Post B53g9OiNGWNsDBEiZs by RnDanger@infosec.exchange
 (DIR) More posts by RnDanger@infosec.exchange
 (DIR) Post #B53fcvvYlDqFf8vK9g by azonenberg@ioc.exchange
       0 likes, 0 repeats
       
       So I've solved the coupler de-embed issue but am even more confused than when I started.I updated to the latest upstream VkFFT and, after fixing a shader caching issue that caused problems the first time I tried this, have it working.The race condition or whatever it was seems to be gone, although I don't have push descriptors working (that's a separate issue).But it seems to now be behaving consistently in the inverted state: forward to tempBuf1 and reverse to tempBuf3, the opposite of what I had originally, now works 100% of the time in my testing.So is it possible that I had a bug all along and that I was winning the race >50% of the time, causing the backwards order to happen more often and thus give correct results with incorrect code? Or is my math wrong somehow, and I have the forward and reverse path signals backwards?I'm gonna push the fix since it's definitely less broken before but I don't like that I don't understand *why* the fix works.
       
 (DIR) Post #B53g9OiNGWNsDBEiZs by RnDanger@infosec.exchange
       0 likes, 0 repeats
       
       @azonenberg I hope someday you will work on something completely unrelated and suddenly this will make sense