Chemgen

This is a repository to generate chemistry for combustion solvers

Testing

python -m pytest tests/

Options for testing:

• --mech_file_ck: Input mechanism file (Chemkin format)
• --therm_file: Thermodynamic data file
• --language: Language for testing (fortran/python)
• --temperature: Temperature in Kelvin
• --pressure: Pressure in Pascal
• --fuel: Fuel composition
• --oxidiser: Oxidiser composition

Usage

Generate GPU coefficients:

python generate_chemistry.py --mode gpu_coef \
    --mech mechanism.inp \
    --therm therm.dat \
    --output output_dir \
    --parallel-level 1

• --parallel-level: GPU parallelization level (1-4)

Generate getrates functions:

python generate_chemistry.py --mode getrates \
    --mech mechanism.inp \
    --therm therm.dat \
    --output getrates.f90 \
    --language fortran \
    --omp

Key options:

• --language: Output language (fortran/python)
• --omp: Enable OpenMP support
• --vector: Generate vectorized version
• --module: Generate a fortran GPU module

Generate mini apps:

./generate_mini_apps.sh <path-to-chem>
# e.g.
./generate_mini_apps.sh ck_files/H2_burke

• resulting a folder named mini_apps_<chem>

Compile mini apps:

# request an interactive node for compilation
salloc --partition=gpu-dev --account=w47-gpu --time=01:30:00 --nodes=1 --gpus-per-node=8 --exclusive
# load the required modules and set enviroment variables
export MPICH_GPU_SUPPORT_ENABLED=1
export OMP_NUM_THREADS=1
module load rocm craype-accel-amd-gfx90a
module load cmake/3.27.7

to compile all miniapps for a mechanism, run:

./compile_mini_apps.sh mini_apps_<chem>
# e.g.
./compile_mini_apps.sh mini_apps_H2_burke  

For a mechanism, run all mini apps to collect performance data:

./compare_mini_apps_write_csv.sh mini_apps_<chem>
#e.g.
./compare_mini_apps_write_csv.sh mini_apps_H2_burke

The performance data will be placed in a csv file naming performance_table_mini_apps_<chem>.csv

./plot_omp_speedup.sh performance_table_mini_apps_<chem>.csv
#e.g.
./plot_omp_speedup.sh performance_table_mini_apps_H2_burke.csv

figures showing how gpu-omp is speeded up with varying problem size (ng) is then generated and placed at a folder naming figures_<chem>