conceptclear / pso_cpu-gpu

An implementation of pso on cpu and gpu

pso_cpu-gpu

An implementation of pso on cpu and gpu

Usage

Listing parameters in PSO must be set before this program:

• maximal iterations
• dimension of the optimization problem
• number of particles
• operation platform (CUDA or CPU)
  parameters listed below can be changed:
• boundary
• c1,c2
• max_velocity
• omega
• max_stall_interations
• tolerance

Fitness function

Fitness funtion can be changed in main.cpp for cpu. In fact, function conducted in cpu could not be translated to cuda. If you want to use a new fitness function in cuda, fitness function defined in pso.cu should be changed.
'max_stall_interations_' is not suitable for cuda because calculating the number of iterations requires memory interaction which decreases the efficiency.

Example

Example fitness function is same as the example used in matlab.

for cpu with 10000 iterations and 1000 particles:

[Time] PSO CPU time: 1026.39ms
Optimization Results:
-0.707079
3.27442e-05

for cuda with 10000 iterations and 1000 particles:

[Time] PSO GPU time: 415.441ms
Optimization Results:
-0.70711
0.000184042