Convert code base to work with CUDA
Opened this issue · 1 comments
nLevin13 commented
Use cuPy, runs numpy on cuda; very slow, unusable results
- parallelizing Cython code (am currently facing issues with GIL, will try to convert code to pure C arrays to avoid the issue)
- Convert repetitive code to for loops for easier parallelization
- Convert pythonic code to static typed numpy arrays for better C performance
- Profile current code to find other areas for performance improvement.