惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

M
MIT News - Artificial intelligence
罗磊的独立博客
Hugging Face - Blog
Hugging Face - Blog
Apple Machine Learning Research
Apple Machine Learning Research
Last Week in AI
Last Week in AI
S
SegmentFault 最新的问题
让小产品的独立变现更简单 - ezindie.com
让小产品的独立变现更简单 - ezindie.com
美团技术团队
人人都是产品经理
人人都是产品经理
WordPress大学
WordPress大学
The Cloudflare Blog
IT之家
IT之家
雷峰网
雷峰网
小众软件
小众软件
博客园 - 叶小钗
博客园 - 聂微东
爱范儿
爱范儿
博客园 - 司徒正美
博客园 - 三生石上(FineUI控件)
V
Visual Studio Blog
博客园 - 【当耐特】
V
V2EX
博客园_首页
T
Tailwind CSS Blog

DEV Community

Authentication Security Deep Dive: From Brute Force to Salted Hashing (With Java Examples) Why AI Systems Don’t Fail — They Drift Spilling beans for how i learn for exam😁"Reinforcement Learning Cheat Sheet" I Replaced Chrome with Safari for AI Browser Automation. Here's What Broke (and What Finally Worked) How Python Borrows Other People's Work The $40 Architecture: Processing 1 Billion API Requests with 99.99% Uptime Vibe Coding: A Workflow Guide (From Zero to SaaS) Most webhook security guides protect the wrong side. The scary part is delivery. Headless CMS for TanStack Start: Build a Blog with Cosmic EU Age Verification App "Hacked in 2 Minutes" — What Actually Happened Comfy Cloud’s delete function does not actually remove files Running AI Models on GPU Cloud Servers: A Beginner Guide Event-driven media intelligence with AWS Step Functions and Bedrock I scored 500 AI prompts across 8 quality dimensions — here's what broke How to Call Google Gemini API from Next.js (Free Tier, No Backend Needed) The Portal Protocol: Reclaiming Human Connection in the Age of AI How to Fix Your Team's Scattered Knowledge Problem With a Self-Hosted Forum Intro to tc Cloud Functors: A Graph-First Mental Model for the Modern Cloud Designing Multi-Tenant Backends With Both Ownership and Team Access I Built a Neumorphic CSS Library with 77+ Components — Here's What I Learned PostgreSQL Performance Optimization: Why Connection Pooling Is Critical at Scale Cómo construí un SaaS multi-rubro para gestionar expensas en Argentina con FastAPI + Vue 3 🚀 I Built an Ethical Hacking Scanner Tool – Open Source Project I Replaced /usage and /context in Claude Code With a Single Statusline A Pythonic Way to Handle Emails (IMAP/SMTP) with Auto-Discovery and AI-Ready Design I Collected 8.9 Million Polymarket Price Points — Here's What I Found About How Markets Really Move EcoTrack AI — Carbon Footprint Tracker & Dashboard Everyone's Using AI. No One Agrees How. 5 self-hosted ebook managers worth trying in 2026 Building Your First AI Agent with LangChain: From Chatbot to Autonomous Assistant
3D Graphics Without Writing Shaders: The Portable GPU API
Shai Almog · 2026-06-20 · via DEV Community

3D Graphics Without Writing Shaders: The Portable GPU API

Cross-platform 3D is one of those problems that looks impossible when you list the constraints. iOS wants Metal and Metal Shading Language. Android wants OpenGL ES and GLSL. The web wants WebGL. Windows wants Direct3D and HLSL. Every one of these has its own shader language, its own pipeline model, and its own buffer semantics. Most portable engines solve this by making you write shaders multiple times, or by adopting a giant dependency that becomes your whole application.

What is Codename One? Codename One is an open-source framework for building native iOS, Android, desktop, and web apps from a single Java or Kotlin codebase. Learn more at codenameone.com.

PR #5151 takes a different path. The new com.codename1.gpu package is a portable 3D API where applications never write shader source at all. You describe a Material: its lighting model, color, texture, shininess. Per-platform generators emit the actual shader, GLSL ES on Android and WebGL, Metal Shading Language on iOS and Mac, HLSL on the new native Windows port. One Java code base, five GPU backends.

A cube in one screen of code

The API centers on a Renderer you implement and a RenderView that hosts it. Here is a complete Phong-lit cube:

RenderView view = new RenderView(new Renderer() {
    private final Camera camera = new Camera();
    private Mesh cube;
    private Material material;

    public void onInit(GraphicsDevice device) {
        cube = Primitives.cube(device, 1.6f);
        material = new Material(Material.Type.PHONG)
                .setColor(0xff3366ff)
                .setShininess(24f);
        camera.setPerspective(45f, 0.1f, 100f)
                .setPosition(2.6f, 2.1f, 3.4f)
                .setTarget(0f, 0f, 0f);
        device.setLight(new Light().setDirection(-0.4f, -1f, -0.55f));
    }

    public void onResize(GraphicsDevice device, int width, int height) {
        camera.setAspect((float) width / Math.max(1, height));
        device.setViewport(0, 0, width, height);
    }

    public void onFrame(GraphicsDevice device) {
        device.clear(0xff101018, true, true);
        device.setCamera(camera);
        float[] model = Matrix4.rotation((float) Math.toRadians(25), 0.35f, 1f, 0.12f);
        device.draw(cube, material, model);
    }

    public void onDispose(GraphicsDevice device) {
    }
});
form.add(BorderLayout.CENTER, view);

RenderView is a regular Codename One component. It participates in layout, sits next to buttons and labels, and the surrounding form keeps working exactly as before. This is a textured variant of the same code running on an iPhone through Metal:

A textured cube rendered through Metal on iOS

Real models, not just primitives

Primitives gives you cubes, spheres, and friends for getting started, but real content comes from 3D tools. The GltfLoader reads binary glTF (.glb), the de-facto standard interchange format that Blender and practically every modern 3D tool export:

GltfLoader.GltfModel model = GltfLoader.loadModel(device,
        Display.getInstance().getResourceAsStream(getClass(), "/boombox.glb"));
Mesh mesh = model.getMesh();
Material material = new Material(Material.Type.PHONG)
        .setTexture(model.getBaseColorTexture());

The model ships as a regular project resource, so the same asset loads on every platform. Here is the Khronos BoomBox sample (about 6,000 triangles with its own base-color texture) rendering on the native Mac target:

The Khronos BoomBox glTF model rendered on the native Mac target

What runs where

Platform Backend
iOS / Mac native Metal (CAMetalLayer), runtime MSL compilation, pipeline cache
Android OpenGL ES 2 via a GLSurfaceView peer
Web (JavaScript port) WebGL on a canvas peer, reusing the core GLSL generator
Windows native Direct3D 11, HLSL generated and compiled at runtime
Simulator (JavaSE) OpenGL via JOGL by default, with a pure-Java software rasterizer as the fallback

The simulator's fallback deserves a note: it is a complete depth-buffered, perspective-correct, textured, lit rasterizer written in plain Java. When OpenGL isn't available, headless CI machines for instance, rendering keeps working and stays deterministic, which is how our screenshot test suite gates 3D output on every platform.

On iOS there is one more trick. Vertex and index buffers are backed by SIMD-aligned arrays, so the mesh data sits at a fixed, aligned C address that is handed to Metal directly with no intermediate copy. Java arrays in, GPU buffers out, nothing in between.

Two levels of API

The hybrid design goes deeper than materials. Most applications stay on the high level: Mesh, Material, Camera, Light, and Primitives, with Matrix4 providing the usual transform helpers (perspective, look-at, rotation, scaling, multiplication) as plain float[16] arrays with no object churn per frame.

Underneath sits a command layer for people who know exactly what they want. VertexBuffer and IndexBuffer hold your geometry, VertexFormat describes its layout through typed VertexAttribute usages (position, normal, texture coordinates, color), and Texture exposes wrap and filter modes. RenderState controls depth testing, culling, and blend modes per draw call. Everything funnels through GraphicsDevice, which is the one object a Renderer ever talks to, and GpuCapabilities reports what the device underneath can actually do, maximum texture size for instance, so you can scale content to the hardware.

Two details matter for performance. Pipelines are cached by descriptor, so a thousand draws with the same material compile one shader, not a thousand. And rendering is on-demand by default: a static scene renders when something changes (requestRender()), while setContinuous(true) switches to a steady frame loop for animation. On-demand is the difference between a 3D product view that sips battery and one that drains it.

Designed to degrade

Not every environment has a GPU backend. RenderView.isSupported() tells you whether real 3D is available, and where it isn't the view renders a placeholder rather than failing. The seam into the platform layer mirrors how BrowserComponent works internally, so ports that don't implement 3D simply report it as unsupported and everything else keeps running.

This is a foundation

A portable 3D API is useful on its own, for product viewers, data visualization, and the occasional spinning logo. But the reason it exists is tomorrow's post: a game development API that builds sprites, scenes, physics, and sound on top of this layer.

If you render something and it doesn't look right on one of the backends, please file an issue with the model or code attached. Five GPU backends mean five ways for an edge case to hide, and real reports are how the generators improve.

The new 3D Graphics and Shaders chapter in the developer guide covers both API levels in depth, including the shader generation model and per-platform notes. Friday's release post has the full index of this week's posts, and builds games on top of this API.