Todd A. Anderson
Todd A. Anderson
Can you post the result of "numba -s"? Thanks.
NumPy now defaults to fastmath so to match performance, please add fastmath to the njit and install/enable SVML.
I'm not sure that the performance runtime label has been proved here. You'd need to do something that uses the created array which will be cheap for existing Numba but...
> Yes, sorry - here you go: > > ``` > import psutil > import numpy as np > from numba import jit, njit, prange > from numba.typed import List...
I don't believe this is a parallel accelerator problem. ``` from numba import njit, prange, guvectorize, int64, typeof from numba.core.types.npytypes import Record import numpy as np mcsparts = [("delta_z", int64)]...
Later I got to thinking that this might be like tuples where guvectorize doesn't support them but there was so much clamor for parallel accelerator support that we put in...
@mscheltienne Thanks for the pull request. We think there needs to be a couple changes. Both trapz and trapezoid exist at the same time in some of the NumPy version...
@hbina Thank you for the PR. This was discussed in the Numba maintainer meeting and we have some questions and comments. First, how do you see this capability being used?...
The PR that fixes this issue is only going to work with namedtuple, not NamedTuple.
I haven't had time to think about why min should be slower. There is perhaps a bit of overhead getting into the parallel side so if it is only the...