\u8bf4\u660e：\u6587\u4e2d\u7684 FPS\u3001\u8017\u65f6\u4e0e\u8bbe\u5907\u8868\u73b0\u7528\u4e8e\u8bf4\u660e\u6392\u67e5\u65b9\u6cd5\u3002\u6700\u7ec8\u7ed3\u8bba\u5e94\u4ee5\u76ee\u6807\u8bbe\u5907\u548c\u771f\u5b9e\u4ea4\u4e92\u7684\u6d4f\u89c8\u5668 Profile \u4e3a\u51c6\u3002<\/p>\n
\u5468\u4e00\u5f00\u5468\u4f1a，\u6709\u524d\u7aef\u5f00\u53d1\u8005\u5c55\u793a\u4e86\u4e00\u5957\u201c\u975e\u5e38\u9ad8\u7ea7\u201d\u7684\u4ea4\u4e92：\u754c\u9762\u6839\u636e AI \u5b9e\u65f6\u9884\u6d4b\u7684\u7528\u6237\u884c\u4e3a，\u52a8\u6001\u8c03\u6574\u5143\u7d20\u5e03\u5c40，\u540c\u65f6\u914d\u5408\u6781\u5ea6\u4e1d\u6ed1\u7684 3D \u6d6e\u52a8 CSS \u52a8\u753b\u3002\u6f14\u793a\u6548\u679c\u7edd\u4f73，\u53ef\u4e00\u4e0a\u6d4b\u8bd5\u673a，\u9875\u9762\u6ed1\u52a8\u76f4\u63a5\u6389\u5230 15 \u5e27，CPU \u5360\u7528\u7387\u98d9\u5230\u4e86 100%，\u624b\u673a\u6e29\u5ea6\u76f4\u7ebf\u4e0a\u5347\u3002<\/p>\n
\u628a AI \u9884\u6d4b\u5efa\u6a21（\u5982\u5b9e\u65f6\u9884\u6d4b\u7528\u6237\u4e0b\u4e00\u6b65\u64cd\u4f5c\u3001\u8f93\u5165\u8865\u5168\u3001\u5f02\u5e38\u63d0\u793a）\u4e0e\u73b0\u4ee3 CSS \u52a8\u753b\u7ed3\u5408\u65f6，\u5f88\u591a\u6280\u672f\u65b9\u6848\u542c\u8d77\u6765\u65e0\u61c8\u53ef\u51fb，\u5b9e\u9645\u843d\u5230 React \u6e32\u67d3\u673a\u5236\u548c\u6d4f\u89c8\u5668\u7684\u5408\u6210\u5c42（Compositor Thread）\u4e0a，\u5168\u662f\u5751\u3002<\/p>\n
AI \u9884\u6d4b\u4e0e CSS \u52a8\u753b\u653e\u5728\u540c\u4e00\u4ea4\u4e92\u94fe\u8def\u65f6，\u6700\u5e38\u89c1\u7684\u6027\u80fd\u95ee\u9898\u96c6\u4e2d\u5728\u4e09\u5904\u3002<\/p>\n
\u4e3a\u4e86\u5728 UI \u4e0a\u5b9e\u65f6\u5c55\u793a AI \u5bf9\u7528\u6237\u884c\u4e3a\u7684\u9884\u6d4b\u6982\u7387，\u6709\u4eba\u559c\u6b22\u5728\u6700\u5916\u5c42\u7684 React Context \u6216\u5168\u5c40 Zustand \u91cc\u76f4\u63a5\u7ed1\u5b9a\u9ad8\u9891\u66f4\u65b0\u7684 State\u3002<\/p>\n
AI \u9884\u6d4b\u5f15\u64ce\u6bcf 50ms \u5410\u51fa\u4e00\u7ec4\u65b0\u7684\u6743\u91cd\u5411\u91cf，React \u6839\u7ec4\u4ef6\u5c31\u91cd\u65b0\u6e32\u67d3\u4e00\u6b21\u3002\u7ed3\u679c\u5c31\u662f\u6574\u68f5 DOM \u6811\u88ab\u75af\u72c2\u6253\u788e\u91cd\u5efa，\u539f\u672c\u5e73\u6ed1\u7684 CSS Transition \u52a8\u753b\u56e0\u4e3a\u4e3b\u7ebf\u7a0b\u7e41\u5fd9，\u88ab\u6253\u65ad\u5f97\u65ad\u65ad\u7eed\u7eed\u3002<\/p>\n
\u4e3a\u4e86\u89e3\u51b3\u52a8\u753b\u6389\u5e27，\u5f88\u591a\u4eba\u7b2c\u4e00\u53cd\u5e94\u662f\u201c\u52a0 GPU \u786c\u4ef6\u52a0\u901f\u201d\u3002\u4ed6\u4eec\u5728\u5168\u5c40 CSS \u91cc\u7ed9\u6240\u6709\u9884\u6d4b\u5361\u7247\u52a0\u4e0a：<\/p>\n
\/* \u770b\u8d77\u6765\u806a\u660e\u7684\u5199\u6cd5，\u5b9e\u9645\u4e0a\u662f\u663e\u5b58\u6740\u624b *\/
\n.ai-predict-card {
\n will-change: transform, opacity, width, height;
\n transform: translateZ(0);
\n}<\/p>\n
\u6d4f\u89c8\u5668\u4e3a\u4e86\u54cd\u5e94 will-change，\u4f1a\u5f3a\u5236\u4e3a\u6bcf\u4e00\u4e2a\u5361\u7247\u5206\u914d\u72ec\u7acb\u7684 Graphics Layer（\u5408\u6210\u5c42）\u3002\u5f53\u9875\u9762\u4e0a\u6709 20 \u4e2a\u5361\u7247\u65f6，\u663e\u5b58\u5360\u7528\u77ac\u95f4\u66b4\u6da8\u6570\u767e\u5146，\u4f4e\u7aef\u673a\u76f4\u63a5\u56e0\u4e3a Out of Memory（OOM）\u5d29\u6e83\u95ea\u9000\u3002<\/p>\n
\u5728\u524d\u7aef\u96c6\u6210\u4e86\u8f7b\u91cf\u7ea7\u7684 Transformer\/ONNX \u6a21\u578b\u8fdb\u884c\u8f93\u5165\u8865\u5168\u6216\u5e03\u5c40\u9884\u6d4b\u65f6，\u628a\u77e9\u9635\u8ba1\u7b97\u3001\u5411\u91cf\u5f52\u4e00\u5316\u7b49 CPU \u5bc6\u96c6\u578b\u903b\u8f91\u76f4\u63a5\u5199\u5728 React useEffect \u6216\u4e8b\u4ef6\u56de\u8c03\u91cc\u3002<\/p>\n
\u77e9\u9635\u8ba1\u7b97\u5360\u7528\u4e3b\u7ebf\u7a0b 200ms，\u6d4f\u89c8\u5668\u5728\u6b64\u671f\u95f4\u65e0\u6cd5\u5904\u7406\u4efb\u4f55\u7ed8\u5236\u8bf7\u6c42（Paint），\u7528\u6237\u7684\u8f93\u5165\u5361\u987f，CSS \u52a8\u753b\u77ac\u95f4\u51bb\u7ed3\u3002<\/p>\n
\u5f53\u9875\u9762\u51fa\u73b0\u672a\u77e5\u5361\u987f\u4e0e\u6389\u5e27\u65f6，\u4e0d\u8981\u778e\u731c\u3002\u6293\u6570\u636e\u662f\u5b9a\u4f4d\u6d4f\u89c8\u5668\u4e3b\u7ebf\u7a0b\u963b\u585e\u4e0e\u663e\u5b58\u6ea2\u51fa\u7684\u552f\u4e00\u6807\u51c6\u3002<\/p>\n
\u9996\u5148，\u5728 Linux\/Mac \u6d4b\u8bd5\u73af\u5883\u4e2d\u4f7f\u7528\u547d\u4ee4\u884c\u89c2\u5bdf\u7cfb\u7edf\u8fdb\u7a0b\u4e0e Node.js SSR\/Puppeteer \u6e32\u67d3\u8fdb\u7a0b\u7684 CPU \u4e0e\u5185\u5b58\u5f00\u9500：<\/p>\n
# \u627e\u5230\u524d\u7aef SSR \u6e32\u67d3\u6216\u672c\u5730\u6d4b\u8bd5\u6d4f\u89c8\u5668\u7684 pid，\u67e5\u770b\u5b50\u7ebf\u7a0b CPU \u6d88\u8017
\ntop -hp $(pgrep -f "chrome|node")<\/p>\n
# \u542f\u52a8 Lighthouse CI \u8fdb\u884c\u6027\u80fd\u95e8\u7981\u68c0\u6d4b，\u6293\u53d6\u52a8\u753b\u5e27\u7387 (FPS) \u4e0e TBT (Total Blocking Time)
\nnpx lighthouserc collect –url=http:\/\/localhost:3000\/predictive-ui<\/p>\n
\u5728 Chrome \u5f00\u53d1\u8005\u5de5\u5177\u91cc，\u901a\u8fc7\u547d\u4ee4\u884c\u542f\u52a8\u5e26\u663e\u5b58\u4e0e\u56fe\u5c42\u76d1\u63a7\u7684\u5206\u6790\u6a21\u5f0f：<\/p>\n
# \u542f\u52a8\u5e26\u6709 GPU \u5185\u5b58\u76d1\u63a7\u6807\u8bb0\u7684 Chrome \u6d4f\u89c8\u5668\u5b9e\u4f8b\u8fdb\u884c\u6027\u80fd\u5256\u6790
\n\/Applications\/Google\\\\ Chrome.app\/Contents\/MacOS\/Google\\\\ Chrome \\\\
\n –enable-gpu-benchmarking \\\\
\n –show-fps-counter \\\\
\n http:\/\/localhost:3000<\/p>\n
\u6253\u5f00 Performance \u9762\u677f\u5f55\u5236 5 \u79d2，\u4f60\u4f1a\u6e05\u695a\u770b\u5230\u4e00\u4e32\u7ea2\u8272\u7684\u201cLong Task\u201d\u7d27\u7d27\u538b\u5728 Main Thread \u4e0a，\u800c Compositor Thread \u53ea\u80fd\u5728\u65c1\u8fb9\u88ab\u52a8\u5e72\u7b49\u3002<\/p>\n
\u53ef\u884c\u7684\u65b9\u5411\u662f\u505a\u8ba1\u7b97\u4e0e\u6e32\u67d3\u9694\u79bb：\u628a\u9002\u5408\u79bb\u7ebf\u6267\u884c\u7684\u9884\u6d4b\u653e\u8fdb Web Worker，\u4f18\u5148\u8ba9 CSS \u52a8\u753b\u8d70\u5408\u6210\u7ebf\u7a0b，React \u53ea\u5904\u7406\u5fc5\u8981\u7684\u72b6\u6001\u4ea4\u63a5\u3002<\/p>\n
\u4e0b\u56fe\u5c55\u793a\u4e86\u91cd\u6784\u540e\u7684\u7ebf\u7a0b\u5206\u5de5\u4e0e\u6e32\u67d3\u7ba1\u9053：<\/p>\n
flowchart LR
\n subgraph Browser Main Thread ["\u6d4f\u89c8\u5668\u4e3b\u7ebf\u7a0b (React)"]
\n A["\u7528\u6237\u4ea4\u4e92\u4e8b\u4ef6 (Input\/Scroll)"] –> B["\u9632\u6296\u6d3e\u53d1\u81f3 Worker"]
\n E["\u63a5\u6536 Worker \u9884\u6d4b\u7ed3\u679c (RAF Batching)"] –> F["\u66f4\u65b0 React \u5c40\u90e8 State"]
\n end<\/p>\n
subgraph Web Worker ["Web Worker (AI \u9884\u6d4b\u7ebf\u7a0b)"]
\n B –> C["\u6267\u884c ONNX\/Tensor \u77e9\u9635\u8ba1\u7b97"]
\n C –> D["\u751f\u6210\u4e0b\u4e00\u6b65 UI \u9884\u6d4b\u5411\u91cf"]
\n D –> E
\n end<\/p>\n
subgraph Compositor Thread ["GPU \u5408\u6210\u7ebf\u7a0b (CSS \u52a8\u753b)"]
\n G["\u53ea\u5904\u7406 transform & opacity"] –> H["\u76f4\u63a5\u4ea4\u7ed9 GPU \u6e32\u67d3\u5c4f (60fps)"]
\n end<\/p>\n
F -. "\u58f0\u660e\u5f0f CSS \u7c7b\u540d\u5207\u6362" .-> G<\/p>\n
\u8fd9\u79cd\u67b6\u6784\u4e0b，\u4e0d\u7ba1 Worker \u91cc\u7684 AI \u9884\u6d4b\u8ba1\u7b97\u82b1\u4e86\u591a\u5c11\u65f6\u95f4，\u4e3b\u7ebf\u7a0b\u7684 CSS \u52a8\u753b\u4f9d\u7136\u80fd\u591f\u51ed\u501f\u786c\u4ef6\u52a0\u901f\u4fdd\u6301 60fps \u6ee1\u5e27\u8fd0\u884c\u3002<\/p>\n
\u4e0b\u9762\u7684\u4ee3\u7801\u5c55\u793a\u4e86\u5982\u4f55\u5728 React \u5168\u6808\u9879\u76ee\u4e2d，\u901a\u8fc7 Web Worker \u9694\u79bb AI \u9884\u6d4b\u8ba1\u7b97，\u5e76\u4f7f\u7528 requestAnimationFrame（RAF）\u4e0e\u4e25\u683c\u63a7\u5236\u7684 CSS \u5408\u6210\u5c42\u5c5e\u6027\u66f4\u65b0 UI\u3002<\/p>\n
\/\/ \u5728\u72ec\u7acb\u7684\u7ebf\u7a0b\u4e2d\u6267\u884c\u8ba1\u7b97，\u7edd\u5bf9\u4e0d\u5360\u4e3b\u7ebf\u7a0b\u65f6\u95f4
\nctx.addEventListener('message', async (event) => {
\n const { inputSequence } = event.data;<\/p>\n
\/\/ \u6a21\u62df\u590d\u6742 AI \u77e9\u9635\u9884\u6d4b\u8ba1\u7b97
\n const startTime = performance.now();
\n let weight = 0;
\n for (let i = 0; i < 1000000; i++) {
\n weight += Math.sin(i) * Math.cos(i);
\n }<\/p>\n
const nextLayoutPrediction = inputSequence.length > 5 ? 'expanded' : 'compact';<\/p>\n
ctx.postMessage({
\n prediction: nextLayoutPrediction,
\n confidence: 0.91,
\n costMs: performance.now() – startTime
\n });
\n});<\/p>\n
export {};<\/p>\n
import { useState, useEffect, useRef, useCallback } from 'react';<\/p>\n
export function usePredictiveAnimation(userInput: string) {
\n const [layoutState, setLayoutState] = useState<'compact' | 'expanded'>('compact');
\n const workerRef = useRef<Worker | null>(null);
\n const rafIdRef = useRef<number | null>(null);<\/p>\n
useEffect(() => {
\n \/\/ \u521d\u59cb\u5316 Worker
\n workerRef.current = new Worker(new URL('.\/ai-predict.worker.ts', import.meta.url));<\/p>\n
workerRef.current.onmessage = (e) => {
\n const { prediction } = e.data;<\/p>\n
\/\/ \u4f7f\u7528 requestAnimationFrame \u6279\u5904\u7406 DOM \u66f4\u65b0，\u907f\u514d\u52a8\u753b\u4ea4\u9519\u5361\u987f
\n if (rafIdRef.current) cancelAnimationFrame(rafIdRef.current);
\n rafIdRef.current = requestAnimationFrame(() => {
\n setLayoutState(prediction);
\n });
\n };<\/p>\n
return () => {
\n workerRef.current?.terminate();
\n if (rafIdRef.current) cancelAnimationFrame(rafIdRef.current);
\n };
\n }, []);<\/p>\n
\/\/ \u952e\u76d8\u8f93\u5165\u9ad8\u9891\u89e6\u53d1\u65f6\u8fdb\u884c\u9632\u6296\u6d3e\u53d1
\n const dispatchInput = useCallback((input: string) => {
\n if (workerRef.current) {
\n workerRef.current.postMessage({ inputSequence: input });
\n }
\n }, []);<\/p>\n
useEffect(() => {
\n dispatchInput(userInput);
\n }, [userInput, dispatchInput]);<\/p>\n
return { layoutState };
\n}<\/p>\n
\/* \u53ea\u6709\u771f\u6b63\u8fd0\u52a8\u7684\u7ec4\u4ef6\u624d\u52a0\u4e0a Composite \u4f18\u5316，\u5e76\u4e14\u4ec5\u4f7f\u7528 transform *\/
\n.cardContainer {
\n transition: transform 0.3s cubic-bezier(0.16, 1, 0.3, 1);
\n \/* \u4e25\u7981\u5bf9 width\/height \u7b49\u4f1a\u89e6\u53d1 Layout \u7684\u5c5e\u6027\u4f7f\u7528 transition *\/
\n}<\/p>\n
.expanded {
\n transform: scale(1.05) translate3d(0, -4px, 0);
\n}<\/p>\n
.compact {
\n transform: scale(1) translate3d(0, 0, 0);
\n}<\/p>\n
\u5199\u4ee3\u7801\u524d，\u62ff\u8fd9\u51e0\u6761\u786c\u89c4\u5219\u5bf9\u7167\u4e00\u4e0b：<\/p>\n
\u6280\u672f\u70ab\u6280\u4e0d\u53ef\u6015，\u53ef\u6015\u7684\u662f\u727a\u7272\u4e86\u6700\u57fa\u7840\u7684\u6d41\u7545\u5ea6\u3002\u7ed9\u4ea4\u4e92\u505a\u51cf\u6cd5，\u8ba9\u8ba1\u7b97\u5f52 Worker，\u8ba9\u52a8\u753b\u5f52 GPU，\u8fd9\u624d\u662f\u73b0\u4ee3\u524d\u7aef\u8be5\u8d70\u7684\u94c1\u8def\u7ebf\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"
React \u52a8\u753b\u63a5 AI \u9884\u6d4b：\u628a\u8ba1\u7b97\u79fb\u51fa\u4e3b\u7ebf\u7a0b，\u522b\u6ee5\u7528 will-change\u8bf4\u660e：\u6587\u4e2d\u7684 FPS\u3001\u8017\u65f6\u4e0e\u8bbe\u5907\u8868\u73b0\u7528\u4e8e\u8bf4\u660e\u6392\u67e5\u65b9\u6cd5\u3002\u6700\u7ec8\u7ed3\u8bba\u5e94\u4ee5\u76ee\u6807\u8bbe\u5907\u548c\u771f\u5b9e\u4ea4\u4e92\u7684\u6d4f\u89c8\u5668 Profile \u4e3a\u51c6\u3002\u5468\u4e00\u5f00\u5468\u4f1a，\u6709\u524d\u7aef\u5f00\u53d1\u8005\u5c55\u793a\u4e86\u4e00\u5957\u201c\u975e\u5e38\u9ad8\u7ea7\u201d\u7684\u4ea4\u4e92：\u754c\u9762\u6839\u636e AI \u5b9e\u65f6\u9884\u6d4b\u7684\u7528\u6237\u884c\u4e3a，\u52a8\u6001\u8c03\u6574\u5143\u7d20\u5e03\u5c40，\u540c\u65f6\u914d\u5408\u6781\u5ea6\u4e1d\u6ed1\u7684 3D \u6d6e\u52a8 CSS \u52a8\u753b\u3002\u6f14\u793a\u6548\u679c\u7edd\u4f73，\u53ef\u4e00<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[10896,50],"topic":[],"class_list":["post-95037","post","type-post","status-publish","format-standard","hentry","category-server","tag-will-change","tag-50"],"yoast_head":"\n