BUG: Fix pocketfft umath strides for powerpc compatibility #29768
Conversation
There was a problem hiding this comment.
Looks good, thanks. ptrdiff_t vs. npy_intp usage differs a bit between functions now, but it doesn't matter.
EDIT: To be clear, the test failures are unrelated, there are some general problems that will be fixed very soon hopefully (maybe it already is).
Unfortunately, tests are failing. It would be nice if you can move the npts_in just to keep things slightly simpler.
numpy/fft/_pocketfft_umath.cpp
Outdated
| #ifndef POCKETFFT_NO_VECTORS | ||
| size_t npts_in = nout / 2 + 1; | ||
|
|
||
| #endif |
There was a problem hiding this comment.
Would be nicer to just move it into the existing #ifndef below, but a small nit, but thanks for fixing the warning!
There was a problem hiding this comment.
Agreed, while you are at it, please do move it inside (it should have been there in the first place, as it is not used outside).
|
@seberg Do we need to change something here at the moment re test failures? |
|
It would be nice to move that one thing, but otherwise should be all good. We can re-trigger CI manually (or close/reopen PR) when things are resolved. |
There was a problem hiding this comment.
Thanks for catching this -- as @seberg noted in the issue, the reason I used ptrdiff_t here was that that was what pocketfft expected. I vaguely recall getting compiler warnings without it, but if this works now without explicit casts, then all fine by me.
numpy/fft/_pocketfft_umath.cpp
Outdated
| #ifndef POCKETFFT_NO_VECTORS | ||
| size_t npts_in = nout / 2 + 1; | ||
|
|
||
| #endif |
There was a problem hiding this comment.
Agreed, while you are at it, please do move it inside (it should have been there in the first place, as it is not used outside).
seberg
changed the title
seberg
changed the title
|
@seberg Thank you for helping here, and sorry for a delay, I did not have much time this weekend. |
* _pocketfft_umath.cpp: fix types Fixes: numpy#29758 Credits to @seberg * _pocketfft_umath.cpp: fix unused variable warning * Move npts_in calculation into #ifndef --------- Co-authored-by: Sebastian Berg <sebastianb@nvidia.com>
BUG: Fix pocketfft umath strides for AIX compatibility (#29768)
5 participants
See: #29758