Join GitHub today
[MRG] Use GEMM in _update_dict #11420
This is the main take away.
In : import numpy as np In : a = np.random.random((5, 6)).copy(order="F") In : b = np.random.random((6, 7)).copy(order="F") In : np.isfortran(np.dot(a, b)) Out: False
This looks good. Out of curiosity could you please run a quick benchmark to evaluate the performance impact of that fix on a typical (smallish) problem of yours?
Benchmarking shows the time change is pretty close to negligible, which is expected since the same operations are done in either case. Though memory usage should be reduced as we are dealing only in Fortran arrays now instead of converting as before thus avoiding a copy.