:
-						z.scale_RGBA_Image_Over(dst, dr, adr, src, sr)
-					case image.YCbCrSubsampleRatio444:
-						z.scale_RGBA_YCbCr444_Over(dst, dr, adr, src, sr)
-					case image.YCbCrSubsampleRatio422:
-						z.scale_RGBA_YCbCr422_Over(dst, dr, adr, src, sr)
-					case image.YCbCrSubsampleRatio420:
-						z.scale_RGBA_YCbCr420_Over(dst, dr, adr, src, sr)
-					case image.YCbCrSubsampleRatio440:
-						z.scale_RGBA_YCbCr440_Over(dst, dr, adr, src, sr)
-					}
 				default:
 					z.scale_RGBA_Image_Over(dst, dr, adr, src, sr)
 				}
@@ -1431,6 +1007,10 @@ func (z ablInterpolator) Transform(dst Image, s2d *f64.Aff3, src image.Image, sr
 	if adr.Empty() || sr.Empty() {
 		return
 	}
+	op := opts.op()
+	if op == Over && opaque(src) { // TODO: also check that opts.SrcMask == nil.
+		op = Src
+	}
 	d2s := invert(s2d)
 	// bias is a translation of the mapping from dst co-ordinates to
 	// src co-ordinates such that the latter temporarily have
@@ -1450,39 +1030,24 @@ func (z ablInterpolator) Transform(dst Image, s2d *f64.Aff3, src image.Image, sr
 	// we cannot use the type-specific fast paths, as they access
 	// the Pix fields directly without bounds checking.
 	if !sr.In(src.Bounds()) {
-		switch opts.op() {
+		switch op {
 		case Over:
 			z.transform_Image_Image_Over(dst, dr, adr, &d2s, src, sr, bias)
 		case Src:
 			z.transform_Image_Image_Src(dst, dr, adr, &d2s, src, sr, bias)
 		}
 	} else if u, ok := src.(*image.Uniform); ok {
-		transform_Uniform(dst, dr, adr, &d2s, u, sr, bias, opts.op())
+		transform_Uniform(dst, dr, adr, &d2s, u, sr, bias, op)
 	} else {
-		switch opts.op() {
+		switch op {
 		case Over:
 			switch dst := dst.(type) {
 			case *image.RGBA:
 				switch src := src.(type) {
-				case *image.Gray:
-					z.transform_RGBA_Gray_Over(dst, dr, adr, &d2s, src, sr, bias)
 				case *image.NRGBA:
 					z.transform_RGBA_NRGBA_Over(dst, dr, adr, &d2s, src, sr, bias)
 				case *image.RGBA:
 					z.transform_RGBA_RGBA_Over(dst, dr, adr, &d2s, src, sr, bias)
-				case *image.YCbCr:
-					switch src.SubsampleRatio {
-					default:
-						z.transform_RGBA_Image_Over(dst, dr, adr, &d2s, src, sr, bias)
-					case image.YCbCrSubsampleRatio444:
-						z.transform_RGBA_YCbCr444_Over(dst, dr, adr, &d2s, src, sr, bias)
-					case image.YCbCrSubsampleRatio422:
-						z.transform_RGBA_YCbCr422_Over(dst, dr, adr, &d2s, src, sr, bias)
-					case image.YCbCrSubsampleRatio420:
-						z.transform_RGBA_YCbCr420_Over(dst, dr, adr, &d2s, src, sr, bias)
-					case image.YCbCrSubsampleRatio440:
-						z.transform_RGBA_YCbCr440_Over(dst, dr, adr, &d2s, src, sr, bias)
-					}
 				default:
 					z.transform_RGBA_Image_Over(dst, dr, adr, &d2s, src, sr, bias)
 				}
@@ -1528,69 +1093,6 @@ func (z ablInterpolator) Transform(dst Image, s2d *f64.Aff3, src image.Image, sr
 	}
 }
 
-func (ablInterpolator) scale_RGBA_Gray_Over(dst *image.RGBA, dr, adr image.Rectangle, src *image.Gray, sr image.Rectangle) {
-	sw := int32(sr.Dx())
-	sh := int32(sr.Dy())
-	yscale := float64(sh) / float64(dr.Dy())
-	xscale := float64(sw) / float64(dr.Dx())
-	swMinus1, shMinus1 := sw-1, sh-1
-
-	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
-		sy := (float64(dy)+0.5)*yscale - 0.5
-		// If sy < 0, we will clamp sy0 to 0 anyway, so it doesn't matter if
-		// we say int32(sy) instead of int32(math.Floor(sy)). Similarly for
-		// sx, below.
-		sy0 := int32(sy)
-		yFrac0 := sy - float64(sy0)
-		yFrac1 := 1 - yFrac0
-		sy1 := sy0 + 1
-		if sy < 0 {
-			sy0, sy1 = 0, 0
-			yFrac0, yFrac1 = 0, 1
-		} else if sy1 > shMinus1 {
-			sy0, sy1 = shMinus1, shMinus1
-			yFrac0, yFrac1 = 1, 0
-		}
-		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
-
-		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx, d = dx+1, d+4 {
-			sx := (float64(dx)+0.5)*xscale - 0.5
-			sx0 := int32(sx)
-			xFrac0 := sx - float64(sx0)
-			xFrac1 := 1 - xFrac0
-			sx1 := sx0 + 1
-			if sx < 0 {
-				sx0, sx1 = 0, 0
-				xFrac0, xFrac1 = 0, 1
-			} else if sx1 > swMinus1 {
-				sx0, sx1 = swMinus1, swMinus1
-				xFrac0, xFrac1 = 1, 0
-			}
-
-			s00i := (sr.Min.Y+int(sy0)-src.Rect.Min.Y)*src.Stride + (sr.Min.X + int(sx0) - src.Rect.Min.X)
-			s00ru := uint32(src.Pix[s00i]) * 0x101
-			s00r := float64(s00ru)
-			s10i := (sr.Min.Y+int(sy0)-src.Rect.Min.Y)*src.Stride + (sr.Min.X + int(sx1) - src.Rect.Min.X)
-			s10ru := uint32(src.Pix[s10i]) * 0x101
-			s10r := float64(s10ru)
-			s10r = xFrac1*s00r + xFrac0*s10r
-			s01i := (sr.Min.Y+int(sy1)-src.Rect.Min.Y)*src.Stride + (sr.Min.X + int(sx0) - src.Rect.Min.X)
-			s01ru := uint32(src.Pix[s01i]) * 0x101
-			s01r := float64(s01ru)
-			s11i := (sr.Min.Y+int(sy1)-src.Rect.Min.Y)*src.Stride + (sr.Min.X + int(sx1) - src.Rect.Min.X)
-			s11ru := uint32(src.Pix[s11i]) * 0x101
-			s11r := float64(s11ru)
-			s11r = xFrac1*s01r + xFrac0*s11r
-			s11r = yFrac1*s10r + yFrac0*s11r
-			out := uint8(uint32(s11r) >> 8)
-			dst.Pix[d+0] = out
-			dst.Pix[d+1] = out
-			dst.Pix[d+2] = out
-			dst.Pix[d+3] = 0xff
-		}
-	}
-}
-
 func (ablInterpolator) scale_RGBA_Gray_Src(dst *image.RGBA, dr, adr image.Rectangle, src *image.Gray, sr image.Rectangle) {
 	sw := int32(sr.Dx())
 	sh := int32(sr.Dy())
@@ -2034,7 +1536,7 @@ func (ablInterpolator) scale_RGBA_RGBA_Src(dst *image.RGBA, dr, adr image.Rectan
 	}
 }
 
-func (ablInterpolator) scale_RGBA_YCbCr444_Over(dst *image.RGBA, dr, adr image.Rectangle, src *image.YCbCr, sr image.Rectangle) {
+func (ablInterpolator) scale_RGBA_YCbCr444_Src(dst *image.RGBA, dr, adr image.Rectangle, src *image.YCbCr, sr image.Rectangle) {
 	sw := int32(sr.Dx())
 	sh := int32(sr.Dy())
 	yscale := float64(sh) / float64(dr.Dy())
@@ -2206,7 +1708,7 @@ func (ablInterpolator) scale_RGBA_YCbCr444_Over(dst *image.RGBA, dr, adr image.R
 	}
 }
 
-func (ablInterpolator) scale_RGBA_YCbCr422_Over(dst *image.RGBA, dr, adr image.Rectangle, src *image.YCbCr, sr image.Rectangle) {
+func (ablInterpolator) scale_RGBA_YCbCr422_Src(dst *image.RGBA, dr, adr image.Rectangle, src *image.YCbCr, sr image.Rectangle) {
 	sw := int32(sr.Dx())
 	sh := int32(sr.Dy())
 	yscale := float64(sh) / float64(dr.Dy())
@@ -2378,7 +1880,7 @@ func (ablInterpolator) scale_RGBA_YCbCr422_Over(dst *image.RGBA, dr, adr image.R
 	}
 }
 
-func (ablInterpolator) scale_RGBA_YCbCr420_Over(dst *image.RGBA, dr, adr image.Rectangle, src *image.YCbCr, sr image.Rectangle) {
+func (ablInterpolator) scale_RGBA_YCbCr420_Src(dst *image.RGBA, dr, adr image.Rectangle, src *image.YCbCr, sr image.Rectangle) {
 	sw := int32(sr.Dx())
 	sh := int32(sr.Dy())
 	yscale := float64(sh) / float64(dr.Dy())
@@ -2550,7 +2052,7 @@ func (ablInterpolator) scale_RGBA_YCbCr420_Over(dst *image.RGBA, dr, adr image.R
 	}
 }
 
-func (ablInterpolator) scale_RGBA_YCbCr440_Over(dst *image.RGBA, dr, adr image.Rectangle, src *image.YCbCr, sr image.Rectangle) {
+func (ablInterpolator) scale_RGBA_YCbCr440_Src(dst *image.RGBA, dr, adr image.Rectangle, src *image.YCbCr, sr image.Rectangle) {
 	sw := int32(sr.Dx())
 	sh := int32(sr.Dy())
 	yscale := float64(sh) / float64(dr.Dy())
@@ -2722,7 +2224,7 @@ func (ablInterpolator) scale_RGBA_YCbCr440_Over(dst *image.RGBA, dr, adr image.R
 	}
 }
 
-func (ablInterpolator) scale_RGBA_YCbCr444_Src(dst *image.RGBA, dr, adr image.Rectangle, src *image.YCbCr, sr image.Rectangle) {
+func (ablInterpolator) scale_RGBA_Image_Over(dst *image.RGBA, dr, adr image.Rectangle, src image.Image, sr image.Rectangle) {
 	sw := int32(sr.Dx())
 	sh := int32(sr.Dy())
 	yscale := float64(sh) / float64(dr.Dy())
@@ -2761,140 +2263,47 @@ func (ablInterpolator) scale_RGBA_YCbCr444_Src(dst *image.RGBA, dr, adr image.Re
 				xFrac0, xFrac1 = 1, 0
 			}
 
-			s00i := (sr.Min.Y+int(sy0)-src.Rect.Min.Y)*src.YStride + (sr.Min.X + int(sx0) - src.Rect.Min.X)
-			s00j := (sr.Min.Y+int(sy0)-src.Rect.Min.Y)*src.CStride + (sr.Min.X + int(sx0) - src.Rect.Min.X)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s00yy1 := int(src.Y[s00i])<<16 + 1<<15
-			s00cb1 := int(src.Cb[s00j]) - 128
-			s00cr1 := int(src.Cr[s00j]) - 128
-			s00ru := (s00yy1 + 91881*s00cr1) >> 8
-			s00gu := (s00yy1 - 22554*s00cb1 - 46802*s00cr1) >> 8
-			s00bu := (s00yy1 + 116130*s00cb1) >> 8
-			if s00ru < 0 {
-				s00ru = 0
-			} else if s00ru > 0xffff {
-				s00ru = 0xffff
-			}
-			if s00gu < 0 {
-				s00gu = 0
-			} else if s00gu > 0xffff {
-				s00gu = 0xffff
-			}
-			if s00bu < 0 {
-				s00bu = 0
-			} else if s00bu > 0xffff {
-				s00bu = 0xffff
-			}
-
+			s00ru, s00gu, s00bu, s00au := src.At(sr.Min.X+int(sx0), sr.Min.Y+int(sy0)).RGBA()
 			s00r := float64(s00ru)
 			s00g := float64(s00gu)
 			s00b := float64(s00bu)
-			s10i := (sr.Min.Y+int(sy0)-src.Rect.Min.Y)*src.YStride + (sr.Min.X + int(sx1) - src.Rect.Min.X)
-			s10j := (sr.Min.Y+int(sy0)-src.Rect.Min.Y)*src.CStride + (sr.Min.X + int(sx1) - src.Rect.Min.X)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s10yy1 := int(src.Y[s10i])<<16 + 1<<15
-			s10cb1 := int(src.Cb[s10j]) - 128
-			s10cr1 := int(src.Cr[s10j]) - 128
-			s10ru := (s10yy1 + 91881*s10cr1) >> 8
-			s10gu := (s10yy1 - 22554*s10cb1 - 46802*s10cr1) >> 8
-			s10bu := (s10yy1 + 116130*s10cb1) >> 8
-			if s10ru < 0 {
-				s10ru = 0
-			} else if s10ru > 0xffff {
-				s10ru = 0xffff
-			}
-			if s10gu < 0 {
-				s10gu = 0
-			} else if s10gu > 0xffff {
-				s10gu = 0xffff
-			}
-			if s10bu < 0 {
-				s10bu = 0
-			} else if s10bu > 0xffff {
-				s10bu = 0xffff
-			}
-
+			s00a := float64(s00au)
+			s10ru, s10gu, s10bu, s10au := src.At(sr.Min.X+int(sx1), sr.Min.Y+int(sy0)).RGBA()
 			s10r := float64(s10ru)
 			s10g := float64(s10gu)
 			s10b := float64(s10bu)
+			s10a := float64(s10au)
 			s10r = xFrac1*s00r + xFrac0*s10r
 			s10g = xFrac1*s00g + xFrac0*s10g
 			s10b = xFrac1*s00b + xFrac0*s10b
-			s01i := (sr.Min.Y+int(sy1)-src.Rect.Min.Y)*src.YStride + (sr.Min.X + int(sx0) - src.Rect.Min.X)
-			s01j := (sr.Min.Y+int(sy1)-src.Rect.Min.Y)*src.CStride + (sr.Min.X + int(sx0) - src.Rect.Min.X)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s01yy1 := int(src.Y[s01i])<<16 + 1<<15
-			s01cb1 := int(src.Cb[s01j]) - 128
-			s01cr1 := int(src.Cr[s01j]) - 128
-			s01ru := (s01yy1 + 91881*s01cr1) >> 8
-			s01gu := (s01yy1 - 22554*s01cb1 - 46802*s01cr1) >> 8
-			s01bu := (s01yy1 + 116130*s01cb1) >> 8
-			if s01ru < 0 {
-				s01ru = 0
-			} else if s01ru > 0xffff {
-				s01ru = 0xffff
-			}
-			if s01gu < 0 {
-				s01gu = 0
-			} else if s01gu > 0xffff {
-				s01gu = 0xffff
-			}
-			if s01bu < 0 {
-				s01bu = 0
-			} else if s01bu > 0xffff {
-				s01bu = 0xffff
-			}
-
+			s10a = xFrac1*s00a + xFrac0*s10a
+			s01ru, s01gu, s01bu, s01au := src.At(sr.Min.X+int(sx0), sr.Min.Y+int(sy1)).RGBA()
 			s01r := float64(s01ru)
 			s01g := float64(s01gu)
 			s01b := float64(s01bu)
-			s11i := (sr.Min.Y+int(sy1)-src.Rect.Min.Y)*src.YStride + (sr.Min.X + int(sx1) - src.Rect.Min.X)
-			s11j := (sr.Min.Y+int(sy1)-src.Rect.Min.Y)*src.CStride + (sr.Min.X + int(sx1) - src.Rect.Min.X)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s11yy1 := int(src.Y[s11i])<<16 + 1<<15
-			s11cb1 := int(src.Cb[s11j]) - 128
-			s11cr1 := int(src.Cr[s11j]) - 128
-			s11ru := (s11yy1 + 91881*s11cr1) >> 8
-			s11gu := (s11yy1 - 22554*s11cb1 - 46802*s11cr1) >> 8
-			s11bu := (s11yy1 + 116130*s11cb1) >> 8
-			if s11ru < 0 {
-				s11ru = 0
-			} else if s11ru > 0xffff {
-				s11ru = 0xffff
-			}
-			if s11gu < 0 {
-				s11gu = 0
-			} else if s11gu > 0xffff {
-				s11gu = 0xffff
-			}
-			if s11bu < 0 {
-				s11bu = 0
-			} else if s11bu > 0xffff {
-				s11bu = 0xffff
-			}
-
+			s01a := float64(s01au)
+			s11ru, s11gu, s11bu, s11au := src.At(sr.Min.X+int(sx1), sr.Min.Y+int(sy1)).RGBA()
 			s11r := float64(s11ru)
 			s11g := float64(s11gu)
 			s11b := float64(s11bu)
+			s11a := float64(s11au)
 			s11r = xFrac1*s01r + xFrac0*s11r
 			s11g = xFrac1*s01g + xFrac0*s11g
 			s11b = xFrac1*s01b + xFrac0*s11b
+			s11a = xFrac1*s01a + xFrac0*s11a
 			s11r = yFrac1*s10r + yFrac0*s11r
 			s11g = yFrac1*s10g + yFrac0*s11g
 			s11b = yFrac1*s10b + yFrac0*s11b
+			s11a = yFrac1*s10a + yFrac0*s11a
 			dst.Pix[d+0] = uint8(uint32(s11r) >> 8)
 			dst.Pix[d+1] = uint8(uint32(s11g) >> 8)
 			dst.Pix[d+2] = uint8(uint32(s11b) >> 8)
-			dst.Pix[d+3] = 0xff
+			dst.Pix[d+3] = uint8(uint32(s11a) >> 8)
 		}
 	}
 }
 
-func (ablInterpolator) scale_RGBA_YCbCr422_Src(dst *image.RGBA, dr, adr image.Rectangle, src *image.YCbCr, sr image.Rectangle) {
+func (ablInterpolator) scale_RGBA_Image_Src(dst *image.RGBA, dr, adr image.Rectangle, src image.Image, sr image.Rectangle) {
 	sw := int32(sr.Dx())
 	sh := int32(sr.Dy())
 	yscale := float64(sh) / float64(dr.Dy())
@@ -2933,145 +2342,54 @@ func (ablInterpolator) scale_RGBA_YCbCr422_Src(dst *image.RGBA, dr, adr image.Re
 				xFrac0, xFrac1 = 1, 0
 			}
 
-			s00i := (sr.Min.Y+int(sy0)-src.Rect.Min.Y)*src.YStride + (sr.Min.X + int(sx0) - src.Rect.Min.X)
-			s00j := (sr.Min.Y+int(sy0)-src.Rect.Min.Y)*src.CStride + ((sr.Min.X+int(sx0))/2 - src.Rect.Min.X/2)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s00yy1 := int(src.Y[s00i])<<16 + 1<<15
-			s00cb1 := int(src.Cb[s00j]) - 128
-			s00cr1 := int(src.Cr[s00j]) - 128
-			s00ru := (s00yy1 + 91881*s00cr1) >> 8
-			s00gu := (s00yy1 - 22554*s00cb1 - 46802*s00cr1) >> 8
-			s00bu := (s00yy1 + 116130*s00cb1) >> 8
-			if s00ru < 0 {
-				s00ru = 0
-			} else if s00ru > 0xffff {
-				s00ru = 0xffff
-			}
-			if s00gu < 0 {
-				s00gu = 0
-			} else if s00gu > 0xffff {
-				s00gu = 0xffff
-			}
-			if s00bu < 0 {
-				s00bu = 0
-			} else if s00bu > 0xffff {
-				s00bu = 0xffff
-			}
-
+			s00ru, s00gu, s00bu, s00au := src.At(sr.Min.X+int(sx0), sr.Min.Y+int(sy0)).RGBA()
 			s00r := float64(s00ru)
 			s00g := float64(s00gu)
 			s00b := float64(s00bu)
-			s10i := (sr.Min.Y+int(sy0)-src.Rect.Min.Y)*src.YStride + (sr.Min.X + int(sx1) - src.Rect.Min.X)
-			s10j := (sr.Min.Y+int(sy0)-src.Rect.Min.Y)*src.CStride + ((sr.Min.X+int(sx1))/2 - src.Rect.Min.X/2)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s10yy1 := int(src.Y[s10i])<<16 + 1<<15
-			s10cb1 := int(src.Cb[s10j]) - 128
-			s10cr1 := int(src.Cr[s10j]) - 128
-			s10ru := (s10yy1 + 91881*s10cr1) >> 8
-			s10gu := (s10yy1 - 22554*s10cb1 - 46802*s10cr1) >> 8
-			s10bu := (s10yy1 + 116130*s10cb1) >> 8
-			if s10ru < 0 {
-				s10ru = 0
-			} else if s10ru > 0xffff {
-				s10ru = 0xffff
-			}
-			if s10gu < 0 {
-				s10gu = 0
-			} else if s10gu > 0xffff {
-				s10gu = 0xffff
-			}
-			if s10bu < 0 {
-				s10bu = 0
-			} else if s10bu > 0xffff {
-				s10bu = 0xffff
-			}
-
+			s00a := float64(s00au)
+			s10ru, s10gu, s10bu, s10au := src.At(sr.Min.X+int(sx1), sr.Min.Y+int(sy0)).RGBA()
 			s10r := float64(s10ru)
 			s10g := float64(s10gu)
 			s10b := float64(s10bu)
+			s10a := float64(s10au)
 			s10r = xFrac1*s00r + xFrac0*s10r
 			s10g = xFrac1*s00g + xFrac0*s10g
 			s10b = xFrac1*s00b + xFrac0*s10b
-			s01i := (sr.Min.Y+int(sy1)-src.Rect.Min.Y)*src.YStride + (sr.Min.X + int(sx0) - src.Rect.Min.X)
-			s01j := (sr.Min.Y+int(sy1)-src.Rect.Min.Y)*src.CStride + ((sr.Min.X+int(sx0))/2 - src.Rect.Min.X/2)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s01yy1 := int(src.Y[s01i])<<16 + 1<<15
-			s01cb1 := int(src.Cb[s01j]) - 128
-			s01cr1 := int(src.Cr[s01j]) - 128
-			s01ru := (s01yy1 + 91881*s01cr1) >> 8
-			s01gu := (s01yy1 - 22554*s01cb1 - 46802*s01cr1) >> 8
-			s01bu := (s01yy1 + 116130*s01cb1) >> 8
-			if s01ru < 0 {
-				s01ru = 0
-			} else if s01ru > 0xffff {
-				s01ru = 0xffff
-			}
-			if s01gu < 0 {
-				s01gu = 0
-			} else if s01gu > 0xffff {
-				s01gu = 0xffff
-			}
-			if s01bu < 0 {
-				s01bu = 0
-			} else if s01bu > 0xffff {
-				s01bu = 0xffff
-			}
-
+			s10a = xFrac1*s00a + xFrac0*s10a
+			s01ru, s01gu, s01bu, s01au := src.At(sr.Min.X+int(sx0), sr.Min.Y+int(sy1)).RGBA()
 			s01r := float64(s01ru)
 			s01g := float64(s01gu)
 			s01b := float64(s01bu)
-			s11i := (sr.Min.Y+int(sy1)-src.Rect.Min.Y)*src.YStride + (sr.Min.X + int(sx1) - src.Rect.Min.X)
-			s11j := (sr.Min.Y+int(sy1)-src.Rect.Min.Y)*src.CStride + ((sr.Min.X+int(sx1))/2 - src.Rect.Min.X/2)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s11yy1 := int(src.Y[s11i])<<16 + 1<<15
-			s11cb1 := int(src.Cb[s11j]) - 128
-			s11cr1 := int(src.Cr[s11j]) - 128
-			s11ru := (s11yy1 + 91881*s11cr1) >> 8
-			s11gu := (s11yy1 - 22554*s11cb1 - 46802*s11cr1) >> 8
-			s11bu := (s11yy1 + 116130*s11cb1) >> 8
-			if s11ru < 0 {
-				s11ru = 0
-			} else if s11ru > 0xffff {
-				s11ru = 0xffff
-			}
-			if s11gu < 0 {
-				s11gu = 0
-			} else if s11gu > 0xffff {
-				s11gu = 0xffff
-			}
-			if s11bu < 0 {
-				s11bu = 0
-			} else if s11bu > 0xffff {
-				s11bu = 0xffff
-			}
-
+			s01a := float64(s01au)
+			s11ru, s11gu, s11bu, s11au := src.At(sr.Min.X+int(sx1), sr.Min.Y+int(sy1)).RGBA()
 			s11r := float64(s11ru)
 			s11g := float64(s11gu)
 			s11b := float64(s11bu)
+			s11a := float64(s11au)
 			s11r = xFrac1*s01r + xFrac0*s11r
 			s11g = xFrac1*s01g + xFrac0*s11g
 			s11b = xFrac1*s01b + xFrac0*s11b
+			s11a = xFrac1*s01a + xFrac0*s11a
 			s11r = yFrac1*s10r + yFrac0*s11r
 			s11g = yFrac1*s10g + yFrac0*s11g
 			s11b = yFrac1*s10b + yFrac0*s11b
+			s11a = yFrac1*s10a + yFrac0*s11a
 			dst.Pix[d+0] = uint8(uint32(s11r) >> 8)
 			dst.Pix[d+1] = uint8(uint32(s11g) >> 8)
 			dst.Pix[d+2] = uint8(uint32(s11b) >> 8)
-			dst.Pix[d+3] = 0xff
+			dst.Pix[d+3] = uint8(uint32(s11a) >> 8)
 		}
 	}
 }
 
-func (ablInterpolator) scale_RGBA_YCbCr420_Src(dst *image.RGBA, dr, adr image.Rectangle, src *image.YCbCr, sr image.Rectangle) {
+func (ablInterpolator) scale_Image_Image_Over(dst Image, dr, adr image.Rectangle, src image.Image, sr image.Rectangle) {
 	sw := int32(sr.Dx())
 	sh := int32(sr.Dy())
 	yscale := float64(sh) / float64(dr.Dy())
 	xscale := float64(sw) / float64(dr.Dx())
 	swMinus1, shMinus1 := sw-1, sh-1
+	dstColorRGBA64 := &color.RGBA64{}
+	dstColor := color.Color(dstColorRGBA64)
 
 	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
 		sy := (float64(dy)+0.5)*yscale - 0.5
@@ -3089,9 +2407,8 @@ func (ablInterpolator) scale_RGBA_YCbCr420_Src(dst *image.RGBA, dr, adr image.Re
 			sy0, sy1 = shMinus1, shMinus1
 			yFrac0, yFrac1 = 1, 0
 		}
-		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
 
-		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx, d = dx+1, d+4 {
+		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx++ {
 			sx := (float64(dx)+0.5)*xscale - 0.5
 			sx0 := int32(sx)
 			xFrac0 := sx - float64(sx0)
@@ -3105,145 +2422,55 @@ func (ablInterpolator) scale_RGBA_YCbCr420_Src(dst *image.RGBA, dr, adr image.Re
 				xFrac0, xFrac1 = 1, 0
 			}
 
-			s00i := (sr.Min.Y+int(sy0)-src.Rect.Min.Y)*src.YStride + (sr.Min.X + int(sx0) - src.Rect.Min.X)
-			s00j := ((sr.Min.Y+int(sy0))/2-src.Rect.Min.Y/2)*src.CStride + ((sr.Min.X+int(sx0))/2 - src.Rect.Min.X/2)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s00yy1 := int(src.Y[s00i])<<16 + 1<<15
-			s00cb1 := int(src.Cb[s00j]) - 128
-			s00cr1 := int(src.Cr[s00j]) - 128
-			s00ru := (s00yy1 + 91881*s00cr1) >> 8
-			s00gu := (s00yy1 - 22554*s00cb1 - 46802*s00cr1) >> 8
-			s00bu := (s00yy1 + 116130*s00cb1) >> 8
-			if s00ru < 0 {
-				s00ru = 0
-			} else if s00ru > 0xffff {
-				s00ru = 0xffff
-			}
-			if s00gu < 0 {
-				s00gu = 0
-			} else if s00gu > 0xffff {
-				s00gu = 0xffff
-			}
-			if s00bu < 0 {
-				s00bu = 0
-			} else if s00bu > 0xffff {
-				s00bu = 0xffff
-			}
-
+			s00ru, s00gu, s00bu, s00au := src.At(sr.Min.X+int(sx0), sr.Min.Y+int(sy0)).RGBA()
 			s00r := float64(s00ru)
 			s00g := float64(s00gu)
 			s00b := float64(s00bu)
-			s10i := (sr.Min.Y+int(sy0)-src.Rect.Min.Y)*src.YStride + (sr.Min.X + int(sx1) - src.Rect.Min.X)
-			s10j := ((sr.Min.Y+int(sy0))/2-src.Rect.Min.Y/2)*src.CStride + ((sr.Min.X+int(sx1))/2 - src.Rect.Min.X/2)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s10yy1 := int(src.Y[s10i])<<16 + 1<<15
-			s10cb1 := int(src.Cb[s10j]) - 128
-			s10cr1 := int(src.Cr[s10j]) - 128
-			s10ru := (s10yy1 + 91881*s10cr1) >> 8
-			s10gu := (s10yy1 - 22554*s10cb1 - 46802*s10cr1) >> 8
-			s10bu := (s10yy1 + 116130*s10cb1) >> 8
-			if s10ru < 0 {
-				s10ru = 0
-			} else if s10ru > 0xffff {
-				s10ru = 0xffff
-			}
-			if s10gu < 0 {
-				s10gu = 0
-			} else if s10gu > 0xffff {
-				s10gu = 0xffff
-			}
-			if s10bu < 0 {
-				s10bu = 0
-			} else if s10bu > 0xffff {
-				s10bu = 0xffff
-			}
-
+			s00a := float64(s00au)
+			s10ru, s10gu, s10bu, s10au := src.At(sr.Min.X+int(sx1), sr.Min.Y+int(sy0)).RGBA()
 			s10r := float64(s10ru)
 			s10g := float64(s10gu)
 			s10b := float64(s10bu)
+			s10a := float64(s10au)
 			s10r = xFrac1*s00r + xFrac0*s10r
 			s10g = xFrac1*s00g + xFrac0*s10g
 			s10b = xFrac1*s00b + xFrac0*s10b
-			s01i := (sr.Min.Y+int(sy1)-src.Rect.Min.Y)*src.YStride + (sr.Min.X + int(sx0) - src.Rect.Min.X)
-			s01j := ((sr.Min.Y+int(sy1))/2-src.Rect.Min.Y/2)*src.CStride + ((sr.Min.X+int(sx0))/2 - src.Rect.Min.X/2)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s01yy1 := int(src.Y[s01i])<<16 + 1<<15
-			s01cb1 := int(src.Cb[s01j]) - 128
-			s01cr1 := int(src.Cr[s01j]) - 128
-			s01ru := (s01yy1 + 91881*s01cr1) >> 8
-			s01gu := (s01yy1 - 22554*s01cb1 - 46802*s01cr1) >> 8
-			s01bu := (s01yy1 + 116130*s01cb1) >> 8
-			if s01ru < 0 {
-				s01ru = 0
-			} else if s01ru > 0xffff {
-				s01ru = 0xffff
-			}
-			if s01gu < 0 {
-				s01gu = 0
-			} else if s01gu > 0xffff {
-				s01gu = 0xffff
-			}
-			if s01bu < 0 {
-				s01bu = 0
-			} else if s01bu > 0xffff {
-				s01bu = 0xffff
-			}
-
+			s10a = xFrac1*s00a + xFrac0*s10a
+			s01ru, s01gu, s01bu, s01au := src.At(sr.Min.X+int(sx0), sr.Min.Y+int(sy1)).RGBA()
 			s01r := float64(s01ru)
 			s01g := float64(s01gu)
 			s01b := float64(s01bu)
-			s11i := (sr.Min.Y+int(sy1)-src.Rect.Min.Y)*src.YStride + (sr.Min.X + int(sx1) - src.Rect.Min.X)
-			s11j := ((sr.Min.Y+int(sy1))/2-src.Rect.Min.Y/2)*src.CStride + ((sr.Min.X+int(sx1))/2 - src.Rect.Min.X/2)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s11yy1 := int(src.Y[s11i])<<16 + 1<<15
-			s11cb1 := int(src.Cb[s11j]) - 128
-			s11cr1 := int(src.Cr[s11j]) - 128
-			s11ru := (s11yy1 + 91881*s11cr1) >> 8
-			s11gu := (s11yy1 - 22554*s11cb1 - 46802*s11cr1) >> 8
-			s11bu := (s11yy1 + 116130*s11cb1) >> 8
-			if s11ru < 0 {
-				s11ru = 0
-			} else if s11ru > 0xffff {
-				s11ru = 0xffff
-			}
-			if s11gu < 0 {
-				s11gu = 0
-			} else if s11gu > 0xffff {
-				s11gu = 0xffff
-			}
-			if s11bu < 0 {
-				s11bu = 0
-			} else if s11bu > 0xffff {
-				s11bu = 0xffff
-			}
-
+			s01a := float64(s01au)
+			s11ru, s11gu, s11bu, s11au := src.At(sr.Min.X+int(sx1), sr.Min.Y+int(sy1)).RGBA()
 			s11r := float64(s11ru)
 			s11g := float64(s11gu)
 			s11b := float64(s11bu)
+			s11a := float64(s11au)
 			s11r = xFrac1*s01r + xFrac0*s11r
 			s11g = xFrac1*s01g + xFrac0*s11g
 			s11b = xFrac1*s01b + xFrac0*s11b
+			s11a = xFrac1*s01a + xFrac0*s11a
 			s11r = yFrac1*s10r + yFrac0*s11r
 			s11g = yFrac1*s10g + yFrac0*s11g
 			s11b = yFrac1*s10b + yFrac0*s11b
-			dst.Pix[d+0] = uint8(uint32(s11r) >> 8)
-			dst.Pix[d+1] = uint8(uint32(s11g) >> 8)
-			dst.Pix[d+2] = uint8(uint32(s11b) >> 8)
-			dst.Pix[d+3] = 0xff
+			s11a = yFrac1*s10a + yFrac0*s11a
+			dstColorRGBA64.R = uint16(s11r)
+			dstColorRGBA64.G = uint16(s11g)
+			dstColorRGBA64.B = uint16(s11b)
+			dstColorRGBA64.A = uint16(s11a)
+			dst.Set(dr.Min.X+int(dx), dr.Min.Y+int(dy), dstColor)
 		}
 	}
 }
 
-func (ablInterpolator) scale_RGBA_YCbCr440_Src(dst *image.RGBA, dr, adr image.Rectangle, src *image.YCbCr, sr image.Rectangle) {
+func (ablInterpolator) scale_Image_Image_Src(dst Image, dr, adr image.Rectangle, src image.Image, sr image.Rectangle) {
 	sw := int32(sr.Dx())
 	sh := int32(sr.Dy())
 	yscale := float64(sh) / float64(dr.Dy())
 	xscale := float64(sw) / float64(dr.Dx())
 	swMinus1, shMinus1 := sw-1, sh-1
+	dstColorRGBA64 := &color.RGBA64{}
+	dstColor := color.Color(dstColorRGBA64)
 
 	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
 		sy := (float64(dy)+0.5)*yscale - 0.5
@@ -3261,9 +2488,8 @@ func (ablInterpolator) scale_RGBA_YCbCr440_Src(dst *image.RGBA, dr, adr image.Re
 			sy0, sy1 = shMinus1, shMinus1
 			yFrac0, yFrac1 = 1, 0
 		}
-		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
 
-		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx, d = dx+1, d+4 {
+		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx++ {
 			sx := (float64(dx)+0.5)*xscale - 0.5
 			sx0 := int32(sx)
 			xFrac0 := sx - float64(sx0)
@@ -3277,184 +2503,165 @@ func (ablInterpolator) scale_RGBA_YCbCr440_Src(dst *image.RGBA, dr, adr image.Re
 				xFrac0, xFrac1 = 1, 0
 			}
 
-			s00i := (sr.Min.Y+int(sy0)-src.Rect.Min.Y)*src.YStride + (sr.Min.X + int(sx0) - src.Rect.Min.X)
-			s00j := ((sr.Min.Y+int(sy0))/2-src.Rect.Min.Y/2)*src.CStride + (sr.Min.X + int(sx0) - src.Rect.Min.X)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s00yy1 := int(src.Y[s00i])<<16 + 1<<15
-			s00cb1 := int(src.Cb[s00j]) - 128
-			s00cr1 := int(src.Cr[s00j]) - 128
-			s00ru := (s00yy1 + 91881*s00cr1) >> 8
-			s00gu := (s00yy1 - 22554*s00cb1 - 46802*s00cr1) >> 8
-			s00bu := (s00yy1 + 116130*s00cb1) >> 8
-			if s00ru < 0 {
-				s00ru = 0
-			} else if s00ru > 0xffff {
-				s00ru = 0xffff
-			}
-			if s00gu < 0 {
-				s00gu = 0
-			} else if s00gu > 0xffff {
-				s00gu = 0xffff
-			}
-			if s00bu < 0 {
-				s00bu = 0
-			} else if s00bu > 0xffff {
-				s00bu = 0xffff
-			}
-
+			s00ru, s00gu, s00bu, s00au := src.At(sr.Min.X+int(sx0), sr.Min.Y+int(sy0)).RGBA()
 			s00r := float64(s00ru)
 			s00g := float64(s00gu)
 			s00b := float64(s00bu)
-			s10i := (sr.Min.Y+int(sy0)-src.Rect.Min.Y)*src.YStride + (sr.Min.X + int(sx1) - src.Rect.Min.X)
-			s10j := ((sr.Min.Y+int(sy0))/2-src.Rect.Min.Y/2)*src.CStride + (sr.Min.X + int(sx1) - src.Rect.Min.X)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s10yy1 := int(src.Y[s10i])<<16 + 1<<15
-			s10cb1 := int(src.Cb[s10j]) - 128
-			s10cr1 := int(src.Cr[s10j]) - 128
-			s10ru := (s10yy1 + 91881*s10cr1) >> 8
-			s10gu := (s10yy1 - 22554*s10cb1 - 46802*s10cr1) >> 8
-			s10bu := (s10yy1 + 116130*s10cb1) >> 8
-			if s10ru < 0 {
-				s10ru = 0
-			} else if s10ru > 0xffff {
-				s10ru = 0xffff
-			}
-			if s10gu < 0 {
-				s10gu = 0
-			} else if s10gu > 0xffff {
-				s10gu = 0xffff
-			}
-			if s10bu < 0 {
-				s10bu = 0
-			} else if s10bu > 0xffff {
-				s10bu = 0xffff
-			}
-
+			s00a := float64(s00au)
+			s10ru, s10gu, s10bu, s10au := src.At(sr.Min.X+int(sx1), sr.Min.Y+int(sy0)).RGBA()
 			s10r := float64(s10ru)
 			s10g := float64(s10gu)
 			s10b := float64(s10bu)
+			s10a := float64(s10au)
 			s10r = xFrac1*s00r + xFrac0*s10r
 			s10g = xFrac1*s00g + xFrac0*s10g
 			s10b = xFrac1*s00b + xFrac0*s10b
-			s01i := (sr.Min.Y+int(sy1)-src.Rect.Min.Y)*src.YStride + (sr.Min.X + int(sx0) - src.Rect.Min.X)
-			s01j := ((sr.Min.Y+int(sy1))/2-src.Rect.Min.Y/2)*src.CStride + (sr.Min.X + int(sx0) - src.Rect.Min.X)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s01yy1 := int(src.Y[s01i])<<16 + 1<<15
-			s01cb1 := int(src.Cb[s01j]) - 128
-			s01cr1 := int(src.Cr[s01j]) - 128
-			s01ru := (s01yy1 + 91881*s01cr1) >> 8
-			s01gu := (s01yy1 - 22554*s01cb1 - 46802*s01cr1) >> 8
-			s01bu := (s01yy1 + 116130*s01cb1) >> 8
-			if s01ru < 0 {
-				s01ru = 0
-			} else if s01ru > 0xffff {
-				s01ru = 0xffff
-			}
-			if s01gu < 0 {
-				s01gu = 0
-			} else if s01gu > 0xffff {
-				s01gu = 0xffff
-			}
-			if s01bu < 0 {
-				s01bu = 0
-			} else if s01bu > 0xffff {
-				s01bu = 0xffff
-			}
-
+			s10a = xFrac1*s00a + xFrac0*s10a
+			s01ru, s01gu, s01bu, s01au := src.At(sr.Min.X+int(sx0), sr.Min.Y+int(sy1)).RGBA()
 			s01r := float64(s01ru)
 			s01g := float64(s01gu)
 			s01b := float64(s01bu)
-			s11i := (sr.Min.Y+int(sy1)-src.Rect.Min.Y)*src.YStride + (sr.Min.X + int(sx1) - src.Rect.Min.X)
-			s11j := ((sr.Min.Y+int(sy1))/2-src.Rect.Min.Y/2)*src.CStride + (sr.Min.X + int(sx1) - src.Rect.Min.X)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s11yy1 := int(src.Y[s11i])<<16 + 1<<15
-			s11cb1 := int(src.Cb[s11j]) - 128
-			s11cr1 := int(src.Cr[s11j]) - 128
-			s11ru := (s11yy1 + 91881*s11cr1) >> 8
-			s11gu := (s11yy1 - 22554*s11cb1 - 46802*s11cr1) >> 8
-			s11bu := (s11yy1 + 116130*s11cb1) >> 8
-			if s11ru < 0 {
-				s11ru = 0
-			} else if s11ru > 0xffff {
-				s11ru = 0xffff
-			}
-			if s11gu < 0 {
-				s11gu = 0
-			} else if s11gu > 0xffff {
-				s11gu = 0xffff
-			}
-			if s11bu < 0 {
-				s11bu = 0
-			} else if s11bu > 0xffff {
-				s11bu = 0xffff
-			}
-
+			s01a := float64(s01au)
+			s11ru, s11gu, s11bu, s11au := src.At(sr.Min.X+int(sx1), sr.Min.Y+int(sy1)).RGBA()
 			s11r := float64(s11ru)
 			s11g := float64(s11gu)
 			s11b := float64(s11bu)
+			s11a := float64(s11au)
 			s11r = xFrac1*s01r + xFrac0*s11r
 			s11g = xFrac1*s01g + xFrac0*s11g
 			s11b = xFrac1*s01b + xFrac0*s11b
+			s11a = xFrac1*s01a + xFrac0*s11a
 			s11r = yFrac1*s10r + yFrac0*s11r
 			s11g = yFrac1*s10g + yFrac0*s11g
 			s11b = yFrac1*s10b + yFrac0*s11b
-			dst.Pix[d+0] = uint8(uint32(s11r) >> 8)
-			dst.Pix[d+1] = uint8(uint32(s11g) >> 8)
-			dst.Pix[d+2] = uint8(uint32(s11b) >> 8)
-			dst.Pix[d+3] = 0xff
+			s11a = yFrac1*s10a + yFrac0*s11a
+			dstColorRGBA64.R = uint16(s11r)
+			dstColorRGBA64.G = uint16(s11g)
+			dstColorRGBA64.B = uint16(s11b)
+			dstColorRGBA64.A = uint16(s11a)
+			dst.Set(dr.Min.X+int(dx), dr.Min.Y+int(dy), dstColor)
 		}
 	}
 }
 
-func (ablInterpolator) scale_RGBA_Image_Over(dst *image.RGBA, dr, adr image.Rectangle, src image.Image, sr image.Rectangle) {
-	sw := int32(sr.Dx())
-	sh := int32(sr.Dy())
-	yscale := float64(sh) / float64(dr.Dy())
-	xscale := float64(sw) / float64(dr.Dx())
-	swMinus1, shMinus1 := sw-1, sh-1
-
+func (ablInterpolator) transform_RGBA_Gray_Src(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.Gray, sr image.Rectangle, bias image.Point) {
 	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
-		sy := (float64(dy)+0.5)*yscale - 0.5
-		// If sy < 0, we will clamp sy0 to 0 anyway, so it doesn't matter if
-		// we say int32(sy) instead of int32(math.Floor(sy)). Similarly for
-		// sx, below.
-		sy0 := int32(sy)
-		yFrac0 := sy - float64(sy0)
-		yFrac1 := 1 - yFrac0
-		sy1 := sy0 + 1
-		if sy < 0 {
-			sy0, sy1 = 0, 0
-			yFrac0, yFrac1 = 0, 1
-		} else if sy1 > shMinus1 {
-			sy0, sy1 = shMinus1, shMinus1
-			yFrac0, yFrac1 = 1, 0
-		}
+		dyf := float64(dr.Min.Y+int(dy)) + 0.5
 		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
+		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx, d = dx+1, d+4 {
+			dxf := float64(dr.Min.X+int(dx)) + 0.5
+			sx := d2s[0]*dxf + d2s[1]*dyf + d2s[2]
+			sy := d2s[3]*dxf + d2s[4]*dyf + d2s[5]
+			if !(image.Point{int(sx) + bias.X, int(sy) + bias.Y}).In(sr) {
+				continue
+			}
+
+			sx -= 0.5
+			sx0 := int(sx)
+			xFrac0 := sx - float64(sx0)
+			xFrac1 := 1 - xFrac0
+			sx0 += bias.X
+			sx1 := sx0 + 1
+			if sx0 < sr.Min.X {
+				sx0, sx1 = sr.Min.X, sr.Min.X
+				xFrac0, xFrac1 = 0, 1
+			} else if sx1 >= sr.Max.X {
+				sx0, sx1 = sr.Max.X-1, sr.Max.X-1
+				xFrac0, xFrac1 = 1, 0
+			}
+
+			sy -= 0.5
+			sy0 := int(sy)
+			yFrac0 := sy - float64(sy0)
+			yFrac1 := 1 - yFrac0
+			sy0 += bias.Y
+			sy1 := sy0 + 1
+			if sy0 < sr.Min.Y {
+				sy0, sy1 = sr.Min.Y, sr.Min.Y
+				yFrac0, yFrac1 = 0, 1
+			} else if sy1 >= sr.Max.Y {
+				sy0, sy1 = sr.Max.Y-1, sr.Max.Y-1
+				yFrac0, yFrac1 = 1, 0
+			}
+
+			s00i := (sy0-src.Rect.Min.Y)*src.Stride + (sx0 - src.Rect.Min.X)
+			s00ru := uint32(src.Pix[s00i]) * 0x101
+			s00r := float64(s00ru)
+			s10i := (sy0-src.Rect.Min.Y)*src.Stride + (sx1 - src.Rect.Min.X)
+			s10ru := uint32(src.Pix[s10i]) * 0x101
+			s10r := float64(s10ru)
+			s10r = xFrac1*s00r + xFrac0*s10r
+			s01i := (sy1-src.Rect.Min.Y)*src.Stride + (sx0 - src.Rect.Min.X)
+			s01ru := uint32(src.Pix[s01i]) * 0x101
+			s01r := float64(s01ru)
+			s11i := (sy1-src.Rect.Min.Y)*src.Stride + (sx1 - src.Rect.Min.X)
+			s11ru := uint32(src.Pix[s11i]) * 0x101
+			s11r := float64(s11ru)
+			s11r = xFrac1*s01r + xFrac0*s11r
+			s11r = yFrac1*s10r + yFrac0*s11r
+			out := uint8(uint32(s11r) >> 8)
+			dst.Pix[d+0] = out
+			dst.Pix[d+1] = out
+			dst.Pix[d+2] = out
+			dst.Pix[d+3] = 0xff
+		}
+	}
+}
 
+func (ablInterpolator) transform_RGBA_NRGBA_Over(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.NRGBA, sr image.Rectangle, bias image.Point) {
+	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
+		dyf := float64(dr.Min.Y+int(dy)) + 0.5
+		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
 		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx, d = dx+1, d+4 {
-			sx := (float64(dx)+0.5)*xscale - 0.5
-			sx0 := int32(sx)
+			dxf := float64(dr.Min.X+int(dx)) + 0.5
+			sx := d2s[0]*dxf + d2s[1]*dyf + d2s[2]
+			sy := d2s[3]*dxf + d2s[4]*dyf + d2s[5]
+			if !(image.Point{int(sx) + bias.X, int(sy) + bias.Y}).In(sr) {
+				continue
+			}
+
+			sx -= 0.5
+			sx0 := int(sx)
 			xFrac0 := sx - float64(sx0)
 			xFrac1 := 1 - xFrac0
+			sx0 += bias.X
 			sx1 := sx0 + 1
-			if sx < 0 {
-				sx0, sx1 = 0, 0
+			if sx0 < sr.Min.X {
+				sx0, sx1 = sr.Min.X, sr.Min.X
 				xFrac0, xFrac1 = 0, 1
-			} else if sx1 > swMinus1 {
-				sx0, sx1 = swMinus1, swMinus1
+			} else if sx1 >= sr.Max.X {
+				sx0, sx1 = sr.Max.X-1, sr.Max.X-1
 				xFrac0, xFrac1 = 1, 0
 			}
 
-			s00ru, s00gu, s00bu, s00au := src.At(sr.Min.X+int(sx0), sr.Min.Y+int(sy0)).RGBA()
+			sy -= 0.5
+			sy0 := int(sy)
+			yFrac0 := sy - float64(sy0)
+			yFrac1 := 1 - yFrac0
+			sy0 += bias.Y
+			sy1 := sy0 + 1
+			if sy0 < sr.Min.Y {
+				sy0, sy1 = sr.Min.Y, sr.Min.Y
+				yFrac0, yFrac1 = 0, 1
+			} else if sy1 >= sr.Max.Y {
+				sy0, sy1 = sr.Max.Y-1, sr.Max.Y-1
+				yFrac0, yFrac1 = 1, 0
+			}
+
+			s00i := (sy0-src.Rect.Min.Y)*src.Stride + (sx0-src.Rect.Min.X)*4
+			s00au := uint32(src.Pix[s00i+3]) * 0x101
+			s00ru := uint32(src.Pix[s00i+0]) * s00au / 0xff
+			s00gu := uint32(src.Pix[s00i+1]) * s00au / 0xff
+			s00bu := uint32(src.Pix[s00i+2]) * s00au / 0xff
 			s00r := float64(s00ru)
 			s00g := float64(s00gu)
 			s00b := float64(s00bu)
 			s00a := float64(s00au)
-			s10ru, s10gu, s10bu, s10au := src.At(sr.Min.X+int(sx1), sr.Min.Y+int(sy0)).RGBA()
+			s10i := (sy0-src.Rect.Min.Y)*src.Stride + (sx1-src.Rect.Min.X)*4
+			s10au := uint32(src.Pix[s10i+3]) * 0x101
+			s10ru := uint32(src.Pix[s10i+0]) * s10au / 0xff
+			s10gu := uint32(src.Pix[s10i+1]) * s10au / 0xff
+			s10bu := uint32(src.Pix[s10i+2]) * s10au / 0xff
 			s10r := float64(s10ru)
 			s10g := float64(s10gu)
 			s10b := float64(s10bu)
@@ -3463,12 +2670,20 @@ func (ablInterpolator) scale_RGBA_Image_Over(dst *image.RGBA, dr, adr image.Rect
 			s10g = xFrac1*s00g + xFrac0*s10g
 			s10b = xFrac1*s00b + xFrac0*s10b
 			s10a = xFrac1*s00a + xFrac0*s10a
-			s01ru, s01gu, s01bu, s01au := src.At(sr.Min.X+int(sx0), sr.Min.Y+int(sy1)).RGBA()
+			s01i := (sy1-src.Rect.Min.Y)*src.Stride + (sx0-src.Rect.Min.X)*4
+			s01au := uint32(src.Pix[s01i+3]) * 0x101
+			s01ru := uint32(src.Pix[s01i+0]) * s01au / 0xff
+			s01gu := uint32(src.Pix[s01i+1]) * s01au / 0xff
+			s01bu := uint32(src.Pix[s01i+2]) * s01au / 0xff
 			s01r := float64(s01ru)
 			s01g := float64(s01gu)
 			s01b := float64(s01bu)
 			s01a := float64(s01au)
-			s11ru, s11gu, s11bu, s11au := src.At(sr.Min.X+int(sx1), sr.Min.Y+int(sy1)).RGBA()
+			s11i := (sy1-src.Rect.Min.Y)*src.Stride + (sx1-src.Rect.Min.X)*4
+			s11au := uint32(src.Pix[s11i+3]) * 0x101
+			s11ru := uint32(src.Pix[s11i+0]) * s11au / 0xff
+			s11gu := uint32(src.Pix[s11i+1]) * s11au / 0xff
+			s11bu := uint32(src.Pix[s11i+2]) * s11au / 0xff
 			s11r := float64(s11ru)
 			s11g := float64(s11gu)
 			s11b := float64(s11bu)
@@ -3489,51 +2704,60 @@ func (ablInterpolator) scale_RGBA_Image_Over(dst *image.RGBA, dr, adr image.Rect
 	}
 }
 
-func (ablInterpolator) scale_RGBA_Image_Src(dst *image.RGBA, dr, adr image.Rectangle, src image.Image, sr image.Rectangle) {
-	sw := int32(sr.Dx())
-	sh := int32(sr.Dy())
-	yscale := float64(sh) / float64(dr.Dy())
-	xscale := float64(sw) / float64(dr.Dx())
-	swMinus1, shMinus1 := sw-1, sh-1
-
+func (ablInterpolator) transform_RGBA_NRGBA_Src(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.NRGBA, sr image.Rectangle, bias image.Point) {
 	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
-		sy := (float64(dy)+0.5)*yscale - 0.5
-		// If sy < 0, we will clamp sy0 to 0 anyway, so it doesn't matter if
-		// we say int32(sy) instead of int32(math.Floor(sy)). Similarly for
-		// sx, below.
-		sy0 := int32(sy)
-		yFrac0 := sy - float64(sy0)
-		yFrac1 := 1 - yFrac0
-		sy1 := sy0 + 1
-		if sy < 0 {
-			sy0, sy1 = 0, 0
-			yFrac0, yFrac1 = 0, 1
-		} else if sy1 > shMinus1 {
-			sy0, sy1 = shMinus1, shMinus1
-			yFrac0, yFrac1 = 1, 0
-		}
+		dyf := float64(dr.Min.Y+int(dy)) + 0.5
 		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
-
 		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx, d = dx+1, d+4 {
-			sx := (float64(dx)+0.5)*xscale - 0.5
-			sx0 := int32(sx)
+			dxf := float64(dr.Min.X+int(dx)) + 0.5
+			sx := d2s[0]*dxf + d2s[1]*dyf + d2s[2]
+			sy := d2s[3]*dxf + d2s[4]*dyf + d2s[5]
+			if !(image.Point{int(sx) + bias.X, int(sy) + bias.Y}).In(sr) {
+				continue
+			}
+
+			sx -= 0.5
+			sx0 := int(sx)
 			xFrac0 := sx - float64(sx0)
 			xFrac1 := 1 - xFrac0
+			sx0 += bias.X
 			sx1 := sx0 + 1
-			if sx < 0 {
-				sx0, sx1 = 0, 0
+			if sx0 < sr.Min.X {
+				sx0, sx1 = sr.Min.X, sr.Min.X
 				xFrac0, xFrac1 = 0, 1
-			} else if sx1 > swMinus1 {
-				sx0, sx1 = swMinus1, swMinus1
+			} else if sx1 >= sr.Max.X {
+				sx0, sx1 = sr.Max.X-1, sr.Max.X-1
 				xFrac0, xFrac1 = 1, 0
 			}
 
-			s00ru, s00gu, s00bu, s00au := src.At(sr.Min.X+int(sx0), sr.Min.Y+int(sy0)).RGBA()
+			sy -= 0.5
+			sy0 := int(sy)
+			yFrac0 := sy - float64(sy0)
+			yFrac1 := 1 - yFrac0
+			sy0 += bias.Y
+			sy1 := sy0 + 1
+			if sy0 < sr.Min.Y {
+				sy0, sy1 = sr.Min.Y, sr.Min.Y
+				yFrac0, yFrac1 = 0, 1
+			} else if sy1 >= sr.Max.Y {
+				sy0, sy1 = sr.Max.Y-1, sr.Max.Y-1
+				yFrac0, yFrac1 = 1, 0
+			}
+
+			s00i := (sy0-src.Rect.Min.Y)*src.Stride + (sx0-src.Rect.Min.X)*4
+			s00au := uint32(src.Pix[s00i+3]) * 0x101
+			s00ru := uint32(src.Pix[s00i+0]) * s00au / 0xff
+			s00gu := uint32(src.Pix[s00i+1]) * s00au / 0xff
+			s00bu := uint32(src.Pix[s00i+2]) * s00au / 0xff
 			s00r := float64(s00ru)
 			s00g := float64(s00gu)
 			s00b := float64(s00bu)
 			s00a := float64(s00au)
-			s10ru, s10gu, s10bu, s10au := src.At(sr.Min.X+int(sx1), sr.Min.Y+int(sy0)).RGBA()
+			s10i := (sy0-src.Rect.Min.Y)*src.Stride + (sx1-src.Rect.Min.X)*4
+			s10au := uint32(src.Pix[s10i+3]) * 0x101
+			s10ru := uint32(src.Pix[s10i+0]) * s10au / 0xff
+			s10gu := uint32(src.Pix[s10i+1]) * s10au / 0xff
+			s10bu := uint32(src.Pix[s10i+2]) * s10au / 0xff
 			s10r := float64(s10ru)
 			s10g := float64(s10gu)
 			s10b := float64(s10bu)
@@ -3542,12 +2766,20 @@ func (ablInterpolator) scale_RGBA_Image_Src(dst *image.RGBA, dr, adr image.Recta
 			s10g = xFrac1*s00g + xFrac0*s10g
 			s10b = xFrac1*s00b + xFrac0*s10b
 			s10a = xFrac1*s00a + xFrac0*s10a
-			s01ru, s01gu, s01bu, s01au := src.At(sr.Min.X+int(sx0), sr.Min.Y+int(sy1)).RGBA()
+			s01i := (sy1-src.Rect.Min.Y)*src.Stride + (sx0-src.Rect.Min.X)*4
+			s01au := uint32(src.Pix[s01i+3]) * 0x101
+			s01ru := uint32(src.Pix[s01i+0]) * s01au / 0xff
+			s01gu := uint32(src.Pix[s01i+1]) * s01au / 0xff
+			s01bu := uint32(src.Pix[s01i+2]) * s01au / 0xff
 			s01r := float64(s01ru)
 			s01g := float64(s01gu)
 			s01b := float64(s01bu)
 			s01a := float64(s01au)
-			s11ru, s11gu, s11bu, s11au := src.At(sr.Min.X+int(sx1), sr.Min.Y+int(sy1)).RGBA()
+			s11i := (sy1-src.Rect.Min.Y)*src.Stride + (sx1-src.Rect.Min.X)*4
+			s11au := uint32(src.Pix[s11i+3]) * 0x101
+			s11ru := uint32(src.Pix[s11i+0]) * s11au / 0xff
+			s11gu := uint32(src.Pix[s11i+1]) * s11au / 0xff
+			s11bu := uint32(src.Pix[s11i+2]) * s11au / 0xff
 			s11r := float64(s11ru)
 			s11g := float64(s11gu)
 			s11b := float64(s11bu)
@@ -3568,133 +2800,60 @@ func (ablInterpolator) scale_RGBA_Image_Src(dst *image.RGBA, dr, adr image.Recta
 	}
 }
 
-func (ablInterpolator) scale_Image_Image_Over(dst Image, dr, adr image.Rectangle, src image.Image, sr image.Rectangle) {
-	sw := int32(sr.Dx())
-	sh := int32(sr.Dy())
-	yscale := float64(sh) / float64(dr.Dy())
-	xscale := float64(sw) / float64(dr.Dx())
-	swMinus1, shMinus1 := sw-1, sh-1
-	dstColorRGBA64 := &color.RGBA64{}
-	dstColor := color.Color(dstColorRGBA64)
-
+func (ablInterpolator) transform_RGBA_RGBA_Over(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.RGBA, sr image.Rectangle, bias image.Point) {
 	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
-		sy := (float64(dy)+0.5)*yscale - 0.5
-		// If sy < 0, we will clamp sy0 to 0 anyway, so it doesn't matter if
-		// we say int32(sy) instead of int32(math.Floor(sy)). Similarly for
-		// sx, below.
-		sy0 := int32(sy)
-		yFrac0 := sy - float64(sy0)
-		yFrac1 := 1 - yFrac0
-		sy1 := sy0 + 1
-		if sy < 0 {
-			sy0, sy1 = 0, 0
-			yFrac0, yFrac1 = 0, 1
-		} else if sy1 > shMinus1 {
-			sy0, sy1 = shMinus1, shMinus1
-			yFrac0, yFrac1 = 1, 0
-		}
+		dyf := float64(dr.Min.Y+int(dy)) + 0.5
+		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
+		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx, d = dx+1, d+4 {
+			dxf := float64(dr.Min.X+int(dx)) + 0.5
+			sx := d2s[0]*dxf + d2s[1]*dyf + d2s[2]
+			sy := d2s[3]*dxf + d2s[4]*dyf + d2s[5]
+			if !(image.Point{int(sx) + bias.X, int(sy) + bias.Y}).In(sr) {
+				continue
+			}
 
-		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx++ {
-			sx := (float64(dx)+0.5)*xscale - 0.5
-			sx0 := int32(sx)
+			sx -= 0.5
+			sx0 := int(sx)
 			xFrac0 := sx - float64(sx0)
 			xFrac1 := 1 - xFrac0
+			sx0 += bias.X
 			sx1 := sx0 + 1
-			if sx < 0 {
-				sx0, sx1 = 0, 0
+			if sx0 < sr.Min.X {
+				sx0, sx1 = sr.Min.X, sr.Min.X
 				xFrac0, xFrac1 = 0, 1
-			} else if sx1 > swMinus1 {
-				sx0, sx1 = swMinus1, swMinus1
+			} else if sx1 >= sr.Max.X {
+				sx0, sx1 = sr.Max.X-1, sr.Max.X-1
 				xFrac0, xFrac1 = 1, 0
 			}
 
-			s00ru, s00gu, s00bu, s00au := src.At(sr.Min.X+int(sx0), sr.Min.Y+int(sy0)).RGBA()
-			s00r := float64(s00ru)
-			s00g := float64(s00gu)
-			s00b := float64(s00bu)
-			s00a := float64(s00au)
-			s10ru, s10gu, s10bu, s10au := src.At(sr.Min.X+int(sx1), sr.Min.Y+int(sy0)).RGBA()
-			s10r := float64(s10ru)
-			s10g := float64(s10gu)
-			s10b := float64(s10bu)
-			s10a := float64(s10au)
-			s10r = xFrac1*s00r + xFrac0*s10r
-			s10g = xFrac1*s00g + xFrac0*s10g
-			s10b = xFrac1*s00b + xFrac0*s10b
-			s10a = xFrac1*s00a + xFrac0*s10a
-			s01ru, s01gu, s01bu, s01au := src.At(sr.Min.X+int(sx0), sr.Min.Y+int(sy1)).RGBA()
-			s01r := float64(s01ru)
-			s01g := float64(s01gu)
-			s01b := float64(s01bu)
-			s01a := float64(s01au)
-			s11ru, s11gu, s11bu, s11au := src.At(sr.Min.X+int(sx1), sr.Min.Y+int(sy1)).RGBA()
-			s11r := float64(s11ru)
-			s11g := float64(s11gu)
-			s11b := float64(s11bu)
-			s11a := float64(s11au)
-			s11r = xFrac1*s01r + xFrac0*s11r
-			s11g = xFrac1*s01g + xFrac0*s11g
-			s11b = xFrac1*s01b + xFrac0*s11b
-			s11a = xFrac1*s01a + xFrac0*s11a
-			s11r = yFrac1*s10r + yFrac0*s11r
-			s11g = yFrac1*s10g + yFrac0*s11g
-			s11b = yFrac1*s10b + yFrac0*s11b
-			s11a = yFrac1*s10a + yFrac0*s11a
-			dstColorRGBA64.R = uint16(s11r)
-			dstColorRGBA64.G = uint16(s11g)
-			dstColorRGBA64.B = uint16(s11b)
-			dstColorRGBA64.A = uint16(s11a)
-			dst.Set(dr.Min.X+int(dx), dr.Min.Y+int(dy), dstColor)
-		}
-	}
-}
-
-func (ablInterpolator) scale_Image_Image_Src(dst Image, dr, adr image.Rectangle, src image.Image, sr image.Rectangle) {
-	sw := int32(sr.Dx())
-	sh := int32(sr.Dy())
-	yscale := float64(sh) / float64(dr.Dy())
-	xscale := float64(sw) / float64(dr.Dx())
-	swMinus1, shMinus1 := sw-1, sh-1
-	dstColorRGBA64 := &color.RGBA64{}
-	dstColor := color.Color(dstColorRGBA64)
-
-	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
-		sy := (float64(dy)+0.5)*yscale - 0.5
-		// If sy < 0, we will clamp sy0 to 0 anyway, so it doesn't matter if
-		// we say int32(sy) instead of int32(math.Floor(sy)). Similarly for
-		// sx, below.
-		sy0 := int32(sy)
-		yFrac0 := sy - float64(sy0)
-		yFrac1 := 1 - yFrac0
-		sy1 := sy0 + 1
-		if sy < 0 {
-			sy0, sy1 = 0, 0
-			yFrac0, yFrac1 = 0, 1
-		} else if sy1 > shMinus1 {
-			sy0, sy1 = shMinus1, shMinus1
-			yFrac0, yFrac1 = 1, 0
-		}
-
-		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx++ {
-			sx := (float64(dx)+0.5)*xscale - 0.5
-			sx0 := int32(sx)
-			xFrac0 := sx - float64(sx0)
-			xFrac1 := 1 - xFrac0
-			sx1 := sx0 + 1
-			if sx < 0 {
-				sx0, sx1 = 0, 0
-				xFrac0, xFrac1 = 0, 1
-			} else if sx1 > swMinus1 {
-				sx0, sx1 = swMinus1, swMinus1
-				xFrac0, xFrac1 = 1, 0
+			sy -= 0.5
+			sy0 := int(sy)
+			yFrac0 := sy - float64(sy0)
+			yFrac1 := 1 - yFrac0
+			sy0 += bias.Y
+			sy1 := sy0 + 1
+			if sy0 < sr.Min.Y {
+				sy0, sy1 = sr.Min.Y, sr.Min.Y
+				yFrac0, yFrac1 = 0, 1
+			} else if sy1 >= sr.Max.Y {
+				sy0, sy1 = sr.Max.Y-1, sr.Max.Y-1
+				yFrac0, yFrac1 = 1, 0
 			}
 
-			s00ru, s00gu, s00bu, s00au := src.At(sr.Min.X+int(sx0), sr.Min.Y+int(sy0)).RGBA()
+			s00i := (sy0-src.Rect.Min.Y)*src.Stride + (sx0-src.Rect.Min.X)*4
+			s00ru := uint32(src.Pix[s00i+0]) * 0x101
+			s00gu := uint32(src.Pix[s00i+1]) * 0x101
+			s00bu := uint32(src.Pix[s00i+2]) * 0x101
+			s00au := uint32(src.Pix[s00i+3]) * 0x101
 			s00r := float64(s00ru)
 			s00g := float64(s00gu)
 			s00b := float64(s00bu)
 			s00a := float64(s00au)
-			s10ru, s10gu, s10bu, s10au := src.At(sr.Min.X+int(sx1), sr.Min.Y+int(sy0)).RGBA()
+			s10i := (sy0-src.Rect.Min.Y)*src.Stride + (sx1-src.Rect.Min.X)*4
+			s10ru := uint32(src.Pix[s10i+0]) * 0x101
+			s10gu := uint32(src.Pix[s10i+1]) * 0x101
+			s10bu := uint32(src.Pix[s10i+2]) * 0x101
+			s10au := uint32(src.Pix[s10i+3]) * 0x101
 			s10r := float64(s10ru)
 			s10g := float64(s10gu)
 			s10b := float64(s10bu)
@@ -3703,12 +2862,20 @@ func (ablInterpolator) scale_Image_Image_Src(dst Image, dr, adr image.Rectangle,
 			s10g = xFrac1*s00g + xFrac0*s10g
 			s10b = xFrac1*s00b + xFrac0*s10b
 			s10a = xFrac1*s00a + xFrac0*s10a
-			s01ru, s01gu, s01bu, s01au := src.At(sr.Min.X+int(sx0), sr.Min.Y+int(sy1)).RGBA()
+			s01i := (sy1-src.Rect.Min.Y)*src.Stride + (sx0-src.Rect.Min.X)*4
+			s01ru := uint32(src.Pix[s01i+0]) * 0x101
+			s01gu := uint32(src.Pix[s01i+1]) * 0x101
+			s01bu := uint32(src.Pix[s01i+2]) * 0x101
+			s01au := uint32(src.Pix[s01i+3]) * 0x101
 			s01r := float64(s01ru)
 			s01g := float64(s01gu)
 			s01b := float64(s01bu)
 			s01a := float64(s01au)
-			s11ru, s11gu, s11bu, s11au := src.At(sr.Min.X+int(sx1), sr.Min.Y+int(sy1)).RGBA()
+			s11i := (sy1-src.Rect.Min.Y)*src.Stride + (sx1-src.Rect.Min.X)*4
+			s11ru := uint32(src.Pix[s11i+0]) * 0x101
+			s11gu := uint32(src.Pix[s11i+1]) * 0x101
+			s11bu := uint32(src.Pix[s11i+2]) * 0x101
+			s11au := uint32(src.Pix[s11i+3]) * 0x101
 			s11r := float64(s11ru)
 			s11g := float64(s11gu)
 			s11b := float64(s11bu)
@@ -3721,16 +2888,15 @@ func (ablInterpolator) scale_Image_Image_Src(dst Image, dr, adr image.Rectangle,
 			s11g = yFrac1*s10g + yFrac0*s11g
 			s11b = yFrac1*s10b + yFrac0*s11b
 			s11a = yFrac1*s10a + yFrac0*s11a
-			dstColorRGBA64.R = uint16(s11r)
-			dstColorRGBA64.G = uint16(s11g)
-			dstColorRGBA64.B = uint16(s11b)
-			dstColorRGBA64.A = uint16(s11a)
-			dst.Set(dr.Min.X+int(dx), dr.Min.Y+int(dy), dstColor)
+			dst.Pix[d+0] = uint8(uint32(s11r) >> 8)
+			dst.Pix[d+1] = uint8(uint32(s11g) >> 8)
+			dst.Pix[d+2] = uint8(uint32(s11b) >> 8)
+			dst.Pix[d+3] = uint8(uint32(s11a) >> 8)
 		}
 	}
 }
 
-func (ablInterpolator) transform_RGBA_Gray_Over(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.Gray, sr image.Rectangle, bias image.Point) {
+func (ablInterpolator) transform_RGBA_RGBA_Src(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.RGBA, sr image.Rectangle, bias image.Point) {
 	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
 		dyf := float64(dr.Min.Y+int(dy)) + 0.5
 		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
@@ -3770,31 +2936,63 @@ func (ablInterpolator) transform_RGBA_Gray_Over(dst *image.RGBA, dr, adr image.R
 				yFrac0, yFrac1 = 1, 0
 			}
 
-			s00i := (sy0-src.Rect.Min.Y)*src.Stride + (sx0 - src.Rect.Min.X)
-			s00ru := uint32(src.Pix[s00i]) * 0x101
+			s00i := (sy0-src.Rect.Min.Y)*src.Stride + (sx0-src.Rect.Min.X)*4
+			s00ru := uint32(src.Pix[s00i+0]) * 0x101
+			s00gu := uint32(src.Pix[s00i+1]) * 0x101
+			s00bu := uint32(src.Pix[s00i+2]) * 0x101
+			s00au := uint32(src.Pix[s00i+3]) * 0x101
 			s00r := float64(s00ru)
-			s10i := (sy0-src.Rect.Min.Y)*src.Stride + (sx1 - src.Rect.Min.X)
-			s10ru := uint32(src.Pix[s10i]) * 0x101
+			s00g := float64(s00gu)
+			s00b := float64(s00bu)
+			s00a := float64(s00au)
+			s10i := (sy0-src.Rect.Min.Y)*src.Stride + (sx1-src.Rect.Min.X)*4
+			s10ru := uint32(src.Pix[s10i+0]) * 0x101
+			s10gu := uint32(src.Pix[s10i+1]) * 0x101
+			s10bu := uint32(src.Pix[s10i+2]) * 0x101
+			s10au := uint32(src.Pix[s10i+3]) * 0x101
 			s10r := float64(s10ru)
+			s10g := float64(s10gu)
+			s10b := float64(s10bu)
+			s10a := float64(s10au)
 			s10r = xFrac1*s00r + xFrac0*s10r
-			s01i := (sy1-src.Rect.Min.Y)*src.Stride + (sx0 - src.Rect.Min.X)
-			s01ru := uint32(src.Pix[s01i]) * 0x101
+			s10g = xFrac1*s00g + xFrac0*s10g
+			s10b = xFrac1*s00b + xFrac0*s10b
+			s10a = xFrac1*s00a + xFrac0*s10a
+			s01i := (sy1-src.Rect.Min.Y)*src.Stride + (sx0-src.Rect.Min.X)*4
+			s01ru := uint32(src.Pix[s01i+0]) * 0x101
+			s01gu := uint32(src.Pix[s01i+1]) * 0x101
+			s01bu := uint32(src.Pix[s01i+2]) * 0x101
+			s01au := uint32(src.Pix[s01i+3]) * 0x101
 			s01r := float64(s01ru)
-			s11i := (sy1-src.Rect.Min.Y)*src.Stride + (sx1 - src.Rect.Min.X)
-			s11ru := uint32(src.Pix[s11i]) * 0x101
+			s01g := float64(s01gu)
+			s01b := float64(s01bu)
+			s01a := float64(s01au)
+			s11i := (sy1-src.Rect.Min.Y)*src.Stride + (sx1-src.Rect.Min.X)*4
+			s11ru := uint32(src.Pix[s11i+0]) * 0x101
+			s11gu := uint32(src.Pix[s11i+1]) * 0x101
+			s11bu := uint32(src.Pix[s11i+2]) * 0x101
+			s11au := uint32(src.Pix[s11i+3]) * 0x101
 			s11r := float64(s11ru)
+			s11g := float64(s11gu)
+			s11b := float64(s11bu)
+			s11a := float64(s11au)
 			s11r = xFrac1*s01r + xFrac0*s11r
+			s11g = xFrac1*s01g + xFrac0*s11g
+			s11b = xFrac1*s01b + xFrac0*s11b
+			s11a = xFrac1*s01a + xFrac0*s11a
 			s11r = yFrac1*s10r + yFrac0*s11r
-			out := uint8(uint32(s11r) >> 8)
-			dst.Pix[d+0] = out
-			dst.Pix[d+1] = out
-			dst.Pix[d+2] = out
-			dst.Pix[d+3] = 0xff
+			s11g = yFrac1*s10g + yFrac0*s11g
+			s11b = yFrac1*s10b + yFrac0*s11b
+			s11a = yFrac1*s10a + yFrac0*s11a
+			dst.Pix[d+0] = uint8(uint32(s11r) >> 8)
+			dst.Pix[d+1] = uint8(uint32(s11g) >> 8)
+			dst.Pix[d+2] = uint8(uint32(s11b) >> 8)
+			dst.Pix[d+3] = uint8(uint32(s11a) >> 8)
 		}
 	}
 }
 
-func (ablInterpolator) transform_RGBA_Gray_Src(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.Gray, sr image.Rectangle, bias image.Point) {
+func (ablInterpolator) transform_RGBA_YCbCr444_Src(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.YCbCr, sr image.Rectangle, bias image.Point) {
 	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
 		dyf := float64(dr.Min.Y+int(dy)) + 0.5
 		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
@@ -3834,127 +3032,140 @@ func (ablInterpolator) transform_RGBA_Gray_Src(dst *image.RGBA, dr, adr image.Re
 				yFrac0, yFrac1 = 1, 0
 			}
 
-			s00i := (sy0-src.Rect.Min.Y)*src.Stride + (sx0 - src.Rect.Min.X)
-			s00ru := uint32(src.Pix[s00i]) * 0x101
-			s00r := float64(s00ru)
-			s10i := (sy0-src.Rect.Min.Y)*src.Stride + (sx1 - src.Rect.Min.X)
-			s10ru := uint32(src.Pix[s10i]) * 0x101
-			s10r := float64(s10ru)
-			s10r = xFrac1*s00r + xFrac0*s10r
-			s01i := (sy1-src.Rect.Min.Y)*src.Stride + (sx0 - src.Rect.Min.X)
-			s01ru := uint32(src.Pix[s01i]) * 0x101
-			s01r := float64(s01ru)
-			s11i := (sy1-src.Rect.Min.Y)*src.Stride + (sx1 - src.Rect.Min.X)
-			s11ru := uint32(src.Pix[s11i]) * 0x101
-			s11r := float64(s11ru)
-			s11r = xFrac1*s01r + xFrac0*s11r
-			s11r = yFrac1*s10r + yFrac0*s11r
-			out := uint8(uint32(s11r) >> 8)
-			dst.Pix[d+0] = out
-			dst.Pix[d+1] = out
-			dst.Pix[d+2] = out
-			dst.Pix[d+3] = 0xff
-		}
-	}
-}
+			s00i := (sy0-src.Rect.Min.Y)*src.YStride + (sx0 - src.Rect.Min.X)
+			s00j := (sy0-src.Rect.Min.Y)*src.CStride + (sx0 - src.Rect.Min.X)
 
-func (ablInterpolator) transform_RGBA_NRGBA_Over(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.NRGBA, sr image.Rectangle, bias image.Point) {
-	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
-		dyf := float64(dr.Min.Y+int(dy)) + 0.5
-		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
-		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx, d = dx+1, d+4 {
-			dxf := float64(dr.Min.X+int(dx)) + 0.5
-			sx := d2s[0]*dxf + d2s[1]*dyf + d2s[2]
-			sy := d2s[3]*dxf + d2s[4]*dyf + d2s[5]
-			if !(image.Point{int(sx) + bias.X, int(sy) + bias.Y}).In(sr) {
-				continue
+			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
+			s00yy1 := int(src.Y[s00i])<<16 + 1<<15
+			s00cb1 := int(src.Cb[s00j]) - 128
+			s00cr1 := int(src.Cr[s00j]) - 128
+			s00ru := (s00yy1 + 91881*s00cr1) >> 8
+			s00gu := (s00yy1 - 22554*s00cb1 - 46802*s00cr1) >> 8
+			s00bu := (s00yy1 + 116130*s00cb1) >> 8
+			if s00ru < 0 {
+				s00ru = 0
+			} else if s00ru > 0xffff {
+				s00ru = 0xffff
+			}
+			if s00gu < 0 {
+				s00gu = 0
+			} else if s00gu > 0xffff {
+				s00gu = 0xffff
+			}
+			if s00bu < 0 {
+				s00bu = 0
+			} else if s00bu > 0xffff {
+				s00bu = 0xffff
 			}
 
-			sx -= 0.5
-			sx0 := int(sx)
-			xFrac0 := sx - float64(sx0)
-			xFrac1 := 1 - xFrac0
-			sx0 += bias.X
-			sx1 := sx0 + 1
-			if sx0 < sr.Min.X {
-				sx0, sx1 = sr.Min.X, sr.Min.X
-				xFrac0, xFrac1 = 0, 1
-			} else if sx1 >= sr.Max.X {
-				sx0, sx1 = sr.Max.X-1, sr.Max.X-1
-				xFrac0, xFrac1 = 1, 0
-			}
-
-			sy -= 0.5
-			sy0 := int(sy)
-			yFrac0 := sy - float64(sy0)
-			yFrac1 := 1 - yFrac0
-			sy0 += bias.Y
-			sy1 := sy0 + 1
-			if sy0 < sr.Min.Y {
-				sy0, sy1 = sr.Min.Y, sr.Min.Y
-				yFrac0, yFrac1 = 0, 1
-			} else if sy1 >= sr.Max.Y {
-				sy0, sy1 = sr.Max.Y-1, sr.Max.Y-1
-				yFrac0, yFrac1 = 1, 0
-			}
-
-			s00i := (sy0-src.Rect.Min.Y)*src.Stride + (sx0-src.Rect.Min.X)*4
-			s00au := uint32(src.Pix[s00i+3]) * 0x101
-			s00ru := uint32(src.Pix[s00i+0]) * s00au / 0xff
-			s00gu := uint32(src.Pix[s00i+1]) * s00au / 0xff
-			s00bu := uint32(src.Pix[s00i+2]) * s00au / 0xff
 			s00r := float64(s00ru)
 			s00g := float64(s00gu)
 			s00b := float64(s00bu)
-			s00a := float64(s00au)
-			s10i := (sy0-src.Rect.Min.Y)*src.Stride + (sx1-src.Rect.Min.X)*4
-			s10au := uint32(src.Pix[s10i+3]) * 0x101
-			s10ru := uint32(src.Pix[s10i+0]) * s10au / 0xff
-			s10gu := uint32(src.Pix[s10i+1]) * s10au / 0xff
-			s10bu := uint32(src.Pix[s10i+2]) * s10au / 0xff
+			s10i := (sy0-src.Rect.Min.Y)*src.YStride + (sx1 - src.Rect.Min.X)
+			s10j := (sy0-src.Rect.Min.Y)*src.CStride + (sx1 - src.Rect.Min.X)
+
+			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
+			s10yy1 := int(src.Y[s10i])<<16 + 1<<15
+			s10cb1 := int(src.Cb[s10j]) - 128
+			s10cr1 := int(src.Cr[s10j]) - 128
+			s10ru := (s10yy1 + 91881*s10cr1) >> 8
+			s10gu := (s10yy1 - 22554*s10cb1 - 46802*s10cr1) >> 8
+			s10bu := (s10yy1 + 116130*s10cb1) >> 8
+			if s10ru < 0 {
+				s10ru = 0
+			} else if s10ru > 0xffff {
+				s10ru = 0xffff
+			}
+			if s10gu < 0 {
+				s10gu = 0
+			} else if s10gu > 0xffff {
+				s10gu = 0xffff
+			}
+			if s10bu < 0 {
+				s10bu = 0
+			} else if s10bu > 0xffff {
+				s10bu = 0xffff
+			}
+
 			s10r := float64(s10ru)
 			s10g := float64(s10gu)
 			s10b := float64(s10bu)
-			s10a := float64(s10au)
 			s10r = xFrac1*s00r + xFrac0*s10r
 			s10g = xFrac1*s00g + xFrac0*s10g
 			s10b = xFrac1*s00b + xFrac0*s10b
-			s10a = xFrac1*s00a + xFrac0*s10a
-			s01i := (sy1-src.Rect.Min.Y)*src.Stride + (sx0-src.Rect.Min.X)*4
-			s01au := uint32(src.Pix[s01i+3]) * 0x101
-			s01ru := uint32(src.Pix[s01i+0]) * s01au / 0xff
-			s01gu := uint32(src.Pix[s01i+1]) * s01au / 0xff
-			s01bu := uint32(src.Pix[s01i+2]) * s01au / 0xff
+			s01i := (sy1-src.Rect.Min.Y)*src.YStride + (sx0 - src.Rect.Min.X)
+			s01j := (sy1-src.Rect.Min.Y)*src.CStride + (sx0 - src.Rect.Min.X)
+
+			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
+			s01yy1 := int(src.Y[s01i])<<16 + 1<<15
+			s01cb1 := int(src.Cb[s01j]) - 128
+			s01cr1 := int(src.Cr[s01j]) - 128
+			s01ru := (s01yy1 + 91881*s01cr1) >> 8
+			s01gu := (s01yy1 - 22554*s01cb1 - 46802*s01cr1) >> 8
+			s01bu := (s01yy1 + 116130*s01cb1) >> 8
+			if s01ru < 0 {
+				s01ru = 0
+			} else if s01ru > 0xffff {
+				s01ru = 0xffff
+			}
+			if s01gu < 0 {
+				s01gu = 0
+			} else if s01gu > 0xffff {
+				s01gu = 0xffff
+			}
+			if s01bu < 0 {
+				s01bu = 0
+			} else if s01bu > 0xffff {
+				s01bu = 0xffff
+			}
+
 			s01r := float64(s01ru)
 			s01g := float64(s01gu)
 			s01b := float64(s01bu)
-			s01a := float64(s01au)
-			s11i := (sy1-src.Rect.Min.Y)*src.Stride + (sx1-src.Rect.Min.X)*4
-			s11au := uint32(src.Pix[s11i+3]) * 0x101
-			s11ru := uint32(src.Pix[s11i+0]) * s11au / 0xff
-			s11gu := uint32(src.Pix[s11i+1]) * s11au / 0xff
-			s11bu := uint32(src.Pix[s11i+2]) * s11au / 0xff
+			s11i := (sy1-src.Rect.Min.Y)*src.YStride + (sx1 - src.Rect.Min.X)
+			s11j := (sy1-src.Rect.Min.Y)*src.CStride + (sx1 - src.Rect.Min.X)
+
+			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
+			s11yy1 := int(src.Y[s11i])<<16 + 1<<15
+			s11cb1 := int(src.Cb[s11j]) - 128
+			s11cr1 := int(src.Cr[s11j]) - 128
+			s11ru := (s11yy1 + 91881*s11cr1) >> 8
+			s11gu := (s11yy1 - 22554*s11cb1 - 46802*s11cr1) >> 8
+			s11bu := (s11yy1 + 116130*s11cb1) >> 8
+			if s11ru < 0 {
+				s11ru = 0
+			} else if s11ru > 0xffff {
+				s11ru = 0xffff
+			}
+			if s11gu < 0 {
+				s11gu = 0
+			} else if s11gu > 0xffff {
+				s11gu = 0xffff
+			}
+			if s11bu < 0 {
+				s11bu = 0
+			} else if s11bu > 0xffff {
+				s11bu = 0xffff
+			}
+
 			s11r := float64(s11ru)
 			s11g := float64(s11gu)
 			s11b := float64(s11bu)
-			s11a := float64(s11au)
 			s11r = xFrac1*s01r + xFrac0*s11r
 			s11g = xFrac1*s01g + xFrac0*s11g
 			s11b = xFrac1*s01b + xFrac0*s11b
-			s11a = xFrac1*s01a + xFrac0*s11a
 			s11r = yFrac1*s10r + yFrac0*s11r
 			s11g = yFrac1*s10g + yFrac0*s11g
 			s11b = yFrac1*s10b + yFrac0*s11b
-			s11a = yFrac1*s10a + yFrac0*s11a
 			dst.Pix[d+0] = uint8(uint32(s11r) >> 8)
 			dst.Pix[d+1] = uint8(uint32(s11g) >> 8)
 			dst.Pix[d+2] = uint8(uint32(s11b) >> 8)
-			dst.Pix[d+3] = uint8(uint32(s11a) >> 8)
+			dst.Pix[d+3] = 0xff
 		}
 	}
 }
 
-func (ablInterpolator) transform_RGBA_NRGBA_Src(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.NRGBA, sr image.Rectangle, bias image.Point) {
+func (ablInterpolator) transform_RGBA_YCbCr422_Src(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.YCbCr, sr image.Rectangle, bias image.Point) {
 	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
 		dyf := float64(dr.Min.Y+int(dy)) + 0.5
 		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
@@ -3994,255 +3205,140 @@ func (ablInterpolator) transform_RGBA_NRGBA_Src(dst *image.RGBA, dr, adr image.R
 				yFrac0, yFrac1 = 1, 0
 			}
 
-			s00i := (sy0-src.Rect.Min.Y)*src.Stride + (sx0-src.Rect.Min.X)*4
-			s00au := uint32(src.Pix[s00i+3]) * 0x101
-			s00ru := uint32(src.Pix[s00i+0]) * s00au / 0xff
-			s00gu := uint32(src.Pix[s00i+1]) * s00au / 0xff
-			s00bu := uint32(src.Pix[s00i+2]) * s00au / 0xff
+			s00i := (sy0-src.Rect.Min.Y)*src.YStride + (sx0 - src.Rect.Min.X)
+			s00j := (sy0-src.Rect.Min.Y)*src.CStride + ((sx0)/2 - src.Rect.Min.X/2)
+
+			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
+			s00yy1 := int(src.Y[s00i])<<16 + 1<<15
+			s00cb1 := int(src.Cb[s00j]) - 128
+			s00cr1 := int(src.Cr[s00j]) - 128
+			s00ru := (s00yy1 + 91881*s00cr1) >> 8
+			s00gu := (s00yy1 - 22554*s00cb1 - 46802*s00cr1) >> 8
+			s00bu := (s00yy1 + 116130*s00cb1) >> 8
+			if s00ru < 0 {
+				s00ru = 0
+			} else if s00ru > 0xffff {
+				s00ru = 0xffff
+			}
+			if s00gu < 0 {
+				s00gu = 0
+			} else if s00gu > 0xffff {
+				s00gu = 0xffff
+			}
+			if s00bu < 0 {
+				s00bu = 0
+			} else if s00bu > 0xffff {
+				s00bu = 0xffff
+			}
+
 			s00r := float64(s00ru)
 			s00g := float64(s00gu)
 			s00b := float64(s00bu)
-			s00a := float64(s00au)
-			s10i := (sy0-src.Rect.Min.Y)*src.Stride + (sx1-src.Rect.Min.X)*4
-			s10au := uint32(src.Pix[s10i+3]) * 0x101
-			s10ru := uint32(src.Pix[s10i+0]) * s10au / 0xff
-			s10gu := uint32(src.Pix[s10i+1]) * s10au / 0xff
-			s10bu := uint32(src.Pix[s10i+2]) * s10au / 0xff
+			s10i := (sy0-src.Rect.Min.Y)*src.YStride + (sx1 - src.Rect.Min.X)
+			s10j := (sy0-src.Rect.Min.Y)*src.CStride + ((sx1)/2 - src.Rect.Min.X/2)
+
+			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
+			s10yy1 := int(src.Y[s10i])<<16 + 1<<15
+			s10cb1 := int(src.Cb[s10j]) - 128
+			s10cr1 := int(src.Cr[s10j]) - 128
+			s10ru := (s10yy1 + 91881*s10cr1) >> 8
+			s10gu := (s10yy1 - 22554*s10cb1 - 46802*s10cr1) >> 8
+			s10bu := (s10yy1 + 116130*s10cb1) >> 8
+			if s10ru < 0 {
+				s10ru = 0
+			} else if s10ru > 0xffff {
+				s10ru = 0xffff
+			}
+			if s10gu < 0 {
+				s10gu = 0
+			} else if s10gu > 0xffff {
+				s10gu = 0xffff
+			}
+			if s10bu < 0 {
+				s10bu = 0
+			} else if s10bu > 0xffff {
+				s10bu = 0xffff
+			}
+
 			s10r := float64(s10ru)
 			s10g := float64(s10gu)
 			s10b := float64(s10bu)
-			s10a := float64(s10au)
 			s10r = xFrac1*s00r + xFrac0*s10r
 			s10g = xFrac1*s00g + xFrac0*s10g
 			s10b = xFrac1*s00b + xFrac0*s10b
-			s10a = xFrac1*s00a + xFrac0*s10a
-			s01i := (sy1-src.Rect.Min.Y)*src.Stride + (sx0-src.Rect.Min.X)*4
-			s01au := uint32(src.Pix[s01i+3]) * 0x101
-			s01ru := uint32(src.Pix[s01i+0]) * s01au / 0xff
-			s01gu := uint32(src.Pix[s01i+1]) * s01au / 0xff
-			s01bu := uint32(src.Pix[s01i+2]) * s01au / 0xff
+			s01i := (sy1-src.Rect.Min.Y)*src.YStride + (sx0 - src.Rect.Min.X)
+			s01j := (sy1-src.Rect.Min.Y)*src.CStride + ((sx0)/2 - src.Rect.Min.X/2)
+
+			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
+			s01yy1 := int(src.Y[s01i])<<16 + 1<<15
+			s01cb1 := int(src.Cb[s01j]) - 128
+			s01cr1 := int(src.Cr[s01j]) - 128
+			s01ru := (s01yy1 + 91881*s01cr1) >> 8
+			s01gu := (s01yy1 - 22554*s01cb1 - 46802*s01cr1) >> 8
+			s01bu := (s01yy1 + 116130*s01cb1) >> 8
+			if s01ru < 0 {
+				s01ru = 0
+			} else if s01ru > 0xffff {
+				s01ru = 0xffff
+			}
+			if s01gu < 0 {
+				s01gu = 0
+			} else if s01gu > 0xffff {
+				s01gu = 0xffff
+			}
+			if s01bu < 0 {
+				s01bu = 0
+			} else if s01bu > 0xffff {
+				s01bu = 0xffff
+			}
+
 			s01r := float64(s01ru)
 			s01g := float64(s01gu)
 			s01b := float64(s01bu)
-			s01a := float64(s01au)
-			s11i := (sy1-src.Rect.Min.Y)*src.Stride + (sx1-src.Rect.Min.X)*4
-			s11au := uint32(src.Pix[s11i+3]) * 0x101
-			s11ru := uint32(src.Pix[s11i+0]) * s11au / 0xff
-			s11gu := uint32(src.Pix[s11i+1]) * s11au / 0xff
-			s11bu := uint32(src.Pix[s11i+2]) * s11au / 0xff
-			s11r := float64(s11ru)
-			s11g := float64(s11gu)
-			s11b := float64(s11bu)
-			s11a := float64(s11au)
-			s11r = xFrac1*s01r + xFrac0*s11r
-			s11g = xFrac1*s01g + xFrac0*s11g
-			s11b = xFrac1*s01b + xFrac0*s11b
-			s11a = xFrac1*s01a + xFrac0*s11a
-			s11r = yFrac1*s10r + yFrac0*s11r
-			s11g = yFrac1*s10g + yFrac0*s11g
-			s11b = yFrac1*s10b + yFrac0*s11b
-			s11a = yFrac1*s10a + yFrac0*s11a
-			dst.Pix[d+0] = uint8(uint32(s11r) >> 8)
-			dst.Pix[d+1] = uint8(uint32(s11g) >> 8)
-			dst.Pix[d+2] = uint8(uint32(s11b) >> 8)
-			dst.Pix[d+3] = uint8(uint32(s11a) >> 8)
-		}
-	}
-}
+			s11i := (sy1-src.Rect.Min.Y)*src.YStride + (sx1 - src.Rect.Min.X)
+			s11j := (sy1-src.Rect.Min.Y)*src.CStride + ((sx1)/2 - src.Rect.Min.X/2)
 
-func (ablInterpolator) transform_RGBA_RGBA_Over(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.RGBA, sr image.Rectangle, bias image.Point) {
-	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
-		dyf := float64(dr.Min.Y+int(dy)) + 0.5
-		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
-		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx, d = dx+1, d+4 {
-			dxf := float64(dr.Min.X+int(dx)) + 0.5
-			sx := d2s[0]*dxf + d2s[1]*dyf + d2s[2]
-			sy := d2s[3]*dxf + d2s[4]*dyf + d2s[5]
-			if !(image.Point{int(sx) + bias.X, int(sy) + bias.Y}).In(sr) {
-				continue
-			}
-
-			sx -= 0.5
-			sx0 := int(sx)
-			xFrac0 := sx - float64(sx0)
-			xFrac1 := 1 - xFrac0
-			sx0 += bias.X
-			sx1 := sx0 + 1
-			if sx0 < sr.Min.X {
-				sx0, sx1 = sr.Min.X, sr.Min.X
-				xFrac0, xFrac1 = 0, 1
-			} else if sx1 >= sr.Max.X {
-				sx0, sx1 = sr.Max.X-1, sr.Max.X-1
-				xFrac0, xFrac1 = 1, 0
-			}
-
-			sy -= 0.5
-			sy0 := int(sy)
-			yFrac0 := sy - float64(sy0)
-			yFrac1 := 1 - yFrac0
-			sy0 += bias.Y
-			sy1 := sy0 + 1
-			if sy0 < sr.Min.Y {
-				sy0, sy1 = sr.Min.Y, sr.Min.Y
-				yFrac0, yFrac1 = 0, 1
-			} else if sy1 >= sr.Max.Y {
-				sy0, sy1 = sr.Max.Y-1, sr.Max.Y-1
-				yFrac0, yFrac1 = 1, 0
-			}
-
-			s00i := (sy0-src.Rect.Min.Y)*src.Stride + (sx0-src.Rect.Min.X)*4
-			s00ru := uint32(src.Pix[s00i+0]) * 0x101
-			s00gu := uint32(src.Pix[s00i+1]) * 0x101
-			s00bu := uint32(src.Pix[s00i+2]) * 0x101
-			s00au := uint32(src.Pix[s00i+3]) * 0x101
-			s00r := float64(s00ru)
-			s00g := float64(s00gu)
-			s00b := float64(s00bu)
-			s00a := float64(s00au)
-			s10i := (sy0-src.Rect.Min.Y)*src.Stride + (sx1-src.Rect.Min.X)*4
-			s10ru := uint32(src.Pix[s10i+0]) * 0x101
-			s10gu := uint32(src.Pix[s10i+1]) * 0x101
-			s10bu := uint32(src.Pix[s10i+2]) * 0x101
-			s10au := uint32(src.Pix[s10i+3]) * 0x101
-			s10r := float64(s10ru)
-			s10g := float64(s10gu)
-			s10b := float64(s10bu)
-			s10a := float64(s10au)
-			s10r = xFrac1*s00r + xFrac0*s10r
-			s10g = xFrac1*s00g + xFrac0*s10g
-			s10b = xFrac1*s00b + xFrac0*s10b
-			s10a = xFrac1*s00a + xFrac0*s10a
-			s01i := (sy1-src.Rect.Min.Y)*src.Stride + (sx0-src.Rect.Min.X)*4
-			s01ru := uint32(src.Pix[s01i+0]) * 0x101
-			s01gu := uint32(src.Pix[s01i+1]) * 0x101
-			s01bu := uint32(src.Pix[s01i+2]) * 0x101
-			s01au := uint32(src.Pix[s01i+3]) * 0x101
-			s01r := float64(s01ru)
-			s01g := float64(s01gu)
-			s01b := float64(s01bu)
-			s01a := float64(s01au)
-			s11i := (sy1-src.Rect.Min.Y)*src.Stride + (sx1-src.Rect.Min.X)*4
-			s11ru := uint32(src.Pix[s11i+0]) * 0x101
-			s11gu := uint32(src.Pix[s11i+1]) * 0x101
-			s11bu := uint32(src.Pix[s11i+2]) * 0x101
-			s11au := uint32(src.Pix[s11i+3]) * 0x101
-			s11r := float64(s11ru)
-			s11g := float64(s11gu)
-			s11b := float64(s11bu)
-			s11a := float64(s11au)
-			s11r = xFrac1*s01r + xFrac0*s11r
-			s11g = xFrac1*s01g + xFrac0*s11g
-			s11b = xFrac1*s01b + xFrac0*s11b
-			s11a = xFrac1*s01a + xFrac0*s11a
-			s11r = yFrac1*s10r + yFrac0*s11r
-			s11g = yFrac1*s10g + yFrac0*s11g
-			s11b = yFrac1*s10b + yFrac0*s11b
-			s11a = yFrac1*s10a + yFrac0*s11a
-			dst.Pix[d+0] = uint8(uint32(s11r) >> 8)
-			dst.Pix[d+1] = uint8(uint32(s11g) >> 8)
-			dst.Pix[d+2] = uint8(uint32(s11b) >> 8)
-			dst.Pix[d+3] = uint8(uint32(s11a) >> 8)
-		}
-	}
-}
-
-func (ablInterpolator) transform_RGBA_RGBA_Src(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.RGBA, sr image.Rectangle, bias image.Point) {
-	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
-		dyf := float64(dr.Min.Y+int(dy)) + 0.5
-		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
-		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx, d = dx+1, d+4 {
-			dxf := float64(dr.Min.X+int(dx)) + 0.5
-			sx := d2s[0]*dxf + d2s[1]*dyf + d2s[2]
-			sy := d2s[3]*dxf + d2s[4]*dyf + d2s[5]
-			if !(image.Point{int(sx) + bias.X, int(sy) + bias.Y}).In(sr) {
-				continue
+			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
+			s11yy1 := int(src.Y[s11i])<<16 + 1<<15
+			s11cb1 := int(src.Cb[s11j]) - 128
+			s11cr1 := int(src.Cr[s11j]) - 128
+			s11ru := (s11yy1 + 91881*s11cr1) >> 8
+			s11gu := (s11yy1 - 22554*s11cb1 - 46802*s11cr1) >> 8
+			s11bu := (s11yy1 + 116130*s11cb1) >> 8
+			if s11ru < 0 {
+				s11ru = 0
+			} else if s11ru > 0xffff {
+				s11ru = 0xffff
 			}
-
-			sx -= 0.5
-			sx0 := int(sx)
-			xFrac0 := sx - float64(sx0)
-			xFrac1 := 1 - xFrac0
-			sx0 += bias.X
-			sx1 := sx0 + 1
-			if sx0 < sr.Min.X {
-				sx0, sx1 = sr.Min.X, sr.Min.X
-				xFrac0, xFrac1 = 0, 1
-			} else if sx1 >= sr.Max.X {
-				sx0, sx1 = sr.Max.X-1, sr.Max.X-1
-				xFrac0, xFrac1 = 1, 0
+			if s11gu < 0 {
+				s11gu = 0
+			} else if s11gu > 0xffff {
+				s11gu = 0xffff
 			}
-
-			sy -= 0.5
-			sy0 := int(sy)
-			yFrac0 := sy - float64(sy0)
-			yFrac1 := 1 - yFrac0
-			sy0 += bias.Y
-			sy1 := sy0 + 1
-			if sy0 < sr.Min.Y {
-				sy0, sy1 = sr.Min.Y, sr.Min.Y
-				yFrac0, yFrac1 = 0, 1
-			} else if sy1 >= sr.Max.Y {
-				sy0, sy1 = sr.Max.Y-1, sr.Max.Y-1
-				yFrac0, yFrac1 = 1, 0
+			if s11bu < 0 {
+				s11bu = 0
+			} else if s11bu > 0xffff {
+				s11bu = 0xffff
 			}
 
-			s00i := (sy0-src.Rect.Min.Y)*src.Stride + (sx0-src.Rect.Min.X)*4
-			s00ru := uint32(src.Pix[s00i+0]) * 0x101
-			s00gu := uint32(src.Pix[s00i+1]) * 0x101
-			s00bu := uint32(src.Pix[s00i+2]) * 0x101
-			s00au := uint32(src.Pix[s00i+3]) * 0x101
-			s00r := float64(s00ru)
-			s00g := float64(s00gu)
-			s00b := float64(s00bu)
-			s00a := float64(s00au)
-			s10i := (sy0-src.Rect.Min.Y)*src.Stride + (sx1-src.Rect.Min.X)*4
-			s10ru := uint32(src.Pix[s10i+0]) * 0x101
-			s10gu := uint32(src.Pix[s10i+1]) * 0x101
-			s10bu := uint32(src.Pix[s10i+2]) * 0x101
-			s10au := uint32(src.Pix[s10i+3]) * 0x101
-			s10r := float64(s10ru)
-			s10g := float64(s10gu)
-			s10b := float64(s10bu)
-			s10a := float64(s10au)
-			s10r = xFrac1*s00r + xFrac0*s10r
-			s10g = xFrac1*s00g + xFrac0*s10g
-			s10b = xFrac1*s00b + xFrac0*s10b
-			s10a = xFrac1*s00a + xFrac0*s10a
-			s01i := (sy1-src.Rect.Min.Y)*src.Stride + (sx0-src.Rect.Min.X)*4
-			s01ru := uint32(src.Pix[s01i+0]) * 0x101
-			s01gu := uint32(src.Pix[s01i+1]) * 0x101
-			s01bu := uint32(src.Pix[s01i+2]) * 0x101
-			s01au := uint32(src.Pix[s01i+3]) * 0x101
-			s01r := float64(s01ru)
-			s01g := float64(s01gu)
-			s01b := float64(s01bu)
-			s01a := float64(s01au)
-			s11i := (sy1-src.Rect.Min.Y)*src.Stride + (sx1-src.Rect.Min.X)*4
-			s11ru := uint32(src.Pix[s11i+0]) * 0x101
-			s11gu := uint32(src.Pix[s11i+1]) * 0x101
-			s11bu := uint32(src.Pix[s11i+2]) * 0x101
-			s11au := uint32(src.Pix[s11i+3]) * 0x101
 			s11r := float64(s11ru)
 			s11g := float64(s11gu)
 			s11b := float64(s11bu)
-			s11a := float64(s11au)
 			s11r = xFrac1*s01r + xFrac0*s11r
 			s11g = xFrac1*s01g + xFrac0*s11g
 			s11b = xFrac1*s01b + xFrac0*s11b
-			s11a = xFrac1*s01a + xFrac0*s11a
 			s11r = yFrac1*s10r + yFrac0*s11r
 			s11g = yFrac1*s10g + yFrac0*s11g
 			s11b = yFrac1*s10b + yFrac0*s11b
-			s11a = yFrac1*s10a + yFrac0*s11a
 			dst.Pix[d+0] = uint8(uint32(s11r) >> 8)
 			dst.Pix[d+1] = uint8(uint32(s11g) >> 8)
 			dst.Pix[d+2] = uint8(uint32(s11b) >> 8)
-			dst.Pix[d+3] = uint8(uint32(s11a) >> 8)
+			dst.Pix[d+3] = 0xff
 		}
 	}
 }
 
-func (ablInterpolator) transform_RGBA_YCbCr444_Over(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.YCbCr, sr image.Rectangle, bias image.Point) {
+func (ablInterpolator) transform_RGBA_YCbCr420_Src(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.YCbCr, sr image.Rectangle, bias image.Point) {
 	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
 		dyf := float64(dr.Min.Y+int(dy)) + 0.5
 		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
@@ -4283,7 +3379,7 @@ func (ablInterpolator) transform_RGBA_YCbCr444_Over(dst *image.RGBA, dr, adr ima
 			}
 
 			s00i := (sy0-src.Rect.Min.Y)*src.YStride + (sx0 - src.Rect.Min.X)
-			s00j := (sy0-src.Rect.Min.Y)*src.CStride + (sx0 - src.Rect.Min.X)
+			s00j := ((sy0)/2-src.Rect.Min.Y/2)*src.CStride + ((sx0)/2 - src.Rect.Min.X/2)
 
 			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
 			s00yy1 := int(src.Y[s00i])<<16 + 1<<15
@@ -4312,7 +3408,7 @@ func (ablInterpolator) transform_RGBA_YCbCr444_Over(dst *image.RGBA, dr, adr ima
 			s00g := float64(s00gu)
 			s00b := float64(s00bu)
 			s10i := (sy0-src.Rect.Min.Y)*src.YStride + (sx1 - src.Rect.Min.X)
-			s10j := (sy0-src.Rect.Min.Y)*src.CStride + (sx1 - src.Rect.Min.X)
+			s10j := ((sy0)/2-src.Rect.Min.Y/2)*src.CStride + ((sx1)/2 - src.Rect.Min.X/2)
 
 			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
 			s10yy1 := int(src.Y[s10i])<<16 + 1<<15
@@ -4344,7 +3440,7 @@ func (ablInterpolator) transform_RGBA_YCbCr444_Over(dst *image.RGBA, dr, adr ima
 			s10g = xFrac1*s00g + xFrac0*s10g
 			s10b = xFrac1*s00b + xFrac0*s10b
 			s01i := (sy1-src.Rect.Min.Y)*src.YStride + (sx0 - src.Rect.Min.X)
-			s01j := (sy1-src.Rect.Min.Y)*src.CStride + (sx0 - src.Rect.Min.X)
+			s01j := ((sy1)/2-src.Rect.Min.Y/2)*src.CStride + ((sx0)/2 - src.Rect.Min.X/2)
 
 			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
 			s01yy1 := int(src.Y[s01i])<<16 + 1<<15
@@ -4373,7 +3469,7 @@ func (ablInterpolator) transform_RGBA_YCbCr444_Over(dst *image.RGBA, dr, adr ima
 			s01g := float64(s01gu)
 			s01b := float64(s01bu)
 			s11i := (sy1-src.Rect.Min.Y)*src.YStride + (sx1 - src.Rect.Min.X)
-			s11j := (sy1-src.Rect.Min.Y)*src.CStride + (sx1 - src.Rect.Min.X)
+			s11j := ((sy1)/2-src.Rect.Min.Y/2)*src.CStride + ((sx1)/2 - src.Rect.Min.X/2)
 
 			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
 			s11yy1 := int(src.Y[s11i])<<16 + 1<<15
@@ -4415,7 +3511,7 @@ func (ablInterpolator) transform_RGBA_YCbCr444_Over(dst *image.RGBA, dr, adr ima
 	}
 }
 
-func (ablInterpolator) transform_RGBA_YCbCr422_Over(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.YCbCr, sr image.Rectangle, bias image.Point) {
+func (ablInterpolator) transform_RGBA_YCbCr440_Src(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.YCbCr, sr image.Rectangle, bias image.Point) {
 	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
 		dyf := float64(dr.Min.Y+int(dy)) + 0.5
 		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
@@ -4456,7 +3552,7 @@ func (ablInterpolator) transform_RGBA_YCbCr422_Over(dst *image.RGBA, dr, adr ima
 			}
 
 			s00i := (sy0-src.Rect.Min.Y)*src.YStride + (sx0 - src.Rect.Min.X)
-			s00j := (sy0-src.Rect.Min.Y)*src.CStride + ((sx0)/2 - src.Rect.Min.X/2)
+			s00j := ((sy0)/2-src.Rect.Min.Y/2)*src.CStride + (sx0 - src.Rect.Min.X)
 
 			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
 			s00yy1 := int(src.Y[s00i])<<16 + 1<<15
@@ -4485,7 +3581,7 @@ func (ablInterpolator) transform_RGBA_YCbCr422_Over(dst *image.RGBA, dr, adr ima
 			s00g := float64(s00gu)
 			s00b := float64(s00bu)
 			s10i := (sy0-src.Rect.Min.Y)*src.YStride + (sx1 - src.Rect.Min.X)
-			s10j := (sy0-src.Rect.Min.Y)*src.CStride + ((sx1)/2 - src.Rect.Min.X/2)
+			s10j := ((sy0)/2-src.Rect.Min.Y/2)*src.CStride + (sx1 - src.Rect.Min.X)
 
 			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
 			s10yy1 := int(src.Y[s10i])<<16 + 1<<15
@@ -4517,7 +3613,7 @@ func (ablInterpolator) transform_RGBA_YCbCr422_Over(dst *image.RGBA, dr, adr ima
 			s10g = xFrac1*s00g + xFrac0*s10g
 			s10b = xFrac1*s00b + xFrac0*s10b
 			s01i := (sy1-src.Rect.Min.Y)*src.YStride + (sx0 - src.Rect.Min.X)
-			s01j := (sy1-src.Rect.Min.Y)*src.CStride + ((sx0)/2 - src.Rect.Min.X/2)
+			s01j := ((sy1)/2-src.Rect.Min.Y/2)*src.CStride + (sx0 - src.Rect.Min.X)
 
 			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
 			s01yy1 := int(src.Y[s01i])<<16 + 1<<15
@@ -4546,7 +3642,7 @@ func (ablInterpolator) transform_RGBA_YCbCr422_Over(dst *image.RGBA, dr, adr ima
 			s01g := float64(s01gu)
 			s01b := float64(s01bu)
 			s11i := (sy1-src.Rect.Min.Y)*src.YStride + (sx1 - src.Rect.Min.X)
-			s11j := (sy1-src.Rect.Min.Y)*src.CStride + ((sx1)/2 - src.Rect.Min.X/2)
+			s11j := ((sy1)/2-src.Rect.Min.Y/2)*src.CStride + (sx1 - src.Rect.Min.X)
 
 			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
 			s11yy1 := int(src.Y[s11i])<<16 + 1<<15
@@ -4588,7 +3684,7 @@ func (ablInterpolator) transform_RGBA_YCbCr422_Over(dst *image.RGBA, dr, adr ima
 	}
 }
 
-func (ablInterpolator) transform_RGBA_YCbCr420_Over(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.YCbCr, sr image.Rectangle, bias image.Point) {
+func (ablInterpolator) transform_RGBA_Image_Over(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src image.Image, sr image.Rectangle, bias image.Point) {
 	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
 		dyf := float64(dr.Min.Y+int(dy)) + 0.5
 		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
@@ -4628,140 +3724,47 @@ func (ablInterpolator) transform_RGBA_YCbCr420_Over(dst *image.RGBA, dr, adr ima
 				yFrac0, yFrac1 = 1, 0
 			}
 
-			s00i := (sy0-src.Rect.Min.Y)*src.YStride + (sx0 - src.Rect.Min.X)
-			s00j := ((sy0)/2-src.Rect.Min.Y/2)*src.CStride + ((sx0)/2 - src.Rect.Min.X/2)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s00yy1 := int(src.Y[s00i])<<16 + 1<<15
-			s00cb1 := int(src.Cb[s00j]) - 128
-			s00cr1 := int(src.Cr[s00j]) - 128
-			s00ru := (s00yy1 + 91881*s00cr1) >> 8
-			s00gu := (s00yy1 - 22554*s00cb1 - 46802*s00cr1) >> 8
-			s00bu := (s00yy1 + 116130*s00cb1) >> 8
-			if s00ru < 0 {
-				s00ru = 0
-			} else if s00ru > 0xffff {
-				s00ru = 0xffff
-			}
-			if s00gu < 0 {
-				s00gu = 0
-			} else if s00gu > 0xffff {
-				s00gu = 0xffff
-			}
-			if s00bu < 0 {
-				s00bu = 0
-			} else if s00bu > 0xffff {
-				s00bu = 0xffff
-			}
-
+			s00ru, s00gu, s00bu, s00au := src.At(sx0, sy0).RGBA()
 			s00r := float64(s00ru)
 			s00g := float64(s00gu)
 			s00b := float64(s00bu)
-			s10i := (sy0-src.Rect.Min.Y)*src.YStride + (sx1 - src.Rect.Min.X)
-			s10j := ((sy0)/2-src.Rect.Min.Y/2)*src.CStride + ((sx1)/2 - src.Rect.Min.X/2)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s10yy1 := int(src.Y[s10i])<<16 + 1<<15
-			s10cb1 := int(src.Cb[s10j]) - 128
-			s10cr1 := int(src.Cr[s10j]) - 128
-			s10ru := (s10yy1 + 91881*s10cr1) >> 8
-			s10gu := (s10yy1 - 22554*s10cb1 - 46802*s10cr1) >> 8
-			s10bu := (s10yy1 + 116130*s10cb1) >> 8
-			if s10ru < 0 {
-				s10ru = 0
-			} else if s10ru > 0xffff {
-				s10ru = 0xffff
-			}
-			if s10gu < 0 {
-				s10gu = 0
-			} else if s10gu > 0xffff {
-				s10gu = 0xffff
-			}
-			if s10bu < 0 {
-				s10bu = 0
-			} else if s10bu > 0xffff {
-				s10bu = 0xffff
-			}
-
+			s00a := float64(s00au)
+			s10ru, s10gu, s10bu, s10au := src.At(sx1, sy0).RGBA()
 			s10r := float64(s10ru)
 			s10g := float64(s10gu)
 			s10b := float64(s10bu)
+			s10a := float64(s10au)
 			s10r = xFrac1*s00r + xFrac0*s10r
 			s10g = xFrac1*s00g + xFrac0*s10g
 			s10b = xFrac1*s00b + xFrac0*s10b
-			s01i := (sy1-src.Rect.Min.Y)*src.YStride + (sx0 - src.Rect.Min.X)
-			s01j := ((sy1)/2-src.Rect.Min.Y/2)*src.CStride + ((sx0)/2 - src.Rect.Min.X/2)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s01yy1 := int(src.Y[s01i])<<16 + 1<<15
-			s01cb1 := int(src.Cb[s01j]) - 128
-			s01cr1 := int(src.Cr[s01j]) - 128
-			s01ru := (s01yy1 + 91881*s01cr1) >> 8
-			s01gu := (s01yy1 - 22554*s01cb1 - 46802*s01cr1) >> 8
-			s01bu := (s01yy1 + 116130*s01cb1) >> 8
-			if s01ru < 0 {
-				s01ru = 0
-			} else if s01ru > 0xffff {
-				s01ru = 0xffff
-			}
-			if s01gu < 0 {
-				s01gu = 0
-			} else if s01gu > 0xffff {
-				s01gu = 0xffff
-			}
-			if s01bu < 0 {
-				s01bu = 0
-			} else if s01bu > 0xffff {
-				s01bu = 0xffff
-			}
-
+			s10a = xFrac1*s00a + xFrac0*s10a
+			s01ru, s01gu, s01bu, s01au := src.At(sx0, sy1).RGBA()
 			s01r := float64(s01ru)
 			s01g := float64(s01gu)
 			s01b := float64(s01bu)
-			s11i := (sy1-src.Rect.Min.Y)*src.YStride + (sx1 - src.Rect.Min.X)
-			s11j := ((sy1)/2-src.Rect.Min.Y/2)*src.CStride + ((sx1)/2 - src.Rect.Min.X/2)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s11yy1 := int(src.Y[s11i])<<16 + 1<<15
-			s11cb1 := int(src.Cb[s11j]) - 128
-			s11cr1 := int(src.Cr[s11j]) - 128
-			s11ru := (s11yy1 + 91881*s11cr1) >> 8
-			s11gu := (s11yy1 - 22554*s11cb1 - 46802*s11cr1) >> 8
-			s11bu := (s11yy1 + 116130*s11cb1) >> 8
-			if s11ru < 0 {
-				s11ru = 0
-			} else if s11ru > 0xffff {
-				s11ru = 0xffff
-			}
-			if s11gu < 0 {
-				s11gu = 0
-			} else if s11gu > 0xffff {
-				s11gu = 0xffff
-			}
-			if s11bu < 0 {
-				s11bu = 0
-			} else if s11bu > 0xffff {
-				s11bu = 0xffff
-			}
-
+			s01a := float64(s01au)
+			s11ru, s11gu, s11bu, s11au := src.At(sx1, sy1).RGBA()
 			s11r := float64(s11ru)
 			s11g := float64(s11gu)
 			s11b := float64(s11bu)
+			s11a := float64(s11au)
 			s11r = xFrac1*s01r + xFrac0*s11r
 			s11g = xFrac1*s01g + xFrac0*s11g
 			s11b = xFrac1*s01b + xFrac0*s11b
+			s11a = xFrac1*s01a + xFrac0*s11a
 			s11r = yFrac1*s10r + yFrac0*s11r
 			s11g = yFrac1*s10g + yFrac0*s11g
 			s11b = yFrac1*s10b + yFrac0*s11b
+			s11a = yFrac1*s10a + yFrac0*s11a
 			dst.Pix[d+0] = uint8(uint32(s11r) >> 8)
 			dst.Pix[d+1] = uint8(uint32(s11g) >> 8)
 			dst.Pix[d+2] = uint8(uint32(s11b) >> 8)
-			dst.Pix[d+3] = 0xff
+			dst.Pix[d+3] = uint8(uint32(s11a) >> 8)
 		}
 	}
 }
 
-func (ablInterpolator) transform_RGBA_YCbCr440_Over(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.YCbCr, sr image.Rectangle, bias image.Point) {
+func (ablInterpolator) transform_RGBA_Image_Src(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src image.Image, sr image.Rectangle, bias image.Point) {
 	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
 		dyf := float64(dr.Min.Y+int(dy)) + 0.5
 		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
@@ -4801,144 +3804,52 @@ func (ablInterpolator) transform_RGBA_YCbCr440_Over(dst *image.RGBA, dr, adr ima
 				yFrac0, yFrac1 = 1, 0
 			}
 
-			s00i := (sy0-src.Rect.Min.Y)*src.YStride + (sx0 - src.Rect.Min.X)
-			s00j := ((sy0)/2-src.Rect.Min.Y/2)*src.CStride + (sx0 - src.Rect.Min.X)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s00yy1 := int(src.Y[s00i])<<16 + 1<<15
-			s00cb1 := int(src.Cb[s00j]) - 128
-			s00cr1 := int(src.Cr[s00j]) - 128
-			s00ru := (s00yy1 + 91881*s00cr1) >> 8
-			s00gu := (s00yy1 - 22554*s00cb1 - 46802*s00cr1) >> 8
-			s00bu := (s00yy1 + 116130*s00cb1) >> 8
-			if s00ru < 0 {
-				s00ru = 0
-			} else if s00ru > 0xffff {
-				s00ru = 0xffff
-			}
-			if s00gu < 0 {
-				s00gu = 0
-			} else if s00gu > 0xffff {
-				s00gu = 0xffff
-			}
-			if s00bu < 0 {
-				s00bu = 0
-			} else if s00bu > 0xffff {
-				s00bu = 0xffff
-			}
-
+			s00ru, s00gu, s00bu, s00au := src.At(sx0, sy0).RGBA()
 			s00r := float64(s00ru)
 			s00g := float64(s00gu)
 			s00b := float64(s00bu)
-			s10i := (sy0-src.Rect.Min.Y)*src.YStride + (sx1 - src.Rect.Min.X)
-			s10j := ((sy0)/2-src.Rect.Min.Y/2)*src.CStride + (sx1 - src.Rect.Min.X)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s10yy1 := int(src.Y[s10i])<<16 + 1<<15
-			s10cb1 := int(src.Cb[s10j]) - 128
-			s10cr1 := int(src.Cr[s10j]) - 128
-			s10ru := (s10yy1 + 91881*s10cr1) >> 8
-			s10gu := (s10yy1 - 22554*s10cb1 - 46802*s10cr1) >> 8
-			s10bu := (s10yy1 + 116130*s10cb1) >> 8
-			if s10ru < 0 {
-				s10ru = 0
-			} else if s10ru > 0xffff {
-				s10ru = 0xffff
-			}
-			if s10gu < 0 {
-				s10gu = 0
-			} else if s10gu > 0xffff {
-				s10gu = 0xffff
-			}
-			if s10bu < 0 {
-				s10bu = 0
-			} else if s10bu > 0xffff {
-				s10bu = 0xffff
-			}
-
+			s00a := float64(s00au)
+			s10ru, s10gu, s10bu, s10au := src.At(sx1, sy0).RGBA()
 			s10r := float64(s10ru)
 			s10g := float64(s10gu)
 			s10b := float64(s10bu)
+			s10a := float64(s10au)
 			s10r = xFrac1*s00r + xFrac0*s10r
 			s10g = xFrac1*s00g + xFrac0*s10g
 			s10b = xFrac1*s00b + xFrac0*s10b
-			s01i := (sy1-src.Rect.Min.Y)*src.YStride + (sx0 - src.Rect.Min.X)
-			s01j := ((sy1)/2-src.Rect.Min.Y/2)*src.CStride + (sx0 - src.Rect.Min.X)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s01yy1 := int(src.Y[s01i])<<16 + 1<<15
-			s01cb1 := int(src.Cb[s01j]) - 128
-			s01cr1 := int(src.Cr[s01j]) - 128
-			s01ru := (s01yy1 + 91881*s01cr1) >> 8
-			s01gu := (s01yy1 - 22554*s01cb1 - 46802*s01cr1) >> 8
-			s01bu := (s01yy1 + 116130*s01cb1) >> 8
-			if s01ru < 0 {
-				s01ru = 0
-			} else if s01ru > 0xffff {
-				s01ru = 0xffff
-			}
-			if s01gu < 0 {
-				s01gu = 0
-			} else if s01gu > 0xffff {
-				s01gu = 0xffff
-			}
-			if s01bu < 0 {
-				s01bu = 0
-			} else if s01bu > 0xffff {
-				s01bu = 0xffff
-			}
-
+			s10a = xFrac1*s00a + xFrac0*s10a
+			s01ru, s01gu, s01bu, s01au := src.At(sx0, sy1).RGBA()
 			s01r := float64(s01ru)
 			s01g := float64(s01gu)
 			s01b := float64(s01bu)
-			s11i := (sy1-src.Rect.Min.Y)*src.YStride + (sx1 - src.Rect.Min.X)
-			s11j := ((sy1)/2-src.Rect.Min.Y/2)*src.CStride + (sx1 - src.Rect.Min.X)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s11yy1 := int(src.Y[s11i])<<16 + 1<<15
-			s11cb1 := int(src.Cb[s11j]) - 128
-			s11cr1 := int(src.Cr[s11j]) - 128
-			s11ru := (s11yy1 + 91881*s11cr1) >> 8
-			s11gu := (s11yy1 - 22554*s11cb1 - 46802*s11cr1) >> 8
-			s11bu := (s11yy1 + 116130*s11cb1) >> 8
-			if s11ru < 0 {
-				s11ru = 0
-			} else if s11ru > 0xffff {
-				s11ru = 0xffff
-			}
-			if s11gu < 0 {
-				s11gu = 0
-			} else if s11gu > 0xffff {
-				s11gu = 0xffff
-			}
-			if s11bu < 0 {
-				s11bu = 0
-			} else if s11bu > 0xffff {
-				s11bu = 0xffff
-			}
-
+			s01a := float64(s01au)
+			s11ru, s11gu, s11bu, s11au := src.At(sx1, sy1).RGBA()
 			s11r := float64(s11ru)
 			s11g := float64(s11gu)
 			s11b := float64(s11bu)
+			s11a := float64(s11au)
 			s11r = xFrac1*s01r + xFrac0*s11r
 			s11g = xFrac1*s01g + xFrac0*s11g
 			s11b = xFrac1*s01b + xFrac0*s11b
+			s11a = xFrac1*s01a + xFrac0*s11a
 			s11r = yFrac1*s10r + yFrac0*s11r
 			s11g = yFrac1*s10g + yFrac0*s11g
 			s11b = yFrac1*s10b + yFrac0*s11b
+			s11a = yFrac1*s10a + yFrac0*s11a
 			dst.Pix[d+0] = uint8(uint32(s11r) >> 8)
 			dst.Pix[d+1] = uint8(uint32(s11g) >> 8)
 			dst.Pix[d+2] = uint8(uint32(s11b) >> 8)
-			dst.Pix[d+3] = 0xff
+			dst.Pix[d+3] = uint8(uint32(s11a) >> 8)
 		}
 	}
 }
 
-func (ablInterpolator) transform_RGBA_YCbCr444_Src(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.YCbCr, sr image.Rectangle, bias image.Point) {
+func (ablInterpolator) transform_Image_Image_Over(dst Image, dr, adr image.Rectangle, d2s *f64.Aff3, src image.Image, sr image.Rectangle, bias image.Point) {
+	dstColorRGBA64 := &color.RGBA64{}
+	dstColor := color.Color(dstColorRGBA64)
 	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
 		dyf := float64(dr.Min.Y+int(dy)) + 0.5
-		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
-		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx, d = dx+1, d+4 {
+		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx++ {
 			dxf := float64(dr.Min.X+int(dx)) + 0.5
 			sx := d2s[0]*dxf + d2s[1]*dyf + d2s[2]
 			sy := d2s[3]*dxf + d2s[4]*dyf + d2s[5]
@@ -4974,144 +3885,53 @@ func (ablInterpolator) transform_RGBA_YCbCr444_Src(dst *image.RGBA, dr, adr imag
 				yFrac0, yFrac1 = 1, 0
 			}
 
-			s00i := (sy0-src.Rect.Min.Y)*src.YStride + (sx0 - src.Rect.Min.X)
-			s00j := (sy0-src.Rect.Min.Y)*src.CStride + (sx0 - src.Rect.Min.X)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s00yy1 := int(src.Y[s00i])<<16 + 1<<15
-			s00cb1 := int(src.Cb[s00j]) - 128
-			s00cr1 := int(src.Cr[s00j]) - 128
-			s00ru := (s00yy1 + 91881*s00cr1) >> 8
-			s00gu := (s00yy1 - 22554*s00cb1 - 46802*s00cr1) >> 8
-			s00bu := (s00yy1 + 116130*s00cb1) >> 8
-			if s00ru < 0 {
-				s00ru = 0
-			} else if s00ru > 0xffff {
-				s00ru = 0xffff
-			}
-			if s00gu < 0 {
-				s00gu = 0
-			} else if s00gu > 0xffff {
-				s00gu = 0xffff
-			}
-			if s00bu < 0 {
-				s00bu = 0
-			} else if s00bu > 0xffff {
-				s00bu = 0xffff
-			}
-
+			s00ru, s00gu, s00bu, s00au := src.At(sx0, sy0).RGBA()
 			s00r := float64(s00ru)
 			s00g := float64(s00gu)
 			s00b := float64(s00bu)
-			s10i := (sy0-src.Rect.Min.Y)*src.YStride + (sx1 - src.Rect.Min.X)
-			s10j := (sy0-src.Rect.Min.Y)*src.CStride + (sx1 - src.Rect.Min.X)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s10yy1 := int(src.Y[s10i])<<16 + 1<<15
-			s10cb1 := int(src.Cb[s10j]) - 128
-			s10cr1 := int(src.Cr[s10j]) - 128
-			s10ru := (s10yy1 + 91881*s10cr1) >> 8
-			s10gu := (s10yy1 - 22554*s10cb1 - 46802*s10cr1) >> 8
-			s10bu := (s10yy1 + 116130*s10cb1) >> 8
-			if s10ru < 0 {
-				s10ru = 0
-			} else if s10ru > 0xffff {
-				s10ru = 0xffff
-			}
-			if s10gu < 0 {
-				s10gu = 0
-			} else if s10gu > 0xffff {
-				s10gu = 0xffff
-			}
-			if s10bu < 0 {
-				s10bu = 0
-			} else if s10bu > 0xffff {
-				s10bu = 0xffff
-			}
-
+			s00a := float64(s00au)
+			s10ru, s10gu, s10bu, s10au := src.At(sx1, sy0).RGBA()
 			s10r := float64(s10ru)
 			s10g := float64(s10gu)
 			s10b := float64(s10bu)
+			s10a := float64(s10au)
 			s10r = xFrac1*s00r + xFrac0*s10r
 			s10g = xFrac1*s00g + xFrac0*s10g
 			s10b = xFrac1*s00b + xFrac0*s10b
-			s01i := (sy1-src.Rect.Min.Y)*src.YStride + (sx0 - src.Rect.Min.X)
-			s01j := (sy1-src.Rect.Min.Y)*src.CStride + (sx0 - src.Rect.Min.X)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s01yy1 := int(src.Y[s01i])<<16 + 1<<15
-			s01cb1 := int(src.Cb[s01j]) - 128
-			s01cr1 := int(src.Cr[s01j]) - 128
-			s01ru := (s01yy1 + 91881*s01cr1) >> 8
-			s01gu := (s01yy1 - 22554*s01cb1 - 46802*s01cr1) >> 8
-			s01bu := (s01yy1 + 116130*s01cb1) >> 8
-			if s01ru < 0 {
-				s01ru = 0
-			} else if s01ru > 0xffff {
-				s01ru = 0xffff
-			}
-			if s01gu < 0 {
-				s01gu = 0
-			} else if s01gu > 0xffff {
-				s01gu = 0xffff
-			}
-			if s01bu < 0 {
-				s01bu = 0
-			} else if s01bu > 0xffff {
-				s01bu = 0xffff
-			}
-
+			s10a = xFrac1*s00a + xFrac0*s10a
+			s01ru, s01gu, s01bu, s01au := src.At(sx0, sy1).RGBA()
 			s01r := float64(s01ru)
 			s01g := float64(s01gu)
 			s01b := float64(s01bu)
-			s11i := (sy1-src.Rect.Min.Y)*src.YStride + (sx1 - src.Rect.Min.X)
-			s11j := (sy1-src.Rect.Min.Y)*src.CStride + (sx1 - src.Rect.Min.X)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s11yy1 := int(src.Y[s11i])<<16 + 1<<15
-			s11cb1 := int(src.Cb[s11j]) - 128
-			s11cr1 := int(src.Cr[s11j]) - 128
-			s11ru := (s11yy1 + 91881*s11cr1) >> 8
-			s11gu := (s11yy1 - 22554*s11cb1 - 46802*s11cr1) >> 8
-			s11bu := (s11yy1 + 116130*s11cb1) >> 8
-			if s11ru < 0 {
-				s11ru = 0
-			} else if s11ru > 0xffff {
-				s11ru = 0xffff
-			}
-			if s11gu < 0 {
-				s11gu = 0
-			} else if s11gu > 0xffff {
-				s11gu = 0xffff
-			}
-			if s11bu < 0 {
-				s11bu = 0
-			} else if s11bu > 0xffff {
-				s11bu = 0xffff
-			}
-
+			s01a := float64(s01au)
+			s11ru, s11gu, s11bu, s11au := src.At(sx1, sy1).RGBA()
 			s11r := float64(s11ru)
 			s11g := float64(s11gu)
 			s11b := float64(s11bu)
+			s11a := float64(s11au)
 			s11r = xFrac1*s01r + xFrac0*s11r
 			s11g = xFrac1*s01g + xFrac0*s11g
 			s11b = xFrac1*s01b + xFrac0*s11b
+			s11a = xFrac1*s01a + xFrac0*s11a
 			s11r = yFrac1*s10r + yFrac0*s11r
 			s11g = yFrac1*s10g + yFrac0*s11g
 			s11b = yFrac1*s10b + yFrac0*s11b
-			dst.Pix[d+0] = uint8(uint32(s11r) >> 8)
-			dst.Pix[d+1] = uint8(uint32(s11g) >> 8)
-			dst.Pix[d+2] = uint8(uint32(s11b) >> 8)
-			dst.Pix[d+3] = 0xff
+			s11a = yFrac1*s10a + yFrac0*s11a
+			dstColorRGBA64.R = uint16(s11r)
+			dstColorRGBA64.G = uint16(s11g)
+			dstColorRGBA64.B = uint16(s11b)
+			dstColorRGBA64.A = uint16(s11a)
+			dst.Set(dr.Min.X+int(dx), dr.Min.Y+int(dy), dstColor)
 		}
 	}
 }
 
-func (ablInterpolator) transform_RGBA_YCbCr422_Src(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.YCbCr, sr image.Rectangle, bias image.Point) {
+func (ablInterpolator) transform_Image_Image_Src(dst Image, dr, adr image.Rectangle, d2s *f64.Aff3, src image.Image, sr image.Rectangle, bias image.Point) {
+	dstColorRGBA64 := &color.RGBA64{}
+	dstColor := color.Color(dstColorRGBA64)
 	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
 		dyf := float64(dr.Min.Y+int(dy)) + 0.5
-		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
-		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx, d = dx+1, d+4 {
+		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx++ {
 			dxf := float64(dr.Min.X+int(dx)) + 0.5
 			sx := d2s[0]*dxf + d2s[1]*dyf + d2s[2]
 			sy := d2s[3]*dxf + d2s[4]*dyf + d2s[5]
@@ -5147,1887 +3967,608 @@ func (ablInterpolator) transform_RGBA_YCbCr422_Src(dst *image.RGBA, dr, adr imag
 				yFrac0, yFrac1 = 1, 0
 			}
 
-			s00i := (sy0-src.Rect.Min.Y)*src.YStride + (sx0 - src.Rect.Min.X)
-			s00j := (sy0-src.Rect.Min.Y)*src.CStride + ((sx0)/2 - src.Rect.Min.X/2)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s00yy1 := int(src.Y[s00i])<<16 + 1<<15
-			s00cb1 := int(src.Cb[s00j]) - 128
-			s00cr1 := int(src.Cr[s00j]) - 128
-			s00ru := (s00yy1 + 91881*s00cr1) >> 8
-			s00gu := (s00yy1 - 22554*s00cb1 - 46802*s00cr1) >> 8
-			s00bu := (s00yy1 + 116130*s00cb1) >> 8
-			if s00ru < 0 {
-				s00ru = 0
-			} else if s00ru > 0xffff {
-				s00ru = 0xffff
-			}
-			if s00gu < 0 {
-				s00gu = 0
-			} else if s00gu > 0xffff {
-				s00gu = 0xffff
-			}
-			if s00bu < 0 {
-				s00bu = 0
-			} else if s00bu > 0xffff {
-				s00bu = 0xffff
-			}
-
+			s00ru, s00gu, s00bu, s00au := src.At(sx0, sy0).RGBA()
 			s00r := float64(s00ru)
 			s00g := float64(s00gu)
 			s00b := float64(s00bu)
-			s10i := (sy0-src.Rect.Min.Y)*src.YStride + (sx1 - src.Rect.Min.X)
-			s10j := (sy0-src.Rect.Min.Y)*src.CStride + ((sx1)/2 - src.Rect.Min.X/2)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s10yy1 := int(src.Y[s10i])<<16 + 1<<15
-			s10cb1 := int(src.Cb[s10j]) - 128
-			s10cr1 := int(src.Cr[s10j]) - 128
-			s10ru := (s10yy1 + 91881*s10cr1) >> 8
-			s10gu := (s10yy1 - 22554*s10cb1 - 46802*s10cr1) >> 8
-			s10bu := (s10yy1 + 116130*s10cb1) >> 8
-			if s10ru < 0 {
-				s10ru = 0
-			} else if s10ru > 0xffff {
-				s10ru = 0xffff
-			}
-			if s10gu < 0 {
-				s10gu = 0
-			} else if s10gu > 0xffff {
-				s10gu = 0xffff
-			}
-			if s10bu < 0 {
-				s10bu = 0
-			} else if s10bu > 0xffff {
-				s10bu = 0xffff
-			}
-
+			s00a := float64(s00au)
+			s10ru, s10gu, s10bu, s10au := src.At(sx1, sy0).RGBA()
 			s10r := float64(s10ru)
 			s10g := float64(s10gu)
 			s10b := float64(s10bu)
+			s10a := float64(s10au)
 			s10r = xFrac1*s00r + xFrac0*s10r
 			s10g = xFrac1*s00g + xFrac0*s10g
 			s10b = xFrac1*s00b + xFrac0*s10b
-			s01i := (sy1-src.Rect.Min.Y)*src.YStride + (sx0 - src.Rect.Min.X)
-			s01j := (sy1-src.Rect.Min.Y)*src.CStride + ((sx0)/2 - src.Rect.Min.X/2)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s01yy1 := int(src.Y[s01i])<<16 + 1<<15
-			s01cb1 := int(src.Cb[s01j]) - 128
-			s01cr1 := int(src.Cr[s01j]) - 128
-			s01ru := (s01yy1 + 91881*s01cr1) >> 8
-			s01gu := (s01yy1 - 22554*s01cb1 - 46802*s01cr1) >> 8
-			s01bu := (s01yy1 + 116130*s01cb1) >> 8
-			if s01ru < 0 {
-				s01ru = 0
-			} else if s01ru > 0xffff {
-				s01ru = 0xffff
-			}
-			if s01gu < 0 {
-				s01gu = 0
-			} else if s01gu > 0xffff {
-				s01gu = 0xffff
-			}
-			if s01bu < 0 {
-				s01bu = 0
-			} else if s01bu > 0xffff {
-				s01bu = 0xffff
-			}
-
+			s10a = xFrac1*s00a + xFrac0*s10a
+			s01ru, s01gu, s01bu, s01au := src.At(sx0, sy1).RGBA()
 			s01r := float64(s01ru)
 			s01g := float64(s01gu)
 			s01b := float64(s01bu)
-			s11i := (sy1-src.Rect.Min.Y)*src.YStride + (sx1 - src.Rect.Min.X)
-			s11j := (sy1-src.Rect.Min.Y)*src.CStride + ((sx1)/2 - src.Rect.Min.X/2)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s11yy1 := int(src.Y[s11i])<<16 + 1<<15
-			s11cb1 := int(src.Cb[s11j]) - 128
-			s11cr1 := int(src.Cr[s11j]) - 128
-			s11ru := (s11yy1 + 91881*s11cr1) >> 8
-			s11gu := (s11yy1 - 22554*s11cb1 - 46802*s11cr1) >> 8
-			s11bu := (s11yy1 + 116130*s11cb1) >> 8
-			if s11ru < 0 {
-				s11ru = 0
-			} else if s11ru > 0xffff {
-				s11ru = 0xffff
-			}
-			if s11gu < 0 {
-				s11gu = 0
-			} else if s11gu > 0xffff {
-				s11gu = 0xffff
-			}
-			if s11bu < 0 {
-				s11bu = 0
-			} else if s11bu > 0xffff {
-				s11bu = 0xffff
-			}
-
+			s01a := float64(s01au)
+			s11ru, s11gu, s11bu, s11au := src.At(sx1, sy1).RGBA()
 			s11r := float64(s11ru)
 			s11g := float64(s11gu)
 			s11b := float64(s11bu)
+			s11a := float64(s11au)
 			s11r = xFrac1*s01r + xFrac0*s11r
 			s11g = xFrac1*s01g + xFrac0*s11g
 			s11b = xFrac1*s01b + xFrac0*s11b
+			s11a = xFrac1*s01a + xFrac0*s11a
 			s11r = yFrac1*s10r + yFrac0*s11r
 			s11g = yFrac1*s10g + yFrac0*s11g
 			s11b = yFrac1*s10b + yFrac0*s11b
-			dst.Pix[d+0] = uint8(uint32(s11r) >> 8)
-			dst.Pix[d+1] = uint8(uint32(s11g) >> 8)
-			dst.Pix[d+2] = uint8(uint32(s11b) >> 8)
-			dst.Pix[d+3] = 0xff
+			s11a = yFrac1*s10a + yFrac0*s11a
+			dstColorRGBA64.R = uint16(s11r)
+			dstColorRGBA64.G = uint16(s11g)
+			dstColorRGBA64.B = uint16(s11b)
+			dstColorRGBA64.A = uint16(s11a)
+			dst.Set(dr.Min.X+int(dx), dr.Min.Y+int(dy), dstColor)
 		}
 	}
 }
 
-func (ablInterpolator) transform_RGBA_YCbCr420_Src(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.YCbCr, sr image.Rectangle, bias image.Point) {
-	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
-		dyf := float64(dr.Min.Y+int(dy)) + 0.5
-		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
-		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx, d = dx+1, d+4 {
-			dxf := float64(dr.Min.X+int(dx)) + 0.5
-			sx := d2s[0]*dxf + d2s[1]*dyf + d2s[2]
-			sy := d2s[3]*dxf + d2s[4]*dyf + d2s[5]
-			if !(image.Point{int(sx) + bias.X, int(sy) + bias.Y}).In(sr) {
-				continue
-			}
+func (z *kernelScaler) Scale(dst Image, dr image.Rectangle, src image.Image, sr image.Rectangle, opts *Options) {
+	if z.dw != int32(dr.Dx()) || z.dh != int32(dr.Dy()) || z.sw != int32(sr.Dx()) || z.sh != int32(sr.Dy()) {
+		z.kernel.Scale(dst, dr, src, sr, opts)
+		return
+	}
+	// adr is the affected destination pixels, relative to dr.Min.
+	adr := dst.Bounds().Intersect(dr).Sub(dr.Min)
+	if adr.Empty() || sr.Empty() {
+		return
+	}
+	op := opts.op()
+	if op == Over && opaque(src) { // TODO: also check that opts.SrcMask == nil.
+		op = Src
+	}
 
-			sx -= 0.5
-			sx0 := int(sx)
-			xFrac0 := sx - float64(sx0)
-			xFrac1 := 1 - xFrac0
-			sx0 += bias.X
-			sx1 := sx0 + 1
-			if sx0 < sr.Min.X {
-				sx0, sx1 = sr.Min.X, sr.Min.X
-				xFrac0, xFrac1 = 0, 1
-			} else if sx1 >= sr.Max.X {
-				sx0, sx1 = sr.Max.X-1, sr.Max.X-1
-				xFrac0, xFrac1 = 1, 0
-			}
+	if _, ok := src.(*image.Uniform); ok && sr.In(src.Bounds()) {
+		Draw(dst, dr, src, src.Bounds().Min, op)
+		return
+	}
 
-			sy -= 0.5
-			sy0 := int(sy)
-			yFrac0 := sy - float64(sy0)
-			yFrac1 := 1 - yFrac0
-			sy0 += bias.Y
-			sy1 := sy0 + 1
-			if sy0 < sr.Min.Y {
-				sy0, sy1 = sr.Min.Y, sr.Min.Y
-				yFrac0, yFrac1 = 0, 1
-			} else if sy1 >= sr.Max.Y {
-				sy0, sy1 = sr.Max.Y-1, sr.Max.Y-1
-				yFrac0, yFrac1 = 1, 0
-			}
+	// Create a temporary buffer:
+	// scaleX distributes the source image's columns over the temporary image.
+	// scaleY distributes the temporary image's rows over the destination image.
+	var tmp [][4]float64
+	if z.pool.New != nil {
+		tmpp := z.pool.Get().(*[][4]float64)
+		defer z.pool.Put(tmpp)
+		tmp = *tmpp
+	} else {
+		tmp = z.makeTmpBuf()
+	}
 
-			s00i := (sy0-src.Rect.Min.Y)*src.YStride + (sx0 - src.Rect.Min.X)
-			s00j := ((sy0)/2-src.Rect.Min.Y/2)*src.CStride + ((sx0)/2 - src.Rect.Min.X/2)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s00yy1 := int(src.Y[s00i])<<16 + 1<<15
-			s00cb1 := int(src.Cb[s00j]) - 128
-			s00cr1 := int(src.Cr[s00j]) - 128
-			s00ru := (s00yy1 + 91881*s00cr1) >> 8
-			s00gu := (s00yy1 - 22554*s00cb1 - 46802*s00cr1) >> 8
-			s00bu := (s00yy1 + 116130*s00cb1) >> 8
-			if s00ru < 0 {
-				s00ru = 0
-			} else if s00ru > 0xffff {
-				s00ru = 0xffff
-			}
-			if s00gu < 0 {
-				s00gu = 0
-			} else if s00gu > 0xffff {
-				s00gu = 0xffff
-			}
-			if s00bu < 0 {
-				s00bu = 0
-			} else if s00bu > 0xffff {
-				s00bu = 0xffff
-			}
-
-			s00r := float64(s00ru)
-			s00g := float64(s00gu)
-			s00b := float64(s00bu)
-			s10i := (sy0-src.Rect.Min.Y)*src.YStride + (sx1 - src.Rect.Min.X)
-			s10j := ((sy0)/2-src.Rect.Min.Y/2)*src.CStride + ((sx1)/2 - src.Rect.Min.X/2)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s10yy1 := int(src.Y[s10i])<<16 + 1<<15
-			s10cb1 := int(src.Cb[s10j]) - 128
-			s10cr1 := int(src.Cr[s10j]) - 128
-			s10ru := (s10yy1 + 91881*s10cr1) >> 8
-			s10gu := (s10yy1 - 22554*s10cb1 - 46802*s10cr1) >> 8
-			s10bu := (s10yy1 + 116130*s10cb1) >> 8
-			if s10ru < 0 {
-				s10ru = 0
-			} else if s10ru > 0xffff {
-				s10ru = 0xffff
-			}
-			if s10gu < 0 {
-				s10gu = 0
-			} else if s10gu > 0xffff {
-				s10gu = 0xffff
-			}
-			if s10bu < 0 {
-				s10bu = 0
-			} else if s10bu > 0xffff {
-				s10bu = 0xffff
-			}
-
-			s10r := float64(s10ru)
-			s10g := float64(s10gu)
-			s10b := float64(s10bu)
-			s10r = xFrac1*s00r + xFrac0*s10r
-			s10g = xFrac1*s00g + xFrac0*s10g
-			s10b = xFrac1*s00b + xFrac0*s10b
-			s01i := (sy1-src.Rect.Min.Y)*src.YStride + (sx0 - src.Rect.Min.X)
-			s01j := ((sy1)/2-src.Rect.Min.Y/2)*src.CStride + ((sx0)/2 - src.Rect.Min.X/2)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s01yy1 := int(src.Y[s01i])<<16 + 1<<15
-			s01cb1 := int(src.Cb[s01j]) - 128
-			s01cr1 := int(src.Cr[s01j]) - 128
-			s01ru := (s01yy1 + 91881*s01cr1) >> 8
-			s01gu := (s01yy1 - 22554*s01cb1 - 46802*s01cr1) >> 8
-			s01bu := (s01yy1 + 116130*s01cb1) >> 8
-			if s01ru < 0 {
-				s01ru = 0
-			} else if s01ru > 0xffff {
-				s01ru = 0xffff
-			}
-			if s01gu < 0 {
-				s01gu = 0
-			} else if s01gu > 0xffff {
-				s01gu = 0xffff
-			}
-			if s01bu < 0 {
-				s01bu = 0
-			} else if s01bu > 0xffff {
-				s01bu = 0xffff
-			}
-
-			s01r := float64(s01ru)
-			s01g := float64(s01gu)
-			s01b := float64(s01bu)
-			s11i := (sy1-src.Rect.Min.Y)*src.YStride + (sx1 - src.Rect.Min.X)
-			s11j := ((sy1)/2-src.Rect.Min.Y/2)*src.CStride + ((sx1)/2 - src.Rect.Min.X/2)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s11yy1 := int(src.Y[s11i])<<16 + 1<<15
-			s11cb1 := int(src.Cb[s11j]) - 128
-			s11cr1 := int(src.Cr[s11j]) - 128
-			s11ru := (s11yy1 + 91881*s11cr1) >> 8
-			s11gu := (s11yy1 - 22554*s11cb1 - 46802*s11cr1) >> 8
-			s11bu := (s11yy1 + 116130*s11cb1) >> 8
-			if s11ru < 0 {
-				s11ru = 0
-			} else if s11ru > 0xffff {
-				s11ru = 0xffff
-			}
-			if s11gu < 0 {
-				s11gu = 0
-			} else if s11gu > 0xffff {
-				s11gu = 0xffff
-			}
-			if s11bu < 0 {
-				s11bu = 0
-			} else if s11bu > 0xffff {
-				s11bu = 0xffff
-			}
-
-			s11r := float64(s11ru)
-			s11g := float64(s11gu)
-			s11b := float64(s11bu)
-			s11r = xFrac1*s01r + xFrac0*s11r
-			s11g = xFrac1*s01g + xFrac0*s11g
-			s11b = xFrac1*s01b + xFrac0*s11b
-			s11r = yFrac1*s10r + yFrac0*s11r
-			s11g = yFrac1*s10g + yFrac0*s11g
-			s11b = yFrac1*s10b + yFrac0*s11b
-			dst.Pix[d+0] = uint8(uint32(s11r) >> 8)
-			dst.Pix[d+1] = uint8(uint32(s11g) >> 8)
-			dst.Pix[d+2] = uint8(uint32(s11b) >> 8)
-			dst.Pix[d+3] = 0xff
-		}
-	}
-}
-
-func (ablInterpolator) transform_RGBA_YCbCr440_Src(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.YCbCr, sr image.Rectangle, bias image.Point) {
-	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
-		dyf := float64(dr.Min.Y+int(dy)) + 0.5
-		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
-		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx, d = dx+1, d+4 {
-			dxf := float64(dr.Min.X+int(dx)) + 0.5
-			sx := d2s[0]*dxf + d2s[1]*dyf + d2s[2]
-			sy := d2s[3]*dxf + d2s[4]*dyf + d2s[5]
-			if !(image.Point{int(sx) + bias.X, int(sy) + bias.Y}).In(sr) {
-				continue
-			}
-
-			sx -= 0.5
-			sx0 := int(sx)
-			xFrac0 := sx - float64(sx0)
-			xFrac1 := 1 - xFrac0
-			sx0 += bias.X
-			sx1 := sx0 + 1
-			if sx0 < sr.Min.X {
-				sx0, sx1 = sr.Min.X, sr.Min.X
-				xFrac0, xFrac1 = 0, 1
-			} else if sx1 >= sr.Max.X {
-				sx0, sx1 = sr.Max.X-1, sr.Max.X-1
-				xFrac0, xFrac1 = 1, 0
-			}
-
-			sy -= 0.5
-			sy0 := int(sy)
-			yFrac0 := sy - float64(sy0)
-			yFrac1 := 1 - yFrac0
-			sy0 += bias.Y
-			sy1 := sy0 + 1
-			if sy0 < sr.Min.Y {
-				sy0, sy1 = sr.Min.Y, sr.Min.Y
-				yFrac0, yFrac1 = 0, 1
-			} else if sy1 >= sr.Max.Y {
-				sy0, sy1 = sr.Max.Y-1, sr.Max.Y-1
-				yFrac0, yFrac1 = 1, 0
-			}
-
-			s00i := (sy0-src.Rect.Min.Y)*src.YStride + (sx0 - src.Rect.Min.X)
-			s00j := ((sy0)/2-src.Rect.Min.Y/2)*src.CStride + (sx0 - src.Rect.Min.X)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s00yy1 := int(src.Y[s00i])<<16 + 1<<15
-			s00cb1 := int(src.Cb[s00j]) - 128
-			s00cr1 := int(src.Cr[s00j]) - 128
-			s00ru := (s00yy1 + 91881*s00cr1) >> 8
-			s00gu := (s00yy1 - 22554*s00cb1 - 46802*s00cr1) >> 8
-			s00bu := (s00yy1 + 116130*s00cb1) >> 8
-			if s00ru < 0 {
-				s00ru = 0
-			} else if s00ru > 0xffff {
-				s00ru = 0xffff
-			}
-			if s00gu < 0 {
-				s00gu = 0
-			} else if s00gu > 0xffff {
-				s00gu = 0xffff
-			}
-			if s00bu < 0 {
-				s00bu = 0
-			} else if s00bu > 0xffff {
-				s00bu = 0xffff
-			}
-
-			s00r := float64(s00ru)
-			s00g := float64(s00gu)
-			s00b := float64(s00bu)
-			s10i := (sy0-src.Rect.Min.Y)*src.YStride + (sx1 - src.Rect.Min.X)
-			s10j := ((sy0)/2-src.Rect.Min.Y/2)*src.CStride + (sx1 - src.Rect.Min.X)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s10yy1 := int(src.Y[s10i])<<16 + 1<<15
-			s10cb1 := int(src.Cb[s10j]) - 128
-			s10cr1 := int(src.Cr[s10j]) - 128
-			s10ru := (s10yy1 + 91881*s10cr1) >> 8
-			s10gu := (s10yy1 - 22554*s10cb1 - 46802*s10cr1) >> 8
-			s10bu := (s10yy1 + 116130*s10cb1) >> 8
-			if s10ru < 0 {
-				s10ru = 0
-			} else if s10ru > 0xffff {
-				s10ru = 0xffff
-			}
-			if s10gu < 0 {
-				s10gu = 0
-			} else if s10gu > 0xffff {
-				s10gu = 0xffff
-			}
-			if s10bu < 0 {
-				s10bu = 0
-			} else if s10bu > 0xffff {
-				s10bu = 0xffff
-			}
-
-			s10r := float64(s10ru)
-			s10g := float64(s10gu)
-			s10b := float64(s10bu)
-			s10r = xFrac1*s00r + xFrac0*s10r
-			s10g = xFrac1*s00g + xFrac0*s10g
-			s10b = xFrac1*s00b + xFrac0*s10b
-			s01i := (sy1-src.Rect.Min.Y)*src.YStride + (sx0 - src.Rect.Min.X)
-			s01j := ((sy1)/2-src.Rect.Min.Y/2)*src.CStride + (sx0 - src.Rect.Min.X)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s01yy1 := int(src.Y[s01i])<<16 + 1<<15
-			s01cb1 := int(src.Cb[s01j]) - 128
-			s01cr1 := int(src.Cr[s01j]) - 128
-			s01ru := (s01yy1 + 91881*s01cr1) >> 8
-			s01gu := (s01yy1 - 22554*s01cb1 - 46802*s01cr1) >> 8
-			s01bu := (s01yy1 + 116130*s01cb1) >> 8
-			if s01ru < 0 {
-				s01ru = 0
-			} else if s01ru > 0xffff {
-				s01ru = 0xffff
-			}
-			if s01gu < 0 {
-				s01gu = 0
-			} else if s01gu > 0xffff {
-				s01gu = 0xffff
-			}
-			if s01bu < 0 {
-				s01bu = 0
-			} else if s01bu > 0xffff {
-				s01bu = 0xffff
-			}
-
-			s01r := float64(s01ru)
-			s01g := float64(s01gu)
-			s01b := float64(s01bu)
-			s11i := (sy1-src.Rect.Min.Y)*src.YStride + (sx1 - src.Rect.Min.X)
-			s11j := ((sy1)/2-src.Rect.Min.Y/2)*src.CStride + (sx1 - src.Rect.Min.X)
-
-			// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-			s11yy1 := int(src.Y[s11i])<<16 + 1<<15
-			s11cb1 := int(src.Cb[s11j]) - 128
-			s11cr1 := int(src.Cr[s11j]) - 128
-			s11ru := (s11yy1 + 91881*s11cr1) >> 8
-			s11gu := (s11yy1 - 22554*s11cb1 - 46802*s11cr1) >> 8
-			s11bu := (s11yy1 + 116130*s11cb1) >> 8
-			if s11ru < 0 {
-				s11ru = 0
-			} else if s11ru > 0xffff {
-				s11ru = 0xffff
-			}
-			if s11gu < 0 {
-				s11gu = 0
-			} else if s11gu > 0xffff {
-				s11gu = 0xffff
-			}
-			if s11bu < 0 {
-				s11bu = 0
-			} else if s11bu > 0xffff {
-				s11bu = 0xffff
-			}
-
-			s11r := float64(s11ru)
-			s11g := float64(s11gu)
-			s11b := float64(s11bu)
-			s11r = xFrac1*s01r + xFrac0*s11r
-			s11g = xFrac1*s01g + xFrac0*s11g
-			s11b = xFrac1*s01b + xFrac0*s11b
-			s11r = yFrac1*s10r + yFrac0*s11r
-			s11g = yFrac1*s10g + yFrac0*s11g
-			s11b = yFrac1*s10b + yFrac0*s11b
-			dst.Pix[d+0] = uint8(uint32(s11r) >> 8)
-			dst.Pix[d+1] = uint8(uint32(s11g) >> 8)
-			dst.Pix[d+2] = uint8(uint32(s11b) >> 8)
-			dst.Pix[d+3] = 0xff
-		}
-	}
-}
-
-func (ablInterpolator) transform_RGBA_Image_Over(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src image.Image, sr image.Rectangle, bias image.Point) {
-	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
-		dyf := float64(dr.Min.Y+int(dy)) + 0.5
-		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
-		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx, d = dx+1, d+4 {
-			dxf := float64(dr.Min.X+int(dx)) + 0.5
-			sx := d2s[0]*dxf + d2s[1]*dyf + d2s[2]
-			sy := d2s[3]*dxf + d2s[4]*dyf + d2s[5]
-			if !(image.Point{int(sx) + bias.X, int(sy) + bias.Y}).In(sr) {
-				continue
-			}
-
-			sx -= 0.5
-			sx0 := int(sx)
-			xFrac0 := sx - float64(sx0)
-			xFrac1 := 1 - xFrac0
-			sx0 += bias.X
-			sx1 := sx0 + 1
-			if sx0 < sr.Min.X {
-				sx0, sx1 = sr.Min.X, sr.Min.X
-				xFrac0, xFrac1 = 0, 1
-			} else if sx1 >= sr.Max.X {
-				sx0, sx1 = sr.Max.X-1, sr.Max.X-1
-				xFrac0, xFrac1 = 1, 0
-			}
-
-			sy -= 0.5
-			sy0 := int(sy)
-			yFrac0 := sy - float64(sy0)
-			yFrac1 := 1 - yFrac0
-			sy0 += bias.Y
-			sy1 := sy0 + 1
-			if sy0 < sr.Min.Y {
-				sy0, sy1 = sr.Min.Y, sr.Min.Y
-				yFrac0, yFrac1 = 0, 1
-			} else if sy1 >= sr.Max.Y {
-				sy0, sy1 = sr.Max.Y-1, sr.Max.Y-1
-				yFrac0, yFrac1 = 1, 0
-			}
-
-			s00ru, s00gu, s00bu, s00au := src.At(sx0, sy0).RGBA()
-			s00r := float64(s00ru)
-			s00g := float64(s00gu)
-			s00b := float64(s00bu)
-			s00a := float64(s00au)
-			s10ru, s10gu, s10bu, s10au := src.At(sx1, sy0).RGBA()
-			s10r := float64(s10ru)
-			s10g := float64(s10gu)
-			s10b := float64(s10bu)
-			s10a := float64(s10au)
-			s10r = xFrac1*s00r + xFrac0*s10r
-			s10g = xFrac1*s00g + xFrac0*s10g
-			s10b = xFrac1*s00b + xFrac0*s10b
-			s10a = xFrac1*s00a + xFrac0*s10a
-			s01ru, s01gu, s01bu, s01au := src.At(sx0, sy1).RGBA()
-			s01r := float64(s01ru)
-			s01g := float64(s01gu)
-			s01b := float64(s01bu)
-			s01a := float64(s01au)
-			s11ru, s11gu, s11bu, s11au := src.At(sx1, sy1).RGBA()
-			s11r := float64(s11ru)
-			s11g := float64(s11gu)
-			s11b := float64(s11bu)
-			s11a := float64(s11au)
-			s11r = xFrac1*s01r + xFrac0*s11r
-			s11g = xFrac1*s01g + xFrac0*s11g
-			s11b = xFrac1*s01b + xFrac0*s11b
-			s11a = xFrac1*s01a + xFrac0*s11a
-			s11r = yFrac1*s10r + yFrac0*s11r
-			s11g = yFrac1*s10g + yFrac0*s11g
-			s11b = yFrac1*s10b + yFrac0*s11b
-			s11a = yFrac1*s10a + yFrac0*s11a
-			dst.Pix[d+0] = uint8(uint32(s11r) >> 8)
-			dst.Pix[d+1] = uint8(uint32(s11g) >> 8)
-			dst.Pix[d+2] = uint8(uint32(s11b) >> 8)
-			dst.Pix[d+3] = uint8(uint32(s11a) >> 8)
-		}
-	}
-}
-
-func (ablInterpolator) transform_RGBA_Image_Src(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src image.Image, sr image.Rectangle, bias image.Point) {
-	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
-		dyf := float64(dr.Min.Y+int(dy)) + 0.5
-		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
-		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx, d = dx+1, d+4 {
-			dxf := float64(dr.Min.X+int(dx)) + 0.5
-			sx := d2s[0]*dxf + d2s[1]*dyf + d2s[2]
-			sy := d2s[3]*dxf + d2s[4]*dyf + d2s[5]
-			if !(image.Point{int(sx) + bias.X, int(sy) + bias.Y}).In(sr) {
-				continue
-			}
-
-			sx -= 0.5
-			sx0 := int(sx)
-			xFrac0 := sx - float64(sx0)
-			xFrac1 := 1 - xFrac0
-			sx0 += bias.X
-			sx1 := sx0 + 1
-			if sx0 < sr.Min.X {
-				sx0, sx1 = sr.Min.X, sr.Min.X
-				xFrac0, xFrac1 = 0, 1
-			} else if sx1 >= sr.Max.X {
-				sx0, sx1 = sr.Max.X-1, sr.Max.X-1
-				xFrac0, xFrac1 = 1, 0
-			}
-
-			sy -= 0.5
-			sy0 := int(sy)
-			yFrac0 := sy - float64(sy0)
-			yFrac1 := 1 - yFrac0
-			sy0 += bias.Y
-			sy1 := sy0 + 1
-			if sy0 < sr.Min.Y {
-				sy0, sy1 = sr.Min.Y, sr.Min.Y
-				yFrac0, yFrac1 = 0, 1
-			} else if sy1 >= sr.Max.Y {
-				sy0, sy1 = sr.Max.Y-1, sr.Max.Y-1
-				yFrac0, yFrac1 = 1, 0
-			}
-
-			s00ru, s00gu, s00bu, s00au := src.At(sx0, sy0).RGBA()
-			s00r := float64(s00ru)
-			s00g := float64(s00gu)
-			s00b := float64(s00bu)
-			s00a := float64(s00au)
-			s10ru, s10gu, s10bu, s10au := src.At(sx1, sy0).RGBA()
-			s10r := float64(s10ru)
-			s10g := float64(s10gu)
-			s10b := float64(s10bu)
-			s10a := float64(s10au)
-			s10r = xFrac1*s00r + xFrac0*s10r
-			s10g = xFrac1*s00g + xFrac0*s10g
-			s10b = xFrac1*s00b + xFrac0*s10b
-			s10a = xFrac1*s00a + xFrac0*s10a
-			s01ru, s01gu, s01bu, s01au := src.At(sx0, sy1).RGBA()
-			s01r := float64(s01ru)
-			s01g := float64(s01gu)
-			s01b := float64(s01bu)
-			s01a := float64(s01au)
-			s11ru, s11gu, s11bu, s11au := src.At(sx1, sy1).RGBA()
-			s11r := float64(s11ru)
-			s11g := float64(s11gu)
-			s11b := float64(s11bu)
-			s11a := float64(s11au)
-			s11r = xFrac1*s01r + xFrac0*s11r
-			s11g = xFrac1*s01g + xFrac0*s11g
-			s11b = xFrac1*s01b + xFrac0*s11b
-			s11a = xFrac1*s01a + xFrac0*s11a
-			s11r = yFrac1*s10r + yFrac0*s11r
-			s11g = yFrac1*s10g + yFrac0*s11g
-			s11b = yFrac1*s10b + yFrac0*s11b
-			s11a = yFrac1*s10a + yFrac0*s11a
-			dst.Pix[d+0] = uint8(uint32(s11r) >> 8)
-			dst.Pix[d+1] = uint8(uint32(s11g) >> 8)
-			dst.Pix[d+2] = uint8(uint32(s11b) >> 8)
-			dst.Pix[d+3] = uint8(uint32(s11a) >> 8)
-		}
-	}
-}
-
-func (ablInterpolator) transform_Image_Image_Over(dst Image, dr, adr image.Rectangle, d2s *f64.Aff3, src image.Image, sr image.Rectangle, bias image.Point) {
-	dstColorRGBA64 := &color.RGBA64{}
-	dstColor := color.Color(dstColorRGBA64)
-	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
-		dyf := float64(dr.Min.Y+int(dy)) + 0.5
-		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx++ {
-			dxf := float64(dr.Min.X+int(dx)) + 0.5
-			sx := d2s[0]*dxf + d2s[1]*dyf + d2s[2]
-			sy := d2s[3]*dxf + d2s[4]*dyf + d2s[5]
-			if !(image.Point{int(sx) + bias.X, int(sy) + bias.Y}).In(sr) {
-				continue
-			}
-
-			sx -= 0.5
-			sx0 := int(sx)
-			xFrac0 := sx - float64(sx0)
-			xFrac1 := 1 - xFrac0
-			sx0 += bias.X
-			sx1 := sx0 + 1
-			if sx0 < sr.Min.X {
-				sx0, sx1 = sr.Min.X, sr.Min.X
-				xFrac0, xFrac1 = 0, 1
-			} else if sx1 >= sr.Max.X {
-				sx0, sx1 = sr.Max.X-1, sr.Max.X-1
-				xFrac0, xFrac1 = 1, 0
-			}
-
-			sy -= 0.5
-			sy0 := int(sy)
-			yFrac0 := sy - float64(sy0)
-			yFrac1 := 1 - yFrac0
-			sy0 += bias.Y
-			sy1 := sy0 + 1
-			if sy0 < sr.Min.Y {
-				sy0, sy1 = sr.Min.Y, sr.Min.Y
-				yFrac0, yFrac1 = 0, 1
-			} else if sy1 >= sr.Max.Y {
-				sy0, sy1 = sr.Max.Y-1, sr.Max.Y-1
-				yFrac0, yFrac1 = 1, 0
-			}
-
-			s00ru, s00gu, s00bu, s00au := src.At(sx0, sy0).RGBA()
-			s00r := float64(s00ru)
-			s00g := float64(s00gu)
-			s00b := float64(s00bu)
-			s00a := float64(s00au)
-			s10ru, s10gu, s10bu, s10au := src.At(sx1, sy0).RGBA()
-			s10r := float64(s10ru)
-			s10g := float64(s10gu)
-			s10b := float64(s10bu)
-			s10a := float64(s10au)
-			s10r = xFrac1*s00r + xFrac0*s10r
-			s10g = xFrac1*s00g + xFrac0*s10g
-			s10b = xFrac1*s00b + xFrac0*s10b
-			s10a = xFrac1*s00a + xFrac0*s10a
-			s01ru, s01gu, s01bu, s01au := src.At(sx0, sy1).RGBA()
-			s01r := float64(s01ru)
-			s01g := float64(s01gu)
-			s01b := float64(s01bu)
-			s01a := float64(s01au)
-			s11ru, s11gu, s11bu, s11au := src.At(sx1, sy1).RGBA()
-			s11r := float64(s11ru)
-			s11g := float64(s11gu)
-			s11b := float64(s11bu)
-			s11a := float64(s11au)
-			s11r = xFrac1*s01r + xFrac0*s11r
-			s11g = xFrac1*s01g + xFrac0*s11g
-			s11b = xFrac1*s01b + xFrac0*s11b
-			s11a = xFrac1*s01a + xFrac0*s11a
-			s11r = yFrac1*s10r + yFrac0*s11r
-			s11g = yFrac1*s10g + yFrac0*s11g
-			s11b = yFrac1*s10b + yFrac0*s11b
-			s11a = yFrac1*s10a + yFrac0*s11a
-			dstColorRGBA64.R = uint16(s11r)
-			dstColorRGBA64.G = uint16(s11g)
-			dstColorRGBA64.B = uint16(s11b)
-			dstColorRGBA64.A = uint16(s11a)
-			dst.Set(dr.Min.X+int(dx), dr.Min.Y+int(dy), dstColor)
-		}
-	}
-}
-
-func (ablInterpolator) transform_Image_Image_Src(dst Image, dr, adr image.Rectangle, d2s *f64.Aff3, src image.Image, sr image.Rectangle, bias image.Point) {
-	dstColorRGBA64 := &color.RGBA64{}
-	dstColor := color.Color(dstColorRGBA64)
-	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
-		dyf := float64(dr.Min.Y+int(dy)) + 0.5
-		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx++ {
-			dxf := float64(dr.Min.X+int(dx)) + 0.5
-			sx := d2s[0]*dxf + d2s[1]*dyf + d2s[2]
-			sy := d2s[3]*dxf + d2s[4]*dyf + d2s[5]
-			if !(image.Point{int(sx) + bias.X, int(sy) + bias.Y}).In(sr) {
-				continue
-			}
-
-			sx -= 0.5
-			sx0 := int(sx)
-			xFrac0 := sx - float64(sx0)
-			xFrac1 := 1 - xFrac0
-			sx0 += bias.X
-			sx1 := sx0 + 1
-			if sx0 < sr.Min.X {
-				sx0, sx1 = sr.Min.X, sr.Min.X
-				xFrac0, xFrac1 = 0, 1
-			} else if sx1 >= sr.Max.X {
-				sx0, sx1 = sr.Max.X-1, sr.Max.X-1
-				xFrac0, xFrac1 = 1, 0
-			}
-
-			sy -= 0.5
-			sy0 := int(sy)
-			yFrac0 := sy - float64(sy0)
-			yFrac1 := 1 - yFrac0
-			sy0 += bias.Y
-			sy1 := sy0 + 1
-			if sy0 < sr.Min.Y {
-				sy0, sy1 = sr.Min.Y, sr.Min.Y
-				yFrac0, yFrac1 = 0, 1
-			} else if sy1 >= sr.Max.Y {
-				sy0, sy1 = sr.Max.Y-1, sr.Max.Y-1
-				yFrac0, yFrac1 = 1, 0
-			}
-
-			s00ru, s00gu, s00bu, s00au := src.At(sx0, sy0).RGBA()
-			s00r := float64(s00ru)
-			s00g := float64(s00gu)
-			s00b := float64(s00bu)
-			s00a := float64(s00au)
-			s10ru, s10gu, s10bu, s10au := src.At(sx1, sy0).RGBA()
-			s10r := float64(s10ru)
-			s10g := float64(s10gu)
-			s10b := float64(s10bu)
-			s10a := float64(s10au)
-			s10r = xFrac1*s00r + xFrac0*s10r
-			s10g = xFrac1*s00g + xFrac0*s10g
-			s10b = xFrac1*s00b + xFrac0*s10b
-			s10a = xFrac1*s00a + xFrac0*s10a
-			s01ru, s01gu, s01bu, s01au := src.At(sx0, sy1).RGBA()
-			s01r := float64(s01ru)
-			s01g := float64(s01gu)
-			s01b := float64(s01bu)
-			s01a := float64(s01au)
-			s11ru, s11gu, s11bu, s11au := src.At(sx1, sy1).RGBA()
-			s11r := float64(s11ru)
-			s11g := float64(s11gu)
-			s11b := float64(s11bu)
-			s11a := float64(s11au)
-			s11r = xFrac1*s01r + xFrac0*s11r
-			s11g = xFrac1*s01g + xFrac0*s11g
-			s11b = xFrac1*s01b + xFrac0*s11b
-			s11a = xFrac1*s01a + xFrac0*s11a
-			s11r = yFrac1*s10r + yFrac0*s11r
-			s11g = yFrac1*s10g + yFrac0*s11g
-			s11b = yFrac1*s10b + yFrac0*s11b
-			s11a = yFrac1*s10a + yFrac0*s11a
-			dstColorRGBA64.R = uint16(s11r)
-			dstColorRGBA64.G = uint16(s11g)
-			dstColorRGBA64.B = uint16(s11b)
-			dstColorRGBA64.A = uint16(s11a)
-			dst.Set(dr.Min.X+int(dx), dr.Min.Y+int(dy), dstColor)
-		}
-	}
-}
-
-func (z *kernelScaler) Scale(dst Image, dr image.Rectangle, src image.Image, sr image.Rectangle, opts *Options) {
-	if z.dw != int32(dr.Dx()) || z.dh != int32(dr.Dy()) || z.sw != int32(sr.Dx()) || z.sh != int32(sr.Dy()) {
-		z.kernel.Scale(dst, dr, src, sr, opts)
-		return
-	}
-	// adr is the affected destination pixels, relative to dr.Min.
-	adr := dst.Bounds().Intersect(dr).Sub(dr.Min)
-	if adr.Empty() || sr.Empty() {
-		return
-	}
-
-	if _, ok := src.(*image.Uniform); ok && sr.In(src.Bounds()) {
-		Draw(dst, dr, src, src.Bounds().Min, opts.op())
-		return
-	}
-
-	// Create a temporary buffer:
-	// scaleX distributes the source image's columns over the temporary image.
-	// scaleY distributes the temporary image's rows over the destination image.
-	var tmp [][4]float64
-	if z.pool.New != nil {
-		tmpp := z.pool.Get().(*[][4]float64)
-		defer z.pool.Put(tmpp)
-		tmp = *tmpp
-	} else {
-		tmp = z.makeTmpBuf()
-	}
-
-	// sr is the source pixels. If it extends beyond the src bounds,
-	// we cannot use the type-specific fast paths, as they access
-	// the Pix fields directly without bounds checking.
-	if !sr.In(src.Bounds()) {
-		z.scaleX_Image(tmp, src, sr)
-	} else {
-		switch src := src.(type) {
-		case *image.Gray:
-			z.scaleX_Gray(tmp, src, sr)
-		case *image.NRGBA:
-			z.scaleX_NRGBA(tmp, src, sr)
-		case *image.RGBA:
-			z.scaleX_RGBA(tmp, src, sr)
-		case *image.YCbCr:
-			switch src.SubsampleRatio {
-			default:
-				z.scaleX_Image(tmp, src, sr)
-			case image.YCbCrSubsampleRatio444:
-				z.scaleX_YCbCr444(tmp, src, sr)
-			case image.YCbCrSubsampleRatio422:
-				z.scaleX_YCbCr422(tmp, src, sr)
-			case image.YCbCrSubsampleRatio420:
-				z.scaleX_YCbCr420(tmp, src, sr)
-			case image.YCbCrSubsampleRatio440:
-				z.scaleX_YCbCr440(tmp, src, sr)
-			}
-		default:
-			z.scaleX_Image(tmp, src, sr)
-		}
-	}
-
-	switch opts.op() {
-	case Over:
-		switch dst := dst.(type) {
-		case *image.RGBA:
-			z.scaleY_RGBA_Over(dst, dr, adr, tmp)
-		default:
-			z.scaleY_Image_Over(dst, dr, adr, tmp)
-		}
-	case Src:
-		switch dst := dst.(type) {
-		case *image.RGBA:
-			z.scaleY_RGBA_Src(dst, dr, adr, tmp)
-		default:
-			z.scaleY_Image_Src(dst, dr, adr, tmp)
-		}
-	}
-}
-
-func (q *Kernel) Transform(dst Image, s2d *f64.Aff3, src image.Image, sr image.Rectangle, opts *Options) {
-	dr := transformRect(s2d, &sr)
-	// adr is the affected destination pixels.
-	adr := dst.Bounds().Intersect(dr)
-	if adr.Empty() || sr.Empty() {
-		return
-	}
-	d2s := invert(s2d)
-	// bias is a translation of the mapping from dst co-ordinates to
-	// src co-ordinates such that the latter temporarily have
-	// non-negative X and Y co-ordinates. This allows us to write
-	// int(f) instead of int(math.Floor(f)), since "round to zero" and
-	// "round down" are equivalent when f >= 0, but the former is much
-	// cheaper. The X-- and Y-- are because the TransformLeaf methods
-	// have a "sx -= 0.5" adjustment.
-	bias := transformRect(&d2s, &adr).Min
-	bias.X--
-	bias.Y--
-	d2s[2] -= float64(bias.X)
-	d2s[5] -= float64(bias.Y)
-	// Make adr relative to dr.Min.
-	adr = adr.Sub(dr.Min)
-
-	if u, ok := src.(*image.Uniform); ok && sr.In(src.Bounds()) {
-		transform_Uniform(dst, dr, adr, &d2s, u, sr, bias, opts.op())
-		return
-	}
-
-	xscale := abs(d2s[0])
-	if s := abs(d2s[1]); xscale < s {
-		xscale = s
-	}
-	yscale := abs(d2s[3])
-	if s := abs(d2s[4]); yscale < s {
-		yscale = s
-	}
-
-	// sr is the source pixels. If it extends beyond the src bounds,
-	// we cannot use the type-specific fast paths, as they access
-	// the Pix fields directly without bounds checking.
-	if !sr.In(src.Bounds()) {
-		switch opts.op() {
-		case Over:
-			q.transform_Image_Image_Over(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
-		case Src:
-			q.transform_Image_Image_Src(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
-		}
-	} else {
-		switch opts.op() {
-		case Over:
-			switch dst := dst.(type) {
-			case *image.RGBA:
-				switch src := src.(type) {
-				case *image.Gray:
-					q.transform_RGBA_Gray_Over(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
-				case *image.NRGBA:
-					q.transform_RGBA_NRGBA_Over(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
-				case *image.RGBA:
-					q.transform_RGBA_RGBA_Over(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
-				case *image.YCbCr:
-					switch src.SubsampleRatio {
-					default:
-						q.transform_RGBA_Image_Over(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
-					case image.YCbCrSubsampleRatio444:
-						q.transform_RGBA_YCbCr444_Over(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
-					case image.YCbCrSubsampleRatio422:
-						q.transform_RGBA_YCbCr422_Over(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
-					case image.YCbCrSubsampleRatio420:
-						q.transform_RGBA_YCbCr420_Over(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
-					case image.YCbCrSubsampleRatio440:
-						q.transform_RGBA_YCbCr440_Over(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
-					}
-				default:
-					q.transform_RGBA_Image_Over(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
-				}
-			default:
-				switch src := src.(type) {
-				default:
-					q.transform_Image_Image_Over(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
-				}
-			}
-		case Src:
-			switch dst := dst.(type) {
-			case *image.RGBA:
-				switch src := src.(type) {
-				case *image.Gray:
-					q.transform_RGBA_Gray_Src(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
-				case *image.NRGBA:
-					q.transform_RGBA_NRGBA_Src(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
-				case *image.RGBA:
-					q.transform_RGBA_RGBA_Src(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
-				case *image.YCbCr:
-					switch src.SubsampleRatio {
-					default:
-						q.transform_RGBA_Image_Src(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
-					case image.YCbCrSubsampleRatio444:
-						q.transform_RGBA_YCbCr444_Src(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
-					case image.YCbCrSubsampleRatio422:
-						q.transform_RGBA_YCbCr422_Src(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
-					case image.YCbCrSubsampleRatio420:
-						q.transform_RGBA_YCbCr420_Src(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
-					case image.YCbCrSubsampleRatio440:
-						q.transform_RGBA_YCbCr440_Src(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
-					}
-				default:
-					q.transform_RGBA_Image_Src(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
-				}
-			default:
-				switch src := src.(type) {
-				default:
-					q.transform_Image_Image_Src(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
-				}
-			}
-		}
-	}
-}
-
-func (z *kernelScaler) scaleX_Gray(tmp [][4]float64, src *image.Gray, sr image.Rectangle) {
-	t := 0
-	for y := int32(0); y < z.sh; y++ {
-		for _, s := range z.horizontal.sources {
-			var pr float64
-			for _, c := range z.horizontal.contribs[s.i:s.j] {
-				pi := (sr.Min.Y+int(y)-src.Rect.Min.Y)*src.Stride + (sr.Min.X + int(c.coord) - src.Rect.Min.X)
-				pru := uint32(src.Pix[pi]) * 0x101
-				pr += float64(pru) * c.weight
-			}
-			pr *= s.invTotalWeightFFFF
-			tmp[t] = [4]float64{
-				pr,
-				pr,
-				pr,
-				1,
-			}
-			t++
-		}
-	}
-}
-
-func (z *kernelScaler) scaleX_NRGBA(tmp [][4]float64, src *image.NRGBA, sr image.Rectangle) {
-	t := 0
-	for y := int32(0); y < z.sh; y++ {
-		for _, s := range z.horizontal.sources {
-			var pr, pg, pb, pa float64
-			for _, c := range z.horizontal.contribs[s.i:s.j] {
-				pi := (sr.Min.Y+int(y)-src.Rect.Min.Y)*src.Stride + (sr.Min.X+int(c.coord)-src.Rect.Min.X)*4
-				pau := uint32(src.Pix[pi+3]) * 0x101
-				pru := uint32(src.Pix[pi+0]) * pau / 0xff
-				pgu := uint32(src.Pix[pi+1]) * pau / 0xff
-				pbu := uint32(src.Pix[pi+2]) * pau / 0xff
-				pr += float64(pru) * c.weight
-				pg += float64(pgu) * c.weight
-				pb += float64(pbu) * c.weight
-				pa += float64(pau) * c.weight
-			}
-			tmp[t] = [4]float64{
-				pr * s.invTotalWeightFFFF,
-				pg * s.invTotalWeightFFFF,
-				pb * s.invTotalWeightFFFF,
-				pa * s.invTotalWeightFFFF,
-			}
-			t++
-		}
-	}
-}
-
-func (z *kernelScaler) scaleX_RGBA(tmp [][4]float64, src *image.RGBA, sr image.Rectangle) {
-	t := 0
-	for y := int32(0); y < z.sh; y++ {
-		for _, s := range z.horizontal.sources {
-			var pr, pg, pb, pa float64
-			for _, c := range z.horizontal.contribs[s.i:s.j] {
-				pi := (sr.Min.Y+int(y)-src.Rect.Min.Y)*src.Stride + (sr.Min.X+int(c.coord)-src.Rect.Min.X)*4
-				pru := uint32(src.Pix[pi+0]) * 0x101
-				pgu := uint32(src.Pix[pi+1]) * 0x101
-				pbu := uint32(src.Pix[pi+2]) * 0x101
-				pau := uint32(src.Pix[pi+3]) * 0x101
-				pr += float64(pru) * c.weight
-				pg += float64(pgu) * c.weight
-				pb += float64(pbu) * c.weight
-				pa += float64(pau) * c.weight
-			}
-			tmp[t] = [4]float64{
-				pr * s.invTotalWeightFFFF,
-				pg * s.invTotalWeightFFFF,
-				pb * s.invTotalWeightFFFF,
-				pa * s.invTotalWeightFFFF,
-			}
-			t++
-		}
-	}
-}
-
-func (z *kernelScaler) scaleX_YCbCr444(tmp [][4]float64, src *image.YCbCr, sr image.Rectangle) {
-	t := 0
-	for y := int32(0); y < z.sh; y++ {
-		for _, s := range z.horizontal.sources {
-			var pr, pg, pb float64
-			for _, c := range z.horizontal.contribs[s.i:s.j] {
-				pi := (sr.Min.Y+int(y)-src.Rect.Min.Y)*src.YStride + (sr.Min.X + int(c.coord) - src.Rect.Min.X)
-				pj := (sr.Min.Y+int(y)-src.Rect.Min.Y)*src.CStride + (sr.Min.X + int(c.coord) - src.Rect.Min.X)
-
-				// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-				pyy1 := int(src.Y[pi])<<16 + 1<<15
-				pcb1 := int(src.Cb[pj]) - 128
-				pcr1 := int(src.Cr[pj]) - 128
-				pru := (pyy1 + 91881*pcr1) >> 8
-				pgu := (pyy1 - 22554*pcb1 - 46802*pcr1) >> 8
-				pbu := (pyy1 + 116130*pcb1) >> 8
-				if pru < 0 {
-					pru = 0
-				} else if pru > 0xffff {
-					pru = 0xffff
-				}
-				if pgu < 0 {
-					pgu = 0
-				} else if pgu > 0xffff {
-					pgu = 0xffff
-				}
-				if pbu < 0 {
-					pbu = 0
-				} else if pbu > 0xffff {
-					pbu = 0xffff
-				}
-
-				pr += float64(pru) * c.weight
-				pg += float64(pgu) * c.weight
-				pb += float64(pbu) * c.weight
-			}
-			tmp[t] = [4]float64{
-				pr * s.invTotalWeightFFFF,
-				pg * s.invTotalWeightFFFF,
-				pb * s.invTotalWeightFFFF,
-				1,
-			}
-			t++
-		}
-	}
-}
-
-func (z *kernelScaler) scaleX_YCbCr422(tmp [][4]float64, src *image.YCbCr, sr image.Rectangle) {
-	t := 0
-	for y := int32(0); y < z.sh; y++ {
-		for _, s := range z.horizontal.sources {
-			var pr, pg, pb float64
-			for _, c := range z.horizontal.contribs[s.i:s.j] {
-				pi := (sr.Min.Y+int(y)-src.Rect.Min.Y)*src.YStride + (sr.Min.X + int(c.coord) - src.Rect.Min.X)
-				pj := (sr.Min.Y+int(y)-src.Rect.Min.Y)*src.CStride + ((sr.Min.X+int(c.coord))/2 - src.Rect.Min.X/2)
-
-				// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-				pyy1 := int(src.Y[pi])<<16 + 1<<15
-				pcb1 := int(src.Cb[pj]) - 128
-				pcr1 := int(src.Cr[pj]) - 128
-				pru := (pyy1 + 91881*pcr1) >> 8
-				pgu := (pyy1 - 22554*pcb1 - 46802*pcr1) >> 8
-				pbu := (pyy1 + 116130*pcb1) >> 8
-				if pru < 0 {
-					pru = 0
-				} else if pru > 0xffff {
-					pru = 0xffff
-				}
-				if pgu < 0 {
-					pgu = 0
-				} else if pgu > 0xffff {
-					pgu = 0xffff
-				}
-				if pbu < 0 {
-					pbu = 0
-				} else if pbu > 0xffff {
-					pbu = 0xffff
-				}
-
-				pr += float64(pru) * c.weight
-				pg += float64(pgu) * c.weight
-				pb += float64(pbu) * c.weight
-			}
-			tmp[t] = [4]float64{
-				pr * s.invTotalWeightFFFF,
-				pg * s.invTotalWeightFFFF,
-				pb * s.invTotalWeightFFFF,
-				1,
-			}
-			t++
-		}
-	}
-}
-
-func (z *kernelScaler) scaleX_YCbCr420(tmp [][4]float64, src *image.YCbCr, sr image.Rectangle) {
-	t := 0
-	for y := int32(0); y < z.sh; y++ {
-		for _, s := range z.horizontal.sources {
-			var pr, pg, pb float64
-			for _, c := range z.horizontal.contribs[s.i:s.j] {
-				pi := (sr.Min.Y+int(y)-src.Rect.Min.Y)*src.YStride + (sr.Min.X + int(c.coord) - src.Rect.Min.X)
-				pj := ((sr.Min.Y+int(y))/2-src.Rect.Min.Y/2)*src.CStride + ((sr.Min.X+int(c.coord))/2 - src.Rect.Min.X/2)
-
-				// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-				pyy1 := int(src.Y[pi])<<16 + 1<<15
-				pcb1 := int(src.Cb[pj]) - 128
-				pcr1 := int(src.Cr[pj]) - 128
-				pru := (pyy1 + 91881*pcr1) >> 8
-				pgu := (pyy1 - 22554*pcb1 - 46802*pcr1) >> 8
-				pbu := (pyy1 + 116130*pcb1) >> 8
-				if pru < 0 {
-					pru = 0
-				} else if pru > 0xffff {
-					pru = 0xffff
-				}
-				if pgu < 0 {
-					pgu = 0
-				} else if pgu > 0xffff {
-					pgu = 0xffff
-				}
-				if pbu < 0 {
-					pbu = 0
-				} else if pbu > 0xffff {
-					pbu = 0xffff
-				}
-
-				pr += float64(pru) * c.weight
-				pg += float64(pgu) * c.weight
-				pb += float64(pbu) * c.weight
-			}
-			tmp[t] = [4]float64{
-				pr * s.invTotalWeightFFFF,
-				pg * s.invTotalWeightFFFF,
-				pb * s.invTotalWeightFFFF,
-				1,
-			}
-			t++
-		}
-	}
-}
-
-func (z *kernelScaler) scaleX_YCbCr440(tmp [][4]float64, src *image.YCbCr, sr image.Rectangle) {
-	t := 0
-	for y := int32(0); y < z.sh; y++ {
-		for _, s := range z.horizontal.sources {
-			var pr, pg, pb float64
-			for _, c := range z.horizontal.contribs[s.i:s.j] {
-				pi := (sr.Min.Y+int(y)-src.Rect.Min.Y)*src.YStride + (sr.Min.X + int(c.coord) - src.Rect.Min.X)
-				pj := ((sr.Min.Y+int(y))/2-src.Rect.Min.Y/2)*src.CStride + (sr.Min.X + int(c.coord) - src.Rect.Min.X)
-
-				// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-				pyy1 := int(src.Y[pi])<<16 + 1<<15
-				pcb1 := int(src.Cb[pj]) - 128
-				pcr1 := int(src.Cr[pj]) - 128
-				pru := (pyy1 + 91881*pcr1) >> 8
-				pgu := (pyy1 - 22554*pcb1 - 46802*pcr1) >> 8
-				pbu := (pyy1 + 116130*pcb1) >> 8
-				if pru < 0 {
-					pru = 0
-				} else if pru > 0xffff {
-					pru = 0xffff
-				}
-				if pgu < 0 {
-					pgu = 0
-				} else if pgu > 0xffff {
-					pgu = 0xffff
-				}
-				if pbu < 0 {
-					pbu = 0
-				} else if pbu > 0xffff {
-					pbu = 0xffff
-				}
-
-				pr += float64(pru) * c.weight
-				pg += float64(pgu) * c.weight
-				pb += float64(pbu) * c.weight
-			}
-			tmp[t] = [4]float64{
-				pr * s.invTotalWeightFFFF,
-				pg * s.invTotalWeightFFFF,
-				pb * s.invTotalWeightFFFF,
-				1,
-			}
-			t++
-		}
-	}
-}
-
-func (z *kernelScaler) scaleX_Image(tmp [][4]float64, src image.Image, sr image.Rectangle) {
-	t := 0
-	for y := int32(0); y < z.sh; y++ {
-		for _, s := range z.horizontal.sources {
-			var pr, pg, pb, pa float64
-			for _, c := range z.horizontal.contribs[s.i:s.j] {
-				pru, pgu, pbu, pau := src.At(sr.Min.X+int(c.coord), sr.Min.Y+int(y)).RGBA()
-				pr += float64(pru) * c.weight
-				pg += float64(pgu) * c.weight
-				pb += float64(pbu) * c.weight
-				pa += float64(pau) * c.weight
-			}
-			tmp[t] = [4]float64{
-				pr * s.invTotalWeightFFFF,
-				pg * s.invTotalWeightFFFF,
-				pb * s.invTotalWeightFFFF,
-				pa * s.invTotalWeightFFFF,
-			}
-			t++
-		}
-	}
-}
-
-func (z *kernelScaler) scaleY_RGBA_Over(dst *image.RGBA, dr, adr image.Rectangle, tmp [][4]float64) {
-	for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx++ {
-		d := (dr.Min.Y+adr.Min.Y-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+int(dx)-dst.Rect.Min.X)*4
-		for _, s := range z.vertical.sources[adr.Min.Y:adr.Max.Y] {
-			var pr, pg, pb, pa float64
-			for _, c := range z.vertical.contribs[s.i:s.j] {
-				p := &tmp[c.coord*z.dw+dx]
-				pr += p[0] * c.weight
-				pg += p[1] * c.weight
-				pb += p[2] * c.weight
-				pa += p[3] * c.weight
-			}
-			dst.Pix[d+0] = uint8(ftou(pr*s.invTotalWeight) >> 8)
-			dst.Pix[d+1] = uint8(ftou(pg*s.invTotalWeight) >> 8)
-			dst.Pix[d+2] = uint8(ftou(pb*s.invTotalWeight) >> 8)
-			dst.Pix[d+3] = uint8(ftou(pa*s.invTotalWeight) >> 8)
-			d += dst.Stride
-		}
-	}
-}
-
-func (z *kernelScaler) scaleY_RGBA_Src(dst *image.RGBA, dr, adr image.Rectangle, tmp [][4]float64) {
-	for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx++ {
-		d := (dr.Min.Y+adr.Min.Y-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+int(dx)-dst.Rect.Min.X)*4
-		for _, s := range z.vertical.sources[adr.Min.Y:adr.Max.Y] {
-			var pr, pg, pb, pa float64
-			for _, c := range z.vertical.contribs[s.i:s.j] {
-				p := &tmp[c.coord*z.dw+dx]
-				pr += p[0] * c.weight
-				pg += p[1] * c.weight
-				pb += p[2] * c.weight
-				pa += p[3] * c.weight
+	// sr is the source pixels. If it extends beyond the src bounds,
+	// we cannot use the type-specific fast paths, as they access
+	// the Pix fields directly without bounds checking.
+	if !sr.In(src.Bounds()) {
+		z.scaleX_Image(tmp, src, sr)
+	} else {
+		switch src := src.(type) {
+		case *image.Gray:
+			z.scaleX_Gray(tmp, src, sr)
+		case *image.NRGBA:
+			z.scaleX_NRGBA(tmp, src, sr)
+		case *image.RGBA:
+			z.scaleX_RGBA(tmp, src, sr)
+		case *image.YCbCr:
+			switch src.SubsampleRatio {
+			default:
+				z.scaleX_Image(tmp, src, sr)
+			case image.YCbCrSubsampleRatio444:
+				z.scaleX_YCbCr444(tmp, src, sr)
+			case image.YCbCrSubsampleRatio422:
+				z.scaleX_YCbCr422(tmp, src, sr)
+			case image.YCbCrSubsampleRatio420:
+				z.scaleX_YCbCr420(tmp, src, sr)
+			case image.YCbCrSubsampleRatio440:
+				z.scaleX_YCbCr440(tmp, src, sr)
 			}
-			dst.Pix[d+0] = uint8(ftou(pr*s.invTotalWeight) >> 8)
-			dst.Pix[d+1] = uint8(ftou(pg*s.invTotalWeight) >> 8)
-			dst.Pix[d+2] = uint8(ftou(pb*s.invTotalWeight) >> 8)
-			dst.Pix[d+3] = uint8(ftou(pa*s.invTotalWeight) >> 8)
-			d += dst.Stride
+		default:
+			z.scaleX_Image(tmp, src, sr)
 		}
 	}
-}
 
-func (z *kernelScaler) scaleY_Image_Over(dst Image, dr, adr image.Rectangle, tmp [][4]float64) {
-	dstColorRGBA64 := &color.RGBA64{}
-	dstColor := color.Color(dstColorRGBA64)
-	for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx++ {
-		for dy, s := range z.vertical.sources[adr.Min.Y:adr.Max.Y] {
-			var pr, pg, pb, pa float64
-			for _, c := range z.vertical.contribs[s.i:s.j] {
-				p := &tmp[c.coord*z.dw+dx]
-				pr += p[0] * c.weight
-				pg += p[1] * c.weight
-				pb += p[2] * c.weight
-				pa += p[3] * c.weight
-			}
-			dstColorRGBA64.R = ftou(pr * s.invTotalWeight)
-			dstColorRGBA64.G = ftou(pg * s.invTotalWeight)
-			dstColorRGBA64.B = ftou(pb * s.invTotalWeight)
-			dstColorRGBA64.A = ftou(pa * s.invTotalWeight)
-			dst.Set(dr.Min.X+int(dx), dr.Min.Y+int(adr.Min.Y+dy), dstColor)
+	switch op {
+	case Over:
+		switch dst := dst.(type) {
+		case *image.RGBA:
+			z.scaleY_RGBA_Over(dst, dr, adr, tmp)
+		default:
+			z.scaleY_Image_Over(dst, dr, adr, tmp)
 		}
-	}
-}
-
-func (z *kernelScaler) scaleY_Image_Src(dst Image, dr, adr image.Rectangle, tmp [][4]float64) {
-	dstColorRGBA64 := &color.RGBA64{}
-	dstColor := color.Color(dstColorRGBA64)
-	for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx++ {
-		for dy, s := range z.vertical.sources[adr.Min.Y:adr.Max.Y] {
-			var pr, pg, pb, pa float64
-			for _, c := range z.vertical.contribs[s.i:s.j] {
-				p := &tmp[c.coord*z.dw+dx]
-				pr += p[0] * c.weight
-				pg += p[1] * c.weight
-				pb += p[2] * c.weight
-				pa += p[3] * c.weight
-			}
-			dstColorRGBA64.R = ftou(pr * s.invTotalWeight)
-			dstColorRGBA64.G = ftou(pg * s.invTotalWeight)
-			dstColorRGBA64.B = ftou(pb * s.invTotalWeight)
-			dstColorRGBA64.A = ftou(pa * s.invTotalWeight)
-			dst.Set(dr.Min.X+int(dx), dr.Min.Y+int(adr.Min.Y+dy), dstColor)
+	case Src:
+		switch dst := dst.(type) {
+		case *image.RGBA:
+			z.scaleY_RGBA_Src(dst, dr, adr, tmp)
+		default:
+			z.scaleY_Image_Src(dst, dr, adr, tmp)
 		}
 	}
 }
 
-func (q *Kernel) transform_RGBA_Gray_Over(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.Gray, sr image.Rectangle, bias image.Point, xscale, yscale float64) {
-	// When shrinking, broaden the effective kernel support so that we still
-	// visit every source pixel.
-	xHalfWidth, xKernelArgScale := q.Support, 1.0
-	if xscale > 1 {
-		xHalfWidth *= xscale
-		xKernelArgScale = 1 / xscale
+func (q *Kernel) Transform(dst Image, s2d *f64.Aff3, src image.Image, sr image.Rectangle, opts *Options) {
+	dr := transformRect(s2d, &sr)
+	// adr is the affected destination pixels.
+	adr := dst.Bounds().Intersect(dr)
+	if adr.Empty() || sr.Empty() {
+		return
 	}
-	yHalfWidth, yKernelArgScale := q.Support, 1.0
-	if yscale > 1 {
-		yHalfWidth *= yscale
-		yKernelArgScale = 1 / yscale
+	op := opts.op()
+	if op == Over && opaque(src) { // TODO: also check that opts.SrcMask == nil.
+		op = Src
 	}
+	d2s := invert(s2d)
+	// bias is a translation of the mapping from dst co-ordinates to
+	// src co-ordinates such that the latter temporarily have
+	// non-negative X and Y co-ordinates. This allows us to write
+	// int(f) instead of int(math.Floor(f)), since "round to zero" and
+	// "round down" are equivalent when f >= 0, but the former is much
+	// cheaper. The X-- and Y-- are because the TransformLeaf methods
+	// have a "sx -= 0.5" adjustment.
+	bias := transformRect(&d2s, &adr).Min
+	bias.X--
+	bias.Y--
+	d2s[2] -= float64(bias.X)
+	d2s[5] -= float64(bias.Y)
+	// Make adr relative to dr.Min.
+	adr = adr.Sub(dr.Min)
 
-	xWeights := make([]float64, 1+2*int(math.Ceil(xHalfWidth)))
-	yWeights := make([]float64, 1+2*int(math.Ceil(yHalfWidth)))
-
-	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
-		dyf := float64(dr.Min.Y+int(dy)) + 0.5
-		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
-		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx, d = dx+1, d+4 {
-			dxf := float64(dr.Min.X+int(dx)) + 0.5
-			sx := d2s[0]*dxf + d2s[1]*dyf + d2s[2]
-			sy := d2s[3]*dxf + d2s[4]*dyf + d2s[5]
-			if !(image.Point{int(sx) + bias.X, int(sy) + bias.Y}).In(sr) {
-				continue
-			}
-
-			// TODO: adjust the bias so that we can use int(f) instead
-			// of math.Floor(f) and math.Ceil(f).
-			sx += float64(bias.X)
-			sx -= 0.5
-			ix := int(math.Floor(sx - xHalfWidth))
-			if ix < sr.Min.X {
-				ix = sr.Min.X
-			}
-			jx := int(math.Ceil(sx + xHalfWidth))
-			if jx > sr.Max.X {
-				jx = sr.Max.X
-			}
-
-			totalXWeight := 0.0
-			for kx := ix; kx < jx; kx++ {
-				xWeight := 0.0
-				if t := abs((sx - float64(kx)) * xKernelArgScale); t < q.Support {
-					xWeight = q.At(t)
-				}
-				xWeights[kx-ix] = xWeight
-				totalXWeight += xWeight
-			}
-			for x := range xWeights[:jx-ix] {
-				xWeights[x] /= totalXWeight
-			}
-
-			sy += float64(bias.Y)
-			sy -= 0.5
-			iy := int(math.Floor(sy - yHalfWidth))
-			if iy < sr.Min.Y {
-				iy = sr.Min.Y
-			}
-			jy := int(math.Ceil(sy + yHalfWidth))
-			if jy > sr.Max.Y {
-				jy = sr.Max.Y
-			}
-
-			totalYWeight := 0.0
-			for ky := iy; ky < jy; ky++ {
-				yWeight := 0.0
-				if t := abs((sy - float64(ky)) * yKernelArgScale); t < q.Support {
-					yWeight = q.At(t)
-				}
-				yWeights[ky-iy] = yWeight
-				totalYWeight += yWeight
-			}
-			for y := range yWeights[:jy-iy] {
-				yWeights[y] /= totalYWeight
-			}
-
-			var pr float64
-			for ky := iy; ky < jy; ky++ {
-				if yWeight := yWeights[ky-iy]; yWeight != 0 {
-					for kx := ix; kx < jx; kx++ {
-						if w := xWeights[kx-ix] * yWeight; w != 0 {
-							pi := (ky-src.Rect.Min.Y)*src.Stride + (kx - src.Rect.Min.X)
-							pru := uint32(src.Pix[pi]) * 0x101
-							pr += float64(pru) * w
-						}
-					}
-				}
-			}
-			out := uint8(fffftou(pr) >> 8)
-			dst.Pix[d+0] = out
-			dst.Pix[d+1] = out
-			dst.Pix[d+2] = out
-			dst.Pix[d+3] = 0xff
-		}
+	if u, ok := src.(*image.Uniform); ok && sr.In(src.Bounds()) {
+		transform_Uniform(dst, dr, adr, &d2s, u, sr, bias, op)
+		return
 	}
-}
 
-func (q *Kernel) transform_RGBA_Gray_Src(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.Gray, sr image.Rectangle, bias image.Point, xscale, yscale float64) {
-	// When shrinking, broaden the effective kernel support so that we still
-	// visit every source pixel.
-	xHalfWidth, xKernelArgScale := q.Support, 1.0
-	if xscale > 1 {
-		xHalfWidth *= xscale
-		xKernelArgScale = 1 / xscale
+	xscale := abs(d2s[0])
+	if s := abs(d2s[1]); xscale < s {
+		xscale = s
 	}
-	yHalfWidth, yKernelArgScale := q.Support, 1.0
-	if yscale > 1 {
-		yHalfWidth *= yscale
-		yKernelArgScale = 1 / yscale
+	yscale := abs(d2s[3])
+	if s := abs(d2s[4]); yscale < s {
+		yscale = s
 	}
 
-	xWeights := make([]float64, 1+2*int(math.Ceil(xHalfWidth)))
-	yWeights := make([]float64, 1+2*int(math.Ceil(yHalfWidth)))
-
-	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
-		dyf := float64(dr.Min.Y+int(dy)) + 0.5
-		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
-		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx, d = dx+1, d+4 {
-			dxf := float64(dr.Min.X+int(dx)) + 0.5
-			sx := d2s[0]*dxf + d2s[1]*dyf + d2s[2]
-			sy := d2s[3]*dxf + d2s[4]*dyf + d2s[5]
-			if !(image.Point{int(sx) + bias.X, int(sy) + bias.Y}).In(sr) {
-				continue
-			}
-
-			// TODO: adjust the bias so that we can use int(f) instead
-			// of math.Floor(f) and math.Ceil(f).
-			sx += float64(bias.X)
-			sx -= 0.5
-			ix := int(math.Floor(sx - xHalfWidth))
-			if ix < sr.Min.X {
-				ix = sr.Min.X
-			}
-			jx := int(math.Ceil(sx + xHalfWidth))
-			if jx > sr.Max.X {
-				jx = sr.Max.X
-			}
-
-			totalXWeight := 0.0
-			for kx := ix; kx < jx; kx++ {
-				xWeight := 0.0
-				if t := abs((sx - float64(kx)) * xKernelArgScale); t < q.Support {
-					xWeight = q.At(t)
+	// sr is the source pixels. If it extends beyond the src bounds,
+	// we cannot use the type-specific fast paths, as they access
+	// the Pix fields directly without bounds checking.
+	if !sr.In(src.Bounds()) {
+		switch op {
+		case Over:
+			q.transform_Image_Image_Over(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
+		case Src:
+			q.transform_Image_Image_Src(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
+		}
+	} else {
+		switch op {
+		case Over:
+			switch dst := dst.(type) {
+			case *image.RGBA:
+				switch src := src.(type) {
+				case *image.NRGBA:
+					q.transform_RGBA_NRGBA_Over(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
+				case *image.RGBA:
+					q.transform_RGBA_RGBA_Over(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
+				default:
+					q.transform_RGBA_Image_Over(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
 				}
-				xWeights[kx-ix] = xWeight
-				totalXWeight += xWeight
-			}
-			for x := range xWeights[:jx-ix] {
-				xWeights[x] /= totalXWeight
-			}
-
-			sy += float64(bias.Y)
-			sy -= 0.5
-			iy := int(math.Floor(sy - yHalfWidth))
-			if iy < sr.Min.Y {
-				iy = sr.Min.Y
-			}
-			jy := int(math.Ceil(sy + yHalfWidth))
-			if jy > sr.Max.Y {
-				jy = sr.Max.Y
-			}
-
-			totalYWeight := 0.0
-			for ky := iy; ky < jy; ky++ {
-				yWeight := 0.0
-				if t := abs((sy - float64(ky)) * yKernelArgScale); t < q.Support {
-					yWeight = q.At(t)
+			default:
+				switch src := src.(type) {
+				default:
+					q.transform_Image_Image_Over(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
 				}
-				yWeights[ky-iy] = yWeight
-				totalYWeight += yWeight
-			}
-			for y := range yWeights[:jy-iy] {
-				yWeights[y] /= totalYWeight
 			}
-
-			var pr float64
-			for ky := iy; ky < jy; ky++ {
-				if yWeight := yWeights[ky-iy]; yWeight != 0 {
-					for kx := ix; kx < jx; kx++ {
-						if w := xWeights[kx-ix] * yWeight; w != 0 {
-							pi := (ky-src.Rect.Min.Y)*src.Stride + (kx - src.Rect.Min.X)
-							pru := uint32(src.Pix[pi]) * 0x101
-							pr += float64(pru) * w
-						}
+		case Src:
+			switch dst := dst.(type) {
+			case *image.RGBA:
+				switch src := src.(type) {
+				case *image.Gray:
+					q.transform_RGBA_Gray_Src(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
+				case *image.NRGBA:
+					q.transform_RGBA_NRGBA_Src(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
+				case *image.RGBA:
+					q.transform_RGBA_RGBA_Src(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
+				case *image.YCbCr:
+					switch src.SubsampleRatio {
+					default:
+						q.transform_RGBA_Image_Src(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
+					case image.YCbCrSubsampleRatio444:
+						q.transform_RGBA_YCbCr444_Src(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
+					case image.YCbCrSubsampleRatio422:
+						q.transform_RGBA_YCbCr422_Src(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
+					case image.YCbCrSubsampleRatio420:
+						q.transform_RGBA_YCbCr420_Src(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
+					case image.YCbCrSubsampleRatio440:
+						q.transform_RGBA_YCbCr440_Src(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
 					}
+				default:
+					q.transform_RGBA_Image_Src(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
+				}
+			default:
+				switch src := src.(type) {
+				default:
+					q.transform_Image_Image_Src(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale)
 				}
 			}
-			out := uint8(fffftou(pr) >> 8)
-			dst.Pix[d+0] = out
-			dst.Pix[d+1] = out
-			dst.Pix[d+2] = out
-			dst.Pix[d+3] = 0xff
 		}
 	}
 }
 
-func (q *Kernel) transform_RGBA_NRGBA_Over(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.NRGBA, sr image.Rectangle, bias image.Point, xscale, yscale float64) {
-	// When shrinking, broaden the effective kernel support so that we still
-	// visit every source pixel.
-	xHalfWidth, xKernelArgScale := q.Support, 1.0
-	if xscale > 1 {
-		xHalfWidth *= xscale
-		xKernelArgScale = 1 / xscale
-	}
-	yHalfWidth, yKernelArgScale := q.Support, 1.0
-	if yscale > 1 {
-		yHalfWidth *= yscale
-		yKernelArgScale = 1 / yscale
-	}
-
-	xWeights := make([]float64, 1+2*int(math.Ceil(xHalfWidth)))
-	yWeights := make([]float64, 1+2*int(math.Ceil(yHalfWidth)))
-
-	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
-		dyf := float64(dr.Min.Y+int(dy)) + 0.5
-		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
-		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx, d = dx+1, d+4 {
-			dxf := float64(dr.Min.X+int(dx)) + 0.5
-			sx := d2s[0]*dxf + d2s[1]*dyf + d2s[2]
-			sy := d2s[3]*dxf + d2s[4]*dyf + d2s[5]
-			if !(image.Point{int(sx) + bias.X, int(sy) + bias.Y}).In(sr) {
-				continue
-			}
-
-			// TODO: adjust the bias so that we can use int(f) instead
-			// of math.Floor(f) and math.Ceil(f).
-			sx += float64(bias.X)
-			sx -= 0.5
-			ix := int(math.Floor(sx - xHalfWidth))
-			if ix < sr.Min.X {
-				ix = sr.Min.X
-			}
-			jx := int(math.Ceil(sx + xHalfWidth))
-			if jx > sr.Max.X {
-				jx = sr.Max.X
-			}
-
-			totalXWeight := 0.0
-			for kx := ix; kx < jx; kx++ {
-				xWeight := 0.0
-				if t := abs((sx - float64(kx)) * xKernelArgScale); t < q.Support {
-					xWeight = q.At(t)
-				}
-				xWeights[kx-ix] = xWeight
-				totalXWeight += xWeight
-			}
-			for x := range xWeights[:jx-ix] {
-				xWeights[x] /= totalXWeight
-			}
-
-			sy += float64(bias.Y)
-			sy -= 0.5
-			iy := int(math.Floor(sy - yHalfWidth))
-			if iy < sr.Min.Y {
-				iy = sr.Min.Y
+func (z *kernelScaler) scaleX_Gray(tmp [][4]float64, src *image.Gray, sr image.Rectangle) {
+	t := 0
+	for y := int32(0); y < z.sh; y++ {
+		for _, s := range z.horizontal.sources {
+			var pr float64
+			for _, c := range z.horizontal.contribs[s.i:s.j] {
+				pi := (sr.Min.Y+int(y)-src.Rect.Min.Y)*src.Stride + (sr.Min.X + int(c.coord) - src.Rect.Min.X)
+				pru := uint32(src.Pix[pi]) * 0x101
+				pr += float64(pru) * c.weight
 			}
-			jy := int(math.Ceil(sy + yHalfWidth))
-			if jy > sr.Max.Y {
-				jy = sr.Max.Y
+			pr *= s.invTotalWeightFFFF
+			tmp[t] = [4]float64{
+				pr,
+				pr,
+				pr,
+				1,
 			}
+			t++
+		}
+	}
+}
 
-			totalYWeight := 0.0
-			for ky := iy; ky < jy; ky++ {
-				yWeight := 0.0
-				if t := abs((sy - float64(ky)) * yKernelArgScale); t < q.Support {
-					yWeight = q.At(t)
-				}
-				yWeights[ky-iy] = yWeight
-				totalYWeight += yWeight
+func (z *kernelScaler) scaleX_NRGBA(tmp [][4]float64, src *image.NRGBA, sr image.Rectangle) {
+	t := 0
+	for y := int32(0); y < z.sh; y++ {
+		for _, s := range z.horizontal.sources {
+			var pr, pg, pb, pa float64
+			for _, c := range z.horizontal.contribs[s.i:s.j] {
+				pi := (sr.Min.Y+int(y)-src.Rect.Min.Y)*src.Stride + (sr.Min.X+int(c.coord)-src.Rect.Min.X)*4
+				pau := uint32(src.Pix[pi+3]) * 0x101
+				pru := uint32(src.Pix[pi+0]) * pau / 0xff
+				pgu := uint32(src.Pix[pi+1]) * pau / 0xff
+				pbu := uint32(src.Pix[pi+2]) * pau / 0xff
+				pr += float64(pru) * c.weight
+				pg += float64(pgu) * c.weight
+				pb += float64(pbu) * c.weight
+				pa += float64(pau) * c.weight
 			}
-			for y := range yWeights[:jy-iy] {
-				yWeights[y] /= totalYWeight
+			tmp[t] = [4]float64{
+				pr * s.invTotalWeightFFFF,
+				pg * s.invTotalWeightFFFF,
+				pb * s.invTotalWeightFFFF,
+				pa * s.invTotalWeightFFFF,
 			}
+			t++
+		}
+	}
+}
 
+func (z *kernelScaler) scaleX_RGBA(tmp [][4]float64, src *image.RGBA, sr image.Rectangle) {
+	t := 0
+	for y := int32(0); y < z.sh; y++ {
+		for _, s := range z.horizontal.sources {
 			var pr, pg, pb, pa float64
-			for ky := iy; ky < jy; ky++ {
-				if yWeight := yWeights[ky-iy]; yWeight != 0 {
-					for kx := ix; kx < jx; kx++ {
-						if w := xWeights[kx-ix] * yWeight; w != 0 {
-							pi := (ky-src.Rect.Min.Y)*src.Stride + (kx-src.Rect.Min.X)*4
-							pau := uint32(src.Pix[pi+3]) * 0x101
-							pru := uint32(src.Pix[pi+0]) * pau / 0xff
-							pgu := uint32(src.Pix[pi+1]) * pau / 0xff
-							pbu := uint32(src.Pix[pi+2]) * pau / 0xff
-							pr += float64(pru) * w
-							pg += float64(pgu) * w
-							pb += float64(pbu) * w
-							pa += float64(pau) * w
-						}
-					}
-				}
+			for _, c := range z.horizontal.contribs[s.i:s.j] {
+				pi := (sr.Min.Y+int(y)-src.Rect.Min.Y)*src.Stride + (sr.Min.X+int(c.coord)-src.Rect.Min.X)*4
+				pru := uint32(src.Pix[pi+0]) * 0x101
+				pgu := uint32(src.Pix[pi+1]) * 0x101
+				pbu := uint32(src.Pix[pi+2]) * 0x101
+				pau := uint32(src.Pix[pi+3]) * 0x101
+				pr += float64(pru) * c.weight
+				pg += float64(pgu) * c.weight
+				pb += float64(pbu) * c.weight
+				pa += float64(pau) * c.weight
 			}
-			dst.Pix[d+0] = uint8(fffftou(pr) >> 8)
-			dst.Pix[d+1] = uint8(fffftou(pg) >> 8)
-			dst.Pix[d+2] = uint8(fffftou(pb) >> 8)
-			dst.Pix[d+3] = uint8(fffftou(pa) >> 8)
+			tmp[t] = [4]float64{
+				pr * s.invTotalWeightFFFF,
+				pg * s.invTotalWeightFFFF,
+				pb * s.invTotalWeightFFFF,
+				pa * s.invTotalWeightFFFF,
+			}
+			t++
 		}
 	}
 }
 
-func (q *Kernel) transform_RGBA_NRGBA_Src(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.NRGBA, sr image.Rectangle, bias image.Point, xscale, yscale float64) {
-	// When shrinking, broaden the effective kernel support so that we still
-	// visit every source pixel.
-	xHalfWidth, xKernelArgScale := q.Support, 1.0
-	if xscale > 1 {
-		xHalfWidth *= xscale
-		xKernelArgScale = 1 / xscale
-	}
-	yHalfWidth, yKernelArgScale := q.Support, 1.0
-	if yscale > 1 {
-		yHalfWidth *= yscale
-		yKernelArgScale = 1 / yscale
-	}
-
-	xWeights := make([]float64, 1+2*int(math.Ceil(xHalfWidth)))
-	yWeights := make([]float64, 1+2*int(math.Ceil(yHalfWidth)))
-
-	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
-		dyf := float64(dr.Min.Y+int(dy)) + 0.5
-		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
-		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx, d = dx+1, d+4 {
-			dxf := float64(dr.Min.X+int(dx)) + 0.5
-			sx := d2s[0]*dxf + d2s[1]*dyf + d2s[2]
-			sy := d2s[3]*dxf + d2s[4]*dyf + d2s[5]
-			if !(image.Point{int(sx) + bias.X, int(sy) + bias.Y}).In(sr) {
-				continue
-			}
-
-			// TODO: adjust the bias so that we can use int(f) instead
-			// of math.Floor(f) and math.Ceil(f).
-			sx += float64(bias.X)
-			sx -= 0.5
-			ix := int(math.Floor(sx - xHalfWidth))
-			if ix < sr.Min.X {
-				ix = sr.Min.X
-			}
-			jx := int(math.Ceil(sx + xHalfWidth))
-			if jx > sr.Max.X {
-				jx = sr.Max.X
-			}
+func (z *kernelScaler) scaleX_YCbCr444(tmp [][4]float64, src *image.YCbCr, sr image.Rectangle) {
+	t := 0
+	for y := int32(0); y < z.sh; y++ {
+		for _, s := range z.horizontal.sources {
+			var pr, pg, pb float64
+			for _, c := range z.horizontal.contribs[s.i:s.j] {
+				pi := (sr.Min.Y+int(y)-src.Rect.Min.Y)*src.YStride + (sr.Min.X + int(c.coord) - src.Rect.Min.X)
+				pj := (sr.Min.Y+int(y)-src.Rect.Min.Y)*src.CStride + (sr.Min.X + int(c.coord) - src.Rect.Min.X)
 
-			totalXWeight := 0.0
-			for kx := ix; kx < jx; kx++ {
-				xWeight := 0.0
-				if t := abs((sx - float64(kx)) * xKernelArgScale); t < q.Support {
-					xWeight = q.At(t)
+				// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
+				pyy1 := int(src.Y[pi])<<16 + 1<<15
+				pcb1 := int(src.Cb[pj]) - 128
+				pcr1 := int(src.Cr[pj]) - 128
+				pru := (pyy1 + 91881*pcr1) >> 8
+				pgu := (pyy1 - 22554*pcb1 - 46802*pcr1) >> 8
+				pbu := (pyy1 + 116130*pcb1) >> 8
+				if pru < 0 {
+					pru = 0
+				} else if pru > 0xffff {
+					pru = 0xffff
+				}
+				if pgu < 0 {
+					pgu = 0
+				} else if pgu > 0xffff {
+					pgu = 0xffff
+				}
+				if pbu < 0 {
+					pbu = 0
+				} else if pbu > 0xffff {
+					pbu = 0xffff
 				}
-				xWeights[kx-ix] = xWeight
-				totalXWeight += xWeight
-			}
-			for x := range xWeights[:jx-ix] {
-				xWeights[x] /= totalXWeight
-			}
 
-			sy += float64(bias.Y)
-			sy -= 0.5
-			iy := int(math.Floor(sy - yHalfWidth))
-			if iy < sr.Min.Y {
-				iy = sr.Min.Y
+				pr += float64(pru) * c.weight
+				pg += float64(pgu) * c.weight
+				pb += float64(pbu) * c.weight
 			}
-			jy := int(math.Ceil(sy + yHalfWidth))
-			if jy > sr.Max.Y {
-				jy = sr.Max.Y
+			tmp[t] = [4]float64{
+				pr * s.invTotalWeightFFFF,
+				pg * s.invTotalWeightFFFF,
+				pb * s.invTotalWeightFFFF,
+				1,
 			}
+			t++
+		}
+	}
+}
 
-			totalYWeight := 0.0
-			for ky := iy; ky < jy; ky++ {
-				yWeight := 0.0
-				if t := abs((sy - float64(ky)) * yKernelArgScale); t < q.Support {
-					yWeight = q.At(t)
+func (z *kernelScaler) scaleX_YCbCr422(tmp [][4]float64, src *image.YCbCr, sr image.Rectangle) {
+	t := 0
+	for y := int32(0); y < z.sh; y++ {
+		for _, s := range z.horizontal.sources {
+			var pr, pg, pb float64
+			for _, c := range z.horizontal.contribs[s.i:s.j] {
+				pi := (sr.Min.Y+int(y)-src.Rect.Min.Y)*src.YStride + (sr.Min.X + int(c.coord) - src.Rect.Min.X)
+				pj := (sr.Min.Y+int(y)-src.Rect.Min.Y)*src.CStride + ((sr.Min.X+int(c.coord))/2 - src.Rect.Min.X/2)
+
+				// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
+				pyy1 := int(src.Y[pi])<<16 + 1<<15
+				pcb1 := int(src.Cb[pj]) - 128
+				pcr1 := int(src.Cr[pj]) - 128
+				pru := (pyy1 + 91881*pcr1) >> 8
+				pgu := (pyy1 - 22554*pcb1 - 46802*pcr1) >> 8
+				pbu := (pyy1 + 116130*pcb1) >> 8
+				if pru < 0 {
+					pru = 0
+				} else if pru > 0xffff {
+					pru = 0xffff
 				}
-				yWeights[ky-iy] = yWeight
-				totalYWeight += yWeight
+				if pgu < 0 {
+					pgu = 0
+				} else if pgu > 0xffff {
+					pgu = 0xffff
+				}
+				if pbu < 0 {
+					pbu = 0
+				} else if pbu > 0xffff {
+					pbu = 0xffff
+				}
+
+				pr += float64(pru) * c.weight
+				pg += float64(pgu) * c.weight
+				pb += float64(pbu) * c.weight
 			}
-			for y := range yWeights[:jy-iy] {
-				yWeights[y] /= totalYWeight
+			tmp[t] = [4]float64{
+				pr * s.invTotalWeightFFFF,
+				pg * s.invTotalWeightFFFF,
+				pb * s.invTotalWeightFFFF,
+				1,
 			}
+			t++
+		}
+	}
+}
 
-			var pr, pg, pb, pa float64
-			for ky := iy; ky < jy; ky++ {
-				if yWeight := yWeights[ky-iy]; yWeight != 0 {
-					for kx := ix; kx < jx; kx++ {
-						if w := xWeights[kx-ix] * yWeight; w != 0 {
-							pi := (ky-src.Rect.Min.Y)*src.Stride + (kx-src.Rect.Min.X)*4
-							pau := uint32(src.Pix[pi+3]) * 0x101
-							pru := uint32(src.Pix[pi+0]) * pau / 0xff
-							pgu := uint32(src.Pix[pi+1]) * pau / 0xff
-							pbu := uint32(src.Pix[pi+2]) * pau / 0xff
-							pr += float64(pru) * w
-							pg += float64(pgu) * w
-							pb += float64(pbu) * w
-							pa += float64(pau) * w
-						}
-					}
+func (z *kernelScaler) scaleX_YCbCr420(tmp [][4]float64, src *image.YCbCr, sr image.Rectangle) {
+	t := 0
+	for y := int32(0); y < z.sh; y++ {
+		for _, s := range z.horizontal.sources {
+			var pr, pg, pb float64
+			for _, c := range z.horizontal.contribs[s.i:s.j] {
+				pi := (sr.Min.Y+int(y)-src.Rect.Min.Y)*src.YStride + (sr.Min.X + int(c.coord) - src.Rect.Min.X)
+				pj := ((sr.Min.Y+int(y))/2-src.Rect.Min.Y/2)*src.CStride + ((sr.Min.X+int(c.coord))/2 - src.Rect.Min.X/2)
+
+				// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
+				pyy1 := int(src.Y[pi])<<16 + 1<<15
+				pcb1 := int(src.Cb[pj]) - 128
+				pcr1 := int(src.Cr[pj]) - 128
+				pru := (pyy1 + 91881*pcr1) >> 8
+				pgu := (pyy1 - 22554*pcb1 - 46802*pcr1) >> 8
+				pbu := (pyy1 + 116130*pcb1) >> 8
+				if pru < 0 {
+					pru = 0
+				} else if pru > 0xffff {
+					pru = 0xffff
+				}
+				if pgu < 0 {
+					pgu = 0
+				} else if pgu > 0xffff {
+					pgu = 0xffff
+				}
+				if pbu < 0 {
+					pbu = 0
+				} else if pbu > 0xffff {
+					pbu = 0xffff
 				}
+
+				pr += float64(pru) * c.weight
+				pg += float64(pgu) * c.weight
+				pb += float64(pbu) * c.weight
 			}
-			dst.Pix[d+0] = uint8(fffftou(pr) >> 8)
-			dst.Pix[d+1] = uint8(fffftou(pg) >> 8)
-			dst.Pix[d+2] = uint8(fffftou(pb) >> 8)
-			dst.Pix[d+3] = uint8(fffftou(pa) >> 8)
+			tmp[t] = [4]float64{
+				pr * s.invTotalWeightFFFF,
+				pg * s.invTotalWeightFFFF,
+				pb * s.invTotalWeightFFFF,
+				1,
+			}
+			t++
 		}
 	}
 }
 
-func (q *Kernel) transform_RGBA_RGBA_Over(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.RGBA, sr image.Rectangle, bias image.Point, xscale, yscale float64) {
-	// When shrinking, broaden the effective kernel support so that we still
-	// visit every source pixel.
-	xHalfWidth, xKernelArgScale := q.Support, 1.0
-	if xscale > 1 {
-		xHalfWidth *= xscale
-		xKernelArgScale = 1 / xscale
-	}
-	yHalfWidth, yKernelArgScale := q.Support, 1.0
-	if yscale > 1 {
-		yHalfWidth *= yscale
-		yKernelArgScale = 1 / yscale
-	}
-
-	xWeights := make([]float64, 1+2*int(math.Ceil(xHalfWidth)))
-	yWeights := make([]float64, 1+2*int(math.Ceil(yHalfWidth)))
+func (z *kernelScaler) scaleX_YCbCr440(tmp [][4]float64, src *image.YCbCr, sr image.Rectangle) {
+	t := 0
+	for y := int32(0); y < z.sh; y++ {
+		for _, s := range z.horizontal.sources {
+			var pr, pg, pb float64
+			for _, c := range z.horizontal.contribs[s.i:s.j] {
+				pi := (sr.Min.Y+int(y)-src.Rect.Min.Y)*src.YStride + (sr.Min.X + int(c.coord) - src.Rect.Min.X)
+				pj := ((sr.Min.Y+int(y))/2-src.Rect.Min.Y/2)*src.CStride + (sr.Min.X + int(c.coord) - src.Rect.Min.X)
 
-	for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
-		dyf := float64(dr.Min.Y+int(dy)) + 0.5
-		d := (dr.Min.Y+int(dy)-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+adr.Min.X-dst.Rect.Min.X)*4
-		for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx, d = dx+1, d+4 {
-			dxf := float64(dr.Min.X+int(dx)) + 0.5
-			sx := d2s[0]*dxf + d2s[1]*dyf + d2s[2]
-			sy := d2s[3]*dxf + d2s[4]*dyf + d2s[5]
-			if !(image.Point{int(sx) + bias.X, int(sy) + bias.Y}).In(sr) {
-				continue
-			}
+				// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
+				pyy1 := int(src.Y[pi])<<16 + 1<<15
+				pcb1 := int(src.Cb[pj]) - 128
+				pcr1 := int(src.Cr[pj]) - 128
+				pru := (pyy1 + 91881*pcr1) >> 8
+				pgu := (pyy1 - 22554*pcb1 - 46802*pcr1) >> 8
+				pbu := (pyy1 + 116130*pcb1) >> 8
+				if pru < 0 {
+					pru = 0
+				} else if pru > 0xffff {
+					pru = 0xffff
+				}
+				if pgu < 0 {
+					pgu = 0
+				} else if pgu > 0xffff {
+					pgu = 0xffff
+				}
+				if pbu < 0 {
+					pbu = 0
+				} else if pbu > 0xffff {
+					pbu = 0xffff
+				}
 
-			// TODO: adjust the bias so that we can use int(f) instead
-			// of math.Floor(f) and math.Ceil(f).
-			sx += float64(bias.X)
-			sx -= 0.5
-			ix := int(math.Floor(sx - xHalfWidth))
-			if ix < sr.Min.X {
-				ix = sr.Min.X
+				pr += float64(pru) * c.weight
+				pg += float64(pgu) * c.weight
+				pb += float64(pbu) * c.weight
 			}
-			jx := int(math.Ceil(sx + xHalfWidth))
-			if jx > sr.Max.X {
-				jx = sr.Max.X
+			tmp[t] = [4]float64{
+				pr * s.invTotalWeightFFFF,
+				pg * s.invTotalWeightFFFF,
+				pb * s.invTotalWeightFFFF,
+				1,
 			}
+			t++
+		}
+	}
+}
 
-			totalXWeight := 0.0
-			for kx := ix; kx < jx; kx++ {
-				xWeight := 0.0
-				if t := abs((sx - float64(kx)) * xKernelArgScale); t < q.Support {
-					xWeight = q.At(t)
-				}
-				xWeights[kx-ix] = xWeight
-				totalXWeight += xWeight
+func (z *kernelScaler) scaleX_Image(tmp [][4]float64, src image.Image, sr image.Rectangle) {
+	t := 0
+	for y := int32(0); y < z.sh; y++ {
+		for _, s := range z.horizontal.sources {
+			var pr, pg, pb, pa float64
+			for _, c := range z.horizontal.contribs[s.i:s.j] {
+				pru, pgu, pbu, pau := src.At(sr.Min.X+int(c.coord), sr.Min.Y+int(y)).RGBA()
+				pr += float64(pru) * c.weight
+				pg += float64(pgu) * c.weight
+				pb += float64(pbu) * c.weight
+				pa += float64(pau) * c.weight
 			}
-			for x := range xWeights[:jx-ix] {
-				xWeights[x] /= totalXWeight
+			tmp[t] = [4]float64{
+				pr * s.invTotalWeightFFFF,
+				pg * s.invTotalWeightFFFF,
+				pb * s.invTotalWeightFFFF,
+				pa * s.invTotalWeightFFFF,
 			}
+			t++
+		}
+	}
+}
 
-			sy += float64(bias.Y)
-			sy -= 0.5
-			iy := int(math.Floor(sy - yHalfWidth))
-			if iy < sr.Min.Y {
-				iy = sr.Min.Y
+func (z *kernelScaler) scaleY_RGBA_Over(dst *image.RGBA, dr, adr image.Rectangle, tmp [][4]float64) {
+	for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx++ {
+		d := (dr.Min.Y+adr.Min.Y-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+int(dx)-dst.Rect.Min.X)*4
+		for _, s := range z.vertical.sources[adr.Min.Y:adr.Max.Y] {
+			var pr, pg, pb, pa float64
+			for _, c := range z.vertical.contribs[s.i:s.j] {
+				p := &tmp[c.coord*z.dw+dx]
+				pr += p[0] * c.weight
+				pg += p[1] * c.weight
+				pb += p[2] * c.weight
+				pa += p[3] * c.weight
 			}
-			jy := int(math.Ceil(sy + yHalfWidth))
-			if jy > sr.Max.Y {
-				jy = sr.Max.Y
+			dst.Pix[d+0] = uint8(ftou(pr*s.invTotalWeight) >> 8)
+			dst.Pix[d+1] = uint8(ftou(pg*s.invTotalWeight) >> 8)
+			dst.Pix[d+2] = uint8(ftou(pb*s.invTotalWeight) >> 8)
+			dst.Pix[d+3] = uint8(ftou(pa*s.invTotalWeight) >> 8)
+			d += dst.Stride
+		}
+	}
+}
+
+func (z *kernelScaler) scaleY_RGBA_Src(dst *image.RGBA, dr, adr image.Rectangle, tmp [][4]float64) {
+	for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx++ {
+		d := (dr.Min.Y+adr.Min.Y-dst.Rect.Min.Y)*dst.Stride + (dr.Min.X+int(dx)-dst.Rect.Min.X)*4
+		for _, s := range z.vertical.sources[adr.Min.Y:adr.Max.Y] {
+			var pr, pg, pb, pa float64
+			for _, c := range z.vertical.contribs[s.i:s.j] {
+				p := &tmp[c.coord*z.dw+dx]
+				pr += p[0] * c.weight
+				pg += p[1] * c.weight
+				pb += p[2] * c.weight
+				pa += p[3] * c.weight
 			}
+			dst.Pix[d+0] = uint8(ftou(pr*s.invTotalWeight) >> 8)
+			dst.Pix[d+1] = uint8(ftou(pg*s.invTotalWeight) >> 8)
+			dst.Pix[d+2] = uint8(ftou(pb*s.invTotalWeight) >> 8)
+			dst.Pix[d+3] = uint8(ftou(pa*s.invTotalWeight) >> 8)
+			d += dst.Stride
+		}
+	}
+}
 
-			totalYWeight := 0.0
-			for ky := iy; ky < jy; ky++ {
-				yWeight := 0.0
-				if t := abs((sy - float64(ky)) * yKernelArgScale); t < q.Support {
-					yWeight = q.At(t)
-				}
-				yWeights[ky-iy] = yWeight
-				totalYWeight += yWeight
-			}
-			for y := range yWeights[:jy-iy] {
-				yWeights[y] /= totalYWeight
+func (z *kernelScaler) scaleY_Image_Over(dst Image, dr, adr image.Rectangle, tmp [][4]float64) {
+	dstColorRGBA64 := &color.RGBA64{}
+	dstColor := color.Color(dstColorRGBA64)
+	for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx++ {
+		for dy, s := range z.vertical.sources[adr.Min.Y:adr.Max.Y] {
+			var pr, pg, pb, pa float64
+			for _, c := range z.vertical.contribs[s.i:s.j] {
+				p := &tmp[c.coord*z.dw+dx]
+				pr += p[0] * c.weight
+				pg += p[1] * c.weight
+				pb += p[2] * c.weight
+				pa += p[3] * c.weight
 			}
+			dstColorRGBA64.R = ftou(pr * s.invTotalWeight)
+			dstColorRGBA64.G = ftou(pg * s.invTotalWeight)
+			dstColorRGBA64.B = ftou(pb * s.invTotalWeight)
+			dstColorRGBA64.A = ftou(pa * s.invTotalWeight)
+			dst.Set(dr.Min.X+int(dx), dr.Min.Y+int(adr.Min.Y+dy), dstColor)
+		}
+	}
+}
 
+func (z *kernelScaler) scaleY_Image_Src(dst Image, dr, adr image.Rectangle, tmp [][4]float64) {
+	dstColorRGBA64 := &color.RGBA64{}
+	dstColor := color.Color(dstColorRGBA64)
+	for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx++ {
+		for dy, s := range z.vertical.sources[adr.Min.Y:adr.Max.Y] {
 			var pr, pg, pb, pa float64
-			for ky := iy; ky < jy; ky++ {
-				if yWeight := yWeights[ky-iy]; yWeight != 0 {
-					for kx := ix; kx < jx; kx++ {
-						if w := xWeights[kx-ix] * yWeight; w != 0 {
-							pi := (ky-src.Rect.Min.Y)*src.Stride + (kx-src.Rect.Min.X)*4
-							pru := uint32(src.Pix[pi+0]) * 0x101
-							pgu := uint32(src.Pix[pi+1]) * 0x101
-							pbu := uint32(src.Pix[pi+2]) * 0x101
-							pau := uint32(src.Pix[pi+3]) * 0x101
-							pr += float64(pru) * w
-							pg += float64(pgu) * w
-							pb += float64(pbu) * w
-							pa += float64(pau) * w
-						}
-					}
-				}
+			for _, c := range z.vertical.contribs[s.i:s.j] {
+				p := &tmp[c.coord*z.dw+dx]
+				pr += p[0] * c.weight
+				pg += p[1] * c.weight
+				pb += p[2] * c.weight
+				pa += p[3] * c.weight
 			}
-			dst.Pix[d+0] = uint8(fffftou(pr) >> 8)
-			dst.Pix[d+1] = uint8(fffftou(pg) >> 8)
-			dst.Pix[d+2] = uint8(fffftou(pb) >> 8)
-			dst.Pix[d+3] = uint8(fffftou(pa) >> 8)
+			dstColorRGBA64.R = ftou(pr * s.invTotalWeight)
+			dstColorRGBA64.G = ftou(pg * s.invTotalWeight)
+			dstColorRGBA64.B = ftou(pb * s.invTotalWeight)
+			dstColorRGBA64.A = ftou(pa * s.invTotalWeight)
+			dst.Set(dr.Min.X+int(dx), dr.Min.Y+int(adr.Min.Y+dy), dstColor)
 		}
 	}
 }
 
-func (q *Kernel) transform_RGBA_RGBA_Src(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.RGBA, sr image.Rectangle, bias image.Point, xscale, yscale float64) {
+func (q *Kernel) transform_RGBA_Gray_Src(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.Gray, sr image.Rectangle, bias image.Point, xscale, yscale float64) {
 	// When shrinking, broaden the effective kernel support so that we still
 	// visit every source pixel.
 	xHalfWidth, xKernelArgScale := q.Support, 1.0
@@ -7105,33 +4646,28 @@ func (q *Kernel) transform_RGBA_RGBA_Src(dst *image.RGBA, dr, adr image.Rectangl
 				yWeights[y] /= totalYWeight
 			}
 
-			var pr, pg, pb, pa float64
+			var pr float64
 			for ky := iy; ky < jy; ky++ {
 				if yWeight := yWeights[ky-iy]; yWeight != 0 {
 					for kx := ix; kx < jx; kx++ {
 						if w := xWeights[kx-ix] * yWeight; w != 0 {
-							pi := (ky-src.Rect.Min.Y)*src.Stride + (kx-src.Rect.Min.X)*4
-							pru := uint32(src.Pix[pi+0]) * 0x101
-							pgu := uint32(src.Pix[pi+1]) * 0x101
-							pbu := uint32(src.Pix[pi+2]) * 0x101
-							pau := uint32(src.Pix[pi+3]) * 0x101
+							pi := (ky-src.Rect.Min.Y)*src.Stride + (kx - src.Rect.Min.X)
+							pru := uint32(src.Pix[pi]) * 0x101
 							pr += float64(pru) * w
-							pg += float64(pgu) * w
-							pb += float64(pbu) * w
-							pa += float64(pau) * w
 						}
 					}
 				}
 			}
-			dst.Pix[d+0] = uint8(fffftou(pr) >> 8)
-			dst.Pix[d+1] = uint8(fffftou(pg) >> 8)
-			dst.Pix[d+2] = uint8(fffftou(pb) >> 8)
-			dst.Pix[d+3] = uint8(fffftou(pa) >> 8)
+			out := uint8(fffftou(pr) >> 8)
+			dst.Pix[d+0] = out
+			dst.Pix[d+1] = out
+			dst.Pix[d+2] = out
+			dst.Pix[d+3] = 0xff
 		}
 	}
 }
 
-func (q *Kernel) transform_RGBA_YCbCr444_Over(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.YCbCr, sr image.Rectangle, bias image.Point, xscale, yscale float64) {
+func (q *Kernel) transform_RGBA_NRGBA_Over(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.NRGBA, sr image.Rectangle, bias image.Point, xscale, yscale float64) {
 	// When shrinking, broaden the effective kernel support so that we still
 	// visit every source pixel.
 	xHalfWidth, xKernelArgScale := q.Support, 1.0
@@ -7209,40 +4745,20 @@ func (q *Kernel) transform_RGBA_YCbCr444_Over(dst *image.RGBA, dr, adr image.Rec
 				yWeights[y] /= totalYWeight
 			}
 
-			var pr, pg, pb float64
+			var pr, pg, pb, pa float64
 			for ky := iy; ky < jy; ky++ {
 				if yWeight := yWeights[ky-iy]; yWeight != 0 {
 					for kx := ix; kx < jx; kx++ {
 						if w := xWeights[kx-ix] * yWeight; w != 0 {
-							pi := (ky-src.Rect.Min.Y)*src.YStride + (kx - src.Rect.Min.X)
-							pj := (ky-src.Rect.Min.Y)*src.CStride + (kx - src.Rect.Min.X)
-
-							// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-							pyy1 := int(src.Y[pi])<<16 + 1<<15
-							pcb1 := int(src.Cb[pj]) - 128
-							pcr1 := int(src.Cr[pj]) - 128
-							pru := (pyy1 + 91881*pcr1) >> 8
-							pgu := (pyy1 - 22554*pcb1 - 46802*pcr1) >> 8
-							pbu := (pyy1 + 116130*pcb1) >> 8
-							if pru < 0 {
-								pru = 0
-							} else if pru > 0xffff {
-								pru = 0xffff
-							}
-							if pgu < 0 {
-								pgu = 0
-							} else if pgu > 0xffff {
-								pgu = 0xffff
-							}
-							if pbu < 0 {
-								pbu = 0
-							} else if pbu > 0xffff {
-								pbu = 0xffff
-							}
-
+							pi := (ky-src.Rect.Min.Y)*src.Stride + (kx-src.Rect.Min.X)*4
+							pau := uint32(src.Pix[pi+3]) * 0x101
+							pru := uint32(src.Pix[pi+0]) * pau / 0xff
+							pgu := uint32(src.Pix[pi+1]) * pau / 0xff
+							pbu := uint32(src.Pix[pi+2]) * pau / 0xff
 							pr += float64(pru) * w
 							pg += float64(pgu) * w
 							pb += float64(pbu) * w
+							pa += float64(pau) * w
 						}
 					}
 				}
@@ -7250,12 +4766,12 @@ func (q *Kernel) transform_RGBA_YCbCr444_Over(dst *image.RGBA, dr, adr image.Rec
 			dst.Pix[d+0] = uint8(fffftou(pr) >> 8)
 			dst.Pix[d+1] = uint8(fffftou(pg) >> 8)
 			dst.Pix[d+2] = uint8(fffftou(pb) >> 8)
-			dst.Pix[d+3] = 0xff
+			dst.Pix[d+3] = uint8(fffftou(pa) >> 8)
 		}
 	}
 }
 
-func (q *Kernel) transform_RGBA_YCbCr422_Over(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.YCbCr, sr image.Rectangle, bias image.Point, xscale, yscale float64) {
+func (q *Kernel) transform_RGBA_NRGBA_Src(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.NRGBA, sr image.Rectangle, bias image.Point, xscale, yscale float64) {
 	// When shrinking, broaden the effective kernel support so that we still
 	// visit every source pixel.
 	xHalfWidth, xKernelArgScale := q.Support, 1.0
@@ -7333,40 +4849,20 @@ func (q *Kernel) transform_RGBA_YCbCr422_Over(dst *image.RGBA, dr, adr image.Rec
 				yWeights[y] /= totalYWeight
 			}
 
-			var pr, pg, pb float64
+			var pr, pg, pb, pa float64
 			for ky := iy; ky < jy; ky++ {
 				if yWeight := yWeights[ky-iy]; yWeight != 0 {
 					for kx := ix; kx < jx; kx++ {
 						if w := xWeights[kx-ix] * yWeight; w != 0 {
-							pi := (ky-src.Rect.Min.Y)*src.YStride + (kx - src.Rect.Min.X)
-							pj := (ky-src.Rect.Min.Y)*src.CStride + ((kx)/2 - src.Rect.Min.X/2)
-
-							// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-							pyy1 := int(src.Y[pi])<<16 + 1<<15
-							pcb1 := int(src.Cb[pj]) - 128
-							pcr1 := int(src.Cr[pj]) - 128
-							pru := (pyy1 + 91881*pcr1) >> 8
-							pgu := (pyy1 - 22554*pcb1 - 46802*pcr1) >> 8
-							pbu := (pyy1 + 116130*pcb1) >> 8
-							if pru < 0 {
-								pru = 0
-							} else if pru > 0xffff {
-								pru = 0xffff
-							}
-							if pgu < 0 {
-								pgu = 0
-							} else if pgu > 0xffff {
-								pgu = 0xffff
-							}
-							if pbu < 0 {
-								pbu = 0
-							} else if pbu > 0xffff {
-								pbu = 0xffff
-							}
-
+							pi := (ky-src.Rect.Min.Y)*src.Stride + (kx-src.Rect.Min.X)*4
+							pau := uint32(src.Pix[pi+3]) * 0x101
+							pru := uint32(src.Pix[pi+0]) * pau / 0xff
+							pgu := uint32(src.Pix[pi+1]) * pau / 0xff
+							pbu := uint32(src.Pix[pi+2]) * pau / 0xff
 							pr += float64(pru) * w
 							pg += float64(pgu) * w
 							pb += float64(pbu) * w
+							pa += float64(pau) * w
 						}
 					}
 				}
@@ -7374,12 +4870,12 @@ func (q *Kernel) transform_RGBA_YCbCr422_Over(dst *image.RGBA, dr, adr image.Rec
 			dst.Pix[d+0] = uint8(fffftou(pr) >> 8)
 			dst.Pix[d+1] = uint8(fffftou(pg) >> 8)
 			dst.Pix[d+2] = uint8(fffftou(pb) >> 8)
-			dst.Pix[d+3] = 0xff
+			dst.Pix[d+3] = uint8(fffftou(pa) >> 8)
 		}
 	}
 }
 
-func (q *Kernel) transform_RGBA_YCbCr420_Over(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.YCbCr, sr image.Rectangle, bias image.Point, xscale, yscale float64) {
+func (q *Kernel) transform_RGBA_RGBA_Over(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.RGBA, sr image.Rectangle, bias image.Point, xscale, yscale float64) {
 	// When shrinking, broaden the effective kernel support so that we still
 	// visit every source pixel.
 	xHalfWidth, xKernelArgScale := q.Support, 1.0
@@ -7457,40 +4953,20 @@ func (q *Kernel) transform_RGBA_YCbCr420_Over(dst *image.RGBA, dr, adr image.Rec
 				yWeights[y] /= totalYWeight
 			}
 
-			var pr, pg, pb float64
+			var pr, pg, pb, pa float64
 			for ky := iy; ky < jy; ky++ {
 				if yWeight := yWeights[ky-iy]; yWeight != 0 {
 					for kx := ix; kx < jx; kx++ {
 						if w := xWeights[kx-ix] * yWeight; w != 0 {
-							pi := (ky-src.Rect.Min.Y)*src.YStride + (kx - src.Rect.Min.X)
-							pj := ((ky)/2-src.Rect.Min.Y/2)*src.CStride + ((kx)/2 - src.Rect.Min.X/2)
-
-							// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-							pyy1 := int(src.Y[pi])<<16 + 1<<15
-							pcb1 := int(src.Cb[pj]) - 128
-							pcr1 := int(src.Cr[pj]) - 128
-							pru := (pyy1 + 91881*pcr1) >> 8
-							pgu := (pyy1 - 22554*pcb1 - 46802*pcr1) >> 8
-							pbu := (pyy1 + 116130*pcb1) >> 8
-							if pru < 0 {
-								pru = 0
-							} else if pru > 0xffff {
-								pru = 0xffff
-							}
-							if pgu < 0 {
-								pgu = 0
-							} else if pgu > 0xffff {
-								pgu = 0xffff
-							}
-							if pbu < 0 {
-								pbu = 0
-							} else if pbu > 0xffff {
-								pbu = 0xffff
-							}
-
+							pi := (ky-src.Rect.Min.Y)*src.Stride + (kx-src.Rect.Min.X)*4
+							pru := uint32(src.Pix[pi+0]) * 0x101
+							pgu := uint32(src.Pix[pi+1]) * 0x101
+							pbu := uint32(src.Pix[pi+2]) * 0x101
+							pau := uint32(src.Pix[pi+3]) * 0x101
 							pr += float64(pru) * w
 							pg += float64(pgu) * w
 							pb += float64(pbu) * w
+							pa += float64(pau) * w
 						}
 					}
 				}
@@ -7498,12 +4974,12 @@ func (q *Kernel) transform_RGBA_YCbCr420_Over(dst *image.RGBA, dr, adr image.Rec
 			dst.Pix[d+0] = uint8(fffftou(pr) >> 8)
 			dst.Pix[d+1] = uint8(fffftou(pg) >> 8)
 			dst.Pix[d+2] = uint8(fffftou(pb) >> 8)
-			dst.Pix[d+3] = 0xff
+			dst.Pix[d+3] = uint8(fffftou(pa) >> 8)
 		}
 	}
 }
 
-func (q *Kernel) transform_RGBA_YCbCr440_Over(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.YCbCr, sr image.Rectangle, bias image.Point, xscale, yscale float64) {
+func (q *Kernel) transform_RGBA_RGBA_Src(dst *image.RGBA, dr, adr image.Rectangle, d2s *f64.Aff3, src *image.RGBA, sr image.Rectangle, bias image.Point, xscale, yscale float64) {
 	// When shrinking, broaden the effective kernel support so that we still
 	// visit every source pixel.
 	xHalfWidth, xKernelArgScale := q.Support, 1.0
@@ -7581,40 +5057,20 @@ func (q *Kernel) transform_RGBA_YCbCr440_Over(dst *image.RGBA, dr, adr image.Rec
 				yWeights[y] /= totalYWeight
 			}
 
-			var pr, pg, pb float64
+			var pr, pg, pb, pa float64
 			for ky := iy; ky < jy; ky++ {
 				if yWeight := yWeights[ky-iy]; yWeight != 0 {
 					for kx := ix; kx < jx; kx++ {
 						if w := xWeights[kx-ix] * yWeight; w != 0 {
-							pi := (ky-src.Rect.Min.Y)*src.YStride + (kx - src.Rect.Min.X)
-							pj := ((ky)/2-src.Rect.Min.Y/2)*src.CStride + (kx - src.Rect.Min.X)
-
-							// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
-							pyy1 := int(src.Y[pi])<<16 + 1<<15
-							pcb1 := int(src.Cb[pj]) - 128
-							pcr1 := int(src.Cr[pj]) - 128
-							pru := (pyy1 + 91881*pcr1) >> 8
-							pgu := (pyy1 - 22554*pcb1 - 46802*pcr1) >> 8
-							pbu := (pyy1 + 116130*pcb1) >> 8
-							if pru < 0 {
-								pru = 0
-							} else if pru > 0xffff {
-								pru = 0xffff
-							}
-							if pgu < 0 {
-								pgu = 0
-							} else if pgu > 0xffff {
-								pgu = 0xffff
-							}
-							if pbu < 0 {
-								pbu = 0
-							} else if pbu > 0xffff {
-								pbu = 0xffff
-							}
-
+							pi := (ky-src.Rect.Min.Y)*src.Stride + (kx-src.Rect.Min.X)*4
+							pru := uint32(src.Pix[pi+0]) * 0x101
+							pgu := uint32(src.Pix[pi+1]) * 0x101
+							pbu := uint32(src.Pix[pi+2]) * 0x101
+							pau := uint32(src.Pix[pi+3]) * 0x101
 							pr += float64(pru) * w
 							pg += float64(pgu) * w
 							pb += float64(pbu) * w
+							pa += float64(pau) * w
 						}
 					}
 				}
@@ -7622,7 +5078,7 @@ func (q *Kernel) transform_RGBA_YCbCr440_Over(dst *image.RGBA, dr, adr image.Rec
 			dst.Pix[d+0] = uint8(fffftou(pr) >> 8)
 			dst.Pix[d+1] = uint8(fffftou(pg) >> 8)
 			dst.Pix[d+2] = uint8(fffftou(pb) >> 8)
-			dst.Pix[d+3] = 0xff
+			dst.Pix[d+3] = uint8(fffftou(pa) >> 8)
 		}
 	}
 }
diff --git a/draw/scale.go b/draw/scale.go
index 39520f4..00ef1d3 100644
--- a/draw/scale.go
+++ b/draw/scale.go
@@ -431,3 +431,10 @@ func transform_Uniform(dst Image, dr, adr image.Rectangle, d2s *f64.Aff3, src *i
 		}
 	}
 }
+
+func opaque(m image.Image) bool {
+	o, ok := m.(interface {
+		Opaque() bool
+	})
+	return ok && o.Opaque()
+}