How extremeflow/neko works
Neko is a portable, high-order spectral-element framework for flow simulations. The documented user workflow is to provide a case (and optionally user code), then run Neko—often under MPI—to perform a simulation. The supplied tree emphasizes accelerator kernels and supporting parallel-I/O and physics modules; it does not expose the main Fortran runtime entry point or enough call sites to reconstruct a complete solver pipeline. The graph therefore includes documented capabilities and indexed kernel dependencies, while leaving unsampled wiring out.
Main entry points: cuda/ax_helm_kernel.h CUDA Helmholtz · hip/ax_helm_kernel.h HIP Helmholtz · dmma_tma_kernel.h CUDA TMA · src Neko runtime
Components 19
Solver numerics
- Neko runtime
src - SEM coefficients
src/sem/bcknd/device/opencl/coef.c - Math operators
src/math/bcknd/device/opencl/math.c
Flow extensions
- Moving mesh
src/ale/bcknd/device/opencl/ale_kinematics.c - Compressible flow
src/fluid/bcknd/device/opencl/compressible_res.c - LES models
src/les/bcknd/device/opencl/dynamic_smagorinsky_nut.c - Particle tracking
src/lpt/bcknd/device/opencl/lpt_wall_collision.c - Scalar residuals
src/scalar/bcknd/device/opencl/scalar_residual.c
Parallel and I/O
- Communication
src/comm/comm_wrapper.c - ADIOS2 streaming
src/io/nek_adios2.cpp
Accelerator compute
- CUDA Helmholtz
src/math/bcknd/device/cuda/ax_helm_kernel.h - CUDA DMMA
src/math/bcknd/device/cuda/dmma_kernel.h - CUDA TMA
src/math/bcknd/device/cuda/dmma_tma_kernel.h - CUDA element blocks
src/math/bcknd/device/cuda/elem_block.h - HIP Helmholtz
src/math/bcknd/device/hip/ax_helm_kernel.h - HIP element blocks
src/math/bcknd/device/hip/elem_block.h - HIP MFMA
src/math/bcknd/device/hip/mfma_kernel.h
Other components
- Simulation user
- Case input
Connections 9
- Simulation user → Case input: provides case
README.md - Simulation user → Neko runtime: runs
README.md - Case input → Neko runtime: configures
README.md - CUDA Helmholtz → CUDA DMMA: uses
src/math/bcknd/device/cuda/ax_helm_kernel.h - CUDA Helmholtz → CUDA TMA: uses
src/math/bcknd/device/cuda/ax_helm_kernel.h - CUDA Helmholtz → CUDA element blocks: uses
src/math/bcknd/device/cuda/ax_helm_kernel.h - CUDA TMA → CUDA DMMA: includes
src/math/bcknd/device/cuda/dmma_tma_kernel.h - HIP Helmholtz → HIP element blocks: uses
src/math/bcknd/device/hip/ax_helm_kernel.h - HIP Helmholtz → HIP MFMA: uses
src/math/bcknd/device/hip/ax_helm_kernel.h