>> Once converted to fixed I'm sure speex will be fine on strong arm >> and intel's newer stuff based on that (pxa?).So just how much work are we talking about, here, to convert to fixed-point? Seems like with some clever type substitution and a few inlines here and there it might maybe not be so horrible? But, then, I haven't looked inside of Speex yet, and don't have a sense of scale... --- >8 ---- List archives: http://www.xiph.org/archives/ Ogg project homepage: http://www.xiph.org/ogg/ To unsubscribe from this list, send a message to 'speex-dev-request@xiph.org' containing only the word 'unsubscribe' in the body. No subject is needed. Unsubscribe messages sent to the list will be ignored/filtered.
> So just how much work are we talking about, here, to convert tofixed-point?> Seems like with some clever type substitution and a few inlines here and > there it might maybe not be so horrible? But, then, I haven't lookedinside> of Speex yet, and don't have a sense of scale...My estimate was about 3 weeks of coding [ several hours a day ] ;) So far I have put in a week of work only. Instead of optimizing only the critical parts of the code I started to replace nearly every float with an integer counterpart (except where it doesn't make sense). I did about a third of the encoder so far and verified this for accuracy. I did somewhat less work on the decoder. Christian <p>--- >8 ---- List archives: http://www.xiph.org/archives/ Ogg project homepage: http://www.xiph.org/ogg/ To unsubscribe from this list, send a message to 'speex-dev-request@xiph.org' containing only the word 'unsubscribe' in the body. No subject is needed. Unsubscribe messages sent to the list will be ignored/filtered.
Maybe this is plain stupid or off topic. Could an atleast working fixed-point implementation of Speex be made by using a C++ "faked float" class with overloaded operators? Don't know if this is a common way to solve stuff like this, but my guess is that it would work. //Best Regards, Jonas Tärnström. <p>> > So just how much work are we talking about, here, to convert to fixed-> point? > > Seems like with some clever type substitution and a few inlines hereand> > there it might maybe not be so horrible? But, then, I haven'tlooked> inside > > of Speex yet, and don't have a sense of scale... > > That's not as simple. With floating point, you can take any twonumbers,> multiply them together and not risk an overflow (not quite but it'strue> in the context of Speex). With integers, that's not the case. You have > to know the range for all the inputs and then (usually) do someshifting> according to that. > > My estimation is that it would probably require in the order of 100 > hours for someone (with prior experience) to convert Speex to > fixed-point. Anyone is up to the task? > > Jean-Marc > > -- > Jean-Marc Valin, M.Sc.A., ing. jr. > LABORIUS (http://www.gel.usherb.ca/laborius) > Université de Sherbrooke, Québec, Canada<p>--- >8 ---- List archives: http://www.xiph.org/archives/ Ogg project homepage: http://www.xiph.org/ogg/ To unsubscribe from this list, send a message to 'speex-dev-request@xiph.org' containing only the word 'unsubscribe' in the body. No subject is needed. Unsubscribe messages sent to the list will be ignored/filtered.
> So just how much work are we talking about, here, to convert to fixed-point? > Seems like with some clever type substitution and a few inlines here and > there it might maybe not be so horrible? But, then, I haven't looked inside > of Speex yet, and don't have a sense of scale...That's not as simple. With floating point, you can take any two numbers, multiply them together and not risk an overflow (not quite but it's true in the context of Speex). With integers, that's not the case. You have to know the range for all the inputs and then (usually) do some shifting according to that. My estimation is that it would probably require in the order of 100 hours for someone (with prior experience) to convert Speex to fixed-point. Anyone is up to the task? Jean-Marc -- Jean-Marc Valin, M.Sc.A., ing. jr. LABORIUS (http://www.gel.usherb.ca/laborius) Université de Sherbrooke, Québec, Canada -------------- next part -------------- A non-text attachment was scrubbed... Name: signature.asc Type: application/pgp-signature Size: 190 bytes Desc: Ceci est une partie de message numériquement signée Url : http://lists.xiph.org/pipermail/speex-dev/attachments/20030924/448ff75f/signature-0001.pgp