mirror of
https://github.com/hajimehoshi/ebiten.git
synced 2024-12-27 04:08:53 +01:00
6fafe734da
This change allows increasing ShaderImageCount without breaking the image display. Simply made the uniform processing dynamic based on ShaderImageCount.
530 lines
15 KiB
Go
530 lines
15 KiB
Go
// Copyright 2023 The Ebitengine Authors
|
|
//
|
|
// Licensed under the Apache License, Version 2.0 (the "License");
|
|
// you may not use this file except in compliance with the License.
|
|
// You may obtain a copy of the License at
|
|
//
|
|
// http://www.apache.org/licenses/LICENSE-2.0
|
|
//
|
|
// Unless required by applicable law or agreed to in writing, software
|
|
// distributed under the License is distributed on an "AS IS" BASIS,
|
|
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
// See the License for the specific language governing permissions and
|
|
// limitations under the License.
|
|
|
|
package graphicscommand
|
|
|
|
import (
|
|
"fmt"
|
|
"image"
|
|
"math"
|
|
"sync"
|
|
"sync/atomic"
|
|
|
|
"github.com/hajimehoshi/ebiten/v2/internal/debug"
|
|
"github.com/hajimehoshi/ebiten/v2/internal/graphics"
|
|
"github.com/hajimehoshi/ebiten/v2/internal/graphicsdriver"
|
|
"github.com/hajimehoshi/ebiten/v2/internal/shaderir"
|
|
)
|
|
|
|
const (
|
|
is32bit = 1 >> (^uint(0) >> 63)
|
|
is64bit = 1 - is32bit
|
|
|
|
// MaxVertexCount is the maximum number of vertices for one draw call.
|
|
//
|
|
// On 64bit architectures, this value is 2^32-1, as the index type is uint32.
|
|
// This value cannot be exactly 2^32 especially with WebGL 2, as 2^32th vertex is not rendered correctly.
|
|
// See https://registry.khronos.org/webgl/specs/latest/2.0/#5.18 .
|
|
//
|
|
// On 32bit architectures, this value is an adjusted number so that maxVertexFloatCount doesn't overflow int.
|
|
MaxVertexCount = is64bit*math.MaxUint32 + is32bit*(math.MaxInt32/graphics.VertexFloatCount)
|
|
|
|
maxVertexFloatCount = MaxVertexCount * graphics.VertexFloatCount
|
|
)
|
|
|
|
var vsyncEnabled int32 = 1
|
|
|
|
func SetVsyncEnabled(enabled bool, graphicsDriver graphicsdriver.Graphics) {
|
|
if enabled {
|
|
atomic.StoreInt32(&vsyncEnabled, 1)
|
|
} else {
|
|
atomic.StoreInt32(&vsyncEnabled, 0)
|
|
}
|
|
|
|
runOnRenderThread(func() {
|
|
graphicsDriver.SetVsyncEnabled(enabled)
|
|
}, true)
|
|
}
|
|
|
|
// FlushCommands flushes the command queue and present the screen if needed.
|
|
// If endFrame is true, the current screen might be used to present.
|
|
func FlushCommands(graphicsDriver graphicsdriver.Graphics, endFrame bool) error {
|
|
if err := theCommandQueueManager.flush(graphicsDriver, endFrame); err != nil {
|
|
return err
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// commandQueue is a command queue for drawing commands.
|
|
type commandQueue struct {
|
|
// commands is a queue of drawing commands.
|
|
commands []command
|
|
|
|
// vertices represents a vertices data in OpenGL's array buffer.
|
|
vertices []float32
|
|
indices []uint32
|
|
|
|
tmpNumVertexFloats int
|
|
|
|
drawTrianglesCommandPool drawTrianglesCommandPool
|
|
|
|
uint32sBuffer uint32sBuffer
|
|
finalizers []func()
|
|
|
|
err atomic.Value
|
|
}
|
|
|
|
// addFinalizer adds a finalizer function to this queue.
|
|
// A finalizer is executed when the command queue is flushed at the end of the frame.
|
|
func (q *commandQueue) addFinalizer(f func()) {
|
|
q.finalizers = append(q.finalizers, f)
|
|
}
|
|
|
|
func (q *commandQueue) appendIndices(indices []uint32, offset uint32) {
|
|
n := len(q.indices)
|
|
q.indices = append(q.indices, indices...)
|
|
for i := n; i < len(q.indices); i++ {
|
|
q.indices[i] += offset
|
|
}
|
|
}
|
|
|
|
// mustUseDifferentVertexBuffer reports whether a different vertex buffer must be used.
|
|
func mustUseDifferentVertexBuffer(nextNumVertexFloats int) bool {
|
|
return nextNumVertexFloats > maxVertexFloatCount
|
|
}
|
|
|
|
// EnqueueDrawTrianglesCommand enqueues a drawing-image command.
|
|
func (q *commandQueue) EnqueueDrawTrianglesCommand(dst *Image, srcs [graphics.ShaderImageCount]*Image, vertices []float32, indices []uint32, blend graphicsdriver.Blend, dstRegion image.Rectangle, srcRegions [graphics.ShaderImageCount]image.Rectangle, shader *Shader, uniforms []uint32, fillRule graphicsdriver.FillRule) {
|
|
if len(vertices) > maxVertexFloatCount {
|
|
panic(fmt.Sprintf("graphicscommand: len(vertices) must equal to or less than %d but was %d", maxVertexFloatCount, len(vertices)))
|
|
}
|
|
|
|
split := false
|
|
if mustUseDifferentVertexBuffer(q.tmpNumVertexFloats + len(vertices)) {
|
|
q.tmpNumVertexFloats = 0
|
|
split = true
|
|
}
|
|
|
|
// Assume that all the image sizes are same.
|
|
// Assume that the images are packed from the front in the slice srcs.
|
|
q.vertices = append(q.vertices, vertices...)
|
|
q.appendIndices(indices, uint32(q.tmpNumVertexFloats/graphics.VertexFloatCount))
|
|
q.tmpNumVertexFloats += len(vertices)
|
|
|
|
// prependPreservedUniforms not only prepends values to the given slice but also creates a new slice.
|
|
// Allocating a new slice is necessary to make EnqueueDrawTrianglesCommand safe so far.
|
|
// TODO: This might cause a performance issue (#2601).
|
|
uniforms = q.prependPreservedUniforms(uniforms, shader, dst, srcs, dstRegion, srcRegions)
|
|
|
|
// Remove unused uniform variables so that more commands can be merged.
|
|
shader.ir.FilterUniformVariables(uniforms)
|
|
|
|
// TODO: If dst is the screen, reorder the command to be the last.
|
|
if !split && 0 < len(q.commands) {
|
|
if last, ok := q.commands[len(q.commands)-1].(*drawTrianglesCommand); ok {
|
|
if last.CanMergeWithDrawTrianglesCommand(dst, srcs, vertices, blend, shader, uniforms, fillRule) {
|
|
last.setVertices(q.lastVertices(len(vertices) + last.numVertices()))
|
|
if last.dstRegions[len(last.dstRegions)-1].Region == dstRegion {
|
|
last.dstRegions[len(last.dstRegions)-1].IndexCount += len(indices)
|
|
} else {
|
|
last.dstRegions = append(last.dstRegions, graphicsdriver.DstRegion{
|
|
Region: dstRegion,
|
|
IndexCount: len(indices),
|
|
})
|
|
}
|
|
return
|
|
}
|
|
}
|
|
}
|
|
|
|
c := q.drawTrianglesCommandPool.get()
|
|
c.dst = dst
|
|
c.srcs = srcs
|
|
c.vertices = q.lastVertices(len(vertices))
|
|
c.blend = blend
|
|
c.dstRegions = []graphicsdriver.DstRegion{
|
|
{
|
|
Region: dstRegion,
|
|
IndexCount: len(indices),
|
|
},
|
|
}
|
|
c.shader = shader
|
|
c.uniforms = uniforms
|
|
c.fillRule = fillRule
|
|
q.commands = append(q.commands, c)
|
|
}
|
|
|
|
func (q *commandQueue) lastVertices(n int) []float32 {
|
|
return q.vertices[len(q.vertices)-n : len(q.vertices)]
|
|
}
|
|
|
|
// Enqueue enqueues a drawing command other than a draw-triangles command.
|
|
//
|
|
// For a draw-triangles command, use EnqueueDrawTrianglesCommand.
|
|
func (q *commandQueue) Enqueue(command command) {
|
|
// TODO: If dst is the screen, reorder the command to be the last.
|
|
q.commands = append(q.commands, command)
|
|
}
|
|
|
|
// Flush flushes the command queue.
|
|
func (q *commandQueue) Flush(graphicsDriver graphicsdriver.Graphics, endFrame bool) error {
|
|
if err := q.err.Load(); err != nil {
|
|
return err.(error)
|
|
}
|
|
|
|
var sync bool
|
|
// Disable asynchronous rendering when vsync is on, as this causes a rendering delay (#2822).
|
|
if endFrame && atomic.LoadInt32(&vsyncEnabled) != 0 {
|
|
sync = true
|
|
}
|
|
if !sync {
|
|
for _, c := range q.commands {
|
|
if c.NeedsSync() {
|
|
sync = true
|
|
break
|
|
}
|
|
}
|
|
}
|
|
|
|
logger := debug.SwitchLogger()
|
|
|
|
var flushErr error
|
|
runOnRenderThread(func() {
|
|
defer logger.Flush()
|
|
|
|
if err := q.flush(graphicsDriver, endFrame, logger); err != nil {
|
|
if sync {
|
|
flushErr = err
|
|
return
|
|
}
|
|
q.err.Store(err)
|
|
return
|
|
}
|
|
|
|
theCommandQueueManager.putCommandQueue(q)
|
|
}, sync)
|
|
|
|
if sync && flushErr != nil {
|
|
return flushErr
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
// flush must be called the render thread.
|
|
func (q *commandQueue) flush(graphicsDriver graphicsdriver.Graphics, endFrame bool, logger debug.Logger) (err error) {
|
|
// If endFrame is true, Begin/End should be called to ensure the framebuffer is swapped.
|
|
if len(q.commands) == 0 && !endFrame {
|
|
return nil
|
|
}
|
|
|
|
es := q.indices
|
|
vs := q.vertices
|
|
logger.Logf("Graphics commands:\n")
|
|
|
|
if err := graphicsDriver.Begin(); err != nil {
|
|
return err
|
|
}
|
|
|
|
defer func() {
|
|
// Call End even if an error causes, or the graphics driver's state might be stale (#2388).
|
|
if err1 := graphicsDriver.End(endFrame); err1 != nil && err == nil {
|
|
err = err1
|
|
}
|
|
|
|
// Release the commands explicitly (#1803).
|
|
// Apparently, the part of a slice between len and cap-1 still holds references.
|
|
// Then, resetting the length by [:0] doesn't release the references.
|
|
for i, c := range q.commands {
|
|
if c, ok := c.(*drawTrianglesCommand); ok {
|
|
q.drawTrianglesCommandPool.put(c)
|
|
}
|
|
q.commands[i] = nil
|
|
}
|
|
q.commands = q.commands[:0]
|
|
q.vertices = q.vertices[:0]
|
|
q.indices = q.indices[:0]
|
|
q.tmpNumVertexFloats = 0
|
|
|
|
if endFrame {
|
|
q.uint32sBuffer.reset()
|
|
for i, f := range q.finalizers {
|
|
f()
|
|
q.finalizers[i] = nil
|
|
}
|
|
q.finalizers = q.finalizers[:0]
|
|
}
|
|
}()
|
|
|
|
cs := q.commands
|
|
for len(cs) > 0 {
|
|
nv := 0
|
|
ne := 0
|
|
nc := 0
|
|
for _, c := range cs {
|
|
if dtc, ok := c.(*drawTrianglesCommand); ok {
|
|
if nc > 0 && mustUseDifferentVertexBuffer(nv+dtc.numVertices()) {
|
|
break
|
|
}
|
|
nv += dtc.numVertices()
|
|
ne += dtc.numIndices()
|
|
}
|
|
nc++
|
|
}
|
|
if 0 < ne {
|
|
if err := graphicsDriver.SetVertices(vs[:nv], es[:ne]); err != nil {
|
|
return err
|
|
}
|
|
es = es[ne:]
|
|
vs = vs[nv:]
|
|
}
|
|
indexOffset := 0
|
|
for _, c := range cs[:nc] {
|
|
if err := c.Exec(q, graphicsDriver, indexOffset); err != nil {
|
|
return err
|
|
}
|
|
logger.Logf(" %s\n", c)
|
|
// TODO: indexOffset should be reset if the command type is different
|
|
// from the previous one. This fix is needed when another drawing command is
|
|
// introduced than drawTrianglesCommand.
|
|
if dtc, ok := c.(*drawTrianglesCommand); ok {
|
|
indexOffset += dtc.numIndices()
|
|
}
|
|
}
|
|
cs = cs[nc:]
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
type rectangleF32 struct {
|
|
x float32
|
|
y float32
|
|
width float32
|
|
height float32
|
|
}
|
|
|
|
func imageRectangleToRectangleF32(r image.Rectangle) rectangleF32 {
|
|
return rectangleF32{
|
|
x: float32(r.Min.X),
|
|
y: float32(r.Min.Y),
|
|
width: float32(r.Dx()),
|
|
height: float32(r.Dy()),
|
|
}
|
|
}
|
|
|
|
func (q *commandQueue) prependPreservedUniforms(uniforms []uint32, shader *Shader, dst *Image, srcs [graphics.ShaderImageCount]*Image, dstRegion image.Rectangle, srcRegions [graphics.ShaderImageCount]image.Rectangle) []uint32 {
|
|
origUniforms := uniforms
|
|
uniforms = q.uint32sBuffer.alloc(len(origUniforms) + graphics.PreservedUniformUint32Count)
|
|
copy(uniforms[graphics.PreservedUniformUint32Count:], origUniforms)
|
|
|
|
// Set the destination texture size.
|
|
dw, dh := dst.InternalSize()
|
|
uniforms[0] = math.Float32bits(float32(dw))
|
|
uniforms[1] = math.Float32bits(float32(dh))
|
|
uniformIndex := 2
|
|
|
|
for i := 0; i < graphics.ShaderImageCount; i++ {
|
|
var floatW, floatH uint32
|
|
if srcs[i] != nil {
|
|
w, h := srcs[i].InternalSize()
|
|
floatW = math.Float32bits(float32(w))
|
|
floatH = math.Float32bits(float32(h))
|
|
}
|
|
|
|
uniforms[uniformIndex+i*2] = floatW
|
|
uniforms[uniformIndex+1+i*2] = floatH
|
|
}
|
|
uniformIndex += graphics.ShaderImageCount * 2
|
|
|
|
dr := imageRectangleToRectangleF32(dstRegion)
|
|
if shader.unit() == shaderir.Texels {
|
|
dr.x /= float32(dw)
|
|
dr.y /= float32(dh)
|
|
dr.width /= float32(dw)
|
|
dr.height /= float32(dh)
|
|
}
|
|
|
|
// Set the destination region origin.
|
|
uniforms[uniformIndex] = math.Float32bits(dr.x)
|
|
uniforms[uniformIndex+1] = math.Float32bits(dr.y)
|
|
uniformIndex += 2
|
|
|
|
// Set the destination region size.
|
|
uniforms[uniformIndex] = math.Float32bits(dr.width)
|
|
uniforms[uniformIndex+1] = math.Float32bits(dr.height)
|
|
uniformIndex += 2
|
|
|
|
var srs [graphics.ShaderImageCount]rectangleF32
|
|
for i, r := range srcRegions {
|
|
srs[i] = imageRectangleToRectangleF32(r)
|
|
}
|
|
if shader.unit() == shaderir.Texels {
|
|
for i, src := range srcs {
|
|
if src == nil {
|
|
continue
|
|
}
|
|
w, h := src.InternalSize()
|
|
srs[i].x /= float32(w)
|
|
srs[i].y /= float32(h)
|
|
srs[i].width /= float32(w)
|
|
srs[i].height /= float32(h)
|
|
}
|
|
}
|
|
|
|
// Set the source region origins.
|
|
for i := 0; i < graphics.ShaderImageCount; i++ {
|
|
uniforms[uniformIndex+i*2] = math.Float32bits(srs[i].x)
|
|
uniforms[uniformIndex+1+i*2] = math.Float32bits(srs[i].y)
|
|
}
|
|
uniformIndex += graphics.ShaderImageCount * 2
|
|
|
|
// Set the source region sizes.
|
|
for i := 0; i < graphics.ShaderImageCount; i++ {
|
|
uniforms[uniformIndex+i*2] = math.Float32bits(srs[i].width)
|
|
uniforms[uniformIndex+1+i*2] = math.Float32bits(srs[i].height)
|
|
}
|
|
uniformIndex += graphics.ShaderImageCount * 2
|
|
|
|
// Set the projection matrix.
|
|
uniforms[uniformIndex] = math.Float32bits(2 / float32(dw))
|
|
uniforms[uniformIndex+1] = 0
|
|
uniforms[uniformIndex+2] = 0
|
|
uniforms[uniformIndex+3] = 0
|
|
uniforms[uniformIndex+4] = 0
|
|
uniforms[uniformIndex+5] = math.Float32bits(2 / float32(dh))
|
|
uniforms[uniformIndex+6] = 0
|
|
uniforms[uniformIndex+7] = 0
|
|
uniforms[uniformIndex+8] = 0
|
|
uniforms[uniformIndex+9] = 0
|
|
uniforms[uniformIndex+10] = math.Float32bits(1)
|
|
uniforms[uniformIndex+11] = 0
|
|
uniforms[uniformIndex+12] = math.Float32bits(-1)
|
|
uniforms[uniformIndex+13] = math.Float32bits(-1)
|
|
uniforms[uniformIndex+14] = 0
|
|
uniforms[uniformIndex+15] = math.Float32bits(1)
|
|
|
|
return uniforms
|
|
}
|
|
|
|
type commandQueuePool struct {
|
|
cache []*commandQueue
|
|
m sync.Mutex
|
|
}
|
|
|
|
func (c *commandQueuePool) get() (*commandQueue, error) {
|
|
c.m.Lock()
|
|
defer c.m.Unlock()
|
|
|
|
if len(c.cache) == 0 {
|
|
return &commandQueue{}, nil
|
|
}
|
|
|
|
for _, q := range c.cache {
|
|
if err := q.err.Load(); err != nil {
|
|
return nil, err.(error)
|
|
}
|
|
}
|
|
|
|
q := c.cache[len(c.cache)-1]
|
|
c.cache[len(c.cache)-1] = nil
|
|
c.cache = c.cache[:len(c.cache)-1]
|
|
return q, nil
|
|
}
|
|
|
|
func (c *commandQueuePool) put(queue *commandQueue) {
|
|
c.m.Lock()
|
|
defer c.m.Unlock()
|
|
|
|
c.cache = append(c.cache, queue)
|
|
}
|
|
|
|
type commandQueueManager struct {
|
|
pool commandQueuePool
|
|
current *commandQueue
|
|
}
|
|
|
|
var theCommandQueueManager commandQueueManager
|
|
|
|
func (c *commandQueueManager) enqueueCommand(command command) {
|
|
if c.current == nil {
|
|
c.current, _ = c.pool.get()
|
|
}
|
|
c.current.Enqueue(command)
|
|
}
|
|
|
|
// put can be called from any goroutines.
|
|
func (c *commandQueueManager) putCommandQueue(commandQueue *commandQueue) {
|
|
c.pool.put(commandQueue)
|
|
}
|
|
|
|
func (c *commandQueueManager) enqueueDrawTrianglesCommand(dst *Image, srcs [graphics.ShaderImageCount]*Image, vertices []float32, indices []uint32, blend graphicsdriver.Blend, dstRegion image.Rectangle, srcRegions [graphics.ShaderImageCount]image.Rectangle, shader *Shader, uniforms []uint32, fillRule graphicsdriver.FillRule) {
|
|
if c.current == nil {
|
|
c.current, _ = c.pool.get()
|
|
}
|
|
c.current.EnqueueDrawTrianglesCommand(dst, srcs, vertices, indices, blend, dstRegion, srcRegions, shader, uniforms, fillRule)
|
|
}
|
|
|
|
func (c *commandQueueManager) flush(graphicsDriver graphicsdriver.Graphics, endFrame bool) error {
|
|
// Switch the command queue.
|
|
prev := c.current
|
|
q, err := c.pool.get()
|
|
if err != nil {
|
|
return err
|
|
}
|
|
c.current = q
|
|
|
|
if prev == nil {
|
|
return nil
|
|
}
|
|
if err := prev.Flush(graphicsDriver, endFrame); err != nil {
|
|
return err
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// uint32sBuffer is a reusable buffer to allocate []uint32.
|
|
type uint32sBuffer struct {
|
|
buf []uint32
|
|
}
|
|
|
|
func roundUpPower2(x int) int {
|
|
p2 := 1
|
|
for p2 < x {
|
|
p2 *= 2
|
|
}
|
|
return p2
|
|
}
|
|
|
|
func max(a, b int) int {
|
|
if a < b {
|
|
return b
|
|
}
|
|
return a
|
|
}
|
|
|
|
func (b *uint32sBuffer) alloc(n int) []uint32 {
|
|
buf := b.buf
|
|
if len(buf)+n > cap(buf) {
|
|
buf = make([]uint32, 0, max(roundUpPower2(len(buf)+n), 16))
|
|
}
|
|
s := buf[len(buf) : len(buf)+n]
|
|
b.buf = buf[:len(buf)+n]
|
|
return s
|
|
}
|
|
|
|
func (b *uint32sBuffer) reset() {
|
|
b.buf = b.buf[:0]
|
|
}
|