Interest in a Fixed Point library?

soshimo

I have some old 8088 assembler routines for fixed point laying around from WAY back in the day and was thinking of converting them to spin assembly as both an exercise to learn the language and as the building block for some NN classes I'm planning on converting from C++ as they have heavy floating point math usage using non-linear activation functions in the hidden layer. I was wondering if there would be an interest in that or if it's been done already (I check obex and couldn't find anything but floating point routines which will be too slow for me). TIA

Andrew E Mileski

I keep meaning to convert my fixed point (S31.32) library from GNU C. The format provides roughly 9 decimal places of precision, which is ample for many purposes.

I originally wrote it for a MP3 decoder (was using 90% CPU with default math handing, and I brought that down to 11%), and it grew from there. My most recent stab at it was writing a parser, though I got distracted by something before I tested it.

Best of luck. Feel free to borrow from my C code if you want.

soshimo

Thanks Andrew - I'll definitely look into this.

Ale

This would be very useful (the routines in PASM) if it could be added to the MATH article at propeller.wikispaces.org/MATH , I can add them (and describethem) if you want.

Ale