aw/golang-image
1
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

draw: RGBA src fast path for scaling.

Browse Source 
benchmark                     old ns/op      new ns/op      delta
BenchmarkScaleSrcRGBA         15124800       2091946        -86.17%

Change-Id: Id8d3088793ebf1d75b929fcf6945987817e87463
Reviewed-on: https://go-review.googlesource.com/6234
Reviewed-by: Rob Pike <r@golang.org>
This commit is contained in:
Nigel Tao 2015-02-27 21:26:47 +11:00
parent 2a40e8dacf
commit b57ddf1b68
2 changed files with 131 additions and 52 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

36
draw/gen.go
View File

@ -284,10 +284,26 @@ func expnDollar(prefix, dollar, suffix string, d *data) string {
switch d.sType { switch d.sType {
default: default:
log.Fatalf("bad sType %q", d.sType) log.Fatalf("bad sType %q", d.sType)
case "image.Image", "*image.NRGBA", "*image.RGBA", "*image.Uniform", "*image.YCbCr": // TODO: separate code for concrete types. case "image.Image", "*image.NRGBA", "*image.Uniform", "*image.YCbCr": // TODO: separate code for concrete types.
fmt.Fprintf(buf, "%sr%s, %sg%s, %sb%s, %sa%s := "+ fmt.Fprintf(buf, "%sr%s, %sg%s, %sb%s, %sa%s := "+
"src.At(sp.X + int(%s), sp.Y+int(%s)).RGBA()\n", "src.At(sp.X + int(%s), sp.Y+int(%s)).RGBA()\n",
lhs, tmp, lhs, tmp, lhs, tmp, lhs, tmp, args[0], args[1]) lhs, tmp, lhs, tmp, lhs, tmp, lhs, tmp,
args[0], args[1],
)
case "*image.RGBA":
// TODO: there's no need to multiply by 0x101 if the next thing
// we're going to do is shift right by 8.
fmt.Fprintf(buf, "%si := src.PixOffset(sp.X + int(%s), sp.Y+int(%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, args[0], args[1],
lhs, tmp, lhs,
lhs, tmp, lhs,
lhs, tmp, lhs,
lhs, tmp, lhs,
)
} }
if dollar == "srcf" { if dollar == "srcf" {
@ -399,8 +415,15 @@ const (
if dr.Empty() { if dr.Empty() {
return return
} }
// 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 := (image.Rectangle{sp, sp.Add(image.Point{int(z.sw), int(z.sh)})}); !sr.In(src.Bounds()) {
z.scale_Image_Image(dst, dp, dr, src, sp)
} else {
$switch z.scale_$dTypeRN_$sTypeRN(dst, dp, dr, src, sp) $switch z.scale_$dTypeRN_$sTypeRN(dst, dp, dr, src, sp)
} }
}
` `
codeNNLeaf = ` codeNNLeaf = `
@ -478,7 +501,16 @@ const (
// scaleY distributes the temporary image's rows over the destination image. // scaleY distributes the temporary image's rows over the destination image.
// TODO: is it worth having a sync.Pool for this temporary buffer? // TODO: is it worth having a sync.Pool for this temporary buffer?
tmp := make([][4]float64, z.dw*z.sh) tmp := make([][4]float64, z.dw*z.sh)
// 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 := (image.Rectangle{sp, sp.Add(image.Point{int(z.sw), int(z.sh)})}); !sr.In(src.Bounds()) {
z.scaleX_Image(tmp, src, sp)
} else {
$switchS z.scaleX_$sTypeRN(tmp, src, sp) $switchS z.scaleX_$sTypeRN(tmp, src, sp)
}
$switchD z.scaleY_$dTypeRN(dst, dp, dr, tmp) $switchD z.scaleY_$dTypeRN(dst, dp, dr, tmp)
} }
` `

59
draw/impl.go
View File

@ -16,6 +16,12 @@ func (z *nnScaler) Scale(dst Image, dp image.Point, src image.Image, sp image.Po
if dr.Empty() { if dr.Empty() {
return return
} }
// 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 := (image.Rectangle{sp, sp.Add(image.Point{int(z.sw), int(z.sh)})}); !sr.In(src.Bounds()) {
z.scale_Image_Image(dst, dp, dr, src, sp)
} else {
switch dst := dst.(type) { switch dst := dst.(type) {
case *image.RGBA: case *image.RGBA:
switch src := src.(type) { switch src := src.(type) {
@ -37,6 +43,7 @@ func (z *nnScaler) Scale(dst Image, dp image.Point, src image.Image, sp image.Po
} }
} }
} }
}
func (z *nnScaler) scale_RGBA_NRGBA(dst *image.RGBA, dp image.Point, dr image.Rectangle, src *image.NRGBA, sp image.Point) { func (z *nnScaler) scale_RGBA_NRGBA(dst *image.RGBA, dp image.Point, dr image.Rectangle, src *image.NRGBA, sp image.Point) {
for dy := int32(dr.Min.Y); dy < int32(dr.Max.Y); dy++ { for dy := int32(dr.Min.Y); dy < int32(dr.Max.Y); dy++ {
@ -60,7 +67,11 @@ func (z *nnScaler) scale_RGBA_RGBA(dst *image.RGBA, dp image.Point, dr image.Rec
d := dst.PixOffset(dp.X+dr.Min.X, dp.Y+int(dy)) d := dst.PixOffset(dp.X+dr.Min.X, dp.Y+int(dy))
for dx := int32(dr.Min.X); dx < int32(dr.Max.X); dx++ { for dx := int32(dr.Min.X); dx < int32(dr.Max.X); dx++ {
sx := (2*uint64(dx) + 1) * uint64(z.sw) / (2 * uint64(z.dw)) sx := (2*uint64(dx) + 1) * uint64(z.sw) / (2 * uint64(z.dw))
pr, pg, pb, pa := src.At(sp.X+int(sx), sp.Y+int(sy)).RGBA() pi := src.PixOffset(sp.X+int(sx), sp.Y+int(sy))
pr := uint32(src.Pix[pi+0]) * 0x101
pg := uint32(src.Pix[pi+1]) * 0x101
pb := uint32(src.Pix[pi+2]) * 0x101
pa := uint32(src.Pix[pi+3]) * 0x101
dst.Pix[d+0] = uint8(uint32(pr) >> 8) dst.Pix[d+0] = uint8(uint32(pr) >> 8)
dst.Pix[d+1] = uint8(uint32(pg) >> 8) dst.Pix[d+1] = uint8(uint32(pg) >> 8)
dst.Pix[d+2] = uint8(uint32(pb) >> 8) dst.Pix[d+2] = uint8(uint32(pb) >> 8)
@ -144,6 +155,12 @@ func (z *ablScaler) Scale(dst Image, dp image.Point, src image.Image, sp image.P
if dr.Empty() { if dr.Empty() {
return return
} }
// 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 := (image.Rectangle{sp, sp.Add(image.Point{int(z.sw), int(z.sh)})}); !sr.In(src.Bounds()) {
z.scale_Image_Image(dst, dp, dr, src, sp)
} else {
switch dst := dst.(type) { switch dst := dst.(type) {
case *image.RGBA: case *image.RGBA:
switch src := src.(type) { switch src := src.(type) {
@ -165,6 +182,7 @@ func (z *ablScaler) Scale(dst Image, dp image.Point, src image.Image, sp image.P
} }
} }
} }
}
func (z *ablScaler) scale_RGBA_NRGBA(dst *image.RGBA, dp image.Point, dr image.Rectangle, src *image.NRGBA, sp image.Point) { func (z *ablScaler) scale_RGBA_NRGBA(dst *image.RGBA, dp image.Point, dr image.Rectangle, src *image.NRGBA, sp image.Point) {
yscale := float64(z.sh) / float64(z.dh) yscale := float64(z.sh) / float64(z.dh)
@ -267,12 +285,20 @@ func (z *ablScaler) scale_RGBA_RGBA(dst *image.RGBA, dp image.Point, dr image.Re
sx1 = sx0 sx1 = sx0
xFrac0, xFrac1 = 1, 0 xFrac0, xFrac1 = 1, 0
} }
s00ru, s00gu, s00bu, s00au := src.At(sp.X+int(sx0), sp.Y+int(sy0)).RGBA() s00i := src.PixOffset(sp.X+int(sx0), sp.Y+int(sy0))
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) s00r := float64(s00ru)
s00g := float64(s00gu) s00g := float64(s00gu)
s00b := float64(s00bu) s00b := float64(s00bu)
s00a := float64(s00au) s00a := float64(s00au)
s10ru, s10gu, s10bu, s10au := src.At(sp.X+int(sx1), sp.Y+int(sy0)).RGBA() s10i := src.PixOffset(sp.X+int(sx1), sp.Y+int(sy0))
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) s10r := float64(s10ru)
s10g := float64(s10gu) s10g := float64(s10gu)
s10b := float64(s10bu) s10b := float64(s10bu)
@ -281,12 +307,20 @@ func (z *ablScaler) scale_RGBA_RGBA(dst *image.RGBA, dp image.Point, dr image.Re
s10g = xFrac1*s00g + xFrac0*s10g s10g = xFrac1*s00g + xFrac0*s10g
s10b = xFrac1*s00b + xFrac0*s10b s10b = xFrac1*s00b + xFrac0*s10b
s10a = xFrac1*s00a + xFrac0*s10a s10a = xFrac1*s00a + xFrac0*s10a
s01ru, s01gu, s01bu, s01au := src.At(sp.X+int(sx0), sp.Y+int(sy1)).RGBA() s01i := src.PixOffset(sp.X+int(sx0), sp.Y+int(sy1))
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) s01r := float64(s01ru)
s01g := float64(s01gu) s01g := float64(s01gu)
s01b := float64(s01bu) s01b := float64(s01bu)
s01a := float64(s01au) s01a := float64(s01au)
s11ru, s11gu, s11bu, s11au := src.At(sp.X+int(sx1), sp.Y+int(sy1)).RGBA() s11i := src.PixOffset(sp.X+int(sx1), sp.Y+int(sy1))
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) s11r := float64(s11ru)
s11g := float64(s11gu) s11g := float64(s11gu)
s11b := float64(s11bu) s11b := float64(s11bu)
@ -607,6 +641,13 @@ func (z *kernelScaler) Scale(dst Image, dp image.Point, src image.Image, sp imag
// scaleY distributes the temporary image's rows over the destination image. // scaleY distributes the temporary image's rows over the destination image.
// TODO: is it worth having a sync.Pool for this temporary buffer? // TODO: is it worth having a sync.Pool for this temporary buffer?
tmp := make([][4]float64, z.dw*z.sh) tmp := make([][4]float64, z.dw*z.sh)
// 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 := (image.Rectangle{sp, sp.Add(image.Point{int(z.sw), int(z.sh)})}); !sr.In(src.Bounds()) {
z.scaleX_Image(tmp, src, sp)
} else {
switch src := src.(type) { switch src := src.(type) {
case *image.NRGBA: case *image.NRGBA:
z.scaleX_NRGBA(tmp, src, sp) z.scaleX_NRGBA(tmp, src, sp)
@ -619,6 +660,8 @@ func (z *kernelScaler) Scale(dst Image, dp image.Point, src image.Image, sp imag
default: default:
z.scaleX_Image(tmp, src, sp) z.scaleX_Image(tmp, src, sp)
} }
}
switch dst := dst.(type) { switch dst := dst.(type) {
case *image.RGBA: case *image.RGBA:
z.scaleY_RGBA(dst, dp, dr, tmp) z.scaleY_RGBA(dst, dp, dr, tmp)
@ -656,7 +699,11 @@ func (z *kernelScaler) scaleX_RGBA(tmp [][4]float64, src *image.RGBA, sp image.P
for _, s := range z.horizontal.sources { for _, s := range z.horizontal.sources {
var pr, pg, pb, pa float64 var pr, pg, pb, pa float64
for _, c := range z.horizontal.contribs[s.i:s.j] { for _, c := range z.horizontal.contribs[s.i:s.j] {
pru, pgu, pbu, pau := src.At(sp.X+int(c.coord), sp.Y+int(y)).RGBA() pi := src.PixOffset(sp.X+int(c.coord), sp.Y+int(y))
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 pr += float64(pru) * c.weight
pg += float64(pgu) * c.weight pg += float64(pgu) * c.weight
pb += float64(pbu) * c.weight pb += float64(pbu) * c.weight