cplx2real function can only be used with fft4r initialized, and the example creates confusion (DSP-147)

[alexbarcelo]

esp-dsp/examples/fft4real/main/dsps_fft4real_main.c

Line 85 in 10d7418

dsps_cplx2real_fc32(x1, N >> 1);

I was trying to follow the example with a codebase that uses not fft4r but only fft2r functions (which are showcased in this example, I understand). I was getting segfault errors and after gazillion checks in the documentation I realized that cplx2real documentation says that fft4r needs to be initialized. In the example may be ok because it is showcasing both, but in a regular scenario it is not ok.

Assuming that this is indeed the problem (my code started working once I changed to cplx2reC_fc32 function, but I have not validated the numerical aspect, so I may have more bugs, and I may have misunderstood the documentation once again), the example should use the proper function to showcase properly which functions to use in which scenarios.

[dmitry1945]

Hi @alexbarcelo

The dsps_cplx2real_fc32 need initialization for FFT4R. You have to call dsps_fft2r_init_fc32 and dsps_fft4r_init_fc32.
Did you call dsps_fft4r_init_fc32?

[alexbarcelo]

Of course everything is due to my ignorance and incompetence. I was assuming that cplx2real and cplx2reC were "similar". Maybe they are not?

I assumed that the example showcases two different "flows": the fft2r_* way and the fft4r_* way. But from your comment, when using dsps_fft2r mehods I need to do the init also for fft4r. Is this the case for dsps_cplx2real_sc16_ansi? Documentation does not mention 4r in that, there is no 4r init, so I was confused.

[dmitry1945]

@alexbarcelo
The cplx2real use a table and additional memory, that's why we have moved the cplx2real to the fft4r init function.
I will check the functions later.