I’m adding float64 Neon support to my HIIR oversampling library. Nice performance improvement, as expected. The functions are matching half the speed of their float32 counterparts.
Picture: performance test on a #RaspberryPi5, 64-bit RPi OS, compiled with Clang 18.
The *NeonOld functions seem now slower than regular *Neon for any number of coefficients, I’ll probably remove them if this is caused by a compiler improvement and not just by a more sophisticated CPU architecture compared to Pi3 or Pi4.
[EDIT: it’s related to hardware so I'll keep them]
[#]dsp #audio #cpp
=> More informations about this toot | More toots from Firesledge@piaille.fr
But I have no idea why I’m doing this, really. I shouldn’t encourage processing of 64-bit audio, this is a pure waste of power.
=> More informations about this toot | More toots from Firesledge@piaille.fr This content has been proxied by September (ba2dc).Proxy Information
text/gemini