beamterm-renderer

beamterm - A WebGL2 Terminal Renderer

A high-performance terminal rendering system for web browsers, targeting sub-millisecond render times. beamterm is a terminal renderer, not a full terminal emulator - it handles the display layer while you provide the terminal logic.

🚀 Live Demos

Check out interactive examples showcasing both pure Rust applications and JavaScript/TypeScript integrations.

Key Features

• Single Draw Call - Renders entire terminal (e.g., 200×80 cells) in one instanced draw

• Zero-Copy Updates - Direct memory mapping for dynamic cell updates

• Unicode and Emoji Support - Complete Unicode support with grapheme clustering

• ASCII Fast Path - Direct bit operations for ASCII characters (no lookups)

• Selection Support - Mouse-driven text selection with clipboard integration (Block/Linear modes)

• Optional JS/TS Bindings - Provides a JavaScript/TypeScript API for easy integration

Performance

beamterm targets sub-millisecond render times across a wide range of hardware:

The screenshot shows Ratzilla's "canvas waves" demo running in a 426×106 terminal (45,156 cells), maintaining sub-millisecond render times on 2019-era hardware (i9-9900K / RTX 2070).

† Includes Ratatui buffer translation, GPU buffer uploads, and draw call execution.

System Architecture

The renderer consists of three crates:

beamterm-atlas - Generates GPU-optimized font atlases from TTF/OTF files. Automatically calculates cell dimensions, supports font styles (normal/bold/italic), and outputs packed texture data.

beamterm-data - Provides shared data structures and efficient binary serialization. Features versioned format with header validation and cross-platform encoding.

beamterm-renderer - The WebGL2 rendering engine. Implements instanced rendering with optimized buffer management and state tracking for consistent sub-millisecond performance.

Architecture Overview

The architecture leverages GPU instancing to reuse a single quad geometry across all terminal cells, with per-instance data providing position, character, and color information. All rendering state is encapsulated in a Vertex Array Object (VAO), enabling single-draw-call rendering with minimal CPU overhead. The 2D texture array maximizes cache efficiency by packing related glyphs into horizontal strips within each layer.

Buffer Management Strategy

The renderer employs several optimization strategies:

1. VAO Encapsulation: All vertex state is captured in a single VAO, minimizing state changes
2. Separate Static/Dynamic: Geometry and positions rarely change; only cell content is dynamic
3. Aligned Packing: All structures use explicit alignment for optimal GPU access
4. Batch Updates: Cell updates are batched and uploaded in a single operation
5. Immutable Storage: 2D texture array uses texStorage3D for driver optimization hints

These strategies combined enable the renderer to achieve consistent sub-millisecond frame times even for large terminals (200×80 cells = 16,000 instances).

Total Memory Requirements

For a 200×80 terminal with 2560 glyphs:

Terminal Renderer API

The renderer provides a high-level Terminal struct that encapsulates the complete rendering system:

Quick Start

use beamterm_renderer::{Terminal, CellData, FontStyle, GlyphEffect};

// Create terminal with default font atlas
let mut terminal = Terminal::builder("#canvas").build()?;

// Update cells and render
let cells: Vec<CellData> = ...;
terminal.update_cells(cells.into_iter())?;
terminal.render_frame()?;

// Handle resize
terminal.resize(new_width, new_height)?;

Selection and Mouse Input

The renderer supports mouse-driven text selection with automatic clipboard integration:

// Enable default selection handler
let terminal = Terminal::builder("#canvas")
    .default_mouse_input_handler(SelectionMode::Linear, true)
    .build()?;

// Or implement custom mouse handling
let terminal = Terminal::builder("#canvas")
    .mouse_input_handler(|event, grid| {
        // Custom handler logic
    })
    .build()?;

TerminalGrid

Main rendering component managing the terminal display. Handles shader programs, cell data, GPU buffers, and rendering state.

FontAtlas

Manages the 2D texture array containing all font glyphs. Provides character-to-glyph ID mapping with fast ASCII optimization. Supports loading default or custom font atlases.

Cell Data Structure

Each terminal cell requires:

• symbol: Character or grapheme to display (&str)
• style: FontStyle enum (Normal, Bold, Italic, BoldItalic)
• effect: GlyphEffect enum (None, Underline, Strikethrough)
• fg/bg: Colors as 32-bit ARGB values (0xAARRGGBB)

Font Atlas 2D Texture Array Architecture

The font atlas uses a WebGL 2D texture array where each layer contains a 16×1 grid of glyphs (16 per layer). This provides optimal memory utilization and cache efficiency while maintaining O(1) coordinate lookups through simple bit operations. The system supports 512 base glyphs × 4 styles + emoji.

2D Texture Array Coordinate System

The font atlas uses a 2D texture array organized as multiple layers, each containing a 16×1 grid of glyphs:

The layers are densely packed, but there might be gaps beween font variants and before the first emoji layer, unless all 512 glyphs are used.

Glyph ID Encoding and Mapping

The glyph ID is a 16-bit value that efficiently packs both the base glyph identifier and style information (such as weight, style flags, etc.) into a single value. This packed representation is passed directly to the GPU.

Glyph ID Bit Layout (16-bit)

ID to 2D Array Position Examples

The consistent modular arithmetic ensures that style variants maintain the same horizontal position within their respective layers, improving texture cache coherence.

ASCII Optimization

ASCII characters (0-127) bypass the HashMap lookup entirely through direct bit manipulation. For ASCII input, the glyph ID is computed as char_code | style_bits, providing zero-overhead character mapping. Non-ASCII characters use a HashMap for flexible Unicode support. This approach optimizes for the common case while maintaining full Unicode capability.

GPU Buffer Architecture

The renderer uses six buffers managed through a Vertex Array Object (VAO) to achieve single-draw-call rendering. Each buffer serves a specific purpose in the instanced rendering pipeline, with careful attention to memory alignment and update patterns.

Buffer Layout Summary

All vertex buffers are encapsulated within a single Vertex Array Object (VAO), enabling state-free rendering with a single draw call.

The Instance Position and Instance Cell buffers are recreated when the terminal size changes,

Vertex Attribute Bindings

Instance Data Packing

The 8-byte CellDynamic structure is tightly packed to minimize bandwidth:

Byte Layout: [0][1][2][3][4][5][6][7]
              └┬─┘  └──┬──┘  └──┬──┘
           Glyph ID  FG RGB   BG RGB
           (16-bit) (24-bit) (24-bit)

This layout enables the GPU to fetch all cell data in a single 64-bit read, with the glyph ID encoding both the texture coordinate and style information as described in the Glyph ID Bit Layout section.

Memory Layout and Performance

For a typical 12×18 pixel font with 2560 glyphs:

The 16×1 grid layout ensures that adjacent terminal cells often access the same texture layer, maximizing GPU cache hits. ASCII characters (the most common) are packed into the first 8 layers, providing optimal memory locality for typical terminal content.

Shader Pipeline

The renderer uses a branchless shader pipeline optimized for instanced rendering:

Vertex Shader (cell.vert)

Transforms cell geometry from grid space to screen space using per-instance attributes. The shader:

• Calculates cell position by multiplying grid coordinates with cell size
• Applies orthographic projection for pixel-perfect rendering
• Passes packed instance data directly to fragment shader without unpacking

Fragment Shader (cell.frag)

Performs the core rendering logic with efficient 2D array texture lookups:

• Extracts 16-bit glyph ID from packed instance data

• Masks with 0x0FFF to exclude effect flags before computing layer index (glyph_id → layer/position)

• Computes layer index and horizontal position using bit operations

• Samples from 2D texture array using direct layer indexing

• Detects emoji glyphs via bit 11 for special color handling

• Applies underline/strikethrough effects via bits 12-13

• Blends foreground/background colors with glyph alpha for anti-aliasing

WebGL2 Feature Dependencies

The renderer requires WebGL2 for:

• 2D Texture Arrays (TEXTURE_2D_ARRAY, texStorage3D, texSubImage3D)
• Instanced Rendering (drawElementsInstanced, vertexAttribDivisor)
• Advanced Buffers (UNIFORM_BUFFER, vertexAttribIPointer)
• Vertex Array Objects (createVertexArray)

Build and Deployment

Development Setup

# Install Rust toolchain
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
rustup target add wasm32-unknown-unknown

# Install tools
cargo install wasm-pack trunk

# Development server
trunk serve

# Production build
trunk build --release

Design Decisions

Why 16×1 Grid Per Layer?

• GPU compatibility: Single-row layout maximizes horizontal cache coherence and minimizes texture sampling overhead
• Simplified math: Position within layer is just a matter of glyph_id & 0x0F
• Cache efficiency: Sequential glyphs (e.g., ASCII characters) are horizontally contiguous, improving texture cache hit rates

Why Separate Style Encoding?

• Avoids duplicating glyph definitions
• Enables runtime style switching without texture lookups
• Maintains consistent coordinates for style variants

Limitations

• Maximum 512 base glyphs (9-bit addressing)

• Fixed 4 style variants per glyph

• Monospace fonts only

• Single font family and font size per atlas

TODO

• Text Effects: Underline, strikethrough
• Font Variants: Bold, italic, and other font weight support
• Complete Glyph Set: Report (e.g. via logging) when glyphs are missing from the atlas
• Emoji support: Currently renders with only the foreground color

Undecided|Lower Prio Features

• Double Buffering: Are there any benefits to double buffering for terminal rendering?
• Dynamic Atlases: Runtime glyph addition without regeneration
• Partial Updates: Only update changed cells instead of full grid
• Context Loss Recovery: Buffer architecture designed for WebGL context restoration