diff --git a/image.go b/image.go
index 7f51a3e34..146c1a13c 100644
--- a/image.go
+++ b/image.go
@@ -247,7 +247,7 @@ func (i *Image) DrawImage(img *Image, options *DrawImageOptions) {
 	colorm, cr, cg, cb, ca := colorMToScale(options.ColorM.affineColorM())
 	cr, cg, cb, ca = options.ColorScale.apply(cr, cg, cb, ca)
 	vs := i.ensureTmpVertices(4 * graphics.VertexFloatCount)
-	graphics.QuadVertices(vs, float32(sx0), float32(sy0), float32(sx1), float32(sy1), a, b, c, d, tx, ty, cr, cg, cb, ca)
+	graphics.QuadVerticesFromSrcAndMatrix(vs, float32(sx0), float32(sy0), float32(sx1), float32(sy1), a, b, c, d, tx, ty, cr, cg, cb, ca)
 	is := graphics.QuadIndices()
 
 	srcs := [graphics.ShaderSrcImageCount]*ui.Image{img.image}
@@ -829,7 +829,7 @@ func (i *Image) DrawRectShader(width, height int, shader *Shader, options *DrawR
 	vs := i.ensureTmpVertices(4 * graphics.VertexFloatCount)
 
 	// Do not use srcRegions[0].Dx() and srcRegions[0].Dy() as these might be empty.
-	graphics.QuadVertices(vs,
+	graphics.QuadVerticesFromSrcAndMatrix(vs,
 		float32(srcRegions[0].Min.X), float32(srcRegions[0].Min.Y),
 		float32(srcRegions[0].Min.X+width), float32(srcRegions[0].Min.Y+height),
 		a, b, c, d, tx, ty, cr, cg, cb, ca)
diff --git a/internal/atlas/image.go b/internal/atlas/image.go
index da6943789..0aed86376 100644
--- a/internal/atlas/image.go
+++ b/internal/atlas/image.go
@@ -43,16 +43,6 @@ func min(a, b int) int {
 	return b
 }
 
-// quadVertices returns vertices to render a quad. These values are passed to graphicscommand.Image.
-func quadVertices(dx0, dy0, dx1, dy1, sx0, sy0, sx1, sy1, cr, cg, cb, ca float32) []float32 {
-	return []float32{
-		dx0, dy0, sx0, sy0, cr, cg, cb, ca,
-		dx1, dy0, sx1, sy0, cr, cg, cb, ca,
-		dx0, dy1, sx0, sy1, cr, cg, cb, ca,
-		dx1, dy1, sx1, sy1, cr, cg, cb, ca,
-	}
-}
-
 func appendDeferred(f func()) {
 	deferredM.Lock()
 	defer deferredM.Unlock()
@@ -149,7 +139,8 @@ func (b *backend) extendIfNeeded(width, height int) {
 
 	srcs := [graphics.ShaderSrcImageCount]*graphicscommand.Image{b.image}
 	sw, sh := b.image.InternalSize()
-	vs := quadVertices(0, 0, float32(sw), float32(sh), 0, 0, float32(sw), float32(sh), 1, 1, 1, 1)
+	vs := make([]float32, 4*graphics.VertexFloatCount)
+	graphics.QuadVerticesFromDstAndSrc(vs, 0, 0, float32(sw), float32(sh), 0, 0, float32(sw), float32(sh), 1, 1, 1, 1)
 	is := graphics.QuadIndices()
 	dr := image.Rect(0, 0, sw, sh)
 	newImg.DrawTriangles(srcs, vs, is, graphicsdriver.BlendCopy, dr, [graphics.ShaderSrcImageCount]image.Rectangle{}, NearestFilterShader.ensureShader(), nil, graphicsdriver.FillRuleFillAll)
@@ -174,7 +165,8 @@ func newClearedImage(width, height int, screen bool) *graphicscommand.Image {
 }
 
 func clearImage(i *graphicscommand.Image, region image.Rectangle) {
-	vs := quadVertices(float32(region.Min.X), float32(region.Min.Y), float32(region.Max.X), float32(region.Max.Y), 0, 0, 0, 0, 0, 0, 0, 0)
+	vs := make([]float32, 4*graphics.VertexFloatCount)
+	graphics.QuadVerticesFromDstAndSrc(vs, float32(region.Min.X), float32(region.Min.Y), float32(region.Max.X), float32(region.Max.Y), 0, 0, 0, 0, 0, 0, 0, 0)
 	is := graphics.QuadIndices()
 	i.DrawTriangles([graphics.ShaderSrcImageCount]*graphicscommand.Image{}, vs, is, graphicsdriver.BlendClear, region, [graphics.ShaderSrcImageCount]image.Rectangle{}, clearShader.ensureShader(), nil, graphicsdriver.FillRuleFillAll)
 }
@@ -353,7 +345,7 @@ func (i *Image) ensureIsolatedFromSource(backends []*backend) {
 
 	w, h := float32(i.width), float32(i.height)
 	vs := make([]float32, 4*graphics.VertexFloatCount)
-	graphics.QuadVertices(vs, 0, 0, w, h, 1, 0, 0, 1, 0, 0, 1, 1, 1, 1)
+	graphics.QuadVerticesFromSrcAndMatrix(vs, 0, 0, w, h, 1, 0, 0, 1, 0, 0, 1, 1, 1, 1)
 	is := graphics.QuadIndices()
 	dr := image.Rect(0, 0, i.width, i.height)
 
@@ -384,7 +376,7 @@ func (i *Image) putOnSourceBackend() {
 
 	w, h := float32(i.width), float32(i.height)
 	vs := make([]float32, 4*graphics.VertexFloatCount)
-	graphics.QuadVertices(vs, 0, 0, w, h, 1, 0, 0, 1, 0, 0, 1, 1, 1, 1)
+	graphics.QuadVerticesFromSrcAndMatrix(vs, 0, 0, w, h, 1, 0, 0, 1, 0, 0, 1, 1, 1, 1)
 	is := graphics.QuadIndices()
 	dr := image.Rect(0, 0, i.width, i.height)
 	newI.drawTriangles([graphics.ShaderSrcImageCount]*Image{i}, vs, is, graphicsdriver.BlendCopy, dr, [graphics.ShaderSrcImageCount]image.Rectangle{}, NearestFilterShader, nil, graphicsdriver.FillRuleFillAll)
diff --git a/internal/atlas/image_test.go b/internal/atlas/image_test.go
index 51d829a38..3c5c351ab 100644
--- a/internal/atlas/image_test.go
+++ b/internal/atlas/image_test.go
@@ -50,12 +50,9 @@ func quadVertices(sw, sh, x, y int, scalex float32) []float32 {
 	sy0 := float32(0)
 	sx1 := float32(sw)
 	sy1 := float32(sh)
-	return []float32{
-		dx0, dy0, sx0, sy0, 1, 1, 1, 1,
-		dx1, dy0, sx1, sy0, 1, 1, 1, 1,
-		dx0, dy1, sx0, sy1, 1, 1, 1, 1,
-		dx1, dy1, sx1, sy1, 1, 1, 1, 1,
-	}
+	vs := make([]float32, 4*graphics.VertexFloatCount)
+	graphics.QuadVerticesFromDstAndSrc(vs, dx0, dy0, dx1, dy1, sx0, sy0, sx1, sy1, 1, 1, 1, 1)
+	return vs
 }
 
 const bigSize = 2049
diff --git a/internal/buffered/image_test.go b/internal/buffered/image_test.go
index cb027129a..2dbd8cd71 100644
--- a/internal/buffered/image_test.go
+++ b/internal/buffered/image_test.go
@@ -52,7 +52,7 @@ func TestUnsyncedPixels(t *testing.T) {
 	// Flush unsynced pixel cache.
 	src := buffered.NewImage(16, 16, atlas.ImageTypeRegular)
 	vs := make([]float32, 4*graphics.VertexFloatCount)
-	graphics.QuadVertices(vs, 0, 0, 16, 16, 1, 0, 0, 1, 0, 0, 1, 1, 1, 1)
+	graphics.QuadVerticesFromSrcAndMatrix(vs, 0, 0, 16, 16, 1, 0, 0, 1, 0, 0, 1, 1, 1, 1)
 	is := graphics.QuadIndices()
 	dr := image.Rect(0, 0, 16, 16)
 	sr := [graphics.ShaderSrcImageCount]image.Rectangle{image.Rect(0, 0, 16, 16)}
diff --git a/internal/graphics/vertex.go b/internal/graphics/vertex.go
index 680c25afa..95d7768b1 100644
--- a/internal/graphics/vertex.go
+++ b/internal/graphics/vertex.go
@@ -50,16 +50,16 @@ func QuadIndices() []uint32 {
 	return quadIndices
 }
 
-// QuadVertices sets a float32 slice for a quadrangle.
-// QuadVertices sets a slice that never overlaps with other slices returned this function,
-// and users can do optimization based on this fact.
-func QuadVertices(dst []float32, sx0, sy0, sx1, sy1 float32, a, b, c, d, tx, ty float32, cr, cg, cb, ca float32) {
+// QuadVerticesFromSrcAndMatrix sets a float32 slice for a quadrangle.
+func QuadVerticesFromSrcAndMatrix(dst []float32, sx0, sy0, sx1, sy1 float32, a, b, c, d, tx, ty float32, cr, cg, cb, ca float32) {
 	x := sx1 - sx0
 	y := sy1 - sy0
 	ax, by, cx, dy := a*x, b*y, c*x, d*y
 	u0, v0, u1, v1 := sx0, sy0, sx1, sy1
 
 	// This function is very performance-sensitive and implement in a very dumb way.
+
+	// Remove the boundary check.
 	dst = dst[:4*VertexFloatCount]
 
 	dst[0] = adjustDestinationPixel(tx)
@@ -99,6 +99,53 @@ func QuadVertices(dst []float32, sx0, sy0, sx1, sy1 float32, a, b, c, d, tx, ty
 	dst[3*VertexFloatCount+7] = ca
 }
 
+// QuadVerticesFromDstAndSrc sets a float32 slice for a quadrangle.
+func QuadVerticesFromDstAndSrc(dst []float32, dx0, dy0, dx1, dy1, sx0, sy0, sx1, sy1, cr, cg, cb, ca float32) {
+	dx0 = adjustDestinationPixel(dx0)
+	dy0 = adjustDestinationPixel(dy0)
+	dx1 = adjustDestinationPixel(dx1)
+	dy1 = adjustDestinationPixel(dy1)
+
+	// Remove the boundary check.
+	dst = dst[:4*VertexFloatCount]
+
+	dst[0] = dx0
+	dst[1] = dy0
+	dst[2] = sx0
+	dst[3] = sy0
+	dst[4] = cr
+	dst[5] = cg
+	dst[6] = cb
+	dst[7] = ca
+
+	dst[VertexFloatCount] = dx1
+	dst[VertexFloatCount+1] = dy0
+	dst[VertexFloatCount+2] = sx1
+	dst[VertexFloatCount+3] = sy0
+	dst[VertexFloatCount+4] = cr
+	dst[VertexFloatCount+5] = cg
+	dst[VertexFloatCount+6] = cb
+	dst[VertexFloatCount+7] = ca
+
+	dst[2*VertexFloatCount] = dx0
+	dst[2*VertexFloatCount+1] = dy1
+	dst[2*VertexFloatCount+2] = sx0
+	dst[2*VertexFloatCount+3] = sy1
+	dst[2*VertexFloatCount+4] = cr
+	dst[2*VertexFloatCount+5] = cg
+	dst[2*VertexFloatCount+6] = cb
+	dst[2*VertexFloatCount+7] = ca
+
+	dst[3*VertexFloatCount] = dx1
+	dst[3*VertexFloatCount+1] = dy1
+	dst[3*VertexFloatCount+2] = sx1
+	dst[3*VertexFloatCount+3] = sy1
+	dst[3*VertexFloatCount+4] = cr
+	dst[3*VertexFloatCount+5] = cg
+	dst[3*VertexFloatCount+6] = cb
+	dst[3*VertexFloatCount+7] = ca
+}
+
 func adjustDestinationPixel(x float32) float32 {
 	// Avoid the center of the pixel, which is problematic (#929, #1171).
 	// Instead, align the vertices with about 1/3 pixels.
diff --git a/internal/graphicscommand/image_test.go b/internal/graphicscommand/image_test.go
index a60175edb..946d9c736 100644
--- a/internal/graphicscommand/image_test.go
+++ b/internal/graphicscommand/image_test.go
@@ -43,12 +43,9 @@ func TestMain(m *testing.M) {
 }
 
 func quadVertices(w, h float32) []float32 {
-	return []float32{
-		0, 0, 0, 0, 1, 1, 1, 1,
-		w, 0, w, 0, 1, 1, 1, 1,
-		0, w, 0, h, 1, 1, 1, 1,
-		w, h, w, h, 1, 1, 1, 1,
-	}
+	vs := make([]float32, 8*graphics.VertexFloatCount)
+	graphics.QuadVerticesFromDstAndSrc(vs, 0, 0, w, h, 0, 0, w, h, 1, 1, 1, 1)
+	return vs
 }
 
 func TestClear(t *testing.T) {
diff --git a/internal/mipmap/mipmap.go b/internal/mipmap/mipmap.go
index 2fe10cca7..2145f2e66 100644
--- a/internal/mipmap/mipmap.go
+++ b/internal/mipmap/mipmap.go
@@ -152,7 +152,7 @@ func (m *Mipmap) level(level int) *buffered.Image {
 	switch {
 	case level == 1:
 		src = m.orig
-		graphics.QuadVertices(vs, 0, 0, float32(m.width), float32(m.height), 0.5, 0, 0, 0.5, 0, 0, 1, 1, 1, 1)
+		graphics.QuadVerticesFromSrcAndMatrix(vs, 0, 0, float32(m.width), float32(m.height), 0.5, 0, 0, 0.5, 0, 0, 1, 1, 1, 1)
 	case level > 1:
 		src = m.level(level - 1)
 		if src == nil {
@@ -161,7 +161,7 @@ func (m *Mipmap) level(level int) *buffered.Image {
 		}
 		w := sizeForLevel(m.width, level-1)
 		h := sizeForLevel(m.height, level-1)
-		graphics.QuadVertices(vs, 0, 0, float32(w), float32(h), 0.5, 0, 0, 0.5, 0, 0, 1, 1, 1, 1)
+		graphics.QuadVerticesFromSrcAndMatrix(vs, 0, 0, float32(w), float32(h), 0.5, 0, 0, 0.5, 0, 0, 1, 1, 1, 1)
 	default:
 		panic(fmt.Sprintf("mipmap: invalid level: %d", level))
 	}
diff --git a/internal/ui/image.go b/internal/ui/image.go
index c49d854cd..987d0bcc3 100644
--- a/internal/ui/image.go
+++ b/internal/ui/image.go
@@ -158,7 +158,7 @@ func (i *Image) Fill(r, g, b, a float32, region image.Rectangle) {
 		i.tmpVerticesForFill = make([]float32, 4*graphics.VertexFloatCount)
 	}
 	// i.tmpVerticesForFill can be reused as this is sent to DrawTriangles immediately.
-	graphics.QuadVertices(
+	graphics.QuadVerticesFromSrcAndMatrix(
 		i.tmpVerticesForFill,
 		1, 1, float32(i.ui.whiteImage.width-1), float32(i.ui.whiteImage.height-1),
 		float32(i.width), 0, 0, float32(i.height), 0, 0,
@@ -235,7 +235,7 @@ func (i *bigOffscreenImage) drawTriangles(srcs [graphics.ShaderSrcImageCount]*Im
 			i.tmpVerticesForCopying = make([]float32, 4*graphics.VertexFloatCount)
 		}
 		// i.tmpVerticesForCopying can be reused as this is sent to DrawTriangles immediately.
-		graphics.QuadVertices(
+		graphics.QuadVerticesFromSrcAndMatrix(
 			i.tmpVerticesForCopying,
 			float32(i.region.Min.X), float32(i.region.Min.Y), float32(i.region.Max.X), float32(i.region.Max.Y),
 			bigOffscreenScale, 0, 0, bigOffscreenScale, 0, 0,
@@ -279,7 +279,7 @@ func (i *bigOffscreenImage) flush() {
 		i.tmpVerticesForFlushing = make([]float32, 4*graphics.VertexFloatCount)
 	}
 	// i.tmpVerticesForFlushing can be reused as this is sent to DrawTriangles in this function.
-	graphics.QuadVertices(
+	graphics.QuadVerticesFromSrcAndMatrix(
 		i.tmpVerticesForFlushing,
 		0, 0, float32(i.region.Dx()*bigOffscreenScale), float32(i.region.Dy()*bigOffscreenScale),
 		1.0/bigOffscreenScale, 0, 0, 1.0/bigOffscreenScale, float32(i.region.Min.X), float32(i.region.Min.Y),