ENH: Add SIMD implementation for deg2rad #19779
Conversation
56a43a8 to
b87609d
Compare
|
When you reach the stage of asking for review, please include a benchmark that shows the impact of this PR. For an example, see the benchmarking done in #18766 |
|
Hi @mattip thank you for pointing out the benchmark issue. I am actually meeting problems on running the benchmark. |
Sure, as long as it has the needed SIMD features to run the code. You may want to throttle the machine to 80% full speed to get more consistent results. |
There was a problem hiding this comment.
similar to #19780, there's no need to implement new universal intrinsics for deg2rad, just implemented directly inside the source file that defines ufunc inner loop.
Note implement a new universal intrinsic to the interface wouldn't affect on the performance, you have to put it in context.
In the case of deg2rad, there's no need to create a new dispatch-able source for it, you can put just your implementation inside https://github.com/numpy/numpy/blob/main/numpy/core/src/umath/loops_unary_fp.dispatch.c.src.
You will have to edit the following files too:
- core/code_generators/generate_umath.py to activate the simd dispatch and the new unfunc type functions.
- add forward declaration for it in here,
numpy/numpy/core/src/umath/loops.h.src
Lines 189 to 191 in 3544fae
No description provided.