WIP: QuadDtype finfo support #165
Conversation
There was a problem hiding this comment.
NOTE: In order for this to pass CI, need to build numpy from source. Will do that in follow-up PR
You can probably just add the nightly build to the index: https://scientific-python.org/specs/spec-0004/#test-with-nightly-wheels and then assume that for building.
I dunno if you are OK with a strict NumPy 2.4 requirement, but I suppose for quaddtype it may be just as well.
| } | ||
|
|
||
| /* Store the Sleef_quad value to the provided pointer */ | ||
| *(Sleef_quad *)ptr = val; |
There was a problem hiding this comment.
If you support unaligned, should use memcpy, unfortunately. (Or we change that we promise to not call it for unaligned stuff, which I can get on-board with.)
There was a problem hiding this comment.
Actually thinking about it, I am not sure if this is true or not :), we needed copyswap anyway due to byte-swapping. I am actually happy to say that the data must be aligned.
Right, I was also thinking to make nightly the default build as right now or maybe even in future this repo will directly be using the upstream features (if some modifications be needed from NumPy side) |
This PR refactors the constants exposure and provides the initial finfo support based on the upstream numpy.
NOTE: In order for this to pass CI, need to build numpy from source. Will do that in follow-up PR