draw: add a fast path for NRGBA src images.
benchmark old ns/op new ns/op delta
BenchmarkScaleSrcNRGBA 14142583 2043782 -85.55%
BenchmarkTformABSrcNRGBA 9846421 1993564 -79.75%
BenchmarkTformCRSrcNRGBA 62041569 13866457 -77.65%
Change-Id: I1edf699dfc6436c0da7e3ab221684406ab1e362f
Reviewed-on: https://go-review.googlesource.com/7793
Reviewed-by: Rob Pike <r@golang.org>
diff --git a/draw/gen.go b/draw/gen.go
index 1450346..4048e19 100644
--- a/draw/gen.go
+++ b/draw/gen.go
@@ -215,7 +215,7 @@
switch d.dType {
default:
return ";"
- case "*image.Gray", "*image.RGBA":
+ case "*image.RGBA":
return "d := dst.PixOffset(dr.Min.X+adr.Min.X, dr.Min.Y+int(dy))"
}
@@ -223,7 +223,7 @@
switch d.dType {
default:
return ";"
- case "*image.Gray", "*image.RGBA":
+ case "*image.RGBA":
return "d := dst.PixOffset(dr.Min.X+int(dx), dr.Min.Y+adr.Min.Y)"
}
@@ -386,7 +386,7 @@
switch d.sType {
default:
log.Fatalf("bad sType %q", d.sType)
- case "image.Image", "*image.NRGBA", "*image.Uniform", "*image.YCbCr": // TODO: separate code for concrete types.
+ case "image.Image", "*image.Uniform", "*image.YCbCr": // TODO: separate code for concrete types.
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,
@@ -400,6 +400,20 @@
lhs, args[0], args[1],
lhs, tmp, lhs,
)
+ case "*image.NRGBA":
+ // 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(%s, %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, args[0], args[1],
+ lhs, tmp, lhs,
+ lhs, tmp, lhs, lhs, tmp,
+ lhs, tmp, lhs, lhs, tmp,
+ lhs, tmp, lhs, lhs, tmp,
+ )
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.
diff --git a/draw/impl.go b/draw/impl.go
index 02eaabf..06b8432 100644
--- a/draw/impl.go
+++ b/draw/impl.go
@@ -139,7 +139,11 @@
d := dst.PixOffset(dr.Min.X+adr.Min.X, dr.Min.Y+int(dy))
for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx, d = dx+1, d+4 {
sx := (2*uint64(dx) + 1) * sw / dw2
- pr, pg, pb, pa := src.At(sr.Min.X+int(sx), sr.Min.Y+int(sy)).RGBA()
+ pi := src.PixOffset(sr.Min.X+int(sx), sr.Min.Y+int(sy))
+ pa := uint32(src.Pix[pi+3]) * 0x101
+ pr := uint32(src.Pix[pi+0]) * pa / 0xff
+ pg := uint32(src.Pix[pi+1]) * pa / 0xff
+ pb := uint32(src.Pix[pi+2]) * pa / 0xff
dst.Pix[d+0] = uint8(uint32(pr) >> 8)
dst.Pix[d+1] = uint8(uint32(pg) >> 8)
dst.Pix[d+2] = uint8(uint32(pb) >> 8)
@@ -341,7 +345,11 @@
if !(image.Point{sx0, sy0}).In(sr) {
continue
}
- pr, pg, pb, pa := src.At(sx0, sy0).RGBA()
+ pi := src.PixOffset(sx0, sy0)
+ pa := uint32(src.Pix[pi+3]) * 0x101
+ pr := uint32(src.Pix[pi+0]) * pa / 0xff
+ pg := uint32(src.Pix[pi+1]) * pa / 0xff
+ pb := uint32(src.Pix[pi+2]) * pa / 0xff
dst.Pix[d+0] = uint8(uint32(pr) >> 8)
dst.Pix[d+1] = uint8(uint32(pg) >> 8)
dst.Pix[d+2] = uint8(uint32(pb) >> 8)
@@ -724,12 +732,20 @@
xFrac0, xFrac1 = 1, 0
}
- s00ru, s00gu, s00bu, s00au := src.At(sr.Min.X+int(sx0), sr.Min.Y+int(sy0)).RGBA()
+ s00i := src.PixOffset(sr.Min.X+int(sx0), sr.Min.Y+int(sy0))
+ 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 := src.PixOffset(sr.Min.X+int(sx1), sr.Min.Y+int(sy0))
+ 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)
@@ -738,12 +754,20 @@
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 := src.PixOffset(sr.Min.X+int(sx0), sr.Min.Y+int(sy1))
+ 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 := src.PixOffset(sr.Min.X+int(sx1), sr.Min.Y+int(sy1))
+ 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)
@@ -1520,12 +1544,20 @@
yFrac0, yFrac1 = 1, 0
}
- s00ru, s00gu, s00bu, s00au := src.At(sx0, sy0).RGBA()
+ s00i := src.PixOffset(sx0, sy0)
+ 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(sx1, sy0).RGBA()
+ s10i := src.PixOffset(sx1, sy0)
+ 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)
@@ -1534,12 +1566,20 @@
s10g = xFrac1*s00g + xFrac0*s10g
s10b = xFrac1*s00b + xFrac0*s10b
s10a = xFrac1*s00a + xFrac0*s10a
- s01ru, s01gu, s01bu, s01au := src.At(sx0, sy1).RGBA()
+ s01i := src.PixOffset(sx0, sy1)
+ 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(sx1, sy1).RGBA()
+ s11i := src.PixOffset(sx1, sy1)
+ 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)
@@ -2371,7 +2411,11 @@
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()
+ pi := src.PixOffset(sr.Min.X+int(c.coord), sr.Min.Y+int(y))
+ 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
@@ -2768,7 +2812,11 @@
for ky := iy; ky < jy; ky++ {
yWeight := yWeights[ky-iy]
for kx := ix; kx < jx; kx++ {
- pru, pgu, pbu, pau := src.At(kx, ky).RGBA()
+ pi := src.PixOffset(kx, ky)
+ 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) * xWeights[kx-ix] * yWeight
pg += float64(pgu) * xWeights[kx-ix] * yWeight
pb += float64(pbu) * xWeights[kx-ix] * yWeight