GRAPHICS PIPELINE • MAY 2026

Global AST Search via WebGPU Compute Shaders.

When you press Cmd+Shift+F to search a massive monorepo, a standard IDE has to iterate through every single file sequentially using the CPU. Even with multi-threading, executing a complex Regular Expression across 100,000 files takes several seconds and completely spikes CPU temperatures.

NitroIDE completely abandons the CPU for global searches. We utilize WebGPU Compute Shaders to offload the text processing to your graphics card, searching thousands of files simultaneously across thousands of parallel GPU cores.

Writing the WGSL Search Kernel

Instead of writing our search algorithm in JavaScript, we write it in WGSL (WebGPU Shading Language). We load the raw byte arrays of your Virtual File System directly into VRAM (Video RAM). We then dispatch a compute shader where each GPU thread is responsible for searching exactly one chunk of text in parallel.

// WGSL Compute Shader for parallel text searching
@group(0) @binding(0) var<storage, read> fileData: array<u32>;
@group(0) @binding(1) var<storage, read_write> matchResults: array<i32>;

@compute @workgroup_size(256)
fn main(@builtin(global_invocation_id) global_id: vec3<u32>) {
  let index = global_id.x;
  // GPU mathematically compares bytes in parallel across 4096 cores
  if (checkPatternMatch(fileData[index])) {
    matchResults[index] = 1; // Flag as a match
  }
}

Storage Buffers: Because WebGPU allows for massive `StorageBuffer` allocations (unlike the restrictive textures of WebGL), we can stream a 1GB project directory into the GPU, execute the search, and map the result buffer back to JavaScript in less than 50 milliseconds.

Instantaneous Regex

This architecture results in a search feature that is completely decoupled from project size. Whether you are searching a 10-file React app or the entire 30-million-line Linux Kernel codebase, the search completes almost instantaneously, bound only by the memory bandwidth of your graphics card.

Search at the Speed of Light.

Import a massive codebase and run a global regex search. Watch the GPU shred it.

Launch Workspace