Not Found

Documentation

Introduction Type Safety Tensor Expressions Tensor Indexing Einsum Einops Autograd Training a Model Profiling & Memory PyTorch Migration Best Practices Runtimes Performance PyTorch Compatibility Benchmarks DType Coverage

wgpu-native.WGPUBuffer

export interface WGPUBuffer {
  /**
   * Size of the buffer in bytes.
   */
  readonly size: number;

  /**
   * Combined {@link BufferUsage} flags specifying allowed operations.
   */
  readonly usage: number;

  /**
   * Current mapping state: 'unmapped', 'pending', or 'mapped'.
   *
   * - 'unmapped': Buffer is not mapped
   * - 'pending': mapAsync() called, waiting for GPU
   * - 'mapped': Buffer is mapped and accessible
   */
  readonly mapState: 'unmapped' | 'pending' | 'mapped';

  /**
   * Opaque native handle to this buffer.
   */
  readonly handle: BufferHandle;

  /**
   * Map buffer for CPU access.
   *
   * @param mode - Map mode (READ or WRITE from {@link MapMode})
   * @param offset - Byte offset into buffer (default: 0)
   * @param size - Number of bytes to map (default: entire buffer)
   * @returns Promise that resolves when buffer is ready for access
   */
  mapAsync(mode: number, offset?: number, size?: number): Promise<void>;

  /**
   * Get a mapped range of the buffer as an ArrayBuffer.
   *
   * Only valid when buffer is in 'mapped' state.
   *
   * @param offset - Byte offset into buffer (default: 0)
   * @param size - Number of bytes (default: entire buffer)
   * @returns ArrayBuffer view of the mapped region
   */
  getMappedRange(offset?: number, size?: number): ArrayBuffer;

  /**
   * Unmap the buffer, making it inaccessible from CPU but usable by GPU again.
   */
  unmap(): void;

  /**
   * Destroy the buffer and free its GPU memory.
   */
  destroy(): void;
}

readonlysize(number)

– Size of the buffer in bytes.

readonlyusage(number)

– Combined BufferUsage flags specifying allowed operations.

readonlymapState('unmapped' | 'pending' | 'mapped')

– Current mapping state: 'unmapped', 'pending', or 'mapped'. - 'unmapped': Buffer is not mapped - 'pending': mapAsync() called, waiting for GPU - 'mapped': Buffer is mapped and accessible

readonlyhandle(BufferHandle)

– Opaque native handle to this buffer.

mapAsync((mode: number, offset?: number, size?: number) => Promise<void>)

– Map buffer for CPU access.

getMappedRange((offset?: number, size?: number) => ArrayBuffer)

– Get a mapped range of the buffer as an ArrayBuffer. Only valid when buffer is in 'mapped' state.

unmap(() => void)

– Unmap the buffer, making it inaccessible from CPU but usable by GPU again.

destroy(() => void)

– Destroy the buffer and free its GPU memory.

GPU buffer resource.

Represents a block of GPU memory that can hold data for tensors, indices, uniforms, etc. Buffers must be created with specific usage flags that determine what operations they can be used for.