API Reference

Complete API documentation for WebGPU Sorting.

Core Classes

GPUContext

Manages WebGPU device lifecycle.

import { GPUContext } from 'webgpu-sorting';

Static Methods

isSupported(): boolean

Check if WebGPU is available in the current environment.

if (GPUContext.isSupported()) {
  // WebGPU is available
}

Constructor

new GPUContext()

Create a new GPU context instance.

const gpu = new GPUContext();

Methods

initialize(config?: GPUContextConfig): Promise<void>

Initialize the WebGPU device.

interface GPUContextConfig {
  powerPreference?: 'low-power' | 'high-performance';
}

await gpu.initialize({
  powerPreference: 'high-performance',
});

getDevice(): GPUDevice

Get the underlying WebGPU device.

const device = gpu.getDevice();

destroy(): void

Release GPU resources.

gpu.destroy();

---

BitonicSorter

GPU-accelerated Bitonic Sort implementation.

import { BitonicSorter } from 'webgpu-sorting';

Constructor

new BitonicSorter(context: GPUContext)

Create a new Bitonic sorter.

const sorter = new BitonicSorter(gpu);

Methods

sort(data: Uint32Array, options?: SortOptions): Promise<SortResult>

Sort an array of unsigned 32-bit integers.

interface SortOptions {
  validate?: boolean; // Verify output is sorted (default: false)
}

interface SortResult {
  sortedData: Uint32Array; // The sorted array
  gpuTimeMs: number; // GPU execution time
  totalTimeMs: number; // Total time including transfers
}

const result = await sorter.sort(data, { validate: true });

destroy(): void

Release GPU resources used by the sorter.

sorter.destroy();

preallocate(maxSize: number): void

Preallocate GPU buffers for sorting arrays up to maxSize. Buffers are reused across multiple sort() calls for better performance in batch sorting scenarios.

sorter.preallocate(1_000_000); // Preallocate for up to 1M elements

// Multiple sorts reuse the same buffers
for (const arr of arrays) {
  const result = await sorter.sort(arr);
}

clearPreallocation(): void

Release preallocated buffers. Buffers will be allocated on-demand for subsequent sorts.

sorter.clearPreallocation();

preallocatedSize: number (readonly)

Returns the current preallocation size, or 0 if not preallocated.

console.log(sorter.preallocatedSize); // 1000000 or 0

---

RadixSorter

GPU-accelerated Radix Sort implementation.

import { RadixSorter } from 'webgpu-sorting';

Constructor

new RadixSorter(context: GPUContext)

Create a new Radix sorter.

const sorter = new RadixSorter(gpu);

Methods

sort(data: Uint32Array, options?: SortOptions): Promise<SortResult>

Sort an array of unsigned 32-bit integers.

const result = await sorter.sort(data);

Best For

Radix Sort is optimized for Uint32Array datasets. For other data types, use BitonicSorter.

destroy(): void

Release GPU resources used by the sorter.

sorter.destroy();

preallocate(maxSize: number): void

Preallocate GPU buffers for sorting arrays up to maxSize. Buffers are reused across multiple sort() calls for better performance in batch sorting scenarios.

sorter.preallocate(1_000_000); // Preallocate for up to 1M elements

// Multiple sorts reuse the same buffers
for (const arr of arrays) {
  const result = await sorter.sort(arr);
}

clearPreallocation(): void

Release preallocated buffers. Buffers will be allocated on-demand for subsequent sorts.

sorter.clearPreallocation();

preallocatedSize: number (readonly)

Returns the current preallocation size, or 0 if not preallocated.

console.log(sorter.preallocatedSize); // 1000000 or 0

---

Error Classes

WebGPUNotSupportedError

Thrown when WebGPU is not available in the browser.

import { WebGPUNotSupportedError } from 'webgpu-sorting';

try {
  await gpu.initialize();
} catch (error) {
  if (error instanceof WebGPUNotSupportedError) {
    // Browser doesn't support WebGPU
  }
}

GPUAdapterError

Thrown when GPU adapter acquisition fails.

import { GPUAdapterError } from 'webgpu-sorting';

GPUDeviceError

Thrown when GPU device acquisition fails.

import { GPUDeviceError } from 'webgpu-sorting';

BufferAllocationError

Thrown when GPU buffer allocation fails.

import { BufferAllocationError } from 'webgpu-sorting';

ShaderCompilationError

Thrown when WGSL shader compilation fails.

import { ShaderCompilationError } from 'webgpu-sorting';

---

Types

SortOptions

interface SortOptions {
  /**
   * Verify the output is correctly sorted after sorting completes.
   * Useful for debugging and testing.
   * @default false
   */
  validate?: boolean;
}

SortResult

interface SortResult {
  /**
   * The sorted array
   */
  sortedData: Uint32Array;

  /**
   * GPU execution time in milliseconds
   */
  gpuTimeMs: number;

  /**
   * Total time including data transfer in milliseconds
   */
  totalTimeMs: number;
}

GPUContextConfig

interface GPUContextConfig {
  /**
   * GPU power preference
   * @default 'high-performance'
   */
  powerPreference?: 'low-power' | 'high-performance';
}

---

Constants

WORKGROUP_SIZE

Default workgroup size for compute shaders.

import { WORKGROUP_SIZE } from 'webgpu-sorting';
// Value: 256

MAX_BUFFER_SIZE

Maximum buffer size supported.

import { MAX_BUFFER_SIZE } from 'webgpu-sorting';
// Value: Device dependent

---

Usage Examples

Basic Sorting

const gpu = new GPUContext();
await gpu.initialize();

const sorter = new BitonicSorter(gpu);
const data = new Uint32Array([5, 2, 8, 1, 9]);
const { sortedData } = await sorter.sort(data);

gpu.destroy();

Multiple Sorts

const gpu = new GPUContext();
await gpu.initialize();

const sorter = new BitonicSorter(gpu);

const arrays = [
  new Uint32Array([3, 1, 4, 1, 5]),
  new Uint32Array([9, 2, 6, 5, 3]),
  new Uint32Array([8, 9, 7, 9, 3]),
];

for (const arr of arrays) {
  const result = await sorter.sort(arr);
  console.log(result.sortedData);
}

gpu.destroy();

Error Handling

async function safeSort(data: Uint32Array): Promise<Uint32Array> {
  if (!GPUContext.isSupported()) {
    return data.slice().sort((a, b) => a - b);
  }

  try {
    const gpu = new GPUContext();
    await gpu.initialize();
    const sorter = new BitonicSorter(gpu);
    const result = await sorter.sort(data);
    gpu.destroy();
    return result.sortedData;
  } catch (error) {
    console.warn('GPU sort failed, using CPU fallback:', error);
    return data.slice().sort((a, b) => a - b);
  }
}

Performance Comparison

async function compareSorters(size: number) {
  const data = new Uint32Array(size);
  for (let i = 0; i < size; i++) data[i] = Math.random() * 1_000_000;

  const gpu = new GPUContext();
  await gpu.initialize();

  const bitonic = new BitonicSorter(gpu);
  const radix = new RadixSorter(gpu);

  const r1 = await bitonic.sort(data);
  const r2 = await radix.sort(data);

  console.log(`Bitonic: ${r1.gpuTimeMs}ms`);
  console.log(`Radix: ${r2.gpuTimeMs}ms`);

  gpu.destroy();
}