Open mdionisio opened 3 years ago
Yeah - looks like that
Anyone reading this other than @mdionisio ?
I can only say that:
if 'ne10_radix_8_butterfly_float32_c' is well implemented my fix is correct.
The previous one was not bugged for result because the generic function is called but is bugger for performance because the output is computed 2 times.
So with my fix the _c version of fft has better performance.
I'm not able to run test on arm with neon now. But in teory if test continue o run correctly it means that all is ok because the neon version is not changed
I'm not sure about the fix and I'm not able now to execute test because I'm on on intel machine. But only reading the code it seams that there is a missing break.