Module cuda_compute 

CUDA backend implementation for tile-based FDTD simulation.

This module implements the TileGpuBackend trait for NVIDIA CUDA GPUs, providing GPU-resident tile buffers with minimal host transfers.

Features

• PTX compilation via NVRTC
• Persistent GPU buffers (pressure stays on GPU)
• Halo-only transfers (64 bytes per edge per step)
• Full grid readback only when rendering

Kernel Source

When the cuda-codegen feature is enabled, kernels are generated from Rust DSL. Otherwise, handwritten CUDA source from shaders/fdtd_tile.cu is used.

Structs

CudaTileBackend

CUDA tile GPU backend.