2c27a34d37
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>
296 lines
9.0 KiB
Go
296 lines
9.0 KiB
Go
// 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.
|
|
|
|
package draw
|
|
|
|
import (
|
|
"bytes"
|
|
"flag"
|
|
"fmt"
|
|
"image"
|
|
"image/color"
|
|
"image/png"
|
|
"math/rand"
|
|
"os"
|
|
"reflect"
|
|
"testing"
|
|
|
|
_ "image/jpeg"
|
|
)
|
|
|
|
var genScaleFiles = flag.Bool("gen_scale_files", false, "whether to generate the TestScaleXxx golden files.")
|
|
|
|
// testScale tests that scaling the source image gives the exact destination
|
|
// image. This is to ensure that any refactoring or optimization of the scaling
|
|
// code doesn't change the scaling behavior. Changing the actual algorithm or
|
|
// kernel used by any particular quality setting will obviously change the
|
|
// resultant pixels. In such a case, use the gen_scale_files flag to regenerate
|
|
// the golden files.
|
|
func testScale(t *testing.T, w int, h int, direction, srcFilename string) {
|
|
f, err := os.Open("../testdata/go-turns-two-" + srcFilename)
|
|
if err != nil {
|
|
t.Fatalf("Open: %v", err)
|
|
}
|
|
defer f.Close()
|
|
src, _, err := image.Decode(f)
|
|
if err != nil {
|
|
t.Fatalf("Decode: %v", err)
|
|
}
|
|
testCases := map[string]Interpolator{
|
|
"nn": NearestNeighbor,
|
|
"ab": ApproxBiLinear,
|
|
"bl": BiLinear,
|
|
"cr": CatmullRom,
|
|
}
|
|
for name, q := range testCases {
|
|
gotFilename := fmt.Sprintf("../testdata/go-turns-two-%s-%s.png", direction, name)
|
|
|
|
got := image.NewRGBA(image.Rect(0, 0, w, h))
|
|
q.Scale(got, got.Bounds(), src, src.Bounds())
|
|
if *genScaleFiles {
|
|
g, err := os.Create(gotFilename)
|
|
if err != nil {
|
|
t.Errorf("Create: %v", err)
|
|
continue
|
|
}
|
|
defer g.Close()
|
|
if err := png.Encode(g, got); err != nil {
|
|
t.Errorf("Encode: %v", err)
|
|
continue
|
|
}
|
|
continue
|
|
}
|
|
|
|
g, err := os.Open(gotFilename)
|
|
if err != nil {
|
|
t.Errorf("Open: %v", err)
|
|
continue
|
|
}
|
|
defer g.Close()
|
|
want, err := png.Decode(g)
|
|
if err != nil {
|
|
t.Errorf("Decode: %v", err)
|
|
continue
|
|
}
|
|
|
|
if !reflect.DeepEqual(got, want) {
|
|
t.Errorf("%s: actual image differs from golden image", gotFilename)
|
|
continue
|
|
}
|
|
}
|
|
}
|
|
|
|
func TestScaleDown(t *testing.T) { testScale(t, 100, 100, "down", "280x360.jpeg") }
|
|
func TestScaleUp(t *testing.T) { testScale(t, 75, 100, "up", "14x18.png") }
|
|
|
|
func fillPix(r *rand.Rand, pixs ...[]byte) {
|
|
for _, pix := range pixs {
|
|
for i := range pix {
|
|
pix[i] = uint8(r.Intn(256))
|
|
}
|
|
}
|
|
}
|
|
|
|
func TestScaleClipCommute(t *testing.T) {
|
|
src := image.NewNRGBA(image.Rect(0, 0, 20, 20))
|
|
fillPix(rand.New(rand.NewSource(0)), src.Pix)
|
|
|
|
outer := image.Rect(1, 1, 8, 5)
|
|
inner := image.Rect(2, 3, 6, 5)
|
|
qs := []Interpolator{
|
|
NearestNeighbor,
|
|
ApproxBiLinear,
|
|
CatmullRom,
|
|
}
|
|
for _, q := range qs {
|
|
dst0 := image.NewRGBA(image.Rect(1, 1, 10, 10))
|
|
dst1 := image.NewRGBA(image.Rect(1, 1, 10, 10))
|
|
for i := range dst0.Pix {
|
|
dst0.Pix[i] = uint8(i / 4)
|
|
dst1.Pix[i] = uint8(i / 4)
|
|
}
|
|
|
|
// Scale then clip.
|
|
q.Scale(dst0, outer, src, src.Bounds())
|
|
dst0 = dst0.SubImage(inner).(*image.RGBA)
|
|
|
|
// Clip then scale.
|
|
dst1 = dst1.SubImage(inner).(*image.RGBA)
|
|
q.Scale(dst1, outer, src, src.Bounds())
|
|
|
|
loop:
|
|
for y := inner.Min.Y; y < inner.Max.Y; y++ {
|
|
for x := inner.Min.X; x < inner.Max.X; x++ {
|
|
if c0, c1 := dst0.RGBAAt(x, y), dst1.RGBAAt(x, y); c0 != c1 {
|
|
t.Errorf("q=%T: at (%d, %d): c0=%v, c1=%v", q, x, y, c0, c1)
|
|
break loop
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
// The fooWrapper types wrap the dst or src image to avoid triggering the
|
|
// type-specific fast path implementations.
|
|
type (
|
|
dstWrapper struct{ Image }
|
|
srcWrapper struct{ image.Image }
|
|
)
|
|
|
|
// TestFastPaths tests that the fast path implementations produce identical
|
|
// results to the generic implementation.
|
|
func TestFastPaths(t *testing.T) {
|
|
drs := []image.Rectangle{
|
|
image.Rect(0, 0, 10, 10), // The dst bounds.
|
|
image.Rect(3, 4, 8, 6), // A strict subset of the dst bounds.
|
|
image.Rect(-3, -5, 2, 4), // Partial out-of-bounds #0.
|
|
image.Rect(4, -2, 6, 12), // Partial out-of-bounds #1.
|
|
image.Rect(12, 14, 23, 45), // Complete out-of-bounds.
|
|
image.Rect(5, 5, 5, 5), // Empty.
|
|
}
|
|
srs := []image.Rectangle{
|
|
image.Rect(0, 0, 12, 9), // The src bounds.
|
|
image.Rect(2, 2, 10, 8), // A strict subset of the src bounds.
|
|
image.Rect(10, 5, 20, 20), // Partial out-of-bounds #0.
|
|
image.Rect(-40, 0, 40, 8), // Partial out-of-bounds #1.
|
|
image.Rect(-8, -8, -4, -4), // Complete out-of-bounds.
|
|
image.Rect(5, 5, 5, 5), // Empty.
|
|
}
|
|
srcfs := []func(image.Rectangle) (image.Image, error){
|
|
srcGray,
|
|
srcNRGBA,
|
|
srcRGBA,
|
|
srcUniform,
|
|
srcYCbCr,
|
|
}
|
|
var srcs []image.Image
|
|
for _, srcf := range srcfs {
|
|
src, err := srcf(srs[0])
|
|
if err != nil {
|
|
t.Fatal(err)
|
|
}
|
|
srcs = append(srcs, src)
|
|
}
|
|
qs := []Interpolator{
|
|
NearestNeighbor,
|
|
ApproxBiLinear,
|
|
CatmullRom,
|
|
}
|
|
blue := image.NewUniform(color.RGBA{0x11, 0x22, 0x44, 0x7f})
|
|
|
|
for _, dr := range drs {
|
|
for _, src := range srcs {
|
|
for _, sr := range srs {
|
|
for _, q := range qs {
|
|
dst0 := image.NewRGBA(drs[0])
|
|
dst1 := image.NewRGBA(drs[0])
|
|
Draw(dst0, dst0.Bounds(), blue, image.Point{}, Src)
|
|
Draw(dstWrapper{dst1}, dst1.Bounds(), srcWrapper{blue}, image.Point{}, Src)
|
|
q.Scale(dst0, dr, src, sr)
|
|
q.Scale(dstWrapper{dst1}, dr, srcWrapper{src}, sr)
|
|
if !bytes.Equal(dst0.Pix, dst1.Pix) {
|
|
t.Errorf("pix differ for dr=%v, src=%T, sr=%v, q=%T", dr, src, sr, q)
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
func srcGray(boundsHint image.Rectangle) (image.Image, error) {
|
|
m := image.NewGray(boundsHint)
|
|
fillPix(rand.New(rand.NewSource(0)), m.Pix)
|
|
return m, nil
|
|
}
|
|
|
|
func srcNRGBA(boundsHint image.Rectangle) (image.Image, error) {
|
|
m := image.NewNRGBA(boundsHint)
|
|
fillPix(rand.New(rand.NewSource(1)), m.Pix)
|
|
return m, nil
|
|
}
|
|
|
|
func srcRGBA(boundsHint image.Rectangle) (image.Image, error) {
|
|
m := image.NewRGBA(boundsHint)
|
|
fillPix(rand.New(rand.NewSource(2)), m.Pix)
|
|
// RGBA is alpha-premultiplied, so the R, G and B values should
|
|
// be <= the A values.
|
|
for i := 0; i < len(m.Pix); i += 4 {
|
|
m.Pix[i+0] = uint8(uint32(m.Pix[i+0]) * uint32(m.Pix[i+3]) / 0xff)
|
|
m.Pix[i+1] = uint8(uint32(m.Pix[i+1]) * uint32(m.Pix[i+3]) / 0xff)
|
|
m.Pix[i+2] = uint8(uint32(m.Pix[i+2]) * uint32(m.Pix[i+3]) / 0xff)
|
|
}
|
|
return m, nil
|
|
}
|
|
|
|
func srcUniform(boundsHint image.Rectangle) (image.Image, error) {
|
|
return image.NewUniform(color.RGBA64{0x1234, 0x5555, 0x9181, 0xbeef}), nil
|
|
}
|
|
|
|
func srcYCbCr(boundsHint image.Rectangle) (image.Image, error) {
|
|
m := image.NewYCbCr(boundsHint, image.YCbCrSubsampleRatio420)
|
|
fillPix(rand.New(rand.NewSource(3)), m.Y, m.Cb, m.Cr)
|
|
return m, nil
|
|
}
|
|
|
|
func srcYCbCrLarge(boundsHint image.Rectangle) (image.Image, error) {
|
|
// 3072 x 2304 is over 7 million pixels at 4:3, comparable to a
|
|
// 2015 smart-phone camera's output.
|
|
return srcYCbCr(image.Rect(0, 0, 3072, 2304))
|
|
}
|
|
|
|
func srcTux(boundsHint image.Rectangle) (image.Image, error) {
|
|
// tux.png is a 386 x 395 image.
|
|
f, err := os.Open("../testdata/tux.png")
|
|
if err != nil {
|
|
return nil, fmt.Errorf("Open: %v", err)
|
|
}
|
|
defer f.Close()
|
|
src, err := png.Decode(f)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("Decode: %v", err)
|
|
}
|
|
return src, nil
|
|
}
|
|
|
|
func benchScale(b *testing.B, srcf func(image.Rectangle) (image.Image, error), w int, h int, q Interpolator) {
|
|
dst := image.NewRGBA(image.Rect(0, 0, w, h))
|
|
src, err := srcf(image.Rect(0, 0, 1024, 768))
|
|
if err != nil {
|
|
b.Fatal(err)
|
|
}
|
|
dr, sr := dst.Bounds(), src.Bounds()
|
|
scaler := Scaler(q)
|
|
if n, ok := q.(interface {
|
|
NewScaler(int, int, int, int) Scaler
|
|
}); ok {
|
|
scaler = n.NewScaler(dr.Dx(), dr.Dy(), sr.Dx(), sr.Dy())
|
|
}
|
|
|
|
b.ResetTimer()
|
|
for i := 0; i < b.N; i++ {
|
|
scaler.Scale(dst, dr, src, sr)
|
|
}
|
|
}
|
|
|
|
func BenchmarkScaleLargeDownNN(b *testing.B) { benchScale(b, srcYCbCrLarge, 200, 150, NearestNeighbor) }
|
|
func BenchmarkScaleLargeDownAB(b *testing.B) { benchScale(b, srcYCbCrLarge, 200, 150, ApproxBiLinear) }
|
|
func BenchmarkScaleLargeDownBL(b *testing.B) { benchScale(b, srcYCbCrLarge, 200, 150, BiLinear) }
|
|
func BenchmarkScaleLargeDownCR(b *testing.B) { benchScale(b, srcYCbCrLarge, 200, 150, CatmullRom) }
|
|
|
|
func BenchmarkScaleDownNN(b *testing.B) { benchScale(b, srcTux, 120, 80, NearestNeighbor) }
|
|
func BenchmarkScaleDownAB(b *testing.B) { benchScale(b, srcTux, 120, 80, ApproxBiLinear) }
|
|
func BenchmarkScaleDownBL(b *testing.B) { benchScale(b, srcTux, 120, 80, BiLinear) }
|
|
func BenchmarkScaleDownCR(b *testing.B) { benchScale(b, srcTux, 120, 80, CatmullRom) }
|
|
|
|
func BenchmarkScaleUpNN(b *testing.B) { benchScale(b, srcTux, 800, 600, NearestNeighbor) }
|
|
func BenchmarkScaleUpAB(b *testing.B) { benchScale(b, srcTux, 800, 600, ApproxBiLinear) }
|
|
func BenchmarkScaleUpBL(b *testing.B) { benchScale(b, srcTux, 800, 600, BiLinear) }
|
|
func BenchmarkScaleUpCR(b *testing.B) { benchScale(b, srcTux, 800, 600, CatmullRom) }
|
|
|
|
func BenchmarkScaleSrcGray(b *testing.B) { benchScale(b, srcGray, 200, 150, ApproxBiLinear) }
|
|
func BenchmarkScaleSrcNRGBA(b *testing.B) { benchScale(b, srcNRGBA, 200, 150, ApproxBiLinear) }
|
|
func BenchmarkScaleSrcRGBA(b *testing.B) { benchScale(b, srcRGBA, 200, 150, ApproxBiLinear) }
|
|
func BenchmarkScaleSrcUniform(b *testing.B) { benchScale(b, srcUniform, 200, 150, ApproxBiLinear) }
|
|
func BenchmarkScaleSrcYCbCr(b *testing.B) { benchScale(b, srcYCbCr, 200, 150, ApproxBiLinear) }
|