linkotec / ffts Goto Github PK

I'm looking into using this library for one project. If this fork is actively maintained I'll likely have at least one PR cleaning up some of the build system.

Am I right about this? Results seem to be wrong if I try to use non-POT 2D transforms.

Second question: performance of chirp-z non-POT is similar to the transform of the next POT? So transforming 1025 elements using chirp-z would take roughly as long as transforming 2048 elements?

Migrated from Anthonix's repo as I don't think it is maintined anymore:

ffts_real_nd.c:96 gives: plan->transform(plan, din + (j * Ms0), buf + (j * (Ms0 / 2 + 1)));
buf here is uint64_t* and is 16 byte aligned. This means that in passing buf + x to a function that expects 16 byte alignment, we must have x%2 == 0 which is not the case resulting in segfaults on my machine whenever j is odd

I submitted a PR for FreeBSD / clang support in CMakeLists.txt to anthonix in 2018. Is there any interest here? I can rebase and resubmit.

Even disabling sse and testing ffts_init_1d without intrinsics have wrong answer.
I just run ffts_init_1d. And ffts_init_1d_real and the result is different.