Hi, I am optimizing GSM AMR reference code. The optimization needs not to be bit-exact, in fact, some quality degradation is permitted. I have done all the obvious things like primitives inlining, autocorrelation approximation, etc. Are there any other mathematical shortcuts to make the codec faster? Right now, the codec is 25% slower that I need it to be. TIA, ~Yaroslav
GSM AMR optimization
Started by ●March 21, 2005
Reply by ●June 20, 20052005-06-20
do you still work with GSM AMR I have some experience about AMR optimizing>Hi, > >I am optimizing GSM AMR reference code. The optimization needs not to >be bit-exact, in fact, some quality degradation is permitted. I have >done all the obvious things like primitives inlining, autocorrelation >approximation, etc. Are there any other mathematical shortcuts to make >the codec faster? > >Right now, the codec is 25% slower that I need it to be. > >TIA, >~Yaroslav >This message was sent using the Comp.DSP web interface on www.DSPRelated.com
Reply by ●August 2, 20052005-08-02
Hi, what is your current speed ?, how powerful your DSP platform is (does it support saturation, round, Q15,..., MAC instruction,...) how many instruction can be grouped to be executed in one CPU Cycle ? I will have a better solution for you based on that information. Thinh Hoang.>Hi, > >I am optimizing GSM AMR reference code. The optimization needs not to >be bit-exact, in fact, some quality degradation is permitted. I have >done all the obvious things like primitives inlining, autocorrelation >approximation, etc. Are there any other mathematical shortcuts to make >the codec faster? > >Right now, the codec is 25% slower that I need it to be. > >TIA, >~Yaroslav >This message was sent using the Comp.DSP web interface on www.DSPRelated.com