CUDA Kernels

Target directory

• src/kineticEQ/cuda_kernel/

Loaders in compile.py

• load_explicit_fused
• load_implicit_fused
• load_gtsv
• load_lo_blocktridiag
• load_implicit_AA

BGK1D1V kernels

explicit_fused/

A fused explicit one-step update. The binding requires CUDA tensors with torch.float64.

implicit_fused/

Provides moments_n_nu_T and build_system_from_moments. The binding also requires CUDA tensors with torch.float64.

gtsv/

Wraps cuSPARSE batched tridiagonal solves.

lo_blocktridiag/

Solves the 3x3 block-tridiagonal systems used by the HOLO low-order layer.

implicit_AA/

Implements Anderson acceleration for implicit Picard iterations.