Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

V3.8.5 batcher #17794

Open
wants to merge 2 commits into
base: v3.8.6
Choose a base branch
from
Open

V3.8.5 batcher #17794

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
2bdba9d
revert bufferData
star-e Oct 31, 2024
d345f3a
use double buffer
star-e Oct 31, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
74 changes: 44 additions & 30 deletions cocos/2d/renderer/mesh-buffer.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -44,6 +44,8 @@ enum MeshBufferSharedBufferView {

const IA_POOL_USED_SCALE = 1 / 2;

const NUM_GFX_BUFFERS = 2;

/**
* @en Mesh buffer used for 2d rendering, used internally and not of concern to the user.
* @zh 2d 渲染使用的网格缓冲数据，内部使用，用户不须关心。
Expand Down Expand Up @@ -179,8 +181,8 @@ export class MeshBuffer {
private _attributes: Attribute[] = null!;

// InputAssembler pools for each mesh buffer, array offset correspondent
private _iaPool: IIARef[] = [];
private _iaInfo: InputAssemblerInfo = null!;
private _iaPool: IIARef[][] = [];
private _iaPoolIdx = 0;
private _nextFreeIAHandle = 0;

//nativeObj
Expand Down Expand Up @@ -258,7 +260,9 @@ export class MeshBuffer {
this.iData = new Uint16Array(this._initIDataCount);
}
// Initialize the first ia
this._iaPool.push(this.createNewIA(device));
for (let i = 0; i < NUM_GFX_BUFFERS; ++i) {
this._iaPool.push([this.createNewIA(device)]);
}
if (JSB) {
this._nativeObj.initialize(attrs);
}
Expand All @@ -277,20 +281,23 @@ export class MeshBuffer {
public destroy (): void {
this.reset();
this._attributes = null!;
this._iaInfo = null!;
this.vData = null!;
this.iData = null!;

// Destroy InputAssemblers
for (let i = 0; i < this._iaPool.length; ++i) {
const iaRef = this._iaPool[i];
if (iaRef.vertexBuffers[0]) {
iaRef.vertexBuffers[0].destroy();
}
if (iaRef.indexBuffer) {
iaRef.indexBuffer.destroy();
for (let k = 0; k < NUM_GFX_BUFFERS; ++k) {
const iaPool = this._iaPool[k];
for (let i = 0; i < iaPool.length; ++i) {
const iaRef = iaPool[i];
if (iaRef.vertexBuffers[0]) {
iaRef.vertexBuffers[0].destroy();
}
if (iaRef.indexBuffer) {
iaRef.indexBuffer.destroy();
}
iaRef.ia.destroy();
}
iaRef.ia.destroy();
iaPool.length = 0;
}
this._iaPool.length = 0;
}
Expand Down Expand Up @@ -319,10 +326,11 @@ export class MeshBuffer {
* @deprecated since v3.7.0, this is an engine private interface that will be removed in the future.
*/
public requireFreeIA (device: Device): InputAssembler {
if (this._iaPool.length <= this._nextFreeIAHandle) {
this._iaPool.push(this.createNewIA(device));
const iaPool = this._iaPool[this._iaPoolIdx];
if (iaPool.length <= this._nextFreeIAHandle) {
iaPool.push(this.createNewIA(device));
}
const ia = this._iaPool[this._nextFreeIAHandle++].ia;
const ia = iaPool[this._nextFreeIAHandle++].ia;
return ia;
}

Expand All @@ -332,7 +340,7 @@ export class MeshBuffer {
* @deprecated since v3.7.0, this is an engine private interface that will be removed in the future.
*/
public recycleIA (ia: InputAssembler): void {
const pool = this._iaPool;
const pool = this._iaPool[this._iaPoolIdx];
for (let i = 0; i < this._nextFreeIAHandle; ++i) {
if (ia === pool[i].ia) {
// Swap to recycle the ia
Expand Down Expand Up @@ -368,15 +376,18 @@ export class MeshBuffer {
return;
}

const iaPool = this._iaPool[this._iaPoolIdx];

// On iOS14, different IAs can not share same GPU buffer, so must submit the same date to different buffers
const iOS14 = sys.__isWebIOS14OrIPadOS14Env;
const submitCount = iOS14 ? this._nextFreeIAHandle : 1;
if (iOS14 && (submitCount / this._iaPool.length < IA_POOL_USED_SCALE)) {
// const iOS14 = sys.__isWebIOS14OrIPadOS14Env;
// const submitCount = iOS14 ? this._nextFreeIAHandle : 1;
const submitCount = this._nextFreeIAHandle;
if (submitCount / iaPool.length < IA_POOL_USED_SCALE) {
const count = submitCount / IA_POOL_USED_SCALE;
const length = this._iaPool.length;
const length = iaPool.length;
// Destroy InputAssemblers
for (let i = length - 1; i >= count; i--) {
const iaRef = this._iaPool[i];
const iaRef = iaPool[i];
if (iaRef.vertexBuffers[0]) {
iaRef.vertexBuffers[0].destroy();
}
Expand All @@ -385,12 +396,12 @@ export class MeshBuffer {
}
iaRef.ia.destroy();
}
this._iaPool.length = count;
iaPool.length = count;
}
const byteCount = this.byteOffset;
const indexCount = this.indexOffset;
for (let i = 0; i < submitCount; ++i) {
const iaRef = this._iaPool[i];
const iaRef = iaPool[i];

const verticesData = new Float32Array(this.vData.buffer, 0, byteCount >> 2);
const indicesData = new Uint16Array(this.iData.buffer, 0, indexCount);
Expand All @@ -406,6 +417,7 @@ export class MeshBuffer {
}
iaRef.indexBuffer.update(indicesData);
}
this._iaPoolIdx = (this._iaPoolIdx + 1) % NUM_GFX_BUFFERS;
this.dirty = false;
}

Expand All @@ -415,7 +427,8 @@ export class MeshBuffer {
let indexBuffer: Buffer;
// HACK: After sharing buffer between drawcalls, the performance degradation a lots on iOS 14 or iPad OS 14 device
// TODO: Maybe it can be removed after Apple fixes it?
if (sys.__isWebIOS14OrIPadOS14Env || !this._iaPool[0]) {
// if (sys.__isWebIOS14OrIPadOS14Env || !this._iaInfo)
{
const vbStride = this._vertexFormatBytes = this._floatsPerVertex * Float32Array.BYTES_PER_ELEMENT;
const ibStride = Uint16Array.BYTES_PER_ELEMENT;
const vertexBuffer = device.createBuffer(new BufferInfo(
Expand All @@ -433,13 +446,14 @@ export class MeshBuffer {

vertexBuffers = [vertexBuffer];
// Reuse purpose for new IAs
this._iaInfo = new InputAssemblerInfo(this._attributes, vertexBuffers, indexBuffer);
ia = device.createInputAssembler(this._iaInfo);
} else {
ia = device.createInputAssembler(this._iaInfo);
vertexBuffers = this._iaInfo.vertexBuffers;
indexBuffer = this._iaInfo.indexBuffer!;
const iaInfo = new InputAssemblerInfo(this._attributes, vertexBuffers, indexBuffer);
ia = device.createInputAssembler(iaInfo);
}
// else {
// ia = device.createInputAssembler(this._iaInfo);
// vertexBuffers = this._iaInfo.vertexBuffers;
// indexBuffer = this._iaInfo.indexBuffer!;
// }
return {
ia,
vertexBuffers,
Expand Down
8 changes: 1 addition & 7 deletions cocos/gfx/webgl/webgl-commands.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -670,13 +670,7 @@
}
}

if (systemInfo.os === OS.IOS && (gpuBuffer.memUsage$ & MemoryUsageBit.HOST) && offset === 0 && size === buff.byteLength) {
// Fix performance issue on iOS.
// TODO(zhouzhenglong): glBufferSubData is faster than glBufferData in most cases.
// We should use multiple buffers to avoid stall (cpu write conflicts with gpu read).
// Before that, we will use glBufferData instead of glBufferSubData.
gl.bufferData(gpuBuffer.glTarget$, buff, gl.DYNAMIC_DRAW);
} else if (size === buff.byteLength) {
if (size === buff.byteLength) {
gl.bufferSubData(gpuBuffer.glTarget$, offset, buff);
} else {
gl.bufferSubData(gpuBuffer.glTarget$, offset, buff.slice(0, size));
Expand Down Expand Up @@ -733,7 +727,7 @@
}
} else {
for (let i = 0; i < gpuTexture.mipLevel$; ++i) {
gl.texImage2D(WebGLConstants.TEXTURE_2D, i, gpuTexture.glInternalFmt$, w, h, 0, gpuTexture.glFormat$, gpuTexture.glType$, null);

Check warning on line 730 in cocos/gfx/webgl/webgl-commands.ts

View workflow job for this annotation

GitHub Actions / Run ESLint 

This line has a length of 152. Maximum allowed is 150
w = max(1, w >> 1);
h = max(1, h >> 1);
}
Expand Down Expand Up @@ -2722,7 +2716,7 @@
gpuTexture.glType$,
pixels,
);
} else if (gpuTexture.glInternalFmt$ !== (WebGLEXT.COMPRESSED_RGB_ETC1_WEBGL as number) && !device.extensions.noCompressedTexSubImage2D$) {

Check warning on line 2719 in cocos/gfx/webgl/webgl-commands.ts

View workflow job for this annotation

GitHub Actions / Run ESLint 

This line has a length of 151. Maximum allowed is 150
gl.compressedTexSubImage2D(
WebGLConstants.TEXTURE_2D,
mipLevel,
Expand Down
26 changes: 4 additions & 22 deletions cocos/gfx/webgl2/webgl2-commands.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -843,13 +843,7 @@ export function WebGL2CmdFuncUpdateBuffer (
cache.glArrayBuffer$ = gpuBuffer.glBuffer$;
}

if (systemInfo.os === OS.IOS && (gpuBuffer.memUsage$ & MemoryUsageBit.HOST) && offset === 0 && size === buff.byteLength) {
// Fix performance issue on iOS.
// TODO(zhouzhenglong): glBufferSubData is faster than glBufferData in most cases.
// We should use multiple buffers to avoid stall (cpu write conflicts with gpu read).
// Before that, we will use glBufferData instead of glBufferSubData.
gl.bufferData(gpuBuffer.glTarget$, buff, gl.DYNAMIC_DRAW);
} else if (size === buff.byteLength) {
if (size === buff.byteLength) {
gl.bufferSubData(gpuBuffer.glTarget$, offset, buff);
} else {
gl.bufferSubData(gpuBuffer.glTarget$, offset, buff.slice(0, size));
Expand All @@ -870,13 +864,7 @@ export function WebGL2CmdFuncUpdateBuffer (
cache.glElementArrayBuffer$ = gpuBuffer.glBuffer$;
}

if (systemInfo.os === OS.IOS && (gpuBuffer.memUsage$ & MemoryUsageBit.HOST) && offset === 0 && size === buff.byteLength) {
// Fix performance issue on iOS.
// TODO(zhouzhenglong): glBufferSubData is faster than glBufferData in most cases.
// We should use multiple buffers to avoid stall (cpu write conflicts with gpu read).
// Before that, we will use glBufferData instead of glBufferSubData.
gl.bufferData(gpuBuffer.glTarget$, buff, gl.DYNAMIC_DRAW);
} else if (size === buff.byteLength) {
if (size === buff.byteLength) {
gl.bufferSubData(gpuBuffer.glTarget$, offset, buff);
} else {
gl.bufferSubData(gpuBuffer.glTarget$, offset, buff.slice(0, size));
Expand All @@ -889,13 +877,7 @@ export function WebGL2CmdFuncUpdateBuffer (
cache.glUniformBuffer$ = gpuBuffer.glBuffer$;
}

if (systemInfo.os === OS.IOS && (gpuBuffer.memUsage$ & MemoryUsageBit.HOST) && offset === 0 && size === buff.byteLength) {
// Fix performance issue on iOS.
// TODO(zhouzhenglong): glBufferSubData is faster than glBufferData in most cases.
// We should use multiple buffers to avoid stall (cpu write conflicts with gpu read).
// Before that, we will use glBufferData instead of glBufferSubData.
gl.bufferData(gpuBuffer.glTarget$, buff, gl.DYNAMIC_DRAW);
} else if (size === buff.byteLength) {
if (size === buff.byteLength) {
gl.bufferSubData(gpuBuffer.glTarget$, offset, buff);
} else {
gl.bufferSubData(gpuBuffer.glTarget$, offset, new Float32Array(buff, 0, size / 4));
Expand Down Expand Up @@ -1101,7 +1083,7 @@ export function WebGL2CmdFuncDestroyTexture (device: WebGL2Device, gpuTexture: I
}

if (gpuTexture.glRenderbuffer$) {
let glRenderbuffer = cache.glRenderbuffer$;
const glRenderbuffer = cache.glRenderbuffer$;
gl.deleteRenderbuffer(gpuTexture.glRenderbuffer$);
if (glRenderbuffer === gpuTexture.glRenderbuffer$) {
gl.bindRenderbuffer(WebGLConstants.RENDERBUFFER, null);
Expand Down
Loading