cguzman95
commented
4 years ago GPU linear solver added using the GPU library provided by CASE, now working on improving it. The linear solver follows the Biconjugate Gradient algorithm. Also, the GPU linear solver alongside the GPU ODE solving is computed now in CAMP instead of CVODE (code in cuda folder)
The idea is change the actual KLU SPARSE linear solver from CVODE for another GPU solver. Guillermo O. will handle some libraries developed at CASE department to test. Like the f() function, we expect achieve some speedup even adding the data transfers. Both times (data movement CPU/GPUand computation) will be measure to evaluate correctly the improvement.
It will be test with CB05 and mock_monarch_1 at the moment.