diff --git a/gpu/compute.go b/gpu/compute.go
index 177a8560..8612cd95 100644
--- a/gpu/compute.go
+++ b/gpu/compute.go
@@ -557,9 +557,9 @@ restart:
 		size := img.Bounds().Size()
 		driver.UploadImage(a.tex, pos, img)
 		rightPadding := image.Pt(padding, size.Y)
-		a.tex.Upload(image.Pt(pos.X+size.X, pos.Y), rightPadding, g.zeros(rightPadding.X*rightPadding.Y*4))
+		a.tex.Upload(image.Pt(pos.X+size.X, pos.Y), rightPadding, g.zeros(rightPadding.X*rightPadding.Y*4), 0)
 		bottomPadding := image.Pt(size.X, padding)
-		a.tex.Upload(image.Pt(pos.X, pos.Y+size.Y), bottomPadding, g.zeros(bottomPadding.X*bottomPadding.Y*4))
+		a.tex.Upload(image.Pt(pos.X, pos.Y+size.Y), bottomPadding, g.zeros(bottomPadding.X*bottomPadding.Y*4), 0)
 	}
 	return nil
 }
diff --git a/gpu/internal/d3d11/d3d11_windows.go b/gpu/internal/d3d11/d3d11_windows.go
index 802ef677..671cdb74 100644
--- a/gpu/internal/d3d11/d3d11_windows.go
+++ b/gpu/internal/d3d11/d3d11_windows.go
@@ -556,8 +556,10 @@ func (b *Backend) DispatchCompute(x, y, z int) {
 	panic("not implemented")
 }
 
-func (t *Texture) Upload(offset, size image.Point, pixels []byte) {
-	stride := size.X * 4
+func (t *Texture) Upload(offset, size image.Point, pixels []byte, stride int) {
+	if stride == 0 {
+		stride = size.X * 4
+	}
 	dst := &d3d11.BOX{
 		Left:   uint32(offset.X),
 		Top:    uint32(offset.Y),
diff --git a/gpu/internal/driver/driver.go b/gpu/internal/driver/driver.go
index a66f9e43..982f8715 100644
--- a/gpu/internal/driver/driver.go
+++ b/gpu/internal/driver/driver.go
@@ -167,7 +167,7 @@ type Timer interface {
 }
 
 type Texture interface {
-	Upload(offset, size image.Point, pixels []byte)
+	Upload(offset, size image.Point, pixels []byte, stride int)
 	Release()
 }
 
@@ -261,11 +261,8 @@ func flipImageY(stride, height int, pixels []byte) {
 func UploadImage(t Texture, offset image.Point, img *image.RGBA) {
 	var pixels []byte
 	size := img.Bounds().Size()
-	if img.Stride != size.X*4 {
-		panic("unsupported stride")
-	}
 	start := img.PixOffset(0, 0)
 	end := img.PixOffset(size.X, size.Y-1)
 	pixels = img.Pix[start:end]
-	t.Upload(offset, size, pixels)
+	t.Upload(offset, size, pixels, img.Stride)
 }
diff --git a/gpu/internal/opengl/opengl.go b/gpu/internal/opengl/opengl.go
index a8305216..e31670f4 100644
--- a/gpu/internal/opengl/opengl.go
+++ b/gpu/internal/opengl/opengl.go
@@ -75,10 +75,11 @@ type glState struct {
 		srcRGB, dstRGB gl.Enum
 		srcA, dstA     gl.Enum
 	}
-	depthTest  bool
-	clearColor [4]float32
-	clearDepth float32
-	viewport   [4]int
+	depthTest         bool
+	clearColor        [4]float32
+	clearDepth        float32
+	viewport          [4]int
+	unpack_row_length int
 }
 
 type state struct {
@@ -270,16 +271,17 @@ func (b *Backend) EndFrame() {
 
 func (b *Backend) queryState() glState {
 	s := glState{
-		prog:       gl.Program(b.funcs.GetBinding(gl.CURRENT_PROGRAM)),
-		arrayBuf:   gl.Buffer(b.funcs.GetBinding(gl.ARRAY_BUFFER_BINDING)),
-		elemBuf:    gl.Buffer(b.funcs.GetBinding(gl.ELEMENT_ARRAY_BUFFER_BINDING)),
-		drawFBO:    gl.Framebuffer(b.funcs.GetBinding(gl.FRAMEBUFFER_BINDING)),
-		depthMask:  b.funcs.GetInteger(gl.DEPTH_WRITEMASK) != gl.FALSE,
-		depthTest:  b.funcs.IsEnabled(gl.DEPTH_TEST),
-		depthFunc:  gl.Enum(b.funcs.GetInteger(gl.DEPTH_FUNC)),
-		clearDepth: b.funcs.GetFloat(gl.DEPTH_CLEAR_VALUE),
-		clearColor: b.funcs.GetFloat4(gl.COLOR_CLEAR_VALUE),
-		viewport:   b.funcs.GetInteger4(gl.VIEWPORT),
+		prog:              gl.Program(b.funcs.GetBinding(gl.CURRENT_PROGRAM)),
+		arrayBuf:          gl.Buffer(b.funcs.GetBinding(gl.ARRAY_BUFFER_BINDING)),
+		elemBuf:           gl.Buffer(b.funcs.GetBinding(gl.ELEMENT_ARRAY_BUFFER_BINDING)),
+		drawFBO:           gl.Framebuffer(b.funcs.GetBinding(gl.FRAMEBUFFER_BINDING)),
+		depthMask:         b.funcs.GetInteger(gl.DEPTH_WRITEMASK) != gl.FALSE,
+		depthTest:         b.funcs.IsEnabled(gl.DEPTH_TEST),
+		depthFunc:         gl.Enum(b.funcs.GetInteger(gl.DEPTH_FUNC)),
+		clearDepth:        b.funcs.GetFloat(gl.DEPTH_CLEAR_VALUE),
+		clearColor:        b.funcs.GetFloat4(gl.COLOR_CLEAR_VALUE),
+		viewport:          b.funcs.GetInteger4(gl.VIEWPORT),
+		unpack_row_length: b.funcs.GetInteger(gl.UNPACK_ROW_LENGTH),
 	}
 	s.blend.enable = b.funcs.IsEnabled(gl.BLEND)
 	s.blend.srcRGB = gl.Enum(b.funcs.GetInteger(gl.BLEND_SRC_RGB))
@@ -358,6 +360,7 @@ func (b *Backend) restoreState(dst glState) {
 	src.bindBuffer(f, gl.ARRAY_BUFFER, dst.arrayBuf)
 	v := dst.viewport
 	src.setViewport(f, v[0], v[1], v[2], v[3])
+	src.pixelStorei(f, gl.UNPACK_ROW_LENGTH, dst.unpack_row_length)
 }
 
 func (s *glState) setVertexAttribArray(f *gl.Functions, idx int, enabled bool) {
@@ -559,6 +562,16 @@ func (s *glState) bindBuffer(f *gl.Functions, target gl.Enum, buf gl.Buffer) {
 	f.BindBuffer(target, buf)
 }
 
+func (s *glState) pixelStorei(f *gl.Functions, pname gl.Enum, val int) {
+	if pname != gl.UNPACK_ROW_LENGTH {
+		panic("unsupported PixelStorei pname")
+	}
+	if val != s.unpack_row_length {
+		f.PixelStorei(pname, val)
+		s.unpack_row_length = val
+	}
+}
+
 func (s *glState) setClearDepth(f *gl.Functions, d float32) {
 	if d != s.clearDepth {
 		f.ClearDepthf(d)
@@ -1250,11 +1263,12 @@ func (t *gpuTexture) Release() {
 	t.backend.glstate.deleteTexture(t.backend.funcs, t.obj)
 }
 
-func (t *gpuTexture) Upload(offset, size image.Point, pixels []byte) {
+func (t *gpuTexture) Upload(offset, size image.Point, pixels []byte, stride int) {
 	if min := size.X * size.Y * 4; min > len(pixels) {
 		panic(fmt.Errorf("size %d larger than data %d", min, len(pixels)))
 	}
 	t.backend.BindTexture(0, t)
+	t.backend.glstate.pixelStorei(t.backend.funcs, gl.UNPACK_ROW_LENGTH, stride/4)
 	t.backend.funcs.TexSubImage2D(gl.TEXTURE_2D, 0, offset.X, offset.Y, size.X, size.Y, t.triple.format, t.triple.typ, pixels)
 }
 
diff --git a/internal/gl/gl.go b/internal/gl/gl.go
index 6b4a3dd7..40da7203 100644
--- a/internal/gl/gl.go
+++ b/internal/gl/gl.go
@@ -106,6 +106,7 @@ const (
 	UNIFORM_BUFFER                        = 0x8A11
 	UNIFORM_BUFFER_BINDING                = 0x8A28
 	UNPACK_ALIGNMENT                      = 0xcf5
+	UNPACK_ROW_LENGTH                     = 0x0CF2
 	UNSIGNED_BYTE                         = 0x1401
 	UNSIGNED_SHORT                        = 0x1403
 	VIEWPORT                              = 0x0BA2
diff --git a/internal/gl/gl_js.go b/internal/gl/gl_js.go
index 1ccec79f..f071be9d 100644
--- a/internal/gl/gl_js.go
+++ b/internal/gl/gl_js.go
@@ -343,7 +343,7 @@ func (f *Functions) IsEnabled(cap Enum) bool {
 func (f *Functions) LinkProgram(p Program) {
 	f.Ctx.Call("linkProgram", js.Value(p))
 }
-func (f *Functions) PixelStorei(pname Enum, param int32) {
+func (f *Functions) PixelStorei(pname Enum, param int) {
 	f.Ctx.Call("pixelStorei", int(pname), param)
 }
 func (f *Functions) MemoryBarrier(barriers Enum) {
diff --git a/internal/gl/gl_unix.go b/internal/gl/gl_unix.go
index 5992d41e..5a7d61e3 100644
--- a/internal/gl/gl_unix.go
+++ b/internal/gl/gl_unix.go
@@ -1101,7 +1101,7 @@ func (f *Functions) LinkProgram(p Program) {
 	C.glLinkProgram(&f.f, C.GLuint(p.V))
 }
 
-func (f *Functions) PixelStorei(pname Enum, param int32) {
+func (f *Functions) PixelStorei(pname Enum, param int) {
 	C.glPixelStorei(&f.f, C.GLenum(pname), C.GLint(param))
 }
 
diff --git a/internal/gl/gl_windows.go b/internal/gl/gl_windows.go
index 434513e3..518f6081 100644
--- a/internal/gl/gl_windows.go
+++ b/internal/gl/gl_windows.go
@@ -409,7 +409,7 @@ func (f *Functions) IsEnabled(cap Enum) bool {
 func (c *Functions) LinkProgram(p Program) {
 	syscall.Syscall(_glLinkProgram.Addr(), 1, uintptr(p.V), 0, 0)
 }
-func (c *Functions) PixelStorei(pname Enum, param int32) {
+func (c *Functions) PixelStorei(pname Enum, param int) {
 	syscall.Syscall(_glPixelStorei.Addr(), 2, uintptr(pname), uintptr(param), 0)
 }
 func (f *Functions) MemoryBarrier(barriers Enum) {