Tiny Universe Simulator System (TUSS)

Algorithm

Paper reference:
Fast Simulations of Gravitational Many-body Problem on RV770 GPU

Executables

# Compile all and run test_core
make

tus (main project)

• Tiny Universe Simulator
• The GPU simulator written in CUDA
• Located in ./src/tus

# Compile only
make tus
# Compile and run
make run_tus
# Run with arguments
make run_tus ARGS="any_args"

cpusim

• CPU SIMulator
• Located in ./src/cpusim

# Compile only
make cpusim
# Compile and run
make run_cpusim
# Run with arguments
make run_cpusim ARGS="any_args"

core

• The core common code
• core_lib
• Located in ./src/core

# Compile and test
make test_core
# With more verbose output
make test_core ARGS=-V
# See more options available
make test_core ARGS=-h

bicgen

• Bodies Initial Condition GENerator
• Load in TIPSY format, and translate into in-house BIN format
• Located in ./scripts/bicgen
• python3 -m scripts.bicgen

tussgui

• Tiny Universe Simulator System GUI
• Visualiation of trajectories via SYSTEM_STATEs
• Input files are expected to be ordered by integer
• Supports Live and Still trajectory mode
• Supports Still snapshot mode
• python3 -m scripts.tussgui

Input/Output Data

Units

No particular units are expected, but simulation uses 1 as the gravitational constant G. If converting from realistic data, make sure the transformed numerical values are converted, such that the gravitational equation with converted values gives identical result, as in its original values with gravitational constant that suits the original units, but with a gravitational constant G value of 1 instead.

An example conversion scheme: mass: [in Msun] * G_solar_mass_parsec_kmps(4.3009e-3) distance: [in ps] velocity: [in km/s]

BODY_STATE

The state of a single body, consists of (POS, VEC, MASS).
The serialization format expands each field, so that looks like: (POS.x,POS.y,POS.z,VEL.x,VEL.y,VEL.z, MASS)

SYSTEM_STATE

A collection of BODY_STATE, each represents individual BODY_STATE.

SYSTEM_STATE in CSV

Print out each BODY_STATE in serialization format.
Each BODY_STATE is represented as individual row with strings representing floating values, with comma to separate each field, and ends with a new line.
This format is not encouraged due to its large file size, and slow deserialization speed due to string processing.

SYSTEM_STATE in BIN

Print out each BODY_STATE in binary serialization format.
Each BODY_STATE is represented as a sequence of bytes, with bytes representing floating values. The specific format looks like the following:

• first 4 bytes: size of floating type (ie., 4 for floating, 8 for double)
• second 4 bytes: number of bodies
• rest: (POS.x,POS.y,POS.z,VEL.x,VEL.y,VEL.z, MASS) for each BODY_STATE

This format is the recommended format.

Demo

Visualization

cpusim

Solar System

mkdir -p ./tmp/solar_sys_cpu_log
make run_cpusim ARGS="-i ./data/ic/solar_system.csv -d 0.05 -n10000 -o ./tmp/solar_sys_cpu_log"
python3 -m scripts.tussgui trajectory_still ./tmp/solar_sys_cpu_log
# Or, some animation
python3 -m scripts.tussgui trajectory_live ./tmp/solar_sys_cpu_log
rm -rf ./tmp/solar_sys_cpu_log

Galaxy

mkdir -p ./tmp/
make run_cpusim ARGS="-i ./data/ic/s0_s112500_g100000_d100000.bin -d 10 -n20 -v -t4 -V1 -o ./tmp --snapshot"
python3 -m scripts.tussgui snapshot ./data/ic/s0_s112500_g100000_d100000.bin
python3 -m scripts.tussgui snapshot ./tmp/s0_s112500_g100000_d100000*
rm -rf ./tmp

tus

Solar System

mkdir -p ./tmp/solar_sys_tus_log
make run_tus ARGS="-i ./data/ic/solar_system.csv -d 0.05 -n10000 -o ./tmp/solar_sys_tus_log"
python3 -m scripts.tussgui trajectory_still ./tmp/solar_sys_tus_log
# Or, some animation
python3 -m scripts.tussgui trajectory_live ./tmp/solar_sys_tus_log
rm -rf ./tmp/solar_sys_tus_log

Galaxy

mkdir -p ./tmp/g
make run_tus ARGS="-i ./data/ic/one_galaxy_138410.bin -d 0.001 -n10000 -o ./tmp/g --snapshot -V4 --len=1 --wid=512 --luf=1024"
python3 -m scripts.tussgui snapshot ./data/ic/one_galaxy_138410.bin
python3 -m scripts.tussgui snapshot ./tmp/g/one_galaxy_138410*

Performance

cpusim

make run_cpusim ARGS="-i ./data/ic/solar_system.csv -d 0.05 -n 10000"
# Base reference
make run_cpusim ARGS="-i ./data/ic/benchmark_100000.bin -d 0.001 -n10 -v -t1 -V0"
# Latest version
make run_cpusim ARGS="-i ./data/ic/benchmark_100000.bin -d 0.001 -n10 -v -t4 -V1"

python3 -m scripts.benchmark cpu

tus

make run_tus ARGS="-i ./data/ic/solar_system.csv -d 0.05 -n 10000"
# Base reference
make run_tus ARGS="-i ./data/ic/benchmark_100000.bin -d 0.001 -n10 -v -V0"
# Latest version
make run_tus ARGS="-i ./data/ic/benchmark_100000.bin -d 0.001 -n10 -v -V1"
make run_tus ARGS="-i ./data/ic/s0_s112500.bin -d 0.001 -n10 -v -V1 -t32"
make run_tus ARGS="-i ./data/ic/s0_s112500.bin -d 0.001 -n10 -v -V1 -t64"
# Best performance
make run_tus ARGS="-i ./data/ic/s0_s112500.bin -d 0.001 -n10 -b 100000 -v -V4 -t32 --len 1 --wid 256 --lur 1024"
# Simple profilier
nvprof ./build/tus/tus_exe -i ./data/ic/s0_s112500.bin -d 0.001 -n10 -b 100000 -v -V4 -t32 --len 1 --wid 256 --lur 1024

# Deprecated
python3 scripts/benchmark/cuda.py
# Preferred
python3 -m scripts.benchmark cuda

Verification

cpusim

make run_cpusim ARGS="-i ./data/ic/solar_system.csv -d 0.05 -n 500 --verify"
make run_cpusim ARGS="-i ./data/ic/benchmark_100000.bin -d 0.001 -n 2 -t 4 --verify"

tus

make run_tus ARGS="-i ./data/ic/solar_system.csv -d 0.05 -n 500 --verify"
make run_tus ARGS="-i ./data/ic/benchmark_100000.bin -d 0.001 -n 2 --verify"

CMake and Makefile

Arg Passing to exe

ARGS="<your_args>"

Enable -ffast-math

CMAKE_ARGS="-DENABLE_FFAST_MATH=ON"