CudaSimulatedBifurcationMachine

Code Source

Overall time complexity: O(sweep*dim)

C code (sbm.c)

This is the c prototype of the algorithm.

Cuda C code (sbm.cu)

Used OS: Ubuntu 20.04.3 LTS

kernel: Linux 5.13.0-51-generic

Used image version: nvidia/cuda:10.0-devel-ubuntu18.04

command to create container:

docker run -it --gpus all nvidia/cuda:10.0-devel-ubuntu18.04 bash

To compile and run the Cuda C code, run in terminal:

nvcc cudaDigitalAnnealing.cu -o ./bin/sbm_cu.o
./bin/sbm_cu.o

This will lead to a test code with dimention = 2, using J matrix = [[0,1],[1,0]],

or simply use the Makefile command to launch the algorithm with python code

make

You can customize different parameters in the python code. More description in "python" section

result:

With dimention = 10,000 and 1000 iterations, the algorithm is able to locate several different local minimums within 3 seconds.
Multiple runs can help locate global minumum of the algorithm. Note that dimention = 40,000 may lead to memory issues?
Also accuracy needs to be tested in the future

python code (main.py)

This is the code that allow users to test different dimentions. to input own data import the main.py file.

Used Python version: 3.6.9 64bit

The code throws a qubo matrix and a state array into the cuda code and calculate the energy