Turbulent Dynamics Lattice Boltzmann (C++)

This is a basic version of the multi-node heterogeneous HPC code to run simulation with hundreds of billions of cells.

Running

#Generate an input file
python3 generate_stirred_tank_input.py -x 100 -f input_debug_gridx100_numSteps20.json
./tdlbcpp --input_file input_debug_gridx100_numSteps20.json

#Load from checkpoint dir
./tdlbcpp --checkpoint_dir checkpoint_2021_9_21__jd7aflakjd

Building

### Dependencies

sudo apt install bazel
cp /location/to/cuda-samples/Common/{helper_*,exception.h} tdlbcpp/src

Generic build

bazel build //tdlbcpp/src:tdlbcpp --verbose_failures -s

GPU build

uses nvcc compiler for cuda files and main.cpp and sets WITH_GPU | WITH_GPU_MEMSHARED define

bazel build --config gpu //tdlbcpp/src:tdlbcpp
bazel build --config gpu_shared //tdlbcpp/src:tdlbcpp

For debug build add -c dbg switch:

bazel build -c dbg --config gpu //tdlbcpp/src:tdlbcpp

Specify capabilities and custom cuda path using the following switches:

--@rules_cuda//cuda:cuda_targets=sm_XY --repo_env=CUDA_PATH=/usr/local/cuda-ZZZ

bazel build -c dbg --config gpu --@rules_cuda//cuda:cuda_targets=sm_75 --repo_env=CUDA_PATH=/usr/local/cuda-11.5 //tdlbcpp/src:tdlbcpp

Tests

bazel test //tdlbcpp/tests/Params:tests

Package Structure

Vector Identification (see Header.h)