Struct CudaPackedBackend 

pub struct CudaPackedBackend { /* private fields */ }

Packed CUDA backend for tile-based FDTD simulation.

All tiles are stored in a single contiguous GPU buffer, enabling GPU-only halo exchange without any host transfers.

Implementations

impl CudaPackedBackend

pub fn new(grid_width: u32, grid_height: u32, tile_size: u32) -> Result<Self>

Create a new packed CUDA backend.

Arguments

• grid_width - Width of the simulation grid in cells
• grid_height - Height of the simulation grid in cells
• tile_size - Size of each tile’s interior (typically 16)

pub fn with_device( ordinal: usize, grid_width: u32, grid_height: u32, tile_size: u32, ) -> Result<Self>

Create a packed CUDA backend on a specific device.

pub fn set_params(&mut self, c2: f32, damping: f32)

Set FDTD parameters.

pub fn upload_tile( &self, tile_x: u32, tile_y: u32, pressure: &[f32], ) -> Result<()>

Upload initial pressure data for a single tile.

pub fn upload_all_tiles(&self, tiles: &[(u32, u32, Vec<f32>)]) -> Result<()>

Upload initial state for all tiles from a Vec of tile buffers.

pub fn step(&mut self) -> Result<()>

Perform one simulation step (GPU-only, no host transfers).

This executes:

1. Halo exchange kernel (copies all edges between tiles)
2. Boundary conditions kernel (handles grid edges)
3. FDTD kernel (computes wave equation for all tiles)
4. Buffer swap (just flips a flag)

pub fn step_batch(&mut self, count: u32) -> Result<()>

Run N steps without any host interaction.

pub fn inject_impulse(&self, x: u32, y: u32, amplitude: f32) -> Result<()>

Inject an impulse at the given global grid position.

pub fn read_pressure_grid(&self) -> Result<Vec<f32>>

Read the full pressure grid for visualization.

This is the only operation that transfers data from GPU to host. Call this only when you need to render the simulation.

pub fn synchronize(&self) -> Result<()>

Synchronize the GPU (wait for all operations to complete).

pub fn grid_size(&self) -> (u32, u32)

Get grid dimensions.

pub fn tile_grid_size(&self) -> (u32, u32)

Get tile grid dimensions.

pub fn tile_count(&self) -> u32

Get total number of tiles.

pub fn halo_copy_count(&self) -> u32

Get number of halo copy operations per step.

Trait Implementations

impl Debug for CudaPackedBackend

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

Formats the value using the given formatter.