Add Splines of order 4 to 7 and support for all heFFTe MPI communication backends #799
Conversation
|
Thanks for the improvements! Do you mind splitting into two PRs? I think I'll want someone else to at least co-review the spline details, but the FFT should be quick to merge |
| @@ -97,13 +95,21 @@ struct is_matching_array< | |||
| : public std::true_type | |||
| { | |||
| }; | |||
| enum class FFTCommPattern : unsigned int | |||
There was a problem hiding this comment.
@sr7cb hey Sanil, what kind of communication options would we expect to interface with in fftx? Trying to make sure our interface is generic enough
There was a problem hiding this comment.
Unfortunately I don't have enough familiarity with our MPI backend to provide a concrete answer. Please go ahead with your process and we can update/revisit down the road.
| @@ -39,7 +39,8 @@ void calculateFFT( bool use_default, bool use_params, | |||
| Experimental::FastFourierTransformParams params; | |||
|
|
|||
| // Set MPI communication | |||
| params.setAllToAll( true ); | |||
| params.setAlltoAll( Cabana::Grid::Experimental::FFTCommPattern::alltoallv ); | |||
| params.setAlltoAll( true ); | |||
There was a problem hiding this comment.
You are right, only one is needed. I will remove the boolean variation. Its behavior and usage is commented at the example example/grid_tutorial/10_fft_heffte/heffte_fast_fourier_transform_example.cpp for backward compatibility
Hi @streeve Sam! We also thank you for maintaining Cabana! I don't mind splitting the PRs. Should I leave this one for the FFT and create a new one for the Splines? I had a look into the actions, and I think I have already addressed it with fixes that I can commit once the PRs are at the places. |
Yes, that sounds great! |
This commits add BSplines of order 4 to 7 following the polynomial definitions on deserno1998*
This also extends the available backends available in heFFTe using a enum class to pick the chosen reshape/communication pattern.
This changes were done in cooperation with @rhalver