We are in the process over the next week to implement the Fixed Point algorithms into a 200 MHz ARM9 CPU with MMU and I was wondering if you had any experience/knowledge with possible pitfalls. We are running Fixed Point compilation on PC based system (P4 processor) and are noticing that the audio gets "noisy" and then comes back to a more normal voice. Are you aware of any issues with the fixed point application? Any input would be appreciated (even where else to look or info regarding Fixed Point Implementation). Wayne Klein V.P. Engineering P.S. Since the description above is not really specific ("noisy"), we could record it as well as have scope print-outs if that helps. This is a big project for our company and we are in the process of getting ready for a demo. -------------- next part -------------- An HTML attachment was scrubbed... URL: http://lists.xiph.org/pipermail/speex-dev/attachments/20080214/f1c0197d/attachment.html
Wayne Klein wrote:> We are in the process over the next week to implement the Fixed Point > algorithms into a 200 MHz ARM9 CPU with MMU and I was wondering if you > had any experience/knowledge with possible pitfalls.All I can say is that if your chip supports the ARM5E DSP-like instructions, you can probably get a good speedup.> We are running > Fixed Point compilation on PC based system (P4 processor) and are > noticing that the audio gets ?noisy? and then comes back to a more > normal voice. Are you aware of any issues with the fixed point application?Nothing I'm aware of, though it could happen if your signal is already clipping or close to the clipping point. Send a (short) sample if you want me to have a look at it. Also, compiling with fixed-point debugging enabled (--enable-fixed-point-debug or FIXED_DEBUG) could tell you if there's an overflow problem somewhere. Cheers, Jean-Marc