600086-Lab-I Parallel Particle Systems Concept
Implement two versions of a simple aerosol paint simulation, one on the CPU and the other on the GPU.
Specification
The scene is composed of a white sheet of paper placed in the centre of a desk. A number of aerosol cans stand on the desk in a circle that surrounds the paper. Each can is initially facing towards the centre of the desk.
The aim of the application is to produce an image on the white paper, by simulating the paint as it is sprayed from each aerosol can, and lands on the paper.
The aerosol spray is represented as a system of particles, where each particle is modelled as a simple projectile subject to both gravity and air resistance.
distance s = v t + 0.5 a t^2 and
acceleration a = g – k v^2 where
t = time g = gravity (9.8 m/s^2) k = drag v = velocity
Initial values for velocity and drag are 1 m/s and 0.05, respectively. These can be adjusted to achieve a good spray pattern.
It is recommended that the initial direction of the spray is parallel to the table and located at least 0.2m above the table.
The spray leaves the aerosol can in a cone. Adjust the cone angle until the majority of the paint hits the paper.
Particle system
The particle system should consist of 1000’s or 10000’s of particles. No collision detection is required between particles.
Each aerosol can contains a single colour of particles.
The colour of a particle is represented as a combination of Red, Green and Blue values in the range 0-1. e.g. Bright Red is (1, 0, 0), Bright Green is (0, 1, 0), Bright Blue is (0, 0, 1), Black is (0, 0, 0) and White (1, 1, 1).
Code Help
The paper is represented as 2D array of at least 1000 x 1000 elements.
When a particle lands on the paper, one element (pixel) of the paper changes colour. The new colour of the paper is calculated as
paper colour = (1 – b) m + b p
m is original paper colour p is the aerosol colour b is the blend coefficient (0.1)
The desk has no role in the simulation other than as a virtual location to rest the paper and aerosol cans.
Once the simulation has completed, the paper array should be saved as a bitmap file or directly displayed on the screen.
You could use the Rust bmp crate to output the image.
Parallel architecture
The threading requirements are specified below. Note that no thread can implement more than one of the listed functions.
CPU implementation
A pipeline architecture is required for the CPU implementation, consisting of 3 stages:
particle creation: Created a batch of particles at the spray can nozzle, per time step.
particle movement: Move a batch of the particles from the spray can nozzle to the virtual table / paper. paper colouring: Colour the paper with any particles from the batch that hit the paper.
A simulation run consists of multiple batches of particles.
Add suitable buffers and duplicate stages to the pipeline (as described in the lectures) to improve performance.
GPU implementation
Particle motion – implemented on 1 thread per particle
Collision of particles with paper and colouring – implemented on 1 thread per collision
(selected by a key-press)
1. Atleast3differentcolouredaerosolsprayslocationindifferentpositionsaroundthepaper. 2. Measuretheperformanceofthesimulation,including:
How long did it take to complete the simulation? How many particles were created?
How many particles hit the paper?
How many particles missed the paper?
Advanced features
(selected by a key-press)
1. VisualizationofaerosolparticlesrenderedusingOpenGLorDirectX,asdiscussedinpreviouslabs (switch off when gathering performance metrics).
2. Morecreativeusesoftheaerosolspraysandpaper
Implementation CPU
A Rust implementation Functionality requiring beta or nightly Rust builds are NOT permitted.
A CUDA implementation. The choice of language from which to host CUDA is left open e.g. C++, C# or Python
Lab book entries
1. GPUdesign-DetailandreflectuponthedesignofyourGPUimplementation.Focusontheparallel aspects of your solution. Clearly explain how you have used threads, mutual exclusion and synchronization in your GPU implementation. (approx. 300-500 words)
2. CPUdesign-DetailandreflectupondesignofyourCPUimplementation.Focusontheparallelaspects of your solution. Clearly explain how you have used threads, mutual exclusion and synchronization in your CPU implementation. (approx. 300-500 words)
3. Performancemetrics-ComparetheperformanceofyourGPUandCPUimplementations,bothWITH and WITHOUT graphics. Include a description of your performance benchmarks. (approx. 300-500 words)
Demonstration
You will be required to demonstrate your two solutions in FEN-177
Submission
All submissions are via Canvas.
程序代写 CS代考 加微信: cstutorcs
Software – CUDA
Your code, in the form of a Visual Studio solution, source code, and working executables/DLLs, along with any required assets.
Software – Rust
Your code, in the form of a Visual Studio Code, source code, and working executables, along with any required assets.
A short narrated video containing showing both the CPU and GPU implementations
The full lab book containing all labs, as a PDF.
Computer Science Tutoring