kevinmcgin / simulation

Simulates the motion of particles on the GPU with Nvidia's CUDA. Renders simulations on OpengGL.

Simulation

Simulates the motion of particles on the CPU or on the GPU with CUDA.

Renders these simulations with OpenGL.

Render Example

2021-12-16.13-08-52.mp4

Road Map

Wiki

Dependencies

• CMake
• CUDA
• OpenGL

Installation

Installation Guide

Configuration

Configuration files are in the config folder.

Git ignored configuration files are generate from *.config.example files after building.

Project

The project config file allows the configuration of the BUILD_PATH and whether to USE_GPU in the simulation engine.

Scripts

Scripts Guide

simulation's Issues

Fix DistributionCircle

Distribution is biased towards an xy-grid
May affect sphere and globe?

Print out progress of simulation

OpenGL on C++

Input may be serialised c++ file

https://stackoverflow.com/questions/9460242/how-to-compile-glut-opengl-project-with-cmake-and-kdevelop-in-linux
https://www3.ntu.edu.sg/home/ehchua/programming/opengl/HowTo_OpenGL_C.html
https://www.khronos.org/opengl/wiki/Getting_Started

use gpu as run time param

instead of as compile time param = compilatios consistent.
main params
test runtie params

OpenGL Real Imposters

May need a choice between real imposters and actual polys

update readme for linux

print progress within laws

blocked by law interaction and collision refactor

Memory leaks

Detect memory leaks and fix.
In test app too.
https://docs.microsoft.com/en-us/visualstudio/debugger/finding-memory-leaks-using-the-crt-library?view=vs-2019

build info above readme

Law Interaction

for (const auto& p1 : particles)
for (const auto& p2 : particles) {
if (p1 != p2) {

		}
	}

timing of main.cpp runs

Update a time field in the cmd line.
Maybe after each pass of a particle in Law, might be slow

OpenGL control camera

End renderer gracefully

https://stackoverflow.com/questions/32303257/looping-over-an-array-in-rapidjson-and-getting-the-object-elements

Runtime args

https://github.com/likle/cargs
https://likle.github.io/cargs/embed.html

Reference frames & gravitational lensing

Investiage how light shifting in space time would affects our simulation renders.

model velocity based on angular momentum (solar/galaxy disk)

code quality

Used SonarCloud

Gitignored config files so as a developer I can safely checkout branches without needing to stash configuration changes in bash run files

Correct radius initialising from mass

New classes:
DistributionDensity {
getMassRadius(double &mass, &radius);
}
DistributionMassRadius(mass, radius) : DistributionDenisty
DistributionMassDensity(mass, density): DistributionDensity
DistrubtuionValue(value) : Distribution

Print Vector3D function

https://stackoverflow.com/questions/23239646/print-function-for-class-c

Bugfix: DistributionGlobe

Is really a cuboid distribution.
Rename classes and create circle, sphere, globe distributions.

Runnable to be Open Source

Project as easy to run as possible.
Last prerequiste to going public!
Should test running on linux usb from scratch. Use knowledge gained to add to readme.

Speed up gravity calculation

acceleration is a function of r and m. Reuse the r part

pure parallel laws

GPU run functions to only manipulate GPU data and not to copy data to and from the CPU. The CPU data is to be copied to and from at the beginning and end of each frame.
Blocked by #64

• GPU data allocated and deallocated between frames only
• only deallocate and reallocate memory if a particle is deleted
• refactor/fix GPU tests
• add NewtonFirstLaw GPU tests

Parrallise Collision

gpuRun function for Collision.
The run function for Collision is segmented into 4 different sections by comments. The frist 3 segments can be put into 3 kernels for the GPU. the final segment can be added as a separte Law that is run on the CPU: DeletionLaw (Those particles who have been marked for deletion shall be cast away!)

wait total frame time - render time between frames

Choosing x particles produces x+1 particles, change to producing x particles

remove duplicate 'random' code in distributions

Maybe all random logic in distribution

add licence

Seq vs CUDA Parallel Calculations

Configure Opengl and Glut on pipeline

CPU only tests on pipelines

refactor collision system (reduce comparisons)

Run tests on GitHub

Need add main executable seperately

https://docs.github.com/en/actions
https://docs.gitlab.com/ee/ci/unit_test_reports.html#googletest

fix gravity tests

Finish Collision Resolver Tests

Vector3D == operator

benchmark GPU vs CPU

Using CPU
Simulation running
101 particles
100% 18.8768s Simulation complete
Using GPU
Simulation running
101 particles
100% 34.4751s Simulation complete

Using CPU
Simulation running
1001 particles
100% 7.04874mins Simulation complete
Using GPU
Simulation running
1001 particles
100% 5.699mins Simulation complete

Using CPU
Simulation running
3001 particles
10.833% 10.5476mins
Using GPU
Simulation running
3001 particles
49.185% 11.5413mins

C++ json input

rename the main as engine

fix non hard compilation failures

automate .obj test files

Vector3D operator==

Then simplify tests

bash command line args

hard compile, test without compile, main without compile, main with render.

https://stackoverflow.com/questions/192249/how-do-i-parse-command-line-arguments-in-bash

Separate Header and Source Files

Parallelise Newton Gravity

• break up runParticle into two parts for cpu and gpu.
• do two single gpu operations: calculate and store radius components in triagnular matrix, apply radius components row by row on the n-1 other particles.

NB: https://stackoverflow.com/questions/40682163/cuda-copy-inherited-class-object-to-device?answertab=oldest#tab-top

Code Coverage

Used SonarCloud

Run renderer automatically

Fix parallelised gravity memory bugs

Collision Bug: deleted big particle