gpu: [compute] add CPU fallback

This change adds a CPU fallback for devices that don't support the old renderer nor have GPU support for compute programs. Most of the hard work is implemented in the gioui.org/cpu module. It uses the SwiftShader project with light modification to output statically compiled CPU .o files for each compute program. The CPU fallback only covers Linux and Android on arm, arm64, amd64 architectures. There is no fundamental reason support can't be extended to other platforms: - macOS and iOS are probably easy, but it's likely that virtually every device has GPU support for compute shaders. - Windows needs a Cgo-less port, or a build constraint to require a C compiler (Gio core doesn't). - FreeBSD and OpenBSD are probably also easy to do because they're so similar to Linux. - The 386 binaries didn't work properly in my tests, so fixes to SwiftShader is probably needed. However, I expect virtually every Intel device can run amd64 binaries. Updates gio#49 Fixes gio#228 Signed-off-by: Elias Naur <mail@eliasnaur.com>
2021-02-04 18:19:51 +01:00
parent f14c151883
commit ea38195e2e
5 changed files with 424 additions and 108 deletions
@@ -8,4 +8,4 @@ require (
 	golang.org/x/sys v0.0.0-20210630005230-0f9fa26af87c
 )

-require golang.org/x/text v0.3.6 // indirect
+require gioui.org/cpu v0.0.0-20210727122813-41509bcd3462
@@ -1,6 +1,8 @@
 cloud.google.com/go v0.26.0/go.mod h1:aQUYkXzVsufM+DwF1aE+0xfcU+56JwCaLick0ClmMTw=
 cloud.google.com/go v0.34.0/go.mod h1:aQUYkXzVsufM+DwF1aE+0xfcU+56JwCaLick0ClmMTw=
 dmitri.shuralyov.com/gpu/mtl v0.0.0-20201218220906-28db891af037/go.mod h1:H6x//7gZCb22OMCxBHrMx7a5I7Hp++hsVxbQ4BYO7hU=
+gioui.org/cpu v0.0.0-20210727122813-41509bcd3462 h1:JZyB+d8tPExZHNZwMiGKeeAVd0mkFTc3Zsmegdn178M=
+gioui.org/cpu v0.0.0-20210727122813-41509bcd3462/go.mod h1:DkhBDuHokSMOUxX5LZQ7IcxyJJzs3OON8Z5ojaXUXxo=
 github.com/BurntSushi/toml v0.3.1/go.mod h1:xHWCNGjB5oqiDr8zfno3MHue2Ht5sIBksp03qcyfWMU=
 github.com/BurntSushi/xgb v0.0.0-20160522181843-27f122750802/go.mod h1:IVnqGOEym/WlBOVXweHU+Q+/VP0lqqI8lqeDx9IjBqo=
 github.com/Knetic/govaluate v3.0.1-0.20171022003610-9aa49832a739+incompatible/go.mod h1:r7JcOSlj0wfOMncg0iLm8Leh48TZaKVeNIfJntJ2wa0=
@@ -14,6 +14,7 @@ import (
 	"io/ioutil"
 	"math"
 	"math/bits"
+	"runtime"
 	"sort"
 	"time"
 	"unsafe"
@@ -28,6 +29,9 @@ import (
 	"gioui.org/layout"
 	"gioui.org/op"
 	"gioui.org/op/clip"
+
+	"gioui.org/cpu"
+	"gioui.org/cpu/piet"
 )

 type compute struct {
@@ -40,16 +44,16 @@ type compute struct {
 	maxTextureDim int

 	programs struct {
-		elements   driver.Program
-		tileAlloc  driver.Program
-		pathCoarse driver.Program
-		backdrop   driver.Program
-		binning    driver.Program
-		coarse     driver.Program
-		kernel4    driver.Program
+		elements   computeProgram
+		tileAlloc  computeProgram
+		pathCoarse computeProgram
+		backdrop   computeProgram
+		binning    computeProgram
+		coarse     computeProgram
+		kernel4    computeProgram
 	}
 	buffers struct {
-		config driver.Buffer
+		config sizedBuffer
 		scene  sizedBuffer
 		state  sizedBuffer
 		memory sizedBuffer
@@ -66,6 +70,8 @@ type compute struct {
 		layerVertices []layerVertex
 		layerAtlases  []*layerAtlas
 		packer        packer
+
+		descriptors *piet.Kernel4DescriptorSetLayout
 	}
 	// images contains ImageOp images packed into a texture atlas.
 	images struct {
@@ -94,6 +100,12 @@ type compute struct {

 		uniforms *materialUniforms
 		uniBuf   driver.Buffer
+
+		// CPU fields
+		cpuTex cpu.ImageDescriptor
+		// regions track new materials in tex, so they can be transferred to cpuTex.
+		regions []image.Rectangle
+		scratch []byte
 	}
 	timers struct {
 		profile string
@@ -103,6 +115,10 @@ type compute struct {
 		blit    *timer
 	}

+	// CPU fallback fields.
+	useCPU     bool
+	dispatcher *dispatcher
+
 	// The following fields hold scratch space to avoid garbage.
 	zeroSlice []byte
 	memHeader *memoryHeader
@@ -124,10 +140,11 @@ type layerPlace struct {
 type layerAtlas struct {
 	// image is the layer atlas texture. Note that it is in RGBA format,
 	// but contains data in sRGB. See blitLayers for more detail.
-	image  driver.Texture
-	fbo    driver.Framebuffer
-	size   image.Point
-	layers int
+	image    driver.Texture
+	fbo      driver.Framebuffer
+	cpuImage cpu.ImageDescriptor
+	size     image.Point
+	layers   int
 }

 type copyUniforms struct {
@@ -274,9 +291,22 @@ type encodeState struct {
 	clip  f32.Rectangle
 }

+// sizedBuffer holds a GPU buffer, or its equivalent CPU memory.
 type sizedBuffer struct {
 	size   int
 	buffer driver.Buffer
+	// cpuBuf is initialized when useCPU is true.
+	cpuBuf cpu.BufferDescriptor
+}
+
+// computeProgram holds a compute program, or its equivalent CPU implementation.
+type computeProgram struct {
+	prog driver.Program
+
+	// CPU fields.
+	progInfo    *cpu.ProgramInfo
+	descriptors unsafe.Pointer
+	buffers     []*cpu.BufferDescriptor
 }

 // config matches Config in setup.h
@@ -332,7 +362,8 @@ const (
 )

 func newCompute(ctx driver.Device) (*compute, error) {
-	maxDim := ctx.Caps().MaxTextureSize
+	caps := ctx.Caps()
+	maxDim := caps.MaxTextureSize
 	// Large atlas textures cause artifacts due to precision loss in
 	// shaders.
 	if cap := 8192; maxDim > cap {
@@ -344,6 +375,35 @@ func newCompute(ctx driver.Device) (*compute, error) {
 		conf:          new(config),
 		memHeader:     new(memoryHeader),
 	}
+	shaders := []struct {
+		prog *computeProgram
+		src  driver.ShaderSources
+		info *cpu.ProgramInfo
+		hash string
+	}{
+		{&g.programs.elements, shader_elements_comp, piet.ElementsProgramInfo, piet.ElementsHash},
+		{&g.programs.tileAlloc, shader_tile_alloc_comp, piet.Tile_allocProgramInfo, piet.Tile_allocHash},
+		{&g.programs.pathCoarse, shader_path_coarse_comp, piet.Path_coarseProgramInfo, piet.Path_coarseHash},
+		{&g.programs.backdrop, shader_backdrop_comp, piet.BackdropProgramInfo, piet.BackdropHash},
+		{&g.programs.binning, shader_binning_comp, piet.BinningProgramInfo, piet.BinningHash},
+		{&g.programs.coarse, shader_coarse_comp, piet.CoarseProgramInfo, piet.CoarseHash},
+		{&g.programs.kernel4, shader_kernel4_comp, piet.Kernel4ProgramInfo, piet.Kernel4Hash},
+	}
+	if !caps.Features.Has(driver.FeatureCompute) {
+		g.useCPU = supportsCPUCompute
+		for _, s := range shaders {
+			if s.src.Hash != s.hash {
+				g.useCPU = false
+				break
+			}
+		}
+		if !g.useCPU {
+			return nil, errors.New("gpu: missing support for compute programs")
+		}
+	}
+	if g.useCPU {
+		g.dispatcher = newDispatcher(runtime.NumCPU())
+	}

 	// Large enough for reasonable fill sizes, yet still spannable by the compute programs.
 	g.output.packer.maxDim = 4096
@@ -397,32 +457,55 @@ func newCompute(ctx driver.Device) (*compute, error) {
 	g.materials.uniBuf = buf
 	g.materials.prog.SetVertexUniforms(buf)

-	buf, err = ctx.NewBuffer(driver.BufferBindingShaderStorage, int(unsafe.Sizeof(config{})))
-	if err != nil {
-		g.Release()
-		return nil, err
-	}
-	g.buffers.config = buf
-
-	shaders := []struct {
-		prog *driver.Program
-		src  driver.ShaderSources
-	}{
-		{&g.programs.elements, shader_elements_comp},
-		{&g.programs.tileAlloc, shader_tile_alloc_comp},
-		{&g.programs.pathCoarse, shader_path_coarse_comp},
-		{&g.programs.backdrop, shader_backdrop_comp},
-		{&g.programs.binning, shader_binning_comp},
-		{&g.programs.coarse, shader_coarse_comp},
-		{&g.programs.kernel4, shader_kernel4_comp},
-	}
 	for _, shader := range shaders {
-		p, err := ctx.NewComputeProgram(shader.src)
-		if err != nil {
-			g.Release()
-			return nil, err
+		if !g.useCPU {
+			p, err := ctx.NewComputeProgram(shader.src)
+			if err != nil {
+				g.Release()
+				return nil, err
+			}
+			shader.prog.prog = p
+		} else {
+			shader.prog.progInfo = shader.info
+		}
+	}
+	if g.useCPU {
+		{
+			desc := new(piet.ElementsDescriptorSetLayout)
+			g.programs.elements.descriptors = unsafe.Pointer(desc)
+			g.programs.elements.buffers = []*cpu.BufferDescriptor{desc.Binding0(), desc.Binding1(), desc.Binding2(), desc.Binding3()}
+		}
+		{
+			desc := new(piet.Tile_allocDescriptorSetLayout)
+			g.programs.tileAlloc.descriptors = unsafe.Pointer(desc)
+			g.programs.tileAlloc.buffers = []*cpu.BufferDescriptor{desc.Binding0(), desc.Binding1()}
+		}
+		{
+			desc := new(piet.Path_coarseDescriptorSetLayout)
+			g.programs.pathCoarse.descriptors = unsafe.Pointer(desc)
+			g.programs.pathCoarse.buffers = []*cpu.BufferDescriptor{desc.Binding0(), desc.Binding1()}
+		}
+		{
+			desc := new(piet.BackdropDescriptorSetLayout)
+			g.programs.backdrop.descriptors = unsafe.Pointer(desc)
+			g.programs.backdrop.buffers = []*cpu.BufferDescriptor{desc.Binding0(), desc.Binding1()}
+		}
+		{
+			desc := new(piet.BinningDescriptorSetLayout)
+			g.programs.binning.descriptors = unsafe.Pointer(desc)
+			g.programs.binning.buffers = []*cpu.BufferDescriptor{desc.Binding0(), desc.Binding1()}
+		}
+		{
+			desc := new(piet.CoarseDescriptorSetLayout)
+			g.programs.coarse.descriptors = unsafe.Pointer(desc)
+			g.programs.coarse.buffers = []*cpu.BufferDescriptor{desc.Binding0(), desc.Binding1()}
+		}
+		{
+			desc := new(piet.Kernel4DescriptorSetLayout)
+			g.programs.kernel4.descriptors = unsafe.Pointer(desc)
+			g.programs.kernel4.buffers = []*cpu.BufferDescriptor{desc.Binding0(), desc.Binding1()}
+			g.output.descriptors = desc
 		}
-		*shader.prog = p
 	}
 	return g, nil
 }
@@ -553,7 +636,7 @@ func (g *compute) compactLayers() error {
 			continue
 		}
 		outputSize := g.output.packer.sizes[0]
-		atlas.ensureSize(g.ctx, outputSize)
+		atlas.ensureSize(g.useCPU, g.ctx, outputSize)
 		for i, l := range layers[:end] {
 			if l.newPlace == l.place {
 				continue
@@ -629,10 +712,10 @@ func (g *compute) renderLayers(viewport image.Point) error {
 			Y: (outputSize.Y + tileHeightPx - 1) / tileHeightPx,
 		}
 		w, h := tileDims.X*tileWidthPx, tileDims.Y*tileHeightPx
-		if err := atlas.ensureSize(g.ctx, image.Pt(w, h)); err != nil {
+		if err := atlas.ensureSize(g.useCPU, g.ctx, image.Pt(w, h)); err != nil {
 			return err
 		}
-		if err := g.render(atlas.image, tileDims); err != nil {
+		if err := g.render(atlas.image, atlas.cpuImage, tileDims, atlas.size.X*4); err != nil {
 			return err
 		}
 	}
@@ -694,7 +777,7 @@ func (g *compute) blitLayers(viewport image.Point) {
 		g.output.uniforms.uvScale = [2]float32{1 / float32(atlas.size.X), 1 / float32(atlas.size.Y)}
 		g.output.uniBuf.Upload(byteslice.Struct(g.output.uniforms))
 		vertexData := byteslice.Slice(g.output.layerVertices)
-		g.output.buffer.ensureCapacity(g.ctx, driver.BufferBindingVertices, len(vertexData))
+		g.output.buffer.ensureCapacity(false, g.ctx, driver.BufferBindingVertices, len(vertexData))
 		g.output.buffer.buffer.Upload(vertexData)
 		g.ctx.BindVertexBuffer(g.output.buffer.buffer, int(unsafe.Sizeof(g.output.layerVertices[0])), 0)
 		g.ctx.BindTexture(0, atlas.image)
@@ -705,6 +788,7 @@ func (g *compute) blitLayers(viewport image.Point) {
 func (g *compute) renderMaterials() error {
 	m := &g.materials
 	m.quads = m.quads[:0]
+	m.regions = m.regions[:0]
 	resize := false
 	reclaimed := false
 restart:
@@ -752,6 +836,10 @@ restart:
 			}
 			m.offsets[op.key] = offset
 			g.enc.setFillImageOffset(op.sceneIdx, offset.Sub(op.off))
+			m.regions = append(m.regions, image.Rectangle{
+				Min: place.Pos,
+				Max: place.Pos.Add(size),
+			})
 		}
 		break
 	}
@@ -768,6 +856,7 @@ restart:
 			m.tex.Release()
 			m.tex = nil
 		}
+		m.cpuTex.Free()
 		handle, err := g.ctx.NewTexture(driver.TextureFormatRGBA8, texSize, texSize,
 			driver.FilterNearest, driver.FilterNearest,
 			driver.BufferBindingShaderStorage|driver.BufferBindingFramebuffer)
@@ -781,6 +870,9 @@ restart:
 		}
 		m.tex = handle
 		m.fbo = fbo
+		if g.useCPU {
+			m.cpuTex = cpu.NewImageRGBA(texSize, texSize)
+		}
 	}
 	// Transform to clip space: [-1, -1] - [1, 1].
 	g.materials.uniforms.scale = [2]float32{2 / float32(texSize), 2 / float32(texSize)}
@@ -788,7 +880,7 @@ restart:
 	g.materials.uniBuf.Upload(byteslice.Struct(g.materials.uniforms))
 	vertexData := byteslice.Slice(m.quads)
 	n := pow2Ceil(len(vertexData))
-	m.buffer.ensureCapacity(g.ctx, driver.BufferBindingVertices, n)
+	m.buffer.ensureCapacity(false, g.ctx, driver.BufferBindingVertices, n)
 	m.buffer.buffer.Upload(vertexData)
 	g.ctx.BindTexture(0, g.images.tex)
 	g.ctx.BindFramebuffer(m.fbo)
@@ -962,7 +1054,7 @@ func (enc *encoder) encodePath(verts []byte) {
 	}
 }

-func (g *compute) render(dst driver.Texture, tileDims image.Point) error {
+func (g *compute) render(dst driver.Texture, cpuDst cpu.ImageDescriptor, tileDims image.Point, stride int) error {
 	const (
 		// wgSize is the largest and most common workgroup size.
 		wgSize = 128
@@ -985,16 +1077,11 @@ func (g *compute) render(dst driver.Texture, tileDims image.Point) error {
 	if s := len(scene); s > g.buffers.scene.size {
 		realloced = true
 		paddedCap := s * 11 / 10
-		if err := g.buffers.scene.ensureCapacity(g.ctx, driver.BufferBindingShaderStorage, paddedCap); err != nil {
+		if err := g.buffers.scene.ensureCapacity(g.useCPU, g.ctx, driver.BufferBindingShaderStorage, paddedCap); err != nil {
 			return err
 		}
 	}
-	g.buffers.scene.buffer.Upload(scene)
-
-	g.ctx.BindImageTexture(kernel4OutputUnit, dst, driver.AccessWrite, driver.TextureFormatRGBA8)
-	if t := g.materials.tex; t != nil {
-		g.ctx.BindImageTexture(kernel4AtlasUnit, t, driver.AccessRead, driver.TextureFormatRGBA8)
-	}
+	g.buffers.scene.upload(scene)

 	// alloc is the number of allocated bytes for static buffers.
 	var alloc uint32
@@ -1027,12 +1114,14 @@ func (g *compute) render(dst driver.Texture, tileDims image.Point) error {
 	if clearSize > g.buffers.state.size {
 		realloced = true
 		paddedCap := clearSize * 11 / 10
-		if err := g.buffers.state.ensureCapacity(g.ctx, driver.BufferBindingShaderStorage, paddedCap); err != nil {
+		if err := g.buffers.state.ensureCapacity(g.useCPU, g.ctx, driver.BufferBindingShaderStorage, paddedCap); err != nil {
 			return err
 		}
 	}

-	g.buffers.config.Upload(byteslice.Struct(g.conf))
+	confData := byteslice.Struct(g.conf)
+	g.buffers.config.ensureCapacity(g.useCPU, g.ctx, driver.BufferBindingShaderStorage, len(confData))
+	g.buffers.config.upload(confData)

 	minSize := int(unsafe.Sizeof(memoryHeader{})) + int(alloc)
 	if minSize > g.buffers.memory.size {
@@ -1040,45 +1129,53 @@ func (g *compute) render(dst driver.Texture, tileDims image.Point) error {
 		// Add space for dynamic GPU allocations.
 		const sizeBump = 4 * 1024 * 1024
 		minSize += sizeBump
-		if err := g.buffers.memory.ensureCapacity(g.ctx, driver.BufferBindingShaderStorage, minSize); err != nil {
+		if err := g.buffers.memory.ensureCapacity(g.useCPU, g.ctx, driver.BufferBindingShaderStorage, minSize); err != nil {
 			return err
 		}
 	}
+
+	if !g.useCPU {
+		g.ctx.BindImageTexture(kernel4OutputUnit, dst, driver.AccessWrite, driver.TextureFormatRGBA8)
+		if t := g.materials.tex; t != nil {
+			g.ctx.BindImageTexture(kernel4AtlasUnit, t, driver.AccessRead, driver.TextureFormatRGBA8)
+		}
+	} else {
+		*g.output.descriptors.Binding2() = cpuDst
+		*g.output.descriptors.Binding3() = g.materials.cpuTex
+	}
+
 	for {
 		*g.memHeader = memoryHeader{
 			mem_offset: alloc,
 		}
-		g.buffers.memory.buffer.Upload(byteslice.Struct(g.memHeader))
-		g.buffers.state.buffer.Upload(g.zeros(clearSize))
+		g.buffers.memory.upload(byteslice.Struct(g.memHeader))
+		g.buffers.state.upload(g.zeros(clearSize))

 		if realloced {
 			realloced = false
 			g.bindBuffers()
 		}
-		g.ctx.MemoryBarrier()
-		g.ctx.BindProgram(g.programs.elements)
-		g.ctx.DispatchCompute(numPartitions, 1, 1)
-		g.ctx.MemoryBarrier()
-		g.ctx.BindProgram(g.programs.tileAlloc)
-		g.ctx.DispatchCompute((enc.npath+wgSize-1)/wgSize, 1, 1)
-		g.ctx.MemoryBarrier()
-		g.ctx.BindProgram(g.programs.pathCoarse)
-		g.ctx.DispatchCompute((enc.npathseg+31)/32, 1, 1)
-		g.ctx.MemoryBarrier()
-		g.ctx.BindProgram(g.programs.backdrop)
-		g.ctx.DispatchCompute((enc.npath+wgSize-1)/wgSize, 1, 1)
+		g.memoryBarrier()
+		g.dispatch(g.programs.elements, numPartitions, 1, 1)
+		g.memoryBarrier()
+		g.dispatch(g.programs.tileAlloc, (enc.npath+wgSize-1)/wgSize, 1, 1)
+		g.memoryBarrier()
+		g.dispatch(g.programs.pathCoarse, (enc.npathseg+31)/32, 1, 1)
+		g.memoryBarrier()
+		g.dispatch(g.programs.backdrop, (enc.npath+wgSize-1)/wgSize, 1, 1)
 		// No barrier needed between backdrop and binning.
-		g.ctx.BindProgram(g.programs.binning)
-		g.ctx.DispatchCompute((enc.npath+wgSize-1)/wgSize, 1, 1)
-		g.ctx.MemoryBarrier()
-		g.ctx.BindProgram(g.programs.coarse)
-		g.ctx.DispatchCompute(widthInBins, heightInBins, 1)
-		g.ctx.MemoryBarrier()
-		g.ctx.BindProgram(g.programs.kernel4)
-		g.ctx.DispatchCompute(tileDims.X, tileDims.Y, 1)
-		g.ctx.MemoryBarrier()
+		g.dispatch(g.programs.binning, (enc.npath+wgSize-1)/wgSize, 1, 1)
+		g.memoryBarrier()
+		g.dispatch(g.programs.coarse, widthInBins, heightInBins, 1)
+		g.memoryBarrier()
+		g.downloadMaterials()
+		g.dispatch(g.programs.kernel4, tileDims.X, tileDims.Y, 1)
+		g.memoryBarrier()
+		if g.useCPU {
+			g.dispatcher.Sync()
+		}

-		if err := g.buffers.memory.buffer.Download(byteslice.Struct(g.memHeader)); err != nil {
+		if err := g.buffers.memory.download(byteslice.Struct(g.memHeader)); err != nil {
 			if err == driver.ErrContentLost {
 				continue
 			}
@@ -1086,12 +1183,16 @@ func (g *compute) render(dst driver.Texture, tileDims image.Point) error {
 		}
 		switch errCode := g.memHeader.mem_error; errCode {
 		case memNoError:
+			if g.useCPU {
+				w, h := tileDims.X*tileWidthPx, tileDims.Y*tileHeightPx
+				dst.Upload(image.Pt(0, 0), image.Pt(w, h), cpuDst.Data(), stride)
+			}
 			return nil
 		case memMallocFailed:
 			// Resize memory and try again.
 			realloced = true
 			sz := g.buffers.memory.size * 15 / 10
-			if err := g.buffers.memory.ensureCapacity(g.ctx, driver.BufferBindingShaderStorage, sz); err != nil {
+			if err := g.buffers.memory.ensureCapacity(g.useCPU, g.ctx, driver.BufferBindingShaderStorage, sz); err != nil {
 				return err
 			}
 			continue
@@ -1101,6 +1202,48 @@ func (g *compute) render(dst driver.Texture, tileDims image.Point) error {
 	}
 }

+func (g *compute) downloadMaterials() {
+	m := &g.materials
+	if !g.useCPU || len(m.regions) == 0 {
+		return
+	}
+	copyFBO := m.fbo
+	data := m.cpuTex.Data()
+	for _, r := range m.regions {
+		dims := r.Size()
+		if n := dims.X * dims.Y * 4; n > len(m.scratch) {
+			m.scratch = make([]byte, n)
+		}
+		copyFBO.ReadPixels(r, m.scratch)
+		stride := m.packer.maxDim * 4
+		col := r.Min.X * 4
+		row := stride * r.Min.Y
+		off := col + row
+		w := dims.X * 4
+		for y := 0; y < dims.Y; y++ {
+			copy(data[off:off+w], m.scratch[y*dims.X*4:])
+			off += stride
+		}
+	}
+}
+
+func (g *compute) memoryBarrier() {
+	if !g.useCPU {
+		g.ctx.MemoryBarrier()
+	} else {
+		g.dispatcher.Barrier()
+	}
+}
+
+func (g *compute) dispatch(p computeProgram, x, y, z int) {
+	if !g.useCPU {
+		g.ctx.BindProgram(p.prog)
+		g.ctx.DispatchCompute(x, y, z)
+	} else {
+		g.dispatcher.Dispatch(p.progInfo, p.descriptors, x, y, z)
+	}
+}
+
 // zeros returns a byte slice with size bytes of zeros.
 func (g *compute) zeros(size int) []byte {
 	if cap(g.zeroSlice) < size {
@@ -1109,7 +1252,7 @@ func (g *compute) zeros(size int) []byte {
 	return g.zeroSlice[:size]
 }

-func (a *layerAtlas) ensureSize(ctx driver.Device, size image.Point) error {
+func (a *layerAtlas) ensureSize(useCPU bool, ctx driver.Device, size image.Point) error {
 	if a.size.X >= size.X && a.size.Y >= size.Y {
 		return nil
 	}
@@ -1122,6 +1265,8 @@ func (a *layerAtlas) ensureSize(ctx driver.Device, size image.Point) error {
 		a.image.Release()
 		a.image = nil
 	}
+	a.cpuImage.Free()
+
 	img, err := ctx.NewTexture(driver.TextureFormatRGBA8, size.X, size.Y,
 		driver.FilterNearest,
 		driver.FilterNearest,
@@ -1136,29 +1281,35 @@ func (a *layerAtlas) ensureSize(ctx driver.Device, size image.Point) error {
 	}
 	a.fbo = fbo
 	a.image = img
+	if useCPU {
+		a.cpuImage = cpu.NewImageRGBA(size.X, size.Y)
+	}
 	a.size = size
 	return nil
 }

 func (g *compute) Release() {
+	if g.useCPU {
+		g.dispatcher.Stop()
+	}
 	type resource interface {
 		Release()
 	}
 	res := []resource{
-		g.programs.elements,
-		g.programs.tileAlloc,
-		g.programs.pathCoarse,
-		g.programs.backdrop,
-		g.programs.binning,
-		g.programs.coarse,
-		g.programs.kernel4,
+		&g.programs.elements,
+		&g.programs.tileAlloc,
+		&g.programs.pathCoarse,
+		&g.programs.backdrop,
+		&g.programs.binning,
+		&g.programs.coarse,
+		&g.programs.kernel4,
 		g.output.blitProg,
 		&g.output.buffer,
 		g.output.uniBuf,
 		&g.buffers.scene,
 		&g.buffers.state,
 		&g.buffers.memory,
-		g.buffers.config,
+		&g.buffers.config,
 		g.images.tex,
 		g.materials.layout,
 		g.materials.prog,
@@ -1168,6 +1319,7 @@ func (g *compute) Release() {
 		g.materials.uniBuf,
 		g.timers.t,
 	}
+	g.materials.cpuTex.Free()
 	for _, r := range res {
 		if r != nil {
 			r.Release()
@@ -1180,48 +1332,82 @@ func (g *compute) Release() {
 		if a.image != nil {
 			a.image.Release()
 		}
+		a.cpuImage.Free()
 	}

 	*g = compute{}
 }

 func (g *compute) bindBuffers() {
-	bindStorageBuffers(g.programs.elements, g.buffers.memory.buffer, g.buffers.config, g.buffers.scene.buffer, g.buffers.state.buffer)
-	bindStorageBuffers(g.programs.tileAlloc, g.buffers.memory.buffer, g.buffers.config)
-	bindStorageBuffers(g.programs.pathCoarse, g.buffers.memory.buffer, g.buffers.config)
-	bindStorageBuffers(g.programs.backdrop, g.buffers.memory.buffer, g.buffers.config)
-	bindStorageBuffers(g.programs.binning, g.buffers.memory.buffer, g.buffers.config)
-	bindStorageBuffers(g.programs.coarse, g.buffers.memory.buffer, g.buffers.config)
-	bindStorageBuffers(g.programs.kernel4, g.buffers.memory.buffer, g.buffers.config)
+	g.bindStorageBuffers(g.programs.elements, g.buffers.memory, g.buffers.config, g.buffers.scene, g.buffers.state)
+	g.bindStorageBuffers(g.programs.tileAlloc, g.buffers.memory, g.buffers.config)
+	g.bindStorageBuffers(g.programs.pathCoarse, g.buffers.memory, g.buffers.config)
+	g.bindStorageBuffers(g.programs.backdrop, g.buffers.memory, g.buffers.config)
+	g.bindStorageBuffers(g.programs.binning, g.buffers.memory, g.buffers.config)
+	g.bindStorageBuffers(g.programs.coarse, g.buffers.memory, g.buffers.config)
+	g.bindStorageBuffers(g.programs.kernel4, g.buffers.memory, g.buffers.config)
+}
+
+func (p *computeProgram) Release() {
+	if p.prog != nil {
+		p.prog.Release()
+	}
+	*p = computeProgram{}
 }

 func (b *sizedBuffer) Release() {
 	if b.buffer == nil {
 		return
 	}
-	b.buffer.Release()
+	b.cpuBuf.Free()
 	*b = sizedBuffer{}
 }

-func (b *sizedBuffer) ensureCapacity(ctx driver.Device, binding driver.BufferBinding, size int) error {
+func (b *sizedBuffer) ensureCapacity(useCPU bool, ctx driver.Device, binding driver.BufferBinding, size int) error {
 	if b.size >= size {
 		return nil
 	}
 	if b.buffer != nil {
 		b.Release()
 	}
-	buf, err := ctx.NewBuffer(binding, size)
-	if err != nil {
-		return err
+	b.cpuBuf.Free()
+	if !useCPU {
+		buf, err := ctx.NewBuffer(binding, size)
+		if err != nil {
+			return err
+		}
+		b.buffer = buf
+	} else {
+		b.cpuBuf = cpu.NewBuffer(size)
 	}
-	b.buffer = buf
 	b.size = size
 	return nil
 }

-func bindStorageBuffers(prog driver.Program, buffers ...driver.Buffer) {
+func (b *sizedBuffer) download(data []byte) error {
+	if b.buffer != nil {
+		return b.buffer.Download(data)
+	} else {
+		copy(data, b.cpuBuf.Data())
+		return nil
+	}
+}
+
+func (b *sizedBuffer) upload(data []byte) {
+	if b.buffer != nil {
+		b.buffer.Upload(data)
+	} else {
+		copy(b.cpuBuf.Data(), data)
+	}
+}
+
+func (g *compute) bindStorageBuffers(prog computeProgram, buffers ...sizedBuffer) {
 	for i, buf := range buffers {
-		prog.SetStorageBuffer(i, buf)
+		if !g.useCPU {
+			prog.prog.SetStorageBuffer(i, buf.buffer)
+		} else {
+			*prog.buffers[i] = buf.cpuBuf
+		}
 	}
 }

@@ -0,0 +1,132 @@
+// SPDX-License-Identifier: Unlicense OR MIT
+
+package gpu
+
+import (
+	"runtime"
+	"unsafe"
+
+	"gioui.org/cpu"
+)
+
+const supportsCPUCompute = runtime.GOARCH == "amd64" || runtime.GOARCH == "arm64" || runtime.GOARCH == "arm"
+
+// This file contains code specific to running compute shaders on the CPU.
+
+// dispatcher dispatches CPU compute programs across multiple goroutines.
+type dispatcher struct {
+	// done is notified when a worker completes its work slice.
+	done chan struct{}
+	// work receives work slice indices. It is closed when the dispatcher is released.
+	work chan work
+	// dispatch receives compute jobs, which is then split among workers.
+	dispatch chan dispatch
+	// sync receives notification when a Sync completes.
+	sync chan struct{}
+}
+
+type work struct {
+	ctx   *cpu.DispatchContext
+	index int
+}
+
+type dispatch struct {
+	_type   jobType
+	program *cpu.ProgramInfo
+	descSet unsafe.Pointer
+	x, y, z int
+}
+
+type jobType uint8
+
+const (
+	jobDispatch jobType = iota
+	jobBarrier
+	jobSync
+)
+
+func newDispatcher(workers int) *dispatcher {
+	d := &dispatcher{
+		work: make(chan work, workers),
+		done: make(chan struct{}, workers),
+		// Leave some room to avoid blocking calls to Dispatch.
+		dispatch: make(chan dispatch, 20),
+		sync:     make(chan struct{}),
+	}
+	for i := 0; i < workers; i++ {
+		go d.worker()
+	}
+	go d.dispatcher()
+	return d
+}
+
+func (d *dispatcher) dispatcher() {
+	defer close(d.work)
+	var free []*cpu.DispatchContext
+	defer func() {
+		for _, ctx := range free {
+			ctx.Free()
+		}
+	}()
+	var used []*cpu.DispatchContext
+	for job := range d.dispatch {
+		switch job._type {
+		case jobDispatch:
+			if len(free) == 0 {
+				free = append(free, cpu.NewDispatchContext())
+			}
+			ctx := free[len(free)-1]
+			free = free[:len(free)-1]
+			used = append(used, ctx)
+			ctx.Prepare(cap(d.work), job.program, job.descSet, job.x, job.y, job.z)
+			for i := 0; i < cap(d.work); i++ {
+				d.work <- work{
+					ctx:   ctx,
+					index: i,
+				}
+			}
+		case jobBarrier:
+			// Wait for all outstanding dispatches to complete.
+			for i := 0; i < len(used)*cap(d.work); i++ {
+				<-d.done
+			}
+			free = append(free, used...)
+			used = used[:0]
+		case jobSync:
+			d.sync <- struct{}{}
+		}
+	}
+}
+
+func (d *dispatcher) worker() {
+	thread := cpu.NewThreadContext()
+	defer thread.Free()
+	for w := range d.work {
+		w.ctx.Dispatch(w.index, thread)
+		d.done <- struct{}{}
+	}
+}
+
+func (d *dispatcher) Barrier() {
+	d.dispatch <- dispatch{_type: jobBarrier}
+}
+
+func (d *dispatcher) Sync() {
+	d.dispatch <- dispatch{_type: jobSync}
+	<-d.sync
+}
+
+func (d *dispatcher) Dispatch(program *cpu.ProgramInfo, descSet unsafe.Pointer, x, y, z int) {
+	d.dispatch <- dispatch{
+		_type:   jobDispatch,
+		program: program,
+		descSet: descSet,
+		x:       x,
+		y:       y,
+		z:       z,
+	}
+}
+
+func (d *dispatcher) Stop() {
+	close(d.dispatch)
+}
@@ -9,7 +9,6 @@ package gpu

 import (
 	"encoding/binary"
-	"errors"
 	"fmt"
 	"image"
 	"image/color"
@@ -364,11 +363,8 @@ func New(api API) (GPU, error) {
 	switch {
 	case !forceCompute && feats.Has(driver.FeatureFloatRenderTargets):
 		return newGPU(d)
-	case feats.Has(driver.FeatureCompute):
-		return newCompute(d)
-	default:
-		return nil, errors.New("gpu: no support for float render targets nor compute")
 	}
+	return newCompute(d)
 }

 func newGPU(ctx driver.Device) (*gpu, error) {