golang-image/draw/gen.go

1388 lines
37 KiB
Go
Raw Normal View History

// Copyright 2015 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
// +build ignore
package main
import (
"bytes"
"flag"
"fmt"
"go/format"
"io/ioutil"
"log"
"os"
"strings"
)
var debug = flag.Bool("debug", false, "")
func main() {
flag.Parse()
w := new(bytes.Buffer)
w.WriteString("// generated by \"go run gen.go\". DO NOT EDIT.\n\n" +
"package draw\n\nimport (\n" +
"\"image\"\n" +
"\"image/color\"\n" +
"\"math\"\n" +
"\n" +
"\"golang.org/x/image/math/f64\"\n" +
")\n")
gen(w, "nnInterpolator", codeNNScaleLeaf, codeNNTransformLeaf)
gen(w, "ablInterpolator", codeABLScaleLeaf, codeABLTransformLeaf)
genKernel(w)
if *debug {
os.Stdout.Write(w.Bytes())
return
}
out, err := format.Source(w.Bytes())
if err != nil {
log.Fatal(err)
}
if err := ioutil.WriteFile("impl.go", out, 0660); err != nil {
log.Fatal(err)
}
}
var (
// dsTypes are the (dst image type, src image type) pairs to generate
// scale_DType_SType implementations for. The last element in the slice
// 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"},
{"*image.RGBA", "*image.RGBA"},
{"*image.RGBA", "*image.YCbCr"},
{"*image.RGBA", "image.Image"},
{"Image", "image.Image"},
}
dTypes, sTypes []string
sTypesForDType = map[string][]string{}
subsampleRatios = []string{
"444",
"422",
"420",
"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() {
dTypesSeen := map[string]bool{}
sTypesSeen := map[string]bool{}
for _, t := range dsTypes {
if !sTypesSeen[t.sType] {
sTypesSeen[t.sType] = true
sTypes = append(sTypes, t.sType)
}
if !dTypesSeen[t.dType] {
dTypesSeen[t.dType] = true
dTypes = append(dTypes, t.dType)
}
sTypesForDType[t.dType] = append(sTypesForDType[t.dType], t.sType)
}
sTypesForDType["anyDType"] = sTypes
}
type data struct {
dType string
sType string
sratio string
receiver string
op string
}
func gen(w *bytes.Buffer, receiver string, codes ...string) {
expn(w, codeRoot, &data{receiver: receiver})
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,
receiver: receiver,
op: op,
})
}
}
}
}
func genKernel(w *bytes.Buffer) {
expn(w, codeKernelRoot, &data{})
for _, sType := range sTypes {
expn(w, codeKernelScaleLeafX, &data{
sType: sType,
})
}
for _, dType := range dTypes {
for _, op := range ops {
expn(w, codeKernelScaleLeafY, &data{
dType: dType,
op: op,
})
}
}
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,
op: op,
})
}
}
}
func expn(w *bytes.Buffer, code string, d *data) {
if d.sType == "*image.YCbCr" && d.sratio == "" {
for _, sratio := range subsampleRatios {
e := *d
e.sratio = sratio
expn(w, code, &e)
}
return
}
for _, line := range strings.Split(code, "\n") {
line = expnLine(line, d)
if line == ";" {
continue
}
fmt.Fprintln(w, line)
}
}
func expnLine(line string, d *data) string {
for {
i := strings.IndexByte(line, '$')
if i < 0 {
break
}
prefix, s := line[:i], line[i+1:]
i = len(s)
for j, c := range s {
if !('A' <= c && c <= 'Z' || 'a' <= c && c <= 'z') {
i = j
break
}
}
dollar, suffix := s[:i], s[i:]
e := expnDollar(prefix, dollar, suffix, d)
if e == "" {
log.Fatalf("couldn't expand %q", line)
}
line = e
}
return line
}
// expnDollar expands a "$foo" fragment in a line of generated code. It returns
// the empty string if there was a problem. It returns ";" if the generated
// code is a no-op.
func expnDollar(prefix, dollar, suffix string, d *data) string {
switch dollar {
case "dType":
return prefix + d.dType + suffix
case "dTypeRN":
return prefix + relName(d.dType) + suffix
case "sratio":
return prefix + d.sratio + suffix
case "sType":
return prefix + d.sType + suffix
case "sTypeRN":
return prefix + relName(d.sType) + suffix
case "receiver":
return prefix + d.receiver + suffix
case "op":
return prefix + d.op + suffix
case "switch":
return expnSwitch("", "", true, suffix)
case "switchD":
return expnSwitch("", "", false, suffix)
case "switchS":
return expnSwitch("", "anyDType", false, suffix)
case "preOuter":
switch d.dType {
default:
return ";"
case "Image":
s := ""
if d.sType == "image.Image" {
s = "srcMask, smp := opts.SrcMask, opts.SrcMaskP\n"
}
return s +
"dstMask, dmp := opts.DstMask, opts.DstMaskP\n" +
"dstColorRGBA64 := &color.RGBA64{}\n" +
"dstColor := color.Color(dstColorRGBA64)"
}
case "preInner":
switch d.dType {
default:
return ";"
case "*image.RGBA":
return "d := " + pixOffset("dst", "dr.Min.X+adr.Min.X", "dr.Min.Y+int(dy)", "*4", "*dst.Stride")
}
case "preKernelOuter":
switch d.sType {
default:
return ";"
case "image.Image":
return "srcMask, smp := opts.SrcMask, opts.SrcMaskP"
}
case "preKernelInner":
switch d.dType {
default:
return ";"
case "*image.RGBA":
return "d := " + pixOffset("dst", "dr.Min.X+int(dx)", "dr.Min.Y+adr.Min.Y", "*4", "*dst.Stride")
}
case "blend":
args, _ := splitArgs(suffix)
if len(args) != 4 {
return ""
}
switch d.sType {
default:
return argf(args, ""+
"$3r = $0*$1r + $2*$3r\n"+
"$3g = $0*$1g + $2*$3g\n"+
"$3b = $0*$1b + $2*$3b\n"+
"$3a = $0*$1a + $2*$3a",
)
case "*image.Gray":
return argf(args, ""+
"$3r = $0*$1r + $2*$3r",
)
case "*image.YCbCr":
return argf(args, ""+
"$3r = $0*$1r + $2*$3r\n"+
"$3g = $0*$1g + $2*$3g\n"+
"$3b = $0*$1b + $2*$3b",
)
}
case "clampToAlpha":
if alwaysOpaque[d.sType] {
return ";"
}
// Go uses alpha-premultiplied color. The naive computation can lead to
// invalid colors, e.g. red > alpha, when some weights are negative.
return `
if pr > pa {
pr = pa
}
if pg > pa {
pg = pa
}
if pb > pa {
pb = pa
}
`
case "convFtou":
args, _ := splitArgs(suffix)
if len(args) != 2 {
return ""
}
switch d.sType {
default:
return argf(args, ""+
"$0r := uint32($1r)\n"+
"$0g := uint32($1g)\n"+
"$0b := uint32($1b)\n"+
"$0a := uint32($1a)",
)
case "*image.Gray":
return argf(args, ""+
"$0r := uint32($1r)",
)
case "*image.YCbCr":
return argf(args, ""+
"$0r := uint32($1r)\n"+
"$0g := uint32($1g)\n"+
"$0b := uint32($1b)",
)
}
case "outputu":
args, _ := splitArgs(suffix)
if len(args) != 3 {
return ""
}
switch d.op {
case "Over":
switch d.dType {
default:
log.Fatalf("bad dType %q", d.dType)
case "Image":
return argf(args, ""+
"qr, qg, qb, qa := dst.At($0, $1).RGBA()\n"+
"if dstMask != nil {\n"+
" _, _, _, ma := dstMask.At(dmp.X + $0, dmp.Y + $1).RGBA()\n"+
" $2r = $2r * ma / 0xffff\n"+
" $2g = $2g * ma / 0xffff\n"+
" $2b = $2b * ma / 0xffff\n"+
" $2a = $2a * ma / 0xffff\n"+
"}\n"+
"$2a1 := 0xffff - $2a\n"+
"dstColorRGBA64.R = uint16(qr*$2a1/0xffff + $2r)\n"+
"dstColorRGBA64.G = uint16(qg*$2a1/0xffff + $2g)\n"+
"dstColorRGBA64.B = uint16(qb*$2a1/0xffff + $2b)\n"+
"dstColorRGBA64.A = uint16(qa*$2a1/0xffff + $2a)\n"+
"dst.Set($0, $1, dstColor)",
)
case "*image.RGBA":
return argf(args, ""+
"$2a1 := (0xffff - $2a) * 0x101\n"+
"dst.Pix[d+0] = uint8((uint32(dst.Pix[d+0])*$2a1/0xffff + $2r) >> 8)\n"+
"dst.Pix[d+1] = uint8((uint32(dst.Pix[d+1])*$2a1/0xffff + $2g) >> 8)\n"+
"dst.Pix[d+2] = uint8((uint32(dst.Pix[d+2])*$2a1/0xffff + $2b) >> 8)\n"+
"dst.Pix[d+3] = uint8((uint32(dst.Pix[d+3])*$2a1/0xffff + $2a) >> 8)",
)
}
case "Src":
switch d.dType {
default:
log.Fatalf("bad dType %q", d.dType)
case "Image":
return argf(args, ""+
"if dstMask != nil {\n"+
" qr, qg, qb, qa := dst.At($0, $1).RGBA()\n"+
" _, _, _, ma := dstMask.At(dmp.X + $0, dmp.Y + $1).RGBA()\n"+
" pr = pr * ma / 0xffff\n"+
" pg = pg * ma / 0xffff\n"+
" pb = pb * ma / 0xffff\n"+
" pa = pa * ma / 0xffff\n"+
" $2a1 := 0xffff - ma\n"+ // Note that this is ma, not $2a.
" dstColorRGBA64.R = uint16(qr*$2a1/0xffff + $2r)\n"+
" dstColorRGBA64.G = uint16(qg*$2a1/0xffff + $2g)\n"+
" dstColorRGBA64.B = uint16(qb*$2a1/0xffff + $2b)\n"+
" dstColorRGBA64.A = uint16(qa*$2a1/0xffff + $2a)\n"+
" dst.Set($0, $1, dstColor)\n"+
"} else {\n"+
" dstColorRGBA64.R = uint16($2r)\n"+
" dstColorRGBA64.G = uint16($2g)\n"+
" dstColorRGBA64.B = uint16($2b)\n"+
" dstColorRGBA64.A = uint16($2a)\n"+
" dst.Set($0, $1, dstColor)\n"+
"}",
)
case "*image.RGBA":
switch d.sType {
default:
return argf(args, ""+
"dst.Pix[d+0] = uint8($2r >> 8)\n"+
"dst.Pix[d+1] = uint8($2g >> 8)\n"+
"dst.Pix[d+2] = uint8($2b >> 8)\n"+
"dst.Pix[d+3] = uint8($2a >> 8)",
)
case "*image.Gray":
return argf(args, ""+
"out := uint8($2r >> 8)\n"+
"dst.Pix[d+0] = out\n"+
"dst.Pix[d+1] = out\n"+
"dst.Pix[d+2] = out\n"+
"dst.Pix[d+3] = 0xff",
)
case "*image.YCbCr":
return argf(args, ""+
"dst.Pix[d+0] = uint8($2r >> 8)\n"+
"dst.Pix[d+1] = uint8($2g >> 8)\n"+
"dst.Pix[d+2] = uint8($2b >> 8)\n"+
"dst.Pix[d+3] = 0xff",
)
}
}
}
case "outputf":
args, _ := splitArgs(suffix)
if len(args) != 5 {
return ""
}
ret := ""
switch d.op {
case "Over":
switch d.dType {
default:
log.Fatalf("bad dType %q", d.dType)
case "Image":
ret = argf(args, ""+
"qr, qg, qb, qa := dst.At($0, $1).RGBA()\n"+
"$3r0 := uint32($2($3r * $4))\n"+
"$3g0 := uint32($2($3g * $4))\n"+
"$3b0 := uint32($2($3b * $4))\n"+
"$3a0 := uint32($2($3a * $4))\n"+
"if dstMask != nil {\n"+
" _, _, _, ma := dstMask.At(dmp.X + $0, dmp.Y + $1).RGBA()\n"+
" $3r0 = $3r0 * ma / 0xffff\n"+
" $3g0 = $3g0 * ma / 0xffff\n"+
" $3b0 = $3b0 * ma / 0xffff\n"+
" $3a0 = $3a0 * ma / 0xffff\n"+
"}\n"+
"$3a1 := 0xffff - $3a0\n"+
"dstColorRGBA64.R = uint16(qr*$3a1/0xffff + $3r0)\n"+
"dstColorRGBA64.G = uint16(qg*$3a1/0xffff + $3g0)\n"+
"dstColorRGBA64.B = uint16(qb*$3a1/0xffff + $3b0)\n"+
"dstColorRGBA64.A = uint16(qa*$3a1/0xffff + $3a0)\n"+
"dst.Set($0, $1, dstColor)",
)
case "*image.RGBA":
ret = argf(args, ""+
"$3r0 := uint32($2($3r * $4))\n"+
"$3g0 := uint32($2($3g * $4))\n"+
"$3b0 := uint32($2($3b * $4))\n"+
"$3a0 := uint32($2($3a * $4))\n"+
"$3a1 := (0xffff - uint32($3a0)) * 0x101\n"+
"dst.Pix[d+0] = uint8((uint32(dst.Pix[d+0])*$3a1/0xffff + $3r0) >> 8)\n"+
"dst.Pix[d+1] = uint8((uint32(dst.Pix[d+1])*$3a1/0xffff + $3g0) >> 8)\n"+
"dst.Pix[d+2] = uint8((uint32(dst.Pix[d+2])*$3a1/0xffff + $3b0) >> 8)\n"+
"dst.Pix[d+3] = uint8((uint32(dst.Pix[d+3])*$3a1/0xffff + $3a0) >> 8)",
)
}
case "Src":
switch d.dType {
default:
log.Fatalf("bad dType %q", d.dType)
case "Image":
ret = argf(args, ""+
"if dstMask != nil {\n"+
" qr, qg, qb, qa := dst.At($0, $1).RGBA()\n"+
" _, _, _, ma := dstMask.At(dmp.X + $0, dmp.Y + $1).RGBA()\n"+
" pr := uint32($2($3r * $4)) * ma / 0xffff\n"+
" pg := uint32($2($3g * $4)) * ma / 0xffff\n"+
" pb := uint32($2($3b * $4)) * ma / 0xffff\n"+
" pa := uint32($2($3a * $4)) * ma / 0xffff\n"+
" pa1 := 0xffff - ma\n"+ // Note that this is ma, not pa.
" dstColorRGBA64.R = uint16(qr*pa1/0xffff + pr)\n"+
" dstColorRGBA64.G = uint16(qg*pa1/0xffff + pg)\n"+
" dstColorRGBA64.B = uint16(qb*pa1/0xffff + pb)\n"+
" dstColorRGBA64.A = uint16(qa*pa1/0xffff + pa)\n"+
" dst.Set($0, $1, dstColor)\n"+
"} else {\n"+
" dstColorRGBA64.R = $2($3r * $4)\n"+
" dstColorRGBA64.G = $2($3g * $4)\n"+
" dstColorRGBA64.B = $2($3b * $4)\n"+
" dstColorRGBA64.A = $2($3a * $4)\n"+
" dst.Set($0, $1, dstColor)\n"+
"}",
)
case "*image.RGBA":
switch d.sType {
default:
ret = argf(args, ""+
"dst.Pix[d+0] = uint8($2($3r * $4) >> 8)\n"+
"dst.Pix[d+1] = uint8($2($3g * $4) >> 8)\n"+
"dst.Pix[d+2] = uint8($2($3b * $4) >> 8)\n"+
"dst.Pix[d+3] = uint8($2($3a * $4) >> 8)",
)
case "*image.Gray":
ret = argf(args, ""+
"out := uint8($2($3r * $4) >> 8)\n"+
"dst.Pix[d+0] = out\n"+
"dst.Pix[d+1] = out\n"+
"dst.Pix[d+2] = out\n"+
"dst.Pix[d+3] = 0xff",
)
case "*image.YCbCr":
ret = argf(args, ""+
"dst.Pix[d+0] = uint8($2($3r * $4) >> 8)\n"+
"dst.Pix[d+1] = uint8($2($3g * $4) >> 8)\n"+
"dst.Pix[d+2] = uint8($2($3b * $4) >> 8)\n"+
"dst.Pix[d+3] = 0xff",
)
}
}
}
return strings.Replace(ret, " * 1)", ")", -1)
case "srcf", "srcu":
lhs, eqOp := splitEq(prefix)
if lhs == "" {
return ""
}
args, extra := splitArgs(suffix)
if len(args) != 2 {
return ""
}
tmp := ""
if dollar == "srcf" {
tmp = "u"
}
// TODO: there's no need to multiply by 0x101 in the switch below if
// the next thing we're going to do is shift right by 8.
buf := new(bytes.Buffer)
switch d.sType {
default:
log.Fatalf("bad sType %q", d.sType)
case "image.Image":
fmt.Fprintf(buf, ""+
"%sr%s, %sg%s, %sb%s, %sa%s := src.At(%s, %s).RGBA()\n",
lhs, tmp, lhs, tmp, lhs, tmp, lhs, tmp, args[0], args[1],
)
if d.dType == "" || d.dType == "Image" {
fmt.Fprintf(buf, ""+
"if srcMask != nil {\n"+
" _, _, _, ma := srcMask.At(smp.X+%s, smp.Y+%s).RGBA()\n"+
" %sr%s = %sr%s * ma / 0xffff\n"+
" %sg%s = %sg%s * ma / 0xffff\n"+
" %sb%s = %sb%s * ma / 0xffff\n"+
" %sa%s = %sa%s * ma / 0xffff\n"+
"}\n",
args[0], args[1],
lhs, tmp, lhs, tmp,
lhs, tmp, lhs, tmp,
lhs, tmp, lhs, tmp,
lhs, tmp, lhs, tmp,
)
}
case "*image.Gray":
fmt.Fprintf(buf, ""+
"%si := %s\n"+
"%sr%s := uint32(src.Pix[%si]) * 0x101\n",
lhs, pixOffset("src", args[0], args[1], "", "*src.Stride"),
lhs, tmp, lhs,
)
case "*image.NRGBA":
fmt.Fprintf(buf, ""+
"%si := %s\n"+
"%sa%s := uint32(src.Pix[%si+3]) * 0x101\n"+
"%sr%s := uint32(src.Pix[%si+0]) * %sa%s / 0xff\n"+
"%sg%s := uint32(src.Pix[%si+1]) * %sa%s / 0xff\n"+
"%sb%s := uint32(src.Pix[%si+2]) * %sa%s / 0xff\n",
lhs, pixOffset("src", args[0], args[1], "*4", "*src.Stride"),
lhs, tmp, lhs,
lhs, tmp, lhs, lhs, tmp,
lhs, tmp, lhs, lhs, tmp,
lhs, tmp, lhs, lhs, tmp,
)
case "*image.RGBA":
fmt.Fprintf(buf, ""+
"%si := %s\n"+
"%sr%s := uint32(src.Pix[%si+0]) * 0x101\n"+
"%sg%s := uint32(src.Pix[%si+1]) * 0x101\n"+
"%sb%s := uint32(src.Pix[%si+2]) * 0x101\n"+
"%sa%s := uint32(src.Pix[%si+3]) * 0x101\n",
lhs, pixOffset("src", args[0], args[1], "*4", "*src.Stride"),
lhs, tmp, lhs,
lhs, tmp, lhs,
lhs, tmp, lhs,
lhs, tmp, lhs,
)
case "*image.YCbCr":
fmt.Fprintf(buf, ""+
"%si := %s\n"+
"%sj := %s\n"+
"%s\n",
lhs, pixOffset("src", args[0], args[1], "", "*src.YStride"),
lhs, cOffset(args[0], args[1], d.sratio),
ycbcrToRGB(lhs, tmp),
)
}
if dollar == "srcf" {
switch d.sType {
default:
fmt.Fprintf(buf, ""+
"%sr %s float64(%sru)%s\n"+
"%sg %s float64(%sgu)%s\n"+
"%sb %s float64(%sbu)%s\n"+
"%sa %s float64(%sau)%s\n",
lhs, eqOp, lhs, extra,
lhs, eqOp, lhs, extra,
lhs, eqOp, lhs, extra,
lhs, eqOp, lhs, extra,
)
case "*image.Gray":
fmt.Fprintf(buf, ""+
"%sr %s float64(%sru)%s\n",
lhs, eqOp, lhs, extra,
)
case "*image.YCbCr":
fmt.Fprintf(buf, ""+
"%sr %s float64(%sru)%s\n"+
"%sg %s float64(%sgu)%s\n"+
"%sb %s float64(%sbu)%s\n",
lhs, eqOp, lhs, extra,
lhs, eqOp, lhs, extra,
lhs, eqOp, lhs, extra,
)
}
}
return strings.TrimSpace(buf.String())
case "tweakD":
if d.dType == "*image.RGBA" {
return "d += dst.Stride"
}
return ";"
case "tweakDx":
if d.dType == "*image.RGBA" {
return strings.Replace(prefix, "dx++", "dx, d = dx+1, d+4", 1)
}
return prefix
case "tweakDy":
if d.dType == "*image.RGBA" {
return strings.Replace(prefix, "for dy, s", "for _, s", 1)
}
return prefix
case "tweakP":
switch d.sType {
case "*image.Gray":
if strings.HasPrefix(strings.TrimSpace(prefix), "pa * ") {
return "1,"
}
return "pr,"
case "*image.YCbCr":
if strings.HasPrefix(strings.TrimSpace(prefix), "pa * ") {
return "1,"
}
}
return prefix
case "tweakPr":
if d.sType == "*image.Gray" {
return "pr *= s.invTotalWeightFFFF"
}
return ";"
case "tweakVarP":
switch d.sType {
case "*image.Gray":
return strings.Replace(prefix, "var pr, pg, pb, pa", "var pr", 1)
case "*image.YCbCr":
return strings.Replace(prefix, "var pr, pg, pb, pa", "var pr, pg, pb", 1)
}
return prefix
}
return ""
}
func expnSwitch(op, dType string, expandBoth bool, template string) string {
if op == "" && dType != "anyDType" {
lines := []string{"switch op {"}
for _, op = range ops {
lines = append(lines,
fmt.Sprintf("case %s:", op),
expnSwitch(op, dType, expandBoth, template),
)
}
lines = append(lines, "}")
return strings.Join(lines, "\n")
}
switchVar := "dst"
if dType != "" {
switchVar = "src"
}
lines := []string{fmt.Sprintf("switch %s := %s.(type) {", switchVar, switchVar)}
fallback, values := "Image", dTypes
if dType != "" {
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 {
lines = append(lines, fmt.Sprintf("case %s:", v))
}
if dType != "" {
if v == "*image.YCbCr" {
lines = append(lines, expnSwitchYCbCr(op, dType, template))
} else {
lines = append(lines, expnLine(template, &data{dType: dType, sType: v, op: op}))
}
} else if !expandBoth {
lines = append(lines, expnLine(template, &data{dType: v, op: op}))
} else {
lines = append(lines, expnSwitch(op, v, false, template))
}
}
lines = append(lines, "}")
return strings.Join(lines, "\n")
}
func expnSwitchYCbCr(op, dType, template string) string {
lines := []string{
"switch src.SubsampleRatio {",
"default:",
expnLine(template, &data{dType: dType, sType: "image.Image", op: op}),
}
for _, sratio := range subsampleRatios {
lines = append(lines,
fmt.Sprintf("case image.YCbCrSubsampleRatio%s:", sratio),
expnLine(template, &data{dType: dType, sType: "*image.YCbCr", sratio: sratio, op: op}),
)
}
lines = append(lines, "}")
return strings.Join(lines, "\n")
}
func argf(args []string, s string) string {
if len(args) > 9 {
panic("too many args")
}
for i, a := range args {
old := fmt.Sprintf("$%d", i)
s = strings.Replace(s, old, a, -1)
}
return s
}
func pixOffset(m, x, y, xstride, ystride string) string {
return fmt.Sprintf("(%s-%s.Rect.Min.Y)%s + (%s-%s.Rect.Min.X)%s", y, m, ystride, x, m, xstride)
}
func cOffset(x, y, sratio string) string {
switch sratio {
case "444":
return fmt.Sprintf("( %s - src.Rect.Min.Y )*src.CStride + ( %s - src.Rect.Min.X )", y, x)
case "422":
return fmt.Sprintf("( %s - src.Rect.Min.Y )*src.CStride + ((%s)/2 - src.Rect.Min.X/2)", y, x)
case "420":
return fmt.Sprintf("((%s)/2 - src.Rect.Min.Y/2)*src.CStride + ((%s)/2 - src.Rect.Min.X/2)", y, x)
case "440":
return fmt.Sprintf("((%s)/2 - src.Rect.Min.Y/2)*src.CStride + ( %s - src.Rect.Min.X )", y, x)
}
return fmt.Sprintf("unsupported sratio %q", sratio)
}
func ycbcrToRGB(lhs, tmp string) string {
s := `
// This is an inline version of image/color/ycbcr.go's YCbCr.RGBA method.
$yy1 := int(src.Y[$i]) * 0x10100
$cb1 := int(src.Cb[$j]) - 128
$cr1 := int(src.Cr[$j]) - 128
$r@ := ($yy1 + 91881*$cr1) >> 8
$g@ := ($yy1 - 22554*$cb1 - 46802*$cr1) >> 8
$b@ := ($yy1 + 116130*$cb1) >> 8
if $r@ < 0 {
$r@ = 0
} else if $r@ > 0xffff {
$r@ = 0xffff
}
if $g@ < 0 {
$g@ = 0
} else if $g@ > 0xffff {
$g@ = 0xffff
}
if $b@ < 0 {
$b@ = 0
} else if $b@ > 0xffff {
$b@ = 0xffff
}
`
s = strings.Replace(s, "$", lhs, -1)
s = strings.Replace(s, "@", tmp, -1)
return s
}
func split(s, sep string) (string, string) {
if i := strings.Index(s, sep); i >= 0 {
return strings.TrimSpace(s[:i]), strings.TrimSpace(s[i+len(sep):])
}
return "", ""
}
func splitEq(s string) (lhs, eqOp string) {
s = strings.TrimSpace(s)
if lhs, _ = split(s, ":="); lhs != "" {
return lhs, ":="
}
if lhs, _ = split(s, "+="); lhs != "" {
return lhs, "+="
}
return "", ""
}
func splitArgs(s string) (args []string, extra string) {
s = strings.TrimSpace(s)
if s == "" || s[0] != '[' {
return nil, ""
}
s = s[1:]
i := strings.IndexByte(s, ']')
if i < 0 {
return nil, ""
}
args, extra = strings.Split(s[:i], ","), s[i+1:]
for i := range args {
args[i] = strings.TrimSpace(args[i])
}
return args, extra
}
func relName(s string) string {
if i := strings.LastIndex(s, "."); i >= 0 {
return s[i+1:]
}
return s
}
const (
codeRoot = `
func (z $receiver) Scale(dst Image, dr image.Rectangle, src image.Image, sr image.Rectangle, op Op, opts *Options) {
var o Options
if opts != nil {
o = *opts
}
// adr is the affected destination pixels.
adr := dst.Bounds().Intersect(dr)
adr, o.DstMask = clipAffectedDestRect(adr, o.DstMask, o.DstMaskP)
draw: make Scale an Interpolator method instead of a function. This means that only Kernel values have a NewScaler method, which re-uses computation when scaling multiple images of the same dst and src dimensions. The NearestNeighbor and ApproxBiLinear scalers don't get any pre-computation to re-use, so don't need a NewScaler method just to satisfy the previous Interpolator interface. As a small bonus, NN.Scale and ABL.Scale should no longer allocate on the fast paths. This change is consistent the upcoming Transformer method, so that the Interpolator interface will be type Interpolator interface { Scale(etc) Transform(etc) } instead of type Interpolator interface { NewScaler(etc) Scaler Transform(etc) } I don't have a good theory for why the "func (ablInterpolator) scale_RGBA_RGBA" benchmark is such a dramatic improvement, but at least it's in the right direction. I'm calling the other benchmark changes as noise. benchmark old ns/op new ns/op delta BenchmarkScaleLargeDownNN 3233406 3169060 -1.99% BenchmarkScaleLargeDownAB 12018178 12011348 -0.06% BenchmarkScaleLargeDownBL 1420827834 1409335695 -0.81% BenchmarkScaleLargeDownCR 2820669690 2795534035 -0.89% BenchmarkScaleDownNN 866628 869241 +0.30% BenchmarkScaleDownAB 3175963 3216041 +1.26% BenchmarkScaleDownBL 26639767 26677003 +0.14% BenchmarkScaleDownCR 51720996 51621628 -0.19% BenchmarkScaleUpNN 42758485 43258611 +1.17% BenchmarkScaleUpAB 156693813 156943367 +0.16% BenchmarkScaleUpBL 69511444 69621698 +0.16% BenchmarkScaleUpCR 124530191 124885601 +0.29% BenchmarkScaleSrcGray 8992205 9129321 +1.52% BenchmarkScaleSrcNRGBA 9807837 9894466 +0.88% BenchmarkScaleSrcRGBA 1333188 1104282 -17.17% BenchmarkScaleSrcUniform 1147788 1162488 +1.28% BenchmarkScaleSrcYCbCr 12164542 12305373 +1.16% Change-Id: I2aee6c392eb7437e843260775aed97ce145b4d47 Reviewed-on: https://go-review.googlesource.com/6556 Reviewed-by: Rob Pike <r@golang.org>
2015-03-03 06:54:53 +01:00
if adr.Empty() || sr.Empty() {
return
}
// Make adr relative to dr.Min.
adr = adr.Sub(dr.Min)
if op == Over && o.SrcMask == nil && opaque(src) {
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.
//
// Similarly, the fast paths assume that the masks are nil.
if o.DstMask != nil || o.SrcMask != nil || !sr.In(src.Bounds()) {
switch op {
case Over:
z.scale_Image_Image_Over(dst, dr, adr, src, sr, &o)
case Src:
z.scale_Image_Image_Src(dst, dr, adr, src, sr, &o)
}
} else if _, ok := src.(*image.Uniform); ok {
Draw(dst, dr, src, src.Bounds().Min, op)
} else {
$switch z.scale_$dTypeRN_$sTypeRN$sratio_$op(dst, dr, adr, src, sr, &o)
}
}
func (z $receiver) Transform(dst Image, s2d f64.Aff3, src image.Image, sr image.Rectangle, op Op, opts *Options) {
var o Options
if opts != nil {
o = *opts
}
dr := transformRect(&s2d, &sr)
// adr is the affected destination pixels.
adr := dst.Bounds().Intersect(dr)
adr, o.DstMask = clipAffectedDestRect(adr, o.DstMask, o.DstMaskP)
if adr.Empty() || sr.Empty() {
return
}
if op == Over && o.SrcMask == nil && opaque(src) {
op = Src
}
d2s := invert(&s2d)
// bias is a translation of the mapping from dst coordinates to src
// coordinates such that the latter temporarily have non-negative X
// and Y coordinates. 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)
// 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.
//
// Similarly, the fast paths assume that the masks are nil.
if o.DstMask != nil || o.SrcMask != nil || !sr.In(src.Bounds()) {
switch op {
case Over:
z.transform_Image_Image_Over(dst, dr, adr, &d2s, src, sr, bias, &o)
case Src:
z.transform_Image_Image_Src(dst, dr, adr, &d2s, src, sr, bias, &o)
}
} else if u, ok := src.(*image.Uniform); ok {
transform_Uniform(dst, dr, adr, &d2s, u, sr, bias, op)
} else {
$switch z.transform_$dTypeRN_$sTypeRN$sratio_$op(dst, dr, adr, &d2s, src, sr, bias, &o)
}
}
`
codeNNScaleLeaf = `
func (nnInterpolator) scale_$dTypeRN_$sTypeRN$sratio_$op(dst $dType, dr, adr image.Rectangle, src $sType, sr image.Rectangle, opts *Options) {
draw: make Scale an Interpolator method instead of a function. This means that only Kernel values have a NewScaler method, which re-uses computation when scaling multiple images of the same dst and src dimensions. The NearestNeighbor and ApproxBiLinear scalers don't get any pre-computation to re-use, so don't need a NewScaler method just to satisfy the previous Interpolator interface. As a small bonus, NN.Scale and ABL.Scale should no longer allocate on the fast paths. This change is consistent the upcoming Transformer method, so that the Interpolator interface will be type Interpolator interface { Scale(etc) Transform(etc) } instead of type Interpolator interface { NewScaler(etc) Scaler Transform(etc) } I don't have a good theory for why the "func (ablInterpolator) scale_RGBA_RGBA" benchmark is such a dramatic improvement, but at least it's in the right direction. I'm calling the other benchmark changes as noise. benchmark old ns/op new ns/op delta BenchmarkScaleLargeDownNN 3233406 3169060 -1.99% BenchmarkScaleLargeDownAB 12018178 12011348 -0.06% BenchmarkScaleLargeDownBL 1420827834 1409335695 -0.81% BenchmarkScaleLargeDownCR 2820669690 2795534035 -0.89% BenchmarkScaleDownNN 866628 869241 +0.30% BenchmarkScaleDownAB 3175963 3216041 +1.26% BenchmarkScaleDownBL 26639767 26677003 +0.14% BenchmarkScaleDownCR 51720996 51621628 -0.19% BenchmarkScaleUpNN 42758485 43258611 +1.17% BenchmarkScaleUpAB 156693813 156943367 +0.16% BenchmarkScaleUpBL 69511444 69621698 +0.16% BenchmarkScaleUpCR 124530191 124885601 +0.29% BenchmarkScaleSrcGray 8992205 9129321 +1.52% BenchmarkScaleSrcNRGBA 9807837 9894466 +0.88% BenchmarkScaleSrcRGBA 1333188 1104282 -17.17% BenchmarkScaleSrcUniform 1147788 1162488 +1.28% BenchmarkScaleSrcYCbCr 12164542 12305373 +1.16% Change-Id: I2aee6c392eb7437e843260775aed97ce145b4d47 Reviewed-on: https://go-review.googlesource.com/6556 Reviewed-by: Rob Pike <r@golang.org>
2015-03-03 06:54:53 +01:00
dw2 := uint64(dr.Dx()) * 2
dh2 := uint64(dr.Dy()) * 2
sw := uint64(sr.Dx())
sh := uint64(sr.Dy())
$preOuter
for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
draw: make Scale an Interpolator method instead of a function. This means that only Kernel values have a NewScaler method, which re-uses computation when scaling multiple images of the same dst and src dimensions. The NearestNeighbor and ApproxBiLinear scalers don't get any pre-computation to re-use, so don't need a NewScaler method just to satisfy the previous Interpolator interface. As a small bonus, NN.Scale and ABL.Scale should no longer allocate on the fast paths. This change is consistent the upcoming Transformer method, so that the Interpolator interface will be type Interpolator interface { Scale(etc) Transform(etc) } instead of type Interpolator interface { NewScaler(etc) Scaler Transform(etc) } I don't have a good theory for why the "func (ablInterpolator) scale_RGBA_RGBA" benchmark is such a dramatic improvement, but at least it's in the right direction. I'm calling the other benchmark changes as noise. benchmark old ns/op new ns/op delta BenchmarkScaleLargeDownNN 3233406 3169060 -1.99% BenchmarkScaleLargeDownAB 12018178 12011348 -0.06% BenchmarkScaleLargeDownBL 1420827834 1409335695 -0.81% BenchmarkScaleLargeDownCR 2820669690 2795534035 -0.89% BenchmarkScaleDownNN 866628 869241 +0.30% BenchmarkScaleDownAB 3175963 3216041 +1.26% BenchmarkScaleDownBL 26639767 26677003 +0.14% BenchmarkScaleDownCR 51720996 51621628 -0.19% BenchmarkScaleUpNN 42758485 43258611 +1.17% BenchmarkScaleUpAB 156693813 156943367 +0.16% BenchmarkScaleUpBL 69511444 69621698 +0.16% BenchmarkScaleUpCR 124530191 124885601 +0.29% BenchmarkScaleSrcGray 8992205 9129321 +1.52% BenchmarkScaleSrcNRGBA 9807837 9894466 +0.88% BenchmarkScaleSrcRGBA 1333188 1104282 -17.17% BenchmarkScaleSrcUniform 1147788 1162488 +1.28% BenchmarkScaleSrcYCbCr 12164542 12305373 +1.16% Change-Id: I2aee6c392eb7437e843260775aed97ce145b4d47 Reviewed-on: https://go-review.googlesource.com/6556 Reviewed-by: Rob Pike <r@golang.org>
2015-03-03 06:54:53 +01:00
sy := (2*uint64(dy) + 1) * sh / dh2
$preInner
for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx++ { $tweakDx
draw: make Scale an Interpolator method instead of a function. This means that only Kernel values have a NewScaler method, which re-uses computation when scaling multiple images of the same dst and src dimensions. The NearestNeighbor and ApproxBiLinear scalers don't get any pre-computation to re-use, so don't need a NewScaler method just to satisfy the previous Interpolator interface. As a small bonus, NN.Scale and ABL.Scale should no longer allocate on the fast paths. This change is consistent the upcoming Transformer method, so that the Interpolator interface will be type Interpolator interface { Scale(etc) Transform(etc) } instead of type Interpolator interface { NewScaler(etc) Scaler Transform(etc) } I don't have a good theory for why the "func (ablInterpolator) scale_RGBA_RGBA" benchmark is such a dramatic improvement, but at least it's in the right direction. I'm calling the other benchmark changes as noise. benchmark old ns/op new ns/op delta BenchmarkScaleLargeDownNN 3233406 3169060 -1.99% BenchmarkScaleLargeDownAB 12018178 12011348 -0.06% BenchmarkScaleLargeDownBL 1420827834 1409335695 -0.81% BenchmarkScaleLargeDownCR 2820669690 2795534035 -0.89% BenchmarkScaleDownNN 866628 869241 +0.30% BenchmarkScaleDownAB 3175963 3216041 +1.26% BenchmarkScaleDownBL 26639767 26677003 +0.14% BenchmarkScaleDownCR 51720996 51621628 -0.19% BenchmarkScaleUpNN 42758485 43258611 +1.17% BenchmarkScaleUpAB 156693813 156943367 +0.16% BenchmarkScaleUpBL 69511444 69621698 +0.16% BenchmarkScaleUpCR 124530191 124885601 +0.29% BenchmarkScaleSrcGray 8992205 9129321 +1.52% BenchmarkScaleSrcNRGBA 9807837 9894466 +0.88% BenchmarkScaleSrcRGBA 1333188 1104282 -17.17% BenchmarkScaleSrcUniform 1147788 1162488 +1.28% BenchmarkScaleSrcYCbCr 12164542 12305373 +1.16% Change-Id: I2aee6c392eb7437e843260775aed97ce145b4d47 Reviewed-on: https://go-review.googlesource.com/6556 Reviewed-by: Rob Pike <r@golang.org>
2015-03-03 06:54:53 +01:00
sx := (2*uint64(dx) + 1) * sw / dw2
p := $srcu[sr.Min.X + int(sx), sr.Min.Y + int(sy)]
$outputu[dr.Min.X + int(dx), dr.Min.Y + int(dy), p]
}
}
}
`
codeNNTransformLeaf = `
func (nnInterpolator) transform_$dTypeRN_$sTypeRN$sratio_$op(dst $dType, dr, adr image.Rectangle, d2s *f64.Aff3, src $sType, sr image.Rectangle, bias image.Point, opts *Options) {
$preOuter
for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
dyf := float64(dr.Min.Y + int(dy)) + 0.5
$preInner
for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx++ { $tweakDx
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
}
p := $srcu[sx0, sy0]
$outputu[dr.Min.X + int(dx), dr.Min.Y + int(dy), p]
}
}
}
`
codeABLScaleLeaf = `
func (ablInterpolator) scale_$dTypeRN_$sTypeRN$sratio_$op(dst $dType, dr, adr image.Rectangle, src $sType, sr image.Rectangle, opts *Options) {
draw: make Scale an Interpolator method instead of a function. This means that only Kernel values have a NewScaler method, which re-uses computation when scaling multiple images of the same dst and src dimensions. The NearestNeighbor and ApproxBiLinear scalers don't get any pre-computation to re-use, so don't need a NewScaler method just to satisfy the previous Interpolator interface. As a small bonus, NN.Scale and ABL.Scale should no longer allocate on the fast paths. This change is consistent the upcoming Transformer method, so that the Interpolator interface will be type Interpolator interface { Scale(etc) Transform(etc) } instead of type Interpolator interface { NewScaler(etc) Scaler Transform(etc) } I don't have a good theory for why the "func (ablInterpolator) scale_RGBA_RGBA" benchmark is such a dramatic improvement, but at least it's in the right direction. I'm calling the other benchmark changes as noise. benchmark old ns/op new ns/op delta BenchmarkScaleLargeDownNN 3233406 3169060 -1.99% BenchmarkScaleLargeDownAB 12018178 12011348 -0.06% BenchmarkScaleLargeDownBL 1420827834 1409335695 -0.81% BenchmarkScaleLargeDownCR 2820669690 2795534035 -0.89% BenchmarkScaleDownNN 866628 869241 +0.30% BenchmarkScaleDownAB 3175963 3216041 +1.26% BenchmarkScaleDownBL 26639767 26677003 +0.14% BenchmarkScaleDownCR 51720996 51621628 -0.19% BenchmarkScaleUpNN 42758485 43258611 +1.17% BenchmarkScaleUpAB 156693813 156943367 +0.16% BenchmarkScaleUpBL 69511444 69621698 +0.16% BenchmarkScaleUpCR 124530191 124885601 +0.29% BenchmarkScaleSrcGray 8992205 9129321 +1.52% BenchmarkScaleSrcNRGBA 9807837 9894466 +0.88% BenchmarkScaleSrcRGBA 1333188 1104282 -17.17% BenchmarkScaleSrcUniform 1147788 1162488 +1.28% BenchmarkScaleSrcYCbCr 12164542 12305373 +1.16% Change-Id: I2aee6c392eb7437e843260775aed97ce145b4d47 Reviewed-on: https://go-review.googlesource.com/6556 Reviewed-by: Rob Pike <r@golang.org>
2015-03-03 06:54:53 +01:00
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
$preOuter
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
}
$preInner
for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx++ { $tweakDx
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
}
s00 := $srcf[sr.Min.X + int(sx0), sr.Min.Y + int(sy0)]
s10 := $srcf[sr.Min.X + int(sx1), sr.Min.Y + int(sy0)]
$blend[xFrac1, s00, xFrac0, s10]
s01 := $srcf[sr.Min.X + int(sx0), sr.Min.Y + int(sy1)]
s11 := $srcf[sr.Min.X + int(sx1), sr.Min.Y + int(sy1)]
$blend[xFrac1, s01, xFrac0, s11]
$blend[yFrac1, s10, yFrac0, s11]
$convFtou[p, s11]
$outputu[dr.Min.X + int(dx), dr.Min.Y + int(dy), p]
}
}
}
`
codeABLTransformLeaf = `
func (ablInterpolator) transform_$dTypeRN_$sTypeRN$sratio_$op(dst $dType, dr, adr image.Rectangle, d2s *f64.Aff3, src $sType, sr image.Rectangle, bias image.Point, opts *Options) {
$preOuter
for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
dyf := float64(dr.Min.Y + int(dy)) + 0.5
$preInner
for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx++ { $tweakDx
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
}
s00 := $srcf[sx0, sy0]
s10 := $srcf[sx1, sy0]
$blend[xFrac1, s00, xFrac0, s10]
s01 := $srcf[sx0, sy1]
s11 := $srcf[sx1, sy1]
$blend[xFrac1, s01, xFrac0, s11]
$blend[yFrac1, s10, yFrac0, s11]
$convFtou[p, s11]
$outputu[dr.Min.X + int(dx), dr.Min.Y + int(dy), p]
}
}
}
`
codeKernelRoot = `
func (z *kernelScaler) Scale(dst Image, dr image.Rectangle, src image.Image, sr image.Rectangle, op Op, opts *Options) {
draw: make Scale an Interpolator method instead of a function. This means that only Kernel values have a NewScaler method, which re-uses computation when scaling multiple images of the same dst and src dimensions. The NearestNeighbor and ApproxBiLinear scalers don't get any pre-computation to re-use, so don't need a NewScaler method just to satisfy the previous Interpolator interface. As a small bonus, NN.Scale and ABL.Scale should no longer allocate on the fast paths. This change is consistent the upcoming Transformer method, so that the Interpolator interface will be type Interpolator interface { Scale(etc) Transform(etc) } instead of type Interpolator interface { NewScaler(etc) Scaler Transform(etc) } I don't have a good theory for why the "func (ablInterpolator) scale_RGBA_RGBA" benchmark is such a dramatic improvement, but at least it's in the right direction. I'm calling the other benchmark changes as noise. benchmark old ns/op new ns/op delta BenchmarkScaleLargeDownNN 3233406 3169060 -1.99% BenchmarkScaleLargeDownAB 12018178 12011348 -0.06% BenchmarkScaleLargeDownBL 1420827834 1409335695 -0.81% BenchmarkScaleLargeDownCR 2820669690 2795534035 -0.89% BenchmarkScaleDownNN 866628 869241 +0.30% BenchmarkScaleDownAB 3175963 3216041 +1.26% BenchmarkScaleDownBL 26639767 26677003 +0.14% BenchmarkScaleDownCR 51720996 51621628 -0.19% BenchmarkScaleUpNN 42758485 43258611 +1.17% BenchmarkScaleUpAB 156693813 156943367 +0.16% BenchmarkScaleUpBL 69511444 69621698 +0.16% BenchmarkScaleUpCR 124530191 124885601 +0.29% BenchmarkScaleSrcGray 8992205 9129321 +1.52% BenchmarkScaleSrcNRGBA 9807837 9894466 +0.88% BenchmarkScaleSrcRGBA 1333188 1104282 -17.17% BenchmarkScaleSrcUniform 1147788 1162488 +1.28% BenchmarkScaleSrcYCbCr 12164542 12305373 +1.16% Change-Id: I2aee6c392eb7437e843260775aed97ce145b4d47 Reviewed-on: https://go-review.googlesource.com/6556 Reviewed-by: Rob Pike <r@golang.org>
2015-03-03 06:54:53 +01:00
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, op, opts)
return
}
var o Options
if opts != nil {
o = *opts
}
// adr is the affected destination pixels.
adr := dst.Bounds().Intersect(dr)
adr, o.DstMask = clipAffectedDestRect(adr, o.DstMask, o.DstMaskP)
draw: make Scale an Interpolator method instead of a function. This means that only Kernel values have a NewScaler method, which re-uses computation when scaling multiple images of the same dst and src dimensions. The NearestNeighbor and ApproxBiLinear scalers don't get any pre-computation to re-use, so don't need a NewScaler method just to satisfy the previous Interpolator interface. As a small bonus, NN.Scale and ABL.Scale should no longer allocate on the fast paths. This change is consistent the upcoming Transformer method, so that the Interpolator interface will be type Interpolator interface { Scale(etc) Transform(etc) } instead of type Interpolator interface { NewScaler(etc) Scaler Transform(etc) } I don't have a good theory for why the "func (ablInterpolator) scale_RGBA_RGBA" benchmark is such a dramatic improvement, but at least it's in the right direction. I'm calling the other benchmark changes as noise. benchmark old ns/op new ns/op delta BenchmarkScaleLargeDownNN 3233406 3169060 -1.99% BenchmarkScaleLargeDownAB 12018178 12011348 -0.06% BenchmarkScaleLargeDownBL 1420827834 1409335695 -0.81% BenchmarkScaleLargeDownCR 2820669690 2795534035 -0.89% BenchmarkScaleDownNN 866628 869241 +0.30% BenchmarkScaleDownAB 3175963 3216041 +1.26% BenchmarkScaleDownBL 26639767 26677003 +0.14% BenchmarkScaleDownCR 51720996 51621628 -0.19% BenchmarkScaleUpNN 42758485 43258611 +1.17% BenchmarkScaleUpAB 156693813 156943367 +0.16% BenchmarkScaleUpBL 69511444 69621698 +0.16% BenchmarkScaleUpCR 124530191 124885601 +0.29% BenchmarkScaleSrcGray 8992205 9129321 +1.52% BenchmarkScaleSrcNRGBA 9807837 9894466 +0.88% BenchmarkScaleSrcRGBA 1333188 1104282 -17.17% BenchmarkScaleSrcUniform 1147788 1162488 +1.28% BenchmarkScaleSrcYCbCr 12164542 12305373 +1.16% Change-Id: I2aee6c392eb7437e843260775aed97ce145b4d47 Reviewed-on: https://go-review.googlesource.com/6556 Reviewed-by: Rob Pike <r@golang.org>
2015-03-03 06:54:53 +01:00
if adr.Empty() || sr.Empty() {
return
}
// Make adr relative to dr.Min.
adr = adr.Sub(dr.Min)
if op == Over && o.SrcMask == nil && opaque(src) {
op = Src
}
if _, ok := src.(*image.Uniform); ok && o.DstMask == nil && o.SrcMask == nil && sr.In(src.Bounds()) {
Draw(dst, dr, src, src.Bounds().Min, 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.
draw: use a sync.Pool for kernel scaling's temporary buffers. benchmark old ns/op new ns/op delta BenchmarkScaleBLLargeDown 257715146 260286012 +1.00% BenchmarkScaleCRLargeDown 426797448 430078734 +0.77% BenchmarkScaleBLDown 4449939 4222542 -5.11% BenchmarkScaleCRDown 8160446 8010056 -1.84% BenchmarkScaleBLUp 22290312 21044122 -5.59% BenchmarkScaleCRUp 33010722 32021468 -3.00% BenchmarkScaleCRSrcGray 13307961 13020192 -2.16% BenchmarkScaleCRSrcNRGBA 40567431 40801939 +0.58% BenchmarkScaleCRSrcRGBA 39892971 40240558 +0.87% BenchmarkScaleCRSrcYCbCr 59020222 59686699 +1.13% benchmark old allocs new allocs delta BenchmarkScaleBLLargeDown 1 1 +0.00% BenchmarkScaleCRLargeDown 1 2 +100.00% BenchmarkScaleBLDown 1 0 -100.00% BenchmarkScaleCRDown 1 0 -100.00% BenchmarkScaleBLUp 1 0 -100.00% BenchmarkScaleCRUp 1 0 -100.00% BenchmarkScaleCRSrcGray 1 0 -100.00% BenchmarkScaleCRSrcNRGBA 1 0 -100.00% BenchmarkScaleCRSrcRGBA 1 0 -100.00% BenchmarkScaleCRSrcYCbCr 1 0 -100.00% benchmark old bytes new bytes delta BenchmarkScaleBLLargeDown 14745600 2949200 -80.00% BenchmarkScaleCRLargeDown 14745600 4915333 -66.67% BenchmarkScaleBLDown 1523712 5079 -99.67% BenchmarkScaleCRDown 1523712 7619 -99.50% BenchmarkScaleBLUp 10117120 101175 -99.00% BenchmarkScaleCRUp 10117120 202350 -98.00% BenchmarkScaleCRSrcGray 4915200 49156 -99.00% BenchmarkScaleCRSrcNRGBA 4915200 163853 -96.67% BenchmarkScaleCRSrcRGBA 4915200 163853 -96.67% BenchmarkScaleCRSrcYCbCr 4915200 245780 -95.00% The increase in BenchmarkScale??LargeDown number of allocs I think is an accounting error due to the low number of iterations: a low denominator. I suspect that there are one or two extra allocs up front for using the sync.Pool, but one fewer alloc per iteration. The number of iterations is only 5 for BL and 3 for CR, for the default timeout. If I increase the -test.benchtime value to 5s, then the reported average (allocs/op) drop from 2 to 0, so the delta should actually be -100% instead of +0 or +100%. Change-Id: I21d9bb0086bdb25517b6a430e8a21bdf3db026f6 Reviewed-on: https://go-review.googlesource.com/8150 Reviewed-by: Rob Pike <r@golang.org>
2015-03-27 01:21:16 +01:00
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.
//
// Similarly, the fast paths assume that the masks are nil.
if o.SrcMask != nil || !sr.In(src.Bounds()) {
z.scaleX_Image(tmp, src, sr, &o)
} else {
$switchS z.scaleX_$sTypeRN$sratio(tmp, src, sr, &o)
}
if o.DstMask != nil {
switch op {
case Over:
z.scaleY_Image_Over(dst, dr, adr, tmp, &o)
case Src:
z.scaleY_Image_Src(dst, dr, adr, tmp, &o)
}
} else {
$switchD z.scaleY_$dTypeRN_$op(dst, dr, adr, tmp, &o)
}
}
func (q *Kernel) Transform(dst Image, s2d f64.Aff3, src image.Image, sr image.Rectangle, op Op, opts *Options) {
var o Options
if opts != nil {
o = *opts
}
dr := transformRect(&s2d, &sr)
// adr is the affected destination pixels.
adr := dst.Bounds().Intersect(dr)
adr, o.DstMask = clipAffectedDestRect(adr, o.DstMask, o.DstMaskP)
if adr.Empty() || sr.Empty() {
return
}
if op == Over && o.SrcMask == nil && opaque(src) {
op = Src
}
d2s := invert(&s2d)
// bias is a translation of the mapping from dst coordinates to src
// coordinates such that the latter temporarily have non-negative X
// and Y coordinates. 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 && o.DstMask != nil && o.SrcMask != nil && sr.In(src.Bounds()) {
transform_Uniform(dst, dr, adr, &d2s, u, sr, bias, 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.
//
// Similarly, the fast paths assume that the masks are nil.
if o.DstMask != nil || o.SrcMask != nil || !sr.In(src.Bounds()) {
switch op {
case Over:
q.transform_Image_Image_Over(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale, &o)
case Src:
q.transform_Image_Image_Src(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale, &o)
}
} else {
$switch q.transform_$dTypeRN_$sTypeRN$sratio_$op(dst, dr, adr, &d2s, src, sr, bias, xscale, yscale, &o)
}
}
`
codeKernelScaleLeafX = `
func (z *kernelScaler) scaleX_$sTypeRN$sratio(tmp [][4]float64, src $sType, sr image.Rectangle, opts *Options) {
t := 0
$preKernelOuter
for y := int32(0); y < z.sh; y++ {
for _, s := range z.horizontal.sources {
var pr, pg, pb, pa float64 $tweakVarP
for _, c := range z.horizontal.contribs[s.i:s.j] {
p += $srcf[sr.Min.X + int(c.coord), sr.Min.Y + int(y)] * c.weight
}
$tweakPr
tmp[t] = [4]float64{
pr * s.invTotalWeightFFFF, $tweakP
pg * s.invTotalWeightFFFF, $tweakP
pb * s.invTotalWeightFFFF, $tweakP
pa * s.invTotalWeightFFFF, $tweakP
}
t++
}
}
}
`
codeKernelScaleLeafY = `
func (z *kernelScaler) scaleY_$dTypeRN_$op(dst $dType, dr, adr image.Rectangle, tmp [][4]float64, opts *Options) {
$preOuter
for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx++ {
$preKernelInner
for dy, s := range z.vertical.sources[adr.Min.Y:adr.Max.Y] { $tweakDy
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
}
$clampToAlpha
$outputf[dr.Min.X + int(dx), dr.Min.Y + int(adr.Min.Y + dy), ftou, p, s.invTotalWeight]
$tweakD
}
}
}
`
codeKernelTransformLeaf = `
func (q *Kernel) transform_$dTypeRN_$sTypeRN$sratio_$op(dst $dType, dr, adr image.Rectangle, d2s *f64.Aff3, src $sType, sr image.Rectangle, bias image.Point, xscale, yscale float64, opts *Options) {
// 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)))
$preOuter
for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
dyf := float64(dr.Min.Y + int(dy)) + 0.5
$preInner
for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx++ { $tweakDx
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, pg, pb, pa float64 $tweakVarP
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 {
p += $srcf[kx, ky] * w
}
}
}
}
$clampToAlpha
$outputf[dr.Min.X + int(dx), dr.Min.Y + int(dy), fffftou, p, 1]
}
}
}
`
)