GPU-PSO algorithm on GPU

An optimized fine-grained implementation of the Particle Swarm Optimization algorithm on GPUs using CUDA programming.

📜Here's the paper with our findings

Running the code

First, load cuda-12.4 using the command module load cuda-12.4

There are 4 files, one requires a common.h header file. Also, all these files have some default arguments set, so you don't need to pass any arguments.

But if you'd like, I've added flag info and sample arguments you could use.

Sequential

Compile using GCC - gcc -o seq cpu_pso.c Sample command - ./seq -m 10000 -d 4 -n 10

Arguments

• -m (max iterations)
• -c (particle count)
• -d (dimensions)
• -v (verbose, to output argument values while running the code)

Coarse grained

Arguments

• -m (max iterations)
• -c (particle count)
• -d (dimensions)
• -t (threads per block)
• -b (blocks)
• -v (verbose, to output argument values when running the code) Compile this file via nvcc -o coarse pso_cuda_coarse.cu Sample command - ./coarse -m 10000 -d 4 -t 64 -c 4096 -v 1

Fine-grained

Compile this file via nvcc -o finegrained pso_cuda_fg.cu Sample command - ./finegrained -m 10000 -d 4 -c 4096 Arguments

• -m (max iterations)
• -c (particle count)
• -d (dimensions)
• -t (threads per block)
• -b (blocks)
• -v (verbose, to output argument values when running the code)

Advanced fine-grained

Compile this file via nvcc -o adv pso_cuda_fg_advanced.cu Sample command - ./adv -m 10000 -c 4096 -d 4 -t 4 -b 2048 -v 1

Arguments

• -m (max iterations)
• -c (particle count)
• -d (dimensions)
• -t (threads per block)
• -b (blocks)
• -v (verbose, to output argument values when running the code)