Struct WasmMemoryManager 

pub struct WasmMemoryManager { /* private fields */ }

Browser-side buffer manager that uses async map_async staging.

This mirrors WebGpuMemoryManager but is designed to work within the single-threaded browser event loop where blocking calls are not allowed.

Implementations

impl WasmMemoryManager

pub fn new(device: Arc<WasmGpuDevice>) -> Self

Create a new WASM memory manager backed by device.

pub fn alloc(&self, bytes: usize) -> WebGpuResult<u64>

Allocate a device buffer of bytes bytes.

pub fn free(&self, handle: u64) -> WebGpuResult<()>

Free the buffer identified by handle.

pub fn copy_htod(&self, handle: u64, src: &[u8]) -> WebGpuResult<()>

Upload host bytes to the device buffer (host-to-device copy).

Uses Queue::write_buffer which is available in both native and WASM.

pub fn copy_dtoh(&self, dst: &mut [u8], handle: u64) -> WebGpuResult<()>

Download a device buffer to host bytes (device-to-host copy).

This is a blocking readback: it maps a staging buffer and waits on the map callback. On a native target (including the wasm feature used for testing) Device::poll drives completion, so this works. On the real wasm32 browser main thread, however, blocking would starve the event loop that delivers the map callback and freeze the tab — so there this method returns WebGpuError::Unsupported and callers must use copy_dtoh_async instead.

pub async fn copy_dtoh_async( &self, dst: &mut [u8], handle: u64, ) -> WebGpuResult<()>

Asynchronously download a device buffer to host bytes.

Unlike copy_dtoh this never blocks the calling thread: it awaits the buffer-map completion via a future resolved from the map_async callback, making it the correct readback path on the single-threaded browser event loop. Like copy_dtoh, an oversized destination is rejected with WebGpuError::InvalidArgument.

Trait Implementations

impl Debug for WasmMemoryManager

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.