internal/graphics: Reuse the vertices backend

This is basically a reland of 9cb631e30f.
This change locks the vertices backend at the end-frame phase to
protect from vertices usages by other goroutines.

Updates #1546
Closes #1681

internal/graphics/vertex.go

@@ -57,32 +57,61 @@ var (
 	theVerticesBackend = &verticesBackend{}
 )
 
+// TODO: The logic is very similar to atlas.temporaryPixels. Unify them.
+
 type verticesBackend struct {
 	backend          []float32
-	head    int
+	pos              int
+	notFullyUsedTime int
+
 	m sync.Mutex
 }
 
+func verticesBackendFloat32Size(size int) int {
+	l := 1024
+	for l < size {
+		l *= 2
+	}
+	return l
+}
+
 func (v *verticesBackend) slice(n int) []float32 {
 	v.m.Lock()
 	defer v.m.Unlock()
 
 	need := n * VertexFloatNum
-	if v.head+need > len(v.backend) {
-		v.backend = nil
-		v.head = 0
+	if len(v.backend) < v.pos+need {
+		v.backend = make([]float32, verticesBackendFloat32Size(v.pos+need))
+		v.pos = 0
 	}
-	if v.backend == nil {
-		l := 1024
-		if n > l {
-			l = n
-		}
-		v.backend = make([]float32, VertexFloatNum*l)
+	s := v.backend[v.pos : v.pos+need]
+	v.pos += need
+	return s
+}
+
+func (v *verticesBackend) lockAndReset(f func() error) error {
+	v.m.Lock()
+	defer v.m.Unlock()
+
+	if err := f(); err != nil {
+		return err
 	}
 
-	s := v.backend[v.head : v.head+need]
-	v.head += need
-	return s
+	const maxNotFullyUsedTime = 60
+	if verticesBackendFloat32Size(v.pos) < len(v.backend) {
+		if v.notFullyUsedTime < maxNotFullyUsedTime {
+			v.notFullyUsedTime++
+		}
+	} else {
+		v.notFullyUsedTime = 0
+	}
+
+	if v.notFullyUsedTime == maxNotFullyUsedTime && len(v.backend) > 0 {
+		v.backend = nil
+	}
+
+	v.pos = 0
+	return nil
 }
 
 // Vertices returns a float32 slice for n vertices.
@@ -92,6 +121,10 @@ func Vertices(n int) []float32 {
 	return theVerticesBackend.slice(n)
 }
 
+func LockAndResetVertices(f func() error) error {
+	return theVerticesBackend.lockAndReset(f)
+}
+
 // QuadVertices returns a float32 slice for a quadrangle.
 // QuadVertices returns a slice that never overlaps with other slices returned this function,
 // and users can do optimization based on this fact.

internal/restorable/image.go

@@ -408,8 +408,8 @@ func (i *Image) appendDrawTrianglesHistory(srcs [graphics.ShaderImageNum]*Image,
 	// All images must be resolved and not stale each after frame.
 	// So we don't have to care if image is stale or not here.
 
-	// vertices is generated at ebiten package and doesn't have to be copied so far.
-	// This depends on the implementation of graphics.QuadVertices.
+	vs := make([]float32, len(vertices))
+	copy(vs, vertices)
 
 	is := make([]uint16, len(indices))
 	copy(is, indices)
@@ -417,7 +417,7 @@ func (i *Image) appendDrawTrianglesHistory(srcs [graphics.ShaderImageNum]*Image,
 	item := &drawTrianglesHistoryItem{
 		images:    srcs,
 		offsets:   offsets,
-		vertices:  vertices,
+		vertices:  vs,
 		indices:   is,
 		colorm:    colorm,
 		mode:      mode,

internal/restorable/images_test.go

@@ -861,3 +861,58 @@ func TestClearPixels(t *testing.T) {
 	// After clearing, the regions will be available again.
 	img.ReplacePixels(make([]byte, 4*8*4), 0, 0, 8, 4)
 }
+
+func TestMutateSlices(t *testing.T) {
+	const w, h = 16, 16
+	dst := NewImage(w, h)
+	src := NewImage(w, h)
+	pix := make([]byte, 4*w*h)
+	for i := 0; i < w*h; i++ {
+		pix[4*i] = byte(i)
+		pix[4*i+1] = byte(i)
+		pix[4*i+2] = byte(i)
+		pix[4*i+3] = 0xff
+	}
+	src.ReplacePixels(pix, 0, 0, w, h)
+
+	vs := quadVertices(w, h, 0, 0)
+	is := make([]uint16, len(graphics.QuadIndices()))
+	copy(is, graphics.QuadIndices())
+	dr := driver.Region{
+		X:      0,
+		Y:      0,
+		Width:  w,
+		Height: h,
+	}
+	dst.DrawTriangles([graphics.ShaderImageNum]*Image{src}, [graphics.ShaderImageNum - 1][2]float32{}, vs, is, nil, driver.CompositeModeSourceOver, driver.FilterNearest, driver.AddressUnsafe, dr, driver.Region{}, nil, nil)
+	for i := range vs {
+		vs[i] = 0
+	}
+	for i := range is {
+		is[i] = 0
+	}
+	if err := ResolveStaleImages(); err != nil {
+		t.Fatal(err)
+	}
+	if err := RestoreIfNeeded(); err != nil {
+		t.Fatal(err)
+	}
+
+	for j := 0; j < h; j++ {
+		for i := 0; i < w; i++ {
+			r, g, b, a, err := src.At(i, j)
+			if err != nil {
+				t.Fatal(err)
+			}
+			want := color.RGBA{r, g, b, a}
+			r, g, b, a, err = dst.At(i, j)
+			if err != nil {
+				t.Fatal(err)
+			}
+			got := color.RGBA{r, g, b, a}
+			if !sameColors(got, want, 1) {
+				t.Errorf("(%d, %d): got %v, want %v", i, j, got, want)
+			}
+		}
+	}
+}

uicontext.go

@@ -24,6 +24,7 @@ import (
 	"github.com/hajimehoshi/ebiten/v2/internal/clock"
 	"github.com/hajimehoshi/ebiten/v2/internal/debug"
 	"github.com/hajimehoshi/ebiten/v2/internal/driver"
+	"github.com/hajimehoshi/ebiten/v2/internal/graphics"
 	"github.com/hajimehoshi/ebiten/v2/internal/hooks"
 )
 
@@ -158,16 +159,22 @@ func (c *uiContext) update(updateCount int) error {
 	if err, ok := c.err.Load().(error); ok && err != nil {
 		return err
 	}
+
 	if err := buffered.BeginFrame(); err != nil {
 		return err
 	}
 	if err := c.updateImpl(updateCount); err != nil {
 		return err
 	}
+
+	// All the vertices data are consumed at the end of the frame, and the data backend can be
+	// available after that. Until then, lock the vertices backend.
+	return graphics.LockAndResetVertices(func() error {
 		if err := buffered.EndFrame(); err != nil {
 			return err
 		}
 		return nil
+	})
 }
 
 func (c *uiContext) updateImpl(updateCount int) error {