On Tue, 2005-01-11 at 02:07 -0500, Jean-Marc Valin wrote:> > noticed there's a 64 bit integer being used in the long term prediction. > > any chance of using a 32 bit integer without incurring some major side-effects? > > A 32-bit integer may overflow (which is why I used a 64-bit int). Of > course, it's possible to emulate it using 32-bit arithmetic. Otherwise, > there's probably a work-around, but it would require rewriting that part > of the search.32 bits should be a concern, mon cher Jean-Marc, lowcosts are on this side where do you leave ? Claude> Jean-Marc > > > cheers, > > tk > > _______________________________________________ > > Speex-dev mailing list > > Speex-dev@xiph.org > > http://lists.xiph.org/mailman/listinfo/speex-dev > >--
hi, jean-marc: am interested in just integer implementation. is the 64 bit used more for possibility of large values rather than accuracy? could give me a rough idea of the range, typical value and the extremes off-hand? i take it your workaround involves some complex manipulations instead of just using the 32bit to represent the more significant bits of 64-bit integer. if yes, could you throw some ideas? thanks! tk On Tue, 11 Jan 2005 08:22:20 +0100, Claude Brisson <claude@renegat.net> wrote:> On Tue, 2005-01-11 at 02:07 -0500, Jean-Marc Valin wrote: > > > noticed there's a 64 bit integer being used in the long term prediction. > > > any chance of using a 32 bit integer without incurring some major side-effects? > > > > A 32-bit integer may overflow (which is why I used a 64-bit int). Of > > course, it's possible to emulate it using 32-bit arithmetic. Otherwise, > > there's probably a work-around, but it would require rewriting that part > > of the search. > > 32 bits should be a concern, mon cher Jean-Marc, lowcosts are on this > side > > where do you leave ? > > Claude > > > Jean-Marc > > > > > cheers, > > > tk > > > _______________________________________________ > > > Speex-dev mailing list > > > Speex-dev@xiph.org > > > http://lists.xiph.org/mailman/listinfo/speex-dev > > > > -- > >
> jean-marc: am interested in just integer implementation. > is the 64 bit used more for possibility of large values rather > than accuracy? could give me a rough idea of the > range, typical value and the extremes off-hand? i take it > your workaround involves some complex manipulations > instead of just using the 32bit to represent the more significant > bits of 64-bit integer. if yes, could you throw some ideas? thanks!Keep in mind that for narrowband/low complexity, this only represents 8000 64-bit additions per second. You can easily emulate that with 32-bit operations without adding anything significant to the total complexity. If you really want to, it's probably possible to either modify the algorithm or slightly decrease the precision, which will lead to lower quality, especially for weak signals. Jean-Marc> tk > > On Tue, 11 Jan 2005 08:22:20 +0100, Claude Brisson <claude@renegat.net> wrote: > > On Tue, 2005-01-11 at 02:07 -0500, Jean-Marc Valin wrote: > > > > noticed there's a 64 bit integer being used in the long term prediction. > > > > any chance of using a 32 bit integer without incurring some major side-effects? > > > > > > A 32-bit integer may overflow (which is why I used a 64-bit int). Of > > > course, it's possible to emulate it using 32-bit arithmetic. Otherwise, > > > there's probably a work-around, but it would require rewriting that part > > > of the search. > > > > 32 bits should be a concern, mon cher Jean-Marc, lowcosts are on this > > side > > > > where do you leave ? > > > > Claude > > > > > Jean-Marc > > > > > > > cheers, > > > > tk > > > > _______________________________________________ > > > > Speex-dev mailing list > > > > Speex-dev@xiph.org > > > > http://lists.xiph.org/mailman/listinfo/speex-dev > > > > > > -- > > > > > _______________________________________________ > Speex-dev mailing list > Speex-dev@xiph.org > http://lists.xiph.org/mailman/listinfo/speex-dev >-- Jean-Marc Valin <Jean-Marc.Valin@USherbrooke.ca> Universit? de Sherbrooke