Graham Markall

From the perspective of numba-cuda I think this is a positive change and a good chance to make the change to the default. Numba does already support PTDS with the...

@ca5446001-tech, as you've noticed, a fix for this this is still blocked on #6707.

I don't know of a way to make this work. I think the environment pointer going missing is also a bug, though it could be that we ought to report...

Following discussion in the triage meeting - closing the gap on making this work would require loading the first class functions and getting their address, then using the binding's `ll.add_symbol()`...

> Note that the "About" section of the GitHub project still links to pydata. That is the Numba homepage. The "Documentation" links on the Numba homepage link to RTD.

Is there a more cohesive way this change can be applied that doesn't require checking `platform.machine` in many places, and defining the int typ accordingly?

Changing `min` to an addition yields slightly different behaviour: ```python import numba import numpy as np from timeit import timeit numba.set_num_threads(1) n = 3 @numba.stencil def min_kernel(a): return (a[0,0,0] +...

Should `x` and `y` be indexed with an empty index in the assignment to make the types stable between the loop body and the block following it? ```python import numpy...

Commit adding the CUDA headers referring to Attachment A of the CUDA EULA: https://github.com/numba/numba/commit/ad931f81e17dd680516177fa6b22fde17f24c52e#diff-c89f9576dacbf6b7fa07d07780942d62a78b9f09357b499154f9886237656c3f https://docs.nvidia.com/cuda/archive/11.2.2/eula/index.html#attachment-a

PR with an idea for a fix in: https://github.com/numba/numba/pull/10275