The solve efficiency of the C++ interface is lower than that of the matlab about qpswift

[abhishek-pandala]

Hi sucrozzg, Thank you for your interest in our solver. As per your question, I have the following suggestions

1. I am not sure how the cpp interface is slower than Matlab. The mex file in matlab calls the cpp interface itself. So I believe there should be some problem with your cpp code. Is the cpp code being run in debug mode? Also are the inputs (dimensions of matrices, number of constraints, etc) given to the qpSWIFT solver in the correct format ?
2. Since the problems you are comparing in matlab and cpp are different, I am wondering if the input matrices provided in the cpp interface are more dense than the ones used in matlab. qpSWIFT performs sparse LDL decomposition and the performance of the solver heavily depends on the sparsity pattern of the input matrices. Have you checked the solve time for RF_MPC using cpp interface ?
3. qpSWIFT does not perform infeasibility detection and hence any infeasible problem is likely to hit the maximum number of iterations. In addition, max iterations are also reached when the input matrices are ill-conditioned. Check to see if these issues arise in your code.

[sucrozzg]

Thank you, professor. Based on your suggestion, I re-installed qpSWIFT following the tutorial on GitHub. This worked well. The solving time of the CPP interface has reached the same level as Matlab. I think maybe there is some omission in my previous installation steps. This is a good product. Thank you!