Tested on:
- Arch Linux 6.8.9 (as for 02/06/24)
- GCC: 14.1.1
- CUDA: 12.4
Compilation:
make
Execution:
Run as ./test N S MODE ITER DEVICE MEM_GPU BS DIST REDUX Select-MU MU SAMPLE MOLECULES
N - length/width of grid
S - number of seeds
MODE - (0) JFA, (1) dJFA, (2) rJFA, (3) drJFA
ITER - number of iterations
DEVICE - Wich device to use, fill with ID of GPU
MEM_GPU - Memory management, 1: manual, 0: auto, anything if DEVICE = 0
BS - block size of GPU (32 max), anything if DEVICE = 0
DIST - distance method, 1: manhattan, 0: euclidean
REDUX - if redux method is used, 0: no, 1: yes
Select MU - election of MU, 1: arbitrary, 0:estimated
MU - value of MU if arbitrary
SAMPLE - 0: rand uniform, 1: lennard jones sample, 2: nbody
MOLECULES - for lennard jones sample {25,50,75,100,125,150,175,200}
DEBUG SIM - if compare against naive, 1: yes, 0: no
Example:
./test $((2**14)) $((2**10)) 0 10 0 0 32 0 0 1 4 0 50 0
N: 16384, N_p(if used): 4096
MU: 4
Sample: 0
Comparison: 0
PBC: 0
Blocksize: 32
Method: 0
Distance used(1: Manhattan, 0:Euclidean): 0
k: 16384, k_m: 2048, k_r: 4096, k_rm: 512
TOTAL TIME REFERENCE: 0.000012, TOTAL TIME PROPOSAL: 3.043226
To compile the 3D version of the program, type the following
make cmp3d
To execute the program, is the same as before, run as test and provide the same arguments, be careful with memory consumption, for N the grid is going to be NxNxN