Closed ast closed 5 years ago
Floats are already at the edge in our IQ correction algorithm (120+ dB image rejection). With fast-math, there will be a numerical stability problem in special signal conditions. I use fast-math elsewhere where it's really needed.
Interesting. I will definitly take a look at that.
By the way, what's the reason you have "re-implemented" complex arithmatic? Is complex.h not availble on some platforms you target or did you experience some other problem?
Of course, platform availability. Have a look on this algo. Fun guaranteed!
Every time I learn about some new nice DSP tricks it's like Christmas!
-ffast-math
is not included i-O3
, and SDR doesn't really need strict IEEE754 complience.Something to think about, perhaps.